Coordination: Its Implications for the Theory of General Linguistics

650 131 128MB

English Pages xii, 318 [328] Year 1968

Report DMCA / Copyright

DOWNLOAD FILE

Polecaj historie

Coordination: Its Implications for the Theory of General Linguistics

Table of contents :
Title
Preface
Contents
1. Introduction
2. Summary
3. The distributional approach to coordinative constructions
4. General properties of coordinations
5. The treatment of coordination in transformational generative grammar
6. The background of the transformational approach to coordination
7. The relation of coordination to constituent structure
8. On the notion 'grammatical function'
9. Outline-sketch of a functional granlmar
10. The description of coordinations in a functional grammar
11. Structural ambiguity in coordinations
12. On the semantics of coordination
Bibliography
Index of subjects
Index of names

Citation preview

ITS IMPLICATIONS FOR THE THEORY OF GENERAL LINGUISTICS

SIMON C. DIK ',, Department of Gtneral Linguistics University of Amsterdam

1968 NORTH-HOLLAND PUBLISHING COMPANY AMSTERDAM

1968 C> NORTH-HOLLAND PUBLISHING CO. -AMSTERDAM No part of this book may be reproduced in any form, by print, photoprlnt, microfilm or any other meaM without written permission from the publisher

PRINTED IN THE NETHERLANDS

To my parents To Annelies



Preface •

I should like to express my sincere thanks to all those who in some way or other have aided me in the preparation of this study, submitted as a doctoral thesis to the Faculty of Arts of the University of Amsterdam. In a way difficult to define in detail this book owes most to Professor Anton Reichling, under whom I had the privilege of working as a student and an assistant over a period of nearly ten years. As supervisor of the thesis Professor Reichling greatly enhanced the value of the text by his acute criticism of earlier versions. I am very much indebted also to Professor A. L. Vos, who gave invaluable advice on the content and style of this study. Professor E. M. Uhlenbeck of the University of Leyden and my colleague Mr. J. G. Kooij also read earlier versions and allowed me to profit by their critical remarks. I am further obliged to those who assisted me by sending reprints of publications, by allowing me to use pre-publication drafts, by providing me with written or personal information on a number of points, or by discussing the subject-matter of this study: Mr. D. M. Bakker, Dr. G. F. Bos, Professor Noam Chomsky, Dr. B. F. Elson, Mr. J. Fidelholtz, Professor Charles J. Fillmore, Mrs. Lila R. Gleitman, Mr. R. D. Huddleston, ..., Mr. W. G. Klooster, Mr. H. Pinkster, Dr. Petr Pitha, Dr. C. J. Ruijgh, Professor Sanford A. Schane, Professor H. Schultink, and Professor William S.-Y. Wang; to the members of the 'Werkgroep Logische Structuur der Grammatica' ('Seminar on the Logical Structure of Gramtnar') for their criticism of an oral presentation of some parts of this study; to Mr. F. Carmiggelt and Mr. H. Pinkster for their assistance in proof-reading; to Mr. D. Betlem for compiling the Index of names: to management and staff of North-Holland Publishing Company for their efficient and rapid production of the book; and to my wife Annelies for unfailing support and assistance. Amsterdam, March 1968

SIMON

c.

DIK

Contents



































vn

1. INTRODUCTION.

































1













1













s













6













11













15

3. THE DISTRIBUTIONAL APPROACH TO COORDINATIVE CONSTRUCTIONS . • • • • • • • • • • • •







17

PREFACE .





1 .1. The study of coordinative constructions . . • 1. 2. On the scope of general linguistics . . . . • 1 . 2 .1. Some historical remarks . . . . . • 1.2 .2. The present-day task of general linguistics

2. SUMMARY









.3 .1. Bloomfield's definition .3 .1 .1. Construction .









































































































17 18 19 19

















20

































21 21

4. GENERAL PROPERTIES OF COORDINATIONS .













25

3 .1. 2. Position . ·3 .1.3. Function • .3 .1 . 4. Form-class 3 .1 . 5. Conclusion .3. 2. Counter-examples to

. . . . . . . . . . . . . . . . . . . . . . . . . . . . Bloomfield's definition

4 .1. .A general definition of'coordination' . • . . • • • • • 4 .1 .1. 'Two or more constituents' . . . . . • • • • • 4 .1 . 2. 'Equivalent as to grammatical function' . • • • • • 4 .1.3. 'At the same level of structural hierarchy' . • • • • • . . . . . . . • • • • • 4 .1 . 4. '.A linking device' 4 .1. 4 .1. Juxtaposition . . . . . . • • • • • 4 .1. 4. 2. Coordinators . . . . . . • • • • • 4 .1 . 4. 2 .1. A criterion for coordinators . . . 4 .1. 4. 2. 2. The patterning of coordinators. . . 4 .1.4 .2 .3. Repetitive, correlative, and heterogeneous coordinators . . . 4 .1. 4. 2. 4. Prepositive vs. postpositive coordinators . . . . . . . 4 .1. 4. 2. 5. Binary and n-ary coordinators. . . 4 .1. 4. 2. 6. The grammatical status of coordinators 4 .1 . 4. 2. 7. The hierarchical status of coordinators 4. 2. The order of the members in a coordination . . . . . . . 4 •.3. Coordination in relation to other types of structure . . . . . 4. 4. Conclusion . . . . • . . . . . . . . . . . .

. .

25 26 27 30 31 32 34 34 41

.

45

. • . . . . .

47 48 51 52 55 58 60

• •

• • •

• •

x

CONTENTS

5. THE TREATMENT OF COORDINATION IN TRANSFORMATIONAL GENERATIVE GRAMMAR

61











Introductory remarks • • • • . . . . • . The inadequacy of constituent structure rules . . . . The introduction of transformational rules . . . . . The transformational principle in its most general form • Inadequacies of· the transformational approach . . . 5. 5 .1. The establishment of derived constituent structure 5. 5. 2. Simplicity . . . . . . . . . . . 5 .5 .3. Coordination and plurals . . . . . . . 5. 5. 4. Referential problems . . . . . . . . 5. 5. 5. Semantic differences . . . . . . . . . 5 . 5 . 6. Irreducible cases . . . . . . . . . . 5. 5. 7. Conclusion . . . . . . • . . . . 5. 6. The use of rule-schemata . . • . . . . . . 5. 6 .1. Rule-schemata in tagmemic theory . . . . . 5. 6. 2. Rule-schemata in transformational grammar . . 5. 7. Recent suggestions for the description of coordinations in transformational generative grammar • . . . . . 5. 7 .1. Sentence conjunction . . . • . . . . . 5. 7 .1 .1. Chomsky . . . . . . . . . 5. 7 .1 .2. Ross &. Lakoff . . . . . . • 5 . 7 .1 . 3. Schane . . . . . . . • . 5. 7 .1 . 4. Koutsoudas . . . . . . . • 5. 7 .1 .5. Conclusion . . . . • . . • 5. 7. 2. Phrasal conjunction • . . . • . . . • 5. 7 .2 .1. Fidelholtz . . . . . . . . • 5. 7. 2. 2. Lakoff &. Peters . . . . . . • 5. 7. 3. Final remarks on the scope of rule-schemata . • 5. 8. Conclusion . . . . . . . . . . . . . •

















































74 74









76

















79 81









87









88

























91 92 93









94

















98 98









































































100 101 102 105 107 109 110 110 112









114

6. THE BACKGROUND OF THE TRANSFORMATIONAL APPROACH TO COORDINATION . . . . . . .









116

5 .1. 5. 2. 5 .3. 5 .4. 5 .5.

6 .1. 6. 2. 6. 3. 6. 4. 6. 5.

Introduction • . . . Aristotle . . . . . Port-Royal • . . . More recent versions of· the Conclusion . . . . .

. . . . . . . . . . . . . . . . . . reduction postulate . . . . . .

61 64

69 72















116





























118 119















124















128

7. THE RELATION OF COORDINATION TO CONSTITUENT STRUCTURE . • • • • • • • • • • • • • •







131

















131 131









134

















7 .1. 7 .2. 7. 3. 7. 4. 7. 5.

The coordinated member = Methodological objections Empirical objections . . Further counter-examples Conclusion . . . . .

single . . . . . . . .

constituent hypothesis . . . . . . • . . . . . . • . . . . . . • . . . . . . •

139 142

CONTENTS

XI

8. ON THE NOTION 'GRAMMATICAL FUNCTION' 8 .1. 8. 2. 8. 3. 8. 4. 8 .5. 8. 6.

.





Bloomfield's view of grammatical function . . • . . . Grammatical function in transformational generative grammar Preliminary conclusions . . . . . . . . . • . Grammatical function in tagmemics . . . . . . . . Grammatical function in Halliday's 'systemic grammlJI'' . . Some general remarks . . . • . . . . . . .



. . . . . .

143

























143 147 154 154 159





160









162









163

































































































164 167 169 170 172 174 175 176 177 178









181

























185 186 187









187









































195 199

10. THE DESCRIPTION OF COORDINATIONS IN A FUNCTIONAL GRAMMAR . . . . . . . . . . . . . . . . •



200





202





204













208 215 223 227

9. OUTLINE-SKETCH OF A FUNCTIONAL GRAMMAR . 9 .1. Basic properties . . . • . • . . • • . . 9. 2. General lay-out of structural descriptions . . • . . 9. 2 .1. Facts accounted for . • . . . . . . . 9. 2. 2. Most inclusive units described . . . . . . 9. 2. 3. Relations between different linguistic expressions . 9. 2. 4. Constituents, categories, and functions . . . 9. 2. 5. The format of structural descriptions . . . . 9. 2. 6. A simplified example . . . . . . . . 9. 2. 7. Categorial vs. functional differences . . . . 9. 3. The selection of functions . . . . . . . . . 9.4. Relative simplicity of a functional grammar . . . . 9. 5. Types of rules needed in a functional grammar. . . . 9. 5 .1. Four basic sets of rules • . • . . . . . 9.5 .2. Continuous vs. discontinuous rules • • • . . 9. 5. 3. Definite vs. indefinite rules (or rule-schemata) . 9. 5. 4. Context-free vs. context-sensitive rules . . . 9. 5. 5. Obligatory vs. optional rules • . . . . . 9. 6. Recursion in a functional grammar . . . . . . . 9. 7. The organization of a functional grammar . . . . . 9. 7 .1. The ordering of the rules and of their application . 9. 7. 2. The notion of 'derivation' in a functional grammar 9. 8. An illustrative example • . • • . . . . . . 9. 9. Conclusion . • . . . . . . . . . . . .

10 .1. 10. 2. 10.3. 10. 4. 10 .5.

A general schema of coordination . . . . . . . . . The incorporation of the schema into the grammar • . . . Some examples . . . . . . . . . . . . . . Some further differences with the transformational description . Some advantages of the functional description of coordinations

11. STRUCTURAL AMBIGUITY IN COORDINATIONS 11 .1. Three types of ambiguity . . . . . . 11 .1 .1. Functional ambiguity . • . . . 11.1.2. Hierarchical ambiguity . . . . 11.1.3. Relational ambiguity . . . . . 11 . 2. Structural ambiguity and functional grammar 11. 2 .1. The inadequacy of IC-analysis . . 11. 2. 2. The functional alternative . . . .

. . . . . . .































































































164 164

190 191 193

228 230 231 236 241 242 246

CONTENTS

XII

12. ON THE SEMANTICS OF COORDINATION .











12 .1. Can semantic aspects be attributed to coordinators? . . . 12 .2. Reichling's theory of semantics . • • . • . . • . 12. 2 .1. Basic principles . . . . . . . . . . . 12 .2 .2. Semantic content and final interpretation . . . . 12 .2 .3. Kinds of linguistic Information • . . • . . . 12. 3. .Application to coordinators . • • . . • . . • . 12 .4. Defining the semantic values of coordinators: logic or linguistics? 12 .5. Semantic values and semantic relations . . • . • . . 12. 5 .1. Semantic values of coordinators . . . . . . . 12. 5. 2. Semantic relations between coordinated members . . 12. 5 . 3. Some final remarks • • . . . . . . . . 12. 6. The relation between grammar and semantics • . . . . 12. 7. Conclusion . • . . • . . . . . . . . . .



. 250 . 250 . 251 . 251

. . . . . . . . . . . . .

. 295

. .

252 251

. 258 . . .

259 270 271

• 281 . 290 .

291































. 296

INDEX OF SUBJEcrs

.































308

INDEX OF NAMES .

































316

BIBLIOGRAPHY .



I. Introduction

1 .1. The study of coordinative constructions

For a variety of reasons the so-called 'coordinative construction' is of special irnportance to general linguistic theory. In the first place, this type of construction seems to be a universal feature of natural languages. Secondly, not only does its existence seem to be universal, but the way in which it is manifested in each particular language also shows a quite general, if not universal pattern. Thirdly, coordination is operative, to a 1 greater or less extent, on many different levels of grammatical structure. Fourthly, this construction imposes some special requirements on the system of rules by which it is to be described, and is therefore directly relevant to the for1nat or model on which grammatical description should be based. Finally, it constitutes one of the sources of the 'immanent pro2 ductivity of natural languages', by virtue of which a potentially infinite nu1nber of sentences can be constructed on the basis of a limited number of structural patterns. This principle of 'creativity', though by no means neglected by earlier scholars, has rightly received special attention in the theory of transformational generative grammar as one of the fundamental characteristics of natural language. · In spite of their importance, coordinative constructions have until quite recently received comparatively little attention. Both in descriptions of particular languages and in the general theory of language, they have generally not been subjected to more than the most casual treatment. Several reasons can be adduced for this neglect. In the first place, the scientific study of syntax as such has developed only recently in modem linguistics. The major achievements of twentiethcentury linguistics lie in the elaboration of the 'lower' levels of the linguistic system, i.e., of phonology and morphology. The ideas developed for the study of syntax were either inadequate to deal with the great 1

I use the term 'grammar' in its ordinary sense of 'morphology and syntax', as opposed to phonology and semantics. 1 For this term and concept, see already Reichling 1952: 77ff., and cf. Reichling 19651 : 85. For the application of this principle in semantics, cf. below section 12. 2. 1.

2

INTRODUCl'ION

complexities met in this field, or turned out to be no more than the tenets of traditional gra1n1nar rephrased in modem ter1ns. The achievement of traditional grammar (i.e. of classical gra1n1nar as it has found its way into the school tradition) should not be underrated. But the somewhat haphazard course of its development, the often non-linguistic disciplines to which it was subordinated, the corresponding vagueness of its defining criteria, its preoccupation with the establishment of the 'parts of speech', the practical airns to which it was usually directed, the restriction first to Greek and Latin and later to a quite Ii1nited group of closely related languages - all these characteristics contribute to the i1npossibility of regarding it as a full-fledged and generally applicable theory of grammar, whatever the merits of certain of its subparts may be. The tendency, therefore, to 'idealize' traditional gra1n1nar, which at present is to be 3 observed in the writings of transfor1national generative gra1n1narians, seems to me just as pointless as the opposite attempts of certain adherents of structuralism to show that it is completely deprived of any scien4 tific value. A second reason for the neglect of the linguistic study of coordinative constructions is the view of certain scholars that coordination and subordination are pri1narily logical or psychological, and not, or only sec5 ondarily, linguistic phenomena. From this point of view, one concentrates on the logical relations pertaining within and between propositions, or on the psychological relations between concepts, judgments etc. rather than on the gra1nmatical relations within linguistic structures. These structures are then interpreted in the light of the (psycho-)logical relations found, but there is no guarantee that there is a one-to-one relation between linguistic structures on the one hand, and logical or psychological organization on the other: items which are logically or psychologically 'coordinated' may be linguistically expressed in subordinative constructions, and vice versa. Consequently, the explanation of the linguistic facts in terms of logical or psychological phenomena almost automatically leads to a division of linguistically similar items and to a grouping together of linguistically diverse ones, on the basis of extra1

See, most recently, Chomsky 1965 and compare Kraak 1966, chapter 1. ' Compare De Saussure's largely negative opinion on the achievements of traditional grammar (19221 : 13, 118, 185-7). 6 The former view is taken by Sandmann 1950; the latter is present, e.g., in the work of several French linguists, notably of Bally and his followers. There are, of course, many individually different positions on this matter, but for the purpose of my argument I need not distinguish and analyse these in detail.

1.1. THE STUDY OF COORDINATIVE CONSTRUCl'IONS

3

linguistic considerations. In contradistinction to these various approaches, I take the view that the terms coordination and subordination can be appropriately defined and meaningfully used in a purely linguistic framework, i.e., can be adequately applied to the description of linguistic structures quite apart from the logical or psychological phenomena which can be presumed to be in some way connected with them. These possible connections cannot be used a priori in the description of the linguistic facts concerned, but should be studied a posteriori when both the linguistic and the logical or psychological facts have been established in ter1ns of their own nature. Logical, psychological, and linguistic phenomena are no doubt closely interrelated, but their areas are far from being coextensive. Any preliminary identification of the three domains cannot but obscure our insight into their own particular features. A third, and quite interesting reason for the neglect of coordinative constructions lies precisely in the characteristic which makes them so important to general linguistic theory, viz. their universal nature. Just because in the coordinative constructions of different languages resemblances exceed differences, it has generally been felt that a detailed analysis of these constructions could be dispensed with. For teaching practice this is indeed justified. In learning a foreign language we need no explicit description of the rules governing coordination, since to a large extent we are able to assi1nilate the foreign patterns ourselves on the basis of our knowledge of our native language, coupled with a limited experience of the constructions of the language learned. But from this it cannot be concluded that these constructions as such pose no problems to grammatical description if the aim is not to provide a practical teaching tool, but a complete, explicit, and theoretically coherent account of a particular language. The only conclusion warranted by this state of affairs is that apparently such constructions belong to the deeper core common to most or all languages and are, therefore, of particular relevance to the general definition of 'natural human language', in other words, to general linguistic 6 theory. It is no doubt a great merit of Chomsky's theory of generative grammar to have clearly for1nulated the final aim of linguistics as the complete and explicit description of all natural languages, and to have emphasized that this aim can only be reached on the basis of a general theory of what •

On this point, see Bierwisch l 966b: 65ff, •

4

INTRODUCTION

language is and how it is organized. Only a complete description will give a full insight into the incredible intricacies of linguistic systems. And only explicit descriptions can hope to achieve this completeness in a scientifically acceptable way, while at the sarne tirne allowing for a precise statement of the problems involved in linguistic description. Only within the framework of complete and explicit descriptions are we able to make reasoned guesses about what is universal in human language, what is common to certain groups of languages, and what is particular to individual specimens of human language. On this basis we may hope to be able to ascertain how 'language' is manifested in 'this particular language', 3 which, in the words of Reichling (1965 : 56), is the ai1n of general linguistic theory. In this way, too, we come to realizetheinfactquitecomplicat.. ed laws governing the structure of seemingly si1nple linguistic units, such as coordinative constructions. It is no coincidence, therefore, that the rise of Chomsky's theory has also occasioned a proliferation of studies on coordination in different languages. Most of these studies elaborate, in different ways, a suggestion contained in Chomsky's Syntactic structures (1957a: 35-6), which I shall have occasion to discuss below. For a survey of the relevant publications I refer to the Bibliography and to the discussion of their various results contained in the body of this book. A careful consideration of these studies shows, among other things, that transformational generative grammarians are far from reaching agreement on the appropriate approach to the desc:fption of coordination. In fact, the latter may well be regarded as a test-case for transfor1national theory. In my opinion, the problems encountered in this field 1night lead to major revisions even within the transformational framework. If, therefore, I take in the present study a prevailingly critical attitude towards the results of these contributions, it is not because I disagree with the aim of complete and explicit linguistic description set up by Chomsky and his followers, but rather because I am of the opinion that transformational theory as developed so far will not be capable of reaching this aim. I fully agree with 'generative grammar', but I am quite 7 critical of many aspects of 'transformational grammar'. The question is, therefore, whether other theories can be developed which have the same or better descriptive power than transformational theory, while lacking •

7

For a further elaboration of this point, cf. my article 1967.

-

-

- - - - . . - - - - - - - - - . . . . . . . . - - - ... ~ ................, ....... -

1.2. ON THE SCOPE OF GENERAL LINGUISTICS

s

the inadequacies of the latter. The present study may be read as an attempt to explore this question. It might be objected against the results of this study that I reject parts of a (partly) formalized theory while not replacing it by a theory which is f or1nalized to the sa111e extent. But in the first place it is doubtful whether the alternative contained in chapters 9 and 10 is indeed inferior to transformational theory in this respect; in the second place, this is, in my opinion, a comparatively uni1nportant matter. For anyone who is aware of the present stage of development of linguistics it must be clear that linguistic theories are not adequate just because they are for1nalized, nor inadequate because they are not. What we want to do is to take certain steps towards a theoretical framework adequate to deal with the facts of language. If this attempt is successful, then for1nalization is only a matter of ti1ne and ingenuity. Premature for1nalization, on the other hand, has certain dangers of its own, since for1nalized systems tend to develop into self-contained organisms, independent of the facts they were designed to account for. The si1nple fact that within such systems a multitude of formal operations and manipulations is possible is then easily taken as a 8 proof of their scientific adequacy. But the only test of the validity of a linguistic theory is whether it provides the means for a satisfactory description and explanation of the facts of language. Furthern1ore, this is a study in general linguistics. Many points of general linguistics, however, can as yet only be discussed in an informal way, since linguistics has not proceeded far enough to make a more formal approach feasible.

1. 2. On the scope of general linguistics It may not be superfluous to add some remarks on general linguistic theory, since the scope of this discipline itself is not yet universally agreed on. The fundamental fact on which general linguistics is based is, of course, that all natural human languages have certain features in common, that they are all variations on one universal theme. The aim of general linguistics is, in the first place, to account for these universal features. '

Cf. the warnings of Black in Jakobson (ed.) 1961: 254, and Bar-Hillel 1966: 399.

p

-· ,-,-........... .

6

INTRODUCfION

1 . 2. 1. Some historical remarks The study of general linguistics, which was firmly established in Europe by De Saussure, has not been the centre of interest of pre-Chomskyan American linguistics, though in practice the work of Boas, Sapir, Bloomfield and their followers has had many invaluable results of a quite general impact. The main emphasis, however, was put on descriptive linguistics, for understandable, though not fully justifiable reasons. The confrontation of the traditional linguistic tools developed for the Indo-European languages with the quite differently structured Amerindian languages soon showed these tools to be inadequate in many respects. As a result of this, the idea developed that the idiosyncracies of human languages so far exceed their common properties that it is a necessary requirement of linguistic methodology that any language should be approached without 9 any presuppositions about language in general. This idea was termed the 'post-Boasian fallacy' by Teeter, in an article containing an illuminating critical analysis of the development of American linguistics since Boas {Teeter 1964). An 'inside' view of certain aspects of this development can be found in Joos' introduction to the collection Readings in linguistics (Joos ed. 1957: v), where it is said, among other things, that 'American linguistics got its decisive direction when it was decided that an indigenous language could be described better without any preexistent scheme of what a language must be than with the usual reliance upon Latin as the model. It is usual to name Franz Boas in this connection ... The abandonment of deduction in favor of induction has never been reversed.' In a comment on one of the papers contained in the collection, Joos writes: 'Trubetzkoy phonology tried to explain everything from articulatory acoustics and a minimum set of phonological laws taken as essentially valid for all languages alike, flatly contradicting the American (Boas) tradition that languages could differ from each other without limit and in unpredictable ways, and offering too much of a phonological explanation where a sober taxonomy would serve as well' (ibid. 96). It is quite possible that Joos has somewhat overstated the prevailing 10 American attitude, but at least his remarks reflect views widely held in American linguistics between Boas and Chomsky. In the same light we • Cf. Reichling 1948 : 16. 10 As Teeter argues (1964: 200 fn. 10), it is not correct to attribute this opinion without qualifications to Boas.

-

--

- - - - - -.

- ,...

- -

· ~·

.

- -

-. - - -r-------..;, ~

..

...

~

= r•. '' .;,. . .

1.2. ON THE SCOPE OF GENERAL LINGUISTICS



!

.... _

~.

7

should read such statements as the foil owing: 'Speech is a human activity that varies without assignable limit from social group to social group' (Sapir 1921: 2), and: 'Hindu grammar described the Sanskrit language completely and in scientific terms, without prepossessions or philosophical intrusions. It was from this model that Western scholars learned, in the course of a few decades, to describe a language in terms of its own structure' (Bloomfield 1939: 2). It was generally thought that it was at least inadvisable to strive for universal definitions of the categories of human language, and the cautious attitude towards generalization prevalent in American linguistics was codified in Bloomfield's famous dictum: 'The only useful generalizations about language are inductive generalizations. Features we think ought to be universal may be absent from the very next language that becomes accessible' (1933: 20). But Bloomfield was careful to add: 'The fact that some features are, at any rate, widespread, is worthy of notice and calls for an explanation; when we have adequate data about many languages, we shall have to return to the problem of general grammar and to explain these similarities and divergences but this study, when it comes, will be not speculative but inductive' (ibid.). Even though I would not wish to accuse pre-Chomskyan American 11 linguistics of 'lack, or neglect of theory', it seems at least warranted to say that it chose the wrong theory, and that this choice is reflected most clearly in its attitude towards generalization. This view of generalization was mistaken in two respects, namely (i) in that it was thought that generalization necessarily came after the complete description of particular facts, and (ii) in that it was judged to be attainable only by purely 12 inductive methods. It is worth noticing that these two points were already forcefully criticized by Reichling in his programmatic inaugural lecture entitled 13 'What is general linguistics?' (1948). General linguistics is, according to Reichling, neither a priori nor a posteriori, but it has to be developed alongside the descriptive study of particular languages. Linguistic description presupposes and results in general insights about the nature of language (cf. ibid. 16). We might express this by saying that there is a 11 11 11

Cf. Chomsky 1965: 193 fn. 1. See also Bach 1966a: 121 ff. Compare Reichling 1935: 4-7.

8

INTRODUCTION

mutual 'feedback' between linguistic theory and linguistic description. Each reinforces, corrects, and develops the other to an ever higher degree of adequacy. The differences between individual languages should not be allowed to o.bscure their underlying unity (ibid. 16-7), and since any individual language comprises the features necessary to language in general, there should be a theoretical possibility to isolate these features 14 even in the analysis of any one particular language (ibid. 17) • In practice, of course, comparison of different languages is the proper method for establishing their common features. The whole matter is succinctly formulated by Reichling in the following words: ' ... every explanatory inquiry of the separate languages must be combined with a previous general linguistic inquiry into the universal categories in the use of language and the universal factors that control the use of language. In this way the explanation of the non-universal categories is given also on the basis of deduction from the universal characteristics of language, characteristics which, however, have themselves been found inductively.' (ibid. 18, my translation). No further argument seems to be needed to refute the two basic misconceptions of 'generalization' noted above sub (i) and (ii). In the mean time, it took some ten more years before the incorrectness of its notion of generalization was forced upon American linguistics, and before it was seen that, if the frarnework of traditional grammar is not fully adequate to deal with all the diversities of natural languages in a general and consistent way, the search for a more adequate general theory of language is far from being a chimerical enterprise, and is even called for if we want to arrive at a real understanding of human language. The break-through of this attitude in American linguistics is largely due to Chomsky, though the work of Greenberg, Pike, Hockett, and others has certainly contributed to this development. It has now become clear that Bloomfield's purely inductive standpoint was too scrupulous and that, though general theoretical statements should of course be based on as large a body of empirical observations as possible, it is also feasible to derive further general laws within the general theory in a purely deductive 15 way, laws which can then be subjected to empirical verification. Just like any other empirical science, therefore, linguistics involves 1'

This point is strikingly paralleled by Chomsky's claim that 'it is surely true - and there is nothing paradoxical in this - that a single language can provide strong evidence for conclusions regarding universal grammar.' ( 1967: 43 7). 15 For recent remarks on this point, see Greenberg 1966: 62.

1.2. ON THE SCOPE OF GENERAL LINGUISTICS

9

inductive and deductive procedures side by side. Both induction and deduction are based on if-then relationships. The difference is that whereas a statement arrived at deductively follows logically (or necessarily) from its premisses, induction presupposes an intellectual 'jump' to a higher level of generalization (cf. Bach l 966a: 11 Sff. ). If by 'inductive' we would mean no more than 'solely based on the observation of the facts', this would be tantamount to excluding induction from the status of a serious (or even a possible) scientific procedure. 'Without the activity of the inquiring mind, the facts are not so kind as to line up in battle-array of their own accord.' (Reichling 1935: 6, my translation). But in the only sensible interpretation of 'induction' the activity of the investigator, his 'creative invention' of the system behind the facts, is already included in its definition. Induction and deduction are not the key-words of two incompatible methodologies, but represent two essential aspects of the only possible method of the empirical sciences. Induction and deduction, furthermore, are used at different levels of generality. Simplifying somewhat, we should distinguish between: (i) induction from fact to statement (pri1nary induction), (ii) induction from statement to statement (secondary induction), (iii) deduction from statement to statement (primary deduction), and (iv) deduction from statement to fact (secondary deduction). The difference between. these steps can be sketched as follows: (i) primary induction: if such and such is the nature of the facts observed, then such and such a (hypothetical) statement can be made about them. (ii) secondary induction: if such and such statements about the facts are true, then such and such a further hypothesis can be set up. (iii) primary deduction: if such and such statements are true, then such and such a further statement must also be true. (iv) secondary deduction: if such and such statements are true, then there must be further linguistic facts of such and such a nature. By endorsing the view of the mutual dependence between theory and description, American linguistics has renewed its connection with the European linguistic tradition. Indeed, Chomsky's remarks on the relation 16 of general linguistic theory to the description of particular languages often read like a restatement of the view which has been implicit in European 'structural' linguistics for several decades now, and which has perhaps been most clearly explicitated in De Saussure's Cours de linguistique •

1



See, e.g., Chomsky 1965: chapter 1, Katz & Postal 1964: l 59ff.

10

INTRODUCTION

generale (1916), in Hjelmslev's Prolegomena to a theory of language (1943, English translation 1953), and in Reichling's inaugural lecture, which I have paraphrased above. An important difference is that in European linguistics the general theory of language is not necessarily viewed as a model of the innate language-learning capacity of the child, but as just what it is: an account of the general properties of the set of objects called natural human languages. Chomsky's equation of these general properties with innate features of the human organism is based on a further, psycholinguistic hypothesis concerning the ontological status of what is described by general linguistic theory. This hypothesis amounts to the idea that the general properties of language are not learned, but known in some sense from the very start. The value of this idea can only be ascertained by psycholinguistic investigation into the nature of language learning. It is independent of general linguistic theory as such, i.e., the general properties of natural languages will be the same no matter whether they are innate or learned• Moreover, there is no single proof so far that linguistic universals indeed constitute properties of the human organism: 'there is no information of any general extralinguistic sort that can be used, at present, to support the assumption that some principle of universal grammar is learned, or that it is innate, or (in some manner) both. If linguistic evidence seems to suggest that some principles are unlearned, there is no reason to find this conclusion paradoxical or surprising.' (Chomsky 1967: 415). But the point is precisely that linguistic evidence can never have any direct bearing on this problem. Linguistic evidence can indeed show that the systems of natural languages are inordinately complex; but it cannot show that the human mind is incapable of coping with such complexity in learning. Chomsky's hypothesis rests on an arbitrarily restrictive view of the creative power of the human intellect. It is much more plausible, to my 1nind, that the human child is born with an extremely powerful capacity of invention, which enables it to 'recreate' the language of its environment for itself in the course of a few years, than that it is inherently gifted with any substantive linguistic knowledge. It is in this vein that Reichling has argued that languageleaming is 'language-creation', i.e., that it consists not in passive assimilation in any behaviourist sense, but in active participation and creative thinking on the part of the learner (1935: chapter 3). At any rate, we can leave Chomsky's speculations aside, since general linguistic theory is in no way affected by not postulating that it describes an innate mental reality. The point at issue here is related to the other psycholinguistic

--~ --------- · ~.. ~ ..... -"""""-·- ·

~--~..

11

1.2. ON THE SCOPE OF GENERAL LINGUISTICS

hypothesis connected with the theory of transfor1national generative gra1n1nar, according to which a linguistic description is a model of the tacit linguistic knowledge (the competence) of a mature native speaker. I return to this point below (section 5.1.). For criticism of the various psychological hypotheses underlying several notions of transforn1ational generative grammar, see HiZ 1967: 71, Harman 1967, Matthews 1967, Putnam 1967, and cf. my remarks in 1967: 358ff.

1. 2. 2. The present-day task of general linguistics 17

Linguistics has to describe and to explain the facts of language. To describe a fact (in a scientific sense) means to make a statement in which all and only the properties of this fact are in some way represented. A description is more adequate or less so, according to the degree in which it succeeds in accounting for all the properties of the fact described, according to its own internal simplicity, and according to its degree of consistency with the total description of which it is a part. In order to be capable of evaluation, a description should be as complete and explicit as the fact described allows. Description is a relation between fact and statement of fact. A complete linguistic description is a set of statements completely covering the facts of a particular language. General linguistic theory is a description of the general properties of human languages. To explain a fact is to show that it is a particular instance or effect of 18 a more general fact. There is no explanation in any absolute sense. Rather, explanation is a relative concept, denoting the establishment of a relation between one fact and another, more general fact. We can explain a feature of a particular language by showing it to be an instance of a general feature of human language. And we can explain the general features of human language if we can show them to be conditioned by general features of the human constitution, in particular, human mental and physiological properties. The latter kind of explanation, however, 17

For description and explanation as the basic tasks of any empirical science, cf. Hempel 1952: 1, 9, 20ff. 11 Cf. Hempel 1952: 9. Description and explanation, if taken in this sense, are fundamentally different, but complementary notions. Joos represents the earlier American attitude when he says: 'When the facts have been stated, it is perverse or childish to demand an explanation into the bargain.' (Joos (ed.) 1957: v). On the contrary: when the facts have been stated, it still remains to show their interrelations and dependences in the light of a general theory of language, i.e., to 'explain' them.

2 .._ ................ ._ .. .

12

INTRODUCTION

is not as such within the competence of the linguist, but should rather be explored by close collaboration between linguists and psychologists. The final aim of linguistics is, on the one hand, to arrive at complete and maximally adequate descriptions of all known human languages, both in their synchronic status and in their historical development, as far as the latter is recoverable. On the other hand, it aims at the development of a systematic body of interconnected statements in which any possible 19 generalization about human language is achieved. A subpart of this system will be constituted by those statements which describe the universal properties of natural languages. We can use the ter1n 'particular linguistics' for the description of particular languages. 'General linguistics', however, can be and has been defined in different ways. According to the narrower definition, general linguistics would only account for the truly universal features of natural languages. On the wider definition it is concerned in principle with any linguistic fact not restricted to just one particular language. Should we adopt the narrower definition, then the question arises how those facts should be accounted for which, though not occurring in all languages, still characterize a proper subset of them with a lower limit of two languages and an upper limit of all languages but one. To illustrate the irnportance of such general, though not universal facts, the following example may be adduced. Suppose we set up the hypothesis that 'the word, defined in such and such a way, is a universal feature of natural languages'. If we now come across a language, say L 1, in which no elements are found to which the definition concerned is applicable, then obviously this universal is falsified. If, furthermore, no definition of the word can be found which is also applicable to the elements of L,, then clearly the conclusion must be that the word is not a universal of natural languages under any possible definition. Still, from this situation it could not be concluded that 'the word should be defined 20 for each language separately'. This would only be true if there were no two different languages to which the same definition of 'the word' could 1•

This program was already sketched by De Saussure (19221 : 20). 1 ° For an example of this widely held opinion, cf. Robins 1964: 193, and my comments in 1966: 406-9. Cf. also Halliday 1961: 252: 'in description, all languages have sentences and all languages have words - but the ''sentenceness'' of the sentence and the ''wordness'' of the word do not derive from the theory of grammar.' From my point of view, this is a contradiction in terms.

1.2. ON THE SCOPE OF GENERAL LINGUISTICS

13

be applied, i.e., if there were no general category 'word' at all. As a matter of fact, however, the situation described still allows for a quite si · cant generalization, viz. 'the word, as defined in such and such a way, is a feature of all languages except L 1'. Should this statement not be for1nulated in the general theory of language, then (i) the theory would be deficient in not having incorporated all possible generalizations, and (ii) no basis would have been established for explaining the 'deviant' structure of L, in this respect. The conclusion to be drawn from this example is that apparently there are si · cant linguistic generalizations which do not represent universals of linguistic structure. These generalizations do not belong as such in the descriptions of particular languages; nor do they belong to the universal theory of human language, if this theory is understood as containing only the properties co1n1non to all languages. Rather, they belong to the intermediate area of statements which, though not applicable to all languages, still exceed the li1nits of particular languages in a significant way. It is for these reasons that 'general linguistics' should be taken in the wider sense noted above. It will then include, as a special case, what may be called 'universal linguistics', equivalent to 'general linguistics' in its 21 narrower definition. · An optirnal description of any one particular language will incorporate not only its particular features, but also its general and universal characteristics in such a way that their general or universal nature is implicitly or explicitly indicated. Just as we can require from general linguistics that it must achieve maximum generalization on any possible point, so we can require a linguistic description of a particular language to be in maxim.urn har1nony with the general theory of language. Only then will such a description show not only how this language is structured, but also to what extent it is a particular manifestation of language in general. The general theory of language is presupposed in any attempt at linguistic description, just as a certain number of linguistic descriptions is presupposed for the establishment of general linguistic theory. The development of linguistics does not, for these mutual dependences, involve vicious circularity, but is rather characterized by the familiar 'spiral' movement characteristic of any empirical science. The fact that certain linguists 11

A similar distinction between 'idiosyncratic', 'general', and 'universal semantics' was proposed by Bar-Hillel in a lecture read in Amsterdam, November 1966.

14

INTRODUCl'ION

concentrate on the general theory of language, while others restrict themselves to the description of (parts of) particular languages is no indication of any discontinuity in these various enterprises, but rather a consequence of the truly im1nense task which linguistics has set itself. The task of general linguistics in its present state is, therefore, to attempt to arrive at as many generalizations about linguistic structure as our present knowledge of particular languages allows; to systematize these generalizations and establish their mutual correlations and dependences; to deduce from the partial systems obtained further generalizations not based on direct extrapolation of existing descriptions, and to test these against such empirical material as is, or can be made available; to deterrnine, on the basis of the results obtained, the general properties of the linguistic descriptions of particular languages in such a way that the latter, while arriving at ever better accounts of the languages concerned, also exhibit their general and universal properties in a maximally adequate way. General linguistics, if viewed in this way, is the condition not only for arriving at ever better linguistic descriptions, but also for establishing a general typology of natural languages, including, as a special case, their possible genetic relationships. For the establishment of the latter, a general theory about the development of natural languages in the course of time is of course indispensable. Such a theory is an integrated subpart of general linguistics, even if the focus is primarily on language systems as utilized as tools of communication within a certain linguistic community, and only secondarily on the historical changes which these systems undergo. It is to various aspects of the general programme sketched above that this study is meant to be a contribution.

2. Summary

Chapter 3 In trying to arrive at a general definition of 'coordinative construction', I start with a critical examination of Bloomfield's essentially distributional approach, and show it to be unsatisfactory under any of its possible interpretations. Chapter 4 I then suggest a tentative definition of 'coordination' and discuss some of its implications. Some criteria are advanced for the recognition of coordinative constructions and a survey is given of possible coordinative patterns. Some further remarks are made on the gra1n1natical status and classification of coordinating particles, and the problem of the hierarchical analysis of coordinations is discussed. This section thus contains a survey of the range of facts to be accounted for. Chapter 5 This chapter contains an extensive criticism of the treatment of coordinations in transformational generative grammar. It is shown that the transformational description of these constructions leads to results which are unsatisfactory in several respects, and that certain types of coordinative construction cannot in principle be accounted for in this way. It is demonstrated that within transformational generative grammar itself there is a gradual development leading away from the purely transformational approach towards a description by means of a quite different type of rules. The conclusion is that the use of these rules can be generalized in such a way as to lead to a completely nontransformational description of coordinations which is not inferior to the transformational one in scope and superior to the latter in that it lacks most of its artificialities and complications. · Chapter 6 In this section it is argued that the transformational treatment of co ordinations can be understood against the background of certain tenets of

16

SUMMARY

(traditional and modern) logic and has, therefore, certain limitations which are due not to linguistic, but to logical presuppositions.

Chapter 7 Next, I investigate the relation between coordinative constructions and constituent analysis, starting from and comrnenting upon Chomsky's hypothesis according to which there is a significant dependence between the two. It is found that no such dependence can be ascertained except on statistical grounds. Chapter 8 After an examination of several of its current interpretations, it is shown that the notion of 'grammatical function', which has gradually emerged in the preceding sections, could be accepted as an irreducible primitive of grammatical theory. Chapter 9 This leads to a theory called 'functional grammar', of which an outlinesketch is provided, including a survey of the different types of rules essential to it, the primitive notions on which it is based, and the character of the structural descriptions which it deter1nines. Chapter JO It is then demonstrated by means of examples how coordinative constructions would be described in a functional gra1nmar. Chapter 11 A general treatment of possible structural ambiguities in coordinative constructions provides, among other things, further arguments for the superiority of a functional grammar to both constituent grammar and transformational grammar. Chapter 12 In this chapter, several problems concerning the semantics of coordinative constructions are discussed. The question 'do coordinators have semantic aspects?' is answered in the affirmative, the question 'does linguistics need logic to describe these semantic aspects?' in the negative. Then follows a general description of the semantic aspects of coordinators and of various semantic relations between the coordinated members. This leads to certain general considerations on the interaction of grammatical and semantic rules within a complete linguistic description.

3. The distributional approach to coordinative constructions

3 .1. Bloomfield's definition Since Bloomfield's Language (1933), coordinative constructions have often been treated as a special case of endocentric constructions. An endocentric construction was defined by Bloomfield (ibid. 194) as a construction belonging 'to the sa1ne for1n-class as one (or more) of the constituents.' Or, somewhat more precisely: 'If a phrase has the same function as one or more of its im1nediate constituents, it is an ENDOCENTRIC PHRASE and has an ENDOCENTRIC CONSTRUCTION' (Bloomfield apud Bloch & Trager 1942: 76). Bloomfield further made the following distinctions: 'Endocentric constructions are of two kinds, co-ordinative (or serial) and subordinative (or attributive). In the former type the resultant phrase belongs to the same form-class as two or more of the constituents. Thus, the phrase boys and girls belongs to the same form-class as the constituents, boys, girls; these constituents are the members of the coordination, and the other constituent is the coordinator. Someti1nes there is no coordinator: books, papers, pens, pencils, blotters (were all lying ... ); sometimes there is one for each member, as in both Bill and John, either Bill or John. There may be minor differences of forrn-class between the resultant phrase and the members; thus Bill and John is plural, while the membersareeachsingular.'

( 1933 : 195). It is clear, therefore, that Bloomfield considered a coordinative construction as an endocentric construction belonging to the same f or1n-class as two or more of its (immediate) constituents. In evaluating this definition, I first draw attention to the fact that the . defining feature by which Bloomfield at first characterizes coordinations (' ... belong to the same forn1-class ... ', 194) seems to be denied to them on the very next page ('There may be minor differences ofform-class ... ', 195). It may be asked, therefore, if this is only a seeming contradiction or a real one. We can only answer this question by a consideration of Bloomfield's concepts of (I) construction, (2) position, (3) function, and (4) form-class,

18

THE DISTRIBUTIONAL APPROACH

which are all involved in this matter. It is my opinion that in spite of Bloomfield's careful attempts to define these notions precisely, each of them contains some inherent obscurities, which, further1nore, carry over from (1) to (2), from (2) to (3), and from (3) to (4) (since they are, in this order, defined in terms of each other), resulting in a cumulation of vagueness in the notion 'forrn-class' on which the contradiction in the definition of coordination is based.

3 . 1 . 1. Construction A construction was defined originally by Bloomfield in ter1ns of the order of constituents in a complex forrn and the (grammatical) meaning corresponding to this order (1926: 157-8). Later, he gave a more advanced and general definition: 'Whenever two (or, rarely, more) forrns are spoken together, as constituents of a complex forrn, the gra1n1natical features by which they are combined, make up a construction.' (1933: 169). These gra1n1natical features comprise order, modulation, phonetic modification, and selection (ibid. 163-5). Among the examples of constructions, there is frequent mention of the English actor-action construction, characterizing such forms as (i) The man sleeps and (ii) The men sleep. But it is easy to see that (i) and (ii), though gra1n1natically similar, also contain differences in gra1111natical features (selection of singular vs. plural noun phrase, difference in agreement, cf. ibid. 190-1). Therefore, according to the definition of 'construction', (i) and (ii) would be different constructions. But then (i) and (ii) are the same construction and they are different constructions, which is a contradiction. Therefore, either the definition of 'construction' is wrong, or the 'actor-action construction' is not a construction but a class of constructions which are equivalent in some way. If The man sleeps and The men sleep are to have the same construction, one cannot define 'construction' as: 'the gra1n1natical features by which two or more forms are combined'. Rather, one would have to resort to some such definition as: 'Two linguistic expressions have the same construction if and only if they have the foil owing grammatical features in common: ... ' Such a definition presupposes a distinction among the grammatical features between those which are relevant and those which are irrelevant to the notion 'construction'. The more grarn1natical features one selects as relevant, the less general are the constructions arrived

3.1. BLOOMFIELD'S DEFINITION

19

at on the basis of these features. Bloomfield, however, does not provide criteria for determining such a distinction.

3 .1.2. Position Bloomfield's notion of 'position' should not be taken as: 'place within a particular form', but as: 'place or slot within a construction' (cf. 1926: 158, 1933: 185, etc.). Thus, he mentions the positions of actor and action in the actor-action construction (ibid.). The notion of 'position', therefore, is affected by the arnbiguity of the notion 'construction' noted above, or, more specifically, by the fact that we do not know exactly at what level of generaliiation we should take this notion. Further1nore, since 'position' is defined in terrns of 'construction', and 'construction' in ter1ns of 'gra1n1natical features' (including order), position is i1nplicitly tied to order. Strictly speaking, therefore, we could not say that John has the sa1ne position in John is reading, Is John reading?, Will John be reading?, etc., though in each of these sentences John is 'actor'.

3 .1.3. Function If 'function' were si1nply equated with 'position', it would already comprise the i1npreciseness of the latter. In fact, however, a further complication arises in that someti1nes one position is taken to define the function of a f or•n, whereas in other cases several positions together make up its function. Originally, Bloomfield si1nply stated that 'The positions in which a for1n occurs are its functions.' (1926: 159). Later on, however, a different, collective use of the term function was added to this: 'The positions in which a for1n can appear are itsfunctions or, collectively, its function.' (1933: 185, cf. 265). Later again, these two uses of the term function were differentiated as 'a function' and 'the function', respectively: 'A function of a for1n is its privilege of appearing in a certain position of a certain construction. The function, collectively, of a form is the sum total of its functions.' (1939: 26; similarly in 1943: 103). We thus have two different notions of function: functiona (a function) being defined as 'possibility of occurring in one particular position', and functionb (the function) as: 'all positions in which a given form can occur'.

20

THE DISTRIBUTlONAL APPROACH

3 .1. 4. Form-class The notion of 'for1n-class' was defined alternatively in ter1ns of position and of function. Thus: 'All for111s which can fill a given position thereby constitute a form-class.' (1933: 185). 'Lexical forms which have any function in common, belong to a com1non form-class.' (ibid. 265; similarly, 1939: 26). But earlier: 'All forms having the same functions constitute a form-class.' (1926: 159). In the first two definitions, for1n-class is defined in ter111s of function° (as containing f or1ns which have a function° in common). In the third definition, it is defined in terms of functionb (as 0 containing forms which have all their functions in common or, equivalently, which have the same functionb). That the latter kind of definition is also implied in Language is evident from Bloomfield's setting up of many form-classes which are defined by more than one function•. Compare, e.g., the sentence 'the functions which define the English for111class of substantive expressions ... are common to an almost unlimited number of words and phrases.' (1933: 265). Therefore, corresponding to the distinction between function• and functionb, we also have two com0 peting notions ofform-class: form-class : the class of all for1ns having a function° in common, andform-classb: the class of all forms having the same function b. Or, in terms of position: 0 0 for1n-class : two or more forms belong to the same form-class if each of them can appear in one particular position. form-classb: two or more forms belong to the same form-classb if in each position in which any one of them can occur, all the others can also occur. It is evident that these two notions ofform-class do not amount to the same thing. According to the first definition the noun phrase the man and 0 the pronoun he belong to the same form-class , since both can occur in such a frame as: (i) ... was late. Furthermore, the man also belongs to the 0 same form-class as the pronoun him, since both can occur in: (ii) I saw ... yesterday. But according to the second definition the noun phrase the man neither belongs to the same f orm-classb as he, nor to the same formclassb as him, since the man can occur both in (i), in which him cannot occur, and in (ii), in which he cannot occur.

3.1. BLOOMFIELD'S DEFINITION

21

3 .1.5. Conclusion The conclusion is that Bloomfield's notion of 'for1n-class' in fact comprises quite different types of categorization. It is further influenced by the lack of clarity in the status of the notion 'position', which is, again, a consequence of the difference between theory and practice regarding the notion 'construction'. What is common to all the difficulties mentioned is that they all involve certain steps of generalization which are, however, not defined as such. The impression is given that all the notions involved are completely analysable in ter1ns of for1nal features, but this does not seem to be true in as far as different undefined abstractive operations are i1nplicitly introduced. . The consequence is that if we take Bloomfield's ter111inology literally, it is not quite possible to determine exactly what is meant by 'belonging to the same form-class'. Correspondingly, it is impossible to give an exact interpretation of 'minor differences ofform-class'. In other words, in Bloomfield's definition of coordination, a further obscurity is added to that of the notions already discussed, in that we are given no criteria to determine at what degree of difference in f or1n-class the definition would cease to be applicable. This is also apparent from the following later formulation: 'If the resultant forn1 agrees as to the major distinctions of f orrn-class with one or more of the constituents, then the construction is said to be endocentric. For instance, the phrase 'bread and butter' has 1 much the same function as the words 'bread', 'butter'.' (1939: 29).

3 .2. Counter-examples to Bloomfield's definition But even if we foil ow Bloomfield in his unspecified use of the terms function, form-class, same, and different, we are bound to find that some coordinative constructions show a quite considerable difference in 'function' with their coordinated members. To take Bloomfield's example: Bill and John, we can list the following environments in which the coordination as a whole, but not Bill or John alone can occur:

Cf. Nida 1949: 94: 'Endocentric constructions are those in which the unit as a whole belongs to substantially the same external distribution class as the nuclear immediate constituent or both immediate constituents.' (my italics).

22

THE DISTRIBUTIONAL APPROACH

(1) a. b. (2) a. b. (3) a. b. (4) a. b. (5) a. b. (6) a. b. (7) a. b.

Bill and John are coining home. *Bill are coining home. Bill and John came home together. *Bill came home together. Those men are Bill and John. *Those men are Bill. Between Bill and John. *Between Bill. Bill and John and their sister Jane. 2 *Bill and their sister Jane. I can never tell Bill and John apart. *I can never tell Bill apart. They answer to the names of Bill and John. *They answer to the na1nes of Bill.

In fact, there seem to be only a few environ1nents in which both Bill and John and Bill or John can occur: (8) a.

b. (9) a.

b. (10) a. b.

Bill and John! Bill! I see Bill and John. I see Bill. They are coining with Bill and John. They are coming with Bill.

The conclusion seems to be, therefore, that no matter how Bloomfield's ter1ns are interpreted, there are in this case not 'minor' but 'major' differences in forn1-class between the coordination as a whole and its members. Together with the general remarks made above, this seems sufficient to reject a general definition of 'coordinative construction' in ter1ns of the endocentric-exocentric dichotomy. The same criticism applies to any other attempt to define coordination on the basis of substitutability of any of the members for the whole construction (see, e.g., Bazel) 1953: 33, Harris 1957: 298, Robins 1964: 234-5, Chatman 1964: 325).

1

their in both cases referring to the same persons as Bill and John.

3.2. COUNTER-EXAMPLES

23

This is even more cogent if, as is often the case in post-Bloomfieldian distributionalism, 'position' is not interpreted as 'place in a construction' (which already presupposes generalization), but as 'place in a particular sentence-frame'. If, in the latter approach, a forrn-class is defined such that two or more forms belong to the sa1ne form-class if there is an environment (taken in this concrete sense) in which each can occur, then probably almost all forms would belong to the same for1n-class. And if 'distribution' is defined as 'the sum total of all the environrnents in which a form can appear', then probably there are no two for ms having the same distribution (cf. Olmsted 1959). If, finally, a form-class is defined in this framework in terrns of complete substitutability in all environ1nents, the result is an empty or a near-empty definition, since probably no two different forms are completely substitutable in all environments in which any one of them can occur. This has been rightly pointed out by Postal (1964a: 55, 58, and fn. 88) in his discussion of Hockett's notion of formclass as given in 1961: 228. One may also compare Hockett 1952: 33 and 1958: 162-3. In the latter passage, Hockett starts by defining 'for1n-class' as follows: 'A class offorms which have sirnilar privileges of occurrence in building larger for1ns is aform-class.' (162). With this definition, the same difficulties arise as with Bloomfield's approach: if the 'privileges of occurrence' need not be strictly identical, but only si1nilar for two forms to belong to the same form-class, how are we to define 'si1nilarity'? Exactly how similar must two or more forms be with respect to their distribution to be included in the same form-class? In the further course of Hockett's argument, however, it appears that the complete substitutability-criterion is again introduced: 'Form-classes are so constituted that, if some member of a given class can occur with some member of a second class, then any member of the first 1night occur with any member of the second.' (ibid. 163). It seems, therefore, that within a purely distributional framework the notion of 'for1n-class' cannot be satisfactorily defined. Correspondingly, the definition of coordinations in terms of form-classes • • • 1s not conv1ncmg. We have seen, so far, that coordinations have grammatical properties of their own which cannot be reduced to properties of their members. In the further course of the argument we shall find many more indications of this phenomenon and we shall see that it is quite crucial to an understanding of the problems involved in the description of coordinations. This is also why the term endocentric is not appropriate in this context,

24

THE DISTRIBUTIONAL APPROACH

since it refers to a particular equivalence relation between a construction 3 as a whole and its constituents. We shall see in chapter 5 that in a different way the 'irreducible' character of coordinative constructions also explains why they cannot be fully accounted for by a transformational description. Cf. Bloomfield's definition quoted in the beginning of this chapter. Perhaps this is the reason why Hall introduces the new ter1n 'universal combination types', setting these apart from endocentric and exocentric constructions (1964: 205).

1

4. General properties of coordinations

4 .1. A general definition of 'coordination' 1

As a tentative general definition of 'coordination', I would like to suggest the following:

Def. A coordination is a construction consisting of two or more members which are equivalent as to gran11natical function, and bound together at the same level of structural hierarchy by means of a linking device. By this definition I restrict the ter1n 'coordination' to a rather narrowly circumscribed class of constructions. Contrast, e.g., Bally's much wider use of the term for a large variety of structures, the com1non denominator 2 of which is not easy to assess (cf. Bally 1944 : 56ff. See the criticism of Antoine 1959: 230ff., 245ff. and Hirschberg 1967: 15). The definition is 3 very close to the one given by Marouzeau 1951 : 63: 'Coordination. Disposition sur le meme plan de plusieurs tern1es ou membres soit simplement juxtaposes ... , soit reunis par une conjonction de coordination.' Ibid. 57: 'Conjonction de coordination: celle qui relic l'un a l'autre deux termes de fonction comparable.' Coordinative constructions, if defined in this way,just constitute a subclass of all possible constructions of a language, without implying anything about the nature of non-coordinative constructions. In particular, I do not have in mind any kind of bipartition of the full set of possible constructions into 'coordinative' and 'subordinative' ones with no residue left. I shall return to this point below (section 4.3.). For ease of reference, I shall now further use the following terminology and symbolism: a coordination (CO) combines several members (M 1 , M 2 , ••• , Mn) by means of a linking device(&). If this device is itself made The terms 'coordination', 'subordination', 'coordinating' and 'subordinating conjunction' are relatively recent acquisitions of the theory of grammar. For a survey of how they came to be used in French grammar in the second half of the nineteenth century, as well as for their later development, see Antoine 1959: 202ff. 1

26

GENERAL PROPERTIES OF COORDINATIONS

up of overt constituents, these are called coordinators (co 1 , co 2 , ••• , con)· The expression &(nM) will be taken to mean: 'a coordination of n members'. Equivalently, I write: &(M 1 , M 2 , ••• ,Mn) or M 1 & M 2 & ... &Mn. In these expressions, the ampersand does not symbolize any particular coordinator(s), but merely denotes the fact that the members are coordinated by some device. I shall further use c 1 , c 2 , ••• , cn for categories or classes, and F 1 , F 2 , ••• , Fn for functions. By the tertn linguistic expression I refer to any arbitrary sequence of constituents in a language. A well-formed linguistic expression is a linguistic expression which is in complete accordance with the phonemic, the grammatical, and the semantic rules of a language. It is thus a linguistic expression which under normal communicative circumstances is completely acceptable to native speakers of the language. The well-formed linguistic expressions comprise well-for1ned sentence-combinations, well-formed sentences, well-formed clauses, well-for1ned phrases, wellformed words, etc. The well-for1ned linguistic expressions of a language are the objects which are deter1nined by the system of that language. In order to explain the general definition advanced here the following remarks must be made.

4 .1 .1. 'Two or more constituents' For some, but not for all coordinations, the number of possible coordinated members is unli1nited, i.e., there is no linguistic boundary to their possible length. Already the ancient grammarians noted this important characteristic: within the general class of conjunctions (O"Uvoeaµor.) Dionysius Thrax distinguished a special subclass of auµ7tA.ex't'r.xo( (coniunctiones copulativae), defined as 'those by which an infinitely expanded 2 expression is bound together' (Ars grammatica § 25, cf. Steinthal 1890-1 , part 2: 322). This characteristic of 'infinite expansibility' is implicitly accounted for in most modern treatments of coordination, and explicitly 2 mentioned in many. It should be noted, however, that since it is not a property of all coordinations, it cannot be used as a general defining criterion for these constructions. As we shall see below, it is this same property which renders a description by means of constituent structure rules 2

Cf. Bally 19442 : 57, Wells 1947: 103, De Groot 1948: 437, 1949: 187, 1962: 69, Bos 1962: 47, Elson & Pickett 1962: 105, Longacre 1964b: 25 fn. 21, 86-7.

4.1. A GENERAL DEFINITION OF 'COORDINATION'

27

as for1nalized in transfor1national generative grammar impossible. I shall return to this matter in chapter 5.

4 .1. 2. 'Equivalent as to grammatical function' In most approaches to coordination it has been emphasized that there should be some sort of equivalence relation between the coordinated members. Some express this by saying that the members should belong to the sarne class or category. The value of this statement depends, however, on the way in which 'class' or 'category' is defined in this context. First of all, we should distinguish between classes set up on the basis of internal grarnrnatical structure, and those based on external grammatical 3 structure, distribution, or valence. By 'internal grarnmatical structure' I refer to the constituents contained in a linguistic element and their hierarchical ordering; 'external grammatical structure' denotes the way in which an element as a whole can be combined with other elements in 4 larger constructions. It has been rightly noted that the members of a coordination need not be equivalent as to internal grammatical structure (Bos 1962: 47-8). In some sense, the equivalence relation should be defined in ter1ns of features proper to the members as a whole. It might be thought, therefore, that they should be equivalent as to external grammatical structure (cf. Nida 1949: 95, Bos 1962: 48, 1964: 213ff.). But the consideration of some examples shows that even this cannot be maintained, since it leads to problems similar to those noted in connection with the distributional definition of coordination. In English, singular nouns or noun phrases have a total distribution 5 different from plural ones. Still, we can say, e.g., (1) Peter and several of his friends arrived late.

Similarly for pronouns vs. nouns or noun phrases (singular or plural):

1

The latter term was originally proposed by Reichling. Cf. De Groot 1948: 437. It is also used by Pittman (1954: 6) and Pike (19671 : 186-7, 283-4). ' A clear distinction between internal and external grammar was made by Wells 1947: 81tr., especially 88. 1 I.e., they belong to different form-classes as detern1ined by Bloomfield's definition 0 of forn1-class (above, section 3. 1 . 4.), or by Hockett's definition ofform-class (section 3. 2.).

28

(2)

GENERAL PROPERTIES OF COORDINATIONS

I saw hi1n and the man who was late yesterday.

For pronouns (or nouns or noun phrases) and complement clauses: (3)

I want to emphasize this point (or: this) and also that you should never forget what your father told you.

For adjectives and certain prepositional phrases: (4)

He felt quite happy and at ease in his new surroundings.

For attributive adjectives and relative clauses in French: (5)

une personne simple de manieres, franche en son langage, et dont le visage ne manquait pas de physiognomie (Balzac, cited in Sandmann 6 1966: 154).

The same in Latin: (6)

maior (sc. frater) et qui prius imperitarat (Livius Ab urbe condita 21: 31) 'The elder (brother) and who previously had reigned'

From these examples it is clear that the equivalence between the members of a coordination cannot be defined in terms of same distribution or valence, unless these notions are interpreted in such a liberal manner as to lose most of the definiteness for which they were originally introduced. I.e., if we want to maintain that the members of a coordination belong to the same class, we will have to set up classes comprising constituents which, by the side of some gra1nmatical similarities, have many i1nportant grammatical differences as well. This criticism applies also to Chomsky's view of coordination, in which the members are also treated as belonging to the same class or category (1957a: 36, 1965: 212 fn. 9, both quoted below, section 5. 3. ). My own solution to this problem, based on the distinction between grammatical functions and classes or categories, will be clarified in chapter 9 below (see especially section 9.2.7.). • For some further French examples see Tesniere (1959: 326), where it is noted that coordinations in which the members are too far apart in internal and external structure are sometimes rejected for stylistic reasons.

4.1. A GENERAL DEFINITION OF 'COORDINATION'

29

If, however, we abandon the search for a purely distributional criterion, we can make, in traditional terms, the following statements about (1)-(6): In ( 1), both Peter and several of his friends are subjects of the sentence; in (2), both him and the man who was late yesterday are objects; in (3), both this point and that you should never forget what your father told you are objects; in (4), both happy and at ease are predicative modifiers (or complements); in (5), simple de manieres,franche en son langage, and dont le visage ne manquait pas de physiognomie are attributive modifiers of une personne; in (6), maior and qui prius imperitarat are (independently used) attributive modifiers. In each case, the members of the coordination, though not equivalent as to internal or external grammatical structure, can be said to be equivalent as to grammatical function. It is this notion of grammatical function which is meant in the general definition given above. It is not to be equated, of course, with Bloomfield's use of the term as outlined in chapter 3, since Bloomfield tried to analyse 'function' exactly in terms of formal characteristics of distribution or valence from which we are dissociating it here. Rather, I should like to subscribe to Longacre's felicitous definition: 'By function is meant the particular office or role of one distinguishable part of a construction type in relation to other parts of the same construction.' (1965: 65). Grammatical function as defined in this way is, in my opinion, a fundamental concept of grammatical theory. Further arguments in favour of this view will be adduced below (chapter 8). We have already seen that Marouzeau's definition of coordination correctly accounts for this functional equivalence. Tesniere, too, emphasizes identity of function between coordinated items (1959: 325ff.). Compare also Strang 1962: 173, where it is stated that coordinators 'function as signs that the structures they link are functioning as equals'. Common function also plays an important part in Pitha's approach. See further Buyssens 1967: 112, where, however, the emphasis is on the functional equivalence between the members and the whole coordination. As has already been indicated in chapter 3, and as will become clearer still in the descriptive framework outlined in chapters 9 and 10, I believe that the relevant functional equivalence should be restricted to the members of the coordination as such.

30

GENERAL PROPERTIES OF COORDINATIONS

4 .1.3. 'At the same level of structural hierarchy'

By this I mean to say that in a coordination all members occupy the same structural level or rank within the total structure in which they are embedded. A structural description of coordinations will necessarily involve multiple constituent analysis or 'multiple branching' (cf. Postal 1964a: 22-5), no more structure being assigned to them than as symbolized in (7)

co &

&

...



M,.

As we shall see later on, it is not excluded that in a coordinative expression different coordinated members occupy different structural levels (cf. below, chapter 11). Thus, an expression like: (8) John and Bill and Harry can be grammatically organized in three different ways, according to the schemas: (9)

co John

and Bill and Harry

(10)

co co John

and

(Bill and Harry)

4.1. A GENERAL DEFINITION OF 'COORDINATION'

31

(11) •.

co co (John and Bill)

and Harry

But this does not violate the rule that the members of a coordinative construction occupy the same level of structure. For only in (9) do we have one coordination; in (10) and (11), we have in each case two different coordinations, one of them being a member of the other. Using the same symbolism as in (7), we could represent (10) as follows:

(12)

co & & where CO', comprising the members M' 1 and M' 2 , is the M 2 of CO. The details of such structures as these will be further treated in chapter 11. Some other views on the hierarchical structure of coordinations, all of them involving more than one structural level, will be discussed in the course of the argument below.

4 .1.4. 'A linking device' It seems that in all languages the linking device can either consist in the mere juxtaposition of the coordinated members (accompanied, in many cases, by a specific intonation-pattern), or in the use of one or more coordinating particles. It is to be observed that the definition of coordination embraces both types of combination; others have restricted this term either to the case of simple juxtaposition of coordinated members, or, alternatively, have required the presence of overt conjunctions for coordination. In the for1ner as in the latter case, the basic similarity be-

32

GENERAL PROPERTIES OF COORDINATIONS

tween both types of structure is obscured (cf. the discussion in Antoine 1959: 213, 254). Tesniere represents the, in my opinion, correct view where he notes that 'jonction' ( coordination) does not require 'jonctifs' ( - coordinators), though jonctifs necessarily presuppose jonction (1959: 82). The ter1n 'linking device' is meant to cover both types of combining principle.

4 .1 . 4 .1. Juxtaposition

If the members of a coordination are combined by juxtaposition, there is, of course, no explicit indication of the relation holding between them. The intrinsic properties of the coordinated members, however, may give 7 rise to the establishment of different relations in the final interpretation. The neutral and most frequent case is that the members are simply combined in a manner which may overtly be expressed by the coordinator and in English. Juxtaposition may, however, also result in alternative, adversative, causal, or consecutive relations (as overtly expressed by or, but,for, 8 and thus or therefore in English). The combinatory case needs no special illustration. For the further possibilities, some examples are: :

Alternative: (13) a. English: Five, six minutes later the bomb exploded. b. Dutch: Morgen, overmorgen, bet maakt 1nij niet uit. 9 'Tomorrow, the day after tomorrow, I don't care'. Serius ocius c. Latin: 'Sooner or later' d. Chinese: Nii chy fann chy miann 'You eat rice eat noodles?' = 'Will you eat rice or 10 noodles?'

1

For this ter1n, and the semantic theory behind it, see chapter 12. Notice that within this theory the aspects mentioned are not regarded as part of the content of the sentences concerned, but rather as added to the content in the interpretation of these sentences as used in actual communicative situations. 8 Cf. Bos 1962: 47, and below chapter 12 for a further elaboration of these various values. 9 For some further examples, see Bos 1964: 225. 10 An example from Chao 1961: 58. Cf. Wang 1965: 460, Hockett 1958: 186.

4.1. A GENERAL DEFINITION OF 'COORDINATIONt

33

Adversative: (14) a. English: It's not cheap, it's expensive! 11 b. Ger1nan: Er bat den Yater instandig; der blieb hart. Iudiciurn hoe omnium mortalium est: fortunam a deo c. Latin: petendam, a se ipso sumendam esse sapientiam. 'This is the opinion of all mortals: fortune must be asked from God, (but) one has to rely on oneself for wisdom.' 12 (Cicero, De Natura Deorum 3: 88). Causal: (15) a. English: We shall stay at home: it's too warm. 13 b. French: Nous ne sortirons pas; ii gele. Consecutive: (16) a. English: It's too warm: we shall stay at home. 13 b. French: II gele; nous ne sortirons pas.

Of course, many more examples might be adduced from many more languages. But the few sentences given here may suffice to indicate the variety of relationships which can characterize the final interpretation of juxtaposed coordinations. In juxtapositions there will often be characteristic features of intonation which, together with the functional equivalence of the juxtaposed members, serve to indicate that a coordination is involved. Thus, the members may be separated by pauses (De Groot 1949: 56, 86) and, no matter whether they are just juxtaposed or combined by overt coordinators, they often have equal stress or tone (De Groot 1949: 58, 60, cf. Gleitman 1965: 272). As often in the case of intonational phenomena, however, it is difficult to determine to what extent these features are essential to the grammatical construction of coordination as such, i.e., to what extent they are part of the definition of 'coordination'. It is clear, for instance, that pauses between coordinated members are by no means 14 indispensable. And even in the absence of any specific intonational 11 11

Cf. Bliimel 1914: 194. In classical grammar this type is of course known as asyndeton adversativum. 11 Cf. Bally 19441 : 56. 1' There is little sense, to my mind, in postulating 'latent pauses' when in fact no pause is present (Bally 19441 : 58).

34

GENERAL PROPERTIES OF COORDINATIONS

pattern, a coordination retains specific properties by which it can be recognized as such. In my opinion, therefore, intonation features are contingent rather than necessary properties of coordinations. A further difficulty with intonation is that with the present stage of language-specific studies it is not easy to arrive at significant general statements about its grammatical relevance. For these reasons I have given only passing attention to intonation features in the present study, though I do not wish to deny the linguistic importance of such features and I realize that any grammatical theory will finally have to account for them in as far as they are indeed gram1natically relevant.

4 .1 . 4. 2. Coordi11ators

Coordinating particles vary in for1n and meaning in different languages. But they show a considerable resemblance in their overall gra1n1natical use and in the set of semantic aspects which they cover in each language. I shall here note some of their general grarnmatical properties. An extensive discussion of their semantic aspects will be given in chapter 12 (see especially section 12. 5 . I.).

4 .1 . 4. 2 .1. A criterion for coordinators One often has difficulty in distinguishing properly between coordinators and other (in particular, subordinative and adverbial) connecting particles. There is not much agreement on this point to be found in the literature, and no general criteria seem to have been established. We can approach this problem by means of the following test, which seems at least to have considerable heuristic value: given a particle which might tentatively be regarded as a coordinator (in a structure like M 1 co? M 2 ), consider the possibility of adding a further particle, the coordinator-status of which has already been established. If this is possible (i.e., if there is a structure like M 1 co co? M 2 ), then co? is not itself a coordinator. If it is impossible (and if there are no further counter-arguments), then co? is itself a coordinator. This test is based, of course, on the premiss that two members can never be coordinated by more than one coordinator.

4.1. A GENERAL DEFINITION OF 'COORDINATION'

35

This may be illustrated with some examples. By the side of: ( 17) John went out and Bill stayed at home. there is nothing like:

(18)

*John went out and Bill stayed at home and and both were happy about this arrangement.

But by the side of: ( 19) He said that he was ill. (where that 1night be regarded as co?), we have:

(20) He said that he was ill and that he could not come. (21) He said either that he was ill or that for some other reason he could not come. Therefore, as far as the test goes, and is a coordinator, but that is not. Compare also: He stayed at home for he was ill. *He stayed at home for he was ill and for his work could be done by others. (24) He stayed at home because he was ill. (25) He stayed at home because he was ill and because his work could be done by others. (22) (23)

Clearly, the test indicates that/or is a coordinator, but that because is not. In the same way, we can distinguish between equivalent pairs in other languages, e.g. between French car and parce que, and between Dutch want and omdat. In each case, there is an objective structural difference between the coordinating and the non-coordinating causal particle, a difference which in some way must be reflected in the structural description of these sentences. This structural difference, to which little attention is paid in the literature on coordination which I have seen, was mentioned in passing already byBlilmel (1914: 193). Bliimel notes the inadvisability of treating Ger-

36

GENERAL PROPERTIES OF COORDINATIONS

man der, do, in such sentences as (26), (27) as coordinating conjunctions:

(26) Wir trafen einen K.naben; der zeigte uns den Weg. (27) Wir verfehlten den Dampfer; da muszten wir einen Kahn nehmen. One of his arguments is: 'Wir konnten ja auch sagen: ... und der ... ; ... und da ... ; waren der, da Verbindungsworter, so batten wir zwei Verbindungsworter zu einem Zweck.' Compare also Sledd's discussion of the difference between conjunctions and conjunctive adverbs (1959: 203). One of his points is that 'conjunctions and ''conjunctive adverbs'' may occur together', as in:

(28) I had given him full instructions, and therefore he knew precisely what to do. The test also provides an argument for the following thesis: while coordinators are not bound up more closely with any one of the members they coordinate (cf. below section 4. 1 . 4. 2. 7. ), subordinating conjunctions rather constitute one unit with the subordinated part, a unit which as a 15 whole can be coordinated with other units. The difference can be seen from (29) and (30): (29)

Coordination.

co co (30)

Subordination.



sub

M 2'

The results of the test suggested above largely coincide with those of a ·Criterion proposed by Gleitman (1965: 274 fn. 2la), according to which a Cf. already Bliimel 1914: 156, 180-1. See also Tesniere 1959: 327, Antoine 1959: 240 fn. 1 and references to Sechehaye, Sandfeld, and Gui1laume. The same structural 1) In this rule-schema, there appears a variable n ranging over the positive integers higher than I. The schema is, therefore, an abbreviation of the infinite set of rules: (45)

co co co co

•M M •M M •M M •M M

M M M

M M

M

• • •

etc. The use of rule-schemata of this type in linguistic description has obvious advantages. They reflect, in a natural way, the fact that in the case of coordinative constructions the linguistic system contains the possibility of indefinite expansions; to be able to use this possibility, however, one does not need to know all its realizations - in fact, this is i1npossible, since the number of these realizations is infinite - but it is sufficient to know the possibility as such (i.e., to be able to act according to it). And this possibility is most naturally formalized by means of rule-schemata such as (44).

5.6. THE USE OF RULE-SCHEMATA

93

The introduction of these rule-schemata into the linguistic description necessitates a slight modification of the requirement mentioned in section 5. 2. above: we can no longer say that a linguistic description must be a system of a finite set of rules specifying an infinite number of linguistic structures. Instead, a linguistic description now contains a finite set of rules and rule-schemata, specifying an infinite set of rules, specifying an infinite set of linguistic structures. In other words, a rule-generating mechanism has in this way been added to the linguistic description. This modification does not seem to meet with any fortnal problems, and, as we shall see below, it makes for a decidedly higher degree of empirical adequacy. It has also obvious repercussions on the 'recursive capacity' of linguistic descriptions, some of which will be noted below. The use of rule-schemata of this general type has been suggested both within the framework of Pike's theory of tagmemics and in the theory of transforrnational generative gra1n1nar itself. I do not differ, therefore, from the latter in recognizing this device, but in giving it a much wider application within the complete linguistic description. I shall now give a detailed survey of the suggestions mentioned.

5. 6 .1. Rule-schemata in tagmemic theory The earliest explicit suggestion in this direction which I have been able to find in tagmernics is in Elson & Pickett 1962: 105. After having defined a 'coordinative or serial phrase' as a kind of phrase 'which includes more than one head tagmeme, with a coordinate relationship indicated either by a coordinating tagmeme or by the mere juxtaposition of the two (or 36 more) head tagmemes or by both.', they proceed as follows: 'The analysis of such constructions as the above is best handled in terms of repeated head tagmemes. Thus +H:n

+(± H:n ... [+C:c ... ] + H:n ... )

37

This formula means that in order to have a coordinate phrase there must be at least two head tagmemes, the dots mean that the tagmeme in ques1• 17

Cf. the definition which I have given above, section 4. 1. In this formula H stands for 'head', C for 'coordination', n for 'noun', or 'nominal', c for 'coordinator', + for 'obligatory', and ± for 'optional'.

94

COORDINATION IN TRANSFORMATIONAL GRAMMAR

. tion can be repeated an indefinite number of times, the brackets around the coordinator tagmeme mean that though it may be repeated the repetitions must be between head tagmemes. Such constructions may be ter111ed open-ended in that there is actually no limit on the number of tagmemes that may occur.' It is clear that, though perhaps less perspicuous, the dot-device is fully equivalent to the use of a variable n. I do not understand, therefore, how Postal can assert that tagmemic writers have not recognized that some 'branchings' are infinitely multiple, adding that 'tagme1nic formulas are incapable of describing unbounded branching correctly, that is, without arbitrary assign1nent of excess structure.' (1964a: 34). And: 'Longacre himself insists on branchings of a degree higher than two. But in his formulation, as in that of tagmemic writers generally, there is some finite upper bound on branching and the problem of infinite branching has not even been faced no less solved.' (ibid. 85 fn. 36). This is only one of the numerous places where Postal gives less credit to the theories he discusses than they deserve. Longacre has further developed the type of rules suggested by Elson & Pickett by introducing a superscript variable n. He adds the following 0 quite pertinent remarks: 'Use of superscript in tagmemic forn1ulae is one manner in which such fortnulae allow for the open-ended flexibility which characterizes actual use of language. Such for1nulae allow for tree structures with an indefinite (and theoretically unlitnited) number of branches ... In view of these features of tagmemic for1nulae it is pointless for advocates of transform grammar to continue to contend that tagmemics cannot handle such problems as infinite branching, infinite nesting, and coordinate constructions admitting formation of infinite series.' (1964: 25 fn. 21). For the working of Longacre's rule-schemata, 38 see ibid. 86-7. 5. 6. 2. Rule-schemata in transformatio11al grammar .

From reading the earlier literature on transformational generative grammar, one would not easily gain the impression that rule-schemata are, in Cf. also Cook 1964: 17: 'In its maximum form it [i.e., the syntagmeme, SD) may contain a (potentially) infinite number of tagmemes in an open-ended construction, or endless stream of tagmemes.' 22: 'In an open·ended construction, the maximum syntagmeme is potentially infinite'. 38

S.6. THE USE OF RULE-SCHEMATA

95

fact, indispensable by the side of constituent structure rules and transf or1national rules even in a transfor111ational grarn1nar. Their necessity is for the first time clearly indicated in Chomsky 1965, though even there the exact range of their application is not explicitly laid down. One has a feeling that, for reasons not clear to me, there has been a certain reluctance on the part of transfor1nationalists to allow for such schemata involving variables. But in the course of the development of the theory, they have forced their way into it, taking over some offices f or•nerly assigned to transfor1national rules. It would seem that they are still gaining in influence. The first published reference to rule-schemata in transforn1ational generative grarnmar which I have been able to find is in Chomsky & Miller 1963: 298. Here, while discussing the transformational description of such sentences as (7), they mention the i1npossibility of describing them by means of constituent structure rules, and then add: 'We might try to meet this problem by extending the notion of constituent-structure grarn1nar to pern1it such rule schemata as Predicate • Adj" and Adj (n ~ 1). Aside from the many difficulties involved in formulating this notion so that descriptive adequacy may be maintained, it is, of course, beside the point in the kind of difficulty that arises in Example 19 [a completely unrelated case, SD]. In general, for each particular kind of difficulty that arises in constituent-structure grammar, it is possible to devise some ad hoe adjustment that might circumvent it. Much to be preferred, obviously, would be a conceptual revision that would succeed in avoiding the mass of these difficulties in a uniforn1 way, while allowing the sirnple constituent-structure grarnrnar to operate without essential alteration for the class of cases for which it is adequate and which initially motivated its development. As far as we know, the theory oftransfor111ational grammar 39 is unique in holding out any hope that this end may be achieved.' What is stated here is clearly this: coordinations can be described both by rule-schemata (the 'many difficulties involved' are not further specified) and by transformational rules. The latter are needed in any case for other types of construction. Therefore, we gain in simplicity if we dispense with rule-schemata altogether. For a similar line of thought, cf. Chomsky 1957a: 41 fn. 6. See also section 5.5 .2. above on the role of simplicity-criteria.

11

96

COORDINATION IN TRANSFORMATIONAL GRAMMAR

In a study on the f or1nal properties of context-free gra1n1nars, in which transfor1national rules are not taken into account, Chomsky & Schi.itzenberger gave a more positive opinion on these rule-schemata: after their introduction into a context-free grammar, 'the gra1n1nar, though still finitely specified by rule-schemata, consists of an infinite number of rules ... This is the natural way to extend CF [ context-free] gra1n1nars to accommodate true coordination, as, e.g., where a string of adjectives of arbitrary length may appear in predicate position with no internal structure defined on them.' (1963: 133). It appears, therefore, that the 'many difficulties' mentioned by Chomsky & Miller (1963: 298) are not difficulties pertaining to the for1nalization of these rule-schemata (nor would we expect this formalization to meet with any real problems), but that they merely mean that a description in terms of these schemata does not answer to the picture they have of what the structural description of coordination should look like, and violates, furthermore, the principle of simplicity. But notice that even if we were to accept the thesis that all cases of coordination can be transfor1nationally derived (which cannot, in fact, be accepted), a problem would remain to which but little attention has been paid. We have seen that to achieve the proper result, the generalized transformations of the earlier version of the theory must be able to operate on an indefinite number of underlying structures. Therefore, if taken strictly, there would not be one coordinating transfor1nation, but an infinite set of such transformations, one for each number of underlying structures on which it operates. This means that in fact the transforrnation could only be given in the f orrn of a transfor1nation-schema of the general for1nat: (46)



T => ds

40





'

0

us = 'underlying structure', T = 'transfor1nation', ds = 'derived structure'.

5.6. THE USE OF RULE-SCHEMATA

97

This is indeed the way in which the conjunction transformation is given in Fidelholtz 1964: 20, and also in Gleitman 1965: 273 (in the latter case, at the suggestion of Chomsky). Should we restrict the application of such transformations to any two underlying structures at a time, then we would automatically get the undesirable multilevel structure which 41 could only be readjusted by ad hoe devices. We see, therefore, that there can be no objection against rule-schemata as such in a complete linguistic description. Nor are they eliminated by 42 the introduction of transformations, but only applied on a different level. This has become more evident still after the later revisions of the theory, in which generalized transfor1nations are eli1ninated entirely, all underlying structures relevant to the ulti1nate sentence being directly generated by the 'base-component' into a generalized phrase-marker (cf. Chomsky 1965: 134ff.). The consequences of this revision for the description of 43 coordinations have not yet been extensively discussed, but the following remarks of Chomsky are relevant to the present argument: discussing the new for1nat of the base component, he suggests that its constituent structure rules might, in part, be replaced by 'rule schemata that go beyond the range of phrase structure rules in strong generative capacity' (1965: 99), and mentioning the problem of incorporating coordinative constructions into the new theory, he says: 'There are certain problems concerning these, but I believe that they can be incorporated quite readily in the present scheme by permitting rule schemata (in the sense of Chomsky and ' 1

An attempt of Koutsoudas to achieve the correct result while still starting from two underlying structures at a time will be discussed below (section 5. 7. 1. 4.). ' 1 Schane (1966: 12-3) has demonstrated that in still quite a different sense the original generalized conjunction transformation as given by Chomsky (1957a: 113) and Gleitman (1965: 273) is not a genuine transfor111ational rule on a par with other such rules, but rather an abbreviation of a set of transformations, since it uses variables in the 'structural description' and the 'structural change' ranging over the different types of constituents which are to be conjoined; in most other cases where variables like X and Y are used in transformational rules, they indicate environmental material which is irrelevant to the transformation as such. For a motivation of the use of these variables in the formulation of transformations, see Koutsoudas 1966: 247-8, 250. For its consequences, see also Bach 1966b. This point has no direct bearing on our argument, since we are concerned only with rule-schemata in which the variable ranges over the number of constituents to be coordinated, not with variables ranging over the constituents (or strings of constituents) themselves. ' 8 In Katz & Postal 1964: 47 it is simply assumed that the account of embedding transformations given there 'can be extended without essential modification to conjunction transformations.' Staal has rightly remarked that this is by no means immediately obvious (1965: 141) and deserves more attention (1966: 71). See the following section for a survey of recent attempts to cope with this problem.

98

COORDINATION IN TRANSFORMATIONAL GRAMMAR

Miller, 1963, p. 298; Chomsky and Schiitzenberger, 1963, p. 133) [both cited above, p. 95-6, SD] introducing coordinated elements that are then modified, rearranged, and appropriately interrelated by singulary transfor1nations.' (ibid. 224 fn. 7). The role of rule-schemata is also clearly brought out in the following recent remark: 'Obviously, the syntactic component must have a finite number of rules (or rule schemata) ... ' (Chomsky 1967: 419). The conclusion is that rule-schemata have now received official status in the theory of transformational grammar. There are no problems concerning their forn1alization, and they cannot be discarded any more as ad hoe devices over against transfor1national rules. We have every reason, therefore, to investigate thoroughly how they could be exploited to their full advantage.

5. 7. Recent suggestions for the description of coordinations in transformational generative grammt1r The latest version of the theory of transfor1national grain mar has not as yet resulted in a uniforn1ly accepted solution for the description of coordinations. Several suggestions have been and are being advanced, not all of which have been laid down in the published literature. I now present a survey of these suggestions, so as to give an i1npression of the direction in which solutions are being sought. Much of what follows, however, should be regarded as tentative and as awaiting a more general consensus 44 among transformationalists.

5. 7 .1. Sentence conjunction Most of these suggestions are explicitly or implicitly restricted to 'sentence conjunction', a term used for those cases in which the reduction to separate 'sentences' is (at least, superficially) possible. All these treatments have in common that they allow for rule-schemata of the general type of (47), by which an indefinite sequence of S's can be initially introduced. I am grateful to Prof. Chomsky and to Mr. J. Fidelholtz of M.I.T. for having provided me with explanation and infor1nation on these developments. They are in no way responsible, of course, for possible flaws in my presentation of these matters.

44

99

5. 7. RECENT SUGGESTIONS

(47)

s • sn

(n ~ 1)

They differ, however, in the particular way in which these S's are reduced in the case of coordination. These differences can be most easily demonstrated with the following example: (48)

John, Bill, and Peter laughed.

For a sentence like (48) the base component would originally introduce a sequence of three S's (according to rule-schema (47)), roughly in the following way: (49)

s

s

s

s NP

VP

John

laughed

NP B.ll l .

VP

NP

VP

laughed

Peter

laughed

The transfor1national problem is to arrive from (49) at (50) in a satisfactory way. (50)

s NP

VP

NP

NP

NP

John

Bill

Peter

laughed

For the achievement of this aim, the follo\ving approaches have been suggested.

100

COORDINATION IN TRANSFORMATIONAL GRAMMAR

5.7.1.1. Chomsky After having suggested the possibility of using rule-schemata (cf. the passage cited above, p. 97-8), Chomsky proceeds as follows: 'If the sugges45 tion of note 9, Chapter 2, is workable, then such rule schemata need not be stated in the grammar at all. Rather, by a general convention we 46 can associate such a schema with each major category. This approach to coordination relies heavily on the filtering effect of transformations, discussed later. Thus wherever we have coordination, some category is coordinated n tirnes in the matrix sentence, and n occurrences of matched sentences are independently generated by the base rules.' (1965: 224 fn. 7). 47 • This must be interpreted as follows. For the given sentence (48), ruleschemata would not only apply to S, but also to any one of the three NPs (say, the first), yielding an arbitrary number of coordinated NPs. Some of these, however, would not be subjected to the insertion of lexical items, but would rather receive so-called 'du1n1ny-elements', symbolized 48 as ~. Now, if (i) the number of NPs equals the number of S's, (ii) the number of dummy elements in the first S equals the number of corresponding NPs in the other S's, and (iii) the structures of the S's are identical save for the coordinated NPs in the first, certain transformations could operate, which would substitute the lexical items dominated by the other NPs for the dum1nies of the first S, and delete all S's (plus further structure dominated by them) but the first. The underlying structure of (48) would, on this proposal, be as symbolized in (51) rather than as in (49):

Cited above, p. 71. For this term see Chomsky 1965: 74: 'a lexical category or a category that dominates a string .. . X .. ., where X is a lexical category, we shall call a major category.'. For the relation 'X dominates Y' cf. above fn. 4. 47 This same suggestion of Chomsky's is also explained in Fidelholtz 1964: 9. 48 For dummy-elements or -symbols cf. Katz & Postal 1964: 48, Chomsk.y 1965: 132.

•5 46

5. 7. RECENT SUGGESTIONS

101

(51)

s

s

s NP l~P

John

NP

VP

NP

VP

NP

VP

NP ~

laughed Bill

laughed Peter

laughed

Underlying structures not meeting conditions (i)-(iii) would not be subjected to these transfor1nations and would not, therefore, result in wellfor1ned sentences (i.e., would be 'filtered out' by the transfor1nations 49 Chomsky 1965: 139, 141, 191).·

5. 7 .1.2. Ross & Lakoff Rather than let rule-schemata apply to other categories than S in the case of sentence conjunction, Ross & Lakoff suggest a special 'conjunction reduction rule' operating on underlying structures such as (49). This involves the following steps: (i) The three identical VPs are disconnected, reduced to one, and put on one level with the highest S under a new dominating S. This results in

The notion of the 'filtering effect of transformations' was developed in the later version of transformational theory, in which the base component generates many structures which do not qualify as deep structures of actual sentences. These pseudodeep structures are not subjected to further transformations, i.e., they are 'filtered out' by the conditions put on the operation of transformations. u

102

COORDINATION IN TRANSFORMATIONAL GRAMMAR

(52)

s VP

s

s

s

NP

NP

NP

John

Bill

Peter

laughed

(ii) S's which dominate only a single category are then deleted (cf. Ross 1966). The result is (53)

s s

VP

NP

NP

NP

John

Bill

Peter

laughed

(iii) The S dominating the three NPs is finally changed to (50).

NP, yielding

5.7.1.3. Schane Schane (1966) suggests a similar approach. He starts, however, by marking the NPs rather than the VPs, and then reproduces a single S-tree with the structure common to the three original S's, as in (54):

5. 7. RECENT SUGGESTIONS

103

(54)

s NP

VP

laughed The three disconnected NPs are then attached to the NP of (54), the result being, again, (50). Schane's work is interesting in that a wealth of phenomena connected with coordination are discussed and the consequences of a rigorous formalization of the description of coordinations are clearly indicated. There is something paradoxical in his solution, however, in that it is presented as a departure from transfor111ational theory, while in fact it is best regarded as one particular elaboration of the principles already given in the later version of this theory, and as quite comparable, in this respect, to the suggestions of Chomsky and Ross & Lakoff. Schane starts by noting that 'Within a generative grammar coordinate constructions are handled awkwardly if they are derived by means of transfor1national rules.' (1966: I). But he goes on to say: 'Yet that many types of coordinate constructions should originate from conjoined full sentences, which are subsequently reduced in some way, seems to be well motivated' (ibid.). He terms his own solution a 'schema' which is 'partly transfor1national and partly non-transfor1national in nature. The nontransforn1ational part - what we call the primary conjunction rules derives a single sentence with conjoined constituents from two ormore coordinate sentences [along the lines indicated he.re, SD]. A set of singulary transforn1ations - the secondary conjunction rules - may then operate on these derived coordinate structures, converting them to related variant for1ns' (ibid. 2). The paradox is that Schane's 'non-transformational part' in fact consists largely of transforn1ational rules. This fact is disguised somewhat by his manner of presentation, in which the 'primary conjunction rules' are at first not given in formalized version, but are rather presented as certain operations on diagrams, just as I have summarized them here (Schane 1966: 13-6). But as soon as these operations are completely formalized, they turn out to involve certain combinations of 'elementary transforma-

104

COORDINATION IN TRANSFORMATIONAL GRAMMAR

tions' (deletions, additions, substitutions). so This is not surprising, since 'reduction of structures from other structures by non-transformational rules' is simply the expression of a contradiction. The transfor1national character of Schane's solution will be more evident still if we rephrase it in the foil owing way:

(i) Mark, in (49), the nodes beneath which the tree structures differ (i.e., the three NPs). (ii) Choose one of them (say, the first), and create a new, identical node dominating it, to yield (55): (55)

s

s NP

John

s VP

NP

laughed Bill

s VP

NP

laughed Peter

VP

laughed

(iii) Adjoin all the remaining marked nodes (and everything beneath) to the newly created one, to obtain (56):

50

In the appendix to Schane's paper by Joyce Friedman (56ff.) the transformational character of his rules is brought out clearly.

5. 7. RECENT SUGGESTIONS

105

(56)

s

s

s NP NP John

NP

VP

s VP

NP

Bill Peter

VP

I

laughed

laughed

laughed

(iv) Delete anything but the first S and everything dominated by it. The result is (50). The only non-transformational feature of Schane's 'primary conjunction rules' is that in accordance with the correct observation quoted above (fn. 42) a 'general principle of conjunction' is added to the transformational rules, specifying the conditions under which these rules can operate. This is not, however, a major revision of the transfor1national framework. From the very start it has been recognized in transformational generative grammar that by the side of constituent structure rules and transforma51 tional rules further 'traffic rules' are needed (though the exact status of these rules has not been fully clarified); in the later version of the theory these traffic rules return (in part) in the form of general conventions specifying the possible form of generalized phrase-markers and the conditions under which transfor1nations can be applied, as well as the manner in which the application takes place. It is to these conventions, which are also relied on in the approaches of Chomsky and Ross & Lakoff, as su1n1narized above, that Schane's 'principle' is fully equivalent.

5. 7 .1 . 4. Koutsoudas It is instructive to compare these various solutions, which all have in common that they use rule-schemata in some way, with the approach 61

For this term cf. Lees 1960: 53-7.

106

COORDINATION IN TRANSFORMATIONAL GRAMMAR

suggested by Koutsoudas (1966: 251-3) which, though quite recent, is still formulated in ter1ns of the earlier version of the transformational theory (i.e., with separate underlying structures instead of generalized phrase-markers as a starting-point). Koutsoudas still treats conjoining transformations as double-base, or binary transfor111ations. As has been noted above (section 5.6.2.), this would lead to the assign1nent of incorrect multi-level derived structures if these transformations were applied more than once, unless some ad hoe readjustment device is intro52 duced. To avoid the incorrect result, Koutsoudas proposes a rule of derived constituent structure which differentiates between the first application of the transforn1ation on the one hand, and all later applications on the other: 'When two constituents are first conjoined by a transformation, they are attached to a node with the same label these constituents had in the source P-marker; i.e., a new node is created. After that, conjoined constituents are attached to the newly created node in accordance with the convention for attachment by adjunction.' (ibid. 253). There are various drawbacks to this proposal, which, among other things, show again that the restriction to two underlying structures in the older version of the theory was arbitrary and has rightly been given up in favour of an indefinite number of underlying 'sentences' in the generalized phrase-marker. (i) In this way, the structure-assignment for a two-member coordination would be different from that of a coordination with a number of members greater than two. (ii) The conjoining transformation, if for1nulated in this way, could not operate automatically on any two conjoinable categories of the sa1ne type, but at each application it would have to be determined whether any one of these categories already had a coordinated structure. No indication is given of how this could be achieved. (iii) While the aim was to arrive at a recursive specification of coordination, we would in this way get two different transfor1nations. A nonrecursive one conjoining single categories, and a recursive one conjoining two categories at least one of which had already been conjoined. 52

For a discussion of earlier similar proposals by Lees, which lead to related problems, cf. Pilha 1966: 206ff.

S. 1. RECENT SUGGESTIONS

107

(iv) As Koutsoudas himself remarks, by this method we could not differentiate between same-level and multi-level coordinations (see the treatment of ambiguity, below chapter 11), so that a sentence like (57) John and Mary, Roger and Helen, and Bill and June went to the show. would be assigned the incorrect structural description (58) (58)

NP

VP

John & Mary Roger & Helen and Bill & June went to the .show instead of the more correct:

(59)

s NP NP

NP

VP

and

NP

John & Mary Roger & Helen Bill & June

went to the show

This latter drawback also affects the other three solutions mentioned. At least, they give no indication of how a structure of type (59) could be arrived at on the basis of an underlying structure in which each of the subjects would be separately combined with the predicate went to the sholt·.

5. 7 .1 .5. Conclusion As far as sentence conjunction is concerned, we may conclude that the later version of the theory certainly has some advantages over the earlier one. Each of the first three solutions, which fit into this later version, how-

108

COORDINATION IN TRANSFORMATIONAL GRAMMAR

ever, is still firmly based on the reduction postulate, and is subject, therefore, to the objections which have been formulated in section 5. 5. against the transformational description of coordination as such. It is to be noted that only Chomsky's solution allows rule-schemata to apply to categories other than S, though on the condition that S's are simultaneously repeated. I shall not go into a detailed analysis of the relative merits of these three approaches here, but I may note in passing that the Ross-Lakoff solution makes a decidedly artificial impression in that both a new S is introduced and old S's are deleted, and in that the conjoined NPs must be labelled as together for111ing an NP by a special ad hoe rule substituting NP for S (cf. Schane 1966: 11 ). At one point in their derivation (52), each NP is also an S, and at another point (53) the three conjoined NPs are an S. Neither of these labellings has any further relevance to the final derived structure, nor to the semantic interpretation of the sentence concerned. If anything is ad hoe, it is this 'conjunction reduction rule'. Schane's solution lacks these vacuous intermediate stages and seems, therefore, to have the advantage of simplicity. It can be regarded as an attempt to reach the result of Chomsky's approach by the methods of Ross & Lakoff. Under the hypothesis of the reduction postulate, however, I regard Chomsky's solution as superior to the other two, since (i) coordinations within a sentence are in this way directly marked in its underlying structure, and (ii) it leads over in a natural way to cases where the reduction to underlying 'sentences' is completely impossible (see the next section). On the other hand, since the rule-schemata applying to S and NP, and the selection of dummies in the latter case are not connected numerically, this approach has the disadvantage of specifying an infinite set of pseudounderlying structures which do not lead to well-for111ed sentences and, indeed, 'relies heavily on the filtering effect of transfor1nations' (cf. section 5. 7. 1 . I.). Thus, we would have to accept the fact that coordinative constructions, which are perfectly perspicuous in themselves, are described by means of a random mechanism which can only lead to the correct result if a host of restrictive conditions are conformed to. Finally, one may well ask: if the NPs which are coordinated in the final sentence are already coordinated in its underlying structure, why need the 'sentences' be also coordinated in this underlying structure, only to be deleted afterwards by this quite complex mechanism? The answer is simple: this

5. 7. RECENT SUGGESTIONS

109

is only because the reduction postulate, which relates complex sentences to si1nple underlying subject-predicate combinations in underlying structures, has become an inevitable dogma of transformational theory. In my opinion, these facts force us to look for another and better solution.

5 . 7. 2. Phrasal conjunction

If, as in Chomsky's proposal, cases of reducible coordination are already allowed to be described by rule-schemata operating on the coordinated categories (albeit in combination with S-repeiition), it is a natural step towards treating the irreducible cases (such as the sentences of (33) and (43)) in the same way without S-repetition. I.e., a sentence like (43i) can then directly be assigned a structure like:

(60)

s NP

NP

NP

NP

and

NP

NP .



John

Paul

George

Ringo

the Beatles

are •

In such cases as this there would simply be a rule-schema which applies directly to the coordinated category, and no further adjustments are necessary. This has indeed been suggested by Fidelholtz (1964), Peters 53 ( 1966), and Lakoff & Peters ( 1966). This type of coordination has received the name of 'phrasal conjunction'. As to the scope and the exact descriptive details of phrasal conjunction there are, again, different • • op1n1ons.

Cf. also the discussion in G leitman 1965: 289-92, where the same idea is attributed to Carlota S. Smith (in an unpublished paper). According to Bierwisch (1966a: 197 fn. 60), a similar approach was suggested by Fillmore and Lees. Though Chomsky does not seem to have expressed his opinion on this matter in his published writings, it appears from personal communication that he would tend to accept some such solution as this for these cases. 63

110

COORDINATION IN TRANSFORMATIONAL GRAMMAR

5. 7. 2 .1. Fidelholtz

Fidelholtz has gone furthest in this direction in at least contemplating the possibility that all so-called 'major categories' might be subjected to a rule-schema of the general type

(61)

X • X(&X)n (n

~

0)

(1964: I, and fn. I). This would mean that not only coordinated nouns and noun phrases could be derived in this way, but also certain cases of coordinated verbs and adjectives and perhaps other categories, 'since this 54 appears to be somewhat more intuitively satisfying.' (ibid. 1).

5. 7. 2. 2. Lakoff & Peters Lakoff & Peters (1966), however, restrict this method to coordinated noun phrases. For these they suggest the following treatment: 'At least in the case of noun phrases, conjunction must occur in the base component. That is, there must be a rule schema of the for111: NP

~

and (NP)

0

,

n

~

2.' (1966: 2).

The working of this rule-schema is explained as follows (ibid. fn. 2): for n - 4, e.g., it generates a string: (62)

and NP NP NP NP

This is then converted by a 'general convention' into: (63)

and NP and NP and NP and NP

Then, the first and is obligatorily deleted by transformation, yielding: 5'

When I had completed this chapter, I had the opportunity to see Yamada & lgarashi 1967. On the basis of Japanese data, these authors come to much the same conclusion regarding the role of rule-schemata in the description of coordinations. In fact, they repeat the (unpublished) proposal of Fidelholtz, viz. to 'associate the coordination rule schemata to each major category' (148). They do not hold, however, that in this way the transformations can be dispensed with altogether.

S. 1. RECENT SUGGESTIONS

(64)

111

NP and NP and NP and NP

Finally, there is an optional transformation deleting all but the last and, the result being: ( 65)

NP NP NP and NP

Thus, in the case of sentence (43i), we would .first get and John Paul George Ringo, then (by general convention) and John and Paul and George and Ringo, then (obligatorily) John and Paul and George and Ringo, and finally (optionally) John Paul George and Ringo. A minor disadvantage of this somewhat roundabout formulation is that stages (62) and (63) seem unnecessary and, again, rather artificial, since (64) can be directly generated by rule-schemata of the Fidelholtz type, such as (61). A second disadvantage is that in Lakoff & Peters' explanation the conjunction plus the foil owing NP are allowed to be dominated by NP, so that the incorrect impression is given that in, e.g., John and Bill, the sequence and Bill is a noun phrase, as symbolized in (66): (66) NP •

NP

NP NP

John

and

Bill

(1966: 25). The correct structure, however, would be as symbolized in 5 (67), just as the same authors give it on p. 11 of their article: s

I must add, however, that I have not been able to obtain Peters 1966, to which we are referred for a motivation of this formulation.

65

112

COO RD INATION IN TRANSFORMATIONAL GRAMMAR

(67) NP

NP John

NP and

Bill

Notice, furthermore, that if we are right in postulating a difference of emphasis between types co 1 M 1 co 2 M 2 , M 1coM 2 , and M 1 M 2 (cf. above, section 4. 1 . 4. 2. 2.), the transformational derivation of all these three types from one preposed and would run counter to the hypothesis of semantic invariance under transformations.

5 . 7. 3. Final remarks on the scope of rule-schemata

As I have noted above, one observes in the development of the theory of transformational generative grammar a certain reluctance to make an extensive use of rule-schemata of the type relevant here. One of the reasons for this is, apparently, that the introduction of such schemata destroys the neat dichotomy of constituent structure rules vs. transformational rules originally proposed in Syntactic structures, in that rule-schemata take over the description of certain phenomena which originally served as (part of) the justification of the transformational approach. This may also explain the fact that once the need and legitimacy of rule-schemata are recognized, there is a marked tendency to restrict the scope of application of this device as far as possible. Another possible reason for this restriction is that one does not want to make the total theory of linguistic description stronger than is necessary for the description of the facts at issue. Thus, Chomsky tries to restrict the application of these rule-schemata exclusively to the category S, as is clear from the following passage: 'In other words, the only way to form new deep structures is to insert ele56 mentary ''propositions'' - technically, base Phrase-markers - in other The use of the term 'proposition' in this context reflects the supposed connection between early logically oriented linguistics as exemplified in the Port-Royal grammar (Lancelot & Arnauld 1660) and the transforn1ational theory. Compare Chomsky 1966c: 42, where in the context of a summary of Port-Royal ideas in modem terms it 68

S. 1. RECENT SUGGESTIONS

113

Phrase-markers. This is by no means a logically necessary feature of phrase structure gram1nars. Notice that the schemata that underlie coordination ... also provide infinite generative capacity, but here too the true recursive property can apparently be litnited to the schema S • S # S # . . . # S, hence to rules introducing ''propositions''.' (1965: 225 fn. 11). He then goes on to say that 'This f or1nulation leaves unexplained some rather marginal pltenomena (e.g., the source of such expressions as ''very, very, ... ,very Adjective'' and some more significant ones (e.g., the possi. bility of iterating Adverbials and various kinds of parenthetic elements, the status of which in general is unclear).' We have seen, however, that there are quite 'central' phenomena which show that Chomsky's restriction cannot be maintained, and that 'phrasal conjunction' for at least some categories other than S is inevitable even within a transformational grammar (cf. above, section 5. 7. 2.). This is probably hinted at in the following recent remark of Chomsky's: 'It may be that the introduction of ''propositional content'' in deep structures by this means [i.e., by the recursive introduction of S's, SD] is the only resursive device in the grammar apart from the rules involved in coordinated constructions, which raise various problems going beyond what we have been discussing here.' (1967: 435). If, however, rule-schemata are allowed to operate on categories other than S (which is inevitable), then the restriction of these 'other categories' to NP (as in Lakoff & Peters 1966) seems to be quite arbitrary. Fidelholtz has correctly noted this (1964). A si1nilar uncertainty as to the scope of rule-schemata can be observed in Wang 1965. For the description of certain Mandarin sentences where only the predicate is coordinated, Wang introduces a rule-schema of type:

(68)

Pred • Predf

(1965: 464). He then adds the foil owing remark: 'It is an important but as yet unanswered question in the theory of grammar whether rules of the is said that 'The base rules allow for the introduction of new propositions (that is, there are rewriting rules of the form A > ••• S ... , where S is the initial symbol of the phrase· structure grammar that constitutes the base); there are no other recursive devices' (my italics). On the, in point of fact, quite questionable relations between Port·Royal and Chomskyan transformationalism, see below chapter 6.

114

COORDINATION IN TRANSFORMATIONAL GRAMMAR

form (68) above should be permitted. In my paper 'Conjoining and dele57 tion in Mandarin syntax.' ••• I took the opposite position ... that only the symbol S (sentence) may be repeated an arbitrary number of times in a grammatical rule.'

5. 8. Conclusion

We are now ready to su1n1narize the critical analysis contained in this chapter. We have seen that transformational rules were introduced to take care of coordinations because of the inherent restrictions of constituent structure rules. The transformational description, however, turned out to create problems of its own, not only for those cases for which the reduction postulate does not hold at all, but also for coordinations for which the reduction to noncoordinated sentences is at least superficially possible. We have also seen that transforn1ationalists are far from being agreed among themselves on the proper method of transforn1ational derivation even in these, most simple cases. No approach has so far been suggested which really achieves the generality, adequacy, and simplicity claimed for transformational descriptions. We have further witnessed the emergence of rule-schemata, which constitute an alternative device to overcome the restrictions of constituent structure rules. The gradual introduction of this device can be su1n1narized as follows: Stage 1. All coordinations were thought to require transformational derivation, for reasons of descriptive adequacy and sirnplicity (Chomsky 1957a: 35-8, Gleitman 1965). Stage 2. The possibility of using rule-schemata was mentioned, but rejected as an ad hoe device, inferior to an overall transfor1national approach (Chomsky & Miller 1963: 298). Stage 3. In the revised version of the theory, the problem arose of specifying an indefinite number of coordinated items in a generalized phrasemarker. Rule-schemata were found to be indispensable after all (Chomsky 1965: 224 f n. 7, 225 f n. 11 ). Stage 4. The irreducible character of some types of coordination was As Prof. Wang was kind enough to inform me, this paper, which was read in 1964 to the Linguistic Society of America, is not yet available in print.

57

5. 8. CONCLUSION

115

noted, and 'phrasal conjunction' (essentially non-transforrnational) was set up as a special case, to be described by rule-schemata (Fidelholtz 1964, Peters 1966, Lakoff & Peters 1966). Stage 5. Rule-schemata have thus been integrated into the transformational theory, but their exact range of application is still a matter of discussion. In general, there is a tendency (which is not motivated by empirical considerations) to restrict their scope as much as possible. A consistent elaboration of 'phrasal conjunction' would show, on the other hand, that for almost any category rule-schemata will have to be f orn1ulated. The line of this development can be drawn to a very simple conclusion. The transfor1national description of coordinations was motivated mainly by two reasons: (i) it was thought to provide a more adequate description, (ii) it was considered to be the most natural answer to the inherent restrictions of constituent structure rules. As to the former point, this does not constitute a decisive argument: there is no a priori sense of 'adequacy of a linguistic description' which can be brought to bear on the selection of its descriptive devices, i.e., there is no 'absolute' sense in which transformational rules can be said to be 'necessary' to linguistic description. As to the second point, this can no longer count as soon as the descriptive power of rule-schemata has been recognized. It is only natural, therefore, to suppose that not only descriptive adequacy, but also generality and si1nplicity will be served if the device of ruleschemata is extended to take care of any type of coordination, leading to the abandonrnent of transfor1national rules in the specification of this type of structure. In chapters 9, 10, and 11, I shall investigate the consequences of this non-transfor1national point of view.

6. The background of the transformational approach to coordination 6 .1. Introduction

In the preceding sections we have found several reasons why a transfor1national description of coordinations can neither achieve the adequacy, nor the completeness, nor the si1nplicity expected from it. For a final evaluation of this theory, it is not irrelevant to consider the roots from which it has developed. . As a point of departure, we can take the foil owing passage from Staal 1966: 71-2, in which the points at issue are succinctly summarized: 'Both in logic and in generative grarnmar, ''and'' and ''or'' are pri1narily connected with propositions and sentences, and only - if at all - secondarily with terms and noun or verb phrases. In logic, for example, ''F(x) A F(y)'' is a well-formed formula, whereas ''F(x A y)'' is not permissible. In generative grammar analogously, ''Jack and Jill went op the hill'' is derived from ''Jack went up the hill'' and ''Jill went up the hill''. Whether this is a characterization of language or of a particular language or group of languages, discovered from the relative simplicity of a system of linguistic description, or whether generative grammar has inherited this from the Western logical tradition, is not immediately clear.' I believe that the second alternative is nearer the truth and that logic has been the main source of inspiration of the transformational approach. To illustrate this point, the following arguments can be adduced. Since antiquity, logic has undergone many important modifications and has received quite different interpretations. But a basic idea which has remained constant is that logic deals, in the first place, with certain terms (subjects), to which certain properties or qualities are assigned (also called attributes, predicates). A fundamental notion, therefore, is the 'simple' or 'elementary proposition' (equivalently, the 'atomic assertion' or 'atomic sentence'), this being the result of assigning one irreducible predicate to one irreducible subject. Now, an assumption which seems to have been quite generally accepted in logic is that any complex or compound proposition or assertion (in particular, any proposition containing coordinated items) can be analysed into, or synthetized from sirnple propositions. If a proposition like 'The subject s has the property P' is written as 'P(s)', then, according to this assumption, any expression of the '

6.1. INTRODUCfION

117

general type of (1) would be fully equivalent to the expression (2), in which only simple propositions appear:

(1) P 1 &P2& ... &Pn (s1&s2& ... &sn) (2) P 1(s 1) & P 1(s 2) & ... & P 1(sn) & P 2(s1) & P2(s2) & ... & P2(sn) & ··· & P n(S1) & P n(s2) & ··· & P nCsn) It other words, in logic no coordinated predicates are recognized which cannot be reduced to si1nple predicates, nor coordinated subjects which 1 cannot be reduced to si1nple subjects. This is the exact parallel of what we have ter1ned the 'reduction postulate' (cf. section S. 4. above). The equation of this logical principle with the view underlying transforrnational theory is reinforced by the fact that the logical analyses which we find in logical treatises are, in many cases, exactly analogous to the syntactic descriptions proposed by transfor1national gra1n1narians. Moreover, the theory of transformational generative gram1nar has clearly been inspired to a great extent by certain developments in modern symbolic logic, while its proponents have also explicitly endorsed some of the earlier logically inspired approaches to language, as we have already seen above (chapter 5, fn. 56), and as will become even more evident below (section 6. 3.). It is hardly possible to deny a fundamental connection between the logical analysis of compound propositions and the transforn1ational 2 description of complex sentences. In the following sections I shall present some evidence for this connection by means of an (admittedly incomplete and somewhat haphazard) anthology from logical and logico-linguistic treatises of different periods. I shall pay some special attention to the views of the so-called PortRoyal grammarians, since it has by now become fashionable in transformational gran11nar to refer to these as the real originators of 'modern

1

Cf. Russell 1962 (1940]: 38: 'The most complete part of logic is the theory of conjunctions. These, as they occur in logic, come only between whole sentences; they give rise to molecular sentences, of which the atoms are separated by the conjunctions. This part of the subject is so fully worked out that we need waste no time on it.' 1 Compare, e.g., the following statements: 'As kernel sentences, one finds it necessary to choose what usually are called simple, active, declarative sentences, with no complex NPs ... or VPs ... ' Chomsky 1957b: 286, cf. 1957a: 106-7. ' ... it may very well be that the kernel always contains only simple, active, declarative, indicative statements.' Lees 1957: 375. Cf. also the use of the term 'proposition' for the 'sentences' introduced in the base component of a transformational grammar (above, chapter 5, fn. 56). On logical influences in transformational theory, cf. also Uhlenbeck 1967: 265 ff.

118 THE BACKGROUND OF THE TRANSFORMATIONAL APPROACH

syntax'. We shall also find that even among the logicians there are some who have voiced their doubts about the value of the reduction postulate.

6. 2. Aristotle In its main outlines the reduction postulate can be traced as far back as Aristotle. In chapter 8 of his De interpretatione, a si1nple proposition is defined as 'an affrrn1ation or a negation signifying one thing about one 3 other thing'. But it is not so much the linguistic expression which, according to Aristotle, determines whether we are dealing with a simple proposition. Linguistic unity of the expressions for subject and predicate is no warrant for unity of propositions: 'Suppose someone were to use the term garment for 'horse and man'. In that case, the proposition 'Gar1nent is white' would itself not be a sirnple affirmation. For to say this is in no way different from saying: 'Horse and man is white', and this, in turn, is in no way different from saying: 'Horse is white and man is white'.' It is, therefore, an extra-linguistic (referential) 'unity' in terms of which simple propositions are defined by Aristotle. Compare also De interpretatione, chapter 11 : 'Saying one thing of many or many things of one (positively or negatively) does not constitute a simple affir1nation or negation, unless that which consists of many is in fact one.' From this latter passage it appears that Aristotle would not apply a reduction such as the one exemplified in 'Horse and man is white' vs. 'Horse is white and man is white' in those cases where a complex linguistic expression in fact refers to one 'subject' or 'predicate' (these ter1ns being taken in their original logical sense). Unfortunately, he gives no examples of irreducible coordinated expressions in this context, though elsewhere he warns us that an expression such as 'Five is two and three' should not be taken as equivalent to 'Five is two and five is three' (De Sophisticis Elenchis 166 a 33-35, cf. below chapter 11 fn. 2). In any case, it is not the structure of the linguistic expressions in which he is primarily interested, but rather the logical structure of 'things said' by means of linguistic expressions. As is evident from the few passages quoted here, he did not 4 regard these two kinds of 'structure' as coinciding.

3

Cf. also De interpretatione 19 b 6-7, Analytica posteriora 93 b 36-7, Poetica 1457 a 28-9. ' Cf. Dinneen 1967: 82, Robins 1951 : 24-5.

6.3. PORT-ROYAL

119

6.3. Port-Royal Even this short survey of Aristotle's ideas shows that much the same position is taken by the authors of the two influential seventeenth-century books which have often been mentioned by Chomsky as his intellectual precursors, viz. the Grammaire genera/e et raisonnee (Lancelot & Arnauld 1660), and the Logique, ou l'art de penser (Arnauld & Nicole 1662). Nor is this very surprising, since all works of this kind are, of course, fir1nly 5 based on Aristotle's Organon. In Lancelot & Arnauld, there are no clear instances of the reduction of coordinations, but that they would have treated them along the lines of the reduction postulate is at once clear from their example (3), a proposition of which they state that it comprises, 'du moins dans notre esprit', the three judgments of (4):

(3) Dieu invisible a cree le monde visible. (4) a. Dieu est invisible. b. II a cree le monde. c. Le monde est visible. 6

(1676: 58-9). Notice, however, that they present this reduction as a purely psychological hypothesis. The tern1s 'judgment' and 'proposition' 7 refer, in their terminology, to purely mental phenomena (ibid. 24-5). In the second place, they present their observations in the context of an attempt to arrive at the principles governing 'les diverses formes de la si · cation des mots' (ibid. 23), i.e., they regard their analyses as illuminating the semantic structure of the complex propositions involved. In particular, they nowhere suggest that they are presenting a syntactic analysis of the complex linguistic structures expressing these propositions, 1

For the general character of the Port-Royal approach and its relations to Aristotelian logic, see Ducrot 1966a: 8. For an evaluation, cf. also Verburg 1952: 327-32. • All page references are to the 1676 edition. 1 The judgment is the product of the second of the three basic operations of the human mind: concevoir,juger, and raisonner (ibid.24). The term 'proposition' isequisignificant, in their tern1inology, to 'judgment': 'Le jugement que nous faisons des choses, comme quand je dis; la terre est ronde, s'appelle PROPOSITION.' (ibid. 25). They do not make the distinction, which Chomsky attributes to them (1966c: 33), between 'judgment' as a form of thought, and 'proposition' as its linguistic expression. Cf. also Amauld & Nicole 1775 (1662): 109: 'Ce jugement s'appelle aussi proposition.'

120 THE BACKGROUND OF THE TRANSFORMATIONAL APPROACH

nor do they suggest that this syntactic structure should in some way involve, or be derived from the simple propositions which are assumed to be present in the mind when complex expressions are used. Finally, and a fortiori, they do not speak of anything like 'transfor1national rules' fo·r deriving the complex structures from underlying sirnple ones. 8 Chomsky's assertion, therefore, that in the Port-Royal grammar we are presented, in nuce, with a transforrnational view of sentence structure is true only in a sense in which it is also true to say that some twenty centuries earlier Aristotle, in fact, discovered the transformational principle. As to the Logique, I must say, paradoxically, that though hailed by Chomsky as corroborating the transforn1ational view (see especially Chomsky 1966c: 35ff.) it contains in fact certain interesting observations which reduce the applicability of the reduction postulate and run counter, therefore, to the transformational approach. Arnauld & Nicole distinguish between simple, complex, and compound propositions. Simple propositions are those in which a single simple predicate is affirmed or denied of a single simple subject. A proposition is compound, when it contains more than one subject, or more than one predicate, or both. It is complex, when it contains only one subject and one predicate, but either subject or predicate or both are not simple but complex, i.e., are already the result of a combination of 'ideas' or concepts (Amauld & Nicole, Part II, chapter5). These distinctions can best be demonstrated by the following examples: Simple propositions: (5) a. Dieu est juste. b. Dieu n'est pas juste. Complex propositions: (6) a. Dieu qui est invisible a cree le monde qui est visible. b. Tout homme qui ne craint rien est Roi. Compound propositions: (7) a. Les biens et les maux, la vie et la mort, la pauvrete et les richesses viennent du Seigneur. b. Alexandre a ete le plus genereux de tous les Rois, et le vainqueur de Darius. 8

Cf. Chomsky 1964: 15-6, 1965: 117-8, 137, 199, 1966b: 16-7, 1966c: 31ff.

-

6.3. PORT-ROYAL

121

The point of view taken by Amauld & Nicole is that for any compound proposition there are just as many corresponding simple propositions as there are combinations of its subjects and predicates. But this is by no means necessarily true of complex propositions. Complex subjects and predicates may in certain cases indeed be said to contain 'subsidiary propositions' (propositions incidentes); but then we use the terrn 'proposition' in a very special sense, for items which are, qualitate qua, part of the subject or the predicate: 'Et la raison de cela est, que les propositions jointes a d'autres par des qui, OU ne sont des propositions que fort imparfaitement ... , ou ne sont pas tant considerees con11ne des propositions que l'on fasse alors, que conune des propositions qui ont ete f ait auparavant, et qu'alors on ne fait plus que concevoir, comme si c'etaient de si1nples idees.' (ibid.). The distinction to which the authors allude with the alternative contained in this quotation is that between explicative and deter1ninative relative clauses, a distinction which is further worked out in the foil owing chapter. Explicative are those relative clauses, which express a property inherent or i1nplied in the antecedent; detern1inative are those which restrict the scope of the antecedent. Thus, in (6a) qui est invisible and qui est visible are both explicative (the latter might also be taken as determinative, if there were another, invisible world). In (6b) qui ne craint rien is deterrninative. Now, in the case of explicative relative clauses, substitution of the antecedent for the relative pronoun results in a true proposition: corresponding to (6a), we have: (8) a. Dieu est invisible. b. Le monde est visible. But in the case of determinative relative clauses, this is not possible: 'Car si, apres avoir dit, les hommes qui sont pieux sont charitables, on vouloit substituer le mot d'hommes au qui, en disant: Les hommes sont pieux, la proposition seroit fausse, parce que ce seroit affirm er le mot de pieux des honunes comme ho1nrnes; mais en disant, Jes hommes qui sont pieux sont charitables, on affirme ni des hommes en general, ni d'aucuns hommes en particulier, qu'il soient pieux; mais l'esprit, joignant ensemble l'idee de pieux avec celle d'hommes, et en faisant une idee totale, juge que l'attribut de charitable convient a cette idee totale; et ainsi tout le jugement qui est exprime dans la proposition incidente est seulement celui par lequel notre

122 THE BACKGROUND OF THE TRANSFORMATIONAL APPROACH

espritjuge que l'idee de pieux n'est pas incompatible avec celle d'homme, et qu'ainsi ii peut les considerer com1ne jointes ensemble, et examiner ensuite ce qui leur convient selon cette union' (ibid. chapter 6). It is evident from these passages that a complex proposition like (9)

Les hommes qui sont pieux sont charitables.

cannot be brought in connection with a proposition like (10)

Les hommes sont pieux.

Consequently, if we use the ter1n 'proposition' for determinative relative clauses, we extend it to units which have only a minimum of similarity with 'real' propositions. Chomsky's summary of these views (1966c: 38) is, in this respect, quite misleading: 'Thus sentences containing explicative as well as restrictive relative clauses are based on systems of propositions (that is, abstract objects constituting the meaning of sentences); but the manner of interconnection is different in the case of an explicative clause, in which the underlying judgment is actually affirmed, and in a deter1ninative clause, in which the propositionformed by replacing the relative pronoun by its antecedent is not affir1ned but rather constitutes a single complex idea together with its noun.' While Arnauld & Nicole are careful to stress the fact that in deter1ninative relative clauses no real 'underlying' propositions are involved, Chomsky nevertheless wants an underlying proposition like (10) to play a role in the syntactic and semantic structure of the complex sentence. This is clearly only because transformationalists have so far always taken the view that all relative clauses are derived by relativization from structures underlying the corresponding simple sentences. One may well challenge them, however, to suggest a satisfactory derivation of (9) from the structures underlying (11) a. Les hommes sont charitables. b. Les hommes sont pieux. Within the framework proposed so far, this is utterly impossible, unless the basic hypothesis that the underlying structure(s) should contain all

6.3. PORT-ROYAL

123

relevant semantic information necessary for the interpretation of the derived sentence is dropped. Nor is it easy to find underlying structures other than those of ( 11 ), on which a satisfactory derivation might be based. To conclude this section, we must say that Chomsky not only attributes to the Port-Royal logicians something which they have never had the intention to propose (namely, a transfor111ational theory of sentence structure), but also that what they say in fact is, in part, incompatible 9 with the transformational theory as it has recently been developed. That I have dealt at such length with the Port-Royal observations on relative clauses is justifiable by the fact that precisely the same problems arise in connection with coordinated constructions. The Port-Royal tenets may be summarized as follows: (i) For certain non-simple propositions there are equivalent sets or systems of simple propositions; (ii) In these cases the latter play a part in the interpretation of the former; (iii) For certain other non-simple propositions there are no such sets or systems of equivalent propositions. Observe that (i) and (iii) are no doubt correct observations, while (ii) is a hypothesis which remains to be proved. Observe, further, that it is in no way suggested that the linguistic expressions corresponding to complex propositions of type (i) should be grammatically described in tern1s of the equivalent simple propositions. The Port-Royal views do not constitute, therefore, a starting-point for a new theory of syntax (as suggested, e.g., in Chomsky 1965: 137), but a refined continuation of the Aristotelian view of logical analysis, couched in psychological terrns. If we translate the Port-Royal views into linguistic ter1ns and apply them to coordinative constructions, the only things we can say are: (i) For some coordinative constructions, there are equivalent sets or systems of non-coordinative constructions; (ii) For other coordinative constructions, there are no such equivalent sets or systems. We have seen that these statements are correct (see above, sections For further doubts about the status of 'Cartesian linguistics' and its present-day theoretical relevance, cf. Lakoff 1966, Hannaford 1967, Mounin 1967: 129-30.

1

124 THE BACKGROUND OF THE TRANSFORMATIONAL APPROACH

5. 5. 4. ff.). This means that the difference between cases (i) and (ii) must be accounted for in an adequate gram1nar. We have also seen that transfor1nationalists express this distinction by differentiating between 'sentence conjunction' and 'phrasal conjunction' (above, section 5. 7. 1. ff.). There is no reason to assume, however, that this is the only or even the best way in which this can be achieved. In fact, we shall see that in the non-transfor1national descriptive system sketched below, this difference can be accounted for in a quite sitnple and generally applicable statement.

6. 4. More recent versions of the reduction postulate In the work of more recent logicians and logically inclined linguists, the reduction postulate, especially as applied to coordinative constructions, is almost unanirnously adhered to. Cases not reducible to simple propositions or sentences are sometirnes mentioned, but only very rarely are they taken as a sign that there is something basically wrong with the reductive approach as such. As an illustration of this situation, consider the foil owing quotations. (1) John Stuart Mill represents an extreme version of the reductionist approach in that he even treats conjunctions like and and but themselves as 'abbreviations of propositions': 'Thus the words, Caesar is dead and Brutus is alive, are equivalent to these: Caesar is dead: Brutus is alive; it is desired that the two preceding propositions should be thought of together.' (1911 [1843]: 52). This reduction is clearly at the same time a reductio ad absurdum of this approach if applied to grammatical description. The expression 'it is desired that the two preceding propositions should be thought of together' can hardly be taken as even approximately accounting for the semantic aspect of and, let alone that it is in any way grammatically relevant to the description of the sentence concerned. This is illustrative of the fact that as soon as we start reducing complex items to underlying simple propositions, it is extremely difficult if not impossible to determine where we have to stop~ In fact, we can develop almost any single word into a proposition. Compare the remarks on Schachter and the quotation from Sandmann below (sub (2) and in section 6. 5.). More directly relevant in the present context is the following statement of Mill's:

6.4. MORE RECENT VERSIONS

125

'For brevity ... and to avoid repetition, the propositions are often blended together: as in this, ''Peter and James preached at Jerusalem and in Galilee," which contains four propositions: Peter preached at Jerusalem, Peter preached in Galilee, James preached at Jerusalem, James preached in Galilee' (ibid.). Notice that unlike the Port-Royal logicians Mill uses the term proposition to refer to linguistic expressions (ibid. 12), and that, whereas they were careful to add 'au moins dans notre esprit', he simply asserts that the complex expression 'contains four propositions'. As has become clear in the preceding chapter, Mill's sentence would receive exactly the same analysis in a transformational grammar. For examples, cf. Fidelholtz 1964: 3 on the sentence (12) John and Bill hit Mary and Sally. and Schane 1966: 40 on the sentence: (13) Sandy and Mary like iguanas and snakes. (2) Schachter (1935) explicitly tried to arrive at a logically inspired description of grammar. No wonder, therefore, that we find such statements as the following: '''Und'' verbindet in unserer Sprache hauptsachlich Satze. Oft sind die einzelnen verbundenen Satze fragmentarisch, so dasz man, dadurch irregefiihrt, die Konjunktion als eine Verbindung zwischen Begriffen auffaszt. Z.B. ''X und Y und Z gehen''. Dieser Satz ist aber zu zerlegen in: ''X geht und Y geht und Z geht''.' (166). '''A und B und C gehen, aber D und E gehen nicht''. Hier sind fiinf Aussagen durch Konjunktionen verbunden.' (168). 'Auch die Disjunktion verbindet zunachst Satze. Tritt sie verbindend zwischen W orten auf, so ist sie in eine SV [ - Satzverbindung, SD] zu iibersetzen.' (171) It should be noted that Schachter goes much further in his reductions than any transformational analysis proposed so far (cf. especially pp. 184, 186). Everything implied in a sentence is explicitated in his analyses. Again, as grammatical descriptions these analyses are hardly useful. But they cannot easily be excluded as soon as the reductionist principle has been accepted.

126 THE BACKGROUND OF THE TRANSFORMATIONAL APPROACH

It is interesting to see, however, that Schachter also came across the irreducible cases of coordinative constructions: 'Die Konjunktion kann sich aber auch auf Be · e beziehen. Dies ist bei vielen ''Grammatischen Satzen'' der Fall, wo die Konjunktion Merkmale (Punkte verschiedener Systeme) verkniipft. Aber auch bei sonstigen Satzen: ''X ist zwischen Y und Z''; ''ich, du und Karl sind drei Leute''; ''zwei und zwei sind vier''. In einem Ladenfenster ist folgender Satz zu lesen: ''Ein X und ein Y kosten einen Schilling'', wobei der Preis so zu verstehen ist, dasz X nur mit Y verkauft wird. Dieser Satz ist also nicht zerlegbar in die beiden Satze: X kostet 50 g und Y kostet 50 g.' (166) But these examples are clearly regarded as exceptional cases, and no clear grammatical consequences are drawn from them.

(3) Russell. In the following passage from Russell, all points discussed return in a particularly telling manner: 'A single gra1n1natical sentence may not be logically single. 'I went out and found it was raining' is logically indistinguishable from the two sentences: 'I went out', 'I found it was raining'. But the sentence 'when I went out I found it was raining' is logically single: it asserts that two occurrences were simultaneous. 'Caesar and Pompey were great generals' is logically two sentences, but 'Caesar and Pompey were alike in being great generals' is logically one. For our purposes, it will be convenient to exclude sentences which are not logically single, but consist of two assertions joined by 'and' or 'but' or 'although' or some such conjunction. A

single sentence, for our purposes, must be one which says something that cannot be said in two separate simpler sentences.' (1962 [1940]: 28). Notice especially the careful distinction between logical and grammatical in this passage, and the repeated restriction 'for our purposes'. (4) Reichenbach. Much less careful is Reichenbach, who simply asserts: 'In conversational language some of these words [i.e., and, or, etc. SD] are sometimes used to combine, not sentences, but words, such as in 'Peter or William will go with you'. We consider this sentence forrn an abbreviation standing for 'Peter will go with you or William will go with you'.' {1947: 23). (5) For a rare dissident among the logicians, we may turn to P. F. Strawson. Again, a full-length quotation seems to be justified, since Strawson summarizes most of the relevant points:

6.4. MORE RECENT VERSIONS

127

''And' can perfor1n many jobs which '.' cannot perfor1n. It can, for instance, be used to couple nouns ('Tom and William arrived'), or adjective~ ('He was hungry and thirsty'), or adverbs ('He walked slowly and painfully'); while'.' can be used only to couple expressions which could appear as separate sentences. One might be tempted to say that sentences in which 'and' coupled words or phrases, were short for sentences in which 'and' coupled clauses; e.g. that 'He was hungry and thirsty' was short for 'He was hungry and he was thirsty'. But this is sirnply false. We do not say, of anyone who uses sentences like 'Tom and William arrived', that he .is speaking elliptically, or using abbreviations. On the contrary, it is one of the functions of 'and', to which there is no counterpart in the case of'.' to form plural subjects or compound predicates. Of course it is true of many statements of the forms 'x and y are/' or 'x is f and g', that they are logically equivalent to the corresponding statements of the form 'x is/ and y is/' or 'x is/ and x is g'. But, first, this is a fact about the use, in certain contexts, of the word 'and', to which there corresponds no rule for the use of '.' And, second, there are countless contexts for which such an equivalence does not hold.' (1952: 79-80). Strawson then goes on to cite such examples as 'Tom and Mary made friends' and the like, which I have treated above in sections 5. 5. 5. and 5. 5. 6. Notice that these remarks of Strawson exactly represent the point of view which I arn defending in this study. I shall have occasion to return to the problem of the exact relation between the logical connectives and the coordinators of natural language in chapter 12 below. (6) Tesniere. The reduction postulate has also had its influence on linguistics in the pre-Chomskyan period. Especially in France the Port-Royal influence on linguistic theory can be assessed. As an example, consider 10 Tesniere 1959. Tesniere characterizes coordinations as resulting from the addition of si1nple sentences: 'Alfred tombe +Bernard tombe Alfred et Bernard tombent' (ibid. 325). But just as in the Port-Royal gra1n1nar, Tesniere does not seem to regard this as a grammatical analysis. In his structural descriptions of coordinations, he rightly concentrates on what is given within these constructions as such by way of relevant structural relations (ibid. 339 ff.). In discussing the sentence:

°

1

For an earlier example, cf. Antoine 1962: 705 fn. 1.

128 THE BACKGROUND OF THE TRANSFORMATIONAL APPROACH

(14)

Les maitres, les pedagogues et les educateurs donnent, repetent et ressassent des avis, des conseils, et des avertissements aux ecoliers, aux collegiens, et aux lyceens.

he notes that there are 81 'simple' sentences corresponding to it and that 'le sujet parlant qui prononce la phrase ci-dessus exprime done en une seule phrase le contenu de 81 phrases differentes.' (ibid. 346). Clearly, the emphasis is on the particular properties of the complex sentence rather than on its reducibility to the 81 corresponding simple sentences, though 11 the relation of (14) to the latter is explicitly indicated by Tesniere.

6. 5. Conclusion The somewhat random survey presented above seems to justify the following systematic comments: The reduction postulate, which has been shown to have originated in logical analysis, can be and has been interpreted in three different ways: (i) logically, (ii) psychologically, and (iii) linguistically. (i) If it is interpreted as sirnply constituting a part of logical systems, and if no direct empirical correlates are clai1ned for these systems, then obviously the postulate can be simply accepted by definition. (ii) If it is taken as a hypothesis concerning the psychological process of interpretation, we can do little more than leave the verification of this hypothesis (if it is verifiable at all) to psychology. I may add that in my opinion such a psychological hypothesis would not have a high degree of initial plausibility. I do not see why the human mind should not be capable of direct apprehension and interpretation of complex 'judgments', without resorting to the simple ones in which these could be analysed. (iii) If the reduction is interpreted as relevant to the linguistic structure of sentences, this can again be done in two ways: (1) semantically, and (2) grammatically. (1) In the former case, the reduction postulate states that the content of the underlying simple sentences is relevant to the content of the cornOn Tesniere's position as compared with transformational generative grammar, see also Grunig 1965 : 5-6.

11

6.S. CONCLUSION

129

plex sentences involved. This can, again, lead to two crucially different • • pos1t1ons: (a) The content of a complex sentence is judged to be equal to the content of the simple sentences to which it can be reduced. (b) The content of a complex sentence is judged to be the total content of these si•nple sentences. (2) If the reduction postulate is interpreted grammatically, it states that the gram1natical structure of the complex sentence should be described in ter1ns of the gra1n1natical structure of the sirnple sentences to which it can be reduced. In the theory of transformational generative grammar, we find this second, grammatical interpretation of the reduction postulate coupled to a semantic interpretation congruent with alternative (lb): the gra1nmatical structure of a complex sentence must be described in terms of the gra1n1natical structure(s) of its underlying sentences, and the content of the complex sentence is the sum total of the contents of these underlying structure(s). We can say, therefore, that in this way an originally logical principle has been carried over into the purely linguistic description of syntax and semantics. The considerations of the preceding chapter have shown, however, that this principle is utterly incapable of accounting for the linguistic facts as they really are. The linguistic interpretation of the reduction postulate creates a host of problems for which no satisfactory solution has as yet been proposed, and leads to consequences which are linguistically unacceptable. Let us close this chapter by quoting a co1nment by Sandmann on Mill's version of the reduction postulate. Sandmann's remarks succinctly summarize some of the problems involved, and are quite in line with the position which I should like to take in this matter: 'The impression that coordination is concerned with a coordination of propositions is, however, misleading. An analysis such as the one offered by Stuart Mill amounts merely to this, that hemakes the direct relationship between concepts explicit for each coordinated term separately. But in this way we get the impression that a statement of the form Peter and James preached at Jerusalem and in Galilee is a combination of coordinated propositions, whereas in fact it is the terms Peter and James, Jerusalem and Galilee which are coordinated. If we attribute to the breaking

130 THE BACKGROUND OF THE TRANSFORMATIONAL APPROACH

up of the quoted statement into smaller propositions any other sense than that of rendering the direct conceptual relationships explicit for each coordinated ter•n, we would hardly ever get anything but coordinated propositions. For the so-called ''simple'' statement Peter preached at Jerusalem could equally well be considered as a combination of propositions: Peter preached+ the preaching was at Jerusalem; Peter preached could be dissolved into there is Peter+ he preached, etc... Coordination in particular is not merely the effect of an act of combining propositions.' (1950: 31-2).

7. The relation of coordination to constituent structure

7 .1. The coordinated member

single constituent hypothesis

In the general definition of coordination (above section 4.1.), I have required that the coordinated members be equivalent as to grammatical function. Quite apart from the question of exactly for1nulating this notion of gram1natical function, to which I return in chapter 8, there is an i1nportant problem behind it, which I shall now first discuss. If the members of a coordination are to be equivalent as to gra1n1natical function, it is, of course, i111plied that they must each have gra1n1natical function. This would most naturally be taken to mean that they must be constituents on some level of structure, the ter111 'constituent' being used in the sense of: a linguistic element (si1nple or complex) which is relevant as a whole on some level of the grammatical hierarchy in the structure of a complex linguistic expression. In other words: in the constituent-analysis of coordinations there should, under this interpretation, always be a level on which all coordinated members are analysed as single constituents. We have seen that this position was taken by Chomsky (1957a: 35-6, 1965: 212 fn. 9, quoted above section 5.3.). Chomsky suggested that an important simplification of the gra1111nar would be achieved if constituents were set up in such a way that each member of a coordination would be a single constituent.

7. 2. Methodological objections On this matter I would first like to make some methodological remarks. In fact, three independent hypotheses are involved: (i) a hypothesis about the constituent structure of complex linguistic • expressions; (ii) a hypothesis about the nature of coordinations and their members; (iii) the hypothesis that the members of a coordination are always single • constituents.

132

COORDINATION AND CONS'l'I'l'UENT STRUCTURE

It see1ns clear that (iii) can only be verified if (i) and (ii) have been established independently. In Chomsky's approach, however, (i) and (ii) are made subordinate to (iii): the suggestion is to set up constituents in such a way as to confor1n to (iii) as closely as possible, in the name of simplicity. In this way, the hypotheses (i), (ii), and (iii) are 1nixed up with each other, and the results acquire an air of circularity. This is particularly evident when sentences which under no constituent analysis con~ form to (iii) are classed as 'semi-gra1n1natical', or even as 'ungrammatical' or 'impossible'. The first course is taken by Chomsky 1957a: 35-6, fn. 2: 'The most reasonable way to describe this situation would seem to be by a description of the following kind: to forrn fully grammatical sentences by conjunction, it is necessary to conjoin single constituents; if we conjoin pairs of constituents, and these are major constituents (i.e., 'high up' in the diagram ... ), the resulting sentences are semi-gra1n1natical; the more completely we violate constituent structure by conjunction, the less gramrnatical is the resulting sentence. This description requires that we generalize the grammatical-ungra1n1natical dichotomy, developing a notion of degree of grammaticalness.' In this way, hypothesis (iii) becomes a postulate accepted by fiat. It is interesting to notice that this postulate has subsequently been used to defend decisions on constituent structure. As is well known, a sentence like (1) The man hit the ball. is analysed by generative grammarians, in accordance with the subjectpredicate analysis of traditional grammar and with the prevalent practice of IC-analysis, into a noun phrase (The man) and a verb phrase (hit the ball) (Chomsky 1957a: 26 etc.). This analysis has been criticized by Reichling (1961: 15) and Uhlenbeck (1963: 7-8, 1967: 275-6) on the ground that the direct grammatical relation obtaining between subject (The man) and finite verb (hit), which in the present tense is manifested 1 by concord, is not accounted for in this way. To refute this criticism, Lees argues as follows (1964: xxxv): 'We may construe predicates as constituents, as in: 1

The same point is made in Grady 1967.

7 .2. METHODOLOGICAL OBJECI'IONS

133

The man hit the ball. The man hit the ball and ==· The man saw the tree. saw the tree., we may construe the subject as a constituent, as in: The man hit the ball. The man and the boy hit > The boy hit .t he ball. the ball., we may construe the object nominal as a constituent, as in: The man hit the ball. The man hit the ball and =::The man hit the tree. the tree., but we may not construe the subject nominal plus verb as a constituent of the sentence, as in: The man hit the ball. The man hit and the boy =>:• The boy saw the ball. saw the ball.' '

This argument of Lees is clearly based on hypothesis (iii) as if it were an incontestable fact; further1nore, it appears that Lees regards the sentence (2)

The man hit and the boy saw the ball.

as ungra1n1natical. The question is, what empirical validity can be assigned to the latter judgment. Is it not, rather, the automatic result of the combined hypotheses (i') that noun phrase and verb phrase are constit~ uents, and (iii) that coordinated members must be single constituents? What value must be assigned to an expression like 'we may not construe x as a constituent' if the result is, according to Chomsky, not an ungrammatical, but a semi-grammatical sentence? Chomsky, in fact, had said no more than that 'many would question the gra1n1naticalness' of sentences like: (3) John enjoyed and my friend liked the play. and that such sentences are 'much less natural' than the alternative: (4) John enjoyed the play and my friend liked it. (1957a: 35 fn. 2). But no matter whether sentences like (2) and (3) are branded as 'semigra1n1natical' or as 'ungrammatical' or 'impossible', the result is a perfect circle if all sentences not conforming to (iii) are simply denied full gra1nmaticalness. In this way, (iii) can in no way be subjected to empirical verification.

134

COORDINATION AND CONSTITUENT STRUCfURE

This point emerges even more clearly in the later phase of the transfor1national theory. In 1957, Chomsky still regarded empirical verification of the notion 'grammaticalness' as possible: 'One way to test the adequacy of a grammar proposed for L is to determine whether or not the sequences that it generates are actually gra1n1natical, i.e., acceptable to a native speaker, etc. We can take certain steps towardsprovidingabehavioral criterion for grammaticalness so that this test of adequacy can be carried out.' (1957a: 13). But in 1965, we are warned that 'The notion ''acceptable'' is not to be confused with ''gra1nmatical'' .' (1965: 11). Sentences may be grammatical, but not or hardly acceptable to native speakers. 'Correspondingly, although one might propose various operational tests for acceptability, it is unlikely that a necessary and sufficient operational criterion might be invented for the much more abstract and far more i1nportant notion of grammaticalness.' (ibid., cf. also 19). In this way, a judgment like 'this sentence is grammatical in this language' cannot be questioned except by an appeal to the over-all framework of the grammatical system proposed. 'Grammatical' becomes equivalent to 'conforming to the rules proposed in the grammar'; consequently, any gramrnar will by definition always describe gra1nmatical sentences, and the notion 'grammatical sentence' will shift with each revision of the grammar. It may very well be that 'gra111matical' should not be equated with 'acceptable to a native speaker'. But as far as I know, no suggestions have been advanced so far to arrive at an interpretation of 'grain1nat2 ical' such that the imminent danger of circularity might be avoided.

7 .3. Empirical objections This would all be very well, if there were no empirical facts bearing on hypothesis (iii). But this is certainly not the case. Consider the following remarks made in this connection by Gleitman (1965: 264-5): 'A more serious difficulty is that certain conjunctions of non-constituent sequences of constituents are uniformly accepted by speakers: 29. 1

I gave the boy a nickel and the girl a dime.

This is confirmed by the detailed analysis of the development of the notion 'grammatical' in Chomsky's writings, presented in Luif 1967. Cf. also the remarks made on this topic by Grunig 1966: 82-4, 91-2.

7. 3. EMPIRICAL OBJECI'IONS

135

The soviets rely on military and on political indications of our intentions. 31. He took John home and Mary to the station. 32. The conjunction of an interrogative and an imperative sentence is excluded. 33. The man was haggard and the girl sick with exhaustion.' 30.

Gleitman further remarks: 'Other conjunctions that cross constituent boundaries are judged awkward but acceptable. Still others are rejected: 35. ?The man saw and the woman heard the shot fired. 3 36. *I want to know why John and when Mary are coming.'

In all these cases combinations of constituents which are not themselves single constituents are coordinated. In the majority of cases this leads to perfectly acceptable sentences. Notice also that sentence 35 was judged 'awkward, but acceptable'. This sentence is of the same type as sentences ,(2) and (3) above, on which I have quoted the opinions of Lees and Chomsky. Despite Gleitman's warnings about the crudeness of the methods used to elicit judgments on the acceptability of these sentences (ibid. 260-2), and her attempts to assign a lower degree of gra1nmaticalness to sentence 4 30 (ibid. 265), it is evident that the completely unobjectionable cases 29, 31, 32, and 33 are sufficient to falsify the coordinated member single constituent hypothesis. Not only can further counter-examples to (iii) be quite easily adduced from English as well as from other languages, but it should also be observed that strict adherence to (iii) in many cases leads to contradictory analyses, unless a host of perfectly acceptable sentences are excluded from being generated by the grammar. Compare, e.g.: ( 5) An old or a young man. 5 (6) The old men and old women are here. 1

? = 'awkward, responses not uniform'. (ibid. 262). ' Cf. also Gleitman's remarks on the solution offered by Chomsky (quoted in the preceding section) (ibid.). ' Schane (1966: 19) calls (6) 'perfectly acceptable' (cf. also ibid. 49-50). Gleitman

136

COORDINATION AND CONSTITUENT STRUCTURE

On the basis of (5) we would want to set uparticle+adjectiveasa constituent; (6) would then be semi- or ungrammatical (this course is suggested by Gleitman 1965: 273 fn. 19, 274). But if we start from (6), we would be forced to take adjective+noun as a constituent (as in Braun 1967: 81), and (5) would be semi- or ungra1n1natical. If, on the other hand, we analyse noun phrases of this type as in (7): (7)

NP ..

Art

Adj

Noun

(an analysis given, e.g., in Lees 1961a: 176, Gleitman 1965: 272, and Schane 1966: 17), then both (5) and (6) would, on hypothesis (iii), be semi- or ungrammatical. Schane does not accept this and tries to solve the dile1nma by restricting the scope of his 'primary conjunction rules' (cf. above section 5. 7 .1. 3.) to major categories which are not also lexical categories (i.e., to S, NP, VP, etc., but not to N, V, etc.). These primary rules will then only describe sentences in which the whole noun phrase is conjoined, as in: (8) (9)

An old man or a young man. The old men and the old women are here.

Cases like (5) and (6) are then derived from the structures underlying (8) and (9), respectively, by means of his 'secondary rules' (i.e., deletiontransformations). But both the sentences described by Schane's primary rules and those derived from these by secondary rules are, in his approach, considered to be fully grammatical, and by this very fact hypothesis 6 (iii) is again shown to be inadequate. More interesting still is Schane's example (1966: 27): glosses a similar example (coordination of adjective+noun) as unacceptable (1965: 264; but cf. the restrictions made on p. 273 fn. 19). If phrases like (6) should tum out to be always unacceptable in English, the argument could be repeated for Dutch, where a phrase like De oude mannen en jonge vrouwen 'The old men and young women' is quite acceptable. • Ross's 'gapping rule' leads to similar results (Ross 1967). Cf. also the suggestion of Koutsoudas ( 1966: 249 fn. 2).

137

7. 3. EMPIRICAL OBJECTIONS

(10) John was in Cambridge and saw Mary on Tuesday.

If in this sentence the phrase on Tuesda.y is taken as modifying both members of the coordination, then it is again counter to (iii), unless it is analysed in such a way that was in Cambridge and saw Mary are each single constituents. In the new format of the first few base component rules, as proposed by Chomsky (1965: 102), this is not the case. These rules specify a structure as indicated in (11):

(11)

s Predicate-Phrase

NP

(Place)

VP

Aux

v

NP

(Time)



If, in this structure, Time is selected (e.g., on Tuesday in (10)), then neither Aux+ VP+ Place (was in Cambridge), nor Aux+ V +NP (saw Mary) is a constituent. For this reason Schane (ibid.) gives a different set of rules, specifying a structure like (12):

(12)

s NP

Predicate-Phrase VP

VB Aux

(NP) V

(Time) (Place)

138

COORDINATION AND CONS'l'I'l'UENT STRUCTURE

This analysis would indeed bring (10) in accordance with hypothesis (iii). It remains to be seen, however, whether Chomsky's arguments leading to the analysis as symbolized in (11) are not so forceful as to prevent Schane's modification. But now consider: (13)

John was in Cambridge and Peter saw Mary on Tuesday.

For this sentence, a different analysis would be necessary again, to yield something like: (14)

s S' NP

(Time)

Predicate-Phrase

so that both John was in Cambridge and Peter saw Mary would constitute a single constituent S'. And in: (15)

John saw Mary on Tuesday and met Harry on Wednesday in Cambridge.

we would rather want the Time adverbial to forn1 one constituent with Aux + V +NP, over against the Place adverbial, while in: (16)

John saw Mary on Tuesday and Harry on Wednesday in Cambridge.

NP+ Time would be desired to be a single constituent. In:

(17)

John saw Mary in Cambridge and Harry in Oxford on Tuesday.

finally, the desired single constituent would rather be NP+ Place. It seems clear that in this way one can go on suggesting new analyses for any new counter-example to (iii) coming to light. Choosing the sentences

7 .4. FURTHER COUNTER-EXAMPLES

139

conforming to any one of these analyses as 'fully gramrnatical' and condernning the others to se1ni- or ungrammaticalness then becomes a quite arbitrary matter. Despite Schane's arguments to the contrary (1966: 21-2), the distribution of these sentences over primary and secondary rules seems hardly less arbitrary. The conclusion is that no matter what course is taken, there will always remain fully gra1n1natical sentences not 7 confor1ning to (iii). For a final and quite striking example of conflicting analyses, treated by Schane in the same vein, consider: (18) John is singing and dancing. In all generative analyses proposed so far, -ing is treated as part of the constituentbe+ing, which is, in tum, part of the constituentAux. Therefore ing +Vis not a constituent, and (18) is again counter to (iii), though a fully grarnmatical sentence if ever there was one. If Aux + V is not a single constituent either, as in Chomsky's analysis (11), then (19) will also be excluded: (19) John is singing and is dancing. Schane therefore proposes to return to the earlier analysis (as given, e.g., in Chomsky 1957a: 39, 111), where Aux+ V together constitute Verb (VB in Schane's rules, cf. (12) above). In this way (19) is accounted for, and (18) can be derived by means of a secondary deletion-rule (Schane 1966: 28-9). But, if there is any difference between (18) and (19), it seems that (19) is 'less natural' than (18). If, therefore, Aux+ V are classed as a single constituent, and if on this basis the coordinations (18) and (19) are described, we are again led to a result which even within the framework of a transformational gramrnar seems hardly satisfactory.

7. 4. Further counter-examples It is easy to find further counter-examples to (iii). I now present a random 7

The same point is made in Braun 1967: 73, 80ff.

140

COORDINATION AND CONSTITUENT STRUCTURE

sample of such cases, collected from the discussions of several linguists or taken from existing written texts. Street (1967: 102-4) mentions the possibility of coordinating members which are not single constituents in Mongolian, and adduces further examples from English: (20) John got a book from his aunt and a bicycle from his parents. (21) Wait till the lights are out and the people asleep. (22) He painted the wall green and the woodwork white. (23) He gave me a book and my brother a bicycle. With inverted order of the constituents within the coordinated members:

(24) I'll give you three dollars for all the books but for one of them no more than fifty cents. Street's proposals to account for these cases within the fra1nework of ICanalysis will be discussed below (section 11 . 2. 1.).

I I

I

Further examples from English:

John got the bike and I the scooter. 8 Smith ftew out to left field or Higgins to center - I forgot which. John will arrive at nine and Bill at ten. 9 John will ar'rive at nine and leave at ten. 10 I floated and he swam across the lake. ... the shift whereby Kepler superseded Ptolemy, or Einstein Newton, 11 or Darwin Aristotle. (31) This study will touch on a variety of topics in syntactic theory and English syntax, a few in some detail, several quite superficially, and 12 none exhaustively. (25) (26) (27) (28) (29) (30)

I

I

. I

I

8

Examples (25) and (26) are from Chatman 1964: 333. • (27) and (28) are from Hudson 1967: 240. 1 ° Koutsoudas 1966: 249 fn. 2. 11 Quine, From a logical point of view (1961), 43. 11 Chomsky 1965: 3.

7 .4. FURTHER COUNTER-EXAMPLES

141

From French: (32) (33) (34) (35) (36)

Raton tire et Bernard croque les marrons. II est bon de par/er et meilleur de se taire. L'un portait sa cuirasse, I' autre son bouclier. L'huttre est pour le juge, les ecailles pour les plaideurs. 13 L'Angleterre prit l'aigle et I' Autriche l'aiglon.

From Ger1nan: (37) (38) (39) (40) (41) (42)

Gestem spielten sie Klavier und sangen sie Lieder. 14 Gestern spielten sie Klavier und sangen Lieder. Herr N. lehrt die Knaben, Herr 0. die Miidchen. Du nim1nst die linke Halfte, ich die rechte. /eh ging voran, er hinter mir. 15 Er ging langsam, sie schnel/.

From Latin: (43)

Ob... debilitatem ani1ni multi parentes, multi amicos, nonnulli patriam, plerique autem se ipsos ... perdiderunt. 'By weakness of character many have ruined their parents, many 16 their friends, some their country, but the majority themselves.' (44) Gallos ah Aquitanis Garunna flumen, a Belgis Matrona et Sequana dividit. 'The Gauls are separatedfrom the Aquitanians by the river Garonne, from the Belgians by the Marne and the Seine.' (45) Gallia est omnis divisa in partes tres, quarum unam incolunt Belgae, aliam Aquitani, tertiam qui ipsorum lingua Celtae, nostra Galli appe/lantur. 'The total area of Gallia is divided into three parts, one of which is inhabited by the Belgians, the second by the Aquitanians, and the 11

Examples (32)-(36) are cited from La Fontaine and Victor Hugo by Tesniere, who pays special attention to this type of construction (1959: 346-51). For further examples from French, see ibid. 1' According to Bliimel 1914: 96, who cites these examples, (38) is more natural than (37). 16 (39)-(42) from Bliimel 1914: 172, 189, 191, 201. Similar examples can easily be found in Dutch. 11 Cicero, De Finibus I : 49.

142

COORDINATION AND CONSTITUENT STRUCTURE

third by those who in their own tongue are called Celts, and in 11 ours Gauls.' The latter example is interesting in that it contains two coordinations of non-constituent sequences: (I) unam .. " Belgae, aliam Aquitani, tertiam qui (... ) appellantur, and (2) ipsorum ... Celtae, nostra Galli. ·

7. 5. Conclusion From the considerations and examples given above we must conclude that the hypothesis according to which coordinated members should always be single constituents on some level of structure cannot be maintained. Not only the coordination of single constituents, but also the coordination of certain pairs, triples, etc. of constituents can result in perfectly well-for1ned sentences. A fortiori, the hypothesis cannot be used to defend or to exclude any type of constituent analysis. Of course, the possibility of coordinating combinations of constituents is not unlimited, and will probably differ from language to language. The question of which (universal, general, or particular) restrictions are operative here is a matter of empirical investigation. I can now also be somewhat more specific about the requirement that the members of a coordination must be equivalent as to grarn1natical function: Given a coordination &(M 1, Ml, ... ,Mn), if M 1 , Ml, and ... Mn are each single constituents, then there must be a gram1natical function F, such that F,(M 1), F 1(M 2 ), ••• , F 1(Mn). if M 1 , M 2 , and ... Mn are each combinations of constituents, then the total set of the grammatical functions of these constituents {F,, F1, ••. , Fn} must be the same for each of the members M 1 , M 2 , ••• Mn. These functions will normally occur in the same order in each of the coordinated members, but from such examples as (24) above we see that under certain circumstances this order may be different. I now turn to the question of how this notion of 'grammatical function' should be interpreted within the total theory of gra1n1natical description.

17

(43)-(44) from Caesar, De Bello Gallico I: 1: 1-2

8. On the notion 'grammatical function'

In the course of the preceding chapters we have found on several occasions that the notion of gramrnatical function seems to be highly relevant to gra1n1natical description in general, and to the description of coordinations in particular. The question is, therefore: what do we mean exactly by the ter1n 'gra1nmatical function'? In this chapter I shall consider several attempts at an interpretation of this term, thus paving the way for the 'functional' system of grammatical description which will be developed in chapter 9. One of the indications of the relevance of functional notions to grammatical description is the fact that nearly every linguist, whatever his own theoretical point of view, makes use of them at some point or other. Since antiquity linguists have been wont to discuss grammatical structure in terms of 'subject', 'predicate', 'object', 'head', 'modifier' etc., ter1ns which refer to phenomena clearly different from such categorial labels as 'noun', 'verb', 'adjective', 'adverb', etc. If, as some hold, the for1ner ter1ns were either not based on linguistic criteria, or in some way redundant over against the categorial tern1s, how is it, we may ask, that nearly everyone continues to use them? In my opinion, the reason for this perseverance lies in the fact that gra1n1natical functions constitute a fundamental di1nension of gra1n1natical structure by the side of constituents and categories of constituents. By this I do not mean to say that the particular way in which these functions have been accounted for in traditional grarn1nar is right in every detail, nor that traditional gramn1ar has provided a complete specification of this functional dimension. I only hold that the basic distinction between functions and classes is correct, and can be elaborated on the basis of the distinctions which have traditionally been made.

8 .1. Bloomfield's view of grammatical function We have seen (above, chapter 3) that in the writings of Bloomfield and

144

ON THE NOTION 'GRAMMATICAL FUNCTION'

his followers the notion of grarnmatical function is completely reduced to position or distribution. It is clear that if this reduction of functions to 'privileges of occurrence' were completely possible, the notion of function would not be a primitive of gra1nmatical theory and could, accordingly, be dispensed with. This is what has indeed happened in most versions of distributional theory. What these have in co1n1non is that they are exclusively or at least for the greater part concerned with constituents, classes of constituents, linear configurations of these classes, classes of such configurations, etc. It is significant that in Harris 1951, which is the most consistent elaboration of these principles, the term 'function' does not even appear. For Bloomfield's view on this matter, we have the following irnportant explicit testimony: 'To earlier students, language appeared to have a third aspect, intermediate between form and meaning; this aspect was usually called function. Thus, a word like apple not only meant a certain kind of fruit, but also functioned as a noun, serving as the subject of verbs, as the object of prepositions and transitive verbs, and so on. Careful study, however, showed that features like these are a part of the f orn1; they are the forrnal features which come into being when two or more forms are combined in a larger form ... A forrn's privilege of occurring in any one position is a function of that form, and all its various functions together make up its function. In sum, the function of a speech for1n consists merely of for1nal features which appear as it serves as part of a more inclusive f orrn.' 1 (1943: 103-4). In Bloomfield 1926: 159, there was still some confusion about the exact status offunctions. Thus, he states: 'The positions in which a form occurs are its functions. Thus, the word John and the phrase the man have the functions of 'actor', 'goal', 'predicate noun', 'goal of preposition', and so on.' But a few lines above, these same labels had been defined not as functions ( = positions), but as functional meanings ( meanings of positions). As is evident from the quotation given above, however, Bloomfield's basic idea was that function can be reduced to formal features of position. And he further held that with each such position there corresponded a unique feature of 'positional meaning'. Thus, a term like 'actor' both refers to a formal featu re of the construction in which a certain ele1

Cf. my comments on this interpretation of 'function' in chapter 3 above.

8.1. BLOOMFIELD'S VIEW OF GRAMMATICAL FUNCTION

145

ment appears, and to the feature of meaning which is presumed to correspond to this feature of forrn. In this way, part of what I regard as functions and functional relations has been taken care of in Bloomfieldian linguistics in ter1ns of constructions over against constructional and functional, or positional meaning. These notions were introduced by Bloomfield in 1926: 157-60, where a construction is defined as a sequence of positions having a certain constructional meaning, the components of which (i.e., the meanings of each 2 position separately) are called 'functional meanings'. According to this view, the two sentences: {I) Richard saw John., and (2) The man is beating the dog.

both 'show the construction of free form plus free f orn1 plus free form meaning 'actor acting on goal'.' (1926: 158). This construction has three positions, the first having the functional meaning 'actor', the second 'action', and the third 'goal' (ibid. 159). Apart from the general difficulties arising from the definitions of the notions 'construction', 'position', and 'function', as treated in chapter 3, I have two main objections against this interpretation of functional notions: (i) its restriction to a certain order of positions makes certain generalizations impossible. E.g., the sentence:

(3) It was John whom Richard saw. contains, in some way, the same functional relations as (I). But this similarity cannot be accounted for if functions are tied to ordered positions within constructions. The same applies to the foil owing Latin sentences: (4) a. b. c. d.

Pater amat filium. Pater filium amat. Filium pater amat. Filium amat pater.

----1

The latter tern1 was replaced by 'positional meaning' in Bloomfield 1933: 267.

146

ON THE NOTION 'GRAMMATICAL FUNCTION'

e. Amat pater filium. f. Amat filium pater. 'The father loves his son'. where the ordering of the constituents has not the slightest influence on the functional relations within the construction, which remain invariant. (ii) the use of the term 'functional meaning' in this context suggests a one-to-one relation between forrnal and semantic features which in fact it is quite hard to assess. It is notoriously difficult to connect any real semantic content with Bloomfield's labels 'actor', 'action', 'goal', etc. (cf. Chomsky 1957a: 100). I cannot agree with Matthews (1967: 133) that for a universal definition of such functional relations as, e.g., subjectverb in ter1ns of such 'notional' categories as 'actor-action' it would not be required that this notional aspect should accompany the functional relation in each particular case. If there is no real one-to-one relationship between formal and semantic features, then neither of them can be used to define the other. What we see in Bloomfield's theory is that his 'functional meanings' are hardly anything more than abbreviations for the formal features in ter1ns of which he defines his functions; and these for1nal features are, as we have seen, themselves of doubtful relevance to the notion of gra1n1natical function. By this I do not wish to suggest that grammatical functions have nothing to do with semantics. But in my opinion they are not themselves semantic notions; they are features of grammatical organization which determine, on the one hand, part of the formal structure of linguistic expressions, and, on the other hand, part of the semantic content of these expressions. While in Bloomfield's system it seems that the notion of construction is completely reducible to order of elements and meaning, Hockett has shown convincingly that at least in some cases it should be regarded as a primitive of grammatical description (1954: 221-2). This is true since two elements having the same phonemic form and the same lexical meaning, and occurring in the same order, may still give rise to different total contents according to whether they have one functional relationship or 3 4 another. Hockett's example is Chinese chau 'to fry' and fan 'rice', which 4 3 combine into chau fan , an expression which can either mean 'to fry rice', or 'fried rice'. In the first case, the construction is 'verb-object', in the second case it is 'attribute-head.' In a later article Hockett goes a long way in recognizing 'constructions' or constructional patterns as distinct

8.2. FUNCTION IN TRANSFORMATIONAL GRAMMAR

147

from 'constituents' (1961: 228), though afterwards he tries to reinterpret 3 the 'constructional markers' themselves as constituents (ibid. 231). To the extent that constructions can be shown to have primitive status over against constituents and order of constituents, these considerations provide arguments in favour of the irreducible character of what I prefer to call 'gran1matical functions'.

8 .2. Grammatical function in transformational generative grammar A different conception of gra1nmatical functions has been advanced in transformational generative grammar. According to this view (most clearly treated in Chomsky 1964: 60-5, Katz & Postal 1964: 33-9, 159, Chomsky 1965: 68-74, 106) grammatical functions like 'subject', 'predicate', 'main verb', 'object', etc. are 'inherently relational notions' (Chomsky 1965: 68), i.e., they merely denote certain relationships or configura- · tions of categories appearing within the same linguistic structure. Thus, a statement like 'John is the subject of the sentence John is reading a book' would mean no more than 'The lexical item John, which is an NP, is directly dominated by S, which dominates the whole sentence John is 4 reading a book'. The definition of the function 'subject-of' is, accordingly, given as follows (Chomsky 1965: 71): (5)

Subject-of: [NP, S].

If this were possible for all grammatical functions, there would, again, be no need to specify these functions separately in the structural description, since all relevant information would be contained already in the categorial representation (the P-marker) of a sentence. Any addition of functional indications to this categorial representation would be redundant. This is precisely the position taken by the writers referred to, who suggest that all grammatical functions can be generally defined in terms of configurations of categories, denouncing the separate indication of these functions in the structural description as superfluous. 5 To this view of grammatical functions I have the foil owing objections: 1

For a more detailed discussion of this point, see Kooij (forthcoming). ' For the relations 'X is a Y' and 'X dominates Y' cf. chapter 5, fn. 4. 1 For similar criticism, see Matthews 1967: 132-5. Some objections are raised also in Fillmore 1966: 19-20.

148

ON THE NOTION 'GRAMMATICAL FUNCfION'

(i) In my opinion, it does not correctly represent the traditional use of the notion 'grammatical function.' When in traditional gram111ar John is called the subject of John is reading a book, this statement not merely explicitates the particular relation obtaining between John and the sentence as a whole, but it for1nulates the 'part' or 'role' played by the item John in this sentence. It is clearly different from the assignment of categories to John ('John is a proper name, a noun'), but it is also different from the mere specification of a relationship between categories. The ter•n 'subject' refers to the value which John has in this sentence-pattern, a value different from those which it has in: (6) a. I see John.

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

I give John a book. This is John. My name is John. How are you, John? John! Don't do that! I received a letter from John.

On the other hand, John has the same value in each of the following sentences: (7) a. John is reading a book. b. Is John reading a book? c. I saw that John was reading a book. d. The book which John was reading was bad. e. When John is reading a book, he always falls asleep. A grammatical function is a plus-value acquired by a constituent when used in a certain pattern. Functional relations are not relations between categories, but grammatical relations are relations between (irreducible) functions. A comparison may perhaps clarify this point. The lexical and grammatical items figuring in a linguistic structure are like the actors in a play. The actors as such are individuals belonging to certain classes (male-female, young-old, good-bad, etc.), irrespective of the play which they are called on to perform. When performing, however, they each represent a certain part, or have a certain function within the total structure of the play. The parts of the play cannot be described in terms of relations

8.2. FUNCfION IN TRANSFORMATIONAL GRAMMAR

149

between the actors, but are rather dictated by the pattern of the play as such. Each of the parts of the play is independent of the actors performing it, though it is usually restricted to certain classes of actors. Many different actors can play the same part, and the total structure of the play is not changed if a completely different cast is selected for its perforn1ance. In just the same way the functional pattern of a linguistic expression is to a great extent independent of the lexical or grammatical items figuring in it. In a linguistic description, therefore, not only the latter must be described in their various classifications, but also the functional patterns which, in various ways, they can represent. (ii) The configurational approach relies heavily, of course, on the transformational view of language structure. The general definitions of grammatical functions are only applicable to underlying P-markers (i.e., to deep structures underlying what are, essentially, systems of simple subject-predicate sentences) and not to derived P-markers or'surface structures'. In Katz & Postal this is made a matter of principle. They simply hold that all grammatical relations can be traced to underlying Pmarkers (1964: 33). Chomsky is somewhat less positive on this point. He does not deny the existence of significant functional relations in 'surface' structures (1964: 65, 1965: 220 fn. 30, 163), though, apart from same stray remarks, he does not indicate how these could be fitted into the configurational approach. The really significant relations, therefore, are defined for the 'deep structure', and the discussion is mainly concerned 6 with these. In this way the traditional distinction between 'logical' subject, object, etc. on the one hand, and 'gra1n1natical' subject, object etc. on the other is revived. It is precisely the 'logical' relations which can be expressed in ter1ns of deep structure in transfor1national grammar, whereas 'grammatical' functions could be approached only in tern1s of surface structure. The restriction of functions to deep structure configurations amounts, 7 therefore, to an over-emphasis of these 'logical' relations, and to the • Cf. Chomsky 1965: 220 fn. 30: 'Observe that the definitions of grammatical relation or grammatical function that have been suggested here refer only to the base of the syntax and not to surface structures of actual sentences in other than the simplest cases. The significant grammatical relations of an actual sentence ... are those which are defined in the basis (deep structure) of this sentence.' For an attempt to extend the definitions of functional relations to surface structures, cf. ibid. 220 fn. 32. This passage, however, which I cannot here quote in full, will be seen to be quite inconclusive. 7 Katz & Postal (1964: 33) attempt to demonstrate the validity of these relations by a

150

ON THE NOTION 'GRAMMATICAL FUNCTION'

neglect of the 'grammatical' relations. This is in marked contrast to the attitude of many linguists who, while doubting the linguistic status of the 'logical' relations, have always recognized at least the 'grammatical' one as important factors in the structural constitution of linguistic expressions. The restriction to deep structure implies that for a sentence like (8) John was persuaded by Bill to leave. the configurational definitions of grammatical functions can, as far as they go, account for the fact that John is the 'logical' object of persuade (to leave) and the 'logical' subject of to leave, while Bill is the 'logical' 8 subject of the sentence. But they cannot account for the functions fulfilled by the elements in question in the actual sentence. E.g., they cannot express the fact that John is, in (8), the grammatical subject of the sentence, etc. This means that according to these definitions there would be no functional difference between (8) and: (9)

Bill persuaded John that John should leave.

Even for those who accept the difference between 'grammatical' and 'logical' subject, object, etc., and agree with Chomsky on the semantic relevance of the latter, this conclusion seems hardly acceptable. We must conclude that in the transformational approach only part of the functional relations can be accounted for. This theory, then, if it were correct, would in any case not be complete. The incongruities involved by this state of affairs come to light quite vividly in certain examples given by Katz & Postal (1964: 35-6). Thus, given the sentence (10) The picture was painted by a new technique. their method allows for the assignment of the subject-function to a 'someone' presumed to be present in the deep structure of the sentence, familiar, but in the absence of any factual proof quite vacuous appeal to 'any speaker of English'. When confronted with the two sentences (i) John drank the milk, and (ii) The milk was drunk by John, they say, 'It is evident to any speaker of Ea . glish that in both (i) and (ii) the relation of both John and the milk to the verb drank/drunk is the same, i.e., in each case John is the 'subject' of this verb while the milk is the 'object'.' 8 See, on this matter, Chomsky 1965: 70.

8.2. FUNCTION IN TRANSFORMATIONAL GRAMMAR

151

but it cannot account for the fact that in the actual sentence the picture is the grammatical subject. In other words, it leads to the selection of a 'subject' for a sentence in which no indication of any 'agent' is given (one can hardly maintain that it is evident to any speaker of English that someone is the 'subject' of (10)), while it does not account for the function of the picture in this sentence, a function which has a direct impact on its further linguistic structure (in determining, e.g., the particular form 9 of the foil owing auxiliary verb). Restricting the grammatical relations to deep structures is simply equivalent to restricting them to a quite si1nple subset of all the actual sentences of a language (i.e., to simple subject-predicate sentences). It is not surprising that under this severe restriction general definitions become possible. If there is the slightest doubt, however, that certain complex structures cannot be completely and adequately described on the basis of these sirnple structures (i.e., if the reduction postulate is inadequate, as we have seen it is), then these definit4ons lose their attractiveness and take on the character of unwarranted over-si1nplifications.

(iii) It is to be observed that the configurational definitions of grammatical functions put a severe restriction on the possible form of underlying P-markers. If, e.g., the function 'subject-of' is defined as in (5), then it is clear that for this definition to be unambiguous there may be no more than one NP directly dominated by S in any base P-marker. This applies, mutatis mutandis, to all functional definitions. The restriction can be generalized, therefore, as follows: if in a base P-marker there is a category c 1 which is directly dominated by a category c,_ 1, then there may be no other category of the same type as c, which is also directly dominated by C1-1·

10

It is not unthinkable that a transformational grammar could be organized in accordance with this restriction (which is explicitly mentioned in Chomsky 1965: 211 fn. 7, 220 fn. 32). But there seem to be no independent arguments in favour of its inclusion, and the following considerations, advanced by Matthews (1967: 135) throw doubt on its feasibility. Within this framework, Matthews notes, 'no language could exhibit a ternary analysis into Subject, 'Verb', and Object or a similar analysis, •

10

For similar criticism, see Matthews 1967: 134. Cf. also Hudson 1967: 232.

I

152

ON THE NOTION 'GRAMMATICAL FUNCTION'

.

within the predicate, into Verb, Direct Object and Indirect Object. These are req11irements for which ... I can think of no independent justification.' This objection might be met by adding to the definitions of functions a condition on the order of the categories involved (cf. Chomsky 1965: 220 fn. 32, Hudson 1967: 232 fn. 3). The function 'subject-of', e.g., could then be defined in terms of 'the first NP directly dominated by S in the base P-marker'. In that case, however, these definitions are made to depend on the hypothesis of a universal order of constituents in base P-markers, which itself is a matter open to debate (Chomsky 1965: 123ff, Staal 1967). (iv) This leads over to a more general point. The configurational definitions of gram1natical functions are meant to be universals of linguistic theory. I.e., they are thought to be applicable to the deep structure of any language (Katz & Postal 1964: 159, Chomsky 1965: 72-3, 1967: 418-9). But far from being based on an empirical investigation of a wide variety of languages, this simply amounts to a projection of both the hierarchical structure and the ordering of constituents devised for a transfor1national grammar of English into the linguistic systems of all other languages. The hypothetical introduction of linguistic universals is a perfectly respectable matter. But there should be some balance between the empirical investigation on which they are based, and the claims they make about hitherto uninvestigated facts. For further critical remarks on these somewhat premature generalizations, see Lyons 1966: 125 and Matthews

1967: 134-5. (v) It is interesting to notice, further, that notwithstanding the presumed reducible character of grammatical functions, some transforrnational grammarians have found it necessary to introduce functional labels into their descriptions, i.e., to posit 'nodes' within constituent structure descriptions which, though formally not differentiated from categories, still carry the names of traditionally recognized functions such as subject, object, etc. Thus, Lees introduces an 'object morpheme' (1960: 46), just like Bierwisch (1963: 40), and Hartung adduces several arguments for recognizing different 'functional' nodes (1964: 20 4). Further, ter1ns like 'predicate' or 'predicate phrase' are used almost universally, and in a way which often gives rise to the question whether strictly categorial or functional notions are involved (cf., on this point, Jacobson 1966: 116). Relevant here, too, is the remark which Katz & Postal append to their

8.2. FUNCI'ION IN TRANSFORMATIONAL GRAMMAR

153

discussion of examples such as those cited above: 'It is of course always possible to provide some kind of similarity in an ad hoe way. For example, one could arbitrarily recog11ize a node Subject dominating the appropriate elements in contrast to a node Object. But this is of no explanatory value, for no independent grammatical motivations can be found for introducing such nodes into the syntactic description.' (1964: 68 fn. 3). As I shall try to show, there are quite a few 'independent grammatical motivations' for introducing functional notions as such into the grarnmar, though these may indeed not be equated to categorial nodes. If these functions are introduced and if the resulting system is elaborated in a systematic manner, there will be nothing arbitrary or ad hoe in this whole procedure. Finally, I may refer to a suggestion of Lyons' (1966: 121). Discussing the possibility of coordinating various types of clauses, he says: 'Traditional grammar, with all its faults and inconsistencies, handled the problem by setting up clauses [read: classes, SD] of 'functionally' different sentence-types (conditional, causal, temporal, etc.) and distinguishing 'formally' different members of each class. It may be that the generative model of syntax should allow for the occurrence of such 'functional' markers, in the deep structure of complex sentences, as 'conditional', 'causal', etc., and for the optional development of these markers in 'for1nally' different ways.' 'I'his is precisely what I shall try to do in a general and systematic manner, but within a non-transformational framework. (vi) The preceding points constitute i1nmanent criticisms within the transformational theory of language structure. If, however, one does not accept this theory in the first place, then the definition of grammatical functions in terms of configurations of categories becomes utterly impossible. We have seen that the definitions proposed refer solely to deep structures, and I believe, in spite of Chomsky's suggestions in this direction (1965: 220 fn. 32) that no non-ad hoe extension of this technique to surface structures is in any way possible. Within a non-transformational framework of language description, however, the notion of 'deep structure' is out of place. The configurational technique, therefore, cannot be applied in any way within such a framework. In other words: within the full set of the linguistic structures of any one language, there does not seem to be a one-to-one correspondence between grammatical functions and configurations of categories. The reduction of the former to the latter is, consequently, excluded in principle.

154

ON THE NOTION 'GRAMMATICAL FUNCflON'

8 .3. Preliminary conclusions From the preceding discussion it follows (I) that grammatical functions cannot be reduced to position or distribution (if taken in the usual sense of place or places within the linear sequence of linguistic structures), and (2) that they cannot be equated with 'relations between categories', since (i) same relations between categories may correspond to different functional relations, and (ii) same functional relations may hold for cate11 gories appearing in different configurations. The conclusion must be that grammatical functions are irreducible aspects of grammatical structure, which can be partly correlated to formal features (special morphemes, words, significant order of elements), but 12 cannot possibly be completely reduced to these. The difficulty experienced by many linguists in accounting for the notion of 'grammatical function' lies in the very fact that it is not always possible to identify it unambiguously with some feature ofform or with a complex of such features, while on the other hand gra1n1natical functions seem to be quite crucial to an understanding of the way in which linguistic structures finally lead to correct interpretations. If, however, we accept both categories and functions as primitives of linguistic theory, and if we require that for each constituent both the category to which it belongs and the function which it fulfils is indicated in the description, we may arrive at a system which, while more systematic and more generally applicable than traditional grammar, is also decidedly more flexible and revealing than modern distributional theories, and, in my opinion, more adequate than the theory of transformational grammar. In chapter 9 we shall see how such a system could be substantiated.

8. 4. Grammatical function in tagmemics It is the great merit of Pike's theory of tagmemics that by the side of grammatical categories it recognizes grammatical functions as irreducible factors, the basic term tagmeme being used for the correlation of a func13 tion (or 'slot') and the class of items by which it can be fulfilled ('fillers'). 11

Cf. Longacre 1965: 67. 12 Heuristically these formal correspondences are of course very important. But a complete definition of functions in terms offorms appears to be utterly impossible. 13 See Pike 19672 : chapter 7 and pp. 490-2. On the relation of tagmemics to transfor-

8.4. GRAMMATICAL FUNCfION IN TAGMEMICS

155

It is important to notice that the ter111 'slot', though suggesting depend2 ence on linear arrangement, is by no means meant as such (cf. Pike 1967 : 220, Elson & Pickett 1962: 57, Longacre 1964b: 15). I may repeat here Longacre's commendable definition of grammatical function quoted above (section 4.1.2.): 'By function is meant the particular office or role of one distinguishable part of a construction type in relation to other parts of the same construction.' (1965: 65). Since the tagmeme comprises a gra1n111atical function and a class of elements fulfilling this function, it is by its very nature related to the other tagmemes of the same construction: the tagmeme is an inherently relational entity. Following the most recent ter1ninology we can say that a tagmeme is necessarily part of a syntag2 meme, i.e., of a unified sequence or pattern of tagmemes (cf. Pike 1967 : 450--1, Longacre l 964b: 15, 1967: 323-4). Transformationalists have voiced two main objections to the notion of the tagmeme: (i) in their eyes it does not properly differentiate between gratrunatical function and grammatical category; (ii) it makesforpointless 14 redundancy in the grammar. As to (i), this objection is based on a misinterpretation of the intentions 15 of tagmemics. Whereas tagmemicists have always emphasized that the tagmeme is a complex entity consisting of a correlation of a function or slot and a class of fillers, both functions and filler-classes are interpreted as categories by Chomsky and Postal. Thus, Chomsky denounces the tagmemic notion of function as inadequate 'since it is a strictly categorial interpretation' (1964: 61). In his later work, he warns us that 'Functional notions like ''Subject'', ''Predicate'' are to be sharply distinguished from categorial notions such as ''Noun Phrase'', ''Verb''.' and points out 'The fundamental error of regarding functional notions as categorial.' (1965: 68-9). This is no doubt correct. But I know of no place in the literature on tagmemics where 'functions' or 'slots' are treated as categories. Indeed, the identifimational grammar, ibid. 495-506. Cf. also Pike 1958, especially p. 276 on the relation between a categorial, a functional, and a functional-categorial approach. A good short introduction to tagmemic theory is Elson & Pickett 1962. See especially 2-3 and 57ff. on the tagmeme. Cf. also Longacre 1964b: 15-6, and his short but quite illuminating survey 1965. Several points are also clarified in Cook 1964. 1' For these points, see Chomsky 1964: 61, Postal 1964a: 33-43, 1966: 94ff., Bach 1964: 42. 16 Cf. Pike 19672 : 496, Longacre 1964b: 15 fn. 11, 1965: 66-7 and fn. 5, 1967: 323-5. See now also Ney 1967: 43-5.

156

ON THE NOTION 'GRAMMATICAL FUNCl'ION'

cation of functions and categories is a mistake which tagmeanics has avoided, by carefully distinguishing the two components within the tagmeme. Postal's criticism of tagmemic functions is based on the same line of thought as Chomsky's, though it is more elaborate. While a tagmemic formula like: (11)

A-B:b+C:c

means: 'the element A consists of the categories b and c having the functions B and C, respectively', Postal reads it as if it asserted: 'The element A consists of the categories B and C; B consists of the category b; C consists of the category c', and then reconstructs this analysis so as to yield a structural description like: (12)

A B

c

b

· c

(Postal 1964a: 33, 1966: 94), where B, C and b, care regarded as labels for categories. This criticism has been answered by Longacre {l964b: 15 fn. 11), who suggests that the tagmeme, as a complex entity, should rather be represented by complex symbols, so as to yield structural descriptions like: (13)

A B:b

C:c

a suggestion which, according to Postal, does not change the inacceptability of the tagmemic approach. In point of fact, however, the difference in representation is immaterial to the issue. The basic fact is that the symbols occupying the nodes in a diagram like (12) do not all stand for categories. I.e., the symbols B, C

8.4. GRAMMATICAL FUNCl'ION IN TAGMEMICS

157

etc. and b, c etc. are not chosen from the sa 1ne set of labels, and their fundamental difference resides in the fact that symbols like B, C etc. necessarily belong to sequences of sirnilar symbols, whereas b, c etc. never belong to such sequences. Instead of confusing functions and categories, this representation faithfully represents the inherently relational character of grarnrnatical functions over against the inherently non-relational character of categories. In terms of rules this means that function-symbols are always introduced by branching rules (one-many rules), while categories are always introduced by non-branching rules (one-one rules). Interpreted in this way tagmemics exactly for1nalizes the difference which Postal emphatically claims it confuses (1966: 95). As to the charge of redundancy (objection (ii) above), this is again less evident than it seems to be. The separate indication of functions in structural descriptions would be redundant if transformationalists were right (i) in distinguishing between surface structure and deep structure and (ii) in their claim that all significant functions can be defined in terms of relations between categories in deep structures. In section 8. 2. we have seen that the latter point has by no means been established and is fully dependent on the first, which itself is subject to serious doubts and criticism. If the premisses (i) and (ii) should prove to be true after all, then the separate indication of functions would indeed be redundant. But within a system which does not accept these premisses, the introduction of func16 tions is not redundant, but inevitable. Though I recognize the basic correctness of tagmemic theory in the fundamental place which it assigns to grammatical functions, I do not simply subscribe to the theory as a whole, since I disagree with some of its other tenets. One of the reasons of my disagreement is the notion of the tagmeme itself, which in Pike's approach is represented as a unit consisting of a function-class correlation. The combination of a function and a class 16

For an interesting comparison of the relative merits of the tagmemic and the transf orn1ational approach to grammatical functions, see Seuren 1966: 206ff. Seuren points out that 'Perhaps, the charge of redundancy is not well founded.' (210). He goes on to show that even a transformational grammar might profit from the direct and explicit introduction of grammatical functions into the base component (a point which had also been made by Hartung 1964: 20 4). I shall go a step further and develop the assumption that a correctly specified introduction of grammatical functions may lead to the abandonment of transformational rules.

158

ON THE NOTION 'GRAMMATICAL FUNCTION'

does not, in my opinion, constitute a 'unit' in any of the natural uses of the latter term. Functions are properties of the grarnmatical patterns of a language, while classes or categories constitute the basic groups of elements which can enter into these patterns. Functions and categories are independent parameters which in any construction can contract a multitude of different mutual relations. A linguistic pattern is, in my opinion, not constituted by a sequence of tagmemes (in the sense of: slot-class correlations), but by a hierarchy of sequences of functions which can be 'filled in' in different ways by certain categories of linguistic elements. It is to be noticed, however, that the manner in which Longacre defines and handles the tagmeme (1964b: 15) comes very close to this point of • view. A second point on which I disagree with tagmemics is the idea that the principle of IC-analysis should be incompatible with a functional description (see especially Longacre 1960). In fact, as soon as it has been recognized that a linguistic structure is not a mere linear sequence of smallest gra1n1natically relevant elements, but is characterized by a certain hier17 archical build-up, we have already accepted the principle of IC-analysis. That the principle of hierarchical structuring is not denied in tagmemics can be seen, e.g., from Elson & Pickett 1962: 58, 83ff., Longacre 1964b: 16-7. In more recent work, Longacre duly recognizes the 'layering or grouping tendencies' within gra1nrnatical structures (1967: 324), but he goes on to say that 'tagmemics holds that much is to be gained by treating this layering as of secondary importance and emphasizing rather the mutual relation of all elements of the clause.' (ibid.). I do not agree that layering is of secondary i111portance. It is true that the mutual relation of all elements within the linguistic structure must be described. But in my opinion this is only possible within the hierarchical structure in which they have their place. On the other hand, Longacre's 'string constituent analysis' is, in fact, one particular version of IC-theory, though the principles governing the choice of I Cs are different from those established by Wells, Bloch, Hockett, and other followers of Bloomfield. The main difference is that Longacre has rightly dropped the unmotivated preference for 'binary' divisions prevalent in the work of the latter (cf. Longacre l 964b: 16). In other words: if IC-analysis is taken in the basic sense in which it was 11

The choice is not between 'ICs or no ICs', but between 'few ICs or many ICs' (Hudson 1967: 225). Hudson gives a good survey of the pros and cons of each of these alternatives.

8.5. FUNCTION IN HALLIDAY'S 'SYSTEMIC GRAMMAR'

159

originally introduced by Bloomfield (1933: 161), viz. as the characterization of the hierarchical constituent-structure of any particular linguistic expression, then it is not only unobjectionable, but even indispensable. One may criticize the particular form which this theory has taken in postBloomfieldian linguistics (cf. below, section 11 . 2. I . ). One may also object to those who hold that IC-analysis is the only thing a grammatical description consists of. But it seems impossible to reject the principle of ICanalysis as such, which is dictated by the very nature of the facts to be accounted for. If it is true that the hierarchical principle is indispensable, then a linguistic expression should not be analysed into strings of tagmemes of different levels scattered over different parts of the gra1nmar, but as a hierarchical whole (representable in a tree-diagra1n) in which on each particular level functions and categories are specified independently. In this respect the 'formulae' of tagmemics are, in my opinion, not capable of revealing the real structure of the linguistic expressions of a language. A final point is that though the nature of the structural descriptions defined by tagme1nics is fairly clear, it is not so evident by what kind of rules these structural descriptions could be generated. Too little attention has been paid to the conception of a grammar as a system of rules rather than as a device for the partial description and generalization of observed regularities. In chapter 9 I shall develop a system of gramn1atical description in which these various points have been incorporated. To a very great extent, however, this system can be viewed as a particular reconstruction oftagmemic notions.

8. 5. Gram1natical function in Halliday' s 'systemic grammar' In Halliday's 'Categories of the theory of gra1n1nar' (1961), terms like 'subject', 'predicator', 'complement', 'adjunct', 'head', 'modifier', 'qualifier', are introduced as 'elements of structure' (256-7). This leads to a functional approach similar to that of tagmemics, even though at that stage Halliday did not use the term function itself in this context. He even rejected the form-function dichotomy in a different passage of the same article (ibid. 261), but this is mainly a matter of terminology. If 'element of structure' is redefined as 'grammatical function', then clearly this sys-

160

ON THE NOTION 'GRAMMATICAL FUNCTION'

tem is to a great extent equivalent to the functional grammar which I am advocating here. This is brought out quite clearly in the later development of Halliday's theory. Cf. 1966a: 57: 'In the representation of syntagmatic relations in language, we may distinguish between a linear sequence of classes, such as 'adjective followed by noun', and a non-linear configuration of functions, such as 'modifier-head relation' or simply 'modification'.' In this same paper, the introduction of function-labels is advocated, though in a way slightly different from what will be suggested below (ibid. 58-9). Halliday would tend to divorce the functional specification from the structural description (in terms of classes), and distinguish between a 'systemic' and a 'structural' component (ibid. 61). The syste1nic component would then contain systems of (functional) features, hierarchically ordered with respect to each other. The question then arises how (i.e., by what kind of rules) these features are ultimately assigned to the relevant constituents in the structure. Though I would not dare to say that this problem is unsolvable, it seems to present great difficulties, and no workable suggestion has so far been advanced. In my own approach, I shall act on the assumption that the functions are necessarily properties of the categories in the structure and should, consequently, be written into it directly. I thus see a much more inti1nate relationship between functional relations and constituent structure than Halliday's approach would suggest. Further details will become apparent in the next chapter. For further impressions on the role of function in a systemic gra1nmar, see the recent work of some of Halliday's collaborators, notably Hudson 1967 and Huddleston & Uren, forthcoming. Hudson arrives at much the same conclusions regarding the introduction of grammatical functions into structural descriptions as those which I shall be advocating below (Hudson 1967: 233ff.). He agrees, however, with Chomsky in the view tl1at grammatical functions could be defined in terms of configurations of categories (ibid. 236-7). I would say, rather, that any search for a definition offunction in terms of (relations between) categories is futile, since functions are irreducible factors of grammatical structure.

8. 6. Some general remarks

Historically, the development of the notion 'grammatical function' in linguistic theory is interesting. Traditional grammar recognized something

8.6. SOME GENERAL REMARKS

161

of the kind, though its precise status between logic and linguistics was never clarified. Bloomfield tried to reduce it to form, and Harris dropped it altogether. The purely categorial view which resulted from this was adopted by transformational grammarians, who added transformational rules to make up for the deficiences of constituent structure grammar; at the same time they attempted to account for grammatical functions in terms of relations between categories. Pike and, in a different way, Halliday reverted to the traditional, irreducible notion of gram1natical 18 function, and accepted this as a primitive of grammatical theory. Though perhaps inviting a charge of over-simplification, we might say that in recent grarnmatical theory we can distinguish three main currents: (i) Distributional, categorial, or constituent structure grammar in the Bloomfield tradition; (ii) Transforrnational grammar; (iii) Functional grammar. The inadequacy of (i) seems to have been established beyond· doubt. Both (ii) and (iii) are attempts to overcome its limitations. The theory of functional grammar can be regarded as an adaptation of 'structuralism' in the light of a reconsideration of the possible merits of traditional grarnmar. The theory of transformational grammar, though incorporating one particular version of structuralism, often exhibits a decidedly hostile attitude towards 'structuralism' as such, and sometimes does not seem to notice that many points objected to are not so much inherent in 'structuralism' itself, but rather in a number of side-issues which have come to be associated with it. Moreover, it is not sufficiently realized that the addition of a functional perspective enriches structuralist theories in a manner which obviates many of the objections levelled against them. It will have become clear that I believe that the functional approach has the better chance to arrive at a really satisfying framework of grammatical description.

----18 Cf. Longacre 1965: 67ff.

9. Outline-sketch of a functional granlmar

In the preceding chapters I have paved the way for the development of a system of functional grammar. I shall now try to give a systematic account of this type of grarrunar, in so far as I am able to specify its properties at present. Since my view of this system stems mainly from the consideration of the problems encountered in the description of coordinative constructions, which constitute only a small subsection of the total grammatical system of any language, while on the other hand the theory proposed here is presented as having a much wider range of applicability, it will be clear that what is contained in this chapter cannot claim to be more than a preliminary attempt at a complete theory of gra1n1nar and that I shall not be able to elaborate this approach in every single detail. I shall try to achieve a maximum amount of specificity by discussing, first, the basic terms and notions essential to this theory, and, second, the different types of rules which will be needed in its irnplementation. In doing so I shall give most attention to those features in which functional gra1nmar differs from other theories, and take for granted those points which have been f or1nulated more or less clearly elsewhere. In this and the following chapters I shall then show what a description along the lines offunctional grammar would look like, both in general and, more particularly, in the case of coordinative constructions. It may be useful to repeat here what I have said earlier (section 5. 1.), viz. that in my view a linguistic description (and consequently a grarn1nar) does not describe what a particular speaker-hearer knows of his language, nor how he uses his language in speaking and interpreting, nor how he acquires his knowledge of the language. Nor does it describe the manner in which the linguist discovers the structure or system of the language. Rather, it describes the system of the language as such, in abstraction from how it is known or used, thus providing an answer to the question: what is the structure of the linguistic expressions of this language? A linguistic description is the product of systematic observation and analysis and there is no reason to suppose that a native speaker 'knows' this product in any sense. Of course, there must be a significant relation between a complete linguistic description and the competence of a native

9.1. BASIC PROPERTIES

163

speaker, but the nature of this relation requires separate (psycholinguistic) investigation. As to the general character of functional grammar as conceived here I have already emphasized that it borrows much of the conceptual framework of Pike's tagme1nics (above section 8. 4.). On the other hand, an extensive use is made of the formalisms developed in Chomsky's theory of generative grammar. The result of this combination is a theory which also reflects, in a systematic manner, the general notions contained in traditional grammar.

9. 1. Basic properties The basic (negative) property of a functional gra111mar of the type I propose is that it does not include transfor1national rules. It is a non-transf or1national theory about the grammatical structure of complex linguistic expressions. It has, therefore, the fundamental characteristic of describing any linguistic expression in ter1ns of its own internal grammatical propperties. No such expression will be treated as a derivative of other struc1 tures of whatever kind. This can also be formulated as follows: though paradigmatic relations between different linguistic expressions are i1nportant to this as to any other type of gra1n1natical theory, these relations will not be built into the description of the expressions as such, as is the case in a transforn1ational grarn1nar (cf. above, section 5. 1.). Thus, an active and a corresponding passive sentence are each described in terms of their own internal structure. This does not mean, however, that this structure is something superficial or unrevealing. Indeed, the addition of a functional perspective enables us to account for many quite subtle differences which cannot be expressed in a purely categorial constituent structure system. There is no room, in this approach, for a difference between 'surface structure' and 'deep structure' as conceived in transforn1ational gram1nar. Any linguistic expression is supposed to have only one gramrnatical structure in which, however, all its relevant grammatical features are specified. As has been clarified in chapter 5, a non-transformational grammar must be powerful enough to overcome the limitations of ('restricted') constit1

Cf. Pike's remarks quoted above in section 5. 5. 2., and Longacre 1964b: 22.

164

OUTLINE-SKETCH OF A FUNCTIONAL GRAMMAR

uent structure grammar as formalized in the theory of transformational grammar. This is achieved by three basic extensions of such constituent structure grammars: (i) Functional gramrnar will contain rules for describing infinite expansions of same-level members; (ii) It will contain the possibility of specifying discontinuous constituents; (iii) For any constituent appearing within a linguistic structure it will not only indicate the category to which this constituent belongs, but also specify the grammatical function which it fulfils in relation to other constituents of the same structure.

9. 2. General lay-out of structural descriptions 9. 2 .1. Facts accounted for

A functional grammar will assign to each linguistic expression a structural description in which at least the following features are specified: (i) The smallest grammatically relevant elements of which it is built up (its ultimate constituents); (ii) The hierarchical structure formed by the different layers of nonultimate constituents or 'groupings' of ultimate constituents. In other words, for any non-ultimate constituent, the constituents of which it is directly built up will be specified (its immediate constituents); (iii) For any non-ultimate constituent, whether it is continuous or discon• ttnuous; (iv) For any constituent (ultimate or non-ultimate), the (sub-)categories to which it belongs; (v) For any constituent, the grammatical function which it fulfils in the linguistic structure in question and the functional relations which it contracts with other constituents on account of its grammatical function.

9. 2. 2. Most inclusive units described

A description along the lines of a functional grammar will not restrict itself to the sentence as the maximum unit described. One need only con-

9.2. LAY-OUT OF STRUCTURAL DESCRIPTIONS

165

sider a few examples to see that many 'sentences' are grarnrnatically connected with preceding or foil owing sentences in different ways:

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

I do not like it at all. But my wife is very pleased with it. I shall not be in to-night. The refore you cannot come. John will not be in to-night. He is going to the theater. How many apples would you like to have? Two please. Who killed Lincoln? Booth.

In (I), we have the construction M 1 but M 2 , but in this case this construction is divided between two successive sentences. In (2), the refore refers to the content of the preceding sentence: its occurrence in this position is conditioned by the preceding sentence. In (3), the occurrence of he is si1nilarly conditioned by the preceding John. In (4), two can be interpreted as 'two apples' because of the preceding question. And in (5), Booth is perfectly clear as an answer because the preceding sentence has already provided the pattern into which it falls. We can label Booth in (5) as 'subject', even though there is no more linguistic material accompanying it in the same sentence. This is most easily accounted for by describing question+ answer as a unified grammatical pattern divided between two sentences (and, usually, between two speakers). I do not wish to suggest that the grarnmatically dependent sentences in (1)-(5) could not, under certain conditions, stand on their own. I only hold that if such sentence combinations as (1)-(5) follow each other directly in this order, each foil owing sentence has a grammatical relationship with the preceding one. These relationships are not accounted for if each of these sentences is only described as an independent unit. Let us approach this matter from another point of view. In his La11guage, Sapir retained the old logical definition of the sentence as 'the linguistic expression of a proposition. It combines a subject of discourse with a statement in regard of this subject.' (1921: 36). Now, if we start from what we usually regard as the 'sentences' of a language, there are two problems connected with this definition: (i) there are sentences which do not have both a subject and a predicate (such as Booth in (5) above), (ii) there are sentences which have more than one subject-predicate combination (as in all so-called complex and compound sentences). The latter point is especially evident in the case of coordinations with clauses as members, as in the example which Sapir adduces himself (ibid. 37 fn. 7): (6)

I shall remain but you may go.

166

OUTLINE-SKETCH OF A FUNCfIONAL GRAMMAR

Sapir remarks that such coordinations 'may only doubtfully be considered as truly unified predications, as true sentences. They are sentences in a stylistic sense rather than from the strictly for1nal linguistic standpoint.' (ibid.). But, if (6) is not a sentence, how do we explain the difference between (6) and: (7)

I shall remain. But you may go.

According to the logical definition there would be no difference between (6) and (7) in sentencehood. Nor is there a difference in grarnmatical construction. The difference between (6) and (7) could then only be described as a rather superficial intonation-feature. A much more satisfactory description is achieved if we define the sentence linguistically in ter1ns of unified intonation-patterns (cf. Reichling 3 1965 : 26-7). Then (6) is one sentence, and (7) is two sentences. But since the grammatical construction is the same in both cases, we are forced to conclude that in (7) two sentences are combined in one gramrnatical construction. . From such examples as these we see that the sentence cannot be defined as 'an independent linguistic for1n, not included by virtue of any gramrnatical construction in any larger linguistic form.' (Bloomfield 1933: 170, foil owing Meillet). There may be quite essential grammatical dependences between sentences foil owing each other in connected discourse. A gra1nmar which only describes sentences, therefore, does not account for the 2 full range of grammatical phenomena. The maximum units to be described by the grammar are those linguistic expressions which are in no way grammatically dependent on their preceding or following linguistic context. Such gran11natically independent linguistic expressions may of course be semantically related to their con~ text, and they may also depend on extra-linguistic factors such as the situation in which they are used. Such dependences, though linguistically

1

Again, Pike is among those who have recognized the fact that sentences only constitute one of the levels (not the highest one) in the grammatical hierarchy. Cf. Pike 19671 : 145-8, 484-6, where a rich bibliographical survey may be found. For a recent practical application of Pike's views on 'units higher than the sentence' see Wheeler 1967 where further references are given.

9.2. LAY-OUT OF STRUCTURAL DESCRIPTIONS

167

quite relevant, are not grammatical dependences. Consequently, they need not be accounted for in the gra1n1nar. We can, for these reasons, take the 'independent linguistic expression' (ile) as a starting-point for gra1nmatical description. By this ter1n, we can refer to the total set of objects to be described by the gra1n1nar. We can further say that any independent linguistic expression is either a sentence or a combination of sentences. A sentence may be declarative, interrogative, i1nperative, excla1native, etc. We need rules, therefore, by which the initial category 'independent linguistic expression' is subcategorized in • vanous ways. A very difficult problem which I shall not try to solve here is the question of how far the grammatical relations between sentences may extend through a running text, i.e., what is the maximum range of an independent linguistic expression. For our present purposes it is sufficient to have shown that there are facts which force us to go beyond the sentence in gra1n1natical description and that any approach to gra1n1natical description would be open to criticism from the very start if it did not include this possibility. I have also demonstrated that this is particularly relevant to coordinations, which often form grammatical 'bridges' over sentence-boundaries. In the system which I shall develop here these facts can in principle be accounted for, though I shall not elaborate their consequences in detail.

9 . 2. 3. Relations between different linguistic expressions

For any linguistic expression, the properties (i)-(v) given in section 9 . 2 . I. should be specified in such a way that a maxi1num degree of generality and coherence is achieved, so that it is i1n1nediately (though implicitly) apparent to what other linguistic expressions it is grammatically related and in what respects. In other words, the description must be such as to indicate which linguistic expressions are identical or similar in functional structure, in categorization, and in constituency. Each of the rules deter1nining these aspects of grammatical structure must be formulated in such a way that it accounts for as many different linguistic expressions as possible. It may be objected to the scheme presented here that such relations as exist, e.g., between active and passive constructions, between declaratives and interrogatives, etc. are not explicitly described (i.e., do not enter into the rules which determine the separate description of each of these con-

168

OUTLINE-SKETCH OF A FUNCTIONAL GRAMMAR

structions), while in a transformational grammar the transforrnations take care of these relations in a particularly elegant way. This is true; these relations between different linguistic expressions are only implicitly characterized in a functional grammar in so far as the same functions or functional relations, the same categories, subcategories, constituents, and lexical or grammatical elements appear in their structural descriptions. But the lack of an explicit description of such relations as these can in no way be counted against a functional description of this type. Indeed, why should the relations between different linguistic expressions be explicitly accounted for in the separate grammatical description of each of these linguistic expressions? Transformationalists argue for this on the ground that native speakers are presumed to 'know' these relations intuitively. But in the first place it is not clear to what extent this observation is based on empirical data. In the second place, if there is such an intuitive knowledge of relations, this must be a knowledge of differences and sirnilarities between different linguistic expressions, each having a structure of its own. In the third place, it seems that the implicit characterization of these differences and similarities provided for in a functional gra1n1nar is quite sufficient to account for any intuitive feeling or knowledge of such relations. I completely agree, in this respect, with the foil owing recent remark of Putnam's: 'It is contended that a gra1n1nar which 'defines' active and passive for1ns separately (this can be done by even a phrase-structure grammar) fails to represent something that every speaker knows, viz. that active and passive f orn1s are related. But why must every relation be mirrorred by syntax? Every 'speaker' of the canonical languages of mathematical logic is aware that each sentence (x) (Fx :::> Gx) is related to a sentence (x) ( Gx :::> Fx); yet the definition of 'well f or1ned formula' fails to 1nirror 'what every speaker knows' in this respect, and is not inadequate on that account.' (1967: 22, fn. 13). I believe indeed that these 'transformational' relations need not be explicitated in a grammar which purports to be a description of the gra1nmatical structure of independent linguistic expressions. This does not mean, however, that relations between linguistic expressions are not described at all. The system of rules constituting a functional grammar itself throws a network of relations over the linguistic expressions described. To what extent these relations are indeed relevant to the knowledge or the use of the language will be a matter of psycholinguistic investigation.

9.2. LAY-OUT OF STRUCTURAL DESCRIPTIONS

169

9. 2. 4. Constituents, categories, and functions

As a general rule, I am assuming here that any constituent of a linguistic expression is assigned to a category of constituents having equivalent gra rnrnatical properties. Of course, one can imagine constituents which are unique in their gra1nmatical properties. Such a constituent can be assigned to a category with only one member. Alternatively, and more economically, category assignment can be dropped in such cases, so that a grammatical function is specified directly for the constituent in question, and not for a category of constituents. Sirnilarly, I assume that any constituent or, equivalently, any category of constituents, has a grammatical function within the linguistic 3 expression in which it appears. It is here that the difference with traditional gram1nar comes to light most clearly. Indeed, traditional grammar has recognized some gran11natical functions, but it has not properly generalized the assign111ent of functions over the whole structural description. This means, among other things, that certain novel functionlabels will have to be introduced. The nomenclature chosen for these labels should not be regarded as too essential a matter. The primary aim of the recognition of functions is to explicitate grammatical differences between linguistic expressions which cannot be accounted for in terms of differences of constituency or categorization. It is the functional differences which count in the first place. The functions can, therefore, be indicated by any device which unambiguously attains the proper differentiation. For the same reason, we can retain such traditional terms as 'subject' and 'object' for grammatical functions, if only we take care to disconnect these from the countless philosophical and logical associations which they have acquired in past centuries, taking them instead as purely linguistic designations of grammatical functions.

1

For our present purposes I act on the assumption that just one function is assigned to any one constituent. Further elaboration of the theory of functional grammar might require us to recognize the possibility of assigning more than one function to a single constituent under certain conditions. Interesting suggestions in this direction have been advanced by Pike (1964). The point is also mentioned by Halliday (1966a: 64), and Huddleston & Uren (forthcoming) indeed provide for multiple function assignment to certain categories. The introduction of this possibility would require a corresponding revision of some of the rule-types which I shall describe below. But since it is not needed in the description of coordinations, I have not included this further complication from the very start.

170

OUTLINE-SKETCH OF A FUNCTIONAL GRAMMAR

9. 2. 5. The f orniat of structural descriptions

If all these considerations are systematically carried into practice, they will result in structural descriptions of a general for1nat which can be represented by the foil owing tree diagram: (8)

ile

F~

F, c,

u

u

u

u

F, C9

u

u

u

u

In this diagram, ile 'independent linguistic expression', c = 'category', F 'function', u - 'ultimate constituent'. From now on, I shall indicate functions by capitals, categories by small Roman type, and ulti1nate constituents by italics. Such a structural description as (8) should be interpreted as follows: the initial category ile is subcategorized as c 1 , and the latter subcategory is

9.2. LAY-OUT OF STRUCTURAL DESCRIPTIONS

171

subcategorized as c 2 • For this latter category, a sequence of functions · F 1 + F 2 + F 3 is selected, one function for each of its immediate constituents. The functional sequence F 1 + F 2 + F 3 is said to be a 'functional pattern' of the category c 2 • For each of the f1Jnctions constituting this pattern, a category is selected which fulfils the function in question. Each of these categories can be further subcategorized (thus, c 3 is subcategorized as c6 ), or receive a further functional pattern (thus, c4 gets the functional pattern F 4 + F 5 ). These procedures are repeated until we have only categories of ultimate constituents (in our example, the categories c 17, c 13 , c 14, c 7 , c 15 ,eg,c 16 , and c 11 ). For each of these 'ulti1nate categories', an appropriate ulti111ate constituent is selected. Functional gram1nar is thus a system comprising the foil owing set of • • • pn1n1t1ves: (1) an initial category ile; (2) the set of all non-initial categories {c 1 , c 2 , ••• , c,.} further divided into: (a) the set of initial subcategories (i.e. subcategories of ile selected before any functional pattern is specified); (b) the set of ultimate categories; (c) the set of all categories which are neither initial subcategories nor ultimate categories. (3) the set of all functions {F 1 , F 2 , ••• , F ,.} ; (4) the set of all ulti1nate constituents {u 1 , u 2 , ••• , u,.}. The ulti1nate constituents are the ter1ninal elements of the system, as against the non-ter1ninal functions and categories. For any independent linguistic expression, .the rules of the grammar determine a structural description of the type symbolized in diagram (8). I repeat that such structural descriptions contain more inforrnation than the 'P-markers' as defined in transfor1national grammar, in that not only (categories of) constituents are assigned to the linguistic expression, but also, to each constituent at any level, the function which it fulfils in relation to other constituents. In such a description the categories take care of (part of) the paradigmatic di1nension. The functions define syntagmatic functional relations between different same-level categories dominated by a single higherlevel category. This latter formulation covers two restrictions which, as will be evident from (8), are incorporated in this version of functional grarnmar: (i) there are no functional relations between included and including

172

OUTLINE-SKETCH OF A FUNCTIONAL GRAMMAR

constituents, but only between same-level constituents. Thus, the assumption is that in a sentence like John came there is a functional relation between John and came, but not between John and John came or between came and John came. (ii) there are no functional relations between constituents which are not parts of the same constituent. Thus, in the sentence The boy came, there are functional relations between The boy and came, and between The and boy, but not between The and came, or between boy and

came. Further evidence for the validity of these two restrictions will be adduced below. Sumrnarizing, we can say that the foil owing different relations are presented in the structural description: (a) category to subcategory; (b) category to functional pattern; (c) function to category; (d) category to ultimate constituent; (e) function to function.

9. 2. 6. A siniplified example The value of these conventions may be illustrated by a si1nple example. Consider the sentence: (9) The man hit the ball. Let us suppose that this sentence receives the following structural de4 scription in a functional gra11unar :

' This description depends on a number of decisions for which no arguments are given here (e.g., the ternary analysis into SUBJECT+PREDICATOR+OBJECf). Moreover, the description is far from complete. It is, however, not the details of the analysis, but the kind of inforrnation presented in a structural description which is relevant here.

9.2. LAY-OUT OF STRUCTURAL DESCRIPTIONS

173

(10) ile

s •

Sdeclarative

SUBJECT

PREDICATOR

np

OBJECT

Ytrans

DET

HEAD

DET

HEAD

art

n

art

n

the

man

the

ball

hit

Such a structural description gives the following information about sentence (9): (i) The man hit the ball. is an independent linguistic expression; it is a sentence; it is a declarative sentence. (ii) Its i1nmediate constituents are the man, hit, and the ball. (iii) the man and the ball are noun phrases, hit is a transitive verb. (iv) The immediate constituents of the man and the ball are the and man, the and ball, respectively. (v) the is an article, man and ball are nouns. (vi) the man has the function SUBJECT, hit has the function PREDICATOR, and the ball has the function OBJECT. There is a SUBJECTPREDICATOR-OBJECT relation between the man, hit, and the ball. The sequence SUBJECT+ PREDICATOR+ OBJECT is a functional pattern of declarative sentences. (vii) the has the function DETERMINER in both the man and the ball; man and ball have the function HEAD in these noun phrases. There is a DETERMINER-HEAD relation between the and man, and

174

OUTLINE-SKETCH OF A FUNCl'IONAL GRAMMAR

between the and ball. The sequence DETERMINER+ HEAD is a functional pattern of noun phrases. There are no other functional relations in this sentence. I.e., there is no functional relation between, e.g., man and hit, or between man and ball, or between the man hit and the ball, etc. This seems, for the ti1ne being, to be a quite natural restriction on the notion 'functional relation' : there are only functional relations between same-level constituents which are part of a single higher-level constituent.

9. 2. 7. Categorial vs. functional differences In a functional grammar of this type, there are three basic di1nensions along which linguistic expressions can be grammatically different: (i) They can have different constituent-analyses; (ii) They can be categorially different; (iii) They can be functionally different. Each of these aspects, constituency, categorization, and functional patterning, thus constitutes an independent parameter in the structural description. Constituency is indicated in the diagram by the branching lines of the tree. Categorization is indicated by the categorial nodes in small Roman type. And functional patterns are symbolized by sequences of same-level nodes in capitals. (i) needs no illustration. As to (ii) and (iii), consider the sentences: (11) (12)

The man hit the ball. The ball hit the man.

If (11) and (12) are analysed as in (10), they have identical constituentanalyses and identical categories in the same order. But they are functionally different. In (11), the man is SUBJECT and the ball is OBJECT; in (12) the ball is SUBJECT and the man is OBJECT. Next, consider sentences (13) and (14): (13) (14)

John hit the ball. He hit the ball.

(13) and (14) have, again, identical constituent-analyses and they are furthermore functionally identical. But they differ in the categories

9. 3. THE SELECTION OF FUNCl'IONS

175

assigned to John and to he, namely proper noun and personal pronoun, respectively. This leads over to another irnportant characteristic of this functional grarn1nar. In such a system there is no need to assign the sa1ne category to such gra1n1natically diverse items as he, John, the man, the three boys, someone, who, what I would like to say, all those nice things which you said to me, etc. (which are all noun phrases in a transfor1national grammar, cf. above, section 4.1.2.). The similarity in the gra101natical behaviour of these items (which, as such, belong to different categories) is accounted for by the fact that they can all fulfil the SUBJECT-function. In a way, any grammatical function (though it is not in itself a category) determines a category of categories which can fulfil this function. Consequently, there is no need in functional grarnmar to extend such a categorial ter1n as 'noun phrase' to single nouns, pronouns, etc., since the sin1ilarities between these categories are accounted for at the functional level. Therefore, when I indicate categories like np (noun phrase), n (noun), pspn (personal pronoun) below, I understand these to constitute disjunct categories, which do not stand in a relation of subcategorization to each other. One can say that a noun phrase always contains one or more nouns, but not that a noun is a noun phrase. As to personal pronouns, these have so many grammatical characteristics of their own that to treat them as included in the noun or noun phrase category is si1nply misleading. Due to these properties, a functional grammar provides for greater flexibility than a purely categorial system, since the relevant gra1nmatical facts are divided between the functional and the categorial di1nension.

9. 3. The selection offunctions

A further, very important point must be made here. If my suggestion that gram1natical functions are prirnitives of grammatical theory is accepted, the question i1n1nediately arises just what and how many functions will be needed in the description of any one particular language. ls not the selection of functions and their assignment to categories in danger of becoming a quite arbitrary matter in the absence of any general governing principle? This point can be answered in terms of a rule and two postulates.

176

OUTLINE-SKETCH OF A FUNCTIONAL GRAMMAR

Together, these for1n the criteria by which a functional description can be evaluated. Rule. At least so many functions are set up for a language as there are grammatical (not semantic) differences between the linguistic expressions of that language which cann.o t be correlated to differences in constituency 5 and/or in categorization. Postulate I. The grammatical functions in any one particular language constitute a finite set. Postulate 2. A subset of the grammatical functions recog11ized for any one particular language is common to all natural languages, and a subset is common to more than one, but not to all languages. I.e., there are universal functions, general functions, and particular (language-specific) functions. Evaluation criteria. Of two grammars G 1 and G 2 , G 1 is superior to G 2 (i) if G 1 accounts more completely for all the differences mentioned in the Rule than G 2 ; (ii) if G 1 requires a smaller number of different grammatical functions than G 2 ; (iii) if G 1 has fewer particular functions than G 2 ; (iv) if, within the set of non-particular functions, G 1 has more universal functions than G 2 • Exactly the same argument can be repeated for categories. Together, these rules and requirements (which seem to irnpose quite sensible conditions on linguistic theory) guarantee (i) the proper restriction of the number of functions figuring in a linguistic description, and (ii) the achievement of maximum generality within such a description.

9. 4. Relative simplicity of a functional grammar The effect of the functional system proposed here is to increase the number of different linguistic patterns ascribed to a language, and to reduce the number of operations performed on these patterns. 5

Cf. above, chapter 8.

9.S. RULES NEEDED IN A FUNCTIONAL GRAMMAR

177

Loss of simplicity can hardly be invoked against such a system, (1) since sirnplicity can only be meaningfully assessed for a complete linguistic description, an ideal from which we are unfortunately too far removed, (2) since si1nplicity considerations can never be valid against an approach which achieves a higher degree of adequacy, i.e., which yields a better picture of the language in question, and (3) since, paradoxically, the abandonrnent of transformational rules itself reduces the total complexity of the gra1nmar in a quite considerable manner. Notice that the overall complexity of a functional gra1n1nar should not be compared with that of the base component of a transforrnational grammar, but with the combined complexity of the base component and the transforn1ational component of such a gra1n1nar. Structural descriptions determined by a functional gramn1ar will generally be more complex than the base P-markers of a transforrnational grammar. But whereas the latter contain only part of the grammatical organization assigned to any one particular linguistic expression, to be completed by the transformational component, the former contain all gra1n1natical infor1nation relevant to a linguistic expression. It is presumed here that a complete functional grarnmar, though more complex than the base component of a transformational grammer, will be decidedly sirnpler than the full syntactic component of such a grammar. As always, the verification of this thesis will have to await a further elaboration of the theory of functional grammar.

9.5. Types of rules needed in a functional grammar

To achieve the correct structural descriptions for linguistic expressions on the basis of the set of prirnitives described above (section 9.2.5.), a functional grammar needs a system of rules among which different types are to be distinguished. But notwithstanding these differences, all rules are of the same basic pattern, which is essentially that of the rewriting rules developed for the base component of transformational grammar (cf. above, section 5. 2. ). Thus, each rule is of the general form: (15)

A

>

z Ix

y

to be read as: 'rewrite A as Z in the context X

Y'. In a functional gram-

178

OUTLINE-SKETCH OF A FUNCTIONAL GRAMMAR

mar A is a single category symbol or a single function symbol. Z is a single category symbol, a single ulti111ate constituent symbol, or a string of function symbols. In the latter case, Z need not be continuous, nor need it be finite. X and Y are arbitrary strings of symbols specifying the context in which A can be rewritten as Z. Any one or both of X and Y 6 may be null. We thus have a single type of rule, which is formallysi1nilartothe constituent structure rules of transforn1ational grammar. But several restrictions imposed on such rules in the latter theory are dropped, and there is more diversity in the rules since a more differentiated set of pri1nitive symbols is involved. Though all of the same basic type of (15), the rules of a functional grammar fall into different subtypes according to their more specific properties. I shall now discuss these differentiating properties, and show how they are related. The main purpose is to demonstrate at least the f orn1al feasibility of the system envisaged here. For this reason, I shall again pay special attention to those features proper to functional grammar which have not been described or defined in the context of other theories.

9. 5 .1. Four basic sets of rules First of all, we must distinguish four basic non-overlapping sets of rules, differentiated by the kinds of symbols which may appear on the left-hand and right-hand side of the arrow, viz.

(i) (ii) (iii) (iv)

subcategorization rules function rules category rules specification rules

These rule-types can be defined as follows:

(i) subcategorization rules

6

Cf. Chomsky 1965: 66, Dingwall 1966: 314.

9.5. RULES NEEDED IN A FUNCnONAL GRAMMAR

179

Given any category, subcategorization rules specify a subcategory included in it. Some examples are:

(16) a. b. c. d.

ile > s np • npsingular personal pronoun > personal pronounfirst person verb • verbtransitive

Left-hand symbols in subcategorization rules are single categories. Righthand symbols are single subcategories of the left-hand categories. Features differentiating a subcategory within a category are written as subscripts after the category-label. A subcategorization rule of the general type A-+ Z thus formalizes the relations 'Z is a subcategory of A', 'Z is included in A', 'A includes Z'. In recent work by Chomsky (1965: 79ff.) and others it has become clear that certain difficult problems are involved in subcategorization. I shall have little to say about these problems here, since I do not need a detailed solution of subcategori:zation for the description of coordinations. It seems to me, however, that any such solution could be readily incorporated in the present scheme.

(ii) function rules Given any non-ultimate category, function rules specify the possible functional patterns for this category. Some examples: (17) a. sentence-combination > QUESTION+ANSWER b. Sdecl • SUBJECT+ PREDICATOR +OBJECT c. np-+ MODIFIER+HEAD In function rules left-hand symbols are single categories. On the righthand side we have sequences of function symbols. A function rule of the general type A • Z thus formalizes the relation 'Z is a functional pattern of A'. For any category, there may be more than one functional pattern. In a gra1n1nar of English, e.g., we would have such function rules as:

180

OUTLINE-SKETCH OF A FUNCTIONAL GRAMMAR

(18) a. sdeci • SUBJ+PRED+OBJECT +INDIRECT OBJECT b. Sdecl • SUBJ+PRED+OBJECT c. Sdecl • SUBJ + PRED

Following the conventions set up in transformational grammar, such rules may be abbreviated to: (19)

sdec1 •

SUBJ+PRED+OBJECT +INDIRECT OBJECT SUBJ + PRED+OBJECT SUBJ+PRED

and this, again, may be abbreviated to: (20)

Sdecl • SUBJ+PRED (OBJECT (INDIRECT OBJECT))

In this way, the most general specification of functional patterns for any category is achieved. At the same time, the 'optional' character of OBJECT and INDIRECT OBJECT is indicated. (iii) category rules Given any function, category rules specify a single category which fulfils this function. Some examples: (21) a. b. c. d.

QUESTION • Sinterrogative SUBJECT , np OBJECT • personal pronoun MODIFIER • adjective

Left-hand symbols are single functions. Right-hand symbols are single categories. A category rule of the general type A • Z thus formalizes the relation 'Z fulfils the function A'. (iv) specification rules Given any ultimate category, specification rules specify a particular ultimate constituent representing this category. Some examples:

9.5. RULES NEEDED IN A FUNCfIONAL GRAMMAR

181

(22) a. nounproper • John b. personal pronOUDrirst person sin1ular • I c. verbtrans third person sin1u Jar past -+ hit Left-hand symbols are single ultimate categories. Right-hand symbols are single ultimate constituents. A specification rule of the general type A • Z thus formalizes the relation 'A is represented by Z'. Subcategorization rules, category rules, and specification rules are oneone rules, i.e., they necessarily have a single symbol on the right-hand side. Only the function rules are one-many (or 'branching') rules, since they must have a sequence of at least two functions (a functional pattern) on the right-hand side. This difference formalizes the fact that functions are inherently relational notions, whereas categories, subcategories and ultimate constituents are inherently non-relational notions (cf. above, section 8. 4.). We can say of the element John that it is a proper noun in abstraction from any particular linguistic expression in which it appears. But we cannot say that it fulfils the function of SUBJECT unless it appears in a linguistic expression which is characterized by a functional pattern comprising the function SUBJECT: a function is necessarily part of a functional pattern, and related to the other functions of this pattern. Function rules may be either continuous or discontinuous, and either definite or indefinite. About these distinctions, the following remarks must be made.

9. 5. 2. Continuous vs. discontinuous rules

Continuous rules are rules in which the symbols on the right-hand side follow each other directly, without interruption. Discontinuous rules are rules in which the symbols on the right-hand side are interrupted, or separated from each other so as to allow for the insertion of other symbols in between. Thus, if a sentence like (23)

I called you up.

182

OUTLINE-SKETCH OF A FUNCTIONAL GRAMMAR

is analysed as in (24): (24) ile

s .



SUBJECT

PREDICATOR

OBJECT

pspn

v

pspn

HEAD

MODIFIER

v

adv

called

I

vou •

up

then there must be a function rule like (25)

vp--+ HEAD+ ... +MODIFIER

This rule must be specified in such a manner that the personal pronoun which functions as an OBJECT is inserted between the HEAD and the MODIFIER. In this case, this must be brought about by a special context-restriction involving the OBJECT. The description must be further diversified to account for the fact that if the OBJECT is a personal pronoun like you, the discontinuity is obligatory. I.e., there is no such sentence as: (26)

*I called up you.

But if the OBJECT is a noun phrase like my father, then the discontinuity is optional. We have:

9.5. RULES NEEDED IN A FUNCTIONAL GRAMMAR

183

(27) a. I called up my father. b. I called my father up. The facts of (23)-{27) could be taken care of by rules of the foil owing kind: (28) a. vp > HEAD+( ... )+MODIFIER / - OBJECT(np)._ b. vp > HEAD+ ... +MODIFIER / OBJECT (pspn)Rule (28a) should be taken as expressing the fact that in the context of an OBJECT fulfilled by an np the functional pattern HEAD+ MODIFIER of the vp may be discontinuous (as symbolized by( ... )); rule (28b) as expressing that in the context of an OBJECT fulfilled by a pspn this functional pattern must be discontinuous. Both rules should be interpreted in such a way that if the functional pattern is discontinuous, the OBJECT (np or pspn) is inserted in between the HEAD and the MODIFIER. As is well known, transformational grammar does not provide for the description of discontinuous constituents in the base component (i.e., in base rules of type ( 15), Z is always a continuous sequence of symbols). Discontinuous constituents are then in all cases derived from continuous sequences in the underlying structure. The points I want to emphasize here are the following: (i) if we have discontinuous constituents in linguistic expressions, why decree a priori that these should always be described in terms of continuous ones'? There seems to be no empirical motivation for this; (ii) transformational grammarians have often recognized that indeed discontinuous constituents could in principle be described by constituent structure rules (see, e.g., Chomsky 1957a: 41 fn. 7 6), though unspecified difficulties are said to arise if this is attempted; (iii) successful attempts have been made by others to develop the formalism needed for describing discontinuities in anon-transformational framework. Thus, Y ngve introduces rules like: (29)

A

B+ ... +C

which, if applied, for example, to a sequence like AQRST, yields a sequence AQCRST (1960: 448-9). Even if, as Postal argues (1964a: 69), Yngve's for1nalization of these rules is 'far from complete', it seems clear 1

Cf. also Postal 1964a: chapter 5.

184

OUTLINE-SKETCH OF A FUNCTIONAL GRAMMAR

that there is no reason to doubt that a complete f orn1alization of such rules as these is possible. We may add that they have been used and found adequate by Harman (1963: 606-7). On the other hand, Yngve's rules must be generalized to allow for more than two discontinuous constituent-parts. I.e., we may in certain cases also need rules of type:

(30)

A • x+ ... + Y+ ... +z

and so on. Moreover, we cannot restrict ourselves to one intervening category or function. E.g., the parts of a discontinuous constituent may in some cases be separated by two elements functioning, e.g., as OBJECT and TIME MODIFIER, such as in Dutch: (31)

Ik belde je gisteren op. lit. 'I called you yesterday up'. = 'I called you up yesterday'.

The necessary formalism for such extensions of discontinuous rules has been developed by G. H. Matthews who formulated discontinuous rules of the most general type conceivable (1963). The fact that Matthews further proves that discontinuous rules do not increase the weak generative capacity of a context-sensitive constituent structure grammar (ibid. 8 146) is not relevant here. This only means that the full set of strings (regardless of their structural descriptions) generated by a system containing discontinuous rules can also in each case be generated by a system not containing such rules. This can be achieved if any part of a discontinuous constituent is itself treated as a constituent. This, however, results in incorrect structural descriptions if we start from the fact that there are discontinuous constituents in natural languages. If this is correct, then any system not accounting for them is inadequate for the description of natural languages in strong generative capacity, no matter what its weak generative capacity may be. It is evident that to be really effective discontinuous rules must be context-sensitive in the sense that they must contain a specification of the elements which may or must be inserted into the 'open space(s)' (cf• •

In using this fact as an argument against such systems, Bierwisch fails to distinguish between strong and weak generative capacity ( l 966b: 42 fn. 12). For this distinction, cf. above section 5. 2.

8

9.S. RULES NEEDED IN A FUNCTIONAL GRAMMAR

185

Rohrer 1967: 53-5). There are several ways in which this can be brought about, one of which has been indicated above. Some further details will be given below (section 9. 7 .1.). In the present context I shall not consider the question of exactly what restrictions must or can be placed on the quite general formulation of discontinuous rules provided by Matthews. This is an empirical question which needs further investigation. What I have tried to establish is the fact that formally there is no objection at all against the direct specification of discontinuous constituents by means of discontinuous constituent structure rules.

9.5.3. De.finite vs. indefinite rules (or rule-schemata) Definite rules are rules in which a definite number of symbols appears on the right-hand side. All one-one rules are of course definite. In the case of function rules, definite ones are those which have a finite number of functions on the right-hand side. All examples of function rules given so far are definite (cf. above, section 9. 5. I.). Indefinite rules or rule-schemata are rules in which an indefinite number of symbols appears on the right-hand side. E.g.:

(32)

A

>

B" (n

~

1)

The properties of these rule-schemata have become sufficiently clear in chapter 5 above (see especially section 5. 6.). I repeat that they are in fact abbreviations of infinite sets of rules. Thus, (32) is an abbreviation of the set of rules: (33)

A A A A

B > B+B > B+B+B > B+B+B+B, etc. >

The inclusion of rule-schemata makes it impossible to require that a gra1111nar must be a finite set of rules. Instead, a grammar must be said to be a finite set of rules and rule-schemata, determining an infinite set of rules, determining an infinite set of linguistic expressions together with their structural descriptions (cf. section 5.6. above).

186

OUTLINE-SKETCH OF A FUNCTIONAL GRAMMAR

Since in a functional gra1n1nar only function rules are one-many, it is evident that all rule-schemata in such a gran1mar must be function ruleschemata. Any rule-schema in a functional grammar is of the general type:

(34) c • Fn(n >I) i.e., in such a rule-schema a functional pattern consisting of an indefinite number of identical functions is assigned to a single category. It should also be noticed that these rule-schemata are only needed in the description of coordinations. I shall therefore leave the elaboration of these schemata to the next chapter, which deals with the description of coordinations in a functional grammar. So far, I have distinguished the four basic types of rules of a functional grammar: subcategorization rules, function rules, category rules, and specification rules. I have further defined the difference between continuous and discontinuous function rules, and I have noted that the latter are necessarily context-sensitive. Finally, I have discussed the distinction between definite function rules and function rule-schemata. To complete this description of the rules of a functional grammar, we must distinguish more properly between context-free and context-sensitive rules, and add the distinction between optional and obligatory rules.

9. 5. 4. Context-free vs. context-sensitive rules The difference between context-free and context-sensitive rules has already been indicated more than once in the preceding sections (cf. above, sections 5. 2., 9. 5., 9. 5. 2.). Context-free rules are rules which can be applied regardless of the context in which the left-hand symbol is embedded (cf. Chomsky 1965: 67). Context-sensitive rules are rules which can be applied if and only if certain conditions on the context of this symbol are fulfilled, i.e., if the context-indication appended to the rewriting rule after the slant line matches the actual context of the particular symbol which is to be rewritten (cf. Bach 1964: 16). It has been proved in the theory of generative grammar that it is absolutely impossible to describe the linguistic expressions of a natural language and their structures correctly by means of systems which only

9.6. RECURSION IN A FUNCfIONAL GRAMMAR

187

include context-free rules. Natural languages are not context-free languages, and the grammars of natural languages are not context-free grammars (see Postal 1964a: 75ff., 1964b: 145ff.). This could hardly have been expected, since it is a well-known fact that the features of structure and the elements appearing in a linguistic expression are interconnected by many intricate dependences, conditioning each others' occurrence in various ways. Such dependences necessarily require context-sensitive rules in the description. In a functional grammar, the representatives of at least the first three basic sets of rules (subcategorization, function, and category rules) may be context-sensitive. Specification rules selecting particular ultimate constituents will probably always be context-free, though it is possible that a further integration of grammar and semantics (cf. below, section 12 . 6.) would require context-sensitive specification rules in which semantic features of the context would be involved. At present, I cannot be very much more definite on this point. Several examples of context-sensitive rules in a functional grammar will be given below (section 9. 8.).

9. 5. 5. Obligatory vs. optional rules An obligatory rule is a rule which must necessarily be applied to a certain non-ter1ninal symbol. An optional rule is a rule which may be applied to such a symbol. If, for a certain non-terminal symbol X, there is but one rule X • Y which can be applied to this symbol at a given point, then this rule is obligatory. If there are at least two rules X • Y, X • Z which can be applied to X, then each of these rules is optional. In a functional grammar, any representative of the four basic sets of rules may be either obligatory or optional. The exact repartition of both types can only be assessed within a completely elaborated description. This completes the preliminary description of the different types of rules constituting a functional grammar. I shall now add some remarks about the occurrence of recursion in such a grammar.

9. 6. Recursion in a functional grammar The recursive property of a grammar has of course been clarified and extensively discussed in the literature on transformational generative gram-

188

OUTLINE-SKETCH OF A FUNCTIONAL GRAMMAR

mar. Let me say something about the status of recursion in a functional grammar. One should distinguish, first of all, between the terms recursive element, recursive rule, and recursive rule-system. If from a symbol X we can derive, by the rules of the gra1n1nar, a string which again contains X, and if from this latter X we can again derive a 9 string containing X, and so on, then the symbol X is a recursive element. A recursive element can be introduced either by a resursive rule or by a recursive rule-system. A recursive rule is a rule which contains the same symbol on the right-hand side and on the left-hand side of the arrow, and which may be reapplied to its own product: (35)

x •... x ...

A resursive rule-system is a system of rules in which a recursive element appears though no recursive rules are included in the system. E.g.: (36) a.

x

b. A c. B

• . .. A



•••

... B ...

• ... x ...

where rules a, b, and c may again be applied to the result of rule c. In connection with recursion, we must mention two fundamental facts concerning functional grammar, viz.: (i) in such a grammar recursive elements will be one of two possible devices for achieving the description of an infinite set of linguistic expressions on the basis of a finite vocabulary. (ii) a functional grammar will contain recursive elements, but no recursive rules. (i) Given a finite set of symbols, say {A, B, C}, there are three different ways to form an infinite set of sequences of these symbols, viz.: (a) the introduction of recursive rules like A • A+ B, (b) the introduction of recursive rule-systems like A • B+C, B • A, (c) the introduction of rule-schemata (or indefinite rules, see above, section 9. 5. 3.) like A > B"+C", specifying any sequence consisting of a finite number of B's foil owed by a finite number of C's. 9

Cf. Chomsky & Mill er 1963 : 290, Bach 1964: 16.

9.6. RECURSION IN A FUNCTIONAL GRAMMAR

189

(ii) The base component of a transfor1national grammar may contain rules of type (a), type (b), and type (c).There are, however, certain for1nal difficulties connected with recursive rules (cf. Bach 1964: 39, 67, Postal 1964a: 16, 23-4). These have even led to the formulation of a restriction by which all recursive rules would be banned (Postal 1964a: 16), though in fact the formal objections do not apply to such rules as: (37)

A • B+A+C

i.e., recursive rules in which the recursive symbol is embedded between other symbols on the right-hand side of the rule (cf. Bach 1964: 39, 67). At any rate, these formal problems do not arise in a functional grammar, since recursive rules are excluded by the very structure of such a grammar: in function rules, category rules, and specification rules, the left and right-hand symbols are chosen from different non-overlapping subsets of the vocabulary. In subcategorization rules, it is evident that left- and right-hand symbols must be different, since otherwise such rules would be vacuous. On the other hand, the introduction of recursive elements is not excluded in a functional grammar. E.g., a certain category to which a functional pattern has been assigned may reappear in the further categorial specification of this functional pattern. For an obvious example, consider: (38)

I said: I do not like this at all.

In (38) we have the expression I do not like this at all, which has the properties of -a sentence, but fulfils the function of OBJECT in the sentence 10 as a whole. This situation could be described in a functional grammar by the rules : 10

As Professor Uhl en beck pointed out to me, this description requires modification because of the fact that I do not like this at all in (38) is a quoted sentence. An adequate grammar must be capable, e.g., of differentiating properly between (a) I said it. and (b) I said: 'it'. There are, in principle, two different methods for achieving this differentiation: we could either assign different functions to it and 'it' in (a) and (b), respectively; or we could describe it and 'it' as belonging to different categories. It seems to me that the second alternative is preferable: the ordinary pronoun it and quoted 'it' are different items, regardless of the grammatical function which they fulfil: we could thus differentiate between a category of pronouns and a category of quoted pronouns; or we could set up a general category of 'quoted expressions' and describe all other

190

OUTLINE-SKETCH OF A FUNCTIONAL GRAMMAR

(39) a. s • SUBJECT+PREDICATOR+OBJECT b. OBJECT • s Indeed this seems to be an appropriate description, since in the context: (40)

I said:_

we can have any sentence, including sentence (38), witness: (41)

I said: I said: I do not like this at all.

Again, I shall refrain from investigating the exact range of recursion in a functional grammar. For our present purposes it is sufficient to have indicated its basic place within such a system, as well as its difference from rule-schemata which in a different way make for the f or1nation of infinite sets of sequences on the basis of a finite set of symbols.

9. 7. The organization of a functional grammar A functional grammar is a system of rules generating all and only the grammatically well-for1ned linguistic expressions of a language together 11 with their structural descriptions. So far, I have described the general fortnat of these structural descriptions, the different types of rules needed to arrive at them, and the set of primitives figuring in these rules. We must now consider the question of how this system should be organized to achieve the proper result. In this connection two basic problems present themselves: (i) the ordering of the rules and of their application: (ii) the nature of the 'derivations' in which the rules are applied.

categories as optional subcategories of this category. At any rate, if I do not like this at all would be a 'quoted sentence' rather than a 'sentence' in (38), it would still be true that any quoted sentence can again contain another quoted sentence. Therefore, the illustrative value of this example for the recursive property of a functional grammar would remain under any of the proposed modifications of the description. 11 Cf. above, section 4. 1. and below, section 12. 6. for other kinds of well-formedness.

9.7. THE ORGANIZATION OF A FUNCflONAL GRAMMAR

191

9.7.1. The ordering of the rules and of their application In part, the ordering of the rules and of their application has become evident in what has been said so far in this chapter. I can now give a more systematic account of these points. I assume a partial ordering of the rules in the grammar, and a set of instructions governing the order of their application. The four basic sets of rules: subcategorization rules, function rules, category rules, and specification rules appear in this order in the grammar. A further ordering of the rules within each of these sets will probably be required, but I shall not consider this point in the present survey. The order of application is as follows: (i) choose a subcategorization rule which has ile as its left-hand symbol; (ii) determine whether further subcategorization of the resulting right-hand category is possible; (iii) as soon as a category has been reached for which no further subcategorization is possible, choose a function rule which has this category on the left-hand side; (iv) for each of the functions resulting from the application of this function rule, choose a category rule applicable to it; (v) return to the subcategorization rules and apply those which are applicable to the categories concerned; (vi) return to the function rules and, if the categories resulting from procedure (v) are non-ultimate categories, apply those function rules which are now applicable; (vii) repeat steps (v) and (vi) until all symbols have been developed into ultimate categories; (viii) for each ulti1nate category, select a suitable specification rule; (ix) apply these until no non-ter1ninal symbols are left. The result is a string of ultimate constituents to which no further rules of the grammar can be applied. This ordering can be represented as follows (the boxes represent the different sets of rules, the arrow the course of a derivation):

192

OUTLINE-SKETCH OF A FUNCTIONAL GRAMMAR

(42) ile

subcategorization rules

function rules ·

category rules

specification rules



ter1n1nal string Of course, this by no means exhausts the problem of the ordering of the rules and of the application of the rules. Further light is thrown on this problem by the foil owing considerations: whereas a context-free rule can be applied at any point in the derivation, a context-sensitive rule can only be applied if the required context is present. Therefore, the rules must be applied in such a way that for each context-sensitive rule all context-free rules must have been applied which are needed for the specification of the proper context. This can be more precisely formulated as follows: for ~ny

9. 7. THE ORGANIZATION OF A FUNCflONAL GRAMMAR

193

context-sensitive rule selected at any point in the derivation, deterrnine if the present context allows of its application. If this is not the case, develop the context further by other rules. After each application of such a rule, determine again whether the resulting context matches the context-restriction of the context-sensitive rule. As soon as this point is reached, apply the context-sensitive rule in question.

9. 7. 2. The notion of'derivation' in a functional grammar This brings us to the problem of how the context is to be specified. I believe that in the present theory of gra1n1nar it will be necessary (i) to refer to context which has already been rewritten by previous rules, and (ii) to refer simultaneously to context features which belong to different levels in terms of structural diagrams such as (10). Thus, there must be a for1nalized equivalent to such a rule as: (43) If the SUBJECT is a noun, a noun phrase, or a third person personal pronoun, then the finite verb is in third person form. This could be formulated in a context-sensitive subcategorization rule as follows: (44)

finite verb • finite verb 3 d

ps /

SUBJECT {n, np, pspn3 d

ps)

The question is, therefore: how do we get derivations in which at each point such contexts as the one given in (44) can be matched? In derivations of the f orrnat suggested by Chomsky for a constituent structure gra1n1nar (1957a: 27, 1965: 66, Bach 1964: 15) this is excluded, since each particular string of such a derivation does not contain any information as to how it was forrned from earlier strings. Thus, if SUBJECT had been rewritten as n, or np, or pspn3 d PS' then there would, in such a derivation, be no string in which both SUBJECT and n or np or pspn 3 d ps could together figure as context for the application of further rules. Notice that the fact that in derivations defined in this way it is impossible to 'look back' to earlier stages of the derivation (i.e., that for each string of such a derivation there is no way to deter1nine how it was arrived at) was regarded by Chomsky as a serious flaw in constituent structure grammars as such, and as an argument for introducing rules of a

194

OUTLINE-SKETCH OF A FUNCTIONAL GRAMMAR

fundamentally different type, viz. transfor1nations (1957a: 37-8). But this restriction is only inherent in his particular definition of the notion 'derivation'; there is no proof that within the theory of constituent structure gramrnar this notion could not be redefined in such a way that this problem does not arise. There are some further problems connected with Chomsky's definition of derivation, which I shall only mention in passing. With this particular definition the requirement that a structural description must be uniquely constructible from a derivation can only be met by certain restrictions on 11 constituent structure rules which are otherwise unnecessary. A further drawback is that the structural description is not automatically given with the derivation as such, but must be 'extracted' from it by certain further conventions. All these problems are solved if we define a derivation not as a sequence of strings of symbols, where each following string is the result of applying a rule to the preceding one, but as a sequence of labelled bracketings of symbols, deterrnined by the following convention: for any rule of type A • X + Y, applied to a string containing A, first rewrite A as X + Y, then put brackets around X + Y, and put A as an index before the first bracket, the result being A(X+ Y). If this procedure is consistently carried out, each particular line of a derivation will contain a complete record of the rules which were applied in its formation. The final line of such a derivation will be a labelled bracketing of ulti111ate constituents in which all grammatically relevant facts about the particular linguistic expression generated will be explicitly indicated. I.e., the final line contains not only the linguistic expression as such, but also the structural description assigned to it by the grammar. This is a very satisfactory state of affairs, since we want a grammar to be a system of rules generating all and only the grammatically well-formed independent linguistic expressions of a language, and assigning a structural description to each of these. Thus, by redefining the notion of derivation along the lines indicated above, we have not only solved the problems connected with derivations in Chomskyan generative granunar, but we have also simplified the theory of constituent structure grammar, and made it more adequate for the achievement of its purpose.

11

Cf. e.g. Bach 1964: 35-41.

9. 8. AN ILLUSTRATIVE EXAMPLE

195

9. 8. An illustrative example To illustrate these various points, let us once more take a simplified example. Consider the sentence ( 45)

The man came.

Let us suppose that the structural description of this sentence is as represented in diagram (46):

(46) ile

s

SUBJECT

PREDICATOR

np

finite verb •

DET

HEAD

fvintr past .

art

fvintr past 3d ps sg

man

the

came

For specifying this structure the following rules would be necessary:

(47) subcategorization rules: a. ile • s b. S • Sdecl c. np •

DPs1

196

OUTLINE-SKETCH OF A FUNCTIONAL GRAMMAR

d. fv • fvintr e. fvintr • fvintr past f. fvintr past • fvintr past 3d ps sa / SUBJECT{np,.) (48) function rules: a. Sdecl • SUBJECT +PREDICATOR b. np11 • DET +HEAD (49) category rules: a. SUBJECT • np b. PREDICATOR • fv c. DET •art d. HEAD • n., I nps,(DET ) (50) specification rules: a. art • the b. Ds1 • man C. fVintr past 3d ps sg • Came Of these rules, all but (47f) and (49d) are context-free. The context-sensitive rules express the facts (i) that the finite verb is in third person singular form if the SUBJECT is a singular noun phrase; (ii) that the noun in a noun phrase is singular if the noun phrase is singular. As in earlier examples, I do not clairn that a functional grammar of English would have to contain the particular rules of (47)-(50). The example is meant only to give an impression of the possibilities of the system as such. A derivation for (45), as redefined above, would have the foil owing form (the number before each line refers to the rule applied to arrive at it): •

DERIVATION I. ile

(47a)

2.

(s) ile

(47b)

3.

(

(sdecl))

ile s

(48a)

4.

(( ile

S Sdecl

(SUBJECT+ PREDICATOR)))

197

9. 8. AN ILLUSTRATIVE EXAMPLE

+ PRED)))

(49a)

5.

( ( ( (np) ile S Sdecl SUBJ

(49b)

6.

(( ( (np) + (fv)))) ile s sdeci S·U BJ PRED

(47c)

7.

(( ( ( (np5J) ile s sdect SUBJ np

+

(fv)))) PRED

(47d)

8.

( ( ( ( (npsJ) ile s sdect SUBJ np

+

(47e)

9.

( ( ( ( (np51)) ile s sdeci SUBJ np

+ +

(

(fvintr))))) PRED f v

( (

(fvintr past))))))

PRED fv fvintr

( (

(

(47f)

10.

( ( ( ( (np51)) ile s sdect SUBJ np past 3d ps sg)))))))

(fvintr PRED fv fvintr fvintr past

(48b)

11.

(( ( ( ( (DET + HEAD)))+ ( ile s Sdecl SUBJ np DPs1 PRED fv ( ( (fvintr past 3d ps sg))))))) fvintr fvintr past

(49c)

12.

( ( ( ( ( ( (art) + HEAD))) ile s sc1ect SUBJ np np51 DET ( ( (fvintr past 3d ps sJ)))))) fvintr fvintr past

+

( PRED fv

(49d)

13.

( ( ( ( ( ( (art) + (n1g)))) + ile s sdect SUBJ np np51 DET HEAD ( ( ( (fvintr past 3d ps sg))))))) PRED fv fvintr fvintr past

(50a)

14.

( { ( (the))+ (n 51)))) + ile s sdect SUBJ np np51 DET art HEAD ( ( ( (fvintr past 3d ps sg))))))) PRED fv fvintr fvintr past ((

(

(

198

OUTLINE-SKETCH OF A FUNCTIONAL GRAMMAR

(50b)

15.

(( ( ( ( ( ( (the)) + ( ile s sdect SUBJ np np51 DET art HEAD n., (man))))) + ( ( ( (fvintr past Jd ps sJ)))))) PRED fv fvintr fvintr past

(50c)

16.

(( ( ( ( ( ( ( (the)) + HEAD n51 ile s Sdecl SUBJ np npSI DET art (man))))) + ( ( ( ( PRED fv fvintr fvintr past fviotr past 3d ps (came))))))))

sg

Such a derivation is admittedly quite complex, but observe that the derivations proposed for a functional grammar do not belong to the gramrnar as such: the grammar is the system of rules completely deterrnining all the well-formed expressions and their structural descriptions. As soon as the rules have been correctly forrnulated and ordered, and the order of their application has been deterrnined in so far as it is of grammatical relevance, the grammatical description is complete. A derivation is only a means to put the formulated rules to work in order to test them, i.e., to determine the expressions generated by the rules and their structural descriptions. Furthermore, unlike derivations in transformational generative grammar, it is only the last line which counts in ·a derivation such as the one given above, since this last line summarizes all the relevant information contained in earlier lines, except for the particular order in which the rules were applied in the derivation. As for this order, there are two possibilities: either it is completely determined in the grammar, or it is irrelevant to the grammatical description. We can say, therefore, that a full derivation such as the one given above does not contain more grammatically relevant information than its final line. On the other hand, this derivation has two undoubted advantages: (i) it provides the basis for a more differentiated context-specification in context-sensitive rules, (ii) its final line contains the complete structural description of the linguistic expression concerned, together with the linguistic expression itself.

9. 9. Conclusion We have thus arrived at the following picture of a functional grammar: a functional grammar consists of a partially ordered set of rules, together

9. 9. CONCLUSION

199

with a set of instructions for applying these rules. Rules and instructions are sufficient to generate an infinite set of complex terminal strings, representing the set of independent linguistic expressions of a language together with their structural descriptions. Any such terminal string can be arrived at in a well-deterrnined manner by means of a derivation. For illustrative purposes such a terminal string can be represented in the form of a tree diagram. Such a diagram can be unambiguously constructed on the basis of the final line of a derivation. It has, however, no further function in the grammar as such. In the preceding pages I have by no means given a complete account of the theory of functional grammar. Further investigation will be needed to establish further details and to evaluate the basic principles which have been presented here. I see no reason, however, why this theory should not be capable of achieving a high degree of adequacy at all possible levels. On the other hand, it does not seem to contain any feature which would not lend itself to a complete formalization or explicitation. Finally, it does not involve us in the many problems which arise in the theory of transformational grammar and leads to simpler structural descriptions in the majority of cases, descriptions which nevertheless contain, in one respect (viz., the functional dimension), more information about the grammatical structure of the linguistic expressions described. It is also evident that the theory offunctional grammar differs in many important respects from any version of 'restricted 'constituent structure grammar. Let us, for the sake of clarity, summarize the main points of difference: (i) independent introduction of grammatical functions. (ii) use of discontinuous rules. (iii) more extensive use of rule-schemata. (iv) no restriction to sentence as highest unit. (v) no recursive rules. (vi) more extensive use of context-sensitive rules. (vii) different notion of 'derivation'. (viii) no problem of 'reconstructing a structural description from a derivation'. (ix) no need for special restrictions on the possible rules of the grammar on account of the definition of 'derivation'. In the next chapter, we shall consider the question of how coordinations can be accounted for within a system of functional grammar.

10. The description of coordinations in a functional grammar

In the description offunctional grammar given in chapter 9 I have not as yet indicated the manner in which coordinations would be incorporated in such a system. Let us now turn to this problem. What we want to achieve is a simple and generally applicable non-transforrnational rule or a set of such rules, capable of accounting for all the various facts which we have noticed about coordinations in the preceding chapters. Let us first surnrnarize some of the minimal requirements which must be met in this attempt. (i) It has become clear that in many instances it is more satisfactory to say that the members of a coordination have the same gra1n1natical function than to say that they belong to the same grarnrnatical category (above, section 4.1.2.). (ii) We have found a variety of structural patterns characterizing coordinations (section 4 .1.4. 2. 2.). (iii) We have seen that not only single constituents, but also certain combinations of constituents can be coordinated (chapter 7). (iv) We have acknowledged the grammatical difference between such sentences as: (I) John and Bill came home, and (2) John and Mary love each other (section 5. 5. 6.). (v) Since a functional grammar as conceived here does not contain transformational rules and does not incorporate any other device for describing 'deletions', it will be evident that in this approach there is no room for describing certain well-formed linguistic expressions as if some constituent(s) had been deleted or left out in their grammatical structure: no wellformed expression is described as 'elliptical' in this sense. In particular, there are no linguistic expressions containing coordinative (sub-)constructions in which any constituent must be supplied or described as if deleted in the final structural description. Only linguistic expressions which are incomplete, interrupted, etc. can be classed as elliptical. But these are not well-formed and are not described by the grammar as such. Thus, in a sequence of question+ answer like:

(3)

Who killed Lincoln? Booth.

..

10.1. A GENERAL SCHEMA OF COORDINATION

201

the expression Booth. is a well-formed English sentence, which need not be described as if derived from Booth (killed Lincoln). This is unnecessary, since in this particular combination the item Booth. is completely sufficient in itself: its grammatical status is completely determined and its semantic content is established unambiguously without the intervention of any further linguistic material (for a contrary opinion, cf. Katz & Postal 1964: 90-1 and fn. 25). Similarly, a sentence like: (4) The man is old and grey.

is perfectly sufficient in itself both in grammatical structure and in semantic content. There is no reason to describe it as if involving in any way the structure of: •

(5) The man is old and (the man is) grey.

(vi) Coordinative constructions can occur at any level of the grammatical hierarchy. We can have coordinated sentences, clauses, phrases, words, stems, and roots. In many cases, the choice of a coordination at a certain point in a structure is optional, and the possibilities are virtually unrestricted: (6) a. I saw John. b. I saw John and Bill. c. I saw John and Bill and Richard. etc. (7) a. A red flag. b. A red and white flag. c. A red, white, and blue flag. etc. In certain cases, the possibilities are restricted in different ways. E.g., in a frame like: (8) I sat between __ . we must either have a plural or a 'collective' noun or noun phrase or a plural pronoun as in: (9) a. I sat between my two friends.

202

COORDINATIONS IN A FUNCTIONAL GRAMMAR

b. I sat between a pair of rogues. c. I sat between them.

or a coordination as in: (10) I sat between Tom and Harry. In other cases again, the choice of a particular coordination is excluded. Thus, we cannot have such sentences as: (11) a. *John and Bill is a very good friend of mine. b. *John or Bill wrote that book together. c. *I saw John but Bill yesterday. etc. All these facts can be accounted for if a general system of rule-schemata is added to the grammar which is sufficiently differentiated to generate all possible coordinative patterns and can be applied at different points in the grammatical hierarchy, its application being restricted by a variety of context-sensitive rules. I shall now first formulate this system, then indicate where it may become operative in the grammar, demonstrate its application to a number of examples, and finally add some remarks on its general corollaries.

10 .1. A general schema of coordination

I start from the assumption that any coordination is a coordination of grammatical functions. First, a coordination of functions is specified and then a category is selected for any of the functions figuring in the coordination. 1

For any coordination of functions &(F), the following choices present themselves: it is either For the symbolism, cf. above section 4. 1. In addition, I use co1 for binary and con for n-ary coordinators (cf. section 4 .1 . 4. 2. 5.). For the patterns described here, cf. section 4.1.4.2.2. 1

10.1. A GENERAL SCHEMA OF COORDINATION

203

( l) a binary coordination &(2F), or (2) an n-ary coordination &(nF). If (I), then it is either (I. I.) a binary coordination by juxtaposition F 1 F 2 , or 2 (1.2.) a coordination by means of a binary coordinator F 1co F 2 • If (2), then it is either (2.1.) an n-ary coordination by juxtaposition F 1 F 2 ••• Fn, or (2. 2.) a coordination by means of an n-ary coordinator con(nF). If (2 . 2. ), then either (2. 2. I.) n occurrences of con are selected: conF 1conF 2 .•. conF n' or (2 . 2. 2.) n-1 occurrences : F 1 conF2 • • • conF "' or (2. 2. 3.) just one occurrence for the last F: F 1 F 2 ••• conF n· This exhausts the (regular) possibilities as presented in chapter 4 above. These choices can be represented in the following diagram:

2

(12) &(F)

&(2F)

&(nF) con(nF)

Thus, any ti1ne a coordination is selected in the grammar, there is in principle a choice from among the six possible basic types given in diagram (12). These choices can of course be added in the form of optional rules to the rules of the grammar as described in chapter 9. We could also say that at certain points in the grammar there is an optional entry to the rules embodying the possibilities given above. If this entry is skipped, then just one function of a certain type is selected. If it is taken, then a coordination of such functions is the result. A further complication arises from the fact that coordinations may be For the sake of simplicity, I do not incorporate the difference between prepositive and postpositive coordinators here.

1

204

COORDINATIONS IN A FUNCTIONAL GRAMMAR

discontinuous. Side by side, we have such sentences as (13a) and (13b), (14a) and (14b): (13) a. b. (14) a. b.

John and Peter and Richard will come to-morrow. John will come to-morrow, and Peter and Richard. He got a bike, a book, and a fountain-pen on his birthday. He got a bike on his birthday, and a book and a· fountain-pen.

To account for these possibilities, the rule-schemata resulting in the functional patterns of (12) will have to provide for optional discontinuities at different points. For some further examples, cf. below section 10.4. 10.2. The incorporation of the schema into the grammar

To achieve the proper result, the system of rule-schemata must be introduced into the grammar in two different ways. In the first place, we want to account for the fact that, e.g., in a sentence like: (15) John, Bill, and Harry came. John, Bill, and Harry can each separately be SUBJECT of came, i.e., that the structure is as represented in:

(16)

s SUBJ

SUBJ

&

SUBJ

v

co John

Bill

and

PREDICATOR

Harry

came

This means that in this case, instead of a single SUBJECT, we want to have a coordination of SUBJECTS. I.e., the coordination-schema must present an option at the function SUBJECT of the function rule: (17)

s • SUBJECT+ PREDICATOR

10.2. THE INCORPORATION INTO THE GRAMMAR

205

such that, when this option is taken, we can get such functional patterns

as: (18) a. SUBJECT & SUBJECT+PREDICATOR b. SUBJECT SUBJECT & SUBJECT+ PREDICATOR, etc. This suggests the first application of the coordination-schema in the grammar: the coordination-schema applies, in principle, to any function specified at any point in the grammar, and leads to the specification of some type of f untion-coordination which replaces the single function figuring in the rule in question. In the second place, we have such cases as: (19) John, Bill, and Harry met. where John, Bill, and Harry are not single SUBJECTS connected with the PREDICATOR, but where the whole coordination is a plural noun phrase which functions as a SUBJECT. Thus: (20)

s SUBJECT

PREDICATOR v

*

*

*

*

























John

Bill

and

Harry

met

From our general principle that any constituent must be assigned a function it follows that John, Bill, and Harry cannot be directly specified as categories in the places marked by asterisks. But their only function in

206

COORDINATIONS IN A FUNCTIONAL GRAMMAR

this structure (in contradistinction to (16)) is to be coordinated MEMBERS within the coordination, which as a whole fulfils the function of SUBJECT. We thus arrive at a description as represented in (21): (21)

s SUBJECT

PREDICATOR

v MEMBER

MEMBER . &

co

John

Bill

and

MEMBER

pr

Harry

met

This suggests the second application of the coordination-schema: at the point where a category is going to be assigned a functional pattern, or is subcategorized, there may be an option to select a coordination instead. In this case, the category concerned is subjected to the coordinationschema. But in contrast to the first application of this schema, all the functions are in this case specified as MEMBERS. For an example, consider a function rule like:

(22)

npP1 -+

DET+HEAD DET + MOD +HEAD • •



In order to achieve the proper result, we must here add an optional entry into the coordination-schema, to yield:

(23)

npp1 -+

DET+HEAD DET + MOD +HEAD • • •

coordination-schema

10.2. THE INCORPORATION INTO THE GRAMMAR

207

where the choice of 'coordination-schema' will then result in any of the six possible coordinative patterns, where all coordinated functions are MEMBERS. In this way the grammatical difference between the verbs came and met in (15) and (19), respectively, is described as a difference in the selection of the SUBJECT; came combines with a coordination of SUBJECTS, and met combines with a single SUBJECT which contains a coordination of MEMBERS. There are two possible objections against this solution: (i) should not the SUBJECTS in (15) and (19) rather be described as identical, the difference being solely attributed to the verb-forms which fulfil the function of PREDICATOR? (ii) does the opposition between (15) and (19) indeed reflect a general grammatical difference, or are we simply dealing with individual grammatical differences due to the particular selection of came and met as PREDICATORS? As to the second point, it will become clear below that the opposition between came and met indeed reflects a general grammatical difference: all comparable cases can in principle be handled in the same way. And for any verb-for1n fulfilling the function of PREDICATOR, there are three and only three possibilities as far as the SUBJECT is concerned: (i) it can combine only with a single SUBJECT, (ii) it can combine only with a coordination of SUBJECTS, (iii) it can combine either with a single SUBJECT or with a coordination of SUBJECTS. In the latter case, there is a possibility of structural ambiguity (cf. chapter 11 below). Further1nore, the two different applications of the coordination-schema are not restricted to the SUBJECT, but can in principle apply to any other function. This approach thus has the advantage that by a quite generally applicable structural principle a great variety of seerningly disparate facts is accounted for. As to the first point, I believe that the impossibility of •John met as against the possibility of John came shows that the difference between such verbs as met and came indeed resides in the different manner in which they combine with the SUBJECT. Should we attribute the difference between (15) and (19) solely to the verbs as such, then in the first place we would lose the generality of the solution proposed here, and in the second place we would be forced to find other means for differentiating properly between the different SUBJECT-selections of the verbs involved. I believe that these arguments are sufficient to justify the present approach. In any case it seems to lead to a general method of describing

208

COORDINATIONS IN A FUNCTIONAL GRAMMAR

coordinations by which the proper differentiation of functions is achieved. This method involves a single general coordination-schema and two different conventions for applying this schema. Correspondingly, we recognize two basically different types of coordinations: coordinations which are themselves part of some functional pattern, i.e., in which one of the functions of a functional pattern is coordinated, such as in (16), and coordinations which are not part of a functional pattern, but constitute a functional pattern by themselves directly do1ninated by some category. In the latter case, the coordinated functions are necessarily MEMBERS. I do not wish to suggest that this method is applicable in every detail to any particular natural language. There are evidently quite a few restrictions on its application. In part, these may be universal restrictions which can be added directly as conditions on the universal rule. In part, they will be general or particular restrictions which must be for1nulated for certain groups of languages or for certain particular languages. 10.3. Some examples

I shall now demonstrate by some examples to what extent the descriptive framework suggested here may be expected to achieve empirical adequacy. Consider the sentences: (24)

a. b. c. d.

John, John, John, John,

Bill, Bill, Bill, Bill,

and and and and

Richard Richard Richard Richard

are a happy trio. carried a stone together. each carried a stone. carried a stone.

In (24a), it is said of John, Bill, and Richard that they constitute a happy trio. There is, therefore, one SUBJECT, which is fulfilled by a coordination of proper nouns. In the description, SUBJECT is specified as npP1, which in turn is specified as MEMBER MEMBER & MEMBER, the final result being John, Bill, and Richard. 'To be a happy trio' simply cannot be said of three persons taken separately. In (24b), the same SUBJECT is combined with carried a stone together. Here it is together which explicitly indicates that there cannot but be a single SUBJECT. In (24c), however, there are three coordinated SUBJECTS, fulfilled by John, Bill, and Richard separately. The presence of each serves to accentuate this fact.

10.3. SOME EXAMPLES

209

In (24d), finally, there are evidently two possibilities, according to whether it is taken as near-equivalent to (24c), or as near-equivalent to (24b). This ambiguity of (24d) is accounted for in the grammar, which assigns two different structural descriptions to this sequence of words: (25)

s SUBJ

SUBJ

&

SUBJ

co

PRED

OBJECT

v

np

~

John

and

Bill

Richard

carried

a stone



(26)

s SUBJECT

MEMBER

John

MEMBER

&

MEMBER

Bill

and

Richard

PRED

OBJECT

v

np

carried

a stone

Next, consider the sentence: (27) John and Bill are carrying four stones . .

Again, sentence (27) is ambiguous in the same way as (24d). John and Bill may each carry four stones, or they may together carry four stones. This is accounted for in the grammar. Notice, however, that there are further situations to which (27) can be applied: (a) John may carry one

210

COORDINATIONS IN A FUNCTIONAL GRAMMAR

stone and Bill three, (b) John two and Bill two, or (c) John three and Bill one. Should not, one might ask, these ambiguities also be accounted for in the grammatical description? In my opinion this is not the case. Indeed, with sentence (27) we can talk about each of these situations, but the exact distribution of the four stones over John and Bill is in no way linguistically indicated in (27). No gra1n1natical difference can be connected with this distribution, nor is (27) semantically different if applied to each of these situations. The meaning of/our is not 'one and one and one and one', 'two and two', 'one and three', but it is 'four'. The different uses of (27) for situations (a), (b), and (c) is a mere matter of reference. The linguistically identical sentence is in each case applied to different 3 extra-linguistic situations. 4

For a further illustration, consider the sentences : (28) (29) (30) (31)

Ice-cream Ice-cream Ice-cream Ice-cream

and and and and

cake are favourite desserts of 1nine. cake are my favourite desserts. cake are my favourite dessert. cake is my favourite dessert.

These sentences can be paraphrased as follows: (28)* 'Ice-cream is a favourite dessert of mine and cake is a favourite dessert of mine'. (29)* 'I have just two favourite desserts: ice-cream and cake'. (30)* 'I have just one favourite dessert, which is composed of two things: ice-cream and cake'. (31)* 'I have just one favourite dessert, which is called 'ice-cream and cake''. Only in the case of sentence (28) can we say that ice-cream and cake both function as separate SUBJECTS. The structure is, therefore, roughly as 5 follows: 8

For a treatment of similar cases, cf. Uhlenbeck 1967: 271-2. For a more precise distinction between grammatical, semantic, and referential (extra-linguistic) phenomena, cf. below sections 12 . 2 . 2. ff. ' The examples are taken, in part, from Fidelholtz 1964. 5 I use the ter1n PREDICATIVE COMPLEMENT (PRED COMPL) for the f11nction of the last term in such patterns as SUBJECT is/are . As always, the distinction of this function is more essential than its particular designation.

10.3. SOME EXAMPLES

211

(32)

s SUBJ

&

SUBJ

n

co

n

ice-cream

and

cake



PRED

PRED COMPL np

are

favourite desserts of mine

The same structure cannot be assigned, however, to any of the sentences (29), (30), or (31 ). None of these is equivalent to something like: (33)

Ice-cream is my favourite dessert and cake is my favourite dessert.

If this is not i1n1nediately clear, compare a sentence like: (34) John and Richard are my best friends. which is not equivalent to: (35) John is my best friend and Richard is my best friend. It is even questionable whether sentences (33) and (35) make any sense under normal conditions, whereas sentences (29) and (34) are perfectly all right. Si1nilarly, we can say: (36) Amsterdam is the largest city in the Netherlands. (37) Amsterdam and Rotterdam are the largest cities in the Netherlands. but not: (38) •Amsterdam is the largest city in the Netherlands and Rotterdam is the largest city in the Netherlands. We have thus found an interesting group of expressions like 'are my favourite desserts', 'are my best friends', 'are the largest cities in the Netherlands', which can only be connected with a single SUBJECT and

212

COORDINATIONS IN A FUNCI10NAL GRAMMAR

not with coordinated SUBJECTS as in (28); this single SUBJECT must either be a plural noun phrase such as in: (39) These boys are my best friends.

or a coordination of nouns or noun phrases, as in (29), (34), and (37). We conclude that in (29), (30), and (31) there is in each case a single SUBJECT, and that the coordination ice-cream and cake occupies some level within this SUBJECT. The basic structure of these sentences is thus like: (40)

s SUBJ

PRED

PRED COMPL

np The problem is, now: how can (29), (30), and (31) be properly differentiated to account for their grarnrnatical differences, manifested in plural vs. singular finite verb, and plural vs. singular noun phrase as a PREDICATIVE COMPLEMENT? These differences can be accounted for as follows. The general rule is that a coordination of nouns or noun phrases counts as plural for the rest of the sentence. Thus, the congruence with a plural finite verb are is as expected. A further general rule of English says that in the context: (41)

SUBJECT( ... pturai) are PRED COMPL(n, np)

the noun or noun phrase fulfilling the PRED COMPL is also plural, unless it has a collective aspect. Sentence (29) conforrns to both these rules and can accordingly be described as represented in:

10.3. SOME EXAMPLES

213

(42)

s SUBJ

MEMBER

&

MEMBER

n

co

n



ice-cream

and

cake -

PRED

PREDCOMPL

are

my favourite desserts

If, however, the noun or noun phrase fulfilling the function of PREDICA'l1VE COMPLEMENT has a 'collective' aspect, it must be singular. Compare: (43) a. John, Bill, and Harry are a happy trio. b. These boys are a good team. c. They are a very happy family. etc. We can say, therefore, that in (30) my favourite dessert is a collective noun phrase functioning as a PREDICATIVE COMPLEMENT. The description would, accordingly, look like:

214

COORDINATIONS IN A FUNCTIONAL GRAMMAR

(44)

s .PRED

SUBJ

MEMBER n •

ice-cream

& co and

MEMBER

PREDCOMPL

npcoll II

n

cake

'

are

my favourite dessert

In sentence (31), finally, the singular is indicates that the noun or noun phrase fulfilling the function of SUBJECT is here treated as singular. Apparently, the peculiar combination ice-cream and cake, though as a coordination of nouns normally plural, here functions as a singular noun: it has the character of a unified idiom with properties foreign to regular coordinations. The singular number of the noun phrase functioning as PRED COMPL is as expected. The description is as represented in (45): (45)

s

PRED

SUBJ

PRED COMPL

DPs1

MEMBER n

&

MEMBER

co

n '

I



ice-cream

and

cake



IS

my favourite dessert

•'

I

I

10.4. SOME FURTHER DIFFERENCES

215

The gra1n1nar thus involves the following congruence-relations (to be described by means of context-sensitive rules): coordination (i) of SUBJECTS

. • plural firute verb • plural PRED COMPL

__., sg._PRED COMPL if (ii) SUBJECT(plural) > plural finite verb it is 'collective' __.,plural PRED COMPL (iii) SUBJECT(sing.) -+ sg. finite verb • sg. PRED COMPL Of course, as elsewhere, the details of the description may require modification in the light of further elaboration of the grammar. What I want to demonstrate here is the fact that in a functional grammar a great many quite subtle grammatical differences can be accounted for in a quite satisfactory manner. In chapter 5 I listed a number of examples which could not be transformationally described (chapter 5, examples (43)). As a careful consideration will show, all these examples can be described in a functional grammar along the lines indicated above. We have, therefore, also accounted for the difference between so-called 'sentence-conjunction' and 'phrasal conjunction' (cf. above, section 5 . 7 . I.). In a functional grammar, this difference is taken care of by the fact that in some cases the functions within a functional pattern are themselves coordinated, while in other cases the coordination appears at some lower level within a certain function. The generality of the description is achieved in two ways: (i) any type of coordination is deter1nined by the same coordination-schema, and (ii) this schema is, in principle, associated with any function, as well as with any (sub-)category. In other words, for any function and for any (sub-) category, there is, in principle, an optional entry into the coordinationschema, which need not be stated in the grammar more than once. Any function and any (sub-)category for which this general rule does not hold can be indicated as such in the grammar.

10. 4. Some further differences with the transformational description ..

In a publication prior to the development of transformational generative gra1nmar (1955a: 40), Chomsky noted the different logical properties of the 'syntactically identical' (his terms) sentences:

216

COORDINATIONS IN A FUNCTIONAL GRAMMAR

(46) The man is tall and thin. (47) The flag is black and white. From (46), we can infer: (48) The man is tall. whereas from (47) we cannot infer:

(49)

The flag is black.

Of course, Chomsky would not now hold that (46) and (47) are syntac6 tically identical. It is instructive, therefore, to consider how their linguistic difference would be accounted for in the transfor•national approach, and to contrast this with the manner it could be dealt with in a functional gra1n1nar. In a transfor1national description, (46) would be described as a case of sentence-conjunction, and derived from the structure underlying: (50)

The man is tall and the man is thin.

(47), however, could not be derived from the structure underlying (51)

The flag is black and the flag is white.

It would therefore have to be regarded as a case of phrasal conjunction, and be directly generated along the lines of:

(52)

The flag is Adj+and+Adj.

According to this description, (46) and (47) would be identical in surface structure, but different in deep structure. Alternatively, (47) could be based on something like:

(53)



The flag is Adj.

Cf. Ducrot 1966b: 136-7 on this point.

10.4. SOME FURTHER DIFFERENCES

217

where black and white should be specified as an adjective in the lexicon. If this course were followed, (46) and (47) would also be different in surface structure. As to the derivation of (46), this is of course subject to all the criticisms listed in chapter 5. Notice further that if (47) is described along the lines of (52), then clearly a rule-schema applying to Adj is needed, which again demonstrates the i1npossibility of restricting phrasal conjunction to NPs (cf. above, section 5.7.3.). If (47) is described along the lines of (53), this will result in a proliferation of adjectives in the lexicon, such as:

(54) a. black and white b. red, white, and blue c. red, white, blue, and orange, etc. In fact, any combination of two or more coordinated colour adjectives would in principle itself have to be entered in the lexicon as an adjective. In that case, the rules underlying their f orrnation would not be explicitated and the description of (47) would accordingly be inadequate. Next, consider the sentence: (55) The cats are black and white. Clearly, this sentence is ambiguous, one of its possible interpretations being congruent with (46), and the other with (47). In the first case, the content is, roughly: 'Some of the cats are completely black, and some of the cats are completely white'. In the second case, it is: 'Each of the cats has a partly black and partly white skin'. Observe, however, that in the first case we cannot subject (55) to a transfor111ational description similar to that of (46). I.e., we cannot derive (55) from the structure underlying: (56) The cats are black and the cats are white. since (56) would be most naturally interpreted as: 'Of two different groups of cats, those belonging to the first group are all black, and those belonging to the second group are all white.' One could only derive (55) transforrnationally from a structure underlying something like:

218

COORDINATIONS IN A FUNCTIONAL GRAMMAR

(57)

Some of the cats are black and some of the cats are white.

It is not clear, however, how such a derivation could be brought about. Let us now turn to the question of how these various facts can be accounted for in a functional grammar. In such a framework it can be stated that sentences (46), (47), and (55) are built on the same basic functional pattern, consisting of SUBJ + PRED + PRED COMPL. The difference between (46) and (47) is that in (46) two PREDICATIVE COMPLEMENTS are coordinated, whereas in (47) there is a coordination within the PRED COMPL. The structures of (46) and (47) are as represented in (58) and (59), respectively:

(58)

s SUBJ

PRED

PRED COMPL

&

PRED COMPL

a~

co

a~

tall

and

thin

np

the man

,



IS

(59)

SUBJ

PRED

PRED COMPL

np

the flag

adjp



lS

MEMBER

&

MEMBER

adj

co

adj

black

and

white

10.4. SOME FURTHER DIFFERENCES

219

The ambiguity of (55) lies in the fact that this sequence represents two 7 sentences, one with a structure analogous to (58), and the other with a structure analogous to (59). . There is, however, a further difference in such sentences as (55), as can ·be seen from: (60) The men are tall and thin. This can either be interpreted as 'There is a group of men each of whom is both tall and thin' (analogous to (46)), or as 'Among a group of men, some are tall and some are thin' (analogous to (55) in one of its interpretations). In the first case both PREDICATIVE COMPLEMENTS apply to each of the members of the SUBJECT. In the second case one of the PREDICATIVE COMPLEMENTS applies to some members of the SUBJECT, whereas the other PRED COMPL applies to the rest of the members of the SUBJECT. I believe, however, that this latter difference is not syntactically based, i.e., that in a structure of type (58) with a plural SUBJECT there is no linguistic indication of the scope of the PRED COMPL in relation to the SUBJECT. The PREDICATIVE COMPLEMENTS apply to the SUBJECT, but it is not indicated whether each of them applies to all or to some members of the SUBJECT. The determina8 tion of this scope is apparently left to the final interpretation of the sentences involved, i.e., depends on factors not given in the sentence as such. In case we want to clarify the exact relations intended, we must choose different constructions in which these relations are explicitly indicated by linguistic means. This case is comparable to that of sentence (27) discussed above. Finally, we can in the present scheme also account for certain phenomena noted by Ducrot {1966b: 137). We can say: (61) The flag is black and white. (62) The flag is black, white, and yellow. (63) The flag is black and white and frayed. 7

I use the term 'sentence' for strings of words plus their grammatical structures. Cf. below, section 11 . 2. 8 For this term, see section 12. 2. 2.

220

COORDINATIONS IN A FUNCTIONAL GRAMMAR

but not: (64)

*The flag is black, white, and frayed.

The fact that the and in (63) must be retained can be easily explained if it is recognized that we are not dealing with one, but with two coordinations on different levels, the complete structure of (63) being:

(65)

s SUBJ

PRED

PRED COMPL

&

PRED COMPL

adjp

co

adj

np

MEMBER & MEMBER

the flag



lS

adj

co

adj

black

and

white

and

frayed

A further illustration of the difference between a functional and a transformational description can be given apropos of certain interesting suggestions made by Kraak & Klooster (1968). They point to the fact that the occurrence of certain clause adverbials correlates with the difference between sentence conjunction and phrasal conjunction. Contrast the foil owing groups of sentences:

(66) a. John came and Peter came. b. c. d. (67) a. b. c. d.

John came and Peter came too. John came and Peter too. . John and Peter came. *John met and Peter met. *John met and Peter met too. *John met and Peter too. John and Peter met.

10.4. SOME FURTHER DIFFERENCES

221

Such words as too, they argue, as genuine clause or sentence adverbials, generally require the presence of a SUBJECT-PREDICATE combination to be used as modifiers. In a sentence like (66c), such a combination does not seem to be present, since we only have the SUBJECT Peter. But too cannot simply modify Peter as such. This provides an argument, therefore, for deriving (66c) from the structure underlying (66b), i.e. from a structure in which Peter is accompanied by a PREDICATE came. I fully agree with these observations, but obviously I cannot accept the (transformational) conclusion. The question is, therefore, how these facts can be accounted for in a functional description. The answer is that, as we have seen above, in the sentences (66a)-(66d) both John and Peter are SUBJECTS, whereas in (67d) the coordination John and Peter constitutes a single SUBJECT. In (66a)-(66d), therefore, John and Peter each contract a functional relation with the PREDICATOR came, no matter whether this word is present twice or only once. Thus, the sentence: ( 68) John came and Peter came could be described as represented in:

(69)

s MEMBER

&

MEMBER

clause

co

clause

SUBJ

SUBJ

PRED

PRED

v

v John

came

and

Peter

Sentence (67d), however, would be described as represented in:

came

222

COORDINATIONS IN A FUNCTIONAL GRAMMAR

(70)

s •

MEMBER

SUBJ

PRED

np

v

&

MEMBER

co John

and

came

Peter

And the sentence: (71) John came, and Peter.

as represented in:

9

(72)

s ~

SUBJ

John

PRED

&

v

co

came

and

SUBJ

Peter

The occurrence of such adverbials as too can now be made to depend on the presence of a SUBJECT-PREDICATOR relation by means of context-sensitive rules. We can then explain in terms of descriptions (69), (70), and (72) why we can have:

' For such descriptions as (72) the coordinative pattern must be specified discontinuously. Cf. section 10. 1.

10.5. SOME ADVANTAGES

(73) a. b. c. d. e.

223

John caane and Peter came too. John came too and Peter came too. John and Peter met too. John came and Peter too. John came too and Peter too.

but not: (74) a. b. c. d.

*John met and Peter met too. *John met too and Peter met too. *John met and Peter too. *John met too and Peter too.

Sentences (74a)-{74d) are sianply excluded because there is no SUBJECTPREDICATOR relation between John and met and Peter and met, but only between John and Peter and met. The facts mentioned by Kraak & K.looster can thus be taken care of in a functional gra1n1nar.

10. 5. Some advantages of the functional description of coordinations It will have become evident that the functional description of coordinations has several i111portant advantages over the transformational approach. (i) Although not as yet completely f oranalized, it seems to yield a decidedly si1npler total description. (ii) There is no problem of 'derived constituent structure', since no transfor1nations are involved. (iii) The problem of referential identity does not arise. In a sentence like: (75) The man is tall and thin. the PREDICATIVE COMPLEMENTS tall and thin are directly assigned to the same SUBJECT The man. And this single SUBJECT The man must necessarily refer to one individual. (iv) There is no difference between 'surface structure' and 'deep structure' in a functional grammar. The grammar assigns to any linguistic expression a structural description in which all its grammatically relevant properties are specified, and which is sufficient to act as a basis for the

224

COORDINATIONS IN A FUNCI'IONAL GRAMMAR 10

rules which assign a semantic content to the expression concemed. (v) The functional description is generally applicable to any kind of coordination. But since the coordination-schema can apply at different levels, the proper differentiation between 'reducible' and 'irreducible' coordinations is immediately indicated. We can, on the basis of the functional description, formulate an important generalization accounting for these differences. 11 Suppose we have a structure where in one functional pattern we find: (76) SUBJECT & SUBJECT +PREDICATOR a functional pattern characterizing, e.g., a sentence like: (77) John and Peter came. We then know i1nmediately that each of the SUBJECTS taken separately contracts a functional relation with the PREDICATOR. 'l'his means that for each such structure we may also expect equivalent structures with the functional pattern: (78) SUBJECT+PREDICATOR such as: (79) a. John came. b. Peter came. We can generalize this as follows. For any functional pattern of type:

we have equivalent functional patterns:

10

On the relation between grammar and semantics, see below section 12. 6. 11 Observe that 'functional pattern' is defined as a sequence of functions dominated by the same categorial node (sections 9. 2. 5., 9. 2. 6.).

10.5. SOME ADVANTAGES

225

(81) a. F 1 + F1: b. F 2 +F1: • • •

In my opinion, this generalization constitutes a meta-grammatical rule formulating a regularity observed in the grammatical description, but not figuring as a grammatical rule in the description as such. In other words, we can, in the general theory of language, formulate certain rules describing regularities observed in grammatical descriptions. These are not themselves rules of the grammar of any language; for this reason, I use the ter1n 'meta-grammatical'. I may here add a remark about the word respectively, which has a property connected with the general rule given above. Consider a sentence like:

(82) John and Bill love Mary and Jane. This sentence may be paraphrased as: (83) 'John loves Mary, John loves Jane, Bill loves Mary, and Bill loves Jane'. This is because the functional pattern of this sentence is:

and, according to the general rule given above, the significant relations are:

(85) a. b. c. d.

SUBJ 1 +PRED+OBJ 1 SUBJ 1 +PRED+OBJ 2 SUBJ 2 +PRED+OBJ 1 SUBJ 2 +PRED+OBJ 2

But now consider: (86) John and Bill love Mary and Jane, respectively.

.I I

226

COORDINATIONS IN A FUNCTIONAL GRAMMAR

This can only be interpreted as paraphrased in: (87) 'John loves Mary and Bill loves Jane'. The word respectively thus has the effect of excluding the relations (85b) and (85d) given above. It restricts the establishment of functional relations in constructions of the general type of (82). We can express this quite generally as follows: in a functional pattern of the general type: (88)

F 1 &F 2 & ... &F,.+( ... )+F' 1 &F' 2 & ... &F',.

the presence of respectively restricts the possible functional relationships to the pairs (F 1 , F' 1), (F2 , F' 2 ), ••. , (F,., F',.). Again, this interesting phenomenon can not so easily be accounted for in a non-functional description.

'l



11. Structural ambiguity in coordinations •

In the preceding chapter I have already had occasion to draw attention to certain cases of structural ambiguity (or constructional homony1nity). In fact, the coordinative construction is a source of several types of structural ambiguity. Let us now consider this matter in a more systematic 1 way. I speak of structural ambiguity if and only if identical sequences of ultimate constituents as represented in phone1nic form (including intonation) correspond to different grammatical descriptions (cf. Chomsky 1957a: 28, 86). 2 Structural ambiguity was perhaps first clearly indicated by Bloomfield, when he wrote: 'Even simple f or1nations may lead to ambiguity because the scope that is, the accompanying constituents on the proper level - of a form are not marked. For instance, 'an apple and a pear or a peach' may mean exactly two pieces of fruit: then the i1nmediate constituents are 'an apple / and / a pear or a peach', and the phrase 'a pear or a peach' and the phrase 'an apple' constitute the scope of the form 'and'. On the other hand, the· phrase may mean either two pieces of fruit or one piece: then the immediate constituents are 'an apple and a pear/ or/ a peach', and the scope of 'and' now consists, on its level, of the phrases 'an apple', 'a pear'.' (1939: 30). The idea that ambiguities of this type are connected with different ICanalyses has been further developed by Wells, whose example old men and women (1947: 93ff.) has since played an i1nportant part in theoretical discussions of this topic. This phrase may either be equivalent in meaning For a full-length treatment of ambiguity, see Kooij (forthcoming). • Notice that Aristotle already has an interesting observation on this point. One might say, he argues, that if 'Five is two and three', then it is both even and odd, and the greater is equal (i.e., if five is two and five is three, then two is three; De Sophisticis Elenchis 166 a 33-35). According to Aristotle, this conclusion would be based on a faulty 'division'. In our terms one might say that in the expression concerned two and three do not each constitute independent PREDICATIVE COMPLEMENTS, but that the expression two and three as a whole is a single PREDICATIVE COMPLEMENT.

1

228

STRUCTURAL AMBIGUITY IN COORDINATIONS

to 'women and old men', or to 'old men and old women'. Corresponding to this meaning-difference there is apparently a difference in construction, which Wells and many others thought to be representable in terms of ICanalysis. In the first interpretation one should, according to this view, 3 analyse the expression as:

(I)

(old men)

(and)

(women)

In the second interpretation, it should be analysed as:

(2)

(old)

(men and women)

In this chapter, I shall first distinguish between the different types of coordinative ambiguity of which the examples of Bloomfield and Wells are representatives. I shall also show that the analysis proposed by Wells and others has to be modified in the light of what we have learned about coordination in the preceding chapters. This will lead, among other things, to a reconsideration of the restrictions inherent in IC-analysis, and to a demonstration of the superiority of functional grammar to any system exclusively based on constituency and categorization.

1 I . 1. Three types of ambiguity

There are three basic sources of structural ambiguity in coordinations. We have already encountered the first in the preceding chapter (sections 10. 3. and 10.4.). We saw there that in an expression like: (3)

John, Bill, and Harry

For further discussion of this example, see Hockett 1954: 217ff, and cf. also Hockett 1961: 225-6. In a context similar to ours, but in an entirely different manner, it is also treated by Schane 1966: 16ff. See further Katz & Postal's discussion of I like little boys and girls (1964: 24), and Braun 1967: 84-7 on similar German examples. a

11.1. THREE TYPES OF AMBIGUITY

229

the three proper nouns may either each have the function of SUBJECT in the functional pattern SUBJECT+PREDICATOR+OBJECT, or be simply MEMBERS within the same SUBJECT. Sirnilar differences were found in the case of ice-cream and cake and, within the PREDICATIVE COMPLEMENT, in expressions like black and white. We can term this type of ambiguity 'functional ambiguity', since in such cases the members of a coordination may have different grammatical functions within the total grammatical structure. The second source of ambiguity in coordinations lies in possible differences in their internal hierarchical organization. I shall tern1 this type 'hierarchical ambiguity'. The third source lies in the fact that certain constituents figuring in a linguistic expression containing a coordination may either be included in one of the coordinated members, or stand outside the coordination as such, either modifying each of its members separately, or the unit constituted by the coordination as a whole. I shall tern1 this latter type 'relational ambiguity', since in this case there is uncertainty about the relation of a certain constituent (or of certain constituents) to the members of the coordination. Functional, hierarchical, and relational ambiguity may combine to produce expressions which are many-ways structurally ambiguous. It seems possible to formulate the conditions for these ambiguities in quite general ter1ns. The same applies to what we could call the 'neutralization' of ambiguity. By this term we refer to the fact that an expression which in itself is potentially x-ways ambiguous is still not open to all x possible interpretations, some of these being precluded by the context or situation in which the expression occurs, or by the co-occurrence restrictions holding between the elements figuring in it. It is even the regular case that just one interpretation is allowed in actual com111unication by these conditions, which explains why these structural ambiguities are tolerated within the linguistic system without in most cases leading to communica4 tive disturbances. ' This point has repeatedly been emphasized by Reichling (University lectures). It is imperative to distinguish carefully between two kinds of neutralization of ambiguity: one brought about by linguistic means (i.e., by contextual features) and the other one determined by extra-linguistic circumstances (knowledge of the situation and further extra-linguistic information). The former belongs properly to the system of language, while the latter is rather a feature of performance (for this distinction, see Kooij (forthcoming)).

230

STRUCTURAL AMBIGUITY IN COORDINATIONS

11 .1 .1. Functional ambiguity

As to functional ambiguity, I need not add very much to what we have observed in chapter 10. We can describe this phenomenon in general ter1ns as follows: given a functional pattern like:

we can have, in relation to, e.g., the function F 1 , two different types of coordination. In the first the function F 1 is itself coordinated to yield a pattern like:

In the second case the category which fulfils the function F 1 comprises a coordination, as represented in: (6)

+ c MEMBER & MEMBER & . . . & MEMBER If it is the case that the functions F 1 of (5) and the MEMBERS of (6) can be ultimately specified by the same ulti1nate constituents (the remainder of the linguistic expressions concerned being identical), then the linguistic expressions containing these constituents are functionally ambiguous as between (5) and (6). This is the case in examples (24d) and (55) of chapter 10: (7) John, Bill, and Richard carried a stone. (8) The cats are black and white. Of course, functional ambiguity automatically involves differences in hierarchical structure. The difference with what I have called 'hierarchical ambiguity' is that in the case of functional ambiguity the question is on what functional level the coordination as a whole is situated, whereas in

r

I

11.1. THREE TYPES OF AMBIGUITY

231

the case of hierarchical ambiguity, as we shall see below, the question is rather whether we are dealing with one coordination or with a system of different coordinations built into each other. It may again be emphasized that in some cases both types of ambiguity combine so that both the hierarchical build-up of the construction and the functional level of the whole construction offer different possibilities.

11 . 1 . 2. Hierarchical ambiguity Since in most cases the members of a coordination can themselves be coordinations, a sequence of several coordinated members will generally allow for different gra1n1natical descriptions, corresponding to the differ5 ent ways in which the members can be hierarchically organized. Thus, a sequence like:

can be organized in three different ways, as represented in the schemas:

(10)

a.

&

&

b.

&

&

c.

&

&

To borrow an example from Lakoff & Peters (1966: 21), an expression like: -·- - - 6 Cf. Antoine 1962: 719ff., G leitman 1965 : 27 5-8, Lakoff & Peters 1966: 21, Schane 1966: 6ff.

232

(11)

STRUCTURAL AMBIGUITY IN COORDINATIONS

Kosygin and Johnson and Rusk

can be taken as: (12) a. (Kosygin) and (Johnson) and (Rusk), or as: b. (Kosygin and Johnson) and (Rusk), or as: c. (Kosygin) and (Johnson and Rusk). As the number of coordinated members increases, the number of potential hierarchical ambiguities is correspondingly augmented, not only since more groupings at one structural level become possible, but also because more different levels can be involved. A general restriction on these possible structures is that any two or more coordinated members should be in uninterrupted sequence. It goes without saying, however, that most of the intricacies which are theoretically possible will seldom or never occur in actual speech. As far as the linking devices are concerned, the possibilities of hierarchical ambiguity are apparently unrestricted in the case of juxtaposition or in the presence of the unemphatic coordinator placed before or after each member except the first. But when the emphatic, repeated coordinator is used (especially when it is correlative) there are certain restrictions. It seems at least difficult to say: (13) Both John and both Mary and Peter. though it is possible to say: (14)

Both John and Mary and Peter.

which can, again, be taken in three different ways:

11.1. THREE TYPES OF AMBIGUITY

233

(15)

a.

Botlz John a11d Mary and Peter

I

I

J

b.

Both John and Mary and Peter

c.

Both John and Mary and Peter

We can further have: (16)

John and both Mary and Peter.

which is unambiguous, and also: (17)

Either John or both Mary and Peter.

We can account for these facts by stating that an emphatic both ... andconstruction cannot be included in another such construction, though it can be included in another emphatic coordination or in an unemphatic coordination. This rule is not restricted to English (the same situation obtains, e.g., in Dutch) and may perhaps allow of a significant generaliza• t1on. There are restrictions, too, if the unemphatic coordinator is placed only once (before or after the last member of the coordination). An expression like: (18)

John, Paul, Peter, and Jack.

is most naturally interpreted as:

234

STRUCTURAL AMBIGUITY IN COORDINATIONS

(19)

John, Paul, Peter, and Jack I J I J

though it could also be taken as : (20) John, Paul, Peter, and Jack

or as: (21)

John, Paul, Peter and Jack

This depends, of course, on whether coordination by juxtaposition is possible in a certain language, and on the relation between juxtaposition and explicit coordination in that language. Finally, there is the case of different coordinators appearing in a coordinated sequence. Here certain possibilities are excluded by a rule which again seems to be of quite general applicability, to the effect that two different coordinators which are not correlative cannot appear on the 6 same structural level in a single coordination. This is why Bloomfield's example: (22) An apple and a pear or a peach. can only be interpreted in the two ways indicated in the beginning of this chapter, and not as: (23)

An apple and a pear or a peach

6

For some exceptions to this rule, see section 4. 1. 4. 2. 3.

11 . 1. THREE 'l'YPES 0 F AMBIGUITY

235

whereas the latter interpretation is possible for: (24) a. An apple and a pear and a peach. b. An apple or a pear or a peach. If a language has different, but (near-)synonymous coordinators, these 7 can, therefore, have the effect of neutralizing hierarchical ambuity. The reverse of this rule is that parts of the emphatic correlative coordinators like both ... and, either ... or, which condition each other, cannot but be at the same level of structure. I.e., if a coordination begins with both ... and only one and follows, then this coordination is unambiguous as far as this relation is concerned, no matter what other coordinations are present. This explains why the expressions (16) and (17) are unequivocal, whereas (14), in which two instances of andfoil ow both, is not. This fact was noted by Quine (1961: 1), who contrasts the ambiguous expressions (25) a. p and q or r b. p or q and r with the unambiguous: (26) a. b. c. d.

either p and q or r. p and either q or r. both p or q and r. p or both q and r.

Quine aptly likens the function (we would rather say one of the functions)8 of both and either to the left-hand parenthesis in a mathematical or logical formula. Indeed, these elements are signs that a coordination of a certain type follows, just as the coordinator in a series of type M 1 M 2 M 3 ••• co Mn is a sign that the coordination in question is going to be ter1ninated. Hierarchical ambiguity can also be neutralized by an intonation-pattern indicating the intended subgrouping (cf. the treatment in Gleitman 1965: 1

For some examples from Ancient Greek, cf. Ruijgh (forthcoming). 8 For comments on the differences between natural languages and formal systems in this respect, cf. Kraak 1966: 168ff.

236

STRUCTURAL AMBIGUITY IN COORDINATIONS

275-8). It seems difficult to determine, however, to what extent such into9 nations are really crucial to an understanding of the structures involved. In actual speech, finally, the exact determination of the subgroupings involved may be quite irrelevant. In certain cases, the speaker may have no clear intentions in this respect. In other cases, the hearer may not be able to reconstruct the speaker's intention. Neither of these situations, however, need have any detrimental effect on the establishment of communication.

11 .1 .3. Relational ambiguity

Less trivial than the preceding cases is the type of ambiguity of which Wells' example (cf. (I) and (2) above) is a representative. The conditions for the occurrence of this type may be formulated in general terms as follows. There is relational ambiguity with respect to a coordination if there is a constituent (or a group of constituents) which may either contract a grammatical relation with a proper subset of the coordinated members, or with all coordinated members separately, or with the unit constituted by the coordination as a whole. I can most easily demonstrate this situation by the following example: (27)

I want a cheap bed and breakfast.

This string of words can be interpreted in three different ways, viz. as: (28) a. 'I want a bed which is cheap, and a breakfast (which need not be cheap)'. b. 'I want a bed which is cheap and a breakfast which is cheap'. c. 'I want something called a 'bed and breakfast', and it must be 10 cheap'.

In (28a), both bed and breakfast have their own independent position in the sentence-structure, and only bed (as HEAD) is modified by cheap: • Cf. Kooij (forthcoming) on this point. Intonation-patterns are in many cases contingent features of linguistic expressions, i.e., features which may, but need not accompany them in all cases (Reichling, University lectures). 1 ° Cf. Schane's remarks on bread and butter, ham and eggs, pencil and paper. (1966: 5 fn. 2).

11.1. THREE TYPES OF AMBIGUITY

237

(29)

a cheap bed and breakfast

In (28b), bed and break/ast are also independent, but both are modified by cheap: (30)

1

I

a cheap bed and breakfast

In (28c), bed and breakfast constitutes a structural unit which as a whole is modified by cheap: (31) .

.

_

_

_

_

,

~

a cheap bed and breakfast . I___

We can now see why Wells' IC-description of old men and women is not satisfactory. In an IC-framework, one does not have the possibility of indicating a double relationship such as occurs in (30). Old men and women can, in such a framework, only be analysed according to (29) and (31). But if the interpretation is 'old men and old women', then it is not correct to act as if the whole combination men and women is modified by old (as is the case in (31)). Rather, men and women are each modified separately by old (as in (30)). In short, whereas IC-analysis can differentiate between (29) and (31), it cannot describe (30) in any way. This is in itself enough to show that IC-analysis can never on its own provide a satisfactory basis for a theory of syntax. I shall return to this point below, after having noted some further complications of relational ambiguity. I shall also show how these phenomena can be accounted for in a functional approach. Of course, coordinations such as (27), which allow of all three possibilities, are relatively rare. The group men and women does not have such internal cohesion as to be capable of modification as a whole by some

238

STRUCTURAL AMBIGUITY IN COORDINATIONS

kind of modifier. This impossibility, however, can in no way affect the basic distinction of three different possible structures for (27). In fact, it is only this distinction which shows that the structure of old men and women, in its second interpretation, is not equivalent to (31), but to (30). In general, therefore, given a string like:

where C is a constituent and M 1-Mn are members of a coordination, relational ambiguity resides in the fact that C can either be related to the first i members (0 < i < n), or to all the members separately, or that it can be related to the totality of the coordination. Even a casual analysis of some spoken or written text shows that the occurrence of coordinations which are potentially ambiguous in this way is by no means rare. In most instances, however, they will pass t1nnoticed since, as I have already emphasized, co-occurrence relations, context, and situation are usually such as to exclude all but one of the possible interpretations. In other cases it will be quite i1n1naterial which particular interpretation is arrived at. Some examples: for one of his articles Uhlenbeck intentionally chose the ambiguous title 'Traditionele zinsontleding en syntaxis' ('Traditional sentence analysis and syntax'), though the opposite of traditional syntax 11 is meant. Compare Bar-Hillel's title 'Logical syntax and semantics', which leaves us in doubt whether 'logical semantics' or 'semantics' tout court is intended. In the quotation from Hockett given above (section 4.1.4.2.6.) a notion of 'structural marker or signal' is introduced. Of course, a 'structural signal' is meant, though this is not explicitly indicated in this expression. In the same way John and Betty Smith may, but need not refer to a 'John Smith'. Relational ambiguity was already noticed by Bliimel (1914: 186). His example is: (33)

11

Dann erhob er sich und ging langsam fort.

Cf. De Groot 1962: 98.

11.1. THREE TYPES OF AMBIGUITY

239

in which dann may either be related to erhob er sich alone, or to erhob er sich and ging langsamfort. Some further random examples: Joyce's vast wealth of verbal sound and association. (Time 24.3. 1967: 4). (35) improvements in efficiency and agressive merchandising. (ibid. Cl). (36) Pity the British theater, and TIME for resorting to the Redgrave menage for a cover story. (ibid. 5; here the comma indicates that for resorting . . . does not go with the British theater). (37) Alfred achete des livres et des cahiers neufs. (Tesniere 1959: 340).

(34)

The same type of ambiguity may be seen to be characteristic of the sentences (10), (13), (15), (16), and (17) discussed in chapter 7 above. Someti1nes this type of ambiguity leads to co1nic results, as in this Dutch sentence from a child's composition: (38)

We hebben eerst de dieren bekeken en gegeten. lit. 'We have first the animals (OBJ) looked at and eaten' = 'We first looked at the animals and then had a meal'. Both bekeken and gegeten can take de dieren as OBJECT. But gegeten is of course meant as intransitive. Considerable complications can arise if several constituents can be correlated to the coordination in various ways. This is the case in the title of an article by Lakoff, which runs: (39)

Stative adjectives and verbs in English.

Either the adjectives or both the adjectives and the verbs can be stative. And either the verbs or both the verbs and the adjectives can be in English. Therefore, the following interpretations are (in abstracto) possible:

(40) a. Stative adjectives in English and stative verbs in English. b. Stative adjectives and stative verbs in English. verbs in English. c. Stative adjectives in English and d. Stative adjectives and verbs in English. More complicated still is:

240

(41)

STRUCTURAL AMBIGUITY IN COORDINATIONS

Nice old men and women from Amsterdam.

which can be taken as: (42) a. Nice old men from Amsterdam and nice old women from Amsterdam. b. Nice old men and nice old women from A1nsterdam. c. Nice old men from Amsterdam and old women from Amsterdam. d. Nice old men and old women from Amsterdam. e. Nice old men from Amsterdam and women from Amsterdam. f. Nice old men and women from Amsterdam. Similar examples are: (43) John read and wrote letters on Tuesday. (Schane 1966: 26). (44) General Motors men and women all over the world. (Time 24. 3. 1967: C 1). On the use of the type adj+noun+and+noun+prepositional phrase in Shakespeare one may also compare Empson 1961 [1930]: 89-101. Especially telling is the foil owing example (Othello I. iii. 249): (45) That I did love the Moor to live with him, my downright violence and storme offortunes may trumpet to the world. In the commentaries on this passage, several different relational interpretations of the second line may be found. If it is true that coordinations are among the structural features common to all languages, if they are indeed structured as described in a functional grammar, and if, as seems probabJe, all languages have constituents which

11.2. STRUCTURAL AMBIGUITY AND FUNCTIONAL GRAMMAR 241

can be related to the members of coordinations in various ways, then the conclusion is that functional, hierarchical, and relational ambiguity are also universal features of natural languages. In that case, we would have found a whole complex of structural features belonging to the com1non pattern on which all natural languages are based.

11. 2. Structural ambiguity and functional grammar

In the preceding sections I have infor1nally presented the different possibilities of structural ambiguity in coordinations. I take it as agreed on that the terrn structural ambiguity indeed applies to these cases, i.e., that with each possible content of the strings concerned, there corresponds a unique gram1natical structure which, however, does not show itself in the outward phonernic form of the string. We can also express this by saying that each such string covers, in fact, just as many different sentences as there are different structures and different corresponding contents. It is for this reason that a string of words like: (46) The old men and women arrived this morning. should not be called a 'sentence' unless it is associated with just one of its possible gra1n1natical structures. In transfor1national generative grammar, at the suggestion of Katz & Postal (1964: 24-5), a different ter1ninology has been adopted. They use the term 'sentence' for a mere string of formatives without its possible structural descriptions; and the ter1n 'sentoid' for a 'sentence' associated with one of its structural descriptions. I regard this ter1ninology as unfortunate, since it runs counter to the natural use of the term 'sentence', and assigns an artificial, derived ter1n to what is in fact the basic phenomenon, viz. the association of a grarnmatical structure with a phonemic form. It appears that this terminology was occasioned by the adoption of the restricted notion of 'derivation' which has been discussed above (section 9. 6.). Indeed, such derivations only result in 'strings of formatives'. But these are only part of the objects to be described by the gra1n1nar, as Katz & Postal rightly 12 remark in the passage referred to. •

11

Cf. Uhlenbeck 1967: 299-300. It seems that Chomsky now also accepts the view that a 'sentence' includes a grammatical structure. Cf. 1967: 406: 'let us consider each abstract ''sentence'' to be a specific pairing of a phonetic representation with an ab-

242

STRUCTURAL AMBIGUl'I'Y IN COORDINATIONS

We can indeed define structural ambiguity as occurring if different grarnrnatical structures fall together in identical phone1nic strings. I would now like to pay some more attention to the question of how these various structural ambiguities can be accounted for in a functional gramrnar. For functional ambiguity this has already been established in chapter 10 above. For hierarchical ambiguity, the matter is self-explanatory. This merely involves the repeated application of the coordinationschema at different levels within a structural description. I shall now concentrate, therefore, on the problems involved in the description of relational ambiguities, which are the most difficult cases for any theory of gra1n1natical description. To clarify the matter further, I shall first make some additional remarks on the inadequacy of IC-analysis in this respect.

11. 2 .1. The inadequacy of IC-analysis

We have seen that a structure like (30) above cannot possibly be accounted for in terms of IC-analysis. This is because this theory of grammatical structure is subject to certain heavy restrictions which prevent it from coping with the full range of gra1n1natical phenomena. The main point relevant here is that IC-description, though capable of indicating the relevant constituents at any level of structure, cannot account for 13 the particular relations holding between these constituents. IC-analysis is, in fact, a parsing-technique by which constituents are established, but which does not allow for the description of further relevant gra1nrnatical properties of these constituents. I can here refer to a recent attempt to remedy this inadequacy within the general framework of IC-theory (Street 1967: 102-4). In explicit

stract structure of some sort (let us call it a deep structure) that incorporates information relevant to semantic interpretation.' The adoption of this definition would mean that a 'sentence' can never be structurally ambiguous. Only a phonemic string in which several sentences accidentally coincide could be so. It may be that I have not interpreted this passage correctly, however, since a little further on in the same study Chomsky again speaks of one sentence having two structures (416). Constrast (419): ' ... an infinite number of pairs (D, S) of deep and surface structures ... , one corresponding to each interpreted sentence.' 13 On this point IC-analysis has been criticized by Reichling (1961, University lectures) and Uhlenbeck (1963, 1967).

11.2. STRUCTURAL AMBIGUITY AND FUNCTIONAL GRAMMAR 243

connection with structures of the type we are dealing with here, Street suggests two alternative devices: (i) The recognition of overlapping ICs (so far only applied in phonology by Hockett (1955)), such that one element may belong to two or more ICs si1nultaneously. Following this principle, 'the ICs of [He has] no special knowledge of, or interest in mathematics might be taken as no special knowledge of mathematics and no interest in mathematics, with or as a marker' (ibid. 102). It seems clear, however, that if this solution is explicitly formulated, some kind of deletion-transforn1ation will be needed to reduce the two occurrences of mathematics to one. The theory of IC-structure has then automatically become a version of transforrnational grammar. (ii) The assig111nent of 'double IC structure' to a certain constitute (= non-ultimate constituent). On this proposal, a sentence like: (47) John got a book from his aunt and a bicycle from his parents. would sirnultaneously have two IC-structures, viz. : (48) a. b. c. d.

14

John/ got a book from his aunt and a bicycle from his parents. John/ got a book ... a bicycle// from his aunt ... from his parents. John/ got///a book ... a bicycle //from his aunt III from his parents. John I got / 11 a book 111/ a bicycle II from his aunt /// from his parents. and:

(49) a. John/ got a book from his aunt and a bicycle from his parents. b. John I got// a book from his aunt and a bicycle from his parents. c. John / got /I a book from his aunt I// (and)/// a bicycle from his parents. This solution (which, it must be added, is presented by Street with some reservations), though interesting as an attempt to overcome the limitations of constituent-analysis in a non-transformational way, does not 1'

and is in both analyses regarded as a marker (cf. above section 4. 1 . 4. 2. 6.).

244

STRUCTURAL AMBIGUITY IN COORDINATIONS

seem to fit very well into the general framework of IC-analysis, nor is it clear how a sentence could be simultaneously provided with more than one structural description in any kind of explicit grarnrnatical statement. Street's suggestion breaks away from the universally accepted presupposition by which a unique structural description is thought to correspond to any sentence. Should his idea be feasible (which I think it is not), then one should at least have to establish the means to deterrnine how the two ICdescriptions interact, i.e., one should in some way be able to integrate them into one structural description. No such means is provided with Street's suggestion. It is to be noted that the li1nitations characterizing IC-description have been carried over into the structural descriptions of transformational grammar which has, in fact, incorporated a particular version of ICtheory. This means that in the P-markers (underlying or derived) of a transformational grammar such structures as (30) can in no way be rep1 resented. s Of course, the differences between (29), (30), and (31) are correlated with differences in description in a roundabout way, in that (29) is derived by transforn1ations from the structure underlying: ( 50)

I want a cheap bed and a breakfast.

whereas (30) is derived from a structure underlying: (51)

I want a cheap bed and a cheap breakfast.

and (31) from a structure underlying something like: (52)

I want a# the bed and breakfast is cheap # bed and breakfast.

The fact remains, however, that these differences in 'transfor1national history' cannot be brought into one-to-one correspondence with differences in derived constituent-structure. (29) would probably have a derived 16 P-marker like:

Cf. on this point also Schane 1966: 16ff. 1• I disregard features irrelevant to the present argument. See Katz & Postal 1964: 24 for structural descriptions equivalent to (53) and (54), assigned to I like little boys and

16

girls.

11.2. STRUCTURALAMBIGUITY ANDFUNCfIONALGRAMMAR 245

(53)

s NP

VP

v

ProN

NP and

NP

DET ADJ a

want

I

cheap

NP

N breakfast

bed

while (31) would have a derived structure like: (54)

s NP ProN

VP

v

NP DET

ADJ

NP

N I

want

a

cheap

bed

and

N

breakfast

There would be no means, however, to differentiate (30) from (29) and (31) in the final derived P-marker. This means that whereas the differences between (29), (30), and (31) would be accounted for by differences in transformational history, the final derived structure of (30) would arbitrarily coincide with either that of (29) or with that of (31). This holds for all similar cases. The conclusion is that one of the basic inadequacies

246

STRUCTURAL AMBIGUITY IN COORDINATIONS

of IC-theory has been carried over into the theory of transfor1national grammar.

11. 2. 2. The functional alternative On this point the superiority of a functional description to any type of constituent-analysis can again be demonstrated. Indeed, a constituentanalysis can account for the manner in which the elements of a linguistic construction are hierarchically ordered into larger wholes. If it is also provided with a categorial di111ension, it can further show to what class each of the constituents at any level belongs. But it has no means of further indicating relationships between (classes of) constituents, beyond • those relations which are inherent in the constituency and the categorization as such. A functional grammar, however, adds a further, functional dimension in assig11ing to each particular class of constituents the grammatical function which it fulfils in the structure as a whole. These functions, among other things, deterrnine the functional relationships between the constituents figuring in the linguistic expression concerned. Within a functional grammar the different types of coordinative ambiguity can be uniformly accounted for as resulting from differences in the scope of functional relations and from differences in the levels on which the coordinations are situated in the complete structural description. I shall demonstrate this by treating the examples discussed above along the lines of such a system. In the differences between (29), (30), and (31), both different functional scopes and different functional levels are involved. In (29), cheap is MODIFIER and bed is HEAD; in (30), cheap is MODIFIER and both bed and breakfast are HEADS; in (31), cheap is MODIFIER and bed and breakfast is HEAD. In (29), there is a coordination of a modified noun and a noun; in (30), there is a coordination of two modified nouns; 17 in (31) there is a coordination of two nouns within a modified HEAD. The three different linguistic expressions will therefore receive structural descriptions as represented in (55), (56), and (57), respectively. 17

Further complications arise if different possible scopes for the indefinite article are recognized. I shall not investigate this possibility here, since it does not add to the theoretically important structural differences.

11.2. STRUCTURAL AMBIGUITY AND FUNCTIONAL GRAMMAR 247

(55) np .

&

MEMBER

np

co

n

HEAD

MOD

DET art

a

MEMBER



adj

n

cheap

bed

breakfast

and

(56) np

DET

MOD .

HEAD

&

HEAD

art

adj

n

co

n

a

cheap

bed

and

breakfast

(57) n

DET

MOD

HEAD

art

adj

np

MEMBER & MEMBER n

a

I

cheap

bed

co

n

and breakfast

248

STRUCTURAL AMBIGUITY IN COORDINATIONS

Some further comments must be made about these structural descriptions. (i) In (55) there is a modification within the first term of a coordination; in (57) there is a coordination_within the last terrri of a modification; in (56) the modification and the coordination are at the same level within the noun phrase: the coordinated items are the HEADS of the MODIFIER. As we have observed earlier in the case of coordinated SUBJECTS and PREDICATIVE COMPLEMENTS, this is a general characteristic of those constructions which cause difficulties to a pure co~stituent­ description: in all these cases the coordination is a part of a certain functional pattern, such that each of the coordinated items on its own contracts functional relations with the remaining functions of this pattern. (ii) In (56) the whole combination a cheap bed and breakfast is described as a constituent directly consisting of the five i1n1nediate constituents a, cheap, bed, and, and breakfast. Such a multiple analysis is unco1n111on in a pure constituent structure system. Still, I believe that a consistent application of the principles of functional grammar justifies this description. If we have the combination a cheap bed as a unit in a sentence structure (e.g., as a SUBJECT or.an OBJECT), then this combination is obviously a unified constituent. But if and breakfast is added to this combination in such a way that the DETERMINER and the MODIFIER are also related to breakfast (as in (56)), then we cannot maintain that a cheap bed is a unit within the complete combination a cheap bed and breakfast, since the basic functional pattern of this combination is an extended one on account of the coordination. Furthermore, no other intermediate constitue11ts can be established within the combination as a whole in this case. Therefore, a direct analysis into five constituents seems the only adequate one. (iii) In the description sketched above the appropriate differentiation between (29), (30), and (31) is achieved, while at the same time the functional relations characterizing them are indicated. In a system in which functions like DETERMINER, MODIFIER, HEAD, and MEMBER are not recognized, such a differentiation cannot be achieved in a meaningful way. The only way to make such a system account for (30) is to reduce it to either (29) or (31 ). This, again, can only be done by postulating a structure for (30) which it does not possess, and by deriving its rea~ structure from this postulated one, e.g., from: '

11.2. STRUCTURAL AMBIGUITY AND FUNCTIONAL GRAMMAR 249

(58) np

DET .

MEMBER

&

MEMBER

np

co

np

MOD

art

adj

n

a

clieap

bed

MOD

HEAD

art

adj

n

a

cheap

breakfast

DET

HEAD

I

I

and

which, as can be easily seen, is very like (55). By deletion of the second occurrence of a cheap, however, we would not arrive at the structure of (30), but at that of (29), as represented in (55). This means that apart from the transformation deleting the second a cheap, we should need further transfor1nations to arrive at an appropriate differentiation of the derived structure of (30). Two points are illustrated by this fact: (i) the nonrecognition of grammatical functions automatically leads to some kind of transformational approach, in which the cases which cannot be accounted for in the too narrow scheme of constituent structure are artificially reduced to cases which can be treated in such a scheme; (ii) though the general manner in which Chomskyan transformational grammar would handle such cases as (30) is clear, it is by no means evident by what transfor111ational operations on the postulated deep structure a correct surface structure could be attained. In this connection I can again refer to the inconclusive suggestions discussed above (section 5. 7.). We may conclude, therefore, that by properly differentiating between the structures of (29), (30), and (31) a functional grammar attains a higher degree of descriptive adequacy than both IC-description and transformational grammar.

12. On the semantics of coordination

12 .1. Can semantic aspects be attributed to coordinators? I have already cited Hockett's opinion that coordinators 'serve not directly as carriers of meaning, but only as markers of the structural relationships between other forms' (1958: 153, cf. 1961: 230). Hockett's point of view is less revolutionary than it might seem to be. The question as to whether particles like prepositions, articles, and connectives have meaning (and if so, what kind of meaning) has been a moot point since antiquity. Hockett's remark reads like a literal paraphrase of the opinion reported by Apollonius, where he opposes 'those who say that the conjunctions do not signify anything, but only serve to bind the expression 2 together' (cf. Steinthal 1891 : part 2, 323). Among the latter, we may rank Aristotle who defined syndesmoi (used by him in a much wider sense than our term 'conjunction') as 'meaningless sounds' (Poetica eh. 20). But what Aristotle called 'meaning' is equivalent to what others have referred to by such terms as 'lexical meaning' or 'independent meaning'. In more recent ti111es, the idea that conjunctions are meaningless recurs, e.g., in Bliimel (1914: 52ff.): 'Streng geno1nrnen miiszten ... solche Worter aus Worterbiichern ausgeschlossen werden' (though later on he ad1nits that 'Verschiedene Bindeworter wie und - oder - aber driicken verschiedene Beziehungen aus.' ibid. 164). It also figures, in different ways, in those writers who stress the basic distinction between 'full words' and 'empty 1 words' or 'form words', or between 'designators' and 'formators', 1

For the distinction between full and empty words, see, e.g., Ullmann 1962: 43-8 Antoine 1959: 313ff., Tesniere 1959: 53-5. Observe that Tesniere, though classing coordinators ('jonctifs') as mots vides, defined as words 'qui ne sont pas charg6s d'une fonction semantique' (ibid. 53), later on proceeds to analyse 'les varietes semantiques du jonctif' (ibid. 331), and speaks about the 'mecanisme semantique des jonctifs qui eclairent le cheminement de la phrase' (ibid. 333). Compare also Martinet's distinction between lexemes and morphemes (1960: 20, cf. 117), which is again based in part upon Vendryes' earlier differentiation between semantemes and morphemes (1921: 86ff.). The latter distinction has been criticized by Reichling (1935: 310ff.) Cf. also Martinet 1966: 41 ff. For a criticism of the opposition of full vs. empty words as a universal criterion for the establishment of word-classes, see Crystal 1966: 32ff. The terms designator and formator, finally, can be found in Weinreich 19661 : 145, 149-50.

12.2. REICHLING'S THEORY OF SEMANTICS

251

and include the coordinators in the latter class of each of these pairs. As is evident from these few remarks, at least part of the answer to the question about the semantic status of coordinators depends on what we understand by 'meaning' or 'semantics'. If, therefore, I want to defend the position that coordinators, while certainly having a grammatical value, also have a semantic aspect, I shall first have to clarify the notion of semantics on which this conviction is based.

12. 2. Reichling's theory of semantics My point of departure is Reichling's theory of semantics, the basis of which was laid in Reichling 1935 and which has been considerably elaborated in the last few years. Part of its most recent form can be found in 3 Reichling 1963 and 1965 : 29-36, 40-7 cf. also Reichling & Uhlenbeck 1964: 168-9, Uhlenbeck 1967: 293-5. Since these presentations are quite compact and for the greater part in Dutch and since, furthermore, many essential points have only been treated in Reichling's University lectures, I shall here give a short summary of this approach to semantics, in so far as it is of direct relevance to the present inquiry.

12. 2. I. Basic principles Reichling's basic idea is that a natural language is not a closed, selfcontained 'code' or system, but an 'open' system in several respects. It cannot but be an open system, since it constitutes a finite means applicable to an infinite variety of different communicative situations. What distinguishes a natural language from artificial (scientific) communicative systems is that while the latter are made up to deal with a certain restricted area of conununication, a natural language must be such as to enable its users to co1n1nunicate about anything that is or becomes part of their experience. This is made possible by the principle of 'immanent productivity' by which any language is characterized (Reichling 1952: 77ff). This principle is not only operative in grammar, where it entails the possibility of arriving at an unlimited number of individually different linguistic expressions on the basis of a finite number of fundamental patterns, but also in semantics where, more specifically, Reichling speaks of the principle of

252

ON THE SEMANTICS OF COORDINATION

'variable symbolization'. It is by virtue of this principle that any semantic aspect which from a synchronic point of view can be regarded as a constant can be applied to different things spoken about. The ter1n 'thing spoken about' should not be taken in the sense of: concrete, physical, observable object, but in the much wider sense of: anything to which we can refer by means of linguistic expressions. The ideal of a natural language in which any tertn would unequivocally refer to just one thing spoken about (an ideal cherished, e.g., by Jespersen) is si1nply unattainable, since it would exceed the limits of human memory and preclude the necessary adaptation to ever new experiences (1935: 233, cf. also Coseriu 1962: 243 f n. 20). The possibility to communicate with the same words about different things and thing-complexes is brought about again by the fact that linguistic expressions are adapted to be used in concrete, specific communicative situations. Any word, phrase, sentence, paragraph etc. as normally used is embedded in a situation in which a great amount of non-linguistic 2 infor1nation is already presupposed (Reichling 1939: 18-9).

12. 2. 2. Semantic content and final interpretation Reichling has elaborated this point as follows: any utterance (i.e., any specific token of a linguistic expression-type) carries in itself a certain amount of linguistic information. But, when spoken in a specific situation, the information which it conveys to a hearer is in the majority of cases (perhaps in all) greater than its own linguistic content. 'l'his is because it is interpreted in function of the vast amount of knowledge which the hearer possesses quite apart from this particular utterance. This knowledge includes all the information amassed by the hearer prior to the occurrence of the utterance, including his knowledge of the particular situation in which the utterance occurs (situational knowledge, including,

1

This has also been felicitously expressed by Buhler who refers to 'die Grundtatsache ... , dasz die natiirliche Sprache iiberall nur andeutet, was 1Jnd wie es gemacht werden soil, und Spielraume filr Kontekstindizien und Stoffhilfen offen liszt.' (1934: 320). Compare also Ducrot 1966a: 16: 'II nous semble ... que la reference implicite a une situation presupposee est un des caracteres les plus f ondamentaux du langage, et non pas une habilete subsidiaire, et qu'elle devrait sans doute apparaitre des la definition de la langue.'

12.2. REICHLING'S THEORY OF SEMANTICS

253

again, information about the particular speaker involved etc.}, and his knowledge about the particular linguistic expressions preceding and/or 3 following this particular utterance (contextual knowledge). The final interpretation, therefore, at which the hearer arrives on the basis of this utterance is a compositional function of the linguistic information contained in it on the one hand, and of any bit of further information not contained in the utterance as such, but relevant to its understanding in this particular context and situation, on the other hand. A fundamental and quite difficult task of semantic analysis is, therefore, to distinguish within the final interpretation between those elements of information which belong to the linguistic content of the expression and those which belong to the non-linguistic knowledge of the hearer. Some illustrations may clarify this point. When, at this moment, I am presented with the expression The president of the United States, I will automatically connect this with 'President Johnson' since I know that the factual situation is at present such as to determine this connection. But there is no question, of course, of 'President Johnson' belonging to the information conveyed by the linguistic expression as such. If this were true, the content of this expression would change at the very moment Johnson's presidency terminated. In this example, the difference between linguistic and non-linguistic information is obvious. The linguistic content of The president of the United States could be paraphrased as: 'The human being who holds the highest civil office in the U.S.A.'. Our non-linguistic knowledge tells us that this description suits President Johnson, and consequently 'President Johnson' will in most cases be an element of the final interpretation, but not of the linguistic content of The president of the United States. The linguistic content of this expression remains constant no matter who is the president of the United States. But its final interpretation was different before November 22, 1963 from what it is today, and it will be different again in the future. The linguistic content of the expression would still be the same if there were no 'president of the United States' at all, but in that case the normal final interpretation would be blocked. We would think that the speaker was making a joke or a mistake, and would react accordingly. Again, this has nothing to do with the linguistic content as such. We The context of a linguistic item consists of the linguistic material preceding and following it. The term situation denotes the non-linguistic phenomena (in the widest possible sense of the tern1) in which a particular utterance is embedded. 1

254

ON THE SEMANTICS OF COORDINATION

1night even imagine a speaker of English ignorant of the fact that President Johnson is the president of the United States. In spite of this lack of knowledge, he might perfectly understand the expression The president of the United States, though his final interpretation of it would be less complete. For a more subtle example of the difference between linguistic and nonlinguistic information, consider the following sentences:

(1) My friends (2) My friends (3) My friends (4) My friends

helped me a great deal when I was at school. help me a great deal nowadays. will help me in the future. always help me a great deal.

In sentences (1)-(4), the identical expression My friends may receive several different final interpretations, e.g.: (1) 'those who were my friends then', (2) 'those who are my friends now', (3) 'those who will be my friends in the future', (4) 'those who are my friends at any ti1ne'. The following additional interpretations are possible: (1) 'those who are my friends now helped me a great deal when I was at school', (3) 'my present friends will help me in the future', (4) 'my present friends always help me a great deal'. One might think, now, that the different possible interpretations of my friends are due to semantic ambiguities of this expression. My friends would then have to be regarded as representing a class of expressions identical in form but different in meaning. These meanings would be differentiated by temporal aspects like 'in the past', 'at present', 'in the future', and 'at any time'. This would mean that any nominal expression would be multiply ambiguous in meaning according to all the possible time-relations in which it could be situated. And for any nominal entry the lexicon would have to list all these different 'meanings'. It is evident, however, that this view will necessarily lead to unacceptable consequences. An expression like my friends does not contain a tirne-indication at all. It is quite neutral as to time. It is 'open', in that various time-relations may be imposed on it in the final interpretation on the basis of situational and contextual and further external information. The different possibilities of interpretation are, in these cases, not due to semantic ambiguities of the expressions involved, but to what 1night be called 'interpretational or referential ambiguity', by virtue of which identical forms with identical semantic aspects can, in their final interpreta-

12.2. REICHLING'S THEORY OF SEMANTICS

255

tion, be applied to a great variety of factual thing-complexes, according to the total 'setting' in which they are used. It seems useful at this point to define the different kinds of ambiguity relevant to linguistic description. Given a unique string of elements in phonemic representation (including intonation), this string is (i) structurally ambiguous if and only if an adequate grammar of the language involved assigns more than one structural description to it (cf. above, chapter 11). (ii) semantically ambiguous if and only if an adequate semantic description of the language involved assigns more than one semantic content to it. (iii) interpretationally ambiguous if and only if, though identical in grammatical structure and semantic content, it gives rise to more than one final interpretation. Only (i) and (ii) must be accounted for as such in the linguistic description. (iii), however, cannot be disregarded, for only by subtracting the effect of (iii) from the final interpretations can we deter1nine to what extent (i) and (ii) are involved. (iii), furthermore, is probably a factor relevant to any linguistic expression. (i) and (ii) are only characteristic of certain subclasses of the linguistic expressions of any language. In terms of the use of language, we can say that in the case of (i) and (ii) the interpreter has to make a choice among different linguistic possibilities; in the case of (iii), the linguistic infor1nation is unique, but a certain choice among non-linguistic possibilities is involved. In the majority of cases, these 'choices' do not depend on any independent computation of the interpreter, since just one final interpretation is uniquely deter1nined by his external (in particular, his contextual and situational) knowledge. I.e., the interpreter is 'set' to just one final interpretation. In other cases just one final interpretation may have much higher probability than other possible interpretations. If, however, two or more final interpretations are left open with comparable degrees of probability, the interpreter may arrive at a final interpretation incongruent with what the speaker meant. In this case, there is a misunderstanding, and the communication is blocked to the extent to which this particular linguistic expression is involved. In positing the irrelevance of external information to linguistic descrip-

256

ON THE SEMANTICS OF COORDINATION

tion, K.raak (1966: 53) denies the special place of linguistically ambiguous expressions in the use of language: 'As far as I see, there is not a single empirical fact, neither in the form of observable behaviour of users of language, nor in the form of introspective data, which corroborates the hypothesis that in the case of homonymy a choice is involved, and that the process of interpretation takes place, in this respect, in an unusual manner' (translation n1ine). To this, I would say (i) that indeed in most cases nothing unusual need be supposed to happen since, whereas the linguistic possibilities leave open certain choices, the choice congruent with the particular setting involved is uniquely deterrnined, (ii) that there is an ample body of empirical data in the form of 1nisunderstandings, puns, jokes, literary effects, etc., which depend directly on the difference between linguistically ambiguous and unambiguous expressions. The basic mistake which can be made in semantics is to treat external, interpretational information as if it belonged to the content of the linguistic expressions as such. This 1nistake can be amply attested from any current dictionary. It is also automatically made by anyone who proceeds as if all information conveyed by a linguistic expression should necessarily form part of the expression as such, as well as by those who neglect the 'openness' of semantic inforrnation by disregarding the fundamental role played in any communicative situation by information which is not 4 inherent in the linguistic expressions as such. Those who hold that such information is irrelevant to gra1n1natical and semantic description sometimes defend this position by referring to what a native speaker can know about a sentence which is presented to hi1n in complete isolation. But in my opinion, this is in itself impossible. One can never present a native speaker of a language with a sentence of that language in such a way that there is any guarantee that he will concentrate on just the infor1nation contained in this sentence. Any such sentence is projected into a whole body of linguistic and non-linguistic knowledge, both being interwoven in a way about which at present not very much seems to be known (for valuable observations on this point, cf. Uhlenbeck 1967). It is only by a careful comparison of the communicative effect of the same sentence in different settings that we can establish the amount of infor1nation internally connected with the sentence as such. By this I do not mean that the linguist would have to wait until he had come across an actual utterance '

For an extensive discussion of these points, cf. Kooij (forthcoming).

12.2. REICHLING'S THEORY OF SEMANTICS

257

to be able to subject this utterance to a semantic analysis. I only mean that for the semantic analysis of any linguistic expression it is all-important to imagine as many possible settings in which it could occur, and to analyse the different final interpretations which it could lead to in these settings.

12 .2 .3. Kinds of linguistic information Within the purely linguistic information contained in a linguistic expression as such we should, following Reichling, carefully distinguish between information relevant to that about which we are speaking and information pertaining to the internal organization of the expression as such. It is advisable to restrict the term 'semantic information' to the former, and to use 'grammatical infor1nation' for the latter. In any natural language, there are few units and expressions which carry only grammatical information; but all units and expressions do carry grarnrnatical information: any element carrying (semantic) information relevant to that about which we are speaking is also by its very nature predisposed to a certain use within the organization of linguistic expressions, and may indicate the way in which the expression is organized. The main reasons why the theory of semantics has not as yet proceeded very far are (i) the fact that the term 'meaning' is in most cases used for both semantic and gra1n1natical infor1nation, and (ii) the circumstance that within the domain of semantic information as such the appropriate distinctions have not been 3 made (cf. Reichling 1963, 1965 : 40). Instead of the ter1n 'meaning', Reichling proposes to use 'semantic aspect' for any kind of semantic information contained in a linguistic element. But this is a general ter1n comprising a great variety of different sorts of semantic information and we must go on im1nediately to make the appropriate distinctions. Most important in the present context is the distinction between 'independent' and 'dependent' semantic aspects. A semantic aspect is independent if it is given directly with the element by which it is carried, in the absence of any further linguistic context. A semantic aspect is dependent if it is only given within a larger whole of which it is a part. Thus, if a speaker of English is confronted with a word like blue, he knows immediately what this word means without any further contextual support. On the other hand, though the final /-z/ of tables certainly has a semantic aspect, this semantic aspect is only es-

258

ON THE SEMANTICS OF COORDINATION

tablished in combination with the semantic aspect of such a stem as table-. When it is not combined with such a noun stem, the element /z/ does not convey any semantic information of its own. The same applies to the semantic aspect of a word like the in a phrase like the man: while man has an independent semantic aspect of its own, the semantic aspect of the necessarily requires a following norninal element to be established. With Reichling, we can now make the following terrninological distinctions: we can use the term content for the independent semantic aspect contained in a phrase, a clause, or a sentence; we can reserve the term meaning for the independent semantic aspect contained in a word; and we can take the terrn semantic value to refer to any kind of dependent semantic aspect. Words like blue and man thus have meanings; a phrase like the old man has a content; and elements like the ending /-z/ in tables or the word the in the old man have semantic values (for a more precise definition of these 3 various tern1s, cf. Reichling 1965 : 31-2).

12 .3. Applications to coordinators

Retur11ing to the question as to whether coordinators have meaning or not, we can rephrase this question as follows: (i) is it true that coordinators only carry grammatical information, i.e., information solely relevant to the internal organization of the linguistic expressions in which they occur, or do they also carry semantic information, i.e., infor1nation relevant to that about which we are speaking? (ii) if the latter is true, what is the nature of this semantic infor1nation? How should it be classified in the light of Reichling's distinctions summarized above? Our answer to these questions is (i) coordinators clearly have grammatical values (just like any other linguistic unit) in so far as they are capable of a certain use in linguistic expressions. But they also have semantic aspects since they specify certain relations within that about which we are speaking. And, or, but, for each specify a certain relation between the semantic aspects of the preceding and foil owing coordinated members. These relations are not purely grammatical, but form an integral part of the semantic information conveyed by the expression as a whole. It is true that and, or, but, for, etc. have different grammatical properties, as is evident from the fact that they cannot each be used to coordinate members be-

12.4. DEFINING THE SEMANTIC VALUES

259

longing to all grammatical classes. But correlating with these gra1nmatical differences there are also semantic differences which come to light most 5 clearly in identical grammatical contexts: (5) (6) (7) (8)

John went John went John went John went

out and Bill stayed at home. out or Bill stayed at home. out but Bill stayed at home. out for Bill stayed at home.

(ii) since, by their very nature, coordinators indicate certain relationships between the semantic aspects of coordinated members, their semantic aspects are clearly dependent, i.e., they are semantic values. They cannot possibly be used on their own to indicate anything about the matter spoken about. Coordinators seemingly used in isolation should either be regarded as different words altogether (as in the Dutch questionparticle en? (and?) used in the sense of Well?, What about it?), or as presupposing coordinated contents either previous]y expressed or understood (as in English But ... , cf. above, section 4.1.4.2. 5.).

12 .4. De.fining the semantic values of coordinators: logic or linguistics?

Those who hold, as I do, that coordinators indeed have semantic values, often tend to suggest that for the description of these values one should rely on logic, since coordinators are the e1ninently 'logical' tools of natural languages. Even in the context of the description of natural language, the coordinators are often classed as 'logical connectives', and treated as if the 'meanings' attached to such connectives in logic are directly relevant to the semantic description of natural language. For a recent example of this point of view, I may cite Staal 1966: 85: 'It seems likely that the use of the logical connectives is mastered relatively late in the process of lear11ing one's mother tongue. It is clear that children who choose, in accordance with Plato, ''A and B'' when asked ''Do you want either A or B ?'', have not yet mastered all features of 6 ''and'' or ''either ... or ... '' and the logical truths determining these. ' Cf. Crystal 1966: 32-4. • More extreme still Schachter 1935: 86: 'Beistriche konnen auch an Stelle von ''und'' auftreten, sie sind dann als logische Partikel aufzufassen: ''Robert, Karl und Max gehen'' = Robert geht und Max geht und Karl geht.' 1

260

ON THE SEMANTICS OF COORDINATION

To this I would like to say (i) that in my opinion children do not master any logical connective in learning their mother tongue, since logical connectives are only learned by means of courses and books on logic, and (ii) that the semantic values of coordinators of natural languages are not, like the logical connectives in logic, deter1nined by 'logical truths'. It is no disparagement of logic (nor of natural language) when I stress the undeniable fundamental differences between the two. Again and again, students of natural language and of logical systems have rightly noted that the two are objects of fundamentally different natures, both in their aims and in their internal properties. This position was quite clearly and eloquently expressed by Coseriu in his treatment of the relations between natural language and logic (1962). The main tenets of this paper, to which I fully subscribe, may be su1n1narized as follows: natural language and logic belong to different, though interrelated fields. Natural language is neither logical nor illogical, but it is neutral in relation to logic. As Aristotle already noted, natural lanl guage is first and foremost characterized as 'meaningful', while logicasystems are largely deter1nined by such properties as truth and falsity. Since only certain sentences of natural language can be correlated with truth..:values, logic is only concerned with a part of natural language, and in a way quite different from its normal co1111nunicative use. Natural language is therefore prior to logic (239), linguistic semantics is not to be equated with logical semantics (240), and the fundamental logistic error is to consider natural language as an object of a logical nature, as a product of logical thinking (238). This may also be expressed as follows. In a quite general way, the ai1n of logic may be described as the construction of systems in which true sentences (or propositions) can be derived from true sentences·(or propositions) in a well-determined way. In order to achieve this aim logic selects certain structures and terms from natural language and modifies these in such a way as to suit the requirements of unequivocal derivation. 'l,his modification involves primarily the definition of a class of 'operators', to which unambiguous ·meanings are assigned, and the setting up of rules which determine the well-formedness of 'sentences' and the validity of 'derivations' within the particular logical system proposed. In contradistinction to what can be observed in natural languages, the 'concrete' meanings of the terms appearing in logical formulas are irrelevant to the

12.4. DEFINING THE SEMANTIC VALUES

261

system as such, while the 'meanings' of the operators involved in these systems are determined by the rules governing well-formedness and validity of derivation, i.e., ulti1nately, by the logical properties of. truth and falsity. Three points should be retained in this connection: (i) natural languages and logical systems have different aims, (ii) logical systems are derived from natural languages by a process of abstraction and modification, (iii) 'meaning' when used in a logical context is fundamentally different from 'meaning' or 'semantic aspect' as found in natural languages, as I shall show below in greater detail. It is undoubtedly an important undertaking to compare the character of natural languages with that of logical systems. I do not deny that in this process certain aspects of natural language may become clearer than they were before (cf. Quine 1961). But the logical systems once constituted cannot be taken as models for understanding natural languages, since these systems only exist after the process of abstraction and modification has taken place. We can learn something about the structure of birds in constructing aeroplanes, but we cannot say that the aeroplane 'explains' the structure of the bird, let alone that we can say, or i1nply, that a bird is, or contains part of, an aeroplane. Or, to borrow a comparison from Strawson (1952: 81): 'It will not do to reproach the logician for his divorce from linguistic realities, any more than it will do to reproach the abstract painter for not being a representational artist; but one may justly reproach him if he claims to be a representational artist.' As has been rightly pointed out by K.raak (1966: 172ff.), it is a careful linguistic analysis which will, in the majority of cases, throw more light on the linguistic facts than any amount of logical analogy. •

For the general point at issue here it is interesting to refer to the discussion between Bar-Hillel (1954) and Chomsky (1955a) on the relevance of logical syntax and semantics for the theory of natural language. In this discussion, it is Bar-Hillel who takes the logician's view, while Chomsky defends that of the linguist. Bar-Hillel started from the conviction, which we have been criticizing here, that logic and linguistics are both 'essentially attempting to construct language systems that stand in some correspondence to natural languages' (1954: 235); that, in other words, their tasks are very much the same, though they have differently oriented interests concerning this task; and that consequently logical syntax and



262

ON THE SEMANTICS OF COORDINATION

semantics as developed by Carnap and others should be made fruitful for linguistics, especially for the clarification of the notions of 'transfor1nation' (taken as directJy connected with the relation of consequence, i.e. of 'logical' derivation and inference) and 'synonymity'. Chomsky rightly opposed this view by stressing the fundamental qualitative difference between artificial language systerns and natural languages, and by noting the fact that in many cases linguistics precisely has to analyse the notions which are taken for granted in logic. What is logical in linguistics is not its object, but linguistic theory as such (just as any scientific theory): 'The correct way to use the insights and techniques of logic is in for1nulating a general theory of Jinguistic structure. But this fact does not tell us what sort of systems for1n the subject matter of linguistics, or how the linguist may find it profitable to describe them. To apply logic in constructing a clear and rigorous linguistic theory is different from expecting logic or any other formal system to be a model for linguistic behavior.' (1955a: 45). It may be asked whether my contention that Chomsky's transformational theory has some of its roots in certain ideas of a more logical than linguistic nature (above chapter 6) is not invalidated by the position which he took explicitly in 1955. To dispel this possible rnisunderstanding, I may stress the following points: (i) I do not regard the transfor1national theory as a logical theory, but as a linguistic one, just as it was intended; (ii) in relation to the description of coordination we have found certain reasons why this theory seems to be linguistically inadequate; (iii) I have isolated a basic principle underlying this theory which I have termed the 'reduction postulate'; (iv) I have shown that the reduction postulate can be traced back through the history of traditional grarnrnar and logic to its ultimate source, viz. to Aristotle; (v) I have established the fact that the reduction postulate which seems to break down on natural languages was fully motivated in the context of traditional logic (cf. above, section 6.5.); (vi) we have therefore found an indirect connection between traditional logic and transformational theory, which may in part explain certain inadequacies of the latter. I may further note that it seems probable that Chomsky's point of view has undergone certain modifications since 1955, bringing hi1n closer 7 to Bar-Hillel's. This might be concluded from his repeated defence of Cf. the pertinent remarks added to the French translation of Chomsky's 1955 article, in Langages 2 (1966) 42 .

7



12.4. DEFINING THE SEMANTIC VALUES

263

logically inspired traditional gra1nmar (cf. his remarks in 1965: 63ff, and elsewhere) and, more specifically, from his use of such terms as 'proposition' in a purely linguistic context (cf. above section 5. 7. 3. and fn. 56). This amounts to saying that in my view Chomsky's transformational theory has, to a certain degree and in a roundabout way, fallen victim to precisely the fault for which he criticized Bar-Hillel in 1955. Finally, it should not be thought that I want to represent Chomsky as asserting the 'logical' nature of coordinating particles. In fact, in this same 1955 article, he makes quite pertinent remarks to the contrary in connection with 'logical particles' such as and, not, all, etc.: 'Actually, the well-known discrepancy between the material conditional of logic and the 'if-then' of English (which seems to be true only if there is some kind of connection in content between antecedent and consequent) should be enough to warn anyone not to make a blind leap from mathematical systems to ordinary linguistic behavior.' (ibid. 39). This discrepancy between logical systems and natural language is ·by no means restricted to the conditional. It is a general characteristic of any feature of logic as compared with any feature of natural language. This has also been forcefully argued by Ryle (1954: 117-8) in a passage which is well worth quoting in full: 'I have spoken as if our ordinary 'and', 'if', 'all', 'some' and so on are identical with the forrnal constants with which the Formal Logician operates. But this is not true. The logician's 'and', 'not', 'all', 'some' and the rest are not our fa1niliar civilian terms; they are conscript terrns, in uniform and under rnilitary discipline, with memories, indeed, of their previous more free and easy civilian lives, though they are not living these lives now. Two instances are enough. If you hear on good authority that she took arsenic and fell ill you will reject the rumour that she fell ill and took arsenic. This familiar use of 'and' carries with it the temporal notion expressed by 'and subsequently' and even the causal notion expressed by 'and in consequence'. The logicians' conscript 'and' does only its appointed duty - a duty in which 'she took arsenic and fell ill' is an absolute paraphrase of 'she fell ill and took arsenic'. This might be called the minimal force of 'and'. In some cases the overlap between the military duties and the civilian work and play of an expression is even slighter. What corresponds in the glossary of Formal Logic to the civilian word 'if' is an expression which plays only a very small, though certainly cardinal part

264

ON THE SEMANTICS OF COORDINATION

of the role or roles of that civilian word ... Forn1al Logic operates ... only with artificial extracts from the selected few topic-neutral expressions of ordinary discourse ... and even these have to be filed down to reduced size and unnatural shape before the Formal Logician will deign to inspect them.' In commenting upon this passage, Staal tries to di1ninish the discrepancy noted by Ryle in the following way: 'It may be noted that Ryle, in describing the 'and' of natural English, uses 'and' in at least two senses: one of these, the 'and' which occurs in 'and subsequently' and in 'and in consequence', seems to be the basic 'and', which in fact is very close to the logical 'and'.' (1966: 81). The confrontation of the opinions of Coseriu, Chomsky, and Ryle on the one hand with those of Bar-Hillel and Staal on the other may have served to clarify the points at issue here. Studies on the mutual relations between linguistics and logic being quite numerous, no complete account of all opinions advanced can be attempted here. But the authors cited 8 may be regarded as exponents of possible approaches to this problem. At one extreme, there is the hyper-logistic point of view of those who hold that anything linguistic is also logical and vice versa; at the other extreme, there is the hyper-linguistic idea that natural language and logic are two completely unrelated domains. In between these two extremes, there are just as many individual positions as there are writers on this topic. My own position lies towards the linguistic end, with due recognition of the heuristic value of logical analysis for linguistics and of the need for logic in the construction of scientific theories; but with grave doubts whenever logical properties are attributed, as such, to the linguistic expressions of natural languages. I can now elaborate the fundamental differences (i) between the coordinators of natural language and the connectives figuring in logical sys8

For an admirable treatment of the differences between the logical connectives and the coordinators of natural language, see also Strawson 1952: 78-93 (quoted in part above, section 6.4.). See further Staal 1963: 9-10, where he explicitly opposes Cbomsky's 1955 view. Some quite interesting remarks on this topic may be found in recent articles by Ducrot, who demonstrates that the attempt to define the coordinators of natural languages in terms of logical characteristics of the expressions in which they figure automatically leads to 'reductionism' (1966a: 3-9, cf. also 1966b). See further Kraak 1966: 168ff. The second issue of the journal Langages presents an extensive bibliography on this subject.

12.4. DEFINING THE SEMANTIC VALUES

265

terns, and (ii) between semantic aspect and interpretation. There is a natural connection between these two points. Those who tend to identify the systems of natural languages with those of logic also tend to diminish or to ignore the role of factors outside the expression as such which rnight have an influence on their interpretation. This is because in logical systems such external factors are excluded by definition. In order to solve the problems arising in such a purely 'internal' approach to the expressions of natural language two courses are open: (i) the role of external factors is sirnply denied; (ii) those external factors which seem unavoidable even in the 'internal' approach are described as properties of the expressions as such, properties which, though not evident from their outward appearance, still belong to their structure on a deeper, latent, or covert level. My point of view is that a linguistic description should not 'over-differentiate' the internal properties of linguistic expressions while denying the role of external interpretative factors. Nor should it overrate the role of such external factors, thus failing to reveal the indeed often quite subtle differences in the internal properties of linguistic expressions. Rather, it should establish the proper balance between the internal and external factors which together lead up to the final interpretation, in such a way that a maximally adequate explanation of human communication is achieved. Considering the sentences: 0

She took arsenic and fell ill. (9) 6 (10) She took arsenic and in consequence she fell ill. 0

6

my first remark is that if indeed and and and were different in sense, and if indeed an