Discussing Language

Dialogues with Wallace L. Chafe, Noam Chomsky, Algirdas J. Greimas, M. A. K. Halliday, Peter Hartmann, George Lakoff, Sy

302 74 8MB

English Pages [441] Year 1974

Report DMCA / Copyright

DOWNLOAD PDF FILE

Table of contents :
INTRODUCTION
CONTENTS
WALLACE L. CHAFE
NOAM CHOMSKY
ALGIRDAS J. GREIMAS
M. A. K. HALLIDAY
PETER HARTMANN
GEORGE LAKOFF
SYDNEY M. LAMB
ANDRE MARTINET
JAMES McCAWLEY
SEBASTIAN K. SAUMJAN
SYNTHESIS
JACQUES BOUVERESSE. I. ON LINGUISTIC METHODOLOGY
JACQUES BOUVERESSE. II. ON THE RATIONALISM VERSUS EMPIRICISM DEBATE
JACQUES BOUVERESSE. III. ON DICHOTOMIES
JACQUES BOUVERESSE. IV. ON THEORIES OF MEANING
JACQUES BOUVERESSE. V. ON PRAGMATICO-SEMANTICS AND NATURAL LOGIC
JACQUES BOUVERESSE. VI. ON PHILOSOPHY OF LANGUAGE
BIO-BIBLIOGRAPHICAL NOTES
BIBLIOGRAPHY
INDEX OF NAMES
INDEX OF SUBJECTS
Recommend Papers

Discussing Language

  • 0 0 0
  • Like this paper and download? You can publish your own PDF file online for free in a few minutes! Sign Up
File loading please wait...
Citation preview

JANUA LINGUARUM STUDIA M E M O R I A E N I C O L A I VAN WIJK D E D I C A T A edenda curai C. H. VAN SCHOONEVELD Indiana University

Series Maior, 93

DISCUSSING

LANGUAGE

DIALOGUES WITH WALLACE L. CHAFE, NOAM CHOMSKY, ALGIRDAS J. GREIMAS, M. A. K. HALLIDAY, PETER HARTMANN, GEORGE LAKOFF, SYDNEY M. LAMB, ANDRÉ MARTINET, JAMES McCAWLEY, SEBASTIAN K. SAUMJAN, AND JACQUES BOUVERESSE

by

HERMAN PARRET Belgian National Science

Foundation

1974

MOUTON THE HAGUE • PARIS

CD Copyright 1974 in The Netherlands and Belgium, Mouton & Co. N.V., Publishers, The Hague, and Herman Parret, Leuven. No part of this book may be translated or reproduced in any form, by print, photoprint, microfilm, or any other means, without written permission from the publishers.

LIBRARY OF CONGRESS CATALOG CARD NUMBER 74-78730

Printed in Belgium by N.I.C.I., Ghent.

To Bea

INTRODUCTION

Discussing Language consists of ten dialogues with linguists and one dialogue with a philosopher of language. The purpose was to present a substantial but easily readable survey of the most characteristic trends in contemporary linguistics, and of the methodological and epistemological problems raised by these trends. When I invited these ten linguists to contribute to this initiative, I considered only the representativeness of their work and the richness of the philosophical problems connected with their linguistic theory. I asked them to draw the genealogy of their work, to make explicit the conceptual apparatus used in it, to build analogies and syntheses, to illustrate their methods, to make suggestions with regard to the possibility of interdisciplinary research, to evaluate competing trends in linguistics, to describe the role and the importance of logic and of language philosophy for linguistics proper. My ultimate wish was to place the often fragmentary and regional problems with which the linguist is concerned in a larger framework and to interpret the theoretical presuppositions of contemporary research in linguistics. I believe that this book is valuable, therefore, not only to linguists who are interested in linguistic theory and in general linguistics, but also and in an equally compelling way to logicians, philosophers of science and of language of any orientation (ontological, epistemological, or analytical); because of the frequent references to other human sciences, anthropologists, psychologists, and sociologists will also find sufficient material in these conversations to interest them. That this book is built in the form of dialogues is not a stylistic originality. First of all, it allowed treating topics of a highly technical and substantial nature in an easily readable form: amplifications and explanations could be requested and analyzed further. Informal conversations - preceded, of course, by a thorough preparation of the subject matter of the dialogue and followed by the editing of the transcripts also had the advantage that they allowed delving into very contemporary and very controversial topics, mostly of a scientific-theoretical or language philosophical nature, which the linguist, because of his typical modesty, does not often or extensively treat. The dialogue form, moreover, made it possible for me to juxtapose alternative theories: I asked some collaborators questions which were suggested by answers of others. In this way, an active discussion could ensue between points of views and

VIII

INTRODUCTION

theoretical stands which otherwise hardly meet each other. The Index of Names and the Index of Subjects, which were added in addition to the bio-bibliographical notes on the authors, and the bibliography (of all works mentioned in the dialogues) can help the reader to continue this confrontation. In a short synthesis and in an extensive interview with the language philosopher Jacques Bouveresse, the most important topics of the dialogues with linguists are treated again - with explicit references to and analyses of the answers given by them - and situated and interpreted in a language philosophical perspective. Discussing Language aims at contributing to a lively and living discussion, not only among linguists but also between linguistics and all human and philosophical disciplines which have language and speaking man as their objects. My greatest thanks go to Ludo and Rita Beheydt, L. R. Mertz, Marc Ooms, and Vic Thaels - all of the Departments of Linguistics and Applied Linguistics at Louvain University - for the demanding and difficult work of transcriptions and translations. In friendship and with great respect, I also thank the eleven contributors to this book, who not only prepared, held, corrected, and annotated these interviews with indisputable seriousness, but also understood the meaning of this initiative in its full scope. June 22, 1973.

CONTENTS

Introduction

VII

Wallace L. Chafe Noam Chomsky Algirdas J. Greimas M. A. K. Halliday Peter Hartmann George Lakoff Sydney M. Lamb André Martinet James McCawley Sebastian K. Saumjan

1 27 55 81 121 151 179 221 249 279

Synthesis

297

Jacques Bouveresse I. On Linguistic Methodology II. On the Rationalism Versus Empiricism Debate III. On Dichotomies IV. On Theories of Meaning V. On Pragmatico-semantics and Natural Logic VI. On Philosophy of Language

701 316 345 361 379 396

Bio-Bibliographical Notes

405

Bibliography

413

Index of Names

423

Index of Subjects

426

WALLACE L. CHAFE

Wallace Chafe, in the beginning of your book Meaning and the Structure of Language 1 you explicitly stated that you consider your theory as a genuine alternative to current trends in American linguistics. Do you still hold this statement ? - I don't believe I said exactly that, but rather that linguistics was in need of new alternatives, and that I hoped to provide at least a few ideas that might be useful along the way. All I would want to say about my own work is that I have constantly been looking for a better understanding of language. I have been trying to pick up whatever is valuable from other sources, but at the same time to put together ideas of my own it is just a continuous struggle. I do not feel I have arrived at anything at this point that I am very satisfied with. About the mid-sixties, the time that Chomsky's Aspects 2 appeared, I thought that it would perhaps be valuable for me to put down in the form of a book the thoughts I had about language. They were quite different from those represented by Aspects. I finished a first draft of my book in 1966 but I continued to work on it for three more years and finally sent it to the publisher in 1969. By the time it actually appeared in print, a lot of points I wanted to make had already come out in other ways: the generative semantic movement had already started at that time. The general point was that the base of language ought to be semantic rather than syntactic. This is the point I tried to make from the early sixties onwards but it only appeared in the open after 1967. When it appeared in my book it had already been in the air for some time from other sources too. I did publish an article, "Language as Symbolization", 3 and a review 4 of Katz' Philosophy of Language in which I tried to make the same kind of point. But the book was an attempt to follow through with this idea in a fairly detailed way. I tried to avoid the sort of formalization of language structure that was developed at M.I.T. and to build one that was more semantically based. I am able to see the book now in a larger framework and to place the form of representation I had in the book against a larger background. After the book was finished I began thinking seriously about discourse structure, i.e., 1

W. L. Chafe, Meaning and the Structure of Language (Chicago, 1970). N. Chomsky, Aspects of the Theory of Syntax (Cambridge, Mass., 1965). 3 W. L. Chafe, "Language as Symbolization", Language 43 (1967), 57-91. 4 W. L. Chafe, "Review of J. J. Katz, The Philosophy of Language", International Journal of American Linguistics 33 (1967), 248-254. 2

2

WALLACE CHAFE

about units which extend beyond the boundaries of sentences. I found that I really needed to get involved in such overwhelmingly large questions as the nature of human knowledge, the extent to which knowledge can be formalized. Those questions arrive in a particularly puzzling way as soon as you start looking at larger structures of language than just simple sentences. Does that mean that your theory is still an alternative to the other rival trends in linguistics ? - I would like to keep the good things in the theory as they stood several years ago but, as I said, I would like to place them within a larger framework. I would like to clarify what exactly is contained in a semantic representation and what aspect of language it is that it represents. By nature, I do not like schools in linguistics - we all are searching for the truth about language. I would not like to have my book become a sort of basis for any kind of dogma, because I would not accept such a dogma myself. However, that is the way things go. This is for example what happened with the various formulations of Chomsky's Aspects and before that of Syntactic Structures, which became sort of a paradigm within which everybody worked. That might be good in some ways but also tends to hamper imaginative progress. I would not particularly care to see my own book used in that way but just taken as possible suggestions for developing a better understanding of language. Your scientific autobiography may be very interesting and clarify certain points. What, originally, were your criticisms against American structuralism ? - As I mentioned in the book, I was dissatisfied with the structuralist model I had available to me for working on American-Indian languages. I happened to work with Iroquois languages, which have an extremely complicated phonology, and I felt particularly pleased by the development of generative phonology that allowed me to account for the facts in terms of rules, in a way that was not possible in the structuralist framework where all we had were phonemes and allophones. That alone did not seem to account for what I found in these languages. In the area of grammar and syntax, it seemed that long words in this polysynthetic language contained items which represented semantic elements - you might have a word that consisted of eight morphemes, each of which seemed to have some kind of meaning - but there were also a lot of idiomatic expressions where morphemes were put together in an arbitrary way to reflect some unitary meaning which we would not anticipate from the meanings of the parts of the expression. This is a familiar problem but it made me think about how this particular language, and of course English too, was used to express meanings. The restriction to phonetic data with just an occasional look at semantic facts was what bothered me very much in structuralism. That really is the heart of the whole problem. It was exactly the same thing that bothered me about the early M.I.T. school, which seemed not to provide any useful ways of dealing with meanings. It surprised me that a lot of people in both psychology and philosophy

WALLACE CHAFE

3

seemed to feel at that time that the M.I.T. trend was a breakthrough, providing a lot of insight into how language dealt with meaning. But this did not seem to me to be the case. So, I needed to look somewhere else and develop ideas of my own. How do you see the dependence of the earlier Chomsky on his predecessors, the American structuralists ? - I guess there was a fairly strong dependence. There seems to have been a line of historical development in linguistics which ran through Zellig Harris to Chomsky. But it is an oversimplification to put it in those terms. Structuralism was concerned with surface structures and Chomsky did linguistics a great service in making clear that surface structures could not be explained in their own terms only, but required reference to something else which lay behind or beneath the surface structure. The real crux of the matter is of course how this should be represented. That really has not been solved. First there were kernel sentences, then there were deep structures, and more recently semantic structures of various kinds. Even that is not the right way to look at things. You oppose your own theory to transformational grammar as semanticism to syntacticism. Is that not a very strong claim ? - In the mid-sixties I thought there was no real justification for positing a level, a stage, something in language which coincided with the so-called deep structure which the transformationalists were working with. I used the term syntacticism simply to refer to that kind of linguistics in which one assumed an autonomous level of syntax. It seemed to me then and still seems to me that there is nothing to be gained from hypothesizing the existence of such a kind of representation in language. Where surface structure may be a well defined level - it is written down on paper in words and sentences - if you try to develop something that underlies surface structure, there is no good reason to think that something like a Chomsky kind of deep structure is very much help. Semanticism hypothesized that things were determined on a level which was closer to meaning; that seems to me to be correct. I have never found a really good demonstration that a syntactic deep structure is a useful thing to have. The solution of an interpretative semantic component as it was sketched by Katz and Fodor 5 and by Katz and Postal6 did not satisfy you ? - No. It still depends on a syntactic level on which the interpretative level is based. I suppose all of this has something to do with the distinction between competence and performance. Whenever one said the kind of thing I was trying to say at that time, he was accused of being concerned with performance rather than competence. Whenever facts were pointed out that seemed to conflict with the syntactic level, it was 5

J. J. Katz and J. Fodor, "The Structure of a Semantic Theory", Language 39 (1963), 170-210. J. J. Katz and P. M. Postal, An Integrated Theory of Linguistic Description (Cambridge, Mass., 1964). 6

4

WALLACE CHAFE

said that deep structure was not concerned with performance but only with competence. I really think that this competence-performance distinction has been a great hindrance in linguistics. It has not helped in our understanding of language at all, in fact it has been used as a rationalization for avoiding all sorts of important, relevant, and interesting questions about language. Dissociating language use from formal structure does not seem to me a profitable line to follow. What I have been concerned with is building a model of language which is related to how language is used and I do not see anything wrong with that. It has been a mistake not to look at language in functional terms rather than in terms of abstract formal structure. A lot of what happened in the sixties happened because many linguists were very fond of formal structure as such; they liked to work with formalism and were much more concerned with that than what was really going on when people talked. But I believe that Chomsky speaks also about semantic competence, not only about syntactic competence. The structure of meaning belongs to competence also ... - Yes, that is true. I don't have too much to say on that point. He may have said that, but he actually has not, as far as I can see, provided very much for a useful approach to questions of that kind. It is really not my intention to be critical of a lot of other linguists. We are all confronted with a very complex and mysterious thing - language. A lot of different avenues for understanding it are open to us. What I object to is the tendency which is so strong in American linguistics to take only one approach to language. Structuralists were extremely narrow and dogmatic and the same has been true of the transformationalists. There is a single line, a sort of party line which has to be followed in order to belong in the field of linguistics. One of the great advantages of your linguistic theory is that it is 'natural'. Meaning is what language is all about. Do you see this common sense definition of language in opposition to the more sophisticated trends in linguistics ? - Naturalness and performance are really comparable terms. They both refer to language as it is actually used when people talk and not to some abstract structure underlying it. Formalism is fine and I am interested in that too. But people have rushed into formalism before they really have known what it is. They have tried to formalize and then they have become slaves of a particular formalism; it can capture a whole generation of scholars. Are your notions of transformation and of derivation of a sentence comparable with the same notions as used by transformationalists ? - 1 am particularly concerned with what is psychologically valid. I certainly have been sceptical about whether many transformations that have been proposed within transformational grammar have any kind of psychological validity. A lot of them seem to be the result of linguists' games, of playing around with the system that has been developed, and do not reflect anything that bears on what is going on in the mind of

5

W A L L A C E CHAFE

the speaker of the language. Language provides a kind of filter for thoughts, if I can put it in those terms. Whether the filter should be described in terms of transformations, as is presently understood, is t o me the important question. I don't really know a satisfactory answer to it right now. You accept levels of deep structure and derivation of sentencesfrom those deep structures. Is this a departure from or a solidarity with the more common transformational opinion ? - In some sense my semantic structures are comparable to the general idea of deep structure or semantic structure in other devices. I am at present much concerned about whether semantic structures of the sort that I represented in the book do represent something that is valid within language. Do linguistic transformations have psychological

reality?

- This is what I am questioning. The same question has arisen within phonology where what is going on is perhaps a little clearer. It becomes much more complex and difficult in the semantic-syntactic area. I suspect that some of these things have psychological reality and others don't. Can one call your present linguistic activity 'psychosemantics' ? - It is a name that I made up. I think of this now as a branch of psycholinguistics. Psycholinguistics has a long history going back to Wundt and Buhler, etc., but there has been a recent resurgence of interest in psycholinguistics since the fifties. It particularly focused at first on attempts to find psychological confirmation for transformational grammar. A lot of work was done in that area which does not seem to have produced anything of lasting interest. Since then the main focus of attention has shifted to child language, language acquisition. It would also be useful and eventually necessary to have another branch of psycholinguistics which might appropriately be called psychosemantics. Psychologists and linguists are trying to get at the same sorts of things at the end; that is, we are trying to understand how the mind works. Linguists do this through language; sometimes they seem to be more interested in language than in the mind, but ultimately I don't think you can understand how language works without understanding how the mind works. The psychological concerns and the linguistic concerns will converge and there really will not be any distinctions between the two except that the psychosemantic approach would look at psychological questions through language, which is a principal avenue to understanding the human mind. Within American linguistics at least this is new. The antimentalism which was so important for a long time destroyed any attempt like this - and I think transformational grammar in its orthodox form also avoided this. Did Chomsky not call linguistics a subpart of psychology ? - That is true.

6

WALLACE CHAFE

What language phenomena do you study in psychosemantics ? - Last year I was particularly concerned with some phenomena involving memory and some involving consciousness; those were special manifestations of a larger concern for this whole area of how language and the mind interact. I see human knowledge as being divisible into two distinct kinds, one of which we may call experiential and the other conceptual. What I mean by experiential are those things in the mind which are most closely related to sensory input, and to emotion also. I do not want to equate experiential knowledge with visual images - I don't exclude auditory imagery, other kinds of sense modalities or emotional things either. This very important component of human thought cannot be formalized in terms of discrete elements and discrete relations. If you put it in terms of computers, it could not be programmed on a digital computer but might conceivably be handled better on an analogue computer, although I don't know of any now that would even begin to be adequate to handle things of this sort. But I wanted to contrast this area of knowledge with another area which I called conceptual. There things are organized in terms of discrete elements, and formalizations are possible in the terms we are used to. It is rather as if the mind had two different parts and one part consisted of a lot of units which were discrete and each of these units was somehow tied to, manifested in, a lot of material in the experiential part of the mind, which was more nebulous and overlapping and vague and so on. Each discrete unit has an experiential content in the form of mental imagery. Whether we can be conscious of concepts, these unitary kinds of things, is an interesting and old question. What is interesting from the point of view of language is that it is only the 'concepts' which are actually symbolized in language - they are used as a way of communication and manipulating a lot of experiential material which is not itself directly amenable to linguistic treatment. So it is as if we have a lot of pictures in our mind which cannot be expressed directly; we have to turn them into some kind of discrete structure before they can be transmitted. Is there any experimental support for this hypothesis ? - 1 have been doing certain investigations lately which I suppose are like psychological experiments. I have been showing movies to people to see how they would talk about them. I had basically three little scenes on movie film which I showed to fairly large groups of people, sixty to seventy at a time. I asked them to describe what they saw. I was interested in seeing how they turned the original visual input into words. I want also to have them describe it on several later occasions to see what role memory plays in all this. What could be very interesting in these psychological experiments is the relevance for linguistics. That is very important and I would like to stress that aspect. At present I don't pretend to understand thoroughly everything I am going to talk about, but I think that linguistics needs to look more in this direction in the future. So, if to any extent my work can be called a real alternative to other work that is being done, I would particularly like to think of it as involving this sort of development.

WALLACE CHAFE

1

Could you give us more data about your experiment ? - I can perhaps describe particularly one very short film that I showed to people. It contained two boys standing facing each other, and one boy gave the other boy a banana; and that is about all there was to it, except that after the second boy had received the banana he changed it from one hand to another. Then he turned towards the viewer, the camera, and smiled. After people have seen this film, they have in their minds a 'concept' of the total scheme of what they saw. This is comparable to some incident one has witnessed; suppose you see something happen on the street as you are walking along. The picture of the total thing forms a sort of holistic unit in your mind; sometimes things of that sort are given names, particularly with a story. Somebody who is about to tell a story has the whole story in his mind, and that in fact has a name. In this case, of course, the movie has no name. But in converting the holistic concept into language it is impossible to do it directly, it has to be broken down into smaller parts. So there seems to be a process going on which I have been calling subconceptualization. By that I mean taking larger concepts which include more of experience and breaking them down into smaller concepts which include smaller amounts of experience. In verbalizing some experience a person subconceptualizes, he performs a conceptual analysis. How did people react to this particular experience ? - People identified one or more events which were included within the total film. There was the giving event, there was also the event of changing the banana from one hand to another, and there was the event of turning and looking at the camera. Now, these events are clearly not equal; in some sense one of them is more important than the others: I have been using the term salient for this. I speak of the event of giving as being more salient than any other event which took place in this film. One way to determine what is salient experimentally is to analyze which part of the film is mentioned by the audience. It turned out, as one would expect in this case, that everybody said something about the giving event, whereas only a fairly small proportion of people mentioned anything else that we included in the film. So that would be a way of measuring or getting some statistical grasp of salience. I am looking for some way to give this notion of salience a more respectable status in investigation. There is also another parameter which I have been calling exhaustiveness, although I don't know whether I like that word too much. But if you took a total description of a scene by one person, and then another description by another person, you might find that one was more exhaustive than the other, that it covered more of what was in the film. Of course, it does not mean that the person who was less exhaustive did not notice these other things, but only that he did not think they were important enough to write down or to verbalize. The speaker who is trying to describe what he has seen goes through the subconceptualization process, first identifying different events and perhaps different background phenomena, etc., and eventually he reaches the point where he will have analyzed events into their component parts - he will have sub-

8

WALLACE CHAFE

conceptualized to the point where he has concepts of particular objects at particular times. How is the subconceptualization verbalized? - He cannot just use some phonological representation for the 'giving event', he has to go further than that. There is this notion of giving, which is a kind of action which includes a beneficiary, an agent, and a patient, and there is the action itself; there is the giver and there is the receiver; there is the object which is given. He has to subconceptualize down to this point. How can those concepts be characterized? - First of all, each of them has coordinates in time and space. Y o u can say "here is a banana", an object you know which exists in a certain place at a certain time. This is part of our everyday experience. The action that this boy performed was localized in space and time as well. It is simply impossible to have words or phonological representations for concepts pinned down in both space and time - I have used the term appearances for these. A n appearance would be either a perception of a particular object or a perception of a particular event or action at a particular time and in a particular place. There is a question of how long an appearance lasts. The notion of consciousness has to be invoked: while one is continuously conscious of a particular thing happening or of a particular object, we can call that a single appearance. N o w there is another kind of concept, which is one step more abstract than this, namely what we might call an individual. A n individual can have a phonological representation because he extends through time - and that would be a proper name. One of the boys might be named "Steve", but the people w h o saw the film would not know that and so could not give a phonological representation to the individuals. So what was necessary then was to go one level higher. One should note that individuals are particular in space but not in time. A s you think of some person or of some house or anything of that sort which you see occasionally but not constantly, you keep seeing different appearances of the same individual. A s soon as you impose particularity in time, then you have what I am calling an appearance. If, on the contrary, you subtract the temporal particularity from an appearance, then what you have is something that is particular only in space and not in time - so the idea of John Jones or A m y Larson would be like that. Some individuals of that sort do have phonological representations like proper names and others don't. I mean the proper name may not be available to the speaker: in that case it is necessary to go to a more abstract level still and this would be the level of what I have called generic concepts. What do you mean by generic concept? - There is a generic concept "boy", for example, which is particular in neither space nor time: it does not have particularity at all. So this you could think of as the third rung on the ladder of abstraction. In the majority of cases it is necessary to go to that

WALLACE CHAFE

9

level before a speaker has a way of communicating because that is the only level at which phonological representation would be available. Generic concepts, at least a vast number of them, do have phonological representations; this is true, too, of some individuals but not of all - that depends very much on the knowledge that the speaker and the hearer hold in common. Appearances, finally, have no phonological representations at all. So, if I sum up, we have the following picture. The speaker would begin with a holistic concept of an entire incident or scene or story, but he subconceptualizes this in a hierarchical fashion until he arrives at the level of appearances, that is, of the smallest concepts that are localized in both space and time. At that point he will then look for ways to communicate them; the appearances are not something he has words for. So he will then have to ask: Is this an appearance of a certain individual? Does that individual have a name? If the answer to that is no, then he will have to ask: Is this individual an instance of some generic concept? And that generic concept then would provide the name. For verbal concepts, that is actions and processes and states, there is nothing at this level of individuals such as there is with nouns - we can have something that is particular in space but not in time, but we do not have the opposite. Something particular in time but not in space would be something like every act of jumping that took place at a particular time everywhere in the universe, and that is just not a coherent concept. I mean it is impossible to imagine a word for something like that; it seems not to be useful. So as a result we do not have proper verbs. I suppose it is because of the nature of human experience - people can be at the same place at different times but you cannot be at different places at the same time; so there is an asymmetry between time and place. At the level of appearances we have both nominal and verbal concepts, at the level of individuals we have only nominal concepts, but then at the generic level we have both kinds again. So if you have a particular appearance like this act of giving in this film, the only way to communicate it is to identify it as an instance of some generic concept which might be the generic concept GIVE. As a matter of fact the majority of people who saw the film called it HAND. I suppose we can say that HANDING is a particular kind of GIVING, and this is another kind of thing that I am interested in. This all leads you as a linguist to focus on two basic questions. One is this question of subconceptualization and what is the most salient way to subconceptualize a larger concept. The second question is, given a particular appearance or individual, how does a speaker choose the proper generic concept for those? Why does he call this HANDING or GIVING or some other thing? A few people said PASSING. All indicate somewhat different interpretations of what was happening. Of course this is a difficult question to answer - how does one pick the proper generic concept? - but I think this is a promising line of investigation. In this scene there was first of all the act of giving or handing, but then there were also the participants in it. For the most part they were called BOYS. One was called the YOUNGER BOY and one the OLDER BOY, but of course some people used other words too, like GUY or KID or YOUNG MAN, and there are any number of ways to find the generic concept that fits.

10

WALLACE CHAFE

Does this entail any consequences for the general linguistic model? - It suggests the following modifications of the model which I had in the book. One would no longer start with what I called semantic structure. If one is trying to model the performance of the speaker, then one would start with the speaker's holistic concept, that is everything he wants to say. Of course this may change as he begins speaking, but at least at the time he begins speaking he has something in mind that he has to say. He needs to subconceptualize that until he gets to the level of appearances, and then find ways of communicating it by means of individual and generic notions. I am presently trying to work out a model which evolves more along that line. The speaker is going to be influenced by three major factors. One of them would be subconceptualization — analyzing experiences into concepts which include smaller concepts. A second would be finding the appropriate ways to communicate these things, and that involves more than just finding the generic concepts. It has to do with lubricating the verbalization process, so as to make it easier for the hearer to identify what appearances the speaker has in mind. And then the third factor would be what I have been calling syntactization, which introduces all those arbitrary elements of a language such as word order, concord, etc., which are presumably there mainly because of the history of a particular language and are not conceptually so relevant. So all three of these things - subconceptualization, and then identifying these small concepts as instances of generic concepts, and syntacticization - all three are intermingled in the process of producing an utterance. Possibly - 1 don't think that it is entirely true - most of the subconceptualization, the really conceptually relevant part of this whole verbalization process, comes near the beginning or tends to come near the beginning; whereas a lot of syntactic processes which are more surface-like will come towards the end: the arrangement of words and then the morphology of words, etc., are all late in the verbalization. I am saying this because it is possible that the kind of representation of semantic structures I had in my book would be sort of midway in this total scheme. The conceptual filtering has already taken place at this point, but the arbitrary syntactic rearrangements haven't taken place yet; that is the kind of thing that I was calling transformation. Another aspect of this which I should bring out is the most crucial, namely that in addition to these processes of verbalization, there is always background knowledge that has been laid down in the mind of the speaker, in his lifetime or in the lifetime of the human race. Verbalization takes place against the background of what is usually called knowledge of the world, human knowledge. We need to distinguish between what is actually verbalized and the knowledge which lies back of that and influences it. This would mean, for example, that what people have formalized in terms ofperformatives (I am speaking of Ross' notions, particularly of performatives) 7 really do not belong to the verbalization process. They are not part of what you say ; 7

J. R. Ross, "On Declarative Sentences", in R. A. Jacobs and P. S. Rosenbaum (eds.), Readings in English Transformational Grammar (Waltham, Mass., 1970), 222-272.

WALLACE CHAFE

11

you do not really say: "I say to you that...". They are a background for the speech act. And that is true of a lot of other things, like presupposition and all kinds of logical considerations which really are not in the verbalization as such but are outside of it. Do we have to represent performatives and presuppositions in the underlying semantic structures or not ? - I am suggesting that there are two kinds of representations. There is the formal linguistic representation, which is what is actually said, and then there are all these other elements which strongly influence what is actually said but are not really a part of it. It seems to me this is one of the things that I understand least and would like most to know more about: how should this background knowledge be formalized or can it be? A lot of it is experiential, consists of mental imagery and emotions, etc., but there are other aspects of it which are more amenable to formalization, I suppose, in logical terms. One immediately thinks that the way to formalize this whole background area is in terms of logic, but I wonder whether that is really the right way to look at it. Logic is just another kind of language; it is a language that leaves out a lot and therefore allows you to do things that you could not do otherwise. Even among the people who use the terms 'performative' and 'presupposition', there is an opposition between those who use them as logical notions and those who consider them to be pragmatic notions... - That is exactly what I am concerned with. The term pragmatic coincides reasonably well with what I have been calling experiential. When I say experiential, I am thinking of the things which lie outside what you would formalize in terms of symbolic logic or whatever. One of the reasons for thinking that a lot of this background knowledge can be formalized in logical terms is that people seem to be able to perform logical operations; people do make inferences and deductions in their minds, apparently. One might then think that at least a large segment of human knowledge ought to be statable in some kind of logical terms. I have been interested in the possibility that reasoning is conducted at least to some extent in terms of imagery. There is a line of investigation that is very interesting from this point of view. People have been asked 8 to perform particular logical operations on what has been called a 'three term series'. This is something where you say "Tom is better than Dick", "Dick is better than Harry", "who is best" ? And then the other person answers who is the best. People who have done this fairly consistently report that they set up a spatial model in their minds, so that if the first sentence is "Tom is better than Dick", they arrange Tom and Dick at the ends of a line, either horizontal or vertical; when they are then told that "Dick is better than Harry", they relate Harry to this first array that they had already constructed. In order to answer who is best, they simply 8

Cf. J. Huttenlocher, "Constructing Spatial Images: a Strategy in Reasoning", Review 75 (1968), 550-560.

Psychological

12

WALLACE CHAFE

look at the total array, where they have three positions, three points, and see which point is on top or furthest left or whatever it is. People report that that's what they do when they introspect about it. Manipulating objects in space, different colored blocks for example, seems to be subject to the same kinds of things. Some operations are easier to do than others, in both the logical problem and in arranging blocks. There seems to be a coincidence between the spatial manipulation and the logical operation. I am personally rather fond of this idea that at least some reasoning is done not in terms of logical formulas but rather in terms of forming visual or spatial images. Maybe a lot of operations are performed in terms of logical symbols. But once you learn logic, you learn how to manipulate symbols and people may in fact have visual images of the symbols that they are manipulating. There may even be auditory images. This is relevant to the question you raised, namely the experiential versus the conceptual or the pragmatic versus the logical part of thinking. But didn't you clearly distinguish in your book semantic units on the one hand and mental images on the other ? - I think that is consistent with what I was saying. Semantic units are like concepts; then there is this whole other area of thought which is in terms of images rather than concepts. So the semantic structures that I was dealing with in the book were divorced from imagery. I say now that imagery plays a very strong role as a sort of background for the building up of these structures. Do the experiments you described yield a hierarchy of lexical items and of syntactic categories ? - Hierarchy, in what sense? For example, in connection with the relation between nouns and verbs, between adjectives and nouns ... - 1 made some point in the book about the centrality of the verb and I still believe that verbs are different from nouns. Verbs are particularized in time, whereas nouns seem to have an existence in space. Verbs are temporally particular and that gives to whole sentences that kind of particularity. The verb has tentacles which pull in nouns of different kinds: an agent, a patient, a beneficiary. So the verb is like the nucleus of the sentence. That does not mean that it is more 'important', because maybe the verb is just there to tell you things about nouns or to bring nouns together; that seems to be its function. As for adjectives, I guess it is because of my background with AmericanIndian languages, particularly Iroquois languages, that I have never thought of them as forming a separate class. In those languages they simply do not form a separate class; there are just verbs and nouns, and if you want to say "Be thick" or "Be big", you just use another verb.

WALLACE CHAFE

13

At what point in the derivation were adjectives brought up in the verbalization experiment that you organized? - 1 think of them in general as coming the same way verbs do, but that would only be true with predicative adjectives as when you want to say "The banana is big". What you had in mind probably were attributive adjectives in particular: "A big banana", "a yellow banana".... That goes together with the question of relative clauses. What I think can be put in these terms: I was talking about finding a generic concept which was adequate to identify some appearance or individual that a person has in mind. Now, it might be that a generic concept like 'boy' would be too vague or broad for the purpose. You would like to narrow down the generic concept. Think of it as a large circle which could be compressed in a smaller circle; adjectives and relative clauses perform that function. You can say "He handed the banana to a small boy". Well, in that context it is better to say that than to say "He handed the banana to a boy". "Small boy" includes less experience than "boy": it is a way of constructing a smaller generic concept than any that you have available, in order to help the hearer to orient things in his own conceptual space. Can you say that lexical insertion and syntacticization have a definite position in the psychosemantic derivation; does lexical insertion always happen before syntacticization ? - No, I don't think so. What I called surface structures in my book did not have any phonological content; they were more like filtered and distorted semantic structures. After such a structure had been formed, then phonological material was substituted for all these surface structure elements all at once. I like to do things all at once when possible because it is easier for human beings to cope with something if they can do it that way instead of doing it in stages. That was the process I called symbolization: symbolization is the substitution of phonological material for semantic or semanticlike material. If I understand lexical insertion to mean the insertion of phonological material, then what I did in my book was to do that only after surface structure had been reached. There is another aspect to this question which is perhaps more important, namely the decomposability of words; that is, whether we should have a structure which contains smaller elements than a word. They might form part of a tree and then you would substitute a word for them. That is what lexical insertion involves. My model is rather the opposite. It is more like going from the larger elements to the smaller rather than from the smaller to the larger. If you had a word for each of those global events, then you would like to be able to communicate with those large concepts. But because of the way knowledge is shared between human beings, it is necessary to analyze things conceptually. However, you would not want to analyze in the production of an utterance below the level where you could communicate. This doesn't mean that you would not know that KILLING means CAUSING SOMEBODY TO BE DEAD; you would know it, but that knowledge would be in the background area and not part of the verbalization process. There is a concept KILL which is very salient conceptually,

14

WALLACE CHAFE

but knowledge which lies in back of that and is related to other things such as being dead and so on is not part of the verbalization itself. Is syntacticization always the last step of the verbalization process ? - No. I was suggesting earlier that a great deal of it comes at the end, but not all. Maybe I can give an example. The syntacticization which is relevant to the formation of sentences, the analysis of larger units into sentences and clauses, comes at an earlier stage. If you are going to tell a fable, you have the whole thing in mind. The first step would be to see that there are two parts of this fable: one is story and one is a moral. Having subconceptualized the fable into two parts, you would then decide to tell the story first and the moral second. This probably is syntacticization, putting things in the right order, syntax at a highest level. Then you get within the story which contains various parts that have to be ordered; the order has some conceptual relevance but is partly arbitrary too. Gradually you construct sentences, you now decide what to put in each sentence, where the sentence boundaries would be, etc. A lot of that is syntax and some of it has to come in fairly early in the verbalization process. Pronominalization, the deletion of certain items, could in fact be thought of as coming as part of the identification of individuals rather than as a late syntactic process. When you have a particular boy in mind, after you have identified him in the discourse, you can call him "he" the next time. One way to look at this is to say that you never enter the item "boy"; what you have in mind is this particular individual and you do not have to worry about finding a generic concept to communicate it after this point because you can just use a pronoun, which is a way of representing this individual directly. This view presents some problems because pronominalization depends in part on the establishment of word-order or of linearity. So at the point where you decide to use a pronoun, you have to know something about the linear arrangement already. That suggests either that you have to anticipate what is going to happen in this process or that it is a kind of back and forth sort of thing. It would be a problem if we had to think of everything in terms of step one, step two, step three, occurring in a rigid order like that. In fact, George Lakoff has introduced the idea of global rules,9 and I think they are psychologically correct. The speaker knows all the time how it is going to come out and he has some idea of the linear arrangment that influences verbalization. Is it possible to handle discourse phenomena with this approach ? - Yes, of course. That is one of the things 1 have especially in mind when I am thinking of, let's say, someone who is going to tell a story. He does not get down to the level of the sentence until he has done a good deal of subconceptualization. Narrative structures or the arrangment of events that you want to tell about, that is all very relevant. This sort of larger structure which you find in a story, etc., is one aspect 9

G. Lakoff, "Global Rules", Language 46 (1970), 627-639.

WALLACE CHAFE

15

of discourse structure. There is another aspect: the elements that have to do with facilitating the communication process and that depend on knowledge which is carried on from one sentence to another, the definite article in English for example, or this whole business of given versus new information. Both of those involve the speaker in assessing what is in the mind of the hearer, in doing certain things that make it easier for the hearer to be able to identify appearances in individuals, etc. Let's talk about the definite article. There are lots of ways in which definiteness can be introduced in a discourse. The main thing is that having introduced early in discourse an instance of some generic notion like "a boy", if one wants to talk about the same boy later, one might again use the same generic concept boy, but this time he will say "the boy". That means he is talking about an instance of boy which the hearer is able to identify on the basis of the context. So at the start you have an individual and you are looking for a generic concept to use, whether you call him a boy or a young man or a kid. You pick a generic concept which is salient for this individual at this point in the discourse. It works also in the other direction: given a particular concept, there may be a particular individual who stands out as most salient within that generic concept. If I mention "boy" later on, there will be one particular instance of boy that I can assume stands out in the hearer's mind. If that is the case, I will say "the". If I wanted to talk about this object in front of us, the most natural generic concept would be "table". I can assume that there is one instance of table that stands out in your mind at this time just because of where we are located; so, of course, I will talk about "the table". Obviously, such phenomena go across the boundaries of sentences. How can you make the notion of context operational in psychosemantic research? Do you need a typology of contexts ? Is there only situational, or also verbal context ? - I would prefer to talk about the linguistic and the extralinguistic context. If I could use "the table" in talking to you about this object, it is not because of anything we have said - we have not talked about this table - but it is because of the situation. If I talk about "the theory", well that of course comes out of the linguistic context. Yes, there seem to be at least those two basic categories. There are also the psychological context, the ideological context... - Sure, the whole sociological setting of discourse is important. All those parameters are part of this background knowledge that I mentioned. They affect what is said in all sorts of different ways, for example the way the speaker and hearer are related, the sociological way and the status and whatever. How to find out the linguistic relevance of this contextuality ? - One could say at least that the knowledge a speaker has, which he may want to communicate, is at least of three different types. One type would be factual knowledge: he has perceived it directly himself. A second type would be things that he has heard from other people, knowledge that he has acquired linguistically. A third kind would

16

WALLACE CHAFE

be knowledge that he has invented within his own mind through imagination or through reasoning or what I might call hypothetical knowledge, including predictions of the future. Maybe we should distinguish other types. Some of the languages that I know something about seem to make distinctions along these lines. It is well known that in some American-Indian languages there are things that are called evidentials, evidential prefixes which have to do with the status of knowledge: you would use a special kind of prefix in talking about something which you heard from somebody else; it is like saying "allegedly" or "people say" rather than communicating something you had directly seen and believed very strongly on that account. That is a good example of how linguistic facts can shed some light on how the mind organizes things. If a language distinguishes between three kinds of knowledge, that is a pretty interesting thing about it, especially if you find it in a number of different languages. In your earlier work, you had a reaction of dismay regarding psychological classifications. Why is it that psychological classifications and linguistic classifications do not coincide ? - I have been reading more in the area of psychology over the last few years and what has struck me more than anything else is the contrast in the methodologies in the two fields. In fact it results more from the history of the two disciplines than from anything that would be inherent in the material. A psychologist is expected to perform experiments all the time; if a linguist performs an experiment then he is thought of as acting like a psychologist. In linguistics there was a fairly explicit methodology during the structuralist period; there was much self-consciousness about how you did things. Transformational grammar has not really been as explicit about methodology: it mainly involves sitting in an armchair and thinking about sentences. So we are dealing with three different methodologies: in psychology, the experimental methodology; pretransformational linguistics, which involved observation of facts, and in that way was very similar to anthropology; and then there is a third kind of activity which transformational linguists have engaged in, and which is much less clearly defined. I have been very much in favor of introspection. There seems to me more variation than I originally thought there might have been; however it is encouraging that when different people perform introspection, they come out with similar results. As William James said about psychology,10 we have to work with introspection, even when we make a lot of mistakes, even if we often might come out with things that are wrong; this is true of any kind of data collecting. We should just realize the fact that we are going to make mistakes, but with good will and constant checking and rechecking, introspection can lead to more useful results than anything else. Do you see any fundamental difference between your criterion of introspection and the transformational notion of grammatical intuition? - I think what they mean by intuition is similar to what I mean by introspection, but II

W. James, The Principles of Psychology (New York, 1890), Vol. I, pp. 191-192.

WALLACE CHAFE

17

at least in a lot of transformational work, I have two feelings about it. One is that the introspection is fairly unsystematic, the formalization is systematic but the observations are not. The other feeling I have is that transformationalists tend to be very insensitive to subtle linguistic differences. People do vary a great deal in sensitivity, I am convinced of that. There is a tremendous variability in verbal consciousness. Is collecting data from a corpus of observable facts not a typical empiricist or behaviorist methodology ? - What structuralists used to do was to go out with a notebook and a pencil, collect a certain amount of data, from that form a corpus, and then try to construct the grammar on that basis. Of course this method contains serious limitations. The ideal thing is to either be a native speaker of the language yourself or to have a verbally sensitive informant who is constantly sitting next to you and whom you can question all the time instead of going off to some place three thousand miles away. Data collecting is also very much influenced by your theory. In the early sixties I worked with an Indian language and collected a lot of material which I have filed away now, but my whole view of what is interesting in language has changed considerably. I find now that I would like to have a lot of data of a different kind, which is no longer so easy to get because a lot of people I worked with are no longer alive. So clearly your theory influences your data. One of the most fundamental assumptions of your model is that the primary function of language is communication... - When I use the term communication, I mean it in the broadest possible sense: the speaker has something in his mind that he wants to introduce into the hearer's mind. However, it does not have to be for the purpose of communicating knowledge in the narrower sense: informing the hearer. There are several reasons why language is used: one of them is just to communicate knowledge in that way, but another is to ask questions or to cause the hearer to perform some action. Language is sometimes used as an act in itself, and I have in my mind performatives like "I promise", "I christen this ship", etc. Language has a variety of functions which I want to include in the word communication because all of them involve some transfer of information from the speaker to other people. Another assumption of your linguistic theory is that language has to be considered as a link between thought and meaning. - I had the view that language provides basically a link between two nervous systems which are not directly connected through neurons; it brings one nervous system closer to another. We have messages travelling around within each of our bodies, but between any two of us, there is a gap. Language provides the bridge. It does it by making use of sound; whatever form knowledge has while it is in an individual, it cannot pass over to another individual in the same form, so it has to be converted to

18

WALLACE CHAFE

some other form. Sound seems to have some advantages over other media, although visual representation has also some advantages that sound does not have, for example in written languages. Language is equally important in thinking, that is in what goes on within a single mind. I do not think that the two aspects of language are really separable, they are just different sides of the same thing: a lot of thinking could amount to manipulating linguistic elements within your own mind. Thinking would first of all include whatever is involved in imagery and second the manipulation of linguistic elements. Interestingly enough, people seem to differ in the amount of each kind that goes on within their heads: some people say they spend all their time verbalizing and others do not. That is what makes introspection difficult, I suppose, because there are different modes of thought. Do you see any continuity between vocalized language and other non-vocalized sign systems, can you consider sign systems, other then vocalized ones, as language? - Of course it all depends on how you use the word language. You can use it to include gesture systems or traffic light systems, but it is probably best to restrict it to what we usually mean by language. The other systems are clearly much more primitive than language - the remarkable thing about language is its tremendous complexity and variety. All other systems that I know are very poor indeed, and the same is true of all animal communication systems. They allow only a very small number of messages to be conveyed, they lack a number of other features of language, and so I prefer not to call them languages. Is there any similarity between gesture and vocalized language, and on the other hand between animal communication and human communication ? - There may be some continuity from the point of view of the evolution of language: prehumans probably had a system which was more or less like what other primates have today, various sounds and gestures which are discretely different from each other and which permit communication of small discrete messages. Somehow the number of possible messages began to grow and what was animal became human. As the number of messages grew, reached a certain threshold, the system itself seems to have changed. The one to one correspondence between message and symbol, message and sound and gesture, was no longer viable. Also with the growth of the system it became possible to combine the large number of concepts within sentences and discourse in a way I do not think other primate systems allow for. I like to think that language has in common with other primate systems only this symbolization aspect, namely that at some point conceptually oriented material is replaced with sound or gesture of some kind. More Cartesian oriented people object that this is only a continuity of the general intelligence but that language involves special psychological faculties distinct from general intelligence ...

WALLACE CHAFE

19

- Language does involve certain specific faculties that probably other animals do not have; the human brain is organized in a way that makes language possible. It is able to cope with this really cumbersome syntactic apparatus that language has, with the fact that things are not represented directly and concepts have to go through all this filtering and distortion that language requires before they ever get turned into sound. Also idioms are possible - I do not know whether other primates could learn to use idioms in the way we do. You stated that semantic units and structures cannot be equated with surface. What do you mean by that and what are the characteristics of this very deep semantic structure ? - One of the big contributions of M.I.T. linguistics has been to point out that surface structures do not convey meanings directly, that a lot of conceptual material is not included in the surface structure. I saw this most clearly in terms of idioms, and it has always surprised me that transformationalists still do not seem to understand how idioms operate. Idioms provide the best example of the discrepancy between conceptual structures and surface structures. If I say "Somebody hit the roof", it means someone became angry. It is clear that the elements in the surface structure do not express directly what is there conceptually. Not only are there lexical idioms but also idioms of an inflectional sort such as the progressive aspect, some forms of BE or HAVE or -ING or -ED at the end of the verb. Those are really idioms too and the surface structure elements do not reflect directly what is present conceptually. If on the other hand you say "He hit a roof", the whole idiomatic meaning evaporates. That is easily explained if you think of a conceptual structure in which there is an item which means something like "become angry", and where there is just no possibility to use the indefinite article because there is no semantic item roof. The idiom is just a first stage symbolization of what is there conceptually. Metaphors are a very difficult problem for structural linguistics but also for transformational grammar. Can they be treated in the framework of your theory ? - In my book I talked about two different kinds of idioms, the first kind being those we have discussed. But you also have single conceptual units which you use in a strange unexpected way in a particular context. "Make the bed", for example, where MAKE has not the usual meaning of "to build, to construct". It has a sort of a idiomatic meaning which you only find in the context of BED. There are a lot of other examples of that sort which I called restricted idioms, which is more or less a synonym for metaphors. However, metaphors can also be used creatively, not only in a sort of frozen way, as in "make the bed". Authors can use them for particular purposes and do not really follow a standard usage. That creative aspect of metaphors presents a real problem and 1 do not know how to cope with it now. You used the very new term 'symbolization'. Is there any difference between on the one hand the relation message and symbol, and on the other hand the old dichotomies of

20

WALLACE CHAFE

content and expression in Hjelmslev, or of signifiant and signifié in Saussure ? What are the fundamental characteristics of a symbol ? - 1 think those are essentially parallel distinctions. We have things going on in our nervous system which we cannot transmit in their original form. They have to be replaced by something which will bridge the gap between individuals. What I meant by 'symbolization' is this replacement of conceptual material by phonetic or phonological material. That really is the same thing as the replacement of signifié by signifiant. Isn't there a more dynamic perspective in that you see symbolization as a process rather than as a static phenomenon ? - Yes, that is true. I have not used the term symbol, rather the term symbolization which is the process from signifié to signifiant. That is the basic difference in the use of the term. Otherwise when I say phonological material, this is the same as expression or signifiant. You introduced the concept of duality. Can you explain how it relates to your conception of language ? - Duality is not really a new idea. The term was introduced by Charles Hockett, 11 and Martinet in France has talked about the double articulation du langage.12 The number of conceptual units in a language is so large that there could not possibly be a different sound for each concept. So you have a sound system consisting of a small number of sounds, and the different arrangements of these sound units are used to symbolize conceptual units. That is what duality involves. In the derivation of an utterance various processes are at work : formation of structures, transformation in other structures and symbolization. Are those processes only formally different or are there substantial differences between them ? - I would say to begin with that I have given up the idea of formation processes all together; this is one of these areas where 1 was misled by M.I.T. linguistics or I wish I had not gone along with them as closely as I did. I was trying to provide ways of generating well-formed semantic structures in sort of an analogous way to what is done in transformational grammar. I do not think it has any validity in language at the present time. What I have in mind now is a generation process which I hope will reflect the actual processes that the speaker goes through in verbalizing his experience. So we are concerned with the three basic processes I mentioned earlier: subconceptualization, and I did not give a word for the second kind but I am thinking of identification, and then finally syntacticization ; a fourth one would be symbolization, which is the introduction of phonological material. All four of them are of different types. The first is a matter of analyzing larger concepts into smaller concepts. The II 12

C. F. Hockett, "The Origin of Speech", Scientific American 102: 3, 89-96. Cf. pp. 236-237 of this book.

WALLACE CHAFE

21

second is looking for ways of communicating appearances and individuals - I could include within that things like definite and given, which are just ways of making these processes easier for the hearer. Syntacticization is the closest thing to orthodox transformation; it would involve several different kinds of subprocesses or subtypes of syntacticization. One of them is simply linearization, that is putting things into the right order. Another is agreement or concord; let's say you move certain features of a noun into a verb, and things of that kind. Symbolization is just a matter of taking a conceptual unit like KILL or a surface structure unit which may or may not be directly functional, and replacing it with a string of phonemes or some matrices of distinctive features. So clearly the four processes are different in kind. Very central for your view on language in the book and afterwards is the notion of concept. Do you consider yourself to be in the tradition of ideational theories or are you identifying concepts with mental images ? - I was in favor of the ideational theory; in the book I was reacting particularly to Alston's criticisms of the ideational view. Alston 1 3 said that you could identify the notion of concept with the notion of mental image, and that you can't have a concept DOG without it being tied to one particular mental image, which is certainly not true. So when I was trying to separate the experiential kind of knowledge from a conceptual kind of knowledge, it was exactly that kind of problem I had in mind. An individual is constantly being bombarded with material from the outside world through his senses. Since he has to interpret this material in some way, there apparently are several things he can do. One is to draw a line around the things that come in. You see something; you delimit various objects. To a certain extent, these lines are already there and we tend to delimit objects according to nature. But nature does not provide everything; there is a good deal of interpretation that goes into this kind of processing. This is what is involved in perception. So that is one thing: you can form a concept of some object at a particular time and place; that is what I was calling an appearance. That is one kind of concept. I think of this as just the idea of that thing at that time, and this kind of concept has spatial and temporal coordinates. Another kind of concept would be generic, which does not have a spatial and temporal coordinate, and that would be my idea of, say, a microphone in general. So there are some concepts which would correspond to mental images and other concepts which would not. I am using the word 'concept' for both of those things, and I hope that does not prove to be confusing. I find it useful, however, because they are each some kind of unit that the mind forms in order to cope with the sensory material that keeps coming into it. They seem to be units of different kinds, but they have in common the fact that they are discrete units. If I understand well: the concept is not always abstract? - That's right. What I was calling appearances are not abstract, whereas generic 13

W. P. Alston, Philosophy of Language (Englewood Cliffs, 1964), 22-25.

22

WALLACE CHAFE

concepts and all individuals are abstract because they are not pinned down in time and space. If by 'abstract' we mean not associated with a mental image, then concepts are not always abstract. Of course, there is the very old question whether you can have images of generic concepts; it goes back to Locke at least, and has been discussed ever since. I tend to think that it is in a sense possible to have mental images of abstract concepts. Do you still think that images are the manifestation of the concepts ? - Actually I use the term 'manifestation' to describe the relation between concepts and experience. We could say that things in the experiential round are manifestations of concepts. It has seemed useful to me to have a term which is the bridge between these two areas. Are concepts to be considered as universals ? - Before Chomsky I was taught that languages could differ from each other, and that conceptualizations could differ from each other in an infinite number of ways. That is wrong. I think it is also clearly wrong to say that conceptual structuring is totally universal. The truth lies in the middle. There are lots of things that all human beings have in common, speakers of the same language have even more in common and so on, but there are still differences everywhere. It is wrong to ignore the differences and it is wrong to exaggerate them. It would be interesting to know whether conceptual universals are autonomous or whether they can be explained in terms of perceptual universals... - They are all tied together. There must be some perceptual things which are built into the organism, and for that reason are universal; this must be true also deeper in the brain. But I think again that this view can be exaggerated. We are born with a certain number of innate mechanisms which determine certain aspects of how we conceptualize, but a lot of the rest of it is dependent on the culture we grow up in. I mentioned in my book the finding about colors by Berlin and Kay. 14 Colors used to be taken as the most obvious case of linguistic relativity, where you could divide the spectrum in almost any possible way in different languages. But it seems now that this is not true. If we have a word for 'red' in different languages, the focus for that is going to be the same or pretty much the same for all human beings. So that seems to be something that is built into the organism. Shouldn't we look at linguistic universals as formal? One knows Chomsky's distinction between substantial and formal universals... - Again I can say the same thing: some of them are and some of them are not. The constraints on conceptual structures seem clearly to depend on larger aspects of human knowledge. That is, whether you can put together two concepts in a certain 14 B. Berlin and P. Kay, Basic Color Terms: Their Universality and Evolution (Berkeley and Los Angeles, 1969).

WALLACE CHAFE

23

larger unit depends on what you know about the world. We can say in general that human knowledge is partly the result of innate mechanisms and partly the result of learning. The same would be true of conceptual constraints. Let us discuss your conception of directionality in language. One of the most important arguments in favor of directionality is that well-formedness of sentences is semantically motivated... - It seems to me that what the speaker does is quite different from what a hearer does. It is wrong to regard the speaker and hearer as being symmetric; the hearer is not just a mirror image of the speaker. What the speaker does is to create utterances: he starts with prelinguistic material and goes through the kinds of processes I was suggesting earlier. Eventually he arrives at the phonetic representation and the sound then passes across to the hearer. The hearer does not have to construct anything, he is presented with something that the speaker has done and reconstructs what the speaker started out with. I think you can say the speaker is like someone who is knitting a sweater; what the hearer does is more like unraveling. But you don't have to know how to make sweaters in order to unravel them, so the hearer's job is easier; all the creative stuff has already been accomplished. Still, what a hearer actually does is poorly understood. I do not have a great deal to say about how the hearer is able to take this sound which the speaker has presented him with, and how he turns it into some kind of knowledge that he integrates in his own mind. I am sure that he uses shortcuts in doing this; he does not go through the whole process the speaker went through. The speaker does a lot of things to make it easier for the hearer to integrate things into his mind as accurately as possible: that is the role of things like the use of a definite article, past tenses, intonation patterns, etc. So well-formedness of sentences is dependent on what the speaker does, obviously not on what the hearer does. If you are concerned with the wall-formedness of utterances, it is something that has a decidedly directional aspect. Ambiguity too goes in one direction. You can have two different meanings which converge in the same phonetic representation; I do not think you find the same thing in the other direction. You do not find a single semantic structure which has a variety of phonetic representations. A lot of people, with this notion of paraphrase, have thought you did. Of course, if paraphrases worked perfectly, they would in fact be alternative phonetic representations of one and the same meaning. However, that seldom occurs. I am fairly convinced that it really never occurs. If things are different phonetically, they are always different semantically, at least to some degree. A lot of people have claimed that there is no directionality in language, that it goes from sound to meaning and meaning to sound equally well. It seems fairly obvious to me that language is not symmetric in that way. Why is the fact that the hearer has to decode, starting from the sound to meaning, not a counterargument to directionality ? - Yes, he has to do that, but he has to perform a different kind of operation than the

24

WALLACE CHAFE

hearer. He doesn't have to worry about well-formedness at all; the formation of the utterances has been accomplished by the speaker. You can start with a semantic structure and, if you apply the proper transformations, you arrive at a well-formed phonetic structure. You could not start with a well-formed phonetic structure and necessarily arrive at a well-formed semantic structure. Is your view on directionality different or parallel with what is said now on this question, for example in generative semantics ? - Lakoff once said 15 in print that he did not believe in directionality, that you can go equally well in both directions. I think this is a misunderstanding of what is involved. I tried to clarify this in my article on "Directionality and Paraphrase". 1 6 I was puzzled by the fact that the generative semanticists seemed to be accepting what Chomsky said on this question, namely that you could go from any level of representation to any other. In itself, of course, that is true, but the underlying question is: where and at what level of representation do you determine what is well-formed and what is not. Chomsky seems to leave that out of his argument completely and these other people seem not to have paid attention to it either. I have to confess some inability to understand why there should be any dispute about this at all. We have to conclude this dialogue. Before leaving I would like to ask you some more general questions. One of the oldest dichotomies in linguistics is the Saussurean dichotomy of langue and parole, language and use of language. Is a dichotomy like that useful and operational in the construction of a linguistic theory today ? - Well, I am not so sure. I earlier criticized the distinction between competence and performance because, as I said, it has been used to avoid a lot of interesting questions. To the extent that competence and performance correspond to langue and parole, I do not think the distinction is a very useful one at the present time. Tt seems to me more of a hindrance than a help. I think Saussure meant several things by langue and parole. It is not a unified dichotomy and there may be other aspects of it that would be more valuable to presserve at the present time. Speakers of a language have some kind of underlying system, and perhaps it is useful to distinguish between the underlying system and particular uses that are made of it on particular occasions. However, as soon as you accept this distinction, you seem to get into trouble very fast because people have the tendency to describe the system as such in isolation from any uses which are made of it, and I think that is what has been the problem. I am not sure that Saussure would have intended that. Maybe what has been made of his distinction has been a kind of injustice to him. 15

G. Lakoff, "On Generative Semantics", in D. D. Steinberg and L. A. Jakobovits (eds.), Semantics: An Interdisciplinary Reader in Philosophy, Linguistics and Psychology (Cambridge, 1971), 236-237. 16 W. L. Chafe, "Directionality and Paraphrase", Language 47 (1971), 1-26.

WALLACE CHAFE

25

Saussure says in his Notes that the study of language has to start from the study of the uses of language... - It would have been better if that insight had been preserved. Another very classic dichotomy is the one Bloomfield made between experiential data and language. Is this dichotomy operational? - I am not so sure about Bloomfield, but post-Bloomfieldians like Trager and also Bloch seem to have believed that there is a kind of self-contained system in language, apart from the experiential background. I certainly am expressing some doubts about that. There is not a kind of system within language that can be dealt with in isolation from the experiential background, since language is used to communicate. In generative semantics, in sociolinguistics of the Halliday type, in psycholinguistics, there is an explicit purpose to transcend the old distinctions made by Morris between syntax, semantics and pragmatics... - Morris' whole scheme indeed seems to have a very decided behaviorist orientation, and I think it does not fit very well with current ideas. Do you feel psychosemantics fits in this trend to transcend these old boundaries ? - Oh yes, very definitely. I am trying to transcend more recent boundaries too, I suppose. My very last question now. What could be the future of your line of investigation ? - What I want to do in the broadest sense is to investigate more deeply this area I have been calling psychosemantics, combining findings in psychology and findings in linguistics, and trying to bring about a closer convergence between the two disciplines. One of the things that needs to be done is to revive a methodology that is acceptable to both the psychologists and the linguists. Maybe it is a matter of doing experiments and observing informants, and also sitting in an armchair and thinking about sentences. My particular concern at the moment is looking further into this relation between the experiential and conceptual parts of human knowledge, and trying to learn more about the interaction between these two ways humans sesm to have of knowing things. I would like to understand the interaction between experience and concepts, and also to know more about whether experiential knowledge can be formally represented in some way amenable to the kind of scientific studies that we are used to. It is quite possible that it is really not amenable to any kind of formal representation of the usual sort, but I would like to know more about this question. I think it is a fundamental question for the kinds of things that are being done in linguistics nowadays. University of California at Berkeley Berkeley, Calif., U.S.A. October 11, 1972

NOAM CHOMSKY

Noam Chomsky, almost everybody considered the publication of Syntactic Structures in 1957 as a revolution in contemporary linguistics. What did it mean in 1957 to say that grammar is generative and what does it mean now? - A generative grammar is nothing but a system of rules and principles that specifies in a perfectly explicit way a set of linguistic objects, in general infinite. Both in the mid-1950s and today, it has been assumed that the linguistic objects generated are structural descriptions of sentences. In the version of generative grammar sketched in Syntactic Structures in 1957,1 the focus of attention, for heuristic reasons, was on the sentences generated. The background assumption, sketched briefly in the introduction, was that general linguistic theory provides a system of levels of representation and a general procedure with the following property: given a permissible grammar and the derivation of a sentence generated by this grammar, the general procedure assigns to this sentence a representation on each linguistic level, the set of these representations constituting its structural description. Later chapters discussed ways in which the assigned structural descriptions might play a role in the use and comprehension of utterances. The background assumptions were sketched out in detail in a still unpublished study called "The Logical Structure of Linguistic Theory". 2 It may be that this formulation was misleading. Thus I have noticed in the literature references to the alleged fact that in the early stages of generative grammar, the topic of study was generation of sentences, and that only later, interest turned to generation of structural descriptions. This is entirely wrong, as the introductory pages of Syntactic Structures and the discussion of explanatory power and of syntax and semantics indicates. Perhaps the matter will be clarified if I point out something not generally known. Syntactic Structures was not written for publication. It is basically a set of lecture notes for an undergraduate course at M.I.T. It would have been less misleading for "The Logical Structure of Linguistic Theory" to have been published, with Syntactic Structures as an informal introduction, perhaps. However, this was impossible under the circumstances of the 1950s.

1

N. Chomsky, Syntactic Structures (The Hague, 1957). N. Chomsky, "The Logical Structure of Linguistic Theory" (M.I.T., Cambridge, Mass. [unpublished, microfilm], 1955). 2

28

NOAM CHOMSKY

Can it be inferred from the infinity of language that grammar itself is a generative device ? - Since a language is infinite, it follows that the grammar that specifies and generates it, being necessarily finite, has recursive devices. The requirement that the grammar be explicit, insofar as possible, derives from an interest in confirmation, disconfirmation and explanation. Combining these requirements, we have generative grammar. Does your notion of creativity cover more than just generativity ? - There has been an unfortunate tendency to confuse what I have called 'the creative aspect of language use' with something quite different, namely, the recursive property of grammars. This is a conceptual confusion, a confusion of performance and competence, in essence. I have used the term 'creative aspect of language use', as the phrase implies, to refer to a property of the use of language, of linguistic behavior: namely, that it is innovative, unbounded in scope, free from the control of external stimuli or detectable physiological states, coherent and appropriate to situations, engendering in the listener thoughts related to those of the speaker, thoughts that the listener might have expressed in a similar use of language. These aspects of language use were noticed and discussed in quite interesting ways in the 17th through 19th century work that I reviewed in Cartesian Linguistics, 3 and elsewhere originally in the context of the problem of existence of other minds. The crucial properties, in this context, were not the infinite scope or freedom from stimulus control - the Cartesians were aware that a mechanical system could exhibit infinite diversity, and a device with a random element, though free from stimulus control, would not exhibit the behavior that would justify attributing to it a mind. Rather, the other properties were crucial in that investigation, and, I believe, they remain so today. The recursive property of generative grammars provides the means for the creative aspect of language use, but it is a gross error to confuse the two, as some linguists do. Is the concept of creativity supported by empirical evidence or is its use based on metatheoretical or ideological assumptions ? - The Cartesians sketched a range of empirical evidence that had bearing on this issue, particularly Cordemoy.4 No one, to my knowledge, has explored the matter systematically. I suspect that the reason is twofold: there have been what might be called 'ideological considerations' that turned attention away from these properties of language use; secondly, once the matter is raised and the issue faced honestly without ideological preconceptions, the point seems rather obvious. Few people are willing to undertake elaborate empirical investigations to demonstrate the obvious. Rather, the more urgent task seems to be to clarify the issues and to try to determine what is implied by the gross facts, which seem fairly clear when one observes the normal use of language without preconceptions, such as those of behaviorism. A 3 4

N. Chomsky, Cartesian Linguistics (New York, 1966). G. de Cordemoy, Discours physique de la parole (1666; 2d ed., 1677; Engl, translation, 1668).

NOAM CHOMSKY

29

scientist approaching the study of language use without such preconceptions would observe at once that even meticulous and extended analysis of 'situations' or 'contingencies of reinforcement' provides him with no insight at all into the 'causation of behavior', understood as the problem of specifying probability of response in terms of the history of reinforcement and present stimulation. He would, if rational, conclude that a rather different and more indirect approach must be taken if there is to be any understanding of these matters. For these reasons, I suppose, no one has undertaken a painstaking and careful empirical investigation of the question whether the ordinary use of language has the properties noted earlier. Does the qualification 'rule-governed'' not neutralize the conceptualpower of'creativity' ? - I don't think so. Quite the contrary. I think it can be argued that 'creativity', either in the sense intended in the expression 'creative aspect of language use' or in the more normal sense, with reference to acts or achievements that have, critically, an additional property of aesthetic or intellectual excellence, presupposes a framework of rule, in part, perhaps, itself a creation of human intelligence and will, in part a reflection of biologically determined mental structure. A sonnet or a string quartet does not lack 'creativity' because it is constructed within-perhaps modifying - strict forms. A painting is no less the product of the creative imagination because it observes certain canons of style and technique. The matter is a commonplace of classical aesthetic theory, often discussed in terms of the relation of genius and rule. If all constraints are abandoned, there can be no creative acts. Thus a person who, say, casually throws paint at a wall is not producing a work of art, whatever else he may be doing. Similar observations apply in the case of intellectual creativity, such as theory construction in science. There are conditions, unknown but surely operative, that determine what Peirce called the 'admissible hypotheses', the theories that count as 'true science'. These are theories that are in some manner matched to the structure of human cognitive capacities. They provide insight and understanding, not merely accurate prediction. The rules of abduction, in Peirce's terminology,5 make possible the scope of human knowledge while also determining its limits; the principles that specify some artistic form or style provide the necessary conditions for artistic creativity, while naturally limiting its scope. That the rules and principles, in both cases, may be subject in part to choice and invention is another matter - correct, but not inconsistent with what I have just been saying, repeating some familiar ideas; it simply raises the discussion to a higher level of abstraction, where the same considerations again apply. It seems to me that the same is true of the more mundane matter of the creative aspect of language use. The same considerations apply. Only rulegoverned behavior can be 'creative' in this narrower sense. This applies to the use of language and also the discovery or 'creation' of grammar by the language-learner, also a rule-governed act. Were it not, all hypotheses compatible with the data would 5

C. S. Peirce, "The Logic of Abduction", in V. Thomas (ed.), Peirce's Essays in the Philosophy of Science (New York, 1957); cf. also, K. T. Fann, Peirce's Theory of Abduction (The Hague, 1970).

30

NOAM CHOMSKY

be available as 'permissible grammars', no learning would be possible, there would be no communication, except purely by accident, among humans with comparable linguistic experience. Is the innateness hypothesis the necessary cornerstone of all rationalistic thinking about language ? - Not at all. Any non-vacuous theory of language, whether rationalist, empiricist, behaviorist, or whatever, must put forth an innateness hypothesis. The interesting questions have to do with the precise character of this hypothesis, not its existence as a necessary component of any non-vacuous theory. According to Searle,6 your conception of innateness implies a definition of man as a 'syntactical animal". Can you agree with this ? - Only in part. In my earliest work on this subject, I insisted on the obvious point that "there are striking correspondences between the structures and elements that are discovered in formal, grammatical analysis and specific semantic functions" and that these correspondences "cannot be ignored" but rather "should be studied in some more general theory of language that will include a theory of linguistic form and a theory of the use of language as subparts". I also suggested that "an investigation of the semantic function of level structure... might be a reasonable step towards a theory of the interconnections between syntax and semantics", and suggested briefly how this might proceed. A fundamental criterion in selecting formal theories, then, must be "their ability to explain and clarify a variety of facts about the way in which sentences are used and understood", their ability "to support semantic description".7 In later work 8 1 pointed out further that it is far from obvious that there is a "boundary separating syntax and semantics", and that the problem of determining such a boundary, "if there is one", "will clearly remain open until these fields are much better understood than they are today", adding that "exactly the same can be said about the boundary separating semantic systems from systems of knowledge and belief". Since 1 have always been quite explicit about this agnosticism, and have always insisted explicitly that a theory of use and understanding must be incorporated in any comprehensive theory of language, I am surprised that Searle and others have commonly assumed that I was advocating a restriction to the study of syntax. Probably the source of the confusion is a misunderstanding of certain technical problems discussed in Syntactic Structures and elsewhere, specifically, the problem of whether it is possible to provide semantic criteria for certain grammatical notions. However, there is no doubt that my own work is primarily on the general and specific properties of syntax and phonology, and I have attempted no serious construc6

J. Searle, "Chomsky's Revolution in Linguistics", The New York Review of Books XVIII (1972): 12, 16-24. 7 From Syntactic Structures. 8 E.g. Aspects of the Theory of Syntax (Cambridge, Mass., 1965).

NOAM CHOMSKY

31

tion of a systematic semantic theory. In this sense, Searle's comment could be rephrased as an accurate description of the limitations of my actual work. Nevertheless, I want to emphasize that I have always insisted that a full theory of language must concern itself with "the substantive and formal constraints on systems of concepts that are constructed by humans on the basis of presented data" 9 and that such efforts lead at once to an 'innateness hypothesis' for semantics. 10 Others, primarily Jerrold Katz, 1 1 have done provocative work in this area. Why does language have such a privileged position among other functions of the mind? What is its specificity in relation to those other functions ? - A serious answer to this question must await some results on other 'faculties of mind'. Unfortunately, there is no work in other domains of study of 'functions of the mind' that has achieved a degree of specificity and articulation comparable to work on language. Therefore, we can only speculate, and I have little to add to classical speculations on this matter, such as those I reviewed in Cartesian Linguistics, Language and Mind, and elsewhere. 12 This leads us to the question of the autonomy of linguistics. What does it mean for you that linguistics is a subpart of psychology and, at the same time, its paradigm ? - Hume set himself the task of discovering "the secret springs and principles, by which the human mind is actuated in its operations". This seems to me the fundamental problem of psychology, though many psychologists would reject the formulation. If so, then investigation of linguistic competence and its role in performance is a subbranch of psychology. It might be regarded as a 'paradigm' in the sense that, for various reasons, it has proceeded well beyond other investigations that might be undertaken. At this point, issues arise that can only be the topic of speculation, given the limits of our understanding: for example, are the properties of language and the innate structures that underlie language acquisition and use specific to a language faculty, or do they have significant analogues in other mental functions? My guess, for what it is worth, is that these properties and structures will be in part specific to language, just as the particular structures that underlie visual perception and learning are specific to this system. Others suspect that 'generalized learning strategies' 13 will be discovered that incorporate the 'language faculty' as a special case. In the absence of any coherent proposals, it is impossible to pursue the matter, I am afraid. 9

From Current Issues in Linguistic Theory (The Hague, 1964). See also Aspects, and further comments in Problems of Knowledge and Freedom (1971), and elsewhere repeatedly. II Cf. J. J. Katz, The Philosophy of Language (New York-London, 1966), and his The Underlying Reality of Language (New York-Evanston, 1971), and Semantic Theory (New York-Evanston, 1972). 12 Cf. Cartesian Linguistics and Language and Mind, enlarged version (New York, 1972). 13 E.g. H. Putnam, "The Innateness Hypothesis and Explanatory Models in Linguistics", in Synthese 17 (1967), 12-22. 10

32

NOAM CHOMSKY

Generative transformational grammar is Cartesian linguistics. Your choice of the Cartesian line of thought and your view on it have often been criticized. Maybe Kant or Leibniz would have been a better option ? - This question is based on a misconception. In Cartesian Linguistics, I discussed several topics, the creative aspect of language use, speculative psychology, the problem of explanation in linguistics, the theory of language form and use as a psychological theory, that were investigated in the 17th, 18th and early 19th centuries within a variety of conflicting and often competitive traditions. I pointed out at the outset that Descartes himself had little to say about language, and some of this I omitted, as irrelevant to the specific issues that concerned me, although some of the minor Cartesians, e.g. Cordemoy, did go into the problems of language more extensively. In particular, the Port-Royal Logic and Grammar, which form a conceptual unity, developed very important ideas about language largely within a Cartesian framework, as did subsequent work in 'philosophical grammar'. I also noted explicitly - a point that has been overlooked by many commentators - that "apart from its Cartesian origins, the Port-Royal theory of language... can be traced to scholastic and renaissance grammar; in particular, to the theory of ellipsis and 'ideal types' that reached its fullest development in Sanctius's Minerva."11 And I emphasized repeatedly that the contributions I was discussing derived from a variety of sources, and that many of those discussed would surely not have regarded themselves as 'Cartesians'. In particular, this is true of Humboldt, to whom I devoted more attention than anyone else in Cartesian Linguistics and elsewhere.15 But I did try to show that the work I was reviewing could reasonably be understood as in large measure an elaboration of Cartesian notions and approaches, though I also specifically referred to quite a number of significant departures. As for Leibniz and Kant, I did make some, surely inadequate reference to Leibniz in the works cited, and also occasional references to Kant 16 noting, as you do in the question, that the omission of a detailed study of Kant was a serious gap, among many others. While there are, as I emphasized, major gaps in this very fragmentary work, I do not believe that some other framework of study would have proven more perspicuous or revealing. Is the history of linguistics a heuristic preliminary for contemporary research in linguistics ? - Certainly not. In fact, my own interest in the work mentioned was aroused only after the basic outlines of most of my work on language were completed, and further work of mine and of many others does not derive from the earlier work reviewed. Nevertheless, I suggested, and believe, that ignorance of this tradition - or more properly, this collection of interpenetrating and conflicting traditions - is unfortunate 14 15 16

Cartesian Linguistics, p. 97. E.g. Current Issues in Linguistic Theory, Aspects of the Theory of Syntax, and later work. Cf. Problems of Knowledge and Freedom, p. 17.

NOAM CHOMSKY

33

and that it is interesting to see how problems that are once again arising in current work were studied in the period before the rise of contemporary linguistics. You view your own theory in the epistemological context of the traditional discussion about empiricism and rationalism. What is the most relevant criterion in order to distinguish between those two positions ? - I've suggested at various times that one may distinguish two approaches to the study of language and other cognitive functions, call them E and R, where E takes the 'operations of the mind' to be primarily a system of 'data-processing' supplemented with a system of properties for the initial analysis of sensory experience, and R, in contrast, proposes that the form of resulting systems of knowledge and belief - in particular, grammar - is predetermined by innate mechanisms and that the role of experience is primarily to differentiate and articulate systems that are predetermined in their general form and operative principles. I've also suggested that it is appropriate to regard E and R as reasonable 'rational reconstructions' of some of the leading ideas in empiricism and rationalism, respectively. I won't review the details, which are sketched in Aspects17 and elsewhere. Taking this approach, empiricist theories of learning - and the narrower varieties developed in the non-vacuous strands of behaviorist psychology - distinctive feature theory, the procedures of analysis developed in their most detailed and richest form in postBloomfieldian American structuralism, Quine's notion of 'quality space' and his associated ideas of 'learning of sentences' in Word and Object18 - but not his later, quite different views - and similar notions, fall within E; whereas rationalist psychology of the sort I discussed in Cartesian Linguistics,19 Aspects,20 and elsewhere, as well as the approach I suggested in Aspects within the framework of the general theory of transformational grammar, fall within R. Is your argumentation in favor of a rationalistic philosophy of language merely based on the existence of complex linguistic universals? - Surely not. Any non-vacuous theory of 'the operations of the mind' will imply certain linguistic universals. Thus suppose Quine is correct, in Word and Object, in describing a language as a system of sentences associated with one another and with stimuli by the mechanism of conditioned response. If so, we can derive at once certain universals. My argument is that the universals implied by the various proposals within E, as outlined in my previous comments and the references cited, are factually incorrect. That is, I have argued that if we take specific varieties of E and R as empirical hypotheses with regard to linguistic universals, formal and substantive, we 17 18 19 20

Aspects of the Theory W.V.O. Quine, Word Cartesian Linguistics, Aspects of the Theory

of Syntax, Part I: Methodological Preliminaries, pp. 3-62. and Object (Cambridge, Mass., 1960), pp. 5 ff., 10-13, 81-85. pp. 59 ff. of Syntax, Chapter 1, Section 8.

34

NOAM CHOMSKY

discover that the conclusions within E are incorrect and those developed within R seem much closer to the truth. Evidently, this is an empirical issue, once the conflicting theories are made precise. Quite a few commentators have argued that by reconstructing the issues as empirical ones, I have removed the whole matter from the traditional context of empiricist and rationalist philosophy. I won't pursue the matter here, except to register a disagreement. The traditional debate involved many issues, among them issues that we would regard as empirical. Thus the theories of perception (of triangles, for example) advanced by Descartes and Cudworth were, in our sense, empirical theories - though of course the philosopher who believed he had demonstrated by reason alone the existence of a G o d who does not deceive us, and who stated in his Principles of Philosophy that "there is no phenomenon in nature which has not been dealt with in this treatise", of course would have seen the matter differently. Hume, when he set himself the task of investigating the operations of the mind, hoped that he would be as successful as the "philosopher...who...from the happiest reasoning, ...determined the laws and forces, by which the revolutions of the planets are governed and directed". A s I read the classical literature, it did not distinguish 'scientific' and 'philosophical' issues, and was surely not concerned solely or even primarily with necessary truths, their origin and character, in our sense of this term. Thus I believe that the rational reconstruction I've suggested does in fact extract certain significant notions from the traditional debates, expressing them, to be sure, in a somewhat different fashion, and leading to conclusions which are in part at variance with all traditional views, as I've often stressed. In contrast, the more common approach which sees the tradition as concerned with topics that are 'paradigmatically philosophical' in some contemporary sense, and that takes the only issue to be the source and character of what we would regard as 'necessary truths', is distorting the traditional discussion. Has structuralism necessarily to coincide with taxonomy and so to be tied to a behavioristic or empiricist approach of language ? Glossematics, for example, seems to be a highly deductive and constructivistic, if not rationalistic, linguistic theory... - I think that behaviorism and structuralism must be sharply distinguished; structuralist theories often went far beyond the level of theory construction that would be permitted by the non-vacuous varieties of behaviorism. I do think that structuralism, in its existing forms, can be regarded as a subvariety of E, in the sense in which I defined this notion, and thus fall within this rational reconstruction of what I take to be leading ideas in empiricism. What structuralism must 'necessarily' be is a different matter, about which I have no opinion. I do not think that the question is welldefined. A s to structuralism and taxonomy, they must also be distinguished. Thus distinctive feature theory is not a variety of taxonomy, in any narrow sense. A s for glossematics, I'm afraid I cannot venture an opinion. I have been unable to discern in glossematics any clear substantive principles of language structure, any explicit empirical hypothesis with regard to the nature or use of language, which can be sub-

NOAM CHOMSKY

35

jected to analysis or verification within the terms that have interested me. As far as I can make out, it is largely an elaborate system of terminology. I say this tentatively. Much of the work in glossematics I simply do not understand. Does your interpretation of European structuralism, and especially of Saussure, not overlook an essential feature of their view on language, namely that they recognize the syntagmatic axis of language, i.e. its discursivity ? - I have never discussed European structuralist conceptions of the syntagmatic axis of language in any detail, the reason being that in the respects that interested me, these conceptions were more primitive and less precisely formulated than the theories of sentence structure developed within post-Bloomfieldian American structuralism. The latter, I have argued, fall within the framework E mentioned before, and are inadequate on empirical grounds. Therefore, the former also fail on the same grounds, a fortiori. I have briefly discussed Saussure's notions with regard to sentence formation. 21 It seems to me that he never escaped the traditional notion that speech consists of a flow of ideas corresponding to a flow of words. This assumption eliminates the subject matter of generative syntax, in effect. Adopting it, philosophical grammar regarded syntax as a rather limited study, since the 'natural order' of words in a sentence was outside its domain. Saussure never clearly challenged this doctrine, but rather reformulated it in his terms. Thus he assigned sentence formation to parole or to some border area of parole and langue, with an obscure status. Possibly his notion of mécanisme de la langue was intended to go beyond these limits, but it is difficult to reach any conclusion on the basis of the few rather vague remarks he left on this subject. In answer to your question, then, I don't feel that I have omitted something essential with regard to European structuralist views concerning the syntagmatic axis. Is the methodological aim of your distinction between competence and performance the same as Saussure's dichotomy between langue and parole ? - There is a close similarity, as I have repeatedly pointed out. There is also a fundamental difference, namely, I would like to regard a generative grammar as an account of competence, not merely a set of structured inventories with certain relations along the 'syntagmatic axis'. Now with regard to your relations to American structuralism, is Sapir present in your scientific 'genealogy' ? - Definitely. In particular, his emphasis on the crucial issue of psychological reality, and his notion of phonological orthography and ideal phonetic pattern, related by rules to the 'phonetic orthography', is taken over directly within generative phonology.

21

E.g. Current Issues in Linguistic Theory, p. 23.

36

NOAM CHOMSKY

Harris is certainly the American structuralist who influenced you most. What is the difference between his and your conception of linguistic transformation ? - The difference lies in the notion of 'generative grammar'. The rudiments of this notion are sketched in Harris' Methods of Structural Linguistics22 where they coexist uneasily with methodological assumptions that seem inconsistent with them. Harris's theory of transformations developed in a different context, namely, as part of the procedure of discourse analysis. Transformations, in this context, are a device for 'normalizing' sentences, that is, restructuring them into patterns that permit application of the substitution procedures of discourse analysis. As Harris's notion of transformation evolved through the 1950s, it remained a relation between sentences or sentence forms, not a device that formed part of a generative grammar. I believe that it would be fair to sketch the basic idea of Harris's theory of transformations in the 1950s as follows: the procedures of Methods are applied to a corpus, fully determining the grammatical structure of the language, its most abstract component being a system of morpheme-to-utterance formulas, a system similar in some respects to a phrase structure grammar for the actual sentences. On the basis of cooccurrence relations between sentence forms characterized in terms of these formulas, transformational relations among sentences are determined. My own approach to transformational grammar, borrowing extensively from Harris, developed within an entirely different general framework. As a student of Harris's in the late 1940s, I was working on a generative grammar of Modern Hebrew that was dissociated from any system of analytic procedures. Its morphophonemic component was a system of dozens of ordered rules, with a depth of ordering of20-30, as 1 tried to show in detail. 23 Its more rudimentary syntactic component was a phrase structure grammar with an elaborate system of indices carried along by phrase structure rules. When the notion 'transformation' began to be developed within the framework of Harris's work on discourse analysis - in which I participated as a student of his - 1 tried to incorporate it within the theory of generative syntax. And a few years later, when I was working primarily on the syntax of English, I worked out the system of transformational grammar that was presented in "Logical Structure of Linguistic Theory". 24 In this system, a transformation is not a relation among sentences or sentence forms, and the phrase structure grammar is not a grammar of surface structures of sentences. In contrast, in Harris's theory, a transformation is a relation among the sentence forms characterized by the morpheme-to-utterance formulas that define a kind of surface structure of actual sentences. Within generative grammar, the phrase structure component specifies a system of abstract forms that are mapped into surface structures by grammatical transformations, which are operations converting phrase-markers, in essence, labelled trees, into other phrase-markers. No sentence is generated by phrase 22

Z. Harris, Methods in Structural Linguistics (Chicago, 1951). N. Chomsky, "Morphophonemics of Modern Hebrew" (University of Pennsylvania, Philadelphia [M.A. thesis, mimeographed], 1951). 24 Cf. note 2. 23

NOAM CHOMSKY

37

structure rules alone, unless, by accident, no obligatory transformations happen to apply in some case. I should add that my own work on these topics, though it diverged in the respects noted from Harris's, was influenced at every point by continual and extensive discussions with him. It would be impossible to trace the innumerable contributions to transformational-generative grammar that derive from his suggestions and comments, even though this was not his central interest and his own work proceeded along somewhat different lines. Do you see a relation between Bloomfield's definition of meaning as an extra-linguistic entity and your own conception of semantics as a mere 'interpretive' component of grammar ? - Not really. Bloomfield's approach to meaning seems to me very different from any approach that has been developed within generative grammar. In Syntactic Structures I assumed a 'use theory' of meaning under the influence largely of Wittgenstein and Oxford philosophy. Subsequent work of Katz, Fodor, Postal and others suggested the possibility of a semantic theory as a more integral part of grammar, and I adopted this view in Aspects. I would question your use of the word 'mere' in referring to semantics as an 'interpretive component' of grammar. The term suggests to me a misunderstanding, common in recent discussion of meaning and form, which I have discussed at some length in my Studies on Semantics in Generative Grammar,25 At the risk of oversimplification, let me try to sketch out the issues as I understand them. Suppose that we accept the notion of 'semantic representation' or 'logical form', a representation of the meaning of a sentence in some universal system of representation analogous to universal phonetics. Suppose we assume further that semantic representation is related by rule to the derivation of a surface structure, where a derivation is a sequence of phrase-markers. This much is common to work within generative grammar of all varieties, so far as I am aware. Various substantive questions then arise as to the nature of semantic representation and the universal system from which it is drawn, and the relation of this representation to other aspects of sentence structure. The theoretical outline developed in Aspects, borrowing heavily from Katz and Postal's An Integrated Theory of Linguistic Descriptions,26 and other work of the period, postulates that deep structure is mapped onto semantic representation by interpretive rules. This is now often called the 'standard theory'. Quite equivalently, we might say that semantic representation is 'independently generated' and then mapped onto deep structure by essentially the same rules, given a different intuitive interpretation. There is no issue here, merely a question of terminology. It is as if someone were to propose that 2 is not the square root of 4, rather 4 = 2 2 . Katz' Semantic Theory27 is a recent exposition of the 'standard theory', in this sense. 25

N. Chomsky, Studies on Semantics in Generative Grammar (The Hague, 1972). J. J. Katz and P. M. Postal, An Integrated Theory of Linguistic Descriptions (Cambridge, Mass., 1964). 27 J. J. Katz, Semantic Theory (New York-Evanston, 1972). 26

38

NOAM CHOMSKY

My own subsequent work, influenced by that of a number of students, in particular Jackendoff,'^8 led to what is now sometimes called the 'extended standard theory', which takes semantic representation to be determined by a mapping of deep structure and phonetically interpreted surface structure; again, the same relations can be described in different ways, if one wishes, with no modification of the empirical assumptions involved. I have argued at some length that so-called 'generative semantics', at least in its clearer forms, adopts the extended standard theory, weakening certain of its constraints and employing a different terminology. Thus LakofT's 'generative semantics' adopts the extended standard theory in effect, but drops the condition that transformational rules relating lexical items to representations of their meanings be segregated from grammatical transformations - i.e., that there is a level of deep structure in the sense of the extended standard theory. He also permits arbitrary relations, derivational constraints, between the phrase-markers constituting a derivation and between these phrase-markers and semantic representation, which includes the 'initial phrase-marker' and other items representing focus, presupposition, and so on. Considering just these three theories, the standard theory and the extended standard theory agree on the existence of deep structure, though as Katz has argued, nothing very profound is at stake here; they differ on the question of whether there is an independent relation between surface or so-called 'shallow' structure and semantic representation. The extended standard theory and generative semantics agree that there is an independent relation between semantic representation and 'more shallow' structures; they differ in that generative semantics is a less restrictive theory, permitting a vastly wider range of possible grammars. There are, then, two fundamental issues: are the standard and extended standard theories correct in postulating deep structure, as an 'intermediate' abstract system of representation with its specific formal properties ? Are the extended standard theory and generative semantics correct in postulating independent relations between semantic representation and 'shallow structure' or phonetically interpreted surface structure? There is a further question: whether generative semantics is correct in relaxing the conditions on derivations, not only permitting lexical insertion freely among transformational rules but permitting arbitrary 'derivational constraints',i.e., arbitrary rules, there being no restrictive characterization of such constraints. But there is no difference among these approaches, or others that have been made at all clear, with regard to the 'centrality' or 'peripherality' of semantic representation. Apart from unformulated intuitions and notational matters, I am not convinced that these various approaches differ with regard to the nature of semantic representation. Where clear, they seem intertranslatable in this regard. Can you understand that you are yourself sometimes criticizedfor having a too empiricist view on language facts? - I recognize that the statement has occasionally been made, but in each instance, 28

R. S. Jackendoff, Semantic Interpretation

in Generative Grammar (Cambridge, Mass., 1972).

NOAM CHOMSKY

39

so far as I am aware, on the basis of a serious misunderstanding of issues such as those I have sketched here. Generative transformational grammar is a grammar of the sentence. Is an explicit definition of the sentence necessary in linguistic theory? How could it be related to the definition of utterance and proposition ? - in Aspects and other formulations of the 'standard theory', it is proposed that the generative grammar of a language provides the systematic definition of sentence. 'Grammatical sentences' are those generated by the grammar; I omit here the matter of deviance and 'semi-grammaticalness', as developed in various ways since the early 1950s. The term utterance, then, belongs naturally to the theory of performance. Utterances are the items that constitute the linguist's corpus, the 'primary linguistic data' for the language-learner. The term proposition might be used within semantic theory, referring to certain sentence-like components of semantic representations. The property of grammaticalness is a theoretical notion, internal to generative grammar. The property of acceptability is determined by observation and experimental test, in principle. Acceptability of utterances is the result of the interaction of a number of factors: grammaticalness and structure as defined by the generative grammar which is a theory of competence, the structure of memory, perceptual strategies, etc. Though there may appear to be some debate about these matters, in my view it is entirely terminological, and has introduced no points of substantive disagreement, a matter that I discussed at greater length in Studies.29 Is the sentence to be seen as a unit of discourse and do we have to postulate grammatical elements within the sentence that depend on the discursive context ? - Actual discourse consists, by definition, of utterances, if we use the terminology just suggested. We might consider ideal discourse that consists of sentences, abstracting away from the factors that interact with grammar in performance, under the comprehensive theory sketched just now. In such an ideal discourse, certain elements of sentences are interpreted by principles that extend beyond the sentence, certain anaphoric processes being the clearest example. Of course, in actual discourse, interpretation involves factors that go beyond the discourse: e.g., presupposed common beliefs, situational context, etc. In my opinion, the reason why some interpretive processes, e.g., those relating to certain types of anaphora, may violate otherwise valid conditions on rules lies in the fact that these are really processes that apply to discourse - hence in particular, to sentences and utterances - not strictly speaking to sentences. What is the goal of linguistic theory: to discover, to describe or to explain ? Do you see, in relation to this, a difference between the aims of linguistic theory and of grammar? - In the terminology suggested in Aspects and earlier work, a grammar achieves the 29

Cf. note 25.

40

NOAM CHOMSKY

level of descriptive adequacy to the extent that it succeeds in characterizing competence, as there defined. A linguistic theory achieves the level of descriptive adequacy insofar as it provides a descriptively adequate grammar for each permissible human language. A theory achieves the higher goal of explanatory adequacy to the extent that it provides a principled basis for selecting a descriptively adequate grammar for each language on the basis of primary linguistic data of the sort that suffices for language acquisition. This theory is explanatory in the quite appropriate sense that it suggests an explanation for the specific judgments made by a person who has acquired knowledge of a language, 'competence' on the basis of primary linguistic data. The goal of explanatory adequacy, in this sense, is the goal of linguistic theory, as I see it. The goal of descriptive grammar is to achieve descriptive adequacy, in the sense just sketched. What, for you, is intuition and introspection, knowing that the data confirmed by intuition have to be translated into scientific generalizations ? - Intuitive and introspective judgments are the primary data for the descriptive grammarian, hence also the linguistic theorist. He may, if he likes, proceed to develop operational tests of one or another sort to try to sharpen, characterize or organize these data. Or he may proceed to construct grammars which form part of the more comprehensive theories that attempt to account for these judgments, and to construct a linguistic theory that aims for explanatory adequacy in the sense just described. Assuming that both courses are pursued, we would hope, ultimately, that the results of the empirical tests would be accounted for by the theory of grammar, and more deeply, by the explanatory linguistic theory. However, it must be recognized that there is no escape from the control exercised by presystematic intuitive and introspective judgments. Thus suppose someone were to propose that 'sentences' are utterances that can be produced in exactly three seconds. It would be a trivial matter to develop operational tests for 'sentencehood' in this sense, and to construct generative grammars that precisely account for the results of these tests. The theory would, then, exactly account for the data. The whole enterprise would be worthless, because the notions characterized would have no interesting relation to the presystematic intuitions and introspective judgments that are the basis for any significant work in linguistics, regarded now as a part of psychology. Dialectal variation is more and more stressed in the recent linguistic literature. Does this imply a development of the notion of grammaticality ? - I see no issue here. I see no reason to modify the proposal in Aspects that linguistic theory deals with an idealization, namely, an ideal speaker-hearer in a homogeneous speech community. Of course, in practice, each actual speaker-hearer acquires his linguistic competence in a speech community that is far from homogeneous. He will be exposed to dialectal and idiolectal variants, and may introduce idiosyncracies of his own. He may be exposed to variant systems that are so different from one another

NOAM CHOMSKY

41

as to be called different languages, though we must be aware of the commonplace and correct observation that the distinction between language and dialect, as these terms are normally used, involves socio-political factors of many sorts; the characterization of a 'language' as a dialect with an army and a navy is not far from the truth. To achieve any significant understanding of the interplay of language types and styles, dialects and idiolects, we must, I believe, proceed within the framework just outlined. In practice, I believe that descriptive linguistics adheres to this general principle, perhaps without conscious adoption of it. We could proceed to develop more complex and intricate notions of grammaticalness and acceptability for the actual real-life non-homogeneous speech communities. I cannot imagine that any issue of principle arises in this connection, nor does the literature, to my knowledge, contain any coherent suggestion that an issue of principle arises. Can we imagine an evolution of grammar such that it would not only be concerned with grammaticaly, but also with appropriateness and acceptability ? - Not only can we imagine this, but in fact all approaches to linguistic theory with which I am familiar attempt to deal with appropriateness and acceptability as well as grammaticalness. Plainly, this is true of the approach developed in Aspects. Much, probably the overwhelming mass of grammatical work in fact, proceeds by abstracting in accordance with the various idealizations just mentioned, on the dual assumption that, on the one hand, the idealizations are legitimate - i.e., it is a deep and important problem to determine the nature of underlying competence - and, on the other, an understanding of appropriateness and acceptability will require a more comprehensive theory that investigates the interaction of competence-grammar and other factors. In Aspects, the introductory discussion deals with acceptability judgments that can be explained by the interplay of grammar and certain perceptual strategies, some of them necessary for any finite temporal processor, some of them probably specific to humans, perhaps even specific to language. Bever has suggested that the principles of adjective order that determine the relative acceptability of "little brown jug" and "brown little jug" can be explained in terms of certain perceptual strategies, interacting with grammar; 3 0 if his quite plausible suggestions are correct, this result further justifies the idealization to competence-grammar, while taking a step towards the more comprehensive theory required for the study of acceptability. And so on. The evolution of language study should, I think, continue along the lines it has always followed, in its clearer expositions at least, namely, towards a study of langue as an abstraction and idealization - now reinterpreting Saussurean langue within the framework of generative grammar - and towards the related study of parole or performance as involving a complex interaction of many factors among them the grammar that is postulated as a theory of the competence of the speakerhearer. 30

Cf., e.g., T. G. Bever, "The Cognitive Basis for Linguistic Structures", in J. R. Hayes (ed.), Cognition and the Development of Language (New York-London, 1970), 279-352.

42

NOAM CHOMSKY

Your distinction between formal and substantive universals is well known. What empirical results have been attained in that connection in the last years ? Can your distinction be maintained as it was ? - The distinction between formal and substantive universals was proposed in Aspects as a useful heuristic distinction, and I think it is valid in the sense in which it was proposed. There has been substantial work, in my opinion, in developing theories of formal and substantive universals. I might mention, in phonology, the proposals developed in Halle and Chomsky's Sound Pattern of English,31 with respect to both substantive and formal universals. In syntax, Bach has made some interesting suggestions with regard to substantive universals,32 and Ross and others have done important work since the appearance of Aspects on conditions on rules and other formal universals.33 Emonds's theory of structure-preserving rules is another contribution in this regard.34 In semantics, the work of Katz, Jackendoff, Fillmore, many 'generative semanticists', and others, has contributed significantly, in my opinion, to understanding of both sorts of universals. I've reviewed sketchily some of the work since Aspects that appears to me significant in the papers in Studies,35 and the other references I have cited here are among the many important works that make substantial contributions in this regard. I see no reason to call the distinction into question - in the form in which it was presented. Though many of my colleagues feel differently, and naturally I respect their judgments, my own feeling is that there has been encouraging progress towards a comprehensive and well-founded theory of language that begins to approach the level of explanatory adequacy in important respects, incorporating explicit proposals concerning formal and substantive universals; and that work on specific languages - English in particular - is progressing towards the goal of descriptive adequacy. I think that there is a tendency in current discussion to misinterpret terminological and notational issues as though they reflected substantive disagreement, and there is a tendency to be overwhelmed by the mass of unsolved problems. Many express the belief that the achievements of recent work are primarily negative, in that they show the inadequacy of previous oversimplified proposals. I disagree with this judgment, in that it seems to me that there has been definite progress, and signs of potential 'convergence', in the reformulations and corrections of earlier work in current contributions to the field.

31

M. Halle and N. Chomsky, Sound Pattern of English (New York, 1968). E. Bach, "On Some Recurrent Types of Transformations", in C. W. Kreidler, Sixteenth Annual Round Table Meeting on Linguistics and Language Studies (= Georgetown U. Monograph Series on Languages and Linguistics, 18 [1965]); "Questions", Linguistic Inquiry 2 (1971), 153-166. 33 J. Ross, "Constraints on Variables in Syntax" (M.I.T., Cambridge, Mass. [unpublished Ph.D. dissertation], 1967). 34 J. Emonds, "Root and Structure Preserving Transformations" (M.I.T., Cambridge, Mass. [unpublished Ph. D. dissertation], 1970). 35 Cf. note 25. 32

NOAM CHOMSKY

43

Competence has been defined as the underlying knowledge of language of a native speaker. How can this knowledge be identified with the knowledge of the grammarian or has the relation to be seen more as a relation of inference than of identity ? - I can only repeat the position developed in Aspects, which seems to me correct. There is a systematic ambiguity in the way in which the term 'grammar' is generally used. We speak of the grammar as the system of rules and principles internalized by the speaker-hearer; the grammar, in this sense, constitutes his competence, leaving open the question as to how this abstract system is realized. In another sense, we refer to the system of rules and principles constructed by the linguist as a grammar; it is, then, a theory of the speaker-hearer's grammar, of his competence. A s long as we are clear about our use of terminology, I see no problem here. We may, of course, be wrong in supposing that the competence of the speaker-hearer has the fundamental properties expressed in our theories, but this is simply a special case of the underdetermination of theory by evidence that is a defining characteristic of empirical science. Similarly, knowing nothing about the physiology of visual perception, a psychologist might proceed to speculate that the visual cortex contains a system of devices that analyze presented figures in terms of line, angle, motion, etc.; in Cartesian Linguistics and elsewhere I have reviewed some of the very interesting speculations of rationalist psychology that relate to this matter. He might outline a theory of such devices and he would, if he takes his work seriously, propose as an empirical hypothesis that such devices actually constitute the system used by the organism. The theory he proposes would be analogous to a grammar, in one sense; the system of devices that he attributes to the organism, with the properties made explicit in his theory, would be analogous to a grammar in the other sense. The neurophysiologist might proceed to show that the organism does or does not actually use a system of the postulated sort, thus confirming or disconfirming the psychologist's theories. Similarly, further observation or introspection might provide useful empirical data. Obviously, all theories will be underdetermined by the evidence, a fact that seems to intrigue a number of philosophers who have written on this subject recently, for reasons that are unclear to me. Is the theory of performance, as it was suggested in Aspects, not a contradictio in terminis ? - I see no problem. If I am interested in understanding why, say, people have difficulty in understanding, repeating or remembering sentences with certain types of embedding, I can proceed to construct a theory of performance that will attribute to the organism linguistic competence - as expressed in a generative grammar certain performance strategies (e.g., the property of being unable to call upon a specific analytic routine, such as the procedure of relative clause analysis, in the course of applying this procedure), certain general properties of any finite temporal processor (e.g., those properties from which it follows that self-embedding is ultimately beyond the capacity of any such device, while arbitrary left-branching, right-branching, and interaction between the two that does not lead to self-embedding can in principle

44

NOAM CHOMSKY

be accommodated). I can then attempt to account for the observed facts of performance in terms of this theory. This is the approach suggested in Aspects.36 It seems to me to embody no self-contradiction. The comprehensive theory proposed would be, in a good sense, a theory of performance, incorporating a theory of competence as one essential part. Not only does this approach seem to me free of any internal contradiction ; what is more, I am aware of no coherent alternative. One can find, in recent literature, a number of references suggesting that the performance-competence distinction must be abandoned, that there is some logical problem of the sort that your question suggests, and so on. I have yet to see a coherent argument, or an alternative that does more than revise terminology. Some examples are discussed in Studies37 and in the extended edition of Language and Mind.z& I might say that similar remarks apply to the vague proposals often made that the exigencies of communication somehow determine the form of grammar. I have yet to see any coherent proposal that differs in principle or even in detail from what I have suggested myself in this regard, and therefore cannot understand these vague remarks as intended, namely, as a criticism of my approach. Nor am I aware of any serious effort to answer the questions that I and others have repeatedly posed: namely, how can you deduce the specific properties of grammars from conditions on communication.39 Until advocates of 'communication theories' take these problems, and their own approach, seriously, I see no way to respond further to what they may have in mind. But I do want to make clear that I would personally welcome, naturally, any result that showed how the properties of language might serve to expedite communication, free thought and expression, or other uses of language; and I have suggested a number of possibilities myself, as have others. Can I have your comment to the following quotation of Searle :40 "a person's knowledge of the meaning of sentences consists in large part in his knowledge of how to use sentences to make statements, ask questions..., in his ability to perform speech acts" ? - I have no objection to Searle's comment, though the conclusions he draws from it are another matter. To be concrete, take the sentence "drinks will be served at 5". The meaning of this sentence, with declarative intonation, is such that a speaker can use it to perform a variety of 'speech acts': it can be used as an assertion, a promise, a prediction, a warning, a threat, an offer, and no doubt much else. I don't see how ii could be used as an order or a question, though perhaps with a bit of imagination one could construct such potential uses as well. Correspondingly, "will you leave the room?" with rising intonation can be used as an order or a question, but not a prediction. The sentences "I'll wash the dishes" or "I promise to wash the dishes" can be used as a promise, as can "I promise to return your book" - different promises, 36 37 38 39 40

Aspects of the Theory of Syntax, pp. 10-15. Cf. note 25. Language and Mind, enlarged version (New York, 1972). E.g. Problems of Knowledge and Freedom (London, 1971). Cf. note 6.

NOAM CHOMSKY

45

by virtue of the intrinsic meaning. I see no reason why a general theory of semantic representations might not be extended to incorporate a theory of the speech acts that can be carried out with sentences having these semantic representations. This is a reasonable program, which might be explored in many ways. On the other hand, the claim that meaning can be reduced to the intentions (or semantic intentions) of speakers seems to me quite unsubstantiated, and there are fundamental difficulties of principle that cannot, so far as I can see, be overcome. If Searle is suggesting otherwise - as I take it he was, in the article from which you quote - I await the evidence. As to his view that speech act theory is in some sense an alternative to, or incompatible with, semantic theory in the sense, say, of Katz, whom he appears to have in mind though he does not mention him, I fail to see that he or anyone else has given any reason to believe that this is the case. However, I will not pursue this matter here, since it would lead me away from the question, as posed. Do connotations and speaker's intentions belong to ''grammatical'' meaning? Can grammar be extended to performative analysis and to the construction of conversational postulates ? - Again, let us be concrete. If I say "drinks will be served at 5", I presumably intend that you will think that drinks will be served at 5, not that there will be a snowstorm tomorrow. In this respect, speakers' intentions are plainly related to grammatical meaning. Perhaps, in making this statement, I also intend that the hearer, an incurable alcoholic, will become incapable of carrying out some alternative action at 5; suppose, in other words, that my statement was intended as a hostile act, in this sense. This intention does not constitute part of the meaning, though of course a theory of meaning, embedded in a broader theory of human action, would contribute to the conclusion that the sentence in question might be used to perform a hostile act of the kind mentioned. I see no problem, in principle, in incorporating grammar within a theory that incorporates intentions of the first sort, and incorporates a good deal of what is often called 'performative analysis'. I do not think that this is accomplished, in any serious way, by postulating underlying structures such as "I assert to you that (warn you that, promise you that, etc.) drinks will be served at 5" for the sentence "drinks will be served at 5". Furthermore, the syntactic arguments that have been adduced for such proposals seem to me extremely weak. Rather, semantic theory should entail that the semantic representation or logical form of this sentence makes it available for certain acts, not others. As to intentions of the second sort, I see no problem in principle in developing a general theory of human action, incorporating grammar, that will try to deal with the fact that "drinks will be served at 5" can be used to perform a certain type of hostile act, but I would be reluctant to extend the term 'grammar' to incorporate such a theory, for obvious reasons. As for conversational postulates - maxims, in Grice's sense41 - these seem to me to constitute 41

H. P. Grice, "The Logic of Conversation", University of California, Berkeley (unpublished paper), 1968.

46

NOAM CHOMSKY

a significant contribution to the theory of successful communication. I welcome it, for its intrinsic interest and because the approach provides a further justification for the idealization that underlies the development of theories of generative grammar, much as Bever's proposal concerning adjective order, 42 which I mentioned before, provides such a justification. Thus assuming Bever's theory, we need not be concerned about the fact that the devices of generative grammar provide no natural way to describe adjective order. This 'failure' is in fact an argument in favor of the postulated devices, given the more comprehensive theory incorporating generative grammar as it stands - with its inability to express this ordering in a natural way - and Bever's postulated perceptual strategy, formulated in quite different terms, though, of course, making use of grammatical devices. Similarly, consider the proposal that a conversational maxim leads us to give a temporal interpretation to the conjunction in such standard examples as "Mary got married and had a baby", as compared with "Mary had a baby and got married". Adopting a comprehensive theory of performance that incorporates the maxim as well as generative grammar, we see that it is no problem for generative grammar that it provides no natural way, without considerable complexity, to express the general temporal interpretation of conjunction. The point is made in Aspects with regard to 'iconic order'. 4 3 And I think it is correct. - Logical semantics is mainly extensionalistic. Do you also, as a grammarian, consider the referential function of language as the most important ? - I don't agree that logical semantics is mainly extensionalistic. Surely this is not true, say, of Katz' semantic theory. Of course, a general semantics and a general performance theory must take into account referential function, as they must also take into account intrinsic connections among predicates - e.g., the relations between "persuade" and "intend", between "uncle" and "male," etc. - the distinction between synonymy and logical equivalence, and much else. I cannot comment on what might be the 'most important' of these problems. The differences in the use of the term 'presupposition' in linguistics are rather confusing. Could you describe your understanding of the term presupposition ? - The term is used in a variety of more or less related ways. In one sense, one speaks of the presupposition of a sentence, within semantic theory, as what must be true for the sentence to have a truth value; various formulations can be given for this general notion. In another sense, one speaks of the presuppositions of the speaker, that is, what he intends the hearer to assume as true when he uses an utterance in one of the permissible ways. The latter sense actually includes a variety of notions. Some clarification and standardization of usage would be a good idea, as a number of people have noted. I've suggested, rather offhandedly, in Studies,44 that there may be some 42 43 44

Cf. note 30. Aspects of the Theory of Syntax, pp. 11 and 225. Studies on Semantics in Generative Grammar, Chapter II.

NOAM CHOMSKY

47

linguistic correlates to the several usages. There are also many open questions, to my mind, as to just how presupposition, in the various senses, relates to theories of performance and competence. You characterized generative semantics as a notational variant of the standard theory and your main criticism of it is that generative semantics is too powerful. What new elements could there be to justify a new definition of the relation between syntax and semantics ? - I didn't, strictly speaking, characterize generative semantics as a notational variant of the standard theory. Rather, I pointed out that all more or less clear versions of generative semantics were in large measure notational variants of the extended standard theory, which, incidentally, was being discussed in lectures in 1965, though there were some significant innovations, two in particular: generative semantics relaxed the requirement that lexical and nonlexical transformations are 'segregated' by a level of structure with the properties of deep structure; and generative semantics allows arbitrary rules, derivational constraints, relating the phrase-markers in a derivation - later work permits even relations between derivations - and relating phrase-markers in a derivation to other aspects of semantic representation, such as focus and presupposition. Furthermore, some of the specific uses of derivational constraints, e.g., with respect to quantifiers, were similar to, though not identical with, certain proposed rules for interpretation of surface structure. Thus as far as I can see, generative semantics simply relaxes some of the conditions on the extended standard theory. As a result, it is a more 'homogeneous' theory, to use Postal's terms, just as any theory that rejects specific detailed assumptions about the form and function and organization of rules, requiring only that there be rules of a more general form, will be a more 'homogeneous' theory. Suppose, say, that linguistic theory were to abandon the conditions that differentiate phonological and transformational rules, requiring merely that there be 'rules'. The theory would be more homogeneous in exactly the sense in which Postal's and Lakoff's proposals with regard to filters and derivational constraints 45 provide a more homogeneous theory. I think that the basic issues involved in this discussion have been seriously misunderstood. Perhaps I can clarify them by reference to the recent volume edited by Stanley Peters, 46 which contains three papers by Postal, Peters, and me - the latter appearing also, without a number of printer's errors, in my Studies on Semantics in Generative Grammar dealing with these issues. Postal argues that the more homogeneous theory that he proposes is a 'better theory', on the grounds that it contains fewer devices, is simpler, and so on. I argue that it is a 'worse theory' on the grounds that it is less falsifiable and permits a far wider range of grammars than the alternatives considered. Peters, on the same grounds, goes still further, suggesting an extreme version of a 'nonhomogeneous' theory. If we sort out the issues properly, it is quite easy to see why 45 46

Cf. the present volume, the dialogues with J. McCawley and G. Lakoff. P. S. Peters (ed.), Goals of Linguistic Theory (Englewood Cliffs, 1972).

48

NOAM CHOMSKY

Peters and I reach one conclusion and Postal the diametrically opposed conclusion with regard to the 'best theory'. A discussion of better and worse theories, on the methodological grounds that Postal suggests, can be undertaken only when the problems to be dealt with, the empirical issues at stake, are fixed. Postal's conclusion would be correct if we were concerned solely with descriptive adequacy. In this case, we would, naturally, prefer the simplest theory that meets the condition of descriptive adequacy for theories. On the other hand, if we are concerned also with explanatory adequacy - with what Peters calls 'the projection problem', the problem of selecting a grammar on the basis of presented data - the matter is considerably more complex. Given two theories T and T', we will be concerned not merely with their simplicity or homogeneity, but also with their restrictiveness. If T and T ' both meet the condition of descriptive adequacy, but T permits only a proper subset of the grammars permitted by T', then we may well prefer T to T ' even if it is more complex, less homogeneous. Postal regards it as obvious that we would prefer T ' to T in this case, but this conclusion is plainly false in general, if our concern extends to explanatory adequacy. Postal confuses the issue by the illustrative examples that he presents. Thus he argues, correctly, that the theory of phrase structure grammar (PSG) was methodologically preferable to transformational grammar (TG), so that P S G could only be rejected on empirical grounds, as it was. But the example is irrelevant to the discussion, since P S G is both simpler, more homogeneous, and more restrictive, narrower in the range of permitted grammars, than T G . Therefore the illustration sheds no light on the only issue, namely, the choice between theories T and T ' where T is simpler and less restrictive than T'. T o take an example of this sort, let T be a theory which merely states that there are rules of phonology and syntax, and let T ' be the familiar theory that distinguishes these rules in terms of their quite different properties and requires that they be segregated in the grammar in some way, rather than freely interspersed. Then T is simpler than T', but far less restrictive. It permits a vast range of possible grammars rejected by T', is less falsifiable, and virtually eliminates any possibility of explanation or of solving the projection problem. Assuming that T ' meets the condition of descriptive adequacy, we would surely prefer it, on methodological grounds alone, to the simpler, more homogeneous and less falsifiable theory T. It is this example, not the irrelevant example of P S G and T G , that is analogous to the case in question. Postal's homogeneous theory H is, as he says, simpler than the 'extended' standard theory ST. It is also less restrictive and less falsifiable, and thus less capable of dealing with the projection problem and the issues raised at the level of explanatory adequacy. If we are concerned solely with descriptive adequacy, we will choose H ; if our concerns extend to explanation and the projection problem, we will prefer ST, assuming that it meets descriptive adequacy. Peters's extreme example of a nonhomogeneous theory would be preferred, on methodological grounds, were it to meet the condition of descriptive adequacy, as perhaps it may. There are, as would be expected, specific respects in which the less homogeneous and more restrictive theories such as ST achieve explanatory adequacy of a sort that escapes

NOAM CHOMSKY

49

the less preferable theory H. A number of these are noted in my paper just cited. Thus postulation of deep structure permits a formulation of the lexicalist hypothesis with regard to nominals of various types, providing an explanation for a conjunction of properties that are a mere accident on the less falsifiable and less restrictive theory H. As I noted, these particular properties are surely describable in less restrictive theories - an observation that has been repeated, irrelevantly, by many people who have misunderstood the problem under discussion. But any other arrangement of these properties would be equally well described within these less restrictive theories, although the alternatives are inconsistent with the narrower theory. Of course it might be argued that we must reject the preferable theories and adopt the worst theory, Postal's H, on empirical grounds. Suppose, say, that we were to discover a language with rules that moved phonological segments in accordance with properties of phrase structure - e.g., a rule that moved the final consonant of the NP subject of a sentence to the beginning of the sentence, thus mapping "the boy is here" into "y the bo is here", etc. - and suppose that these rules were freely interspersed among others. Such observations might lead us reluctantly to abandon the distinction between syntactic and phonological rules, adopting the simpler, more homogeneous, less falsifiable theory that merely postulates the existence of rules (derivational constraints). This new theory would be 'best' in Postal's sense, and 'worst', though, regrettably, the best we can do, if we are concerned as well with explanatory adequacy and the projection problem. Postal claims that there are considerations of this sort that require abandonment of the more restrictive theories that postulate the existence of deep structure. I disagree, for reasons explored in some detail in the paper cited. The issues are clearly empirical; rational debate is possible and welcome. But the methodological issue is badly confused in Postal's account. What is the importance in this connection of the problem of lexical insertion ? - These remarks, perhaps, will clarify the issue of the importance of lexical insertion. The extended standard theory postulates that all lexical insertion operations apply at the level of deep structure, and that all transformations of other sorts are restricted to the mapping relating deep and surface structure. Generative semantics weakens this requirement, permitting the equivalent of lexical insertion - i.e., rules relating a lexical formative to some more elaborate structure that represents the meaning, or part of the meaning - to apply freely among other transformational operations. This is one of the several issues involved in the empirical question: does deep structure exist in the sense of the extended standard theory. As for the relation between syntax and semantics, my view has always been and remains agnostic. The notions seem to me too unclear to permit a satisfactory answer to the question how or whether syntactic and semantic rules are distinguished. As indicated in response to your question concerning Searle's statement about my conception of innateness, 47 my view, frequentCf. above p. 30.

50

NOAM CHOMSKY

ly expressed, is that any attempt to delimit the boundaries of these domains must certainly be quite tentative, and that perhaps we shall discover that the attempt is misguided. The question can only be resolved by further investigation of these systems and their interrelations. I might add that commentators who attribute to me the view that syntax and semantics must in principle be sharply distinguished, or those who go beyond and claim that I have urged that problems of meaning be set aside, are simply unacquainted with what I have written on the subject or are, more likely, misinterpreting the technical discussions mentioned earlier. What is it that makes generative semantics too powerful ? Is it the use of transderivational constraints and global rules ? - To repeat, generative semantics is too powerful, in my opinion, in every respect in which it departs from the extended standard theory; namely, in relaxing the conditions on the existence of deep structure, in the assumption that there are arbitrary rules (derivational constraints), and in the further assumption that there exist transderivational constraints. Of course, my assertion must be backed by empirical arguments and studies demonstrating this. I will not go into the matter here, except to express my belief that the existing literature presents adequate arguments, though many uncertainties remain. Natural languages have to be analyzed in terms of 'natural logic'. The logical form of natural languages is very specific - concepts are fuzzy concepts and truth is always a matter of degree. Is there any hope for such a natural logic ? - That concepts are 'vague' is a commonplace; e.g., there is no sharp criterion that distinguishes bald and nonbald people. I'm afraid I do not see the importance of this observation for semantic theory. Semantic theory, insofar as it is concerned with the analysis of concepts, will specify the relations that exist among these concepts. If they are partially indeterminate, semantic theory will try to express this fact as precisely as possible - vagueness of concepts does not imply imprecision in the analysis of these concepts. It might be argued, as Hilary Putnam has suggested, that concepts relating to 'natural kinds' (e.g., "tiger", "lemon", etc.) should not be analyzed in terms of a system of semantic features, but rather in terms of 'stereotypes' and ranges of variation. Correct or not, this is certainly an intelligible proposal, and it can be worked out in detail. It might turn out to be the case, for example, that the relation between "persuade" and "intend" or "believe", or the relation between "uncle" and "male", can be expressed properly within a system of semantic features and rules; whereas the concepts "lemon" and "tiger" should be characterized in a different way, in terms of stereotype and variation. Similar questions can be raised at many points in the investigation of semantics. The issues require both conceptual clarification and empirical study. As far as I can tell, there are no issues of principle here that distinguish, say, between the standard theory, the extended standard theory, generative semantics, and other more or less clear variants of generative grammar that have been proposed.

NOAM CHOMSKY

51

Can language form be analyzed independently from language function? Do you maintain that language does not have any essential purpose or that there is no connection between its purpose and its structure? - Language can be used to communicate, to express one's thoughts, to clarify one's thoughts, to mislead, to maintain friendly personal relations with little or no concern for communication, for stating honestly what one believes irrespective of what the audience (if any) may believe with regard to the beliefs of the speaker, and so on. I know of no coherent proposal to the effect that one, or some, of these uses is a 'central purpose' of language. It is striking that communication theorists generally hedge, including, say, 'self-communication' or 'communication in the absence of an audience' (or, if pressed, without concern for the audience) as a form of communication. They would also, no doubt, include Malinowski's 'phatic communication' as a form of communication, along with other varieties of language use that can be identified. When pressed to analyze communication in the absence of an audience, they invariably, to my knowledge, introduce some notion equivalent to sentence meaning. In short, I doubt that they hold any position at all, when their views are made precise. Consequently, I do not reject their position. Consider the further question: can we explain the particular characteristics of language in terms of the exigencies of communication? It seems to me that the answer is generally negative, though having isolated certain formal properties of language, we can doubtless inquire into their communicative function - an entirely different matter, obviously. I have discussed this matter 48 and I see no possible answer to the questions that immediately arise. Such properties as structure-dependence of transformational rules, the transformational or phonological cycles, the conditions on the structure and organization and function of rules of various types seem to me plainly not deducible from the communicative function of language. I see no respect in which language would be less efficient as a 'communication system', for an arbitrary organism, if, say, it were possible to form questions in English by the structure-independent rule of inverting the leftmost auxiliary with the string that precedes it. Transformational-generative grammar, in any known variety, rejects such a rule as unformulable, but this property of language (if such it is, as I believe) is not deducible from considerations of communicative efficiency. Language permits many forms of ambiguity ; one can easily imagine alternative systems that rule these out. Again, I see no hope of explaining these facts on grounds of communicative efficiency. The same holds true, so far as I know, wherever we look. For this reason, I think we must conclude that the many claims that somehow communicative function must, or even can, be used to determine properties of language are simply without substance. Let me distinguish these empty claims from others that are, in fact, substantive. It might be argued, say, that the organization of memory in the particular organism, homo sapiens, is such that transformational grammar is a system that meets conditions 48

E.g. Problems of Knowledge and Freedom, pp. 27-32.

52

NOAM CHOMSKY

of communicative efficiency. George Miller and I presented a speculative argument of this sort about 10 years ago. 49 Or, one might suggest, as I have in a forthcoming paper called "Conditions on Transformations", 50 that surface filters relate to communicative efficiency and are partially explicable, perhaps, in these terms. T. G.Bever has made a number of interesting suggestions in this connection. 51 Jerry Fodor has suggested that a careful study of perceptual strategies may replace part, or all, of generative grammar. All of these and related suggestions have some plausibility, and are surely worth investigating. In contrast, I can see no substance in the proposals of Searle, in the article from which you quoted, 52 and others with regard to the alleged interpénétration of the study of form and the study of function. We can only await some substantive proposal. If one is forthcoming, it must surely be evaluated. Thus I would suggest that your question be reformulated. As you put it, it seems to presuppose that language function will contribute in some way to the study of form, and you ask whether it is possible to study form independently of this connection. Analogously, it was common to ask, years ago, whether phonology can be studied independently of considerations of meaning (synonymy). The question was badly put, because it had not been shown - nor, to my knowledge, can it be shown - that considerations of synonymy can be used in any coherent way in the study of phonology. Similarly, until it is shown that the study of language function can somehow contribute to the investigation of language form, one cannot raise the question whether form can be analyzed independently from function. You always stress the expressive function of language - language is the expression of thought. Is the expressive function of language unrelated and opposed to its communicative function ? - If we understand 'communication' in some reasonably narrow sense, I would then merely say the obvious: namely, communication is one of the functions of language. In communicating, we express our thoughts in the hope that the listener understands what we are saying. We may be hoping to persuade him, to inform him that we believe such-and-such, and so on. The function of language for expression of thought is not 'opposed' to its communicative function; rather, it is presupposed by the use of language for the special purposes of communication. I think that these are truisms. What is not a truism is the claim, frequently made, that communication, in some sense sufficiently restricted so that the claim is not tautologous, is a central function of language, and that the concepts of semantic theory can be reduced to considerations of speaker-intention. Not only is this not a truism; it is also, so far as I am aware, quite false. Searle and others have suggested that it is a commonsense view that the 49

G. A. Miller and N. Chomsky, "Finitary Models of Language Users", in R. D. Luce, R. R. Bush, and E. Galanter (eds.), Handbook of Mathematical Psychology (New York, 1963). 80 N. Chomsky, "Conditions on Transformations", M.I.T., Cambridge, Mass. (unpublished paper), 1972. 51 Cf. note 30. 52 Cf. note 6.

NOAM CHOMSKY

53

purpose of language is communication, and that it is paradoxical and perverse for me to deny this commonsense view. As far as I can make out, the proposal that Searle advocates is empty, as he puts it, and becomes false, or at best unsubstantiated, when it is given any clear form. I see nothing in 'common sense' to suggest that a person who spends years in his study writing a book, not caring particularly whether anyone will read or understand it, is not using language in some normal way. I take it that an advocate of 'communication theory' would disagree, but if so, I see no reason to accept his conclusion. Is the relation between language and culture of the same type as the relation between language and thought? Don't we have to accept a social or sociological metatheory in order to do good linguistics ? Can I mention here that from two different angles (Habermas and Hymes) the notion of'communicative competence' has been suggested? - The relation between language and culture is not the same as the relation between language and thought, so far as I can see. Language is a means for the expression of thought, though it can be used for other purposes, and I have no doubt, on the basis of introspection, that I can think without language. Whatever the relation of language and culture may be, it is quite different. I think that very good linguistics is done, and always has been, without any social or sociological metatheory. For this reason, I do not agree that we have to accept a social or sociological metatheory to do good linguistics, fortunately. However, 1 can imagine no objection to the project of studying language in its social and cultural context, or to the project of developing a theory of 'communicative competence', if it can be accomplished. I don't frankly feel that I have anything to contribute to this project, but it does not follow that I would not view it with great interest. The distinction made by Morris between syntax, semantics and pragmatics is well known. How is the domain of linguistics to be defined with regard to this distinction ? How can a notion of linguistic contextuality be correctly defined? - I don't feel that I have anything significant to say in this regard, beyond the tentative remarks in Aspects already cited, and others that are equally uninformative. To conclude this dialogue, a question about the relationship between philosophy and linguistics. Do you agree with Hymes' statement53 that in your earlier period you applied philosophy to linguistics and in the later period linguistics to philosophy as the study of the mind? - I suspect that I am the worst person to answer this question. I understand the point of what Hymes said, and I think it is a tenable interpretation of my published work. However, as I see it myself, from a different view, not necessarily a correct one - there is no guarantee that a person provides the most accurate testimony as to his intellectual development - 1 have always been as much interested in the philosophy of language 53 Dell Hymes, review of J. Lyons, Noam Chomsky 416-427.

(New York, 1970), in Language 48 (1972),

54

NOAM CHOMSKY

and mind as in linguistics. My ideas about all of these subjects have been continually revised, on the basis of my own work and the work of others, and I assume that this will continue to be the case. As a graduate student, I was interested in both philosophy and linguistics. My general feeling in the 1950s was that what I was doing probably wasn't linguistics at all, a feeling shared by editors of linguistics journals, readers for publishers, chairmen of linguistics departments, and most other professional linguists. I always regarded behaviorism as an intellectual scandal, and felt that empiricist theories, interpreted as psychological theories, were rather dubious. There was a tension in my own thinking for many years in this regard. I was firmly convinced that the 'constructionalist' approaches of Harris, Goodman, and others who influenced me very profoundly were fundamentally correct, and I did a good deal of work trying to demonstrate this. At the same time, my own linguistic work, since I was an undergraduate, proceeded along entirely different lines, as I became aware with increasing clarity, particularly through the insistent prodding of Morris Halle. What I am trying to suggest is that my ideas and beliefs about the structure of language and my attitudes towards the philosophy that I was studying and thinking about developed more or less in parallel, the problems, relationships, inconsistencies, and so on, becoming gradually clearer to me as time went on. Looking at what has actually been published, one might have little basis for assuming this, I suppose. Could the grammarian Chomsky say, rather metaphysically perhaps, that generative transformational grammar proved and proves the uniqueness of man ? - I wish that I could give a positive answer to this question, but I cannot. It is a plausible thesis, enshrined in tradition, that humans have unique qualities of mind and that the properties of language somehow reflect these. The work in generative transformational grammar does, I believe, permit a somewhat sharper formation of this traditional view. It does not, however, demonstrate its truth. My own guess - or perhaps 'hope' would be better - is that this traditional view will prove correct, under some modern formulation that departs from certain metaphysical and empirical assumptions that led the tradition astray, in my opinion. That is, further investigation of the constructive achievements of the normal human intellect, and of other organisms, may reveal significant qualitative differences - a different type of intelligence rather than simply a higher degree of intelligence - and may succeed in making this vague notion precise. It may also prove true, as has often been speculated, that a general theory of the human mind will postulate a language faculty with special properties that are in part unique to it, interacting with other faculties. If pressed, I would be inclined to speculate that something of the sort is true. To speak of 'proving', however, is to misconceive the limits of present understanding. Massachusetts Institute of Technology Cambridge, Mass., U. S. A. November 3, 1972

ALGIRDAS J. GREIMAS

Algirdas Julien Greimas, I would like to ask you if you consider yourself to be a structural linguist ? - I answer yes, without any reservation, if, as we do in France, you consider structuralism as a scientific approach to language comparable to that of the exact sciences, which fundamentally consists in a search for relations and tries to establish invariants and variants. I am a structuralist in that sense, although saying that one is a structural linguist is sort of a pleonasm - being a linguist already presupposes that one is a structuralist. So, you distinguish between French and American structuralism ? - When Americans talk about structuralism, they usually refer to the Bloomfieldian school, I think. There is an excessive formalism in American structuralism; it is faulty in other respects too, but the successors of that school stressed this enough so that we do not need to come back to that. On the other hand, contemporary American linguistics might be reproached because it is not structural enough, i.e., that it does not draw all the consequences from the concept of structure. If structure is defined as a relation between terms, either the 'relation' or the 'terms' that are present in the relation can be stressed. If the attention is focused on the terms at the expense of the relations, there it is a constant danger of falling into a 19th century type of atomism. It has probably to be added that American structuralism is asemantic in principle. And there, you do not agree either... - No. However, I consider that the worst is over. Semantics gets its place again, its primary importance is recognized. Not only does this already show positive results but a far-reaching revolution in the conception of linguistics will necessarily follow, a revolution that will unite us, despite some terminological misunderstandings about the use of the word. Even within European linguistics, your position is quite peculiar, quite original, since you believe in a linguistics of discourse. What are, according to you, the possibilities of a linguistics of discourse ? - For me, the expansion of linguistics so as to encompass the discursive organization

56

A. J. GREIMAS

is a matter of course, coming from the very fact that I adopted the semantic point of view right from the beginning. The semantic units that are manifested at the surface level in language, grosso modo correspond, if the structural point of view is maintained, to the utterances as syntactic units. Two terms at least are needed in order to speak of a manifested semantic structure, no matter whether it be a verbal or a nominal 'phrase' (syntagm). If one wants to go further than that and take into account the combinatorics and the production of vaster semantic units, one immediately goes beyond the boundaries of the sentence and arrives in discourse. Those were, in my opinion, the logical consequences that followed from the semantic conception of language I had postulated. Obviously, some other factors played some role as well and, more particularly, the problem of the narrative organization of discourse, the importance of which was shown in the works of Propp and Lévi-Strauss. 1 The convergence of those two problems required a unique solution, of course. I would say even more - the necessity of a discursive linguistics seems to me to follow naturally from the generative attitude. What is, in fact, the generative approach ? It consists in going from the simple to the complex, from fundamental to derived or expanded structures; as a consequence, the generative method, which I often adopt myself, is a deductive procedure quite commonly used, if under various forms, in scientific research. From this point of view, it is quite understandable that the task of linguistics is not only to account for the production of sentences; it has to show how, from a magma of unarticulated or hardly articulated meaning, the syntagmatic chains we know as discourses are produced and articulated in utterances : utterances are no primary data, but, in a sense, the 'terminal' data of the generated discourses. Discourse is the primary datum; through an effort of articulation it explodes into utterances, it is a totality that decomposes in utterances and does not result from their concatenation. Take a trivial example - you have just seen a movie and you are asked to describe it. In answer to such a question some very simple ideas, which serve as a scheme, will come into your mind, organizing your memorized knowledge; only then will you articulate your memories in sequences and utterances. That is exactly the way I see the generation of discourse. Can it be said that a linguistics of discourse is at the same time generative and structural and that the opposition of structuralism to generativism is illusory when one approaches discursive linguistics ? - I am with you on the whole line, if, of course, we first agree on the good usage of the term 'structural'. The essential difference between what I am trying to do and what is being done in the United States lies in a different conception about the form a grammatical theory has to take. 'Being a structuralist' should mean not to consider 1 V. Propp, "Morphology of the Folktale, Part III", in International Journal of American Linguistics 24 (1958): 4 (Publication Ten of the Indiana University Research Center in Anthropology, Folklore and Linguistics, 1958). Cf. C. Lévi-Strauss, "La structure et la forme, réflexions sur un ouvrage de Vladimir Propp", Cahiers de l'Institut de science économique appliquée 99 (1960), 3-36.

A. J. GRÉ1MAS

57

as objects of thought the phenomena, but the relations between phenomena. If, in order to construct a grammar, one takes a stock of any elements and calls that stock a structure, if one then adds to that another pack of generation rules and considers the whole as a grammar - that, for me, is exactly a non-structuralist approach. 'Structure' means first a coherent set of relations, the elements being the termini of those relations - they have no other existence than to be terms of relations; every other attitude is atomistic. Only such a structure, whose elements are all inter-defined, can serve as a starting point for the generation of syntagmatic units. Both attitudes probably resemble each other very much from a generativist point of view, but, at the same time, they are quite different from the structural point of view. Given the Saussurean dichotomy between 7angue' and 'parole', what is the statute of discourse ?'2 In other words, does discourse have to be identified with parole, as Saussure himself sometimes does, or is it a phenomenon of langue ? - I recognize the great heuristic value of the Saussurean dichotomy you have just mentioned, it was fertile soil upon which two generations of linguists reflected. But 1 think the dichotomy has now given all it could give for two reasons. Through greater formalization, it has been decomposed into a certain number of epistemological categories. So you know that Hjelmslev interpreted the opposition between langue and parole as a general opposition, 3 common to all scientific approaches, between system and process, or, in linguistics, between paradigmatic and syntagmatic. This is already a positive result. The opposition concerns both modes of organization of language considered as a semiotic system. But parallel to this, there exists another problem - the mode of existence of semiotic systems; there the Saussurean opposition between langue and parole was expressed, in Europe, by the opposition of the virtual to the actual, which corresponds grosso modo to Chomskyan competence and performance. I say grosso modo because this parallelism, however real, covers more than that - the European attitude reflects a social, collective conception of language, as opposed to the Americans, who have a psychological individualistic conception. This shows that a third opposition has to be distinguished in the Saussurean dichotomy: langue as a system of constraints and parole as idiolectal manifestation, the level where change can take place under pressure from the linguistic praxis. You then ask the question, what is the place of discourse within this set of oppositions? Given Hjelmslev's conception of a syntagm, I would answer in Hjelmslevian terms that discourse is a sign with a peculiar statute. Hjelmslev says that language at the moment of its manifestation is composed of the intermingled levels of expression and content, and the fusion, the correlation, between both levels constitutes a system of signs; but signs must be understood, not only as morphemes, but as every unit that has a level of expression and a level of content: syntagm, utterance, phrase, etc.... 2

Cf. H. Parret, Language and Discourse (The Hague, 1971), especially Chapter IV. Especially L. Hjelmslev, "Langue et Parole", Essais linguistiques (= Travaux du Cercle linguistique de Copenhague 12 [1959]). 3

58

A. J. GREIMAS

Discourse then is a syntagmatic unit with variable dimensions and an organization peculiar to it. The problem, of course, is to know how those macro-signs have to be analyzed. Don't you think the notion of discourse already underlies some of Saussure's intuitions : where he speaks of faculty of language' (faculté de langage) and of linguistic mechanism ? Saussure too sensed the originality of the syntagmatic... - Saussure still greatly inspires philosophical thought, linguistic thought somewhat less, since his successor, Louis Hjelmslev, took over and formalized Saussurism. From a point of view of historical curiosity one can go back from Hjelmslev to Saussure or from Saussure to Hjelmslev and discover more or less the same things, but the concept of 'production', for example, does seem to me to be a Hjelmslevian concept. Generation and production are discussed nowadays with a more materialistic or a more idealistic connotation; the concept of language productivity, however, is already explicit with Hjelmslev. If on the other hand you talk about 'faculty of language', it concerns the concept of competence as it was inherited by the classical psychology of the 17th and 18th centuries. Do you consider yourself as a Hjelmslevian and do you think the Hjelmslevian interpretation of Saussure is the most respectful of the latter's thought ? - Yes to both questions. I think indeed that the interpretation given by Hjelmslev is a step forward when compared to Saussure, just because he formalizes Saussure. I personally have known Hjelmslev only rather late, but I was literally seduced by him. When you ask me the question if and why I am a Hjelmslevian, I would answer frankly: by my sense for estheticism. I consider the Prolegomena4 to be the most beautiful linguistic text I have ever read. I use 'esthetic' in the same sense as a mathematician when he finds a formula beautiful or elegant because of its rigor, its simplicity. Beauty and efficience coincide here. 1 have always wanted to write like Hjelmslev, without ever succeeding, of course. To be a Hjelmslevian or a Chomskyan is the unhappy question we live with, but it should not have to be asked in linguistics : such a division into sects only shows the immaturity of a scientific discipline. To be a Hjelmslevian first means to interpret Saussure in a certain way. To give an example: the thunderbolt for me was when I understood in reading Hjelmslev that language is not a system of signs, or, at least, that it is not only that. It meant a starting-point of very heterodox thinking for me. Another example : among the important things I found with Hjelmslev it is his conception ofform which first shocked me, but then I was seduced : the traditional opposition betweenform and content, which is still often accepted as evident, is definitely destroyed, replaced by a new conception of the signifying form : syntax as the form of the content is a lasting positive point in contemporary linguistics. 4 L. Hjelmslev, Prolegomena to a Theory of Language (translated by J. Whitfield) (Madison, Wisconsin, 1963).

A. J. GREIMAS

59

Up till now, structural linguistics has mainly been concerned with units smaller than the sentence and generative linguistics is a linguistics of the sentence. Is it possible to catch a glimpse of suprasentential linguistics in Hjelmslev's works ? What is, according to you, the relation between the sentential and suprasentential units of language ? - I do not think one could find a theory of discourse in Hjelmslev. In any case, it depends on what is meant by 'to find in such and such an author' ; we will usually recognize nowadays the possibility of various readings of a text. Texts are interpreted in different ways at the same time, so the main thing is that each of us will read texts in his own way. As for me, the concept of 'discourse' was quite normally integrated in my reading of Hjelmslev, even if it cannot be said to have existed as such in Hjelmslev. With regard to the relation between discursive and sentential units, one can approach the problem in various ways - as a semanticist or as a syntactician for example. If you take a part of discourse, a surface utterance like "Adam eats an apple", which has a very concrete conten+, you will observe that this utterance is semantically decomposable - i.e. it can be paraphrased - in a program of underlying processes: the program 'to eat an apple' contains 'to open one's mouth, to shut one's mouth, to chew, to swallow, to make gestures in a certain way', which in fact is a long discourse. So one and the same semantic unit can be manifested either by a single utterance or as a program of utterances. One can thus speak of an equivalence between what you call a sentential unit and a discursive one. Semantics did not recognize the elasticity of language yet, although just to recognize it might be easier than to account for it. From a syntactical point of view I would say that the fundamental problem is to know how we manage to sustain the communicative process, how it is possible that meaning passes from the instance of emission to that of reception. Only if syntax is interpreted as recurrent, as a support for a content to be transmitted - with the help of a bundle of redundant grammatical categories - can the fact of communication be accounted for, in part at least. This syntactic recurrence, in my eyes, is founding the discourse. Now, every discourse contains verbs, substantives, etc., always in a redundant way : that makes it possible for the rest of the meaning content to pass. As far as I am concerned, this grammatical isotopy accounts for the why of syntax as well as for the mode of existence of discourses. Do you agree with the contention that the discourse structure is at the same time semantic and syntactic ? - I would say that is exactly the same thing, since the categories upon which syntax is built are, by definition, semantic. Does that mean that every semantic and discursive fact has to be syntactically verifiable ? - My answer will be double. First there exists an interpénétration of semantic and syntactic facts. I saw how generative grammar was imported to France. People who

60

A. J. GREIMAS

did not want to use in their syntactic analysis categories that would in the slightest way be semantic admitted in the end that there were such categories as 'animate'/ 'inanimate'. It appears that they are syntactically necessary to account for certain types of utterances. Other semantic categories can be introduced in that way and legally recognized: in enlarging the corpus of examined natural languages the number of those semantic or syntactic categories will constantly increase. Theoretically, every semantic category can be syntacticized. And the opposite is also true: every category that is used in the construction of any grammar is a semantic category. I do not see why syntactic categories should have a privileged status, rather, the opposite is true. It is by taking in a certain number of semantic categories and by putting them at the disposal of syntax on behalf of communication that it is possible to understand and to define the grammatical component of language. On the other hand the relations between semantics and syntax depend on the way you conceive the general economy of linguistic or semiotic theory. According as the generation of the particular objects, utterances or discourses, by the linguistic "machine" is conceived in the form of instances or levels (landings of derivation and transformation), the problem of the equivalence of the two levels and of the validation of the deeper level by the surface level will arise. Hence, if one agrees that the deep level is to be a semantic one and the surface level is to be organized as a syntactic form, it will be clear that the concepts of grammaticality or syntactic compatibilities interfere as principles of validation of a more profound semantic representation. You have just used a term which is peculiar to you, 'isotopy of discourse'. What is the operational importance of this concept ? - I already said that a grammar is made up of the combination of a small number, maybe fifteen, of binary grammatical categories, which are recurrent all along the discourse - whether it has to be formulated in rules or defined in structural relations is not important. Grammatical isotopy is thus that collection of categories underlying the whole of discourse and of which the syntagmatic dimensions are not sentential but discursive. Where the semantics of discourse is concerned the problem is somewhat different. The isotopy, on the one hand, can be considered in an interpretative approach as in Katz and Fodor's work, 5 as the principle that allows the semantic concatenation of utterances. Let us take, for example, a series of three utterances: the first utterance allows two distinct readings, the second three, the third four. The reading, common to all three utterances, will then be called 'isotopic reading': only because of it can discourse be considered a unit of meaning. But the real generative approach asks for an inversion of the terms of the problem: the semantic isotopy considered as a thematic coherence of discourse has to be postulated in the first place: the stock of semantic categories, those which are recurrent and underlying the whole 5

J. J. Katz and J. A. Fodor, "The Structure of a Semantic Theory", in J. A. Fodor and J. J. Katz (eds.), The Structure of Language. Readings in the Philosophy of Language (Englewood Cliffs, 1964), 476-518.

A. J. GREIMAS

61

of discourse, has to be taken as the structure ab quo from which discourse with all its variable utterances can be generated. Do you think a typology of discursive isotopies would be possible ? - Maybe not in the absolute sense, because this question can only be asked in the framework of cultural relativism. In effect, natural languages cover original cultures, and cultures constitute semantic universes which are differently articulated; so it is only within a given culture that it is possible to establish a certain number of fundamental cultural isotopies. Of course, we can then examine the problem of a typology of cultures. Lévi-Strauss's work consisted precisely of bringing to light the existence of such isotopies:6 wearing clothes, eating, ... are isotopie facts. The problem can be transposed from the collective level to the individual level and from discourse as a class to discourse as an occurrence: a poem by Mallarmé for example can be read according to three different isotopies. Take his poem "Salut", studied by François Rastier: 7 you first read it on a 'denotative' level, as a ceremony, as the story of an anniversary banquet, but also according to a second isotopy 'navigation' where the "white tablecloth" is read as being "the sea", and finally according to a third isotopy as the whole problem of writing, "tablecloth" being then transformed into "a sheet of white paper". Are there discursive isotopies in infinite number ? - I do not think so. In any case the problem is double. Where the pluri-isotopy of a given text is concerned, I would rather think that the number of possible readings of that text is quite limited and this is due to the psychological limitations of the human mind which cannot generate or understand a large number of terms at the same time - six terms seem to be a reasonable list. If, on the contrary, it is a typology of discourses that is concerned, discourses whose sum constitutes a semantic universe, collective or individual, then the problem is no different from the one of the number of sentences generated by a grammar: the number of types of discourse are determinable, the number of discourse occurrences is infinite. How does one characterize the continuity between linguistics and poetics as the study of poetic discourse ? - Poetics, if it answered the conditions of being scientific, might be considered in the wake of linguistics, since poetics has as its task to formulate the rules of organization and articulation of discourses of a certain type. Of course, poetics is still a vestige of another era and we cannot think of using it as it is now with the goal of scientific research. On the contrary, we have to take the concepts and the rules of poetics one 6

Especially the volumes of Mythologiques (Paris, 1964-1972). Cf. also La pensée sauvage (Paris, 1962). 7 F. Rastier, "Systématique des isotopies", in A. J. Greimas (éd.), Essais de sémiotique poétique (Paris, 1972), 80-106.

62

A. J. GREIMAS

by one and give them linguistic definitions. Otherwise, talking the whole time about metaphors, when one does not know what a metaphor is, can, at best, be an approximation and does not help us very much. However, apart from the poetics which only studies the surface forms of discourse, there exists a deep level of discourse organization and that is narrativity. It is not well-known, not much analyzed, it is a universal discursive form though: the analysis of oral literature shows us in effect that there are no societies without narrative discourses, that the typology of those discourses is possible. On the other hand, what is called mythology in archaic societies corresponds in more developed societies to written literature and history, the important discursive form upon whose pattern man and the world are constructed. Folklore, mythology, literature, history, those are discourses that we can call 'natural' in the same sense as one speaks of natural languages. Their 'literary' forms are linguistic forms that should actually be the object of a linguistics of discourse. Poetics is a discursive linguistics which has as its object, the poetic fact... - I do not agree with you on this definition. Poetics is a discipline concerned with content. It has to analyze the articulations of the form of the content of discourse, in the wake of the syntax of the sentence, as we said above. Poetic discourse has its own specificity because it not only questions the content level but also the level of expression. In this sense poetics is a partial analysis of the poetic fact considered under its sole semantic aspect, whereas what is poetic also asks for the analysis of the signifier. Poetic discourses constitute a subclass of literary discourses. So we have a double problem: how to analyze the two levels of language separately and how then to see the correlation between the two levels - that correlation of a particular type that constitutes the originality of the poetic phenomenon. From that point of view it can be said that research on narrativity has somewhat contributed to the study of poetic fact, of poetry in general. You know that the decisive step here was Jakobson and Levi-Strauss' publication some ten years ago of their analysis of "Les Chats" by Baudelaire. 8 Jakobson's idea was that poetry is sort of a paradigmatization of the syntagmatic, a projection of the paradigmatic relations on the syntagmatic production of discourse. That is true and decisive. Some progress has been made since in recognizing that poetry is not a mere iteration, which it would be if the Jakobsonian definition was accepted in its narrow sense: poetry is not merely a walking in place, however short a poem is it consists at least of a transformation, a paradigmatic inversion, the negation of some content that had first been posed, and the assertion of its opposite.... That is really the definition of narrativity: discourse, whether it be poetic or not, consists of a certain number of manipulations of contents which in fact are the beginning of a very deep discursive syntax. The question has then been raised whether on the level of expression the same paradigmatization could not be followed by comparable transformations. The analysis can be done here as an 8

C. Lévi-Strauss and R. Jakobson, "Les Chats de Baudelaire", L'Homme 2 (1962): 1, 5-21.

A. J. GREIMAS

63

analysis of distinctive features, and one can conceive of a poem which begins in a shadowy atmosphere because of the use of the sounds o, a, u, and then abruptly expresses the joy of life by the use of the sounds i and e. So it is the correlation of the levels of expression and content analyzed in phonological and semantic distinctive features, as well as their concomitant transformations, that account for the poetic phenomenon. The explanation of the 'fusion', the correlation between the two levels, lies in the recognition of one same type of articulation at both levels. This goes back to what Saussure already called motivation:9 if natural language is based on the arbitrariness of signs, poetic discourse moves more in the direction of their motivation and tries to have the level of expression, which always keeps its material, acoustic or graphic aspect, adhere to the level of content, giving the poem a sense of truth, of reality, of authenticity. This search for the truth which is not so different from the search for the beautiful is what characterizes poetry, 'sacred' or 'profane'. This way we come nearer to the structural definition of the rather mysterious phenomenon, the 'fusion of sound and meaning'. In what respect can the levels of expression and content of language be said to be isomorphic ? - I will remind you of what I have already said about my reading of Hjelmslev, which taught me that the analysis of language considered as a system of signs is neither the only nor the best analysis possible. It is true that it took a long time for linguistics to learn to segment the verbal continuum in signs-morphemes; but once the segmentation has been completed, what remains to be done is to leave the level of signs for a separate analysis of the levels of expression and of content. One then has two formal levels of language. It is not astonishing that the analysis procedures that can be applied to them are comparable. I would not dare say that it is in the nature of those two levels to be isomorphic or whether it is our analysis procedures that make them isomorphic. That is a problem for the philosophy of science. It may be sufficient to say that the results of the analysis yield the image of a certain isomorphism. So if, on the level of expression, you take a manifested unit such as the phoneme, you can pursue your analysis either on this side of the manifested unit or beyond it. The phoneme is then decomposed into distinctive features which can be called phemes, but it can also in fact enter into a syntagmatic combinatorics which produces syllables. If you observe the level of content, you will see that the sememe is a unit of meaning that the 'word' manifests when the utterance is interpreted. On this side of the sememe you will find semantic distinctive features, semes, whereas the simplest combination of sememes constitutes the semantic utterance which, from a semantic point of view, constitutes with its predicates and the actors (subject, object, etc.) around it, a unit comparable to the syllable made up by a vowel surrounded by consonants. The isomorphism of the two levels of language is evident. 9

F. de Saussure, Cours de linguistique genérale, published by Ch. Bally and A. Sechehaye (Paris 1968), 180-184. Cf. also H. Parret, Language and Discourse, pp. 75-83.

64

A. J. GREIMAS

One then understands Hjelmslev, who asserts signifier and signified are nothing but operational and interchangeable names; this assertion allows him to account for the constitution of meta-languages in indefinite number. Why then question isomorphism? When the levels of expression and content have been joined, discourse is constituted as a succession of manifested signs, not conjoining one by one the isomorphic units of both levels - it rather produces what Martinet calls the 'double articulation'. 10 Indeed we need several phonemes to cover a sememe, and the semantic unit, isomorphic to the syllable, is manifested under the form of a surface utterance. On the level of the manifestation of signs the isomorphism is lost, and that makes it so difficult. That is one of the fundamental characteristics of human language. There are immediate consequences for poetic discourse because if poetic discourse tries to reproduce the same articulation and the same syntagmatic structures on the level of expression and on the level of content, it then arrives at two isomorphic constructions which are parallel but not concomitant. To the opposition of two semantic units will correspond a comparable opposition of two syllables, but it is evident that the space and the dimensions of their manifestations are not the same. If we examine what has been done in discursive linguistics we see that narrative structures have been privileged. It is in this domain that the results are the most valuable. Why is that so ? - Well, as always in cases like that, there was a meeting of favorable circumstances. Through that high-class international door-to-door salesman of linguistics, Roman Jakobson, we were lucky to discover the existence of a translation of Vladimir Propp, published by Indiana University Press, but ignored in the U. S.11 At a given moment we had four or five persons in Paris who knew of the existence of a certain Propp. Linguistics was in fashion and generally considered to be a pilot science for all other human sciences; this was the background of that appearance of Propp. Also, the narrative analysis of the Russian popular tales was no different from Dumézil's work on the comparative mythology of Indo-European populations,12 nor from LéviStrauss's presentation of the first model of the structural interpretation of myth, in this case the Oedipus myth 13 - all those works had some linguistic air. So it is quite clear what I mean by an exceptional meeting of circumstances. On the other hand, the analysis of oral literature opens considerable possibilities because narrative discourses produced orally are relatively short and relatively easy to analyze; they also offer considerable guarantees of, I would not say, universality, but at least of great generality, since it is possible to pass from European folklore to African or Indian folklore and find discursive structures that are comparable, i.e., 10

Cf. pp. 236-237 of this book. "Morphology of the Folktale, Part III", see note 1. 12 G. Dumézil, VHéritage indo-européen á Rome (Paris, 1949); La Saga de Hadingus: du mythe au román (Paris, 1953). 13 C. Lévi-Strauss, "La structure des mythes", in Anthropologie structurale (Paris, 1958), 227-256. 11

A. J. GREIMAS

65

similar and different, which in a first stage need a distributional type of analysis, but which are a solid ground, a starting-point for a linguistic thought, and at the same time general and diversified in character. One should not think that narrative analysis is an infantile discipline because it is busy with children's tales. Of course, the analysis of a tale is not the analysis of a novel by Proust, but that does not mean that the linguist has first to analyze the work of Proust. The analysis of simple narrative structures allowed us to assemble a number of narrative models which can be used as hypotheses of predictibility for the analysis of more complex discourses. Narrative models can be projected upon texts of a considerable length; those models serve in the first place to segment long and complex narrative discourses. On the other hand, there is a hypotaxic organization of narrative discourses familiar to linguists: micro-stories are taken in charge by stories on a hierarchical superior level, and constitute an organization which is comparable to that of a complex sentence. A new step was the recognition of the 'actantial structure' (structure actantielle), i.e., a certain constant distribution of the dramatis personae. Such a structure is not figurative but abstract: the actors are not necessarily humans or animals or even magical objects, they can be constituted by conceptual content. This allowed the magnification of the narration problem and the recognition, for example in philosophical discourses, of a narrative structure which has an 'actantial' character. Narrativity ceases to be such and such a way of telling a story and appears as a general principle of organization of discourses of all sorts which can be at the origin of the construction of a narrative grammar. So in principle, there would not be any reason to say that the analysis of poetry is more difficult than the analysis of tales. It has always been thought that a poetic text is more marked by the subjectivity of the author than a narrative text. - The differences mainly arise from the way we conceive the various deep levels of language. At the surface, discourse is often covered by a figurative level, made of images of the world and filled with objects, with characters; if the existence of a deeper level underlying the figurative one is accepted, one observes that various changes represented figuratively in narration can get a logical expression on the level of abstract structures. Let us take for example Lévi-Strauss's book Le Cru et le Cuit.u It analyzes a corpus of myths on the origin of cooking, i.e., a number of tales telling how humanity has passed from raw food to cooked. This transition can be formulated on the level of deep structure as a logical transformation situated between two discrete terms raw ->• cooked. But it can also be observed that on the surface of the narration, the discontinuous space can be filled with numerous stories about the quest for fire, about the gods' or the animals' learning to cook, etc.: everything is possible for the human imagination. 14

C. Lévi-Strauss, Le Cru et le Cuit, Mythologiques

I (Paris, 1964).

66

A. J. GRE1MAS

Let us come back to poetry. Poetic discourse, at least from the middle of the 19th century, is characterized by the fact that it almost directly manifests transformations without all the figurative stories that were commonly used in Racinean or Shakespearean poetry. If this difference is taken into account, one sees that poetic discourse is more difficult to analyze since it asks for a direct representation of the abstract level, but at the same time it is easier, if one has the necessary methodological tools. In your structural semantics you distinguish between latent structure and manifest structure of discourse. Can both notions be identified as the deep and the surface structure, in transformational grammar ? - Yes and no. Since one has chosen to explore the depths of language rather than to ascend the very high summits reached in the construction of meta-meta-Ianguages, confusions in that domain get more and more numerous, mainly in Europe, where both dichotomies are used in parallel. First, what is manifestation?1n some way it is the presence of the objects in front of us: in order to be a language each language has its material cover, the signifier. Consequently, only if the two levels of language are conjoined, only if the phonematic component is introduced, can one talk about manifestation. If manifestation defined that way is compared to the dichotomy of surface and deep structure, one sees that deep structure can be manifest as well as surface structure - what is said or written in the logical or mathematical language is also manifest. It is a dangerous confusion to think that manifest structures are surface structures: surface syntactic structures are in fact the forms of the content analyzed previously to their manifestation. The concept of surface does not coincide with manifestation. A new confusion arises in saying not only that manifestation and surface structure coincide, but also that surface structure has to be seen as the denotative level of language, that language is first of all a denotation of things. Hjelmslev and Chomsky are equally responsible for this. This belief in the original innocence of language leads one to admit implicitly that language names things and processes in a correct, univocal and monosemic way, that ambiguities are anomalies that have to be treated as exceptional cases. It is difficult to imagine what would have been the form of generative grammar if its author, instead of taking Descartes' Discours de la Méthode as the model for the 'normal' functioning of language, had taken a modern poetic text.... The third factor of confusion sprouts from the fact that linguistics comes closer to psychoanalysis. European linguistics, while being a guide for other human sciences, has at the same time been influenced by the theories that these sciences propose, for example by the Freudian theory. One of the most demanding representatives of Freud in France, Jacques Lacan, is at the same time the one who uses most linguistic methods to nourish his psychoanalysis.15 The result of it is that the Freudian distinction between what is manifest and what is latent is piled on"top of the rest and raises 15

J. Lacan, Ecrits (Paris, 1967).

A. J. GREIMAS

67

a new terminological ambiguity. What is manifest and immediately given is everyday language - current language will thus be called the signifier, even if one sets as the goal of the research the discovery of a new signified hidden under that signifier, in other words, of a latent meaning level. So again there is a possible confusion between deep structures in linguistics and Freud's latent structures. Obviously in Hjelmslevian terms, the signifier of psychoanalysis is already a language with two levels where the latent signified is in fact a meta-signified. That is dangerous from a double point of view: first if natural language is considered as a signifier, one forgets the whole problem of natural language as such while only exclusively researching the second and so-called 'deeper' meanings. It is as if you said to me "This table is round" and I would refuse the message you addressed to me about the roundness of the table in order only to recall the interrogation "Why the hell does he make that assertion ?" I do not deny the existence of your conscious or unconscious backthoughts ; what puzzles me is that one does not distinguish between, on the one hand, psychoindividual and socio-cultural connotations and, on the other hand, the metasignified itself, which may be assimilated to the semantic deep structures. Across the Atlantic the innocence of language is implicitly admitted; in Europe language is over-guilty, it is the perfect tool to mask the truth. On both sides of the Atlantic there is a sort of contradictory valorization of language now considered as a euphorical object, then as a disphorical object: it should only be considered as a scientific object. A good and methodical transposition of the opposition between manifest and latent in mythological analysis has been made by Lévi-Strauss. 16 He interpreted it as a possible double reading of a myth: a surface reading which is horizontal and syntagmatic, which shows the successive troubles of Oedipus, is compatible with a second reading in depth, vertical or paradigmatic, which shows cultural problems where two contradictory interpretations of parenthood affront each other. Such an analysis which shows how society thinks about itself through its tales raises the problem of linguistic levels in terms of a double semantic structure. What you call the elementary structure of meaning is thus a latent structure. Is this latent structure logico -semantic ? - It is easy to cumulate those terms and to talk about 'logico-semantic' structure. You will say that that is obvious for a semanticist - all logical categories are semantic for him by definition. It has to be stressed that what I call elementary structure of meaning is a construction that uses very abstract semantic categories which can be considered as undefined or nondefinable. They could also be designated as 'universals of language', but the name would be too restrictive because natural languages are not the only ones concerned. This is a problem of general semiotics, one of the essence of meaning. So terms such as 'relation', 'identity', 'difference', 'negation', 'assertion', 16

Cf. note 13.

68

A. J. GREIMAS

'grasping' (saisie), 'generation', 'description' are elementary concepts that are not definable as such. They have mutually interdependent definitions and by these 'inter-definitions' one can construct a model that accounts for the conditions of the grasping and/or the production of meaning. This model can be considered formal because of the very great generality of its categories. As the form of the content, it can get various semantic investments - it serves as a structure ab quo for the generation of complex discourses. On the other hand the categories that constitute the elementary structure can also be called logical categories; if they are rapidly surveyed, one immediately observes that they also serve for the construction of all logical languages. This is not astonishing. General semiotics contains a set of formal languages among which the linguistic is found, next to logico-mathematical languages which serve to describe other less formal languages. The procedure for the construction of the elementary structure is based on the principle of 'interdefinition'. Having posed the concepts that cannot be defined, i.e. the universals, one has to try to organize them in relational networks in such a way that they are mutually defined, the one in relation to the others. In that sense I think it is correct to say that the elementary structure is logico-semantic. It is logical because such an articulation of mutually defined concepts makes it operational, yet it is semantic at the same time because it is the forms of the content that have been used as tools for the construction. Obviously, such an elementary structure constitutes the a priori for all possible axiomatization, for all theoretical grammatical construction.

Would the construction of such an elementary meaning theory no longer belong to linguistics ? - If it does not belong to linguistics, it belongs to semantics, however, or if you like, to a general semiotics. Obviously, there is the question: who is competent to talk about those kinds of problems? Is it the philosopher, the linguist, or the logician who directly needs this operational apparatus? At first sight it is the most fertile soil for epistemology or the thought on the conditions of knowledge and in particular on the conditions of scientific knowledge. However, philosophical exploration is distinct from the semanticist's approach: the latter, instead of contemplating universals disinterestedly, will try to manipulate concepts like relation or term in order to make them operational to achieve a scientific goal. The epistemological object is in both cases the same but the perspective with which the conceptualization will be made is not.

You regularly allude to Merleau-Ponty in your writings. Is the meta-language in which the elementary structure of meaning is expressed psycho-perceptive ? In other words, what is the role of perception for the constitution of meaning in language ? - I will confess to you something that being a linguist I should not say. Of course I have been influenced both by Merleau-Ponty and by Husserl, and by phenomeno-

A. J. GREIMAS

69

logy in general. 17 That is the epistémè of my youth. It is nevertheless clear that linguistics should succeed in putting between brackets the linguists's philosophical a prioris. Let me make my attitude more precise. Linguistic fact, as every phenomenon, indeed has several possible instances of which the grasping can take place. Let us take as an example the level of expression in language. Linguistic analysis can consider in the first place the instance of phonation, to x-ray the movements of the phonation organs and to reconstruct the program of the production of sounds and their succession. It can also consider the acoustic instance yielding a physical analysis of sounds. In the end, it can concentrate on the perceptive instance, which again is of a physiological nature. Have I, when formulating the elementary structure, privileged such and such an instance of seizure to the detriment of the others? It is quite possible, but I would rather say, when talking about the grasping of meaning, that one can use the spatial metaphor in 'seizure' to imagine every subject aiming at the object of his research. I think it is a metaphor common to all sciences of our era, a 'scientific' metaphor, as there are many. Without denying the influence I underwent, I would simply say that the problem of the psychological perceptive 'seizure' is essentially reduced to the recognition of the founding importance of the 'split' (l'écart), to use a Saussurean term. To seize meaning is to seize splits, differences. This definition of the minimal conditions of the appearance of meaning exemplified at the level of perception, can it not be represented in another way ? Maybe. But what is gained in saying, for example, that to produce meaning is to create a split ? The condition of all science is the introduction of discontinuity in continuity. To seize differences only means to produce discontinuity. But the recording of differences immediately raises the problem of the identity: one cannot conceive of a scientific practice without it presupposing the concept of identity, but that concept in its turn presupposes difference : in order to have identity, there must exist at least two identifiable phenomena. So what first appears as a psychological starting-point is immediately transformed into a logical demand. The elementary structure of meaning is materialized, if you like, as a perceptive seizure, but it is also and mainly the basis for a meta-language that can found the logical and semantic operations. What relation is there between linguistics and psychology? Between linguistics and sociology, sociology being the other basic human science ? - Even if I refused psychologism when talking about the structure of meaning, I think that the current trends in linguistics lead us to the middle of psychology. There is a revaluation of traditional 17th and 18th century psychology. This revaluation consists in conferring a semiotic status to psychology. Let us explain that. When the concept of competence is introduced in linguistics, this already means a certain return to the psychology of mental faculties. That is not all. People nowadays 17

Cf. A. J. Greimas, Sémantique structurale (Paris, 1966), pp. 8-10.

70

A. J . G R E I M A S

are more and more interested in the analysis of modalities. But what are modalities then? They are only logical formulations of the concepts which also belong to the psychology of mental faculties and finally to classical philosophy. When I propose for example to consider narrativity an articulation of the subject along the three modalities of volition, knowledge and power, I only try to give a logical form to concepts which indeed can be considered 'psychological'. One rather recycles psychology that way, and this is only possible through its formalization. In a sense it might be said that semiotics is 'good psychology'. There is now a new dichotomy opposing the 'utterance act' (Vénonciation) to the 'utterance' {Vénoncé). It comes from Benveniste, 18 but at the same time alludes to Austin's work and to the speech act.19 Does this not remind you of the phrase often heard in European classes of philosophy about "the thinking thought which thinks the thought thought" ? Then also we have a problem that is philosophical and psychological at the same time and which has a linguistic formulation : the problem of the conception of language which as an act of uttering poses its own utterance as an object. This dichotomy is only efficient because it is formulated in logico-semantic terms. At last I would like to stress the congruence of linguistics and psychology. In the framework of the research on narrative structures we want to bring to light an actantial structure whose parallelism with the psychological reflection on the relations between subject and object and other psychoanalytical instances is evident. I would not go so far as to say that actantial analysis tries to formalize psycho-analysis, but through certain aspects it appears as a 'syntax of the interior life'. You ask me what I think about the relation between linguistics and sociology. European linguistics of Saussurean inspiration remains quite sociological despite the appearance to the contrary. In Chomsky the speaker coincides mutatis mutandis with the psychological subject; European linguists will always see the speaking subject as a collective subject - it is not us who speak the language, it is the language that speaks in us. We bathe in language as a social reality: in a sense language is society. In a philosophical perspective, the problem for Chomskyism is to assure the transition from the psychological to the transcendental subject; our problem would be the transition between the collective subject to the 'universal human mind'. But enough philosophy. So one can understand why French structuralism has been welcomed by anthropology in the first place. The latter has to work on collective representations: linguistics was a sister science sharing the same preoccupations. The interpretation of mythology as the figurative form of the ideology of society is a problem that can be analyzed in both disciplines, for example. This also explains that at one point structuralism was misconstrued in France as a philosophy. They thought it had ambiguous relations 18 19

E. Benveniste, Problèmes de linguistique générale (Paris, 1966), pp. 225-288. Cf. especially J. Searle, Speech Acts (Cambridge, 1969).

A. J. GREIMAS

"71

- relations of interpénétration, but also of opposition - with certain forms of Marxism, which also is an explicatory theory of society. Do you make any difference between what in Saussurean tradition is called the collective subject, and the speaker? A linguistics of the ''utterance acf or the ''speech act'' would perhaps more be a linguistics of the speaker than of the collective subject... - No. I think the problem has to be seen in another light. The scientific attitude consists in only believing in what one sees, obviously in the metaphorical sense of that expression, i.e., to consider language only in and through its manifestations. The utterance act (/'énonciatiori) and the subject of the utterance act only exist in the measure they can be found in the text itself in one way or another. The procedure that allows the assertion of their existence is one of a logical presupposition. If you ask questions about a sentence wondering who is the speaker, you admit that the existence of the utterance presupposes the existence of the utterance act or the speech act. When it is not one single utterance that you analyze but a whole discourse, there is a chance that you will find not only one but several subjects of the utterance act. The subject of the utterance act can be explicit, as it is in stories where the T - f o r m is used; it can also remain implicit, but the explicitation of the speaker immediately raises the question of the new implicit speaker who has to be explicited. Thus, one has to make a decisive choice which will distinguish a regular scientific project from a laxist attitude: scientific means to me an attitude that defines the subject of the utterance act, or the speaker, as the logical presupposition of the utterance, and nothing else. All other differences lead us to identify the speaker with the psychological subject or even the ontological subject. In literary semiotics, for example, one ends in asking questions about the author of a text, his origins, the influences he has undergone, i.e., all the psychological imponderables which were so difficult to discard before. Let us talk about the functions of language. Saussurean linguistics mainly characterizes language through its communicative function, as opposed to Chomskyan linguistics which sees language essentially as expression of thought. - I would first like to stress the polysemy of the term function - it has at least three distinct meanings. First its utilitarian, instrumental sense; to the question "What is language used for?" the answer is obviously, to communicate, to express. In using this criterion, which is not linguistic at all, it will be easy to find a poetic function of language, a descriptive function, a subversive, revolutionary function, etc. I do not think a pertinent classification is possible on the basis of that criterion. There is a second meaning of the term function: its organicist sense. This sense comes from physiology or indirectly from Malinowski's anthropology: society is an organism that is a set of coordinated functions, and language is thus one of the functions of society. Finally, there is a third sense : the logico-mathematical. They have not used it much in linguistics yet. It is the one I use, for example, for the definition of the elementary utterance, where the function is considered to be the logical relation between the actors.

72

A. J. GREIMAS

If by functions of language are meant the instances of the grasping of it, the three well-known functions will be found, no matter what names they will be given: on the one side, the emission or expression, in the middle, the transmission or communication, and at the end of the channel, the reception or meaning. The originality of a linguistic theory consists of accentuating one or the other of those instances. The history of linguistics can be interpreted as a paradigmatic variation of the importance that was successively given to such and such a category or to such and such a term of that category: after a generation of phoneticians you have a generation of semanticists, the generative approach succeeds the interpretative approach, etc.... In principle there is no criterion that permits stressing one approach to the detriment of another: they are all valid, the more so if one postulates, as Chomsky seems to do, the translatability of one instance into another, if one accepts the idea that the processes of encoding and decoding are the same. Musical language for example can be studied on the basis of a score, but also on the basis of an analysis of the gestures of the pianist's hands, or also as a perceptive phenomenon. Are there instances of seizure that are 'more relevant', 'more pertinent' than others ? That is, I think, a question about efficiency, economy of means, perfectioning of the technical tool, etc. - consequently, it is a question of opportunity, not of principle. The paradigmatic variation in the history of linguistics affects of course the theoretical attitudes: but it is extremely fertile because it allows seeing things from one angle in a way that is not possible from another point of view. However, lucid theoreticians should interrogate all instances successively and the club of linguistic scientists should accept all attitudes. The continuity or the absence of continuity between animal language and human language, between articulated and gesture languages, is a question that is very much discussed in linguistic theory. Chomsky''s philosophical position is that language is the essence of man, so there is no continuity. Is your viewpoint as radical? - Maybe less radical. The problem of the origins and the evolution of language does not seem to me to be relevant for linguistics. As a first step, at least, I would replace it with the one of a general typology of languages. The problem of animal language - or of animal languages - is raised now because of the progress made in zoosemiotics. I did not follow it closely enough for my opinion to carry much weight. I would observe, however, that a classification based on transmission or on the sensorial canals used is not sufficient; this 'phonetic' approach has to be completed by a syntactic and semantic analysis of this type of language. At first sight it seems that many animals can form nothing but attributive utterances of a certain type, those where the subject coincides with the 'ontological subject'; they can only talk about themselves, express pain, joy of life, and other perhaps simplified human feelings. It is already different if the animals live in communities: a new type of communication exists which belongs to the group strategy. There are chiefs and subordinates, there is war and peace, problems of aggression and defense,

A . J. GRE1MAS

73

espionage too, in fact a whole strategy of a semiotic order, comparable to the systems of socio-cultural connotations underlying the behavior of humans in society and which belong to sociolinguistics. From a syntactic point of view the forms that animal communication takes in those cases are comparable to descriptive utterances of a mimetic type: the animal makes himself understood by reproducing behaviors and known situations. What is interesting in the production of those mimetic utterances is that they presuppose not only a comprehension of the behavior of others but also an implicit knowledge about the meaning of the behavior. Those are problems of the semiotics of gestures, common for the most part to man and animal. Take a man who digs a hole. That behavior is not only meaningful for himself - he knows the meaning of the program of his gestures - but also for any observer who is not engaged in a process of communication with him. Gestures that would try to imitate, to reproduce this meaningful behavior would transform meaning in communication, presupposing a mutual consensus, sort of a contract according to which the signs are emitted in order to be understood. Thus, all of the essential elements for a fundamental semiotics seem to be present in animal communication. This is to say that the main characteristics of animal languages are present as such in gesture languages of humans, even that the fundamental structure of communication based on a tacit interindividual contract is common to both types of languages. So the opposition you suggested is not posed between animal and human language, but between articulated or so-called 'natural' languages and other types of language. The somatic signifier, the human or the animal body, is not appropriate to produce a sufficiently complex code to create paradigmatic oppositions in order to stand for too long syntagmatic sequences. Phonation may be a complete somatic program of gestures: the signifier only becomes efficient when an acoustic discontinuum is recognized. If one knows that the discontinuity of the signifier corresponds, on the semantic level, to the taxonomic competence, one then sees how this double combinatorics creates the conditions of a relatively loose set of meanings, leading, through sentential syntax, to the production of utterances in indefinite number. However, it is not the 'double articulation' that constitutes alone the specificity of natural language. From the syntactic point of view it is necessary to add another characteristic that is at least as important: the capacity of splitting, through procedures of shifting, the utterance from the utterance act. So a man who is gesticulating or an animal mimicking behaviors never produce utterances with a subject that is different from themselves. Natural languages are superior in that they construct again and again utterances of which subjects are absent from the utterance act or are what is called 'non-persons'. Shifting does not only concern the category of person, but also the categories of time and space: they allow man to construct, on the basis of an arbitrary 'then' or a 'there', an abstract time and space which are logical without direct relation to the speaker and the hie et nunc situation. From a syntactic point of view, that is essentially what can be called the 'humanity' of man.

74

A. J. GRE1MAS

Saussurean structuralism has always been interpreted as a theory of linguistic immanence. I would like to ask you how you see the relation between linguistic meaning and the natural world. - I have been concerned with the natural world in a way different from positivistic linguistics. The latter, as you know, postulates on the one hand the existence of linguistic signs and on the other the existence of extra-linguistic referents. Saussure's principle was that the problem of the referent was not relevant to linguistics, which is concerned with form, not with substance. In the same line I have tried to consider the world of common sense, as it is profiled before us, not as a referent but as a signifier.™ It is a fact that the world signifies for us, that we understand what it is about when we talk about the objects of the world, that we not only attribute physical qualities to them, but also moral ones - look for example at the romantic conception of nature. It is obvious, in a sense, that nature is but the reflection of culture, that the individual invests meanings in it in order to read them afterwards. As a signifier, the world is thus a readable signifier. From that point of view one can speak of the language of the world and of a semiotics of the natural world - 'natural' in the same sense as when one speaks of natural languages. Confronted with this world of meanings, semiotics, which tries to decipher it, seems to be an interpretative semiotics: here, it is rather difficult to imagine the mechanisms of the generation of meaning. However, the world is not only humanized by the meanings one confers upon it but also by an artificial environment that man constitutes for himself. Let us think about the semiotics of environment, a young discipline which treats problems of urbanism. The architect who constructs a new city is confronted with the meanings he has to give to those new forms: the houses, the streets, the open spaces; some even imagine that a new city could give a new meaning or a new sense to the life of the inhabitants. This is, through an abstract ideological model, a real generation of complex semiotic objects which can even have at the limit therapeutic effects.... Of course, we are talking here about domains that are covered by non-linguistic semiotics, but this semiotization of nature, once it is admitted, allows for the comprehension of the statute of so-called situational contexts which often intervene in linguistic discourses. One understands that they are not a foreign element - it rather is the substitution of a discursive sequence belonging to a certain semiotics by an equivalent sequence from a different one. The mixing of codes is a frequent phenomenon in semiotics: the language of movies for example uses four or five intermingled codes of manifestation, which all converge to produce a global meaning.... There is an enormous confusion around the concept of 'contextuality\ Is there an opposition between semiotic and situational context ? - No, certainly not. Human life consists in passing from so-called real situations to 20 A. J. Greimas, "Conditions d'une semiotique du monde naturel", in Du sens. Essais semioliques (Paris, 1970), 49-92.

A. J. GREIMAS

75

narrated or imaginary situations. Take a discourse that reproduces a dialogue. The characters who are supposed to converse in the real world are transposed, described in the discourse. Inversely, it can be said of a conversation that it constitutes one discourse with two or several subjects producing the utterances. The modes of existence of semiotic beings can, of course, be different, but the structures that allow apprehending and describing them are not that different. I already talked about auto-narrativity, stories that man tells himself, upon which it is possible to build sort of a syntax of the interior life. The same is true of the semiotic organization of social or inter-individual relations. Things are not said the same way when a superior addresses an inferior, a man a women, when one talks to someone one knows or someone one does not know: hence, there exists a secondary syntax, a sociolinguistic syntax which again is comparable to narrative syntax. There are three modes of presence of semiotic structures: one on the level of the speakers in the so-called real world, one on the imaginary level of discourse, and finally one on the level of the interiorized structures that constitute the personal life where man places the objects of his desire, where he contradicts, etc. We have passed now to the heart of philosophical concern. The structuralist hypothesis of a closed linguistic universe can be examined as to two aspects: the relation of language to reality but also the relation of language to thought. - The Saussurean tradition is quite clear on this point. What Saussure had in mind was not the negation of thought nor the negation of the so-called physical reality but their exclusion from linguistic concern. Signifier named the contribution of physical substance and signified that of psychic substance, to the constitution of linguistic form: he tried to discard the essences in order to better put into light the semiotic status of those two components of language. So if I called meaning the relation of reciprocal presupposition between signifier and signified, I only interdefine those concepts and I make them operational without any metaphysical involvement. It is important to maintain the autonomy of linguistics. It guarantees serenity in research and sustains a good understanding inside the club of linguists. A physicist is not asked whether he is a believer or an atheist. Why should one ask this to the linguist? Metaphysical problems do not disappear, but if one introduces epistemological postulates that are not absolutely indispensable, the club of scientists inevitably becomes a group of sectarian partisans. I do not think for example that the concept of competence demanded the assertion of the innatist hypothesis. The deontology of the semiotician should consist in being content with an epistemological minimum. Thought and reality are nebulae, as Saussure said - it is better to replace them by operational concepts. The problem of the referent - of a double referent, the physical and the psychic one - disappears at the same time. Whatever ontological status may be given to reality, we are only concerned with it to the degree that it is observable and describable. It is obvious that I do not negate the existence of referential structures, but they should not be confused with semiotic structures. Yes, the world exists as it

76

A. J. GREIMAS

is, but it exists also according to what it signifies. It is the world of common sense, the world of perceptible qualities. More and more logicians, are interested in natural languages, and more and more linguists are concerned with logic. What is your opinion about the neologism of 'natural logic' or of 'natural grammar ' ? Is such a logic or such a grammar possible ? - It has already been a certain time since Lévi-Strauss stressed the existence of a ''concrete logic'. He noted that just as there were botanical, zoological or other great classifications, archaic societies also had classificatory categories which served to frame the animal kingdom, the vegetable kingdom, etc.21 Those categories constitute taxonomies and can be represented in tree diagrams, but they are not abstract and they manifest special and tactile properties of the objects which are to be classified. Ethno-science is a discipline that had been born from this kind of observation. It has been understood that each binary category could have a discriminatory role, that it had not to be abstract in order to be used as a tool for human thought. The same can be said about narration, which is nothing else than a figurative reasoning that is 'concrete' and 'natural', a reasoning which in order to talk about the world uses figures of that world. I once met a Lithuanian-American after a long visit he had made in Russia among people who are wrongly called 'simple people'. I tried to ask him questions about his life there, and I observed quite curiously that, instead of answering me directly, he began each time to tell me a little story of which only the implicit moral, just as in Jesus's parables, constituted an answer to my question. To tell things is to reason figuratively. Mythical stories appear to be what they really are : answers, in a narrative and figurative mode, to the great philosophical interrogations of humanity. In that sense one can speak of a 'concrete' logic, one can call it 'natural'. That is the big contribution of Lévi-Strauss to discursive linguistics. When he proposed his analysis of the Oedipus myth, 22 he showed that narrative reasoning could be formalized and defined, in this precise case, as a correlation of two binary and 'natural' categories, but at the same time those categories are located at such a depth that they are capable of generating all Oedipus myths, including the Freudian version. What a beautiful paradox. You are talking about Lévi-Strauss as about a generativist... - Maybe he did not use the term generation, since generative grammar had not been born yet, but his approach can certainly be considered to be generative: from simple deep structures he accounted for the production of surface discourses which were figurative and multiple. The multiplicity of discourses that sprout from one and the same model is explained by relations of paradigmatic transformations between them. If one tries to reflect on the linguistic research of the whole 19th century and to evaluate its results, one observes that it led to the constitution of a general model of an ideal proto-Indo-European language that had never existed; it is Saussure indeed who 21 22

C. Lévi-Strauss, La pensée sauvage. Cf. note 13.

A. J. GREIMAS

77

achieved the construction of this building.23 This model is a pure construction and accounts for the particular Indo-European languages. The deductive approach - and 'generation' is a deductive approach - has a quite honorable tradition in linguistics. So Lévi-Strauss, when he shows that from simple structures one can generate other structures that are more complex but transformationally related, he just enlarged the problem of 19th century linguistics. He extrapolated comparatism, which was up until then to the level of morphemes, to a typology of discourses. That would not have been possible without the mediation of comparative mythology founded by Georges Dumézil. 24 ... Can we come back once again to the question of the possibility of a ''concrete'' logic ? - As it is found implicit in natural languages and in discourses in natural languages, 'concrete' or 'natural' logic, with regard to the foundations of the logic, is in the same situation as constructed logics. It presupposes the existence of a meta-logic which has principles of identity, difference, excluded third, the relations of negation and assertion, in fact all undefinables that enter in the composition of the elementary structure.... The way in which natural languages use meta-logic can vary; the same is true of modal logics, for example, which, however different they are, are all founded on a few elementary meta-logical categories. However, if there exists a factual difference, not a difference by right, between natural logics and artificially constructed logics, it resides in the fact that constructed logics seem to a large extent to be obliged to account for adequacy, between the judgments they formulate and the referent, whereas natural logics are content with their sole internal coherence. If I understand well, the strains that exist now, with the logicians as well as with the linguists, have no reason to exist. One wants to apply certain logical models directly to linguistics without passing through this meta-logic... - The strain you are talking about comes, I think, from the exaggerated and unmerited belief that linguists and logicians have a priori in the certainties of their neighbors' discipline. Linguistics considers that it has to study all languages whatever they be and could thus very well pretend that the analysis and construction of logical languages appertains to it. In fact, it would only answer the calls of a Russell, a Reichenbach or a Wittgenstein, who all consider linguistics or at least natural languages as the source and the foundation of logic, but whose properly linguistic reasonings do not always seem very convincing. It is difficult to predict what would come out of this. What is sure is that the different logics that constantly confront linguistic facts with their often chimerical reference pose the problems in terms of true or false, whereas the linguist tries to recognize what is and what is not; those logics are therefore not directly

23 F. de Saussure, Mémoire sur le système primitif des voyelles dans les langues indo-européennes (Thesis) (Leipzig, 1879). 24 Cf. note 14.

78

A. J. GREIMAS

applicable to linguistic analysis. If the meta-logic is unique, logics which are constructed or made explicit through natural languages are multiple. All this shows us what you understand by generativity of languages: languages are 'generated' from that meta-logic... - Exactly. Those elementary structures, paradigmatic as well as syntagmatic, are the basis of all generation in grammar or, even more generally, of all meaning structure. It is time to conclude. Linguistics is frequently reproached by philosophers for being reductionist in principle. - That problem is not proper to linguistics. It is the scientific project of linguistics and not linguistics itself that is so accused, because every scientific approach is reductionist by definition-and it cannot be otherwise. Has botany ever been considered to be the description of all plants, and zoology, of all animals, taken one by one? Every scientific approach supposes the choice of a definite level of generality and the treatment of individuals inside classes. Scientific research is first of all a research for invariants. You certainly know the famous examples of Descartes' piece of wax. This piece can take all forms - it can appear to be liquid or solid, and nevertheless it still is wax. Wax is what is invariant in wax. Linguistic objects are of this kind, deprived of all aspects of concrete life. Linguistics has never tried to cheat anybody, and philosophers are responsible for their own illusions. Could you characterize in a few words what you understand by language as a scientific object ? - Your question goes outside the scope of linguistics. If linguistics, instead of being concerned as it is now with natural languages, proclaimed that its object of knowledge is all possible languages, then it could be said that linguistics as a language talking about all other languages is the ultimate scientific object. Since that is not the case, it is the general semiotics, the 'semiology' announced by Saussure, 25 which contains linguistics as a particular semiotics, that will appear the goal of the general scientific project. The terminological change I just introduced stresses the specificity of the scientific object we are talking about: semiotics cannot but be the science of meaning and of meanings - it explores the world in trying to extract from it all the modi significandi. General semiotics would thus be in a first stage the systematic exploration of the general conditions of meaning, taking the form of a theory of meaning. But since there is no such theory, but numerous theories of meaning, general semiology could well become, even if it does not like it, a typology of the theories of meaning, and, at the same time, the place where those theories are evaluated. Simplifying things, we could perhaps say that what science aims at when it poses language as its object is its 25

Cours de linguistique générale, pp. 32-36.

A. J. GRE1MAS

79

definition, if one understands by definition the coherent and exhaustive description of all the properties of the object, which exhausts this object until it coincides with it. Your great originality in this panorama of contemporary linguistics is to have seen that structural semantics is a semiotics, the semiotics. What is the relation between semiotics and linguistics ? - One has to distinguish between the situation de facto and de jure. Theoretically, linguistics is a discipline among others inside general semiotics. In practice, however, one often tends to consider semiotics as the extension of linguistics. This actual situation can be explained for two reasons. First of all, linguistics appears to be an established science with relatively sure methods and a praxis that is recognized and taught: it is normal that the young semiotics turns to linguistics, borrowing its approach and its procedures. But the superiority of linguistics sprouts also from the fact that the natural languages it studies are able to paraphrase the other semiotics, whereas the inverse is rarely possible. And so natural languages are often confused implicitly with the logico-semantic languages that serve as scientific meta-language in semiotic analysis, but which are manifested the same way as natural languages are. This confusion goes even further; at all times it menaces the distinction between what is signified and that with which it is signified, between the languages of manifestation which constitute the object of particular semiotics - pictural semiotics, musical semiotics, cinematographical semiotics etc. - and meaning which passes, as through mazes, through those multiple and often intermingled languages. The fact that the world is human and that this world signifies through all languages confers upon semiotics a particular statute: that is why it is sometimes accused of scientific imperialism, why it is considered as putting its fingers into everything. Actually, what semiotics is searching for is first to propose a scientific methodology to the various domains of the humanities, such as 'literature', or the different 'histories of art', which are deprived of it and the scientific character of which is doubtful. But it mainly wants to bend the approach and the tackling of the various human sciences so as to make their methodology and at the same time their objects of study comparable. That way, semiotics could contribute to the epistemological and methodological unification of the social sciences. The task of linguistics, then, the first among the semiotics, appears capital in this 20th century, which must be the century in which the human sciences take their proper place, as the 17th century was the one in which the sciences of nature were constituted. Ecole Pratique des Hautes Etudes Château de Charbonnières, France July 21, 1972

M. A. K. HALLIDAY

Michael Halliday, you are one of the most representative linguists of what one might call the trend of sociolinguistics. You use terms like sociogrammar and sociosemantics; does that imply a very particular view on the scope of linguistics ? - I would really prefer to leave out the 'socio', if I had the choice. But we probably have to talk about 'sociolinguistics' these days, because of the shift in the meaning of 'linguistics'. When I was a student, with J. R. Firth, linguistics was the study of language in society; it was assumed that one took into account social factors, so linguists never found it necessary to talk about sociolinguistics. But during the last ten or fifteen years the pendulum has swung in the opposite direction, away from the social context towards the study of language from what I would call an 'intraorganism' point of view, or language as knowledge if you like; so that anyone who is concerned with the other, 'inter-organism' aspect of language, with how people talk to each other, has to prefix 'socio' to what he is doing. Hence you have sociolinguistics; and hence, also, 'sociosemantics' or 'sociogrammar'. Let me put it this way: these two perspectives - on the one hand the intra-organism perspective, language as what goes on inside the head (language as knowledge), and on the other hand the inter-organism perspective, language as what goes on between people (language as interaction, or simply as behavior) - are complementary and not contradictory. There tend to be fashions in linguistics, as in many other things. I started in a tradition where the perspective was mainly of the inter-organism kind. Then the pendulum swung the other way, largely through the influence of Chomsky who emphasized the philosophical and psychological links of the subject. And so those wanting to talk about language from the point of view not so much that 'people talk' but that 'people talk to each other' have called what they are doing sociolinguistics. I think both Hymes and Labov have pointed out that the 'socio' is really unnecessary, and I rather agree with them. You wrote that a good linguist has to go outside linguistics. What do you mean by that? - This is a related point. If you look at the writings of linguists in the 1950s, you find great stress laid on the autonomy of linguistics. Linguistics is seeking recognition as a subject in its own right; it has not to be evaluated against other disciplines. Now,

82

M. A. K. HALLIDAY

as long as you concentrate attention on the core of the linguistic system, on linguistic form (grammar and vocabulary), then the interrelationships that you are studying are - or can be treated as if they were - wholly bounded within language, since their immediate points of reference are also within language: on the one hand the semantic system, and on the other hand the phonological system. But once you become concerned with the linguistic system as a whole, including the semantic system, then you have to look outside language for your criteria of idealization. The essential question at issue is this: what are or are not two instances of the same phenomenon, two tokens of the same type ? The moment you include the semantic and phonological systems in your picture, then you are involved in the interfaces between language and something else: in one direction meaning, and in the other direction sound. The two are not symmetrical, of course, because the system is not symmetrical; the classic problem in phonology, the debate over the phoneme, is a debate about the nature of idealization at the 'output' end - in classifying two sounds together as tokens of the same type, do you look 'downwards' and take account of the expression system, or only 'upwards' towards the content? But when we are concerned with the grammatical system our point of reference is clearly 'upwards'. We relate the distinctions that we draw in the grammatical system (grammar and vocabulary) to the semantics. Where then do we find the criteria for distinctions in the semantic system? How do we decide what are or are not instances of the same meaning, tokens of the same semantic type? Only by going outside language. In practice most people, including many linguist?, without even really thinking about this issue quite arbitrarily use the orthographic system as their criterion of idealization. They assume that if two things are written the same they are the same, and if they are written differently they are different. (They are reluctant to accept, for example, that differences of intonation may realize distinctions within the semantic system, distinguishing one semantic type from another in just the same way that different words or structures do.) This in the last resort is circular. You cannot find within language criteria for semantic idealization, criteria for deciding whether two things are the same or are not the same in meaning. You have got to go outside language. The accepted way of doing this is to postulate a conceptual system. One says, in effect, we have a system of concepts, two concepts are the same or different, and that is how we decide whether two linguistic elements are the same or different. If we admit that there is a semantic system, a semantic level of organization within the linguistic system, then the question we are asking is "What is above that?"; and it is at that point that we move outside language. We are regarding semantics as an interface between language and something else, and it is to that something else that we go for our criteria of idealization. In that sense, the linguistic system is not autonomous. Only, once we admit that, we can then take account of the fact that there is more than one direction that we may go outside language. A conceptual system is not the only form that such a higher-level semiotic can take.

M. A. K. HALLIDAY

83

Do you stress the instrumentality of linguistics rather than its autonomy ? - These are not really contradictory. But there are two different issues involved when you talk about autonomy. One is: "To what extent is the subject self-sufficient?" My answer is: "It isn't." (But then what subject is ?) The second is: "To what extent are we studying language for the purpose of throwing light on language or for the purpose of throwing light on something else?" This is a question of goals; it is the question why you are doing it. In this sense the two perspectives are just complementary. Probably most people who have looked at language in functional terms have had a predominantly instrumental approach; they have not been concerned with the nature of language as such so much as with the use of language to explore something else. But I would say that in order to understand the nature of language itself we also have to approach it functionally. S o l would have both prospectives at once. It seems to me that we have to recognize different purposes for which language may be studied. An autonomous linguistics is the study of language for the sake of understanding the linguistic system. An instrumental linguistics is the study of language for understanding something else - the social system, for example. One needs for a relevant linguistic theory other larger theories, behavioral and sociological theories. One can find in your publications many allusions to Bernstein''s sociology. What does Bernstein mean for you ? - If you are interested in inter-organism linguistics, in language as interaction, then you are inevitably led to a consideration of language in the perspective of the social system. What interests me about Bernstein is that he is a theoretical sociologist who builds language into his theory not as an optional extra but as an essential component. 1 In Bernstein's view, in order to understand the social system, how it persists and changes in the course of the transmission of culture from one generation to another, you have to understand the key role that language plays in this. He approaches this first of all through the role that language plays in the socialization process; he then moves on towards a much more general social theory of cultural transmission and the maintenance of the social system, still with language playing the key role. To me as a linguist this is crucial for two reasons, one instrumental and one autonomous if you like. Speaking 'instrumentally', it means that you have in Bernstein's work a theory of the social system with language embedded in it, so that anyone who is asking, as I am, questions such as "What is the role of language in the transmission of culture ? how is it that the ordinary everyday use of the language, in the home, in the neighborhood and so on, acts as an effective channel for communicating the social system?" finds in Bernstein's work a social theory in the context of which one can ask these questions. In the second place, speaking 'autonomously', this then feeds back into our study of the linguistic system, so that we can use the 1

Basil Bernstein, Class, Codes and Control I: Theoretical Studies Towards a Sociology of Language (London, 1971), especially Chapters 7-10.

84

M. A. K. HALLIDAY

insights we get from Bernstein's work to answer the question: why is language as it is ? Language has evolved in a certain way because of its function in the social system. Why this privileged position of language in the socialization process, for Bernstein and for you? - I suppose because, in the processes by which the child becomes a member of society, language does in fact play the central part. Even if you take the most fundamental type of personal relationship, that of the child and its mother, this is largely mediated through language. Bernstein has the notion of critical socializing contexts; there are a small number of situation types, like the regulative context (control of the child's behavior by the parent), which are critical in the socialization of the child. The behavior that takes place within these contexts is largely linguistic behavior. It is the linguistic activity which carries the culture with it. You and Bernstein mean by language vocalized language and not other - Yes, although we would of course agree on the important role systems like gesture. Clearly the more that one can bring these into the more insight one will gain. But nevertheless language, in the natural language in its spoken form, is the key system.

systems of signs ? of paralinguistic the total picture, sense of speech,

Other linguists working in the field of sociolinguistics are Hymes and Labov. Is there again solidarity with these researchers ? - Hymes has adopted, in some of his work at least, an intra-organism perspective on what are essentially inter-organism questions. 2 This is a complex point. Let me put it this way: suppose you are studying language as interaction, you can still embed this in the perspective of language as knowledge. This is what is lying behind Hymes' notion of communicative competence, or competence in use. To link this up with the recent history of the subject, we should mention Chomsky first. The great thing Chomsky achieved was that he was the first to show that natural language could be brought within the scope of formalization; that you could in fact study natural language as a formal system. The cost of this was a very high degree of idealization; obviously, he had to leave out of consideration a great many of those variations and those distinctions that precisely interest those of us who are concerned with the sociological study of language. From this point of view Chomskyan linguistics is a form of reductionism, it is so highly idealized. Now, Chomsky's idealization is expressed in the distinction he draws between competence and performance. Competence (in its original sense) refers to the natural language in its idealized form, performance to everything else - it is a ragbag including physiological side-effects, mental blocks, statistical properties of the system, subtle nuances of meaning and various other things all totally unrelated to each other, as Hymes himself has pointed out. If you are interested in linguistic interaction, you don't want the high level of idealization that is 2

Dell H. Hymes, "Competence and Performance in Linguistic Theory", in Renira Huxley and Elizabeth Ingram (eds.), Language Acquisition: Models and Methods (London-New York, 1971).

M. A. K. HALLIDAY

85

involved in the notion of competence; you can't use it, because most of the distinctions that are important to you are idealized out of the picture. What can you do about this? You can do two things. You can say, in effect, "I accept the distinction, but I will study performance"; you then set up "theories of performance", in which case it is necessary to formulate some concept (which is Hymes' communicative competence) to take account of the speaker's ability to use language in ways that are appropriate to the situation. In other words, you say there is a "sociolinguistic competence" as well as a linguistic competence. Or you can do what I would do, which is to reject the distinction altogether on the grounds that we cannot operate with this degree and this kind of idealization. We accept a much lower level of formalization; instead of rejecting what is messy, we accept the mess and build it into the theory (as Labov does with variation). 3 To put it another way, we don't try to draw a distinction between what is grammatical and what is acceptable. So in an inter-organism perspective there is no place for the dichotomy of competence and performance, opposing what the speaker knows to what he does. There is no need to bring in the question of what the speaker knows; the background to what he does is what he could do - a potential, which is objective, not a competence, which is subjective. Now Hymes is taking an intra-organism ticket to what is actually an inter-organism destination; he is doing 'psycho-sociolinguistics', if you like. There's no reason why he shouldn't; but I find it an unnecessary complication. That is an interesting point here. What according to you is the role of psychology as a background-theory of linguistic theory ? Iam thinking here of Saussure's and Chomsky's view that linguistics is a sub-part of psychology. - I would reject that absolutely; not because I would insist on the autonomy of linguistics, nor would I reject the psychological perspective as one of the meaningful perspectives on language, but because this is an arbitrary selection. If someone is interested in certain particular questions, then for him linguistics is a branch of psychology; fine, I accept that as a statement of his own interests and purposes. But if he tries to tell me that all linguistics has to be a branch of psychology, then I would say no. I am not really interested in the boundaries between disciplines; but if you pressed me for one specific answer, I would have to say that for me linguistics is a branch of sociology. Language is a part of the social system, and there is no need to interpose a psychological level of interpretation. I am not saying this is not a relevant perspective, but it is not a necessary one for the exploration of language. We are now coming to one of the key points: your opinion about the relation between grammar and semantics, and also about that between behavioral potential, meaning potential and grammar. Can you say that there is a progression between to do, to mean and to say, in your perspective? - Yes. First let me say that I adopt the general perspective on the linguistic system 3

William Labov, "The Study of Language in its Social Context", Studium Generate 23 il970).

86

M. A. K. HALLIDAY

you find in Hjelmslev, in the Prague school, with Firth in the London school, with Lamb, and to a certain extent with Pike - language as a basically tristratal system: semantics, grammar, phonology. (Grammar means lexicogrammar; that is, it includes vocabulary.) Now, it is very important to say that each of these systems, semantics, grammar and phonology, is a system of potential, a range of alternatives. If we take the grammatical (lexicogrammatical) system, this is the system of what the speaker can say. This relates back to the previous point we were discussing - it seems to me unnecessary to talk about what the speaker knows; we don't need to be concerned with what is going on in his head, we simply talk about an abstract potential. What the speaker can say, i.e., the lexicogrammatical system as a whole, operates as the realization of the semantic system, which is what the speaker can mean - what I refer to as the 'meaning potential'. I see language essentially as a system of meaning potential. Now, once we go outside language, then we see that this semantic system is itself the realization of something beyond, which is what the speaker can do - I have referred to that as the 'behavior potential'. I want to insist here that there are many different ways of going outside language; this is only one of them. Perhaps it would be better at this point to talk in terms of a general semiotic level: the semantic system, which is the meaning potential embodied in language, is itself the realization of a higher level semiotic which we may define as a behavioral system or more generally as a social semiotic. So when I say can do, I am specifically referring to the behavior potential as a semiotic which can be encoded in language, or of course in other things too. One of your statements is that can mean is a form of can do. - Yes, and that could be confusing, because it is trying to say two things at once in an abbreviated form. To my mind, the key concept is that of realization, language as multiple coding. Just as there is a relation of realization between the semantic system and the lexicogrammatical system, so that can say is the realization of can mean, so also there is a relation of realization between the semantic system and some higherlevel semiotic which we can represent if you like as a behavioral system. It would be better to say that can mean is 'a realization of can do\ or rather 'is one form of the realization of can do\ Now, in the early sixties those of us who were interested in what people do linguistically were labelled 'taxonomic' by the transformationalists, who criticized us f o r being data-oriented, for looking at instances, for dealing with corpuses, and so on. To my knowledge, no linguist has ever simply described a corpus; this is a fiction invented for polemic purposes. The question is, what status do you give to instances of language behavior ? There are many purposes for which we may be interested in the text, in what people actually do and mean and say, in real situations. But in order to make sense of the text, what the speaker actually says, we have to interpret it against the background of what he 'can say'. In other words, we see the text as actualized potential; it is the actual seen against the background of the potential. But note that the actual and the potential are at the same level of abstraction. This is what makes

M. A. K. HALLIDAY

87

it possible to relate the one to the other. They are at the same level of coding within the system, so that any text represents an actualization (a path through the system) at each level: the level of meaning, the level of saying (or wording, to use the folklinguistic term for the lexicogrammatical system), and of course the level of sounding or writing. The key notion is that of realization, in the Hjelmslevian sense: each level is the realization of the lower level? - Rather of the higher level. The earlier tradition usually had the meaning at the top, not at the bottom! If you can speak of a teleology of your whole description, can you say that semantics or sociosemantics is the key to the whole system ? - Well, yes. If I was forced to choose a key, it would be that. This semantic level is structured- you use the term network. Can you explain this term 'semantic network' here? - I would use the term network for all levels, in fact: semantic network, grammatical network, phonological network. It refers simply to a representation of the potential at that level. A network is a network of options, of choices; so for example the semantic system is regarded as a set of options. If we go back to the Hjelmslevian (originally Saussurean) distinction of paradigmatic and syntagmatic, most of modern linguistic theory has given priority to the syntagmatic form of organization. Structure means (abstract) constituency, which is a syntagmatic concept. Lamb treats the two axes together: for him a linguistic stratum is a network embodying both syntagmatic and paradigmatic relations all mixed up together, in patterns of what he calls AND nodes and OR nodes. I take out the paradigmatic relations (Firth's system) and give priority to these; for me the underlying organization at each level is paradigmatic. Each level is a network of paradigmatic relations, of OR's - a range of alternatives, in the sociological sense. This is what I mean by a potential: the semantic system is a network of meaning potential. The network consists very simply of a set of interrelated systems, the system being used here in the Firthian sense, though perhaps slightly more abstract, and making fuller use of his own 'polysystemic' principle. Let me just define it: a system is a set of options, a set of possibilities 'A, B or C' together with a condition of entry. The entry condition states the environment: 'in the environment X, there is a choice among A, B and C.' The choice is obligatory; if the conditions obtain, a choice must be made. The environment is, in fact, another choice (and here I depart from Firth, for whom the environment of a system was a place in structure - the entry condition was syntagmatic, whereas mine is again paradigmatic). It is equivalent to saying 'if you have selected X (out of X and Y), then you must go on to select either A, B or C'. The 'then' expresses logical dependence - there is no real time here - it is a purely abstract model of language as choice, as sets

88

M. A. K. HALLIDAY

of interrelated choices. Hudson's recent book gives an excellent account of system networks in grammar. 4 Now this is what is represented in the network. The network is a representation of options, more particularly of the interrelations among options. Hence, a semantic network is a representation of semantic options, or choices in meaning. Is there any difference between a semantic structure and a grammatical structure ? - We may have some confusion here through the use of the term structure. May I use it in the Firthian sense: just as the system is the form of representation of paradigmatic relations, the structure is the form of representation of syntagmatic relations. The output of any path through the network of systems is a structure. In other words, the structure is the expression of a set of choices made in the system network. We know more or less what the nature of grammatical structures is. We know that constituent structure in some form or other is an adequate form of representation of the structures that are the output of the lexicogrammatical level. Itismuchlessclearwhatis the nature of the structures that are the output of the semantic level. Lamb used to draw a distinction here: he used to say that the semantic structures were networks, while lexicosyntactic structures were trees and morphological structures were strings. I don't think he holds to this any more. If you take the sort of work that Geoffrey Turner has been doing, of the investigation of language development in young children, where we have been using the notion of meaning potential in the form of semantic system networks, in this situation it has been possible to bypass the level of semantic structure and go straight into lexicogrammatical constituent structure. 5 That's all right for certain limited purposes. But there is obviously a limitation here, and when we attempt semantic representation for anything other than these highly restricted fields it is almost certainly going to be necessary to build in some concept of semantic structure. But what it will look like exactly I don't know. I don't think we can tell yet. Probably some form of relational network on the lines that Lamb and Peter Reich are working on. 6 The input of the semantic network is sociological and particular, and the output is linguistic and general. What do you mean by 'particular' on the one hand and 'general' on the other hand? - Let me take an example. Suppose you are interested, in a context of cultural transmission, in the way in which a mother controls the behavior of the child. She is expressing, through the use of language, certain abstract behavioral options, which 4

R. A. Hudson, English Complex Sentences: An Introduction to Systemic Grammar (= North Holland Linguistic Series 4) (Amsterdam-London, 1971). 5 Geoffrey J. Turner, "Social Class and Children's Language of Control at Age Five and Age Seven", in Basil Bernstein (ed.), Class, Codes and Control II: Applied Studies Towards a Sociology of Language (London, 1973). 6 Peter A. Reich, "Relational Networks", Canadian Journal of Linguistics 15 (1970). Cf. David G. Lockwood, Introduction to Stratificational Linguistics (New York, 1972).

M. A. K. HALLIDAY

89

we then characterize in terms which relate them to some model of the social system. In other words, she may be choosing among different forms of control - a simple imperative mode, a positional appeal, a personal appeal or the like, as in Bernstein's work; and when we show how this choice is encoded in language, what we are doing is deriving a set of linguistic categories from options in the social system. Now these will be very general categories, in the linguistic system: forms of transitivity, or forms of modification within the noun phrase, for example. But in order to get to them, we have to go through a network of behavioral options which become highly specific. A linguistic category such as 'clause type: material process, benefactive' appears (among other things) as the expression of some behavioral option that is highly specific in terms of the social theory, such as 'threat of loss of privilege'. The sociological categories which these linguistic ones realize will in relation to the social system be very particular, deriving from particular social contexts. You can relate this to the well-known problem of getting from the 'macro' scale of society to the 'micro' scale of language. This is wrongly posed, in my view; the problem is not one of size, but of level of abstraction. What are for language highly abstract and general categories have to be seen as realizing highly concrete and specific notions in the social structure. The whole difficulty is to define the relation on the one hand between the behavioral potential and the meaning potential and on the other hand between the meaning potential and the grammar. These two relations, that is what your linguistic theory has to define. What are the different conditions for a semantic network in connection with the other two levels of the whole theory? - I would see both these relations as defined by the concept of realization. The semantic network is one level in a system of multiple coding. There are two main trends in thinking about language, aren't there? There is the realizational view, language seen as one system coded in another and then recoded in another; and the combinatorial view, where language is seen as larger units made up of smaller units. Of course both these relations are found in language, but people assign them very varying statuses. If we adopt the first emphasis, which is the Hjelmslevian view, we can extend the realizational concept outside language, so that just as the lexicogrammatical system realizes the semantic system, the semantic system realizes the behavioral system, or the social semiotic. If we then consider any specific part of the semantic system, there are three conditions which our representation must meet. One is that it must associate this part of the system with other parts of the same system on the same level. In other words, we must be able to show what is the total semantic potential within which the particular set of options that we are dealing with operates. But at the same time, we must be able to relate it to the other systems in both directions: both upward and downward. That is, if we claim that we have identified a set of options in meaning, not only do we have to relate these to other sets of options in meaning in a systematic way, but we have also to show, first, how this set of options in meaning realizes an aspect of the social system, and secondly, how it is in turn realized in the lexico-

90

M. A. K. HALLIDAY

grammatical system. This is a very strong demand, in a sense, because if one says that there is a significant choice in meaning in social control situations between, say, moral disapprobation and other forms of disapprobation, as Geoffrey Turner does, or between imperative and obligative types of rule-giving, then one must be able to specify three things. One, exactly how this relates to the other options in meaning that have been set up. Two, how this expresses higher level behavioral options. Three, how this is in turn realized in the grammar. If we claim that a child can interpret the social system by listening to what his mother says, then presumably a linguist should be able to do the same. How can one define the dissimilarity of realization between the semantics and the grammar then ? In other words, what is the definition of grammar ? - Well, I am not very clear on the boundaries here, between lexicogrammar and semantics. I tend to operate with rather fluid boundaries. But it can be defined theoretically, in that the lexicogrammaiical system is the level of internal organization of language, the network of relations of linguistic form. And it is related outside language only indirectly, through an interface. I would also want to define it functionally, in terms of the metafunctions; we haven't come to that yet. Let us just say that it is the purely internal level of organization, the core of the linguistic system. With a grammatical and a lexical part ? - Yes, but - at least in my perspective; one might conceive differently for other purposes - the two are not really different. The lexical system is not something that is slotted in afterwards to a set of slots defined by the grammar. The lexicon - if I may go back to a definition I used many years ago - is simply the most delicate grammar. In other words, there is only one network of lexicogrammatical options. And as these become more and more specific, they tend more and more to be realized by the choice of a lexical item rather than by the choice of a grammatical structure. But it is all part of a single grammatical system. Is syntax also a component of the grammar ? - You notice I am avoiding the term syntax; only for this reason - that it has come into present-day linguistics from two different sources and so it has two different meanings. On the one hand you have syntax in the context of semantics-syntacticspragmatics, where it is defined in terms of a general theory of signs, on criteria which are drawn from outside language. On the other hand, there is the context in which you have semantics-grammar-phonetics, and then within grammar you have the division into syntax-morphology. This is a different sense of the term, where the criteria are within language itself: syntax is that part of the grammatical system which deals with the combination of words into sentences, or phrases into sentences. But I myself am not convinced of the traditional linguistic distinction between syntax and morphology, at least as a general phenomenon; I think it applies to certain languages only, and not

M. A. K. HALLIDAY

91

to all of them, and so I don't feel the need to use syntax in that sense. But I am avoiding using it in the other sense because of the confusion between the two meanings of the term. I would like to come back to the relation between semantics and grammar. Is it possible that a semantic option has more than one realization in the grammar ? - Yes. Well, that's a very good and open question, to my mind. Let me start by saying that I think we must admit theoretically that it is possible. We may have what Lamb calls diversification between levels. What this means is that, in addition to one-to-one relations in the coding system, where one element on one level is realized by one element on another level, you may also have many-to-one and one-to-many. Now here we are talking about one-to-many; in other words, the phenomenon where one element in the semantic system is realized by more than one in the lexicogrammatical system. First, then, we must admit theoretically that this happens, that there is free variation in the grammatical system, with one meaning realized by two or more forms. But then I would add that we should always be suspicious when we find this, because it usually turns out that the distinction in the lexicogrammatical system does in fact express a more delicate distinction in the semantic system that we haven't yet got round to. In other words, let us not go so far as to deny free variation, but let us be highly suspicious of any actual instances of it, because very often it turns out that there is a more subtle or more 'delicate' distinction in the semantic system which is being expressed in this way. Can we go so far as to say that the grammatical system is arbitrary in connection with the meaning differences ? - What do you mean by arbitrary ? In the Saussurean sense the relation between signifiant/signifié is arbitrary. There is no isomorphism between the two levels. This seems to be important because in generative semantics each syntactic difference means at the same time a semantic difference. There is no arbitrary relation between syntax and semantics there. - Well, I would tend to agree with this. When we talk about the arbitrariness of the sign, we are referring to the Saussurean content/expression relation. I believe every linguist must agree that there is arbitrariness at this point. But there is I think just this one point in the whole linguistic system where we can talk about arbitrariness - that is, at the line that is drawn by Hjelmslev between content and expression. The relations across this line are arbitrary; this we must accept. But if we are considering the relation between semantics and grammar, which is all within Hjelmslev's content, then I would say it is not arbitrary. Consider a grammatical structure. A grammatical structure is a configuration of roles, or functions if you like, each of which derives from some option in the semantic system - not one to one, but as a whole. Let us take an example from child language. The child says water on, meaning

92

M. A. K. HALLIDAY

"I want the water tap turned on." We relate this to some general meaning or function for which the child is using language: in this case, the satisfaction of a material desire. We can see that the grammatical structure represents this very clearly. It consists of two elements, one identifying the object of the desire, i.e., water, and the other specifying the nature of the request, i.e., on. We express this by means of structural labels. It is clear that the grammatical structure here is a non-arbitrary configuration of elements which, taken as a whole, represent the function for which language is being used, and each of which constitutes a particular role within that function. Let me say in passing that this was said by Malinowski fifty years ago, when he pointed out that the elementary structures of the child's language represented very clearly the functions that language served for it.7 I agree with this, but I would go further and say that it is also a property of adult language: if you take a grammatical structure, for example a transitivity structure that we represent in terms of categories like agent, process and goal, or a modal structure, each of these grammatical structures represents a configuration that is derived as a whole from the semantic level of which it is the realization. So, in that sense, I would consider that the linguistic system at that point is non-arbitrary. The arbitrariness comes in simply in the relation between the content and the expression. Is it possible to relate all that you said about the scope of linguistics, and about the relationships between behavior, meaning and grammar, to the functional aspect of your theory of language ? - Yes. I would accept the label 'functional' and I think the point that we have just been discussing provides an excellent illustration of this. Consider any sentence of the adult language, for example in English Balbus built a wall. Taking up what I said just now, this represents a configuration of roles, or syntactic functions, a configuration which is not arbitrary since it represents very clearly the meaning of the sentence as a set of options in the semantic system. We can now go on to say that this sentence embodies a number of structures all at the same time; there are represented in that sentence at least three - let us confine ourselves to three - different structural configurations, each one of which corresponds to a different function of language. On the one hand, there is a transitivity structure involved in it; we could characterize this as Agent + Process + Goal of result. Now this configuration represents the function of language expressing a content, what I prefer to call the ideational function: language as expressing the speaker's experience of the external world, and of his own internal world, that of his own consciousness. But on the other hand that clause has structure also in the modal sense, representing what I would call the interpersonal function of language, language as expressing relations among participants in the situation, and the speaker's own intrusion into it. 7 Bronislaw Malinowski, "The Problem of Meaning in Primitive Languages", supplement I to C. K. Ogden and I. A. Richards, The Meaning of Meaning (London, 1923).

M. A. K. HALLIDAY

93

So the clause consists simultaneously of a Modal element plus a Residual element. The Modal element expresses the particular role that the speaker has chosen to adopt in the situation and the role or role options that he has chosen to assign to the hearer. A t the same time the clause has a third structural configuration, that in terms of a Theme and a Rheme, which is its structure as a message in relation to the total communication process - expressing its operational relevance, if you like. The point I want to make is this: in my opinion all these three - and I would be prepared to add one or two more - structural configurations are equally semantic; they are all representations of the meaning of that clause in respect of its different functions, the functions which I have referred to as ideational, interpersonal and textual. So in all these cases the structure is not arbitrary, to link up with what we were saying before. Is there any difference between the typology of the uses of language and the typology of the functions of language ? I believe that you define the function as a discrete area of formalized meaning potential. - Right. I would like to make a distinction between function and use, just as you suggest, and somewhat in these terms. A s far as the adult language is concerned, it is possible to talk about the 'uses' of language, by which I would understand simply the selection of options within the linguistic system in the context of actual situation types: 'use' in its informal everyday sense. In that sense, use is a valuable concept; but we can't really enumerate the uses of language in a very systematic way - the nearest we can come to that is some concept of situation types, of which Bernstein's critical socializing contexts would be an example. N o w I would distinguish that from function, because the whole of the adult language system is organized around a small number of functional components. The linguistic system, that is to say, is made up of a few very large sets of options, each set having strong internal constraints but weak external constraints. By 'strong internal constraints' I mean that there is strong environmental conditioning on choice: if you make a certain selection in one system within that set of options, this will determine up to a point the selection you make in other systems within the same set. Whereas the external constraints are weak; that is to say, the selection does not affect the choices that you make in the other sets of options. Take for instance the structure of the clause. There is one set of options in transitivity representing the type of process you are talking about, the participant roles in this process and so on. This is a tightly organized set of systems, each one interlocking with all the others. And there is another set of options, those of mood, relating to the speaker's assignment of speech roles to himself and to the hearer, and so on; there systems are again tightly organized internally. But there is little mutual constraint between transitivity and mood. What you select in transitivity hardly affects what you select in mood, or vice versa. N o w what are these components ? Fundamentally, they are the components of the language system which correspond to the abstract functions of language - to what I have called 'meta-functions', areas of meaning potential which

94

M. A. K. HALLIDAY

are inherently involved in all uses of language. These are what I am referring to as ideational, interpersonal, and textual; generalized functions which have as it were become built into language, so that they form the basis of the organization of the entire linguistic system. Would you identify the function of language with the structure of language ? - May I make a distinction here between two uses of the term structure? 'Structure' can be used in a sense which is more or less synonymous with 'system', where 'structure of language' means, in effect, the linguistic system. I have been avoiding using the term 'structure' in that sense, in order to avoid confusion; so let me comment on function and system first. The linguistic system is functional in origin and in orientation, so that in order to understand the nature of the linguistic system we have to explain it as having evolved in the context of this set of basic functions. System is not identical with function, but rather the linguistic system is organized around the set of abstract functions of language. I think that is true in the phylogenetic sense in the evolution of language; I am sure it is true in the ontogenetic sense, in the development of language by a child. In other words, the nature of the linguistic system is such that it has to be explained in functional terms. The other sense of structure is the stricter, Firthian sense, where structure is the abstract category for the representation of syntagmatic relations in language. Here I would say that function and structure are also different concepts, and in order to relate them we have to think of function in its other sense of structural functions or roles, like Agent, Actor, Subject, Theme and the rest. A linguistic structure is then a configuration of functions. But this is function in a different sense, though the two are ultimately related. Isn't it the case that you use an extrinsic definition offunction? There is also another definition in the Hjelmslevian sense where function is nothing else than intersystematic relationship. Your definition is an extrinsic definition of function. - Yes; in the first sense I am defining function extrinsically. I am not using the term in its technical Hjelmslevian sense. But I think there is an important connection between this extrinsic sense and the second sense I referred to just now, function used in the meaning of 'grammatical functions' as distinct from 'grammatical classes or categories'. That notion of function refers to an element of structure considered as a role in the total structural configuration. There is a relationship between this meaning of function and the extrinsic sense in which I am using the term: the grammatical functions, in the sense of roles, are derivable from the extrinsic functions of language. There is determination there. The category of function is a very classic one in linguistic theory and has been used since Saussure and Hjelmslev. I assume that the Prague school was inspired and fascinated

M. A. K. HALLIDAY

95

by Biihler's scheme of the different functions of language.8 Do you believe that the Biihler scheme is still valuable, or that Biihler's definition of the expressive, cognitive and referential functions of language is still valid? - I think to a certain extent it is; but remember that Biihler is not attempting to explain the nature of the linguistic system in functional terms. He is using language to investigate something else. His interest is if you like psycholinguistic; and one might compare for example Malinowski's functional theory of language, which is also aimed outside language although in another direction, ethnographic or sociolinguistic as it would be called now. I would consider both these views entirely valid in terms of their own purposes, but I would want myself to adopt a somewhat different (though related) system of functions in order to direct it inwards to explain the nature of the linguistic system. The definition is still extrinsic but the purpose is an intrinsic one. I can explain very simply the relation between the functional framework that I use and that of Biihler. My own ideational corresponds very closely to Biihler's representational, except that I want to introduce the further distinction within it between experiential and logical, which corresponds to a fundamental distinction within language itself. My own interpersonal corresponds more or less to the sum of Biihler's conative and expressive, because in the linguistic system these two are not distinguished. Then I need to add a third function, namely the textual function, which you will not find in Malinowski or Biihler or anywhere else, because it is intrinsic to language: it is the function that language has of creating text, of relating itself to the context - to the situation and the preceding text. So we have the observer function, the intruder function, and the relevance function, to use another terminological framework that I sometimes find helpful as an explanation. To me the significance of a functional system of this kind is that you can use it to explain the nature of language, because you find that language is in fact structured along these three dimensions. So the system is as it were both extrinsic and intrinsic at the same time. It is designed to explain the internal nature of language in such a way as to relate it to its external environment. Could you give a brief description of what you mean by the 'logical' and'experiential' functions of language ? - Within the ideational function, the lexicogrammatical system embodies a clear distinction between an experiential and a logical component in terms of the types of structure by which these are realized. The experiential function, as the name implies, is the 'content' function of language; it is language as the expression of the processes and other phenomena of the external world, including the world of the speaker's own consciousness, the world of thoughts, feelings, and so on. The logical component is distinguished in the linguistic system by the fact that it is expressed through recursive structures whereas all the other functions are expressed through non-recursive structures. In other words, the logical component is that which is represented in the linguis8 Josef Vachek, The Linguistic School of Prague: An Introduction to its Theory and Practice (Bloomington, Indiana, 1966).

96

M. A. K. HALLIDAY

tic system in the form of parataxis and hypotaxis, including such relations as coordination, apposition, condition, reported speech and others. These are the relations which constitute the logic of natural language; including those which derive from the nature of language itself - reported speech is obviously one example of this, and another is apposition, the 'namely' relation. I think it is necessary to distinguish the logical from the experiential, partly because logical meanings are clearly distinct in their realization, having exclusively this linear recursive mode of expression, and partly because one can show that the logical element in the linguistic system, while it is ideational in origin, in that it derives from the speaker's experience of the external world, once it is built into language becomes neutral with respect to the other functions, such that all structures whatever their functional origin can have built into them inner structures of a logical kind. Is the 'ideational' function identical with the 'referential' function of language ? - Weli, I think it includes the referential function, but it is wider. It depends how widely one is is using the term 'referential'. It is certainly wider than the strict definition of referential, but it might be considered as equivalent to referential in the sense in which Hymes uses the term, provided one points out that it has these two sub-components of experiential and logical - I am not sure where Hymes would put the logical element in the linguistic system. Hymes has a basic distinction between referential and socioexpressive; as I understand it, this would correspond pretty closely, his referential to my ideational, noting this query about the logical, and his socioexpressive to my interpersonal. 9 Is it possible in your linguistic theory to elaborate a hierarchy of functions, or is it sufficient to make up the taxonomy of functions ? - Yes, the latter, i would not like to impose a hierarchy of functions, because I believe that there can be hierarchy ad hoc only for the purpose of given investigations. It is noticeable that those whose orientation is primarily psycholinguistic tend to give priority to the ideational function, whereas for those whose orientation is primarily sociolinguistic the priorities are at least equal and possibly the other way - priority might be given to the interpersonal function. This could be reflected in the direction of derivation. If let us say one was working with a functionally based generative semantics, it might well be that for sociological, or rather 'inter-organism', purposes one's generative component would be the interpersonal function, whereas for a more psychologically oriented, 'intra-organism' semantics the generative component would be, as it usually is in generative semantics, the ideational one. I believe that this question of hierarchy of functions is very important in linguistic discussion nowadays. I think for example of the Chomskyan sophistication of the 9 Dell H. Hymes, "Linguistic Theory and the Functions of Speech", in International Days of Sociolinguistics (Rome, 1969).

M. A. K . HALL1DAY

97

expressive function of language. Chomsky defines language as expression of thought and he wouldn't like to see stressed the more communicative features of the semantic structure of language. What do you think of the stress on the expressive function of language ? - I find it unhelpful to isolate any one function as fundamental. I have very much a goal-oriented view, and I can see that for certain types of inquiry it may be useful to single out one function rather than another as prior of fundamental, but I don't find this useful myself. It seems to me important to give equal status in the linguistic system to all functions. And I would point out that our traditional approach to grammar is not nearly as one-sidedly oriented towards the ideational function as sometimes seems to be assumed. For instance, the whole of the mood system in grammar, the distinction between indicative and imperative and, within indicative, between declarative and interrogative - this whole area of grammar has nothing whatever to do with the ideational component. It is not referential at all; it is purely interpersonal, concerned with the social-interactional function of language. It is the speaker taking on a certain role in the speech situation. This has been built into our interpretation of grammar, and I see no reason for departing from this and treating the social meaning of language as some kind of optional extra.

It is very peripheral? - I don't think it is peripheral at all. I don't think we can talk about the functions in these terms of 'central' and 'peripheral'. If you want a model of the production of discourse, I would say that meanings in all functions are generated simultaneously and mapped on to one another; not that we first of all decide on a content and then run it through an interaction machine to decide whether to make it a statement or a question. (I avoid using the term 'expressive' in this discussion simply because there is a confusion here between 'expressive' meaning, expression of thought and 'expressive' in the more usual Buhler sense which is non-representational and corresponds to Hymes's use in 'socio-expressive'.)

Can one say that the communicative function is a kind of super-function or macrofunction, and that the other functions that you mentioned are subfunctions of the communicative function ? - Again I would be unhappy with that. I would want to insist - though always pointing out that it is simply for the purposes of the kinds of investigation I personally am interested in - on the ideational and interpersonal having equal status. The textual function can be distinguished from these two in that it is an enabling function which is intrinsic to language; but as between the first two, I can't see either being more allembracing than the other. All three could be called 'metafunctions' - 'meta-' rather than 'macro-', the point being that they are abstract; they represent functions of language as incorporated into the linguistic system. You notice I am hedging slightly

98

M. A. K. HALLIDAY

on your question, because I am not quite sure how to relate these to what you are are calling 'the communicative function'. But that depends on the definition that you give of the nature of language. Do you see language first of all as communication or as an isomorphic system of logical relations ? - Certainly not as an isomorphic system of logical relations. I suppose therefore I see it as communication, though I would rather say that I see language as a meaning potential. It is a form of human semiotic, in fact the main form of human semiotic; and as such I want to characterize it in terms of the part that it plays in the life of social man. Or, what is the same thing in more abstract terms, I see the linguistic system as a component - an essential component - of the social system. I believe that it is necessary to say that the speaker and the hearer have a certain knowledge of the functions of language. Can you specify this ? - I think that is certainly implied by what I say, but I would make no use of that formulation. Why? - Because it is introducing a level of discourse which is unnecessary in this context. It is certainly true that for a speaker and a hearer to interact linguistically they must have this knowledge; but we only know that they have this knowledge because we see them interact. If therefore it is possible to describe the interaction in the form that I mentioned earlier, that is as the actualization of a system of potentials, then it becomes unnecessary to introduce another level, that of knowledge. This would not be true for example in relation to Lamb's work - I mention Lamb because what he does is entirely compatible with my own. We have very much the same premises about language, but we differ precisely in that he is primarily looking at language intra-organistically and I the other way. For Lamb, of course, the whole point is to find out what it is that the speaker has in his head; therefore he is trying to characterize the knowledge that you have j ust mentioned.10 But I am not. I am trying to characterize human interaction, and it is unnecessary to attempt to interpose a component of what the speaker-hearer knows into the total descriptive framework. Is a functional theory of language such as yours a theory of language as 'language system', as the Saussurean 'langue' ? I believe that your theory of language is a step against the very classic dichotomies 0/'langue'/'parole' or 'competence'11'performance' and so on. - Yes. It is true that I find little use for these dichotomies - though I should point out that this thought is far from being original to me. My former teacher, Firth, 10

Sydney M. Lamb, "Linguistic and Cognitive Networks", in Paul Garvin (ed.), Cognition: Multiple View (New York, 1970). Cf. the discussion with Lamb in the present volume.

A

M. A. K. HALLIDAY

99

himself criticized these very cogently in some of his own writings. 11 He said that he found it unnecessary to operate with mind/body, mentalism/mechanism, word/idea and 'such dualisms'. 1 would agree with Firth - again, always saying that it depends on the purpose for which you are looking at language. I mentioned earlier that, for what we are going to call sociolinguistic purposes for the moment, it is necessary to minimize the distinction between what is grammatical and what is acceptable. If I put this another way I think it will clarify the point here. There will always be idealization in any study of language, or indeed in any systematic inquiry. The point is here that we need to reduce the level of idealization, to make it as low as possible, in order that we can understand the processes of interaction, the sort of phenomena studied from an ethnomethodological standpoint by Sachs, Schegloff and others. 12 We have to impose as low a degree of idealization on the facts as is compatible with a systematic inquiry. This means, in other words, that what is grammatical is defined as what is acceptable. There is no distinction between these two, from which it also follows that there is no place for a distinction between competence and performance or between langue and parole, because the only distinction that remains is that between the actual, and the potential of which it is an actualization.

What is the meaning of one of your statements: "In order to understand the nature of language, it is necessary to start from considerations of its use" ? - Oh yes, this is a very closely related point, and comes back to what I was saying earlier. I think that the use of language can be defined in precisely these terms, namely as the actualization of a potential. Now we want to understand language in use. Why? Partly in order to approach this question of how it is that ordinary everyday language transmits the essential patterns of the culture: systems of knowledge, value systems, the social structure and much else besides. How do we try to understand language in use? By looking at what the speaker says against the background of what he might have said but did not, as an actual in the environment of a potential. Hence the environment is defined paradigmatically: using language means making choices in the environment of other choices. I would then take the next step of saying that, when we investigate the nature of the linguistic system by looking at how these choices that the speaker makes are interrelated to each other in the system, we find that this internal structure is in its turn determined by the functions for which language is used - hence the functional components we were talking about. We then have to take one more step and ask how is it that the linguistic system has evolved in this way, since as we have seen the abstract functional components are, although related to, yet different from the set of concrete uses of language that we actually find in given situations. This can best be approached through studies of language development, 11

J. R. Firth, Papers in Linguistics 1934-1951 (London, 1957), especially Chapter 16. Emanuel A. Schegloff, "Notes on a Conversational Practice: Formulating Place", in D. Sudnow (ed.), Studies in Social Interaction (Glencoe, Illinois, 1971). 12

100

M. A. K. HALLIDAY

through the study of how it is that the child learns the linguistic system. I think when we look at that from a functional point of view, we find some kind of an answer to the question how it is that the linguistic system evolved in contexts of use. What to do with another classic dichotomy, that of ''synchronic' versus 'diachronic' ? - Well, these are different perspectives. I think it would be foolish to deny that this is a real difference. For any system one may look either at its internal structure or at the processes by which it evolved and reached that structure. But I personally am very much in sympathy with the trend which puts these two perspectives closely together, in the sense that either can be used to illuminate the other. I would not like to accept the very rigid separation which some linguists at one time tried to maintain. But - if I may put in parenthesis - one must have a historical perspective on the subject. It is very easy to criticize one's predecessors for having drawn artificial boundaries ; but if you look at the development of knowledge, knowledge usually advances by the drawing of artificial boundaries in order that one can isolate a certain area for study. It is the next generation which sees that the boundaries were artificial and throws them out. Fine; but they would never have got to that stage if their predecessors had not drawn the boundaries in the first place. Is the study of the acquisition of language in the child not a kind of diachronic linguistics ? - Yes, I think in a sense it is, though I think one has got to be careful here. I have been interested in language development in the child from a functional point of view, and I think that one gets a great deal of insight here into the nature of the linguistic system and how it may have evolved. But 1 think one has to be careful and say simply how it "may" have evolved. We cannot know for certain that ontogeny reflects phylogeny. All we can say is that when we examine how a child learns the linguistic system from the functional standpoint, we get a picture which could be a picture of how human language evolved. One very interesting thing that happens, or at least did in the case of the child I was studying, is that you first of all find him creating his own language on what is presumably the phylogenetic model. Then there comes a very sudden discontinuity - at least a discontinuity in the expression, and, more important, in the nature of the system itself - when the child as it were shrugs his shoulders and says, look, this is just too much work creating the whole of human language again from the start; why don't I settle for the readymade language that I hear around me? And he moves into the adult system. Another, the last classic dichotomy that we have to mention here, is the very important distinction between language as system and language as process. This was a very important operational distinction in structural linguistics: language as system, as paradigm, and language as syntagmatic order or discourse. Can you do something with this distinction ? - I think you can. Incidentally there is an interesting contrast here between Firth

M. A. K. HALLIDAY

101

and Hjelmslev. Both of them of course admitted the basic distinction between syntagmatic and paradigmatic relations as formulated (in partly different terms) by Saussure. They then diverged, in that for Firth there was no difference in the level of abstraction between the two; syntagmatic relations were just as abstract as paradigmatic ones. Indeed, in Firth's system the more abstract relations tended to be handled syntagmatically, since Firth was interested not in the potential but in the typical actual, which is rather a different thing. 13 Hjelmslev on the other hand made a very clear distinction between the abstract system, with paradigmatic relations as the underlying ones, and the text as process. Now, I find myself here perhaps somewhat nearer to Hjelmslev than to Firth. The distinction that in practice I find it necessary to operate with - if I may come back again to these terms - is that between the system as a potential, and its actualization. But in considering the system as a potential, I personally find it useful to characterize this entirely in paradigmatic terms, and to regard structure, the structure that underlies the process, as derived from this. Is that not a kind of reduction of the specific discursive dimension in human language ? - No, I don't think it is. Can I try to answer that in a slightly roundabout way, by talking about the nature of text? There is one view of text as a kind of supersentence. This is the notion that text is to be defined by size; a text is something which is bigger than a sentence. I find that rather unhelpful. To me the concept of text is to be defined by level of abstraction, not by size. In other words, text is to semantics what sentence is to grammar. A sentence is to be defined as a fundamental unit of grammar, and we don't define the sentence as a kind of superphoneme. Now, in the same way, the text is to be defined as a fundamental unit of semantics, and we don't define it as a kind of supersentence. It's exactly the same point: you can't define the sentence as something big on the phonological level, because it isn't abstracted at that level at all; equally, you can't define the text as something big on the lexicogrammatical level, because it isn't abstracted on that level - it's on the semantic level. For any level of linguistic structure, semantic, lexicogrammatical or phonological, there will be certain elements and units which can be isolated from the stream, from the process if you like, and which must be isolated from the process if we are to link them to the linguistic system. So it seems to me there are two stages here. One is the syntagmatic relations within the linguistic units themselves of whatever level, which are part of the linguistic system but derivative from the paradigmatic relations within the system. The second is the discourse relations within the text, which include among other things the options that the speaker selects in the light of earlier stages in the same process; in other words, in the light of what has been said before. A very important point seems to me the question of the variation of linguistic units. We have seen in the history of modern linguistics that for instance the morpheme was 13

F. R. Palmer (ed.), Selected Papers of J. R. Firth 1952-59 (London, 1968), especially Chapter 11.

102

M. A. K. HALLIDAY

the privileged unit in structuralism and the sentence in transformational grammar. Do you believe in a privileged position of one of the units ? - I don't believe in the privileged position of any unit at any level. It seems to me that for each of the levels of the linguistic system there is a hierarchy of units. I am not convinced that these units are universals; there may be some universals among them, like syllable perhaps in phonology, sentence in grammar. But just as in phonology one finds linguistic structure embodied in a hierarchy of units - for one language it might be tone-group, stress-group, syllable, phoneme; for another language something different - similarly in the grammar one finds a hierarchy something like sentence, clause, phrase, word, morpheme. Now I wouldn't pick out any of these as having priority. There may be certain properties which are specific to one or another unit, and it may be that some of them are universals, although I'm not quite sure what that would mean since they're really too concrete to be universals. Equally on the semantic level there may be a hierarchy of units; I suppose there is, in some form or other, but here we are at such an abstract level that at the moment we can only really handle the general concept of text. No, we can go further than that; we can recognize 'discourse units' such as episode, narrative, exchange and so on, which I would locate on the semantic level, but these tend to be specific to particular genres or situation types - I'm thinking for example of Sinclair's work on classroom language.14 Is macro-semantics only a part of sociosemantics or is it the whole of sociosemantics ? By macro-semantics, I mean the semantics of units larger than the sentence. - I think it is only a part of it. I am a little worried about the notion of 'macro' because I don't feel that there is any special position to be accorded in semantics to units larger than the sentence. But I would say very definitely that what we are calling 'sociosemantics' involves meaning that is expressed in units smaller than the sentence just as much as that which is expressed in larger units. Note that the sentence in my opinion is not a semantic unit, anyway. What we discussed up to now was very theoretical. It is perhaps a good idea to talk now about your more descriptive work in English. And I would like to ask you a few things about your study of transitivity, as it is a very good example of your functional analysis of some language phenomena,15 In what sense has your study on transitivity been a concretion of your options and opinions about language ? - In this sense, I think: that one can impose a sort of functional grid on the lexicogrammatical system of any language, by which I mean that one can assign any part of that system to one or another functional point of origin. Now transitivity can be defined as the experiential element in the grammar of the clause. Your lexicogramma14 J. McH. Sinclair, I. J. Forsyth, R. M. Coultherd and M. Ashby, The English Used by Teachers and Pupils (University of Birmingham, 1972). 15 M. A. K. Halliday, "Language Structure and Language Function", in John Lyons (ed.), New Horizons in Linguistics (Harmondsworth, 1970).

M. A. K. HALLIDAY

103

tical system can be classified according to a function-rank matrix. So, for the rank of the clause, the experiential function is expressed through transitivity, the interpersonal function is expressed through mood, the textual function is expressed through theme. So transitivity would be an example of what you referred to because it represents a functionally defined component in the grammatical system of the language. Now what does transitivity mean? I understand it to mean the grammar of processes; that is to say, the set of options whereby the speaker expresses the whole range of types of process that are recognized in the semantic system of the language - the process type itself, in English material process, mental process, verbal process, relational process and their subcategories; and the participant roles that enter into these processes. You study transitivy as that part of the grammar that is concerned with the expression of processes as integrated phenomena - and the study of transitivity would lie within the experiential component in the grammar of the clause ? Can you possibly explain this a bit? - Yes, perhaps I could illustrate it. I think that in English there are in the system essentially four main types of process, those that I just named: material processes, mental processes, verbal processes and relational processes, and every clause in the language is the expression of one or other of these four. These four types are characterized by different semantic potentials, for instance different sets of options in voice (systems of active, passive, middle and so on). And different configurations of participant roles are associated with each. Let's take an example of a mental process clause. The main sub-categories of mental process are cognition (thinking, etc.), perception (seeing, etc.) and reaction (liking, etc.) John heard the music and John enjoyed the music are both mental process clauses, John enjoyed the music being reaction and John heard the music perception. These are characterized as configurations of three elements: the Process, here represented by enjoy or hear\ the Phenomenon, here music; and the element affected by the process, which must be human (or quasi-human) incidentally, which we may call the Cognizant. So the 'mental process' is one type of process recognized by the English language, and it is characterized by having a structure in terms of Process, Cognizant and Phenomenon - of course there may be other, circumstantial elements as well. This is different from the structure of something like the boys were throwing stones, which is a material process, not a mental process, and which is a different configuration of participant roles. What are in your example now the lexicogrammatical effects of this semantic structure ? - Well, let's characterize one of these clauses in systemic terms, i.e., in terms of the options that it represents. For instance, John enjoyed the music is partially described as mental process, reaction, middle voice, plus domain. (The last means that the phenomenon is specified.) Each feature that we enumerate in such a list - mental

104

M. A. K. HALLIDAY

process, reaction, and so on - makes its contribution to the lexicogrammatical structure. So this systemic representation as a whole is then represented in the form of a configuration of structural functions, or structural roles, in this case those that I have mentioned: Process, Cognizant, Phenomenon. The way in which we derive these is that each of the features in the systemic network has associated with it a particular realization statement, in terms such as: If the clause has the feature 'mental process', insert the function Cognizant and sub-categorize the Process in a certain way. If the clause has the feature 'middle', then the mental process must be one of the type represented by the verb enjoy, as distinct from a non-middle mental process clause, which is represented for instance by please as in the music pleased John. So that each of the features that figures in the description of the clause in systemic terms makes its contribution to the lexicogrammatical structure, either by inserting an element, by ordering the elements, by inserting a lexical item or a member of a set of lexical items, or something or that sort. In your semantic representation of the transitivity phenomena you don't speak of participants but of participant roles. Is this because the semantic representation must be very abstract? - Yes, it is. Let us consider again an example such as John enjoyed the music. We may describe John in terms such as Cognizant, but in fact John will be the expression of other roles in other structural configurations at the same time. That is to say, not only will he be Cognizant in a mental process clause, but he may also be Subject in a clause of a certain mood, and Theme in a clause of a certain theme-rheme structure, so that any element in the structure is in fact a complex of roles. So although we may talk about participants in the special case of transitivity structures, it is more accurate to say that what enters into the configuration in question is not a participant as such but a participant role. Do you need in your approach to transitivity phenomena the presence of semantic features derived from contextuality ? - When you say 'contextuality', do you mean elements which are circumstantial to the process like place, manner, and so on, or do you mean features of the situational context of the utterance? Of the process. - Yes; the elements of transitivity structure include not only the process itself and the participant roles like actor, goal, agent, cognizant, phenomenon and so on but also circumstantial elements of extent, location, cause and the like. It is possible to specify the total list. And the verbal context, must it beformalized in the semantic representation ? - The verbal context, in fact the whole context of the utterance, situational as well

M. A. K. HALLIDAY

105

as verbal, is not represented in transitivity, or indeed in the experiential component at all, but in the textual component. In other words, it enters in through its part in determining the structure that is derived from the textual functions, and that is embodied in structural configurations such as those of Theme and Rheme, and Given and New. These are not part of transitivity; they are on a different structural dimension. Is there any relationship between the experiential and the textual function of language ? - There is a relationship, in the sense that they are both incorporated in the linguistic system and they are both represented simultaneously in the linguistic structure. But I don't think there is any special relationship. There's not much mutual determination between them. There is a little: I could give you one example where there is determination between them. In English the system of voice: middle or non-middle, and if non-middle then active or passive, is the expression of meanings derived from the textual function. But the potential of any given clause type for options in voice is determined by the transitivity features. In other words, transitivity determines what are the possibilities in terms of active/passive, etc., and the textual function determines which the speaker actually selects. For example, in certain types of material process clause you know that there is a system of five voices: middle, active with goal, active without goal, passive with agent and passive without agent. This is a fact of transitivity, deriving from the experiential component. But the difference in meaning among these five resides not in transitivity but in theme and information, which derive from the textual component, so that the selection of, say, 'passive without agent' represents not a more specific process type but a specific message type. So that is the kind of relationship that you get between the two. Another example of function analysis elaborated by you is about nominality. What is the importance of this phenomenon for your functional conception of language ? - Nominality is another excellent example, because it shows how each of the functions is represented, or rather how meanings derived from each of the functions are represented, in the total structure. If you consider the difference between John enjoyed the music and what John enjoyed was the music, where you have used the resources of nominality, in this case nominalization, the distinction between these two is derived solely from the textual function; again, it is a distinction relating to the nature of the message. What the speaker wants to make as his theme is different in the two cases: in the case of John enjoyed the music the theme is John, while in the case of what John enjoyed was the music, the theme is John enjoyed. Nominalization in English is a device for giving to any particular clause the desired structure in terms of theme and rheme. That of course is only one part of nominality. The whole phenomenon of being a noun, or behaving like a noun, has implications for all functions. In principle the noun is a naming element, and naming in general derives from the experiential func-

106

M. A. K. HALLIDAY

tion, although not entirely, because naming of course embodies connotation as well as denotation: names have an interpersonal component as well as an ideational one, so that in order to derive a particular noun, in the sense of a word of the noun class, we may very well have to go to an origin which is already functionally complex in that it is both ideational and interpersonal. Then, thirdly, there is nominality as a classifying device, which is logical in function. When I build up compound nouns and noun phrases, say I start with station and then I add railway station and then I add suburban railway station and so on, what I am doing every time is introducing a taxonomy, a classification system whose content is experiential but whose form is logical. Hence the structure of these compound noun phrases has a recursive component in it, very well brought out in English by these long strings which are linear recursions. This expresses the fact that the classificatory function of the elements is a logical one, that of forming a taxonomy. But the content of these elements, the meanings 'suburban' as opposed to 'in a city', 'railway' as opposed to 'bus' and so on, are of course experiential. So the noun phrase is a very good example of a complex structure which derives from different functional origins. Do you have any idea whether those structures also exist in other languages ? - Oh, yes, there is no doubt they do. I would certainly claim that the distinction that I am making between the functions is a universal one; and I would also suggest that the structural uniqueness of the logical component that I mentioned earlier, namely the fact that it is represented by means of recursive structures, is at least a candidate for universality. I don't know, but 1 think it's a fair guess. Now of course the structures in question will not necessarily have the same form in other languages as they do in English. But I am fairly confident that there is a nominal taxonomy in all languages, and that the expression of this nominal taxonomy includes some recursion. The recursive structure embodies the fact that the taxonomy is open-ended: you could go on sub-classifying. That is a very formal universal? - Yes. You are one of the few linguists who have elaborated a coherent theory on intonation. 16 It is very difficult to elaborate a theory of intonation within structuralism and transformational grammar. Can you give briefly some characteristics of your conception of linguistic intonation ? - Yes. It's very important to remember that distinctions of intonation have quite different roles in different languages. We know for instance the traditional distinction between tone-languages and intonation-languages; but there's much more to it than that. No doubt all languages use the resource of intonation, but they use it in very 16

M. A. K. Halliday, Intonation and Grammar in British English (The Hague, 1967).

M. A. K. HALLIDAY

107

different ways. I have examined mainly English intonation, and the point that interests me about it is that it is very clear what the role of intonation in English is in functional terms. We can distinguish here between the two main components of intonation, tone or pitch movement on the one hand and tonicity, that is to say the location of tonic stress, on the other hand. Now these are clearly distinct in function in English; there are fuzzy edges, as there always are in language, but in general tone (pitch movement) has an interpersonal function in English - it expresses meanings derived from the interpersonal function, mainly sub-systems of mood - whereas tonicity (the location of tonic stress) has a textual function, that is to say, it expresses meanings derived from the textual component, and more specifically it expresses the information structure. The information structure is that system of the textual component which is concerned with the speaker's distribution of the message into two elements : what he is treating as information that is recoverable to the hearer, which we call the Given part, and what he is treating as information that is not recoverable to the hearer, which is the New part. We can't make a rule to say that what carries the tonic stress is new and everything else is given, because it is not as simple as that. Nevertheless it is quite clear in English how the tonic stress pattern expresses the information structure. The rules are fairly complex, but they're there. I think myself it is quite impossible to give an adequate account of a language without taking cognizance of those meanings which are expressed through intonation. But we have to find out exactly what these are in each case. There may be a general tendency in intonation languages for tone to express textual meanings, but I wouldn't go further than that. What is the place of the focus in this whole system of intonation rules ? - Well, the focus, as I understand it, is simply an element in the information structure. This is what I have been referring to just now. If we say John enjoyed the music with the tonic stress on music, then the focus of information is on the music and the meaning is that either the music or enjoyed the music or the whole of John enjoyed the music may be new information. In other words, at least the final element is new, and, working backwards, any or all of the rest of the clause may be new. If on the other hand we say John enjoyed the music with the focus of information on enjoyed, then the message is that the music is given, that is to say the speaker is treating it as recoverable to the hearer; John may or may not be given, we can't tell (actually we can tell, by a more subtle distinction within the rhythm); but we know that enjoy is new. Now the focus of information is realized by tonic stress; broadly speaking, it is that element in the information unit that the speaker is marking out as being the termination of what is not recoverable to the speaker. Whether or not what precedes it is being specified as not recoverable to the hearer depends on other factors. Anything that follows it is always being specified as recoverable. So, as I said, the rules are fairly complex but they're clear; and we can be quite specific about what 'focus of information' means. Now this is all about English. I do not believe that this is a universal pattern. I am

108

M. A. K. HALLIDAY

quite sure that all languages have textual structures (that is, structures expressing textual meanings, just as they all have ideational structures and interpersonal structures); and probably in some way or other they all make a distinction between the given and the new. But we cannot just assume that there will be a focus of information expressed by intonation in the same way as in English, and having the same meaning as it has in English, in all other languages. Is it possible to identify the focus with some very specific syntactic component of the sentence, or in more logical terms, as some generative semanticists do, with the argument of the sentence ? - It is a syntactic element, though not of the sentence. The information unit - which is a constituent in textual structure, not in ideational structure like the sentence and the clause - is structured in terms of two elements which we are calling Given and New. The term 'focus' refers to the location of the New element; in English this means specifically its terminal point, since the focus is realized phonologically as tonic prominence and the prominence falls on the item that occurs in final position within the New. Given-New is a syntactic structure just like Actor-Process-Goal, SubjectPredicator-Complement, Modifier-Head and so on. This is what Lakoff apparently failed to understand when he pointed out - claiming to refute me - that the focal element is often one which has been mentioned before. 1 7 1 had pointed this out quite explicitly myself, stating that one of the meanings of 'information being treated by the speaker as non-recoverable to the hearer' (this being the semantic category that is realized grammatically by the New) is precisely something that has been mentioned before but is contrastive in the context, for example Have you seen John and Mary anywhere? Vve seen John, with focus on John implying 'but not Mary'. This is what explains how you can have the focus on anaphoric items, which would otherwise be inexplicable. So the focus is a grammatical concept, not a logical one; and it derives from the textual component of the linguistic system. One way of studying linguistics could be to see how people learn to mean. "Learning how to mean" is exactly the title of one of your papers. It is a study of the child's language, and language development and acquisition, topics which are very much at stake nowadays. Can you tell me what this study of learning how to mean has to offer to general linguistics ? - I see this again from a functional perspective. There has been a great deal of study of language development in the past ten or fifteen years, but mainly on the acquisition of syntax seen from a psycholinguistic point of view - which is complementary, again, to a 'sociosemantic' perspective. To me there seem to be two aspects to be stressed here. One is: what is the ontogenesis of the system, in the initial stage before the child 17

George Lakoff, "Presupposition and Relative Wellformedness", in Steinberg and Jakobovits (eds.), Semantics: An Interdisciplinary Reader in Philosophy, Linguistics and Psychology (Cambridge, 1971).

M. A. K. HALLIDAY

109

takes over the mother tongue ? The other is: what are the strategies through which a child moves into the mother tongue and becomes linguistically adult? I would simply make two points here. 1 think by studying child language you get a very good insight into function and use (which for the very young child, as I said, are synonymous). We can postulate a very small set of uses, or functions, for which the child first creates a semiotic system. I have tried this out in relation to one subject, and you can see the child creating a meaning potential from his own vocal resources in which the meanings relate quite specifically to a certain set of functions which we presume to be general to all cultures. He learns for instance that language can be used in a regulatory function, to get people to do what he wants; and within that function he learns to express a small number of meanings, building up a system of content/expression pairs where the expression is derived entirely from his own resources. He creates a language, in functional terms. Then at a certain point he gives up this trail. In the case that I studied, the child dropped the language-creating process at the stage where he had a potential of about four or five functions with some fifty meanings altogether, roughly fifty elements in the system. Anyway the stage comes when he switches and starts taking over the adult system. So there is a discontinuity in the expression; but there is no discontinuity in the content, because to start with he maps the expressions of the adult system on to his own functional framework. He does this, it seems to me, by a gradual separation of the two notions of function and use; on the one hand the original uses of language go on expanding, as he goes on using language in new and other ways, but at the same time he builds in this functional framework into the linguistic system itself. I have tried to describe how he does this; basically I think he does it through internalizing a fundamental distinction between pragmatic uses of language, those which demand a response, and represent a way of participating in the situation, and what I call 'mathematic' uses of language, those which do not demand a response but represent rather a way of observing and of learning as one observes. Now these two come out of his original set of very concrete functions, but they turn into functional components of the linguistic system itself, the interpersonal and the ideational that we were talking about earlier. Are the causes for this change environmental ? - I assume them to be environmental with a biological foundation. The biological conditions must be met, the level of maturation must have been reached. Given that level of maturation, then I would look for environmental causes in the social system. I don't want to get into arguments about the psycholinguistic mechanisms involved, because I don't think this assumes any particular psycholinguistic perspective. Is your point of view not too behavioristic here ? - No, I would say that it is emphatically not behavioristic. It has always seemed to me, and again here I am simply following Firth, that behaviorist models will not account for linguistic interaction or for language development. There is a very curious

110

M. A. K. HALLIDAY

notion that if you are assigning a significant role to the cultural environment in language learning you are a behaviorist. There is no logical connection here at all. We should perhaps demolish the fallacy of the unstructured input. There has been a myth around over the past few years that the child must have a specific innate languagelearning capacity, a built-in grammar, because the data to which he is exposed are insufficient to account for the result. Now that is not true. The language around him is fantastically rich and highly structured; Labov has said this and he is quite right. 18 It is quite unnecessary to postulate a learning device having these highly specific linguistic properties. That doesn't mean it is wrong; it means it isn't necessary. I want to distinguish very sharply here between the particular psychological model which you use and the functional conditions on language learning. These do not presuppose each other in any way. What I am doing is simply studying the child's language development in an interactional perspective, and this has got nothing whatever to do with behaviorist theories of psychology. How does this viewpoint on language development in the child lead us into a sociosemantic approach to language ? - First, it points up the fact that a child who is learning language is learning 'how to mean'; that is, he is developing a semantic potential, in respect of a set of functions in language that are in the last resort social functions. They represent modes of interaction between the child and others. So the child learns how to interact linguistically; and language becomes for him a primary channel of socialization, because these functions are defined by social contexts, in Bernstein's sense as I mentioned earlier. The child's semantics therefore is functionally specific; what he is developing is a 'social semantics' in the sense that it is a meaning potential related to a particular set of primary social functions. And second - though it's a closely related point it is above all through a developmental approach that we can make concrete the notion of language as part of the social semiotic: the concept of the culture as a system of meaning, with language as one of its realizations. Could you explain more concretely your hypothesis about the functional origin of language ? What does the system offunctions look like in this first phase of the development of language in the child? - In this first phase I suggested that the child learns: the instrumental function, which is the 'I want' function of language, language used to satisfy a material need; the regulatory function, which is the 'do as I tell you' function, language used to order people about; the interactional function, 'me and you', which is language used to interact with other people; the personal function, 'here I come', which is language used as the expression of the child's own uniqueness; the heuristic function comes a little while behind, and is language as a means of exploring the environment, the 'tell me 1»

Cf. note 3.

M. A. K. HALL1DAY

111

why' function of language; and finally the imaginative function, 'let's pretend', which is really language for the creation of an environment of one's own. In the case of the particular subject I worked with, these six functions had all appeared in his proto-language: he had developed a semiotic system in respect of all these six functions without making any demand on the adult language at all. The elements of the system were entirely his own invention. Then there came a point at which he switched, and as it were gave up his own special system in favor of that of English. Simultaneously with this switch, he generalized out of his original range of functions this very general functional opposition between what I referred to as the pragmatic and the mathetic. With this child the distinction was very clear, because he developed an interesting strategy of his own, which was absolutely consistent: he used a rising intonation for all pragmatic utterances, and a falling one for all mathetic ones. So he knew exactly what he was doing: either he was using language as an intruder, requiring a response ("I want something", "do something", etc.), which he did on a rising tone; or he was using language as an observer, requiring no response (in the meanings of observation, recall or prediction), and with these there was a falling tone. The pragmatic function evolved here clearly out of the instrumental and regulatory uses of language. The mathetic function evolved in a way that was much less clear; it required a lot of time to trace the history of this, but I think it arises out of the personal and heuristic functions. Language is first used to identify the self, in contra-distinction to the environment; it is then used to explore the environment, and by the same token then to explore the self. This child as I say made a beautiful distinction between the rising tone for the pragmatic or 'doing' function and the falling tone for the mathetic or 'learning' function. Next stage, the adult language, unlike the child's proto-language, gives him the possibility of meaning more than one thing at once. There comes the moment when these functions are incorporated into the linguistic system itself, in the highly abstract form of the meta-functions I mentioned earlier: the pragmatic function into the interpersonal function in the linguistic system and the mathetic function into the ideational function in the linguistic system. Whereas, in the first stage, the functions stand in an 'either... or' relationship - the child is using language either to do this or to do thatthe beauty of the adult linguistic system is that he can do more than one thing at once. In fact he must do more than one thing at once, because now, in the adult stage, every time he opens his mouth he is both observer and intruder at the same time. And this is why human language evolved by putting in between the meaning and the sound a formal level of grammatical structure, because it is the grammatical structure which allows the different functions to be mapped on to one another in a sort of polyphony. I use this metaphor because in polyphonic music the different melodies are mapped on to one another so that any particular chord is at one and the same time an element in a number of different melodies. In the same way, in adult language any element in the syntagm - say a word - is at one and the same time filling a role in a number of different structures. Now you can't do this without a grammar. The child's system

112

M. A. K. HALLIDAY

is a two-level system: it has a content and an expression. The adult system is a threelevel system, of content, form and expression. So this functional plurality makes the difference between adult language and child language ? - Yes; this is what I mean by functional plurality - that any utterance in the adult language operates on more than one level of meaning at once. This is the crucial difference between the adult language and the child's language. Everything you have said up till now proves that your view of the scope of linguistics differs widely from the views we are acquainted with in various other trends. This is perhaps a good occasion to turn back to our starting-point. I would like to ask you to redefine what you mean by linguistics and by sociolinguistics, and what you mean by saying that a good linguist has to go outside linguistics. - Well, I hope I didn't quite put it that way, that a good linguist has to go outside linguistics! Let's go back to the observation that there are two main perspectives on language: one is the intra-organism perspective, the other is the inter-organism perspective. In the intra-organism perspective we see language as what goes on in the head; in the inter-organism perspective it is what goes on between people. Now these two perspectives are complementary, and in my opinion linguistics is in the most healthy state when both are taken seriously. The past ten or fifteen years have been characterized by a very large concentration on intra-organism linguistics, largely under the influence of Chomsky and his 'language as knowledge' or psycholinguistic perspective. I am personally glad to see that there is now a return to the inter-organism perspective in which we take account of the fact that people not only speak, but that they speak to each other. This is the fact that interests me. People often ask, must you make a choice whether you are going to study intra- or inter-organism linguistics, can't you just study language? I would say, up to a point you can. If you are studying the inner areas of the linguistic system, linguistic form in Hjelmslev's sense - the phonological and lexicogrammatical systems - you can be neutral up to a point; but the moment you go into semantics, your criteria of idealization depend on your making a selection. You either say with Chomsky that linguistics is a branch of theoretical psychology, or - which is equally valid - that linguistics is a branch of theoretical sociology. For that matter you could say that linguistics is a branch of theoretical esthetics. What are the implications of your view for the problem of language teaching ? - The type of perspective I have on language naturally relates to my own interests. My interests are, primarily, in language and the social system; and then related to this, in the two areas of language and education, and language and literature. All these have something in common. They make it necessary to be interested in what the speaker does; in the text. Now in order to make sense of 'what the speaker does',

M. A. K. HALLIDAY

113

you have to be able to embed this in the context of 'what the speaker can do'. You've got to see the text as an actualized potential ; which means that you have got to study the potential. As regards language teaching - could I rather say 'language in education', because I am not so much concerned with pedagogical linguistics as with educational linguistics, and with the kinds of presupposition that are made about language in the educational system at present? - here again you need a functional perspective. Let me take one example. Consider the question of literacy, teaching reading and writing: what is learning to read and to write? Fundamentally it is an extension of the functional potential of language. Those children who don't learn to read and write, by and large, are children to whom it doesn't make sense; to whom the functional extension that these media provide has not been made clear, or does not match up with their own expectations of what language is for. Hence if the child has not been oriented towards the types of meaning which the teacher sees as those which are proper to the writing system, then the learning of writing and reading would be out of context, because fundamentally, as in the history of the human race, reading and writing are an extension of the functions of language. This is what they must be for the child equally well. Here is just one instance of a perspective on language in the context of the educational system. In stylistics too the emphasis is on the study of the text, and again there is a functional basis to this. We are interested in what a particular writer has written, against the background of what he might have written - including comparatively, against the background of other things he has written, or that other people have written. If we are interested in what it is about the language of a particular work of literature that has its effect on us as readers, then we shall want to look at not simply the effects of linguistic prominence, which by themselves are rather trivial, but the effects of linguistic prominence in respect of those functions of language which are highlighted in the particular work. I am thinking here of Zumthor's point where he has said that the various genres of literature in different epochs are characterized by differences of emphasis on the different functions of language. 19 I think this is very true. It seems to me that you can only understand the linguistic properties of the text in relation to the orientation of the whole of which it is a part to certain patterns of linguistic function. I have tried to illustrate this in my study of the language of Golding's The Inheritors, where it is very clearly the transitivity system that is at work. 20 The whole book is about transitivity, in a certain sense. There is a highlighting of man's interpretation of the processes of the external world; and therefore it is no accident that there is a highlighting in the language, in the grammar, of certain aspects of the transitivity system. This illustrates once again the same perspective on language. A central position is accorded to the study of the text; no sharp separation is made 19

Paul Zumthor, Poétique Médiévale (Paris, 1972). M. A. K. Halliday, "Linguistic Function and Literary Style: A n Inquiry into the Language of William Golding's The Inheritors", in Seymour Chatman (ed.), Literary Style: A Symposium (New York, 1971). 20

114

M. A. K. HALLIDAY

between competence and performance; the text is seen as an actualization of the total potential, in the context of a functional theory for the interpretation of the potential. 1 see this as the thread which links the social, the educational and the literary perspectives on language. As a conclusion of this dialogue, I would like to ask you two kinds of questions. First I would ask you to thematize your scientific genealogy, if I may call it that. Most people consider you as a representative of the school of English linguistics, namely Firth's London school. Do you agree with this ? - I would be quite happy to accept a designation as Firth's pupil. I think he was a great scholar and I have tried to develop some of his ideas. There seems to be a great deal of richness in Firth's work, and much of it still remains to be taken up. But here I have to admit that I find myself having a slightly different interpretation of the recent history of linguistics from that which is perhaps the most generally accepted. It seems to me there are essentially two main traditions in modern linguistics: one represented by Hjelmslev, the Prague school, Firth and the London school, Sydney Lamb and stratificational grammar, and by and large Pike and tagmemics; the other represented by Bloomfield, the structuralist school, and Chomsky; while the later versions of transformation theory, especially generative semantics, Lakoff, McCawley and others, have moved from the latter towards the former. One symptom of this is the distinction between a tristratal view of language, with the key concept of realization, of language as a multiple-coding system, which is characteristic of the Hjelmslevian view, contrasted with a bistratal view of language based on a combinatory or compositional conception rather than a realizational one. In other words, language in the Bloomfieldian concept is interpreted as consisting primarily of two levels related by composition: a grammar and a phonology, with grammatical units composed out of phonological units. The Prague school and Hjelmslev see language as having a grammar, a phonology and a semantics, with a realizational or coding relation among them. Chomsky essentially takes over the Bloomfieldian view. One is not denying Chomsky's greatness and originality if one says that he belongs to the Bloomfieldian tradition in linguistics. So you've got two traditions within linguistics itself: the combinatorial, compositional one which is Bloomfield and Chomsky, structuralism and the early transformation theory, and the realizational one which is Hjelmslev, Firth, Troubetzkoy and others. In the Bloomfieldian tradition, then, the key concept is that of constituency, Chomsky takes this over, he formalizes constituent structure, and then finds it necessary to introduce a 'deep structure' because constituent structure won't handle semantic relations in language. Out of this comes generative semantics, which is again tristratal and basically a realizational concept. Another way of looking at this is from outside linguistics. If we go back to the 'psycho-' versus 'socio-' perspective, fundamentally Chomsky's links are with the philosophical-logical-psychological tradition in the study of language, whereas the other approach is more associated with

M. A. K. HALLIDAY

115

the ethnographie and sociological study of language. Chomsky for the first time shows that it is possible to formalize natural language. This has opened a whole new era in linguistics; but in doing so Chomsky remained within the Bloomfieldian concept of language and largely within the psychological-philosophical universe of discourse. He imposes a very high degree of idealization, operates with a strong boundary between competence and performance, and so on. The other tradition, which is closer to European structuralism, is more concerned with how people behave, and especially with the interpretation of how they behave. My own background is entirely of this kind: an approach to language which is originally ethnographic, Malinowskian in my case; which emphasizes interaction and which operates with a very low degree of idealization - hence much less formalization, and what there is is of a very different kind, best seen in the work of Greimas, with its origins in Lévi-Strauss.21 Now, obviously all this is over-simplified. But it may perhaps give some answer to your question. This sketch seems very interesting, but still I have some difficulties with it. I would like to formulate two of them here. First of all, I thought that Hjelmslev,s linguistics was especially a study of the intrasystematic relations of language, and that is different from the more extrinsic approach to language as in the Prague school and in Martinet. I clearly distinguish between the formalizing structuralism and the substantializing structuralism. These seem to me very opposite directions within structuralism. - I would have said different elements rather than opposite directions. I agree that Hjelmslev is representing the formalizing tradition in structural linguistics; but this is nevertheless within the framework of a basically inter-organism view of language. Hjelmslev is concerned with language as process; process is text, and text is the interorganism instantiation of language. This is why Hjelmslev is so difficult. It looks as though he is formalizing knowledge about language, but he is not. He is formalizing the linguistic system in a context which is really that of text, that of language as process. Lamb in his article on Hjelmslev criticizes him for this. 22 1 think he's right in showing where Hjelmslev doesn't quite bring it off. Don't you think the time has come for a revaluation of the Prague school? What is the importance of the Prague school in linguistics ? - I am delighted that so many linguists have now taken up the work of the Prague school. I have always considered this of fundamental importance; primarily because they were the first linguists to attempt to build functional theories into the linguistic system instead of imposing them from outside - they seek to interpret the linguistic system in functional terms. Secondly, because they have a basic concept of the nature 21 A. J. Greimas, Sémantique Structurale (Paris, 1966). Also "Éléments pour une théorie de l'interprétation du récit mythique", in Roland Barthes (éd.), Communications 8 (Paris, 1966). Cf. the discussion with Greimas in the present volume. 22 Sydney M. Lamb, "Epilegomena to a Theory of Language", Romance Philology 19 (1966).

116

M. A. K. HALLIDAY

of language which is entirely relevant to the sort of problems I am interested in. I would certainly acknowledge the Prague school as a major contributor to modern linguistics. My second difficulty then is your analogy between Bloomfield and Chomsky. I agree that from the point of view of the constituent structure they are alike. But don't you agree that Chomsky''s notion of creativity is absolutely absent in Bloomfieldian linguistics? Bloomfieldian linguistics is taxonomic and inductive whereas Chomsky's notion of creativity contradicts this. - I doubt this. Of course, I agree that Chomsky has opened a new perspective; but I think he did so in a framework that is Bloomfieldian in its essentials, and not merely from the point of view of constituency. I don't really see that Chomsky was saying anything at all new when he was talking about the creativity of language. This is clearly implicit in Bloomfield, and even more so in Hjelmslev, Firth and others. I think you must look at the thing in a historical perspective. In a sense, Bloomfield was focusing concentration on the surface of language because he saw that linguistics had to become a lot more explicit and that was the only way in which it could do so. That is my point; there is no idea in Bloomfield like deep structure. - No, I agree. Deep structure is new in Chomsky; but it is only necessary if you accept the Bloomfieldian position to start with. You don't need 'deep structure' if you are not a 'structuralist'. Chomsky starts with a two-level system: syntax and phonology, no semantics. If you represent language as structure in the Bloomfieldian sense, then you get to the point where it doesn't work. In other words, it is no longer possible to get any 'deeper' than Bloomfield had already got. Harris showed just how far it is possible to get with an essentially combinatorial or compositional view of language; you can get, roughly speaking, as far as the phrase in grammar. Then you stop, because beyond that point language is so complex that it is impossible to construct more abstract units out of combinations of the lower level elements. There is only one thing you can do once you have gone so far: introduce the concept of transformation, that is, invent more abstract structures and rules for transforming the one kind to the other. This is what Harris did and what Chomsky did. But if you haven't started with a combinatorial view, then there is no need to invent deep structures as the generative semanticists have seen - because you aren't assuming that the whole of the linguistic system can be represented as a combinatorial form. The question is whether, if you keep an IC-type constituent structure, with deep structure as well as surface structure and rules mapping the one on to the other, the deep structure then looks like, or leads on to, a convincing representation of semantics. If I understand you well, a very fundamental distinction in the history of linguistics is that between sociologically oriented linguistics andpsychologically oriented linguistics. - Yes I think it is significant, and it ties up with the other distinction between the concept of realization and that of composition.

M. A. K. HALLIDAY

117

Another way of concluding would be by asking you something about the philosophical background of your theory. Two questions here. Could you survey your opinion about the relation between language and experience ? And how is experience an operational category in linguistic theory? - As I see it, the individual's experience is mediated through language but in the context of the social structure. When we say that the linguistic system is part of the social system, this implies that what is transmitted to us through language, when we learn it as children, is in fact the social system; and that our experience is codified in language through the intermediary of the social structure - there is no such thing as experience that is independent of the social structure. So when we transmit experience through language what we are transmitting is the social system. You see I would take a view which is Whorfian as modified by Bernstein. 23 Yes, so there is no direct relationship between the individual and language but only a direct relationship between a social structure and language ? - That is the direct relationship as I see it, yes. I would like to emphasize that I'm not rejecting the creativity of the individual. On the contrary, 1 am fascinated by it, as anyone interested in stylistics must be. But just as an individual's linguistic creativity is defined by linguistic structure, so also at the next level it is defined by the social structure. The individual is a complex of social roles; he derives his individuality from the social structure, and this is what he expresses in language. Some of our most creative acts, linguistically speaking, are repetitions; and let's do away with this notion that nearly every sentence that is uttered is being uttered for the first time - on the contrary, I would say, nearly every sentence that is uttered is not being uttered for the first time. Creativity does not consist in creating new sentences. Creativity consists in new interpretations of existing behavior in existing social contexts; in new semiotic patterns, however realized. I am not accustomed to formulating this and therefore I am saying it rather badly; but 1 think the whole question of the relationship between the individual and language has to be seen as embedded in the social structure. There is confusion here partly because it is assumed that one must posit a direct relation between language and the individual, otherwise one is denying individual creativity. I don't believe this at all. I think the creativity of the individual is a function of the social system. My last question: the key problem for a philosopher of language is the problem of the relation between language and world. I would like to know your opinion here. To take an easy formulation: is language the mirror of the world? - Here we come to universals, don't we? There are fashions in linguistics, as I said before. The recent fashion has been to emphasize the biological nature of the human being, bringing out those respects in which we are all alike. This needs to be comple23

Basil Bernstein, Class, Codes and Control I, Chapter 7.

118

M. A. K. HALLIDAY

mented by reference to the cultural nature of the human being, bringing out those respects in which we may differ. Language expresses both human biology and human culture. It expresses the unity of the human race and it expresses the diversity of human culture. Experience is a product of both, and experience is encoded in language; but it is experience as processed by the culture and by the sub-culture. This is one of the reasons for sub-cultural misunderstandings, failures of communication (including those on a large scale, that we call educational failures) in the community. In a plural society you have sub-cultural differences which lead to different encodings of experience in language. Even when it is the same language, there may be different meaning systems, different orientations in the meaning potential associated with the different sub-cultures. So biology and culture as it were interact in the linguistic system; and since human experience is a product of the two, and is mediated through language, I would agree in seeing a form of dialectic whereby the individual's experience is transmitted to him through language, and then is expressed by him in language, and so transmitted in turn to others. As I said, I would accept Bernstein's modification of Whorf. The weakness in Whorf's hypothesis seems to be that it has no place for the social structure; it is a hypothesis about language, culture, and the individual. I think Bernstein is right in saying that the relation between language and culture is mediated by the social structure. So I would say that in the last analysis language does mirror the world, but it does so only very indirectly, through the mediation of experience which is itself mediated by the social structure. The result is not so much a mirror but - to vary the metaphor - a multiple recoding, an ongoing working interpretation of the world as it impinges on social man. What are the implications of your opinion about this language-world relation for the problematic issue of universals ? - If we start from the distinction Chomsky made between formal and substantive universals - what Firth used to call 'general' categories and 'universal' categories, the latter of which he rejected - then general linguistic theory is a theory of formal universals; there is no dispute about the existence of such universals, however difficult it may be to characterize them at a suitably abstract level. When you come to substantive universals, which are more concrete - sometimes rather too concrete, perhaps, in the form in which they are offered (for example phonological features) the problem seems to be that the relation between these and the formal universals is quite arbitrary. There is no necessary reason why a system having this particular content should take this particular form. So we interpret the formal universals in biological terms, as given by the structure of the human brain. Now what interests me more are universals of human culture. I should like to know, for example, to what extent we find a universal aspect in the functional origins of the child's linguistic system. After all we know all human groups are biologically endowed with the same brain structure; a few lessons in linguistics has always been enough to demonstrate this. But there is no a priori reason why all human cultures should make identical

M. A. K. HALLIDAY

119

demands on language. Clearly there is a point up to which they do: as Lévi-Strauss has put it, for all cultures "the universe is an object of thought at least as much as it is a means of satisfying needs", and this fact is reflected in language.24 No doubt we can go further than this before we reach the point of specificity where differences begin to emerge and we find different cultures using language in different ways, but we do reach it sooner or later. These questions may give us something to which to relate the substantive universals if we find them. 25 Is grammar that which relativizes the universality of behavior and biological structure, or is it the part which makes universal the particular experiences of the social structure ? - I think both can be true. On the one hand, the universals, in one sense of the term, will be found in what Hjelmslev calls the 'purport'; this is simply because we all have the same physiology and we all live on the same planet. These are already relativized in the semantic system, which is the grid imposed on them by the culture, and also simply by the special nature of the language in question; and the grammatical system relativizes them still further. At the same time - and this is where the other kind of universals, the formal universals, come in - the whole of human experience, which is essentially culture-specific in that it is mediated by the social system, is ultimately representable in terms of a linguistic semiotic having certain highly abstract formal properties; and in that sense grammar 'makes universal', by providing a common framework for the interpretation of experience. But this is only at the most general level ; I would not like to argue from the formal universals of language to the nature of experience itself. The fact that at a sufficiently abstract level we can represent the internal organization of all human languages in terms of a particular formal system tells us nothing about the conditions of human life. I would like to ask you, where is your work leading to ? - Since I am probably rather more modest than I may have sounded sometimes when getting excited, I would like to put this in very tentative terms by saying simply what are the fundamental questions that interest me and which I hope linguistics will continue to pursue among other questions that concern it. There are perhaps two fundamental questions lying behind what I do, one intrinsic and one extrinsic. The first one is, why is language as it is ? Mankind could have evolved an enormous number of different semiotic systems ; why did they evolve a system which has these particular properties that language has? I think that the functional approach is one way of gaining some insight into this question; and it will perhaps allow us to return to an interest in the origins and the evolution of language, which has been somewhat out of fashion for the last hundred years. The second question concerns language and the transmission of culture : how is it that the most ordinary, casual, informal, everyday uses of language, without any kind of instruction and without even any kind of 24 25

Claude Lévi-Strauss, The Savage Mind (London, 1966). Joseph H. Greenberg, Essays in Linguistics (Chicago-London, 1963), Chapter 7.

120

M. A. K. HALLIDAY

explicit understanding behind them, so efficiently transmit to the child the fundamental patterns of the culture, systems of knowledge, social structure, value systems and the like ? I think we have very little understanding of this at the moment, and it is that aspect of language and the social system which interests me a great deal. As a final small parenthesis, I would point out that there is one tremendous gap in our understanding of language which is really fundamental here; that is that we have practically n o idea of the nature of children's peer group speech, the kind of language with which young children interact with each other. We've got to find out more about this among other things, of course - before we can begin to answer the second of these questions, and probably the first as well. Center for Advanced Study in the Behavioral Sciences Stanford, Calif., U. S. A. October 9, 1972

PETER HARTMANN

Peter Hartmann, your work focuses on the presuppositions of all scientific research and of linguistics in particular. That is perhaps your originality among contemporary linguists. Could you explain your point of view about those presuppositions ? - I would say the presuppositions of scientific work are a topic of primary importance to be discussed in every discipline. Linguistics, in this sense, should be integrated into a general framework which should define its goals. I see three subdivisions: science in its social context; science and its social function, where society sets the tasks science has to pursue since society is the environment of science and supports and finances it; and linguistics in this function, that is considering which function linguistics can have in society as it was described above. Perhaps we could analyze each topic separately. It is illusory, anachronistic and irresponsible not to relate the research conducted in one's own discipline to the surrounding context: beyond this context, the research in question would in practice not be possible, and researchers and scientists would not be able to work or even to exist. It is not necessary to talk about that at length. Acquiring one's competence, being trained in science, in scientific skills, is so dependent on society that one should demand, or at least recognize, that science should react to the social environment. By 'react' I would mean, quite specifically, executing a task that would benefit the social environment; in other words, society should be able to put specific demands to a discipline, which would then propose solutions in accordance with the advances made in that discipline. Are those demands of a political nature ? - In principle they are, of course, political as well, but I wanted to be even more abstract and general. The demands do not mean that a specific science should be limited to the wishes or questions of others. Demands concern the expectations that society may have with regard to a discipline which it establishes, supports and, as already said, also finances. Society, specifically a nation or a state, elaborates policies that can bear upon scientific questions as soon as they concern universities. In this context, it does not seem futile to me to stress political questions. It seems evident to me that a policy taking universities into account will open new horizons for the university itself and for each discipline or science, and thus for linguistics as well this discipline alone would never have reached them. Should one require from sciences

122

PETER HARTMANN

that they fulfil demands transcending science? In all cases, yes, because a specialist, an expert - a linguist as well - who is engulfed in scientific thought cannot be expected to transcend the views that are immanent to his discipline in order to look at himself from a distance, to judge or to criticize himself. What is the social function of science? - ' T h e practical consequences of science within society' is an expression that can help explain this. 'Practical consequences' means that one can and must expect from science that its achievements must not be kept for itself Of course, there are achievements that are exclusively or primarily used in science, that are steps in the construction of science as such, but it must not be limited to this. The question of science and its practical application, of theory and practice, of linguistics and its possible fields of application, is a very important, decisive, and, for science itself, a useful question. The social function, defined as the practical consequences for society, would be the answer to what has been said under the first point: science should only be seen in a social context. The repayment for what society has contributed to science is what science contributes to society. The social function of science could thus be defined as its input to the innovation or its value to the emancipation of society. What, in this context, is the peculiar role or function of linguistics ? - Linguistics is a rather specialized discipline since it is a discipline that is taught and learned only at universities. Its practical consequences, however, bear upon the whole process in the educational system. Elementary school creates the native speaker's competence, and the problems it raises are taken seriously today. To a very large extent, training in elementary school and high school depends on language. No science can be transmitted or taught if it is not based on dialogue: this raises the question of the transmission of knowledge, that is, the transmission of noetic or conceptual contents. Other social functions would include contributions of other branches. It is sufficient to look at all of those borderline disciplines which have an integrated training program today: pedagogy, psychology, philosophy, theology, law, physics, and probably others. There are also disciplines where there is no autonomous curriculum, but like communication teaching where the linguist will not be the only one to make a contribution. Is it not a consequence of your position that linguistics has to become a sociolinguistics ? - It is a danger that should be avoided. I call it a danger because typical and genuine linguistic viewpoints would then be subordinated to others; the independence of the linguistic viewpoint is not yet dead in my mind. It is possible, however, to look at linguistic objects, linguistic realizations, etc. as lying within the domain of the social sciences. In a general theory of behavior, performed by individuals or groups, language would then be one peculiar type of behavior, that is, the use of signs in a communicative process. I would not accept a direct link with sociology. Recent trends

PETER HARTMANN

123

show that sociolinguistics and a sociology of language cannot be created by having linguists concern themselves with sociological categories. The opposite is equally impossible. An autonomous theory has to be constructed with its own tendencies, theorems, and structure - a theory that contains both projects and arrives at assertions that are simultaneously true for both disciplines. This requires a new kind of theory. Presently, sociolinguistic assertions, if they are interesting and powerful, are quite limited. That is also true of linguistic assertions that pretend to be relevant to sociology. Authors like Bernstein and others, 1 who are leaders in the field of sociolinguistics, always seem to start from one point of view and explain only one side of the problem. If sociolinguistics means creating a new scientific direction and theory, such that linguistic behaviors and specific practical phenomena may be analyzed simultaneously, then I share your opinion without reservation. It would be one of the first interdisciplinary projects that should be realized. Is it true that the greatest difficulty is the correct formulation of the dialectic relation between immanence and transcendence of linguistics and of science ? - Maybe the difficulty exists only as far as the relation itself between immanence and transcendence has not been thematized. The most necessary book that ought to be written is a systematics of linguistics, i.e., a problem oriented and at the same time operational analysis of the scientific and linguistic processes of acquiring knowledge about language. If that has been done, immanence and transcendence with regard to one's own discipline will necessarily become a theme that will be discussed. It is to be hoped that this dialectics will not simply remain a problem but become a question that should necessarily be analyzed in order to arrive at concrete and interesting assertions. You gave a rather concrete analysis of the aims of linguistics. Perhaps we could now discuss thosefour orfive domains that, according to you, constitute the goals of linguistics. What is text linguistics and what are its possibilities ? - Text linguistics is indeed a good topic to start with - with text linguistics there is a change in the motivation of scientific research. Language presents itself in a text. Every linguistic function observed or ascribed to elementary units until this time - words, sentences, parts of words, etc. - only appears to be clear within a normal language event, i.e. when linguistic expressions appear in the form of a text. Obviously, text linguistics has to remain an open science in that same way, since a text does not contain the whole of communication, but only the linguistic component within the whole communicative process. Thus text linguistics cannot analyze everything that happens and must be described theoretically between sentence and communication. The topics which interest linguistics, however, have sufficiently changed so as to give a certain relevance to this tendency. When, for example, disciplines whose first ob1 Cf. B. Bernstein, Class, Codes and Control I: Theoretical Studies towards a Sociology of Language (London, 1971), and P. P. Giglioli (ed.), Language and Social Context (Harmondsworth, 1972).

124

PETER HARTMANN

ject is not language - pedagogy, psychology, law, theology, literary analysis, etc. - are confronted with language phenomena, these always have the form of a text, not of words, sentences, or grammars. This is the precise distinction between present linguistic observation and linguistic experience outside linguistics proper. When other disciplines also have some linguistic experience or have to refer to it, the collaboration of linguists is only useful if, on the side of linguistics, one operates at the text level. The constitution and function of larger linguistic units which have to be analyzed is in fact a step forward on a much larger front. That is shown, for example, in that the first structure-analytical initiatives of text linguistics were immediately followed by initiatives of text pragmatics. Much remains to be done here, however. Within linguistics, one has come nearer to what could be called a phenomenology of language : the object of linguistics should be language as it really appears. It does not appear in sentences or words, but in such a way that the words in the sentences of texts contribute to a communicative effect. The idea of text linguistics is to get nearer to the real object. Are there texts that are not linguistic ? - It is an independent question to decide whether one wants a text concept that would transcend and contain linguistic text. Siegfried Schmidt has tried to understand productions, that is, ordered sets of all kinds, as texts.2 One can do this but it should be noted that the theoretical - and analytical - value of such a construct is reduced by an excessive broadening of its meaning. Such a 'text', in quotes at least, must then be assumed for music, pieces of art, literary productions and the more so for texts understood as complex productions constituted by ordered elements. It is, for example, possible to talk about a 'text' in a law suit. Therefore, it is possible to construct such a concept of text, but are there analogous efficient methods to analyze such texts? Because those methods should be inspired, partly at least, by the structural methodology of linguistics ; A. J. Greimas, for example, presents something like that for the analysis of narrative structures.3 Which kind of structuralism can be used here and, in particular, whether structural analysis has been transposed on the level of the textfunction must both be analyzed precisely. If one understands texts as all ordered productions, i.e. all ordered sets which have functional value in specific environments, should one not distinguish which kind of signs are contained or used in the text in question? Maybe this could be an important criterion, since notes are not words and dots, lines and colors in a space are signs of a different kind from linguistic signs. If one accepts such a broad definition of text, is text linguistics then different from what was called semiotics or semiology, in classical structuralism ? - I think that linguistics, seen as text linguistics, could be integrated in a broader 2

S. J. Schmidt (ed.), Zur Grundlegung der Literaturwissenschaft (München, 1972), pp. 44 ff.; and "Linguistik und Literaturwissenschaft", Linguistik und Didaktik 1 (1970), 99 ff. 3 A. J. Greimas, Sémantique structurale (Paris, 1967); and Du Sens (Paris, 1970).

PETER HARTMANN

125

context like semiotics. The main criterion used in semiotics is the value of the signs, whereas text linguistics does not come further than to say that the signs which we analyze are linguistic signs. These signs can be treated as such, underly formation recurrences and rules, show invariants and variants, and they can be described in a grammar. Perhaps the question will be asked: 'Why not write the grammar of a play ?' One could talk about the grammar of a sonata with as much sense. I, too, am interested in the questions you ask: 'How far is linguistics independent?' and 'What makes the originality of linguistics?' At this moment, my view could be expressed in this way. In many human productions or actions, ordered sets of interpretable elements appear where the production or action process itself is also ordered, that is, can be described with rules. That is more general, however, than what happens in linguistic productions. Linguistic usage, at some point, is incomparable to other productions, since linguistic signs allow some uses, uses of themselves, which other elements do not, or not in the same way, and which cannot be done with other material. When it is asked: 'Why not unite visual and verbal signs conceptually?' one can refer to the criterion of translatability. In no way is it possible to translate all linguistic signs into visual ones, whereas the content of all visual signs can in some way be expressed on the linguistic level. You characterized text as the 'original language sign'. What do you mean by that ? - To say that a sign is original in language means that it appears in a spontaneous communication. The constituent elements of language within a communication process are texts, not sentences or words. With the term 'original', I meant that when someone who is not a linguist uses language, he uses it in a communicative process, that is, he uses texts to perform the action of communicating. One should stress this disparity with signs that are not directly original in language: sentences, words, parts of words, morphemes, elements, etc.... A great problem for text linguistics is to elaborate an operational description of larger language units. Do you think there is an analogy between smaller linguistic elements within the sentence boundaries and larger ones like texts ? - Is it possible to imagine text grammars as one has had sentence grammars until now? I would agree with Teun van Dijk's opinion that we have to try to write text grammars,4 perhaps in analogy to what was meant before with the grammar of a sonata. That means, the most difficult problem will be to describe linguistic continuity in a text: does one talk in a same, similar, or different way as before? That is one of the decisive differences between sentence and text. A sentence is always analyzed as a unit with its own boundaries - the parallel between text and sentence grammar can only be drawn and attained when the communicative expressions of the text can be observed and structured as closed units. However, that is only one aspect of text 4

T. A. van Dijk, Some Aspects of Text Grammars (The Hague, 1972), especially pp. 1-33.

126

PETER HARTMANN

linguistics. Of course, text linguistics will have to describe systems of signs as they are ordered in an actual text, that is, a linguistic text. On the other hand, it must also describe how texts are generated, actualizing the speaker's competence to produce texts. The latter project is the one that appeals most to me: how may we understand the continuation of a text ? I do not know at the moment whether this has to be done under the form of a grammar, that is, until a specific grammatical form can be elaborated. I think, however, a text grammar is possible. But as long as 'grammar' is considered in its narrow sense, for some time it will only be possible in the case of actual texts which have been understood as a whole. Do you think that sentence grammar is the only possible starting pointfor text grammar ? Must text grammar be opened with a sentence grammar? Is sentence grammar a constitutive component, which is primary and has to come first in a text grammar ? - Starting from the thoughts that led me in the past to ask for text linguistics, I would still answer negatively. One should rather arrive at the result that text characteristics would yield primary constructs and symbols. Basic to text grammar must be the fact that texts are presented as objects and that those texts follow a course. One could refer to Greimas's isotopic levels,5 which seem rather plausible to me since they integrate the aspects of the constitution of texts. It is they who allow one to contract sentences, to make ellipses, to make mistakes, to drop whole sentences, to pause within texts, to introduce gestures, etc. This phenomenon should be analyzed in such a way that a sort of intentional dimension can be ascribed to the text: the communicative form of the text will be the following, the object of the communication will be such and such, this is what the speaker intends with the text, which in its turn determines to relate certain subcomponents like sentences specifically. That is what should be of primary importance if one should construct a linguistically adequate grammar. If I am right, in a generative text grammar, the relational symbols appear 'above' sentential symbols, because the relational levels have to be observed, analyzed, and structured between sentence and text. To achieve this is to construct a text grammar. If one should not succeed, the result would be an additive grammar or, at its best, a grammar of sentence sequences. Teun van Dijk has stressed6 the fact that text grammars must include categories which are not interesting at all for sentence grammars, and they need not be, such as the purely syntactic explanation of a sentence connection. Not only the logical elements which are seen as conjunctional, but, more important, the whole diaphorics of a text, and also its anaphorics and cataphorics, again, not only on the pronominal level, but also on the level of semantemes. The appearance of a word in a so-called context means only that a word gets its meaning because somewhere else there is another word which constitutes a meaning relation and allows for communicating it. Some time ago, we analyzed newspaper texts in order to ascertain how far they were text-bound. It was shown that in independent texts, which were quite 5 6

A. J. Greimas, Sémantique structurale, pp. 69-95. T. A. van Dijk, Some Aspects of Text Grammars, pp. 20-30.

PETER HARTMANN

127

factual - information about new types of cars, or news - hardly a single element could be understood independently, practically every word had a pendant somewhere in the text which was necessary to state, discover, and describe its value fully. In that sense also, 1 think that when we shall be able to do it, a grammatical description of texts will sensibly differ from the sentence grammars that have been constructed to the present time. Is a text typology already possible ? - There are two components in this question: in general, is it already possible to conceive of a text typology, and in a more restricted sense, is it possible to construct a typology that respects current assertions about linguistic theory and methods and still is a plausible text typology? To the first question I can only answer positively. Intuitive observations about the fact that there are various kinds of texts are not recent. Moreover, I think this trivial if formulated in such a global way. The more special question is interesting. I have been concerned with text sorts, mainly with one particular sort of texts, newspapers, and even more specifically with texts whose titles were not well-formed sentences. One can construct immediately a certain text typology using the titles: according to such a criterion, there would be texts without titles, texts with titles, with titles consisting of well-formed sentences, with titles not having the form of well-formed sentences. But that is nothing, really. In a valid text grammar, one could and should include the typical differences between texts. The first question would then be where they can be observed: on the relational level, on the sentential level, on the level of semantemes, or, as I think, on the level of semanteme relations ? The possible occurrence of word contents will influence the classification of text sorts as much as the relationship between the sentences constituting the text. I would have to check this opinion but I think it is valid. A second goal of linguistics is semantics. You have written that there is presently a revalorization of the semantic basis of all linguistic phenomena. What do you mean by that ? - I wanted to characterize an evolution that was taking place at a time: progressively, the observation and analysis of formal linguistic constituents and combinatorics was replaced by introducing semantic content. A clear example is the role played by word content in transformational and generative grammar. It is true, deep structure only contains parts of speech as components, a syntactic word, 'kernels' and specific indicators. In Chomsky's very first work, 7 however, a role played by semantics is already evident. In other words, one could only structure sentences that were understood. When I talk about a revalorization of the semantic basis of all language phenomena, I mean that syntax has become a part of semantics. If that is true, the way something is introduced in a text is important; it depends on the semantic basis of this text, and it must be described with regard to it. 7

N. Chomsky, Syntactic Structures (The Hague, 1957).

128

PETER HARTMANN

Do you consider the classical opposition between syntax and semantics as still valuable ? - The opposition of syntax versus semantics is a methodological analytical instrument. It is not a real belief that can be thematized and would entail assertions of the kind: "this is pure syntax" versus "this is pure semantics". It is an analytical instrument that is used because, in many linguistic recognition procedures, there is some good sense in introducing a separation between combinatorics as such and the combined elements with regard to their content, insofar as one can ascribe a content to them. Obviously, there is no such thing as an isolated, independent content, because words, except in the lexicon, never occur asyntactically. If the opposition is understood as pure combinatorics on the one hand, and what is being combined on the other hand, I do not believe in this opposition any longer. If you ask whether this distinction, in particular the separation of combinatorics and combined elements, has any analytical value or not, I do believe that such a pair can have just that function: the exposure of element boundaries and element relations. In that sense, it is comparable to that old pair in which one does not 'believe' any longer, but with which one would work analytically: form and content. I would say, one must understand such pairs as methodological instruments. They will keep their value if the corresponding analytical procedure is used. Between brackets, I would add that linguistics, text linguistics in particular, needs much heuristic research. One can only guess at what remains to be done. In such a situation, one can only be glad if an analytical tool which has proved valuable somewhere can be used. Before discarding an analytical pair or a methodological instrument, a new one, which is at least as useful as the other one, should be at our disposal. Let us come back to heuristics. Situations will certainly arise, where linguistics will have to introduce something new or where specific questions will have to be answered without recurring to the phrase 'it must still be researched', but it will not be finished before the year 2000! It probably would be too late then. When one enters a heuristic moment, it is important to look at the construction of the tools itself; knowing that one needs a theory will probably not be sufficient. Quite often it will be necessary inadiscipline during periods of heuristic research to work with instruments that can be discarded later on. Here one is right again in saying that syntax versus semantics have fulfilled part of their role as a categorial pair. The function of this pair as an analytical diacritic tool has not disappeared, however, certainly not for ever. Is text linguistics principally oriented towards semantics ? - Semantics will play a very large role in text linguistics. However, it is not yet possible to evaluate the exact importance of semantics. Indeed, what has been analyzed heretofore are texts that were understood, that is, one works with actual texts, which appear in actuality in the communicative processes. When at the same time it is asked that those semantic complexes called texts should be described in the form of a grammar, however, semantics alone will not help very much, because, as stressed above, an

PETER HARTMANN

129

analytical instrument is needed in order to structure continua. Yes, an essential goal of text linguistics is to describe the semantics of texts in a sensible way, but this will only be structurally satisfactory when, besides the fact that everything has 'semantic value' or should contribute to the symbolization of semantic values, it will be possible to describe the object under consideration. Therefore we need combinatorics, which as such are not semantic rules, to combine semantic components. It has to be added, of course, that semantic components in texts may be such that they are much larger than what is usually understood by semantemes. There are many kinds of semantics: philosophical semantics, logical and linguistic ones; even HusserVsphenomenology can be called a semantics - the Logical Investigations, 8 for example. Can you imagine a possible collaboration and confrontation between those various semantics ? - If, for a while, we would keep the various conceptions of semantics you enumerate out of text linguistics, I would prefer it. Not that I do not want to integrate and to use them later on, but because the points of view which were basic to the enumerated semantics may differ from the ones that play the main part in the constitution of texts. Husserl's semantics, for example, like other philosophical considerations about meaning, is in fact agrammatical: it works within the sphere of meaning without considering syntactic phenomena. Besides, non-linguistic semantics is characterized by the fact that it is mostly one-word semantics. An example is Husserl's inkpot. 9 The linguist, however, at least the text linguist, considers one-word semantics as a grotesque abstraction that is used in order to construct lexicons, but whether a lexicon is needed really depends on the kind of description the analyst wants to make of a particular linguistic phenomenon. Of course, lexicons will always be used, it is trivial to learn a language, to translate something; even generative grammars and other quite formalized languages need their lexicons. But the lexicon is now part of grammar: a global conception determines the fact that a lexicon is used. The types of semantics you enumerated contrast with this in that they express a certain idealism or are geisteswissenschaftlich oriented; they seem to be accompanied by 'tendencies' carrying cultural elements. Instead, under the influence of and because of text linguistics, I long for a semantics that would be free from those tendencies, a semantics that might go so far as to recognize one-word semantics no longer. From the side of the one who produces a text, or even from the point of view of semiotics, the semantic organization of a text comprises particular and discrete elements which are instrumental to it. I may refer here to the remarks of Pieter Verburg, 10 who already distinguishes, at the sentence level, between semantics which describes the meanings of semantemes in a sentence and delotics which describes the function of the sentence as a whole. 8

E. Husserl, Logische Untersuchungen (1902 [third edition, Halle, 1921]). E. Husserl, op. cit., II, 2, p. 24, and II, 1, pp. 37-42. 10 P. A. Verburg, "Vicissitudes of Paradigms", Paper 25 (Wenner-Gren Foundation for Anthropological Research, 1964), 30-31. 9

130

PETER HARTMANN

Delotics would lie at the suprasegmental level: for example, someone wants to be insulting with a particular expression, to make a joke, to express himself with or without precision; those tendential components have to be integrated. The kind of semantics I am always thinking of - and I do not how whether it is a very good expression, scientifically speaking - is what could be called normal semantics. That does not mean a semantics of colloquial language, but a pure semantics with few philosophical a prioris about the recognition value of words, etc. More precisely still, I am thinking of a concrete semantics. What I would be interested in is what poetry, among others, tries to do: using language in such a way that the appearance of language as such already carries its own meaning. The appearance of words in texts or discourses should in fact never be described otherwise - it is rarely necessary to think of the meaning of a particular word in order to construct a text. That is all quite unscientific, just expressed as I feel it. I only point toward a semantics that would be free of preconceptions or epistemological ambitions, which always seem suspect to me, whatever corner they are coming from. If one is a philosopher, it is perfectly justified if one wants to construct a philosophically oriented semantics; but a linguistically oriented, or a text linguistically oriented semantics, should contrast with that. Obviously an alternative position is quite possible: for example, the one taken by researchers who from the start stress interdisciplinarity. They will say that linguistics always wants to follow a peculiar evolution and impose its own views but that pedagogues, psychologists, etc. have also achieved something and that linguists should look at that as well. I know that this is a different opinion, and I sometimes sympathize with it. But let us first construct a concrete semantics, a semantics which evolves from the text itself with as few preconceptions as possible. When we have that, then we can see.

Is this concrete semantics not a pragmatics ? I am referring to the distinction Morris makes between syntax, semantics, and pragmatics. - Concrete semantics, as I sketched it above, may come quite near to pragmatics or even be a description of the functions of linguistic elements in pragmatic contexts. I think it is possible. Various semantic interpretations can be seen together; they are needed in the text and are legitimated by it, but at the same time they are relativized as well: discrete contents as such do not play a role in the text, all discrete factors become 'concrete' in that they are used with a pragmatic function, that is, subordinated to a communicative aim, knowing that zero communication, just talking for its own sake, can also be an aim. A third goal of linguistics is to describe the variety of languages in relation to language in general. It is one of the classical problems in linguistics. How do you see this third goal ? - Let us remember all this lies within the scope of the question about the aims and functions of a science in society. Text linguistics as such is a particular aim which

PETER HARTMANN

131

shows how linguistics can fulfil its function within society. The achievements of text linguistics can be applied in many, many fields; many sectors of society will be able to or will have to use the results obtained by text linguistics. Semantics in turn is a subpart of text linguistics, since it has to be developed together with the analysis of text values. The question of language variety is immanent to the discipline and is an eternal problem that existed even before the time of linguistics. What function then can research with regard to language variety have in society ? The answer lies in the domain of translation theory, contrastive grammar, or similar fields; language typology, universal language, and other similar topics are subclassifications or chapters of the science of translation. Universal language would be the language form that allows the description of common elements in various languages - one needs such a tool if one wants to translate. Apparently, universal language is something that exists in the translator's mind as his translation competence. If the translator is a machine, it will, in the frame of its foreign language competence, generate this universal language ad hoc in view of the foreign text through the confrontation with its own language. In that way, 'universal language' is just an idiom for an occasional product that exists when two different forms of symbolization or languages are confronted. Is universal language an abstraction or does it have concrete elements in each language, universals of language ? - I am of the opinion that one should describe sciences in general according to the procedures one uses in practising that science. The alternatives you have given are, according to me, exponents of different procedures. The formulation of such a question, does universal language exist, seems to me the consequence that would be drawn by a scientist who is of the opinion that the common structures he has recognized must have their place somewhere in reality. The question if it is an abstraction, on the contrary, is the consequence drawn by a scientist who is of the opinion that he puts constructs in the world, so to speak, that these constructs have a reality as constructs, which must of course be considered unreal with relation to 'normal' reality. That there is no generality would be formulated as a result by a scientist who is of the opinion that only the individual concrete structure, but not a general structure which is 'nonexistent', is of interest. With the foregoing, I want to express that the alternatives mentioned are conditioned by procedures or method. These three alternatives are used in the science of languages and will continue to occur. It is even possible that one would use one or the other formulation oneself. However, in so doing one should not want to express anything else but the procedures with which one has just operated. Therefore I think that universal language exists in the sense that it is a categorical point of departure or a construct by which the difference between two entities can be thematized on the one hand and the distinction can be controlled on the other. This 'reality' of universal language as construct is not a common form of the reality of objects; it is at the same time 'nonexistent'.

132

PETER HARTMANN

With text linguistics and contemporary semantics, one may observe that old borders are being crossed. Maybe one should ask the question here about the linguisticity of linguistics, entailing the question about interdisciplinary approaches and their possibilities. - With this question one leaves the discussion of contents or task sectors; one is rather discussing whether these sectors are still related to linguistics, whether extralinguistic methods are already playing a big role, or whether linguistics as a science is only beginning to expand its scope of interest. I am of the opinion that linguistics is starting to expand itself in connection with its new problems and that it is rightly doing so. I also think that the question of the 'pure' linguistic method needs to be discussed at this moment. Consider, for example, our introductory remarks to sociolinguistics: what is the linguistic element in a truly sociolinguistic theory ? I believe that, to that extent, linguisticity can be subjected to change. Certain components will survive, namely, those which could be considered the disclosure of linguistic phenomena or the structuring of language reality. It is again an open question how one structures. Someone who looks only for elements as signifiers will structure in a different way from someone who only wants to comprehend the pragmatic effect of certain elements. Structuring texts, for example, will have results across sentences. It will describe sentence structures also, but the structure of the texts will not coincide with that of the sentences. In summary, the linguisticity of linguistics must sometimes be determined, I believe, according to the problems, the state of the research sectors, and the methodological problems in question. There will of course always be cases where, for example, a psychologist and a linguist will be discussing a case of speech therapy, and where one can show what the linguist can contribute and when the psychologist has to take over. However, this seems to me a rather external observation, because only through this cooperation can the linguist gain access to new categories which he could not have obtained by himself. The scope of the linguistic interest can and should shift occasionally, and thus the concept of linguisticity can change also. Certain components seem to be essential, namely, those which yield a rather old description of the linguistic science: that language as such is the object of investigation and that one structures the language continua in a way which may vary but which must render the language phenomena comprehensible in a system. Interdisciplinary cooperation is necessary in principle, but might it also be a question of research organization? - It is indeed possible to present the problems of linguistics more fully, not only with regard to precise themes such as discourse analysis, semantics, theory of translation, but also according to questions such as how linguistic findings can be incorporated for example into the domain of the school in an intelligent way. What can linguistics contribute to the analysis of the therapeutic conversation? How could linguistics comment on the role of language in mass communication or in journalism and so on? Such problems can no longer be sufficiently dealt with, in my view, by a single linguist

PETER HARTMANN

133

or even by a group of pure linguists. The interdisciplinary approach, often just a slogan, may be rightly cited here. Research organization should be topicalized. There will and should be individual research as before, but one can already perceive a categorization of research topics. Some of those topics cannot be treated in individual research; some of them cannot be treated in this way any more as they require structured project tasks and teams; there are also topics which require groups of teams; and, finally, some topics require a fixed institutionalized research unit. 11 This situation is not yet realized everywhere, and it appears that the research categorization which can be distinguished is not yet fully instituted in linguistics. Questions of organization of research, of the combination of individual researchers on a topic that transcends individuality, seem to me very important for the future. This is especially so because there are more and more topics which must have at their disposal more than one point of view, and, as a result, more than one discipline. Interdisciplinary research, if it is really necessary, should be organized into corresponding teams or into larger units. One should know, however, that nothing is as hard as initiating and following through such tasks well. If one has a little experience with it, one notices the important role of the initiation of new co-workers, the assurance of continuity, the maintenance of the work structure, and personal considerations. One must still learn something here. Could you please express your opinion on some special relations among disciplines ? Here in Konstanz, for example, you have work groups for formal linguistics in which mathematics, logics and linguistics cooperate. - The purpose of this interdisciplinary work should be classified under a more general title within the framework of linguistic problems, namely under theory construction. I would amalgamate mathematics, logics and linguistics under this or consider them as components to be merged, as it has always been in our operations here in Konstanz. Questions on the development of theories and probably also on methods should be discussed in such a triangle of disciplines. In 'theory construction' within linguistics today, one should operate concurrently with mathematics, logics, and linguistics. Do you consider automated data analysis of linguistic phenomena possible or significant ? - The experiences of our group have shown that automated data analysis has become a necessary derivative within linguistics from the moment one wants to test whether methods used, analyses, etc., have attained such a degree offormalizability that data analysis can be employed. In other words, the role of data analysis in linguistics is that it obliges linguistics to formulate its problems so that data analysis becomes possible. This sort of formalizing also leads to a distantiating judgment of linguistics, of its own methodology, etc., which would otherwise hardly be possible. It must be observed that automated data analysis is presently and will remain of pressing imporII

Cf. P. Hartmann, Zur Lage der Linguistik in der BRD (Frankfurt, 1972), pp. 43, 50-56, 88ff.

134

PETER HARTMANN

tance for areas which play a relevant role in linguistics but which do not, of course, constitute the whole area of this discipline. We probably agree that the question of the development of methods, of instruments, of theories, is one section. Another section, however, is the production of applicable scientific knowledge for customer areas. Theory construction and methodological questions are very important. Could we talk about that now ? You have written that the scientist must do his work consciously and with critical rationality. I do not know any other linguist who emphasizes methodological questions so strongly. What do you mean by 'critical rationality' ? - I might express it this way: the idea behind my work is to shape a linguistics which deserves the name of a science capable of theory making. In my view, the corresponding metatheoretical level which makes the evaluation of one's own theoretical primitives possible belongs to this theoretical fitness. 'Critical rationality' would belong to this metatheoretical level as the ability to adjust one's own discipline with the problems and expectations which may be anticipated from it and also to adjust this discipline with the expectations one anticipates oneself from this work. 'Critical rationality' should therefore express that linguistics be made flexible when certain constellations, requirements, etc., demand it. This should formulate a distance from the special linguistic science which one has chosen and must cultivate oneself, a distance from pure expert thought or from thought which is only scientifically immanent, a distance from a, let us call it, mere naively rotating or adopted methodology. Perhaps it is now tolerable to offer a statement, which seems to me to be necessary in the present context for the completion of the points of view which are determining for my work. My opinion has been, and still is, that the science of language - and this includes linguistics as one of its special branches - should become more 'philosophical', to assure particularly that the discipline perseveres and develops, and that it conforms to the requirements of distantiated and enlightened intellectuality. This has always been highly regarded by scientists. It must develop in regard to the requirements of methodological knowledge as well. Such a position exposes linguistics, in particular at the level of its fundamental research, to philosophy of language on the one hand, and it enables linguistics to judge and evaluate the results of other disciplines on the other. Finally, it is elevated to the metatheoretical level which is indispensable for interdisciplinary work. It goes without saying that the concomitant perspectives, points of view, and methods of theory construction and theory evaluation should be integrated as fixed parts of courses of study. A linguistics which is characterized by 'critical rationality'' is Grundlagenforschung... - A discipline which claims that it can control itself and develop further and thus also make plans must do 'fundamental research' (Grundlagenforschung) because I would consider it a prerequisite for the attainment of critical rationality. In short, I think that the foundations of a discipline must actually be investigated, that is, speaking concretely, the procedures that lead to knowledge. One could, then, also describe

PETER HARTMANN

135

fundamental research as the criticism of methodology or as the investigation of all those components which together compose the peculiarity of a discipline. Fundamental research is in this respect a somewhat dangerous branch for a science because it must of course also bring to light the components which are actually not peculiar to a certain science, namely that in which all sciences coincide, because they are sciences and thereby not 'typical'. Fundamental research, then, must also make visible what actually makes a science a 'typical' or special discipline. In my opinion, a discipline becomes specific in this way: a bundle of procedures is connected to methods which are employed in certain steps and for the purpose of certain knowledge. The determination and the production of projects would be fundamental research with regard to the contents, and this would relate again to what was said at the outset regarding the tasks of a discipline like linguistics: What tasks does linguistics have ? Where can one recognize projects? How can the projects be regarded as consisting of active functions for others ? This may, of course, also be read - in parentheses - and it is really meant, that a science can have a socialist component: when it does nothing or undertakes nothing in its knowledge without the concomitant idea that this should also benefit others. The production of projects as a component of fundamental research leads somewhat to the border of the discipline. It presupposes that one has contacts with areas from which projects for one's own discipline can be formulated. In summary, fundamental research seems to me indispensable in order to be able to judge the processes of one's own discipline, to relativize its results and, finally, to relativize the discipline itself because it is respectively connected to other disciplines and its environment. A linguistics with a built in critical rationality has theory construction as one of its methodological components. You have written that there are two methodological components: the inductive analysis of material and the theory construction, which are related to each other. - In a linguistics so understood, theory construction is a component in a more precise sense than in other sciences, which are perhaps not so decidedly critically rational. Indeed, it makes theory construction itself an object of investigation, and the two phases you mentioned, heuristics and theory construction, or induction and deduction, are to that extent an object of reflexions and interests proper to the discipline. One could say in general that it must belong to the catalogue of investigation to topicalize, communicate, discuss, simply to teach as such its proper theory construction. I would not consider the relation between inductiveness and deductiveness as a pair, where one decides for one side so that this, let us say, becomes a conviction of faith or opinion. I would consider it a procedural pair that behaves either inductively or deductively according to what has to be done. When one does heuristics, one can for the most part proceed inductively; one even must when, for example, there is no taxonomy as yet. However, when one has to describe, or when the theory should be drafted for descriptions, one can or even must operate deductively. I would, then, separate 'inductive' and 'deductive' for different phases, and I would do this for each

136

PETER HARTMANN

research process no matter how long it is: it may last five minutes or a year. An instrument is invoked again here, the operation of which has to be learned. A t any rate, I believe that one cannot and should not teach anybody the sovereign operation of instruments without a corresponding distance from which one surveys the problem and self-criticism. Is there only one purpose for theory construction or are there several? - There are different kinds of theories depending on the purpose they should fulfil. One should perhaps not say that theory construction itself has several purposes. What purposes can theories have ? One could say that there should be such theories that allow the description of something, more specifically, that they also explain something through the structuration necessary for descriptions. Reflections on text linguistics have shown that one can obviously distinguish types of theories, especially according to what one should do when one has the theory. The following questions arise. Should the theory be formalizable or not? Should the theory only describe completely known areas of objects, or should it also describe incompletely known areas of objects, that is, those that are open or are still in their development? One can add to the former observation that it is significant whether completely understood 'texts' must be comprehended or whether the 'texts' are still in progress. The question of the forms of theory that diverge accordingly is not yet solved. It seems to be a fact, however, that, although the demand of formalizability is higher, it severely restricts the scope of a theory or, let us say, the tolerance of usability of a theory. The question is, where, in what areas of application, does one need which theory ? Our considerations of this are not yet complete. For the moment, we are interested in types of theories, and we have discovered that the so-called theory of science has been provided so far with areas in which the object counterparts (real facts, natural laws, etc.) do not change - theories are primarily offered here, somewhat like calculations etc., which are not necessarily applicable to areas, for example, speech events, in which the appearance of objects change definitely as they can be composed or interpreted differently. One could add the following question as an afterthought: if one assumes that the science of language should place its findings at the disposal of the community for the widest possible application, then a very important concept appears, namely, that of 'simple technology'. This does not refer to a technology or theory which is worth less; it means that the highly ranked theories constructed by scientific experts must be adapted so that non-specialists, non-theoreticians, can work sensibly with these theories or with their results. Let us take a concrete area as an example. Linguistic findings should be made productive for the area of elementary education, an area where one cannot expect that the teachers or educators would be linguists; they should not actually be linguists nor have any expert qualifications. On the contrary, they should rather have the qualifications of non-scientific people, especially since they should be able to deal sensibly with children in this elementary phase. 'Simple technology' seems to me indispensable for such areas of

PETER HARTMANN

137

application, in other words, an adaptation of linguistic findings, so that they can be correctly evaluated and used by educators on the elementary level. This requirement should not be understood as a leveling or deterioration of science, but, on the contrary, as a new dimension of the science in question. In my opinion, this requires much more than what specialists who have learned to construct such theories usually offer. It must again be assumed, as has been summarily mentioned before, that there is a certain distance between one's own theory construction and method as well as the ability to judge for what a theory can be used and what is necessary to employ it correctly. Is there an ideal way of model construction in linguistics ? - I cannot honestly answer this question yet. I would understand the construction of models in this way. A piece of reality is a model of the corresponding theory if that theory is confirmed or proven by that piece of reality. The question of an ideal type or an ideal way of model construction would hardly be answered from this position. For me the question is basically this: is there an ideal way of theory construction in such a way that I can consider arbitrary pieces of reality as its model? This question, however, could only be answered after one understands clearly what the relation is between a non-ideal type of theory and reality. My position is that, in principle, each theory, even incomplete ones or those pseudo-theories afflicted with large tolerances in their constructs, is usable for model construction, insofar as pieces of reality can be coordinated to it. That is, we consider the model as a series of applications of theories to areas of reality. In this light, there could not be any ideal or non-ideal cases for us, as one produces the coordinations for the model. It is clear that the adequacy of the theory plays a role. Dissatisfaction or the fact that it is not ideal comes on account of the theory but not on account of model construction which is understood as a neutral procedure of pair construction. One could understand model construction in another way: so that a model would not be the result of the confrontation of a theory with a piece of reality, but merely consist of generalizations which, as a model, should also be valid for other phenomena. If one agrees with this formulation, one could discuss this also, but even then I would still be unable to answer the question whether there is an ideal way to proceed with generalizations and transfers from the model. For clarity's sake, one should distinguish two cases, model construction I and model construction II. You retain in this formulation your two principal components, the inductive and the deductive phases, and I believe that in yourformulation of model construction, heuristics, for example, plays a role. - One can say, of course, that model construction II, in which generalizations are made for the purpose of transfer to other areas, is that form of model construction which is only possible during a heuristic phase; indeed, the heuristic phase is so characterized that it serves to collect material for which one wants to construct a theory later

138

PETER HARTMANN

on, whereas the deductive or theoretical phase is a later step. One could agree on this; model construction II is the generalization and categorization possible in the heuristic phase. On the contrary, model construction I is the confrontation with reality accomplished after a theory is available. Then the question of the ideal model construction in the case of model construction II could be restricted within heuristics: what must happen in heuristics in order to arrive at transferable structures ? The central epistemological question is the relation between object and description. In that respect, you have made an important procedural distinction between scope and effect in descriptive systems. - Scope and effect should be distinguished, so that by scope the following question may be answered: what does one pay attention to in descriptive systems? I feel that the assessment of significance made descriptions possible, as is still the case, and that it is as such the first or basic procedure for sciences and disciplines. One wants to recognize certain things and one proclaims or finds, therefore, certain characteristics significant. That is, one proclaims them as the decisive components for the investigation or as characteristics in reality. The scope, that is, the trend in the description which concentrates on a certain descriptive landscape causes - that is, one takes care of that oneself - significances to be established. The effect of a description is thus accomplished, that by dint of the significances described, interpretations of these significances can be provided, and the effect is simply that knowledge is realized. The investigation of significances and the knowledge of that by which significances are investigated do not coincide. Do you think it is possible to employ formal languages in the description of natural languages ? - Yes, certainly. Because formal languages are special cases of the descriptive languages which one employs, at first for totally arbitrary descriptions. One should distinguish the object language from the descriptive language and the theoretical language. The descriptive language would be there to refer to the object language; the theoretical language would formulate what has shown up in the relation between descriptive language and object language or what has been explained or theoretically recognized. A s to your question: are formal languages usable in this business ? In any case, yes. For example, certainly for the constitution of theoretical languages as soon as a theory should be so that its statements are formalizable. Whether formal languages are already sensible in the role of a descriptive language, that is, for directly attacking the objects, can only be decided, it seems to me, according to the paiticular instance to be described. Therefore, a formal language for the description of meaning values of individual speech elements does not yet seem totally plausible to me. The question is when a scientific language should be coined in the form o f a calculus. If one is satisfied with a formalization of, say, the first degree, merely with formalizing symbols where formalization is already accomplished by an unambiguous association

PETER HARTMANN

139

of term and object or of symbol and object, then it may already, I agree here with you, result in a usable descriptive language. The distinction between the three levels of language which was drawn above, that is, the formalization and purpose types, does not seem to be unimportant as a whole, and it should be decided from the foundations where or for what reason formality should play a role: should it be used as a theoretical language or should it be part of the descriptive language. After all, it may happen that the descriptive language must refer to occurrences of the object language, for which a formal form or a formal language as a descriptive language is as appropriate as a non-formal. One would, after all, also hope for that. In view of the goal to develop a formal description, for as many phenomena as possible, one would like to know better what is important for the description. That was the important, very difficult methodological problem. Could we now discuss the object of linguistics: language. What is language? You have called a linguistics adequate to its object, a phenomenology of language... 12 - Maybe one could make even more exact distinctions, so that one separates in linguistics, as in every discipline really, the subject areas, that is, why it should make statements, its object area, that is, about what it should make statements, and its object, that is, what often becomes a reference point in the process of a specific scientific act of understanding. I would separate subject and object because it is clear, for example, that the object area of the science of language is language occurrences. The subject area of the science of language, however, is to make statements, the result of which is that this object, language, has been understood scientifically. To the subject area of linguistics also belong the sound laws, structure assessments, etc. The object of the science of language would succintly be called language. Linguistics adequate to its object is a phenomenology of language or a science of the phenomenality of language... - Linguistics as adequate to its object brings us again to text linguistics. It actually seems important to me to constitute a sort of linguistics which so incorporates the object, language, in its scientific screen of knowledge and methodology as it actually occurs. A sort of linguistics does not seem to be adequate to its object when it has already undertaken a selection, that is an abstraction, before it has employed its procedures for gathering knowledge, making a judgment, or forming structuralization. In this case, already reduced phenomena come into its field of vision, as was already mentioned shortly before with relation to the text, sentence and word. Of course one also knows that a pure phenomenology, somewhat in the sense of Husserl, is practically impossible. One should nevertheless strive toward a sort of linguistics which is adequate, a linguistics that would come as close as possible to a phenomenology of language. The linguistic aspect of this phenomenology would be that the phenomena 12

Cf. P. Hartmann, Zur Theorie der Sprachwissenschaft (Assen, 1961), pp. 69 ff., 81 ff., 141 ff.

140

PETER HARTMANN

would be investigated and structured according to linguistically relevant significances. Now, what kind of phenomena should this be? Especially those that really occur, also those which, for example, are ungrammatical, in the sense of generative grammar. Maybe this is a parenthesis here, but do you recognize in your work a certain influence of HusserVs theory ? - This question is somewhat hard for me to answer, notwithstanding 'critical rationality', because I should have had to take myself a little more importantly, in the sense of a self analysis and confrontation with Husserl, than 1 have been willing to do so far. Husserl has, indeed, consciously interested me, in fact, at a very early time; 13 it concerned the treatment of questions of language in his Logical Investigations and in Formal and Transcendental Logic.1* What interested me in Husserl rather more than in other philosophers was the frequent remarks on the theory of meaning and also the interpretation of sentence construction. I admit that when Husserl expresses his ideas of meaning, individual words are 'blown up' philosophically and metaphysically, or at least overemphasized: a situation of naming things or of word use is analyzed, which is really odd in normal language life. Therefore, it is actually a case of mysticlike self observation, consciousness, and exclusion of all other contexts, when he says "Book!", and the whole imaginable 'content' of the word would appear to him. What made Husserl especially interesting to me in those days, however, was that he was the only one, I believe, to concern himself with syntax, the only one who talked about sentence forms and concerned himself with the formation of groups of adjectives, etc. In short, he was probably of interest to me because he was not a pure philosopher in the direction of the Geisteswissenschaften, but he also knew mathematics and logic, as the symbols which were under consideration in the language expressions he investigated were viewed by him as complex and also analyzed. That was a question related to the genealogy of your work. Humboldt and German idealism is also present in your works, as a broader perspective. - To my 'genealogy' belong, indeed, names you have already mentioned as Humboldt, and others: Cassirer, Weisgerber, Saussure, Hjelmslev, and Glinz; in individual works I have appealed directly to these authors. Of special interest to me was what Cassirer really said about the description of names. 15 I have written my own book on Weisgerber16 because I wanted to know what could be found or is still to be found in actuality and for my point of view; a rather long conversation with Weisgerber showed later that he did not refute this work but that he thought that some of the 13

P. Hartmann, "Die Rolle der Sprache in Husserls Lehre von der Konstitution", Der Deutschunterricht 6 (1954), 29-55 (Inaugural lecture). E. Husserl, Logische Untersuchungen (1902); Formale und tranzendentale Logik, Versuch einer Kritik der logischen Vernunft, in: Jahrbuch für Philosophie und phänomenologische Forschung 10 (1929), 1-298. 15 E. Cassirer, Philosophie der symbolischen Formen, I. Die Sprache (Berlin, 1923). 16 P. Hartmann, Wesen und Wirkung der Sprache im Spiegel L. Weisgerbers (Heidelberg, 1958). 14

PETER HARTMANN

141

points discussed in it should be reconsidered. During this phase I tried, so to speak, a hermeneutic procedure concerning representatives in the field of linguistics and philosophy of language. I am now reading, for the same reason, Steinthal,17 for example, which might sound surprising. The history of the science of language, however, is one of our main and supporting interests. We consider this crucial as a discipline without knowledge of its own: the history of the problems is worth only half as much as when it disposes of conscious judgments. Such reading shows, for example, that in addition to educible components of the systems presented by Humboldt, Cassirer, Weisgerber, etc., there is something like 'deep structures' in these individual systems. Moreover, there are other characteristics which do not so much characterize these thoughts as typical but which one can keep using in order to connect or understand something else with them. Thus, those studies or rather efforts for understanding from the past form relatively present hints, which is not the same as thinking like Humboldt. Finally, Humboldt, for example, was not a linguist at all. One can say that in the certainty that one has taken him up first and understood him well. Let us specify one of these themes: language as form, in relation to Humboldt''s 'interior formC (Innerer Form). What do you mean by this definition of 'language as form' ? - 'Language as form' for me was a connection of statements as they can be found in Humboldt, Saussure, but also Hjelmslev. The idea which it expresses is language as an act, more exactly a partial act which represents and also expresses a content, which per se, however, is a usable, tractable, arbitrarily manageable 'purport'. Could your notion o/form be compared with the structuralist concept of structure ? - One could confront it with that. The matter of form, that is, the combinatory potential and constructability of linguistic complexes in view of certain communication goals, is described in its different representations by one's ability to ascertain structures in language occurrences. In this sense there is no identity between the two concepts form and structure. What about the traditional opposition between form and genesis ? Is this opposition irreducible ? - One could rather say that the idea that the form is a combinatory product, made available by language as concomitant action for the main action of the contentoriented utterance-intention and thus for the communicative purpose, was generally behind the statements on 'language as form'. The view was especially that the 'genesis' of such combinatory products and the connection conceived for a perhaps 'genetic' language description announces or offers itself. It must be said that those things were not yet a theme in my past considerations. One can consider the production 17 H. Steinthal, Einleitung in die Psychologie und Sprachwissenschaft (Berlin, 1881); Logik und Psychologie, ihre Prinzipien und ihre Verhältnisse zu einander (Berlin, 1855).

Grammatik,

142

PETER HARTMANN

of actual language occurrences, the production of texts, as the generation of form components of linguistic communicative contents. The total language occurrence obtains, through the method of combination of elements, an articulated appearance of which the recognized structures can be described. I do not think you believe language is something static, rather a 'product'. One can find the following phrase in Saussure: "language produces a form", which is opposed to the sentence "language is a form". - Indeed. Because the variability and combinatory potential of elements should be the formal aspect, that is, that which is available in the language for contents of a different sort and tendency, and that language, insofar as it is designated as form, must be seen as a whole, as a structurable, differently constitutable processual form of sign organization, or, as I would say today, of text constitution. ' Grammaticality' is a key-notion in transformational grammar. I find in your work the concept of 'grammaticity' (Grammatizitat). Do these notions denote the same linguistic phenomenon ? - With the pair 'grammaticality' and 'grammaticity' different levels should be addressed. The distinction should be the following: that each language contains in reality a system of rules, "has a grammar", from which universally for language the result is that there is no language without grammar, without rules for the use of signs grammaticity should designate the fact that languages have grammars. The former idea was that languages have and need a grammar because they operate with discrete symbols, and identical symbols are combined in different relationships and in the form of different combinations; these products are described by grammars. I rather believe Chomsky meant by grammaticality that something coincides with a certain grammar, understood as rule norm; this belongs in any way to the level of particular languages. The discussion of 'grammaticity' has its source in the attempt to prepare a universal grammar and to collect the components for a universal grammar in a partly pretheoretical and chiefly heuristic way, for which a theory would have to be developed later. Such a phase, one could probably call it pre-theoretical or heuristic, still seems to me important and justified, even if today one would perhaps do some things in a different way. Milka Ivic has observed18 that the work I have done in my Theorie der Grammatik,19 elaborating the concept of 'grammaticity', constitutes a first s tep for what was later done in generative grammar. Connected to this is the question of the possibility of a universal grammar. Do you think it is possible to construct a universal grammar ? - One could conceive of a universal grammar as the collection of grammatical data 18

M. Ivic, Trends in Linguistics (The Hague, 1965), translated as Wege der (Munchen, 1971), esp. § 521. 19 P. Hartmann, Theorie der Grammatik (The Hague, 1963).

Sprachwissenschaft

PETER HARTMANN

143

which apply to all languages. That should not mean that one would generalize the accusative or the tenses if they cannot be generalized. What one can ascertain as universally observable should be expressed, for example, that all languages must be able to express time relations. In this respect one can certainly, as I notice now, recognize the influence of Husserl's thought. He asked the question, how the existential statement is expressed or formulated in the different languages. This actually meant that the linguistic form of the existential statement in different languages should be established, as well as the forms of time relations, etc. This practically results in a grammar of possible forms. It is probably somewhat shocking for linguists to acknowledge a grammar of possibilities. If one adds that what actually matters is the description of language forms, then the possibilities of language use become a legitimate topic of research for linguistics. One also obtains a gate into the abnormal or 'impossible' uses of language; all that would fall out of a, let us say, orderly grammar, but which is very important for literary production, extreme situations, child language, etc., could be incorporated. You have characterized language as a determining praxis which is time suspending. In this definition there are two components; determining praxis and time suspension. Could you expand on that ? - Your question refers back to some work I did on questions of syntax. The idea was then that sentence construction, viewed in terms of content and cognition, is a determining praxis, not just an empty, pure act or an activity of usurping or communicating. It was rather a view on what happens - cognitively or conceptually - when several semantemes are combined with the result of a complex determination so that something develops which is commonly understood as the structure of a judgment. It was never quite satisfactory to me to view this process as merely logical, to set it off instantly against the ability to judge, and consequently to make a predicate always formulate a judgment: one does not only have the feeling but one can also determine that often nothing is actually judged. Nevertheless, with predication a procedure is applied, namely, word contents relating to items - with or without reference relation are connected to each other as variables. Thus, a complex class is formed intensionally, that is, by dint of these word contents. If this multiple characterization is classified with regard to its effect, then one can say the stratum of things or contents is determined. 'Determining praxis' should express this. One could say that this is no longer a linguistic term. 'Determining' contains the expression of a form of specifying by dint of cognitive procedures: a sentence content is specified intensionally and for this one uses words. I believe that this is the main task of words or the main effect of word use. Time suspension refers to the combination of determining components formulated free of time. This does not mean that one cannot appeal to a specific time level or time in general; time address, time as reference, is kept totally intact. What becomes irrelevant is that time relations exist between determining elements, that is, between procedural factors as word contents, in the execution of determination.

144

PETER HARTMANN

You have interpreted language in several of its functions:20 language functioning as naming, shaping, distinguishing, determining, understanding, but I do not find the expressive function in this list. - You are right. Expression and expressivity are for me, still now, a somewhat unclear category, in contrast to the functions you mentioned. This impression shows up occasionally, for example, in contact with literature. I feel that the main question still is what the function of texts actually is. For the time being, I think a text has to be defined in the framework of a common communicative context. For me, something like 'assertion effect' would contain the category 'expression'. It can be the effect or the purpose of an assertion to express oneself. I admit, however, that this function, to express oneself or to prove oneself, is in this way perhaps broached rather poorly. One could, however, also 'express' thought... The 'expression of thought' has not become an issue for me so far. Anyway, my concern with the functions was not to collect all recognizable functions of language. With the series: naming, distinguishing, determining, understanding, the cognitive valence of the linguistic sign is brought in as relief, and to express oneself is, I think, not a cognitive valence, maybe as little as it is a cognitive valence when, for example, a theologian or a teacher of religion talks about God as an expression of faith, that is, as a participant. That is not a cognitive effort but it is rather meant to proclaim or engender a certain form of behavior. That is, for me, also a legitimate language function, but I would want it described separately. However, the series mentioned from naming to understanding is clearly coined only on the cognitive valence of the language sign. 'Expression' is lacking there, because self-expression was not understood as cognitive valence. The state of affairs would still have to be made clear for the 'expression of thought'. And what about your statement that language is a useful praxis ? - That contains more of course. 'Language as human praxis' means that it is possible for man to live as an acting being, to have language, and to use and apply it. The stress is on human praxis. It can probably incorporate everything, and thus this formulation could cover in shorthand everything a man can or should transform into his own personal actions. A very extensive discussion, which is nevertheless topicalized in your work, is the problem of the relation between language and thought. - Indeed, this has always interested me. The question is only whether it is or has been well formulated that there exists a parallel between thought and language. What one can accept is that through the association of signs and cognitive operations a parallel can be produced or insured. If that were not possible, nobody could accept after all 20

P. Hartmann, Theorie der Grammatik p. 373; Das Wort als Name (Köln-Opladen, 1958), passim.

PETER HARTMANN

145

that he could express one of his ideas so that a listener can grasp to a certain degree what he has thought. However, one could say that, with relation to the fundamental possibility of drawing a parallel between language and thought, whether it exists or not must still be investigated. From the moment science is concerned, this parallel has to be desired; in normal life, however, it does not have to be wanted. It can also be intentionally undesirable, in diplomacy, for example; this can also be the case with lies or when one jokes, pulls somebody's leg, the relation can be scrutinized more closely. The parallel is adopted as a basic principle or again as a tool; it is, however, not asserted in each individual case as if it would be otherwise impossible. Can one here use the phenomenological distinction between thought as the given data and language ? - I do not think so, as I would not know what the given data (das Vorgegebene) would be. There are cases where 'the given' can be used, where the parallel between thought and language is consciously established, in the scientific idiom, as I said. The question whether one could not describe thought by means of language anyway should not be discarded easily. The question is only: in what circumstances? And are there circumstances in which this is the case, then yes. As one is of the opinion that the parallel should be established as a basic assumption without its being given in each case, however, one can only say that one has to check whether this distinction between 'given data' and language can be applied or not. Thus, it could be that there are no 'given data' in something like literary texts or in purely intensional language use, because here the point concerns exactly the way in which language develops into an expression. The epistemological question of the relation between thought and language is very important if one wants to know whether logical structures are also linguistic structures. - Yes, or the reverse. Or the reverse... - Right. The problem behind this is whether large parts of logic or even mathematics do not turn out to be phenomena of language. This is valid, I think, for meta-mathematics and for the level of the evaluation of mathematics. In that context, it looks as if mathematics, logic, and language can once more be properly joined into a triangle. What was said before, that one would need this triangle for the theory construction of linguistics, appears now in a different context: the three should be specified together and as to their mutual relationships because it is already obvious that language plays a role in logic as well as in mathematics, and probably everywhere where signs are used. It is also clear that mathematics plays a role everywhere rule governed behavior, structurable behavior, is present or to be described. This again can be found in language and logic. Likewise, the logical, that which logic describes, is found in language and in mathematics, in language somehow in the combinatorial

146

PETER HARTMANN

possibility or incompatibility of the contents of semantemes. The three mentioned system forms could actually contribute to one business which emphasizes itself differently in each of the three components. Another language philosophical problem concerns the relation between language and world, or language and experience. / believe that you have somewhere referred to language as 'interworld' (Zwischenwelt). - If I have called language an 'interworld', one must bear in mind in what sense this word has been used. Because the expression Zwischenwelt has been partially determined by the word use of Humboldt, Weisgerber, and probably also Cassirer.21 If one understands language as the existence of ordered symbolic contents, one can imagine these symbolic contents placed between the subject and the object of experience. Before we talked about the pair 'given data' and language. In the situation where one can operate with given data, the following question is legitimate: "Is there a word for it?" In such moments, and there are many cases where this question is asked within science, it is not only consciously investigated, combined, changed, created, etc., but it seems then rather evident that the subject, understood as subject of experience, and its thoughts in no way coincide with the value of the semanteme. The semanteme value and the object of experience do not coincide any more. Sometimes we ask ourselves exactly this question, whether or not they coincide in a certain speech situation or communicative purpose. Always at that time when language use becomes conscious, the separation sketched seems to me to be given. This again is a point where one talks from a metatheoretical point of view about what it actually means that language is a mediator when scientists can make statements in which there are clearly 'given data'. In certain speech situations and purposes it can make sense to say: symbol and reality, symbol and object do not coincide in any way. Then there are also other cases where it is true that, as a matter of course, symbol and reality coincide as one expresses oneself in symbols. Here lies the reason why certain expressive forms seem rather doubtful as matters of investigation: because there the 'life with symbols' plays such a role. It is perhaps especially difficult for a scientist, because he is not a writer or a participant, he lives in symbols, because he is bound to investigate symbols as an observer, and he is happy when he finds a criterion according to which the symbol obtains a characteristic value; as such the value of the language system is viewed independently, and language is granted autonomy. In connection with the relation of language and experience you have said in a plausible formulation that language would be the pairing of two kinds of acts: the inner experience and the outer picture of this experience. - First, one would say here that language is used with the purpose of representing experience. One starts from this point of view so that one has the distance to determine 21

P. Hartmann, Theorie der Grammatik, pp. 126 and 303; Wesen und Wirkung der Sprache im Spiegel der Theorie Leo Weisgerbers, pp. 62 ff. and 134 ff.

PETER HARTMANN

147

that one can attain this purpose in one way or another. Then one adds inner experience, and for this one could probably substitute thought. In addition, there is the realization of the outer picture. From this, two kinds of acts resulted indeed. To have thoughts is not the same as to have signs - it need not be the case anyway. One can consider whether one has a sign or the fitting word for an idea; if both are strongly intertwined, idea and picture take place at the same time. I think that it is possible. However, that there are in principle two kinds of acts results from the fact that it does not always have to be that way. Not always that way? - It does not always have to be that way. Many people would naturally say that 'outer picture' is articulated much too coarsely; and Herder and Weisgerber would probably say that also somebody who does not talk while thinking has language available, of course, because otherwise he would not be able to think at all.22 One should here, however, look more closely into what that would actually mean, and whether somebody like that would really have 'language available' or whether he has available only the capacity for identifying something and covering it with a symbol or a name, no matter of what sort. It is my opinion that one does not need to say that such a thinker 'has language', but rather that it suffices to say that he could if he wanted to; he could give a designation or give reference to a word content, but he does not do it. Interestingly, there is an analogous remark in Husserl: if somebody reads a book, and he reads rather quickly, he does not in any way need to imagine the content of each word, and he does not do it either; if it need be, however, he can bring everything to mind, and, anyhow, much remains an 'empty' formula for him, for example, that Greenland is an island, something no one has surely been able to 'experience' himself yet. If, however, one needs the concrete representation, one stands still, as it were, and one imagines an island or goes on thinking in some other way. But one does not need to do that. I believe that it is generally so when one thinks: it is quite sufficient that one could picture it externally, connect it with words, but one does not do it. Only the potential linguistic competence is sufficient. To that extent one can probably say that someone must be able to handle language in order to conceive certain matters or relations. One can, however, execute conceptual operations without determining reference by means of names. We are arriving now at the 'critical' part of our conversation, the evaluation of the work in contemporary linguistics. We may perhaps talk in a general sense about the research requirements in the science of language. - The research requirements as a proper category are not necessarily a topic within the common academic disciplines. However, as soon as one takes seriously the necessities which were mentioned at the beginning, namely, to view science in certain 22

Cf. P. Hartmann, Wesen und Wirkung der Sprache in Spiegel der Theorie Leo Weisgerbers, pp. 28 ff., 134 ff., and 153 ff.

148

PETER HARTMANN

functions for others, it becomes indispensable to prove how far the scientific effort until now, thus also research, has left gaps, and even more, for what certain basic research would be necessary in order to fulfil these functions. For example, at the moment one wants to clarify the question how linguistics could somehow become relevant to educational problems, it is no longer feasible to construct an even more elegant grammar perhaps, but one should rather simply determine whether it would not first be necessary to make clear as a prerequisite in what form grammars of certain types can be introduced in the school. Many things here are evident. It is clear that, for example, grammars of the transformational or the generative type describe language competence. Thus linguistic competence must be available before grammars of this type can be sensibly used in school. On the other hand, it is clear that the native language is brought to school, but that the foreign language is learned. From this the result is that, for example, in German schools, one should not necessarily make students study French or English by means of a transformational grammar because, in this case, linguistic competence is just being built up, and it is not an available language that is being described. Another sub-question on research requirements which should be discussed briefly is that of research planning. Until now, research planning has normally been undertaken only in large institutes and it is emphasized on supra-regional levels. One will here also distinguish all kinds of plans, which are distinguished: planning of contents, of personnel, of finances, of research, of studies, etc. Looking again at the research requirements within linguistics, one must say in general that certain areas have remained uncultivated. For example, the whole field of the question what language does accomplish in the framework of socialization.... One can specify 'socialization' by establishing that every participation in a special group with a special purpose, for example, going to school, going to college, studying a certain course, etc., is a socialization phenomenon. For all these cases of socialization special language performances are common and possible, and many socialization undertakings remain without result because language questions play a role which are either not understood or not solved. This has been emphasized in the research on language barriers and is often only a question of the specialized language of a certain discipline. In short, the research requirements as a proper topic in the discipline seem to me very important, and I would like to refer to some questions which have already occurred in our conversation. I believe, for example, that the need for fundamental research is larger than what has so far been done in fundamental research. The questions on the application of linguistic findings also form a gap. I may remind you of my remarks on 'simple technology'. I believe that modern investigations, more modern than I could do myself, on questions of cognition and of production of language are still a real problem. The same is true for the analysis of theory types. Further topics are those that were referred to before as interdisciplinary investigations. How are these tasks realized in contemporary linguistics ? - If one turns with such a point of view to the linguistics of today, one must say that

PETER HARTMANN

149

the cases of research requirements mentioned are on the whole hardly focused upon in reality and are also not integrated into a well-defined plan of a material nature. What is the position, for example, of transformational grammar in this context of requirements? I would formulate my impression here: as a typically system immanent and science immanent trend, it does not comply with those need situations that are predominantly extradisciplinarily oriented. For one, it does not at all appeal to them, and it does not or cannot do so; it is not really adequate for them either. The situation would be the same for descriptions of the linguistic generativity as, in an analogous global view, for the structuralist investigation. This is also the case for German language philosophy, which is somewhat in line with Heidegger,23 although there are already implications available concerning the linguistic function of the mind. One would probably refer to all this as 'academic interests'. Contemporary linguistics in Germany, however, must be viewed as rather differentiated as a whole; at this time increased concentration on areas of application can be observed. In addition to this need-oriented characterization, one can bring up science immanent points of view and one can then see the indisputable value of transformational grammar in the fact that it makes the quality of the descriptive tool a topic and that it has made formalizability accessible to a large extent. Consequently, this direction should not only be recognized as a methodological contribution, but also as something that can no longer be pushed aside. The special question of deep and surface structure, it seems to me, cannot be answered unequivocally. It is clear that one can establish deep structures; in my opinion it is true as well that it should be determined under what circumstances the concept of deep structure arises and the deep structure as such becomes recognizable ; for example, when derivative relations should be established in a certain manner. Within data processing, however, it becomes clear that the distinction between deep and surface structure cannot be evaluated as so oppositive, but rather that the surface structure appearances suffice to represent certain things satisfactorily in an equally analytic way and to record them on a machine. But is this distinction not a contribution to the solution of the cognition problem ? - Surely, that is the case, because the basis of the deep structure is a cognitively oriented basis. One can already see that in Hockett's representation of 'deep grammar'. 24 'Deep grammar' is the forerunner of deep structure and in fact the analysis of the cognitive kernel of sentences. The question was rather: does one need the concept of deep structure ? Does one need it perhaps for data processing, which is first of all not a cognitive business, is it? At the moment when one talks about deep structure, however, one uses a cognitive structure. That is right. This is again, however, the result of certain procedures which make sense for certain purposes and precede many statements or ideas. When the purpose changes, it is trivial to say that one does not need them any more. 23

Cf. H. Heidegger, Unterwegs zur Sprache (Pfullingen, 1959). Cf. C. Hockett, A Course in Modern Linguistics (New York, 1958); "Two Models of Grammatical Description", Word 10 (1954), 210-233. 24

150

PETER HARTMANN

What is, according to you, the actual value of structural linguistics in Europe as well as in the United States ? - The value of structuralism in the United States cannot be disregarded any more either. Structuring, as the main linguistic occupation, is considerably older than structuralism in the United States, also older than Saussure; there is, as far as I can see, no language oriented consideration without a structuring basis of some kind. Whether they were Assyrian vocabularies, Greek grammars, medieval ideas about language or grammar, this component is present everywhere. In this way, one could say more generally that the structuring of language occurrences which form a continuum is, if American structuralism is, in that respect, a late brand of a long tradition. Finally, in such a context, the considerations of Hjelmslev are also indispensable now because of the heavy formal and abstract tendency which was already recognizable in it. To say something general about structuralism in Europe is somewhat more difficult because there are clearly different directions, at least for Europeans. Thus, for example, dependency grammar, a peculiar structuralist development, is applied and further developed in several places. To provide an evaluation of German language philosophy is rather difficult. A new work is that by Josef Simon 25 which starts from the point of view of Kantian philosophy and which appeared to me to be quite good. Former reflections of Heidegger, on the other hand, are at the moment no longer in the linguistic discussion. It is questionable to me whether ideas such as the ones Lohmann 26 had in ethnolinguistics are taken up again; ethnolinguistics is hardly practised in Germany. Especially with relation to language philosophy, I have the impression today that on the whole not enough is happening, that there is anyway too little cooperation from the side of linguistics in order to be able to say what value those things have. Doubtless, they have their value in their domain but it is difficult for a linguist to determine whether they also have a value in their further use. With regard to linguistics in Germany, my opinion is that the questions which have been briefly sketched before as research requirements should be more strongly focused upon. One must consider here that the linguistic potential, to put it plainly, the research capacity, is limited in Germany. It is a fact, however, that more and more works are concerned with topics which lie in the direction of the applicability of linguistics. It is not worth mentioning them here individually because a fairly complete picture of the situation has been published recently. 27 Universität Konstanz Konstanz, West Germany August 26, 1972

25 26 27

J. Simon, Philosophie und linguistische Theorie (Berlin-New York, 1971). Cf. the journal Lexis, edited by J. Lohmann, from 1948. P. Hartmann, Zur Lage der Linguistik in der BRD.

GEORGE LAKOFF

George Lakoff, what is the way to do linguistics today? What is language as the object of linguistics ? - I take linguistics to be the study of natural language in all of its manifestations. This is a broad conception of the field, and I think it is an appropriately broad one. It includes not just syntax-semantics, phonetics-phonology, historical linguistics, anthropological linguistics, etc., which form the core of most academic programs in this country, but also the role of language in social interaction, in literature, in ritual, and in propaganda, and as well the study of the relationship between language and thought, speech production and perception, linguistic disorders, etc. One might object, of course, and propose that linguistics be considered a narrower discipline, on the grounds that one can only arrive at a coherent formal theory by limiting one's sights. I think it is true that many of the advances made in the period of transformational grammar, for example, could not have been made had Chomsky not limited the object of his theory to the study of something that does not exist in the real world, namely, the abstract set of rules of grammar (excluding meaning and use, and production and perception mechanisms) internalized in the mind of an ideal speaker-hearer in a homogeneous speech community. Perhaps the most interesting result to come out of transformational grammar is that such a limitation of the discipline is impossible - no coherent linguistic theory results from narrowing one's sights in this way. This is an interesting result, because it is not a priori true. In chemistry, for instance, there is a useful theory of ideal cases. It was thought that one might have been able to come up with a coherent theory of such a very limited domain in linguistics - 'ideal grammar'. The fact that the study of grammar ultimately had to take into account the study of meaning and use makes me wary of any artificial limitations of the domain of linguistics. For this reason, I would not presume to tell anyone how to do linguistics; I don't think there is any one way. It is important to distinguish between what one considers a profitable research strategy at a given point in history and what is the domain of the field. I think that from the late 1950s to the mid 1960s transformational grammar provided a very profitable research strategy for certain people, though not others. An enormous amount of knowledge was gained: but by the late 1960s transformational grammar had pretty much outlived its usefulness as a research strategy. By 1967

152

GEORGE LAKOFF

it was no longer possible for those working seriously in transformational grammar to maintain that there was a syntactic deep structure that was distinct from logical structure, that fully determined meaning, contained all lexical items, and was the input to all transformational rules. Which new research strategies were proposed at that time ? - Chomsky, in his extended standard theory, created a new and very different notion of deep structure, giving up any significant claims to its semantic relevance, but keeping it as a level prior to the application of any transformations and containing all lexical items. Such a deep structure, if it could be shown to exist, would be very close to suxface structure. Generative semantics, on the other hand, gave up the idea of deep structure altogether, letting logical structure play the role of underlying syntactic structure. While Chomsky tried to maintain the idea that syntax was independent of meaning and use, generative semanticists suggested the opposite. Of course, the two uses of the term 'syntax' here are not at all comparable, since they are meant to cover entirely different ranges of phenomena. In generative semantics, 'syntax' is taken to be the study of what strings of words can express what meanings in what context. In this respect, generative semanticists are not doing generative grammar. Generative grammar assumes that strings of words can be determined to be syntactically well-formed or ill-tormed in isolation, and sees as its goal providing a set of rules that can generate the well-formed strings. We reject the assumption that syntactic well-formedness in isolation is a viable concept. Generative semantics sees rules of syntax not as generating strings of words, but rather as generating relations between strings of words and what they mean relative to given contexts. It should be borne in mind that in adopting generative semantics as a research strategy, we are not attempting to provide a complete theory of language. For example, we are not attempting to account for the facts of speech production or speech perception or the use of language in ritual or in literature. Instead we are trying to provide a theory of a subpart of linguistics, the relationship between sentences and what they mean in limited sorts of contexts. We are operating under the gratuitous assumption that such a theory can be constructed without taking into account the actual processes of speech production and perception, among other things. I have the creepy feeling that such an assumption will turn out to be wrong, just as the generative grammarians were wrong in assuming that a coherent theory of syntax was possible without taking meaning and use into account. But for the present I think it is wise, at least for me, to stick to generative semantics, since it has turned out over the past six years to be very fruitful as a mode of inquiry and has not yet outlived its usefulness. At the same time I am glad that competent investigators are studying various other aspects of language that generative semanticists are not looking at, since such studies will bring us closer to the day when an integrated theory of language in all of its manifestations will be possible. I should also say that I do not think that theory construction and verification is the

GEORGE LAKOFF

153

only or even the most important mode of doing linguistics. Theorizing is more glamorous these days than doing careful descriptive work. I think that is unfortunate. Linguistic description is still an art, and is not likely to become a science for a long time to come. Unfortunately it is an art that has begun to die just at the time when it should be flourishing most. The reason is that it is still widely believed that linguistic descriptions of little-known languages should be formal and should follow some particular theory. But it has become clear in the past decade that no linguistic theory is anywhere near adequate to deal with most facts. What is wrong with formal descriptions is that they only allow for those facts that happen to be able to be dealt with by the given formalism. At this time in history, any description of a language that adheres strictly to some formal theory will not describe most of what is in the language. Moreover, as formal theories become outmoded, as is happening at an ever-increasing rate, descriptions of exotic languages made on the basis of those theories become increasingly less useful. I think the time has come for a return to the tradition of informal descriptions of exotic languages, written whenever possible in clear prose rather than in formal rules, so that such descriptions will still be useful and informative when present theories are long forgotten. In the past ten years the domain of known linguistic facts has spread far beyond the reaches of any foreseeable formal theories. In order to cope with these facts, we need all the help we can get. For this reason, it is important to recognize that there is no one particular 'way' to do linguistics. What there are are various research strategies, some more productive at present than others. You have to redefine the concept of linguistic competence and the opposition of competence and performance used by Chomsky, but also by structural linguists... - Chomsky has used the terms 'competence' and 'performance' in different ways at different times. The only consistent way in which I can understand his use of the term 'performance' is that he takes it to be a wastebasket for all the phenomena that cannot be accomodated by whatever theory he happens to be maintaining at a given time. Let me give some examples of the disparate ways in which he has used the term. Sometimes 1 he uses the term 'performance' to means what a person actually does, with 'competence' being whatever mental abilities enable him to do what he does. Call these concepts 'performance-1'and 'competence-1'. At other times, 2 Chomsky uses 'performance' to include perceptual strategies and psychological processing abilities. Since a set of perceptual strategies is not part of what one does, but is rather part of one's mental abilities, this notion, call it 'performance^', is part of competence-1. In performance-2, processing abilities and perceptual strategies are taken to be part of a human being's general mental abilities, rather than being part of some particular natural language, like Mohawk or Djirbal or Vietnamese. The abilities required to speak a particular natural language would come under the rubric of competence-2. 1 2

N. Chomsky, Aspects of the Theory of Syntax (Cambridge, Mass., 1965), p. 4. Ibid., pp. 10-13.

154

GEORGE LAKOFF

However, C h o m s k y also speaks o f particular languages as having "performance rules", f o r e x a m p l e , 3 h e s p e a k s o f f r e e w o r d o r d e r a s b e i n g d e t e r m i n e d b y "rules o f p e r f o r m a n c e " . T h i s t h e n is p e r f o r m a n c e - 3 , in w h i c h certain

not-clearly-specified

l a n g u a g e - p a r t i c u l a r rules - p a r t o f t h e g r a m m a r o f s o m e l a n g u a g e like L a t i n o r N a v a h o - are c o n s i d e r e d a s p a r t o f ' p e r f o r m a n c e ' rather t h a n ' c o m p e t e n c e ' . C o m p e t e n c e - 3 w o u l d t h e n c o v e r c e r t a i n l a n g u a g e - p a r t i c u l a r rules, b u t n o t others.

Where

c a n o n e r e a s o n a b l y d r a w t h e l i n e ? I find t h e p e r f o r m a n c e - 3 / c o m p e t e n c e - 3 d i s t i n c t i o n particularly h a r d t o c o m p r e h e n d . I f C h o m s k y t a k e s l i n g u i s t i c s t o b e t h e s t u d y o f linguistic competence, then taking competence to b e competence-3, the study o f w o r d free o r d e r ( w h i c h is n e v e r c o m p l e t e l y free, b u t h a s c o n s t r a i n t s v a r y i n g f r o m l a n g u a g e t o l a n g u a g e ) w o u l d not b e part o f t h e s t u d y o f linguistics. V e r y s t r a n g e i n d e e d . W h a t is stranger is t h a t C h o m s k y w o u l d , I b e l i e v e , c l a s s i f y m a n y o f t h e m o s t interesting p h e n o m e n a currently b e i n g s t u d i e d b y l i n g u i s t s a s p e r f o r m a n c e . W e c a n m a k e a brief list: R o s s ' w o r k o n f u z z y g r a m m a r ; 4 t h e w o r k o f L a b o v a n d o t h e r s o n v a r i a b l e rules; R o b i n L a k o f f ' s w o r k o n t h e r e l a t i o n b e t w e e n g r a m m a r a n d t h e u s e o f l a n g u a g e in s o c i a l a n d cultural c o n t e x t s ; 5 F i l l m o r e ' s w o r k o n d e i x i s ; 6 M o r g a n ' s s t u d y

of

s e n t e n c e f r a g m e n t s ; 7 t h e s t u d i e s o f literal v e r s u s i n d i r e c t m e a n i n g d o n e b y G o r d o n , m y s e l f , S a d o c k , H e r i n g e r , G r e e n a n d C o l e , 8 w h i c h is b a s e d o n t h e s t u d y o f s p e e c h a c t s b y Searle a n d i m p l i c a t u r e s b y G r i c e ; 9 w o r k o n n a t u r a l l o g i c b e i n g d o n e b y m y s e l f ,

3

Ibid., p. 127. J. Ross, "The Category Squish : Endstation Hauptwort", Papers from the Eighth Regional Meeting Chicago Linguistic Society 8 (further as CLS) (1972), 316-328; J. Ross, "A Fake N P Squish", in C. J. N. Bailey and R. Shuy (eds.), New Ways of Analyzing Variation in English (Georgetown, 1973) ; J. Ross, "Nouniness", in preparation ; G. Lakoff, "Fuzzy Grammar and the Competence/Performance Terminology Game", CLS 9 (1973). 5 R. Lakoff, "Some Reasons Why There Can't Be Any Some-any Rule", Language 45 (1969), 608-615 ; R. Lakoff, "Tense and its Relation to Participants", Language 46 (1970), 838-844 ; R. Lakoff, "If's, And's, and But's about Conjunction", in Ch. Fillmore and D. T. Langendoen (eds.), Studies in Linguistic Semantics (New York, 1971), 115-150; R. Lakoff, "Passive Resistance", CLS 7 (1971), 149-162; R. Lakoff, "The Pragmatics of Modality", CLS 8 (1972), 229-246; R. Lakoff, "Language in Context", Language 48 (1972), 907-927; R. Lakoff, "Language and Woman's Place", Language and Society (1973); R. Lakoff, "The Logic of Politeness : or, Minding your P's and Q's", CLS 9 (1973); R. Lakoff, "Questionable answers and answerable questions", in B. Kachru et al. (eds.), Papers in Linguistics in Honor of Henry and Renée Kahane (University of Illinois Press, 1973). 6 Ch. Fillmore, "Santa Cruz Lectures on Deixis" (1971, unpublished). 7 J. Morgan, "Sentence Fragments", in B. Kachru et al. (eds.), Papers in Linguistics in Honor of Henry and Renée Kahane (University of Illinois Press, 1973). 8 D. Gordon and G. Lakoff, "Conversational Postulates", CLS 7 (1971), 63-84; J. Sadock, "Queclaratives", CLS, 7 (1971), 223-231; J. Sadock, "Speech Acts Idioms", CLS 8 (1972), 229-239; J. Heringer, "Some Grammatical Correlates of Felicity Conditions and Presuppositions", Working Papers in Linguistics 11 (The Ohio State University Department of Linguistics, 1972); G. Green, "How to Get People to D o Things with Words", Georgetown Roundtable (Georgetown, 1972); G. Green, Semantics and Syntactic Regularity (Cambridge, 1973); P. Cole, "Conversational Implicature and Syntactic Rules", in C. Bailey and R. Shuy (eds.), New Ways of Analyzing Variation in English (Georgetown, 1973). 9 J. Searle, Speech Acts (Cambridge, 1969); H. P. Grice, "Logic and Conversation" (1967, unpublished). 4

GEORGE LAKOFF

155

Karttunen, McCawley, Horn, Dahl, and Keenan, including the study of hedges and fuzzy concepts being done by Zadeh and myself. 10 It is often thought that the principal difference between Chomsky and the generative semanticists resides in their differing conceptions of the relation between syntax and semantics. There are certainly great differences there, but the biggest difference has to do with the question of what is the scope of linguistics. We consider the work mentioned above as central to the study of linguistics, that is, as investigating important phenomena that any linguistic theory with claims to any adequacy at all must deal with. So far as I can tell, Chomsky seems to believe that these are all outside the scope of linguistic theory. In short, generative semantics is trying to come to grips with a much wider domain of facts than either Chomsky's standard or extended standard theories are set up to deal with. It is partly for this reason that Chomsky's claim that the two theories are notational variants is utterly crazy. How can two theories be notational variants if they are about two very different domains of facts ? In one sense the question of what one calls 'competence' and what one calls 'performance' is a piddling issue of terminology - the facts are the same, call them what you will. In another sense, though, it is a matter of critical importance, if, like Chomsky, one uses the terms 'performance' and 'competence' to characterize what kinds of facts you feel a linguistic theory should be responsible for. In the latter case, such a decision can have an effect on whether a given fact is a crucial counterexample, or 'merely a matter of performance' which can be brushed under the rug. Chomsky's shifting definitions of performance provide him with a rug big enough to cover the Himalayas. How to define then the relation between acceptability and grammaticality ? - I don't think that such a distinction makes sense. A number of concepts that were basic to transformational grammar, such as grammaticality, were more artifacts not real natural language concepts, but artificial concepts that Chomsky needed to make it look like that theory had a chance of working. As I have suggested elsewhere, 11 I don't think that one can in general say that sentences in isolation are grammatical or not. Instead one has to ask whether a given sentence can be paired with a given logical structure in a given context. If one views grammars as generating relations 10

G. Lakoff, "Linguistics and Natural Logic", in D. Davidson and G. Harman (eds.), Semantics of Natural Language (Dordrecht, 1972), 545-665; L. Karttunen, "Some Observations on Factivity", Papers in Linguistics 4 (1971), 1; L. Karttunen, "Presuppositions of Complex Sentences", Linguistic Inquiry 4 (1973); L. Karttunen, "Remarks on Presuppositions", in A. Rogers et al. (eds.), Performatives, Presuppositions and Implicatures (Center for Applied Linguistics, Washington D.C., to appear); J. McCawley, Selected Papers on Grammar and Meaning (Seminar Press, 1973); L. Horn, "On the Semantic Properties of Logical Operators in English" (UCLA Diss., 1972); E. Keenan, "Two Kinds of Presupposition", in Ch. Fillmore and D. T. Langendoen (eds.), Studies in Linguistic Semantics (New York, 1971), 45-54; G. Lakoff, "Hedges", CLS 8 (1972), 183-228. II G. Lakoff, "Presuppositions and Relative Grammaticality", in D. D. Steinberg and L. A. Jakobovits (eds.), Semantics. An Interdhciplinary Reader in Philosophy, Linguistics and Psychology (Cambridge, 1971), 329-340.

156

GEORGE LAKOFF

between sentences, contexts and logical structures, the notion 'grammatically' has no meaning. In its place there is the concept of the degree of well-formedness of triples of the form [sentence, logical structure, context]. Don't you needfor doing linguistics some sharp methodological notions and oppositions ? - I take it you are referring to the question of what constitutes empirical evidence for a given analysis, that is, what kind of data we deal with. Back in the days of transformational grammar, linguistic data was taken to consist of intuitive judgements of grammaticality. Everybody has given up on that idea, including Chomsky and his students, who, like those of us in generative semantics, found no way of distinguishing 'ungrammatical' from 'semantically anomalous' sentences. They maintained the distinction, though it seems to have lost any empirical status. Chomsky's dictum is: let theoretical considerations decide what is ungrammatical and what is semantically anomalous. Within the extended standard theory there is no way of deciding on empirical grounds whether a given phenomenon is to be handled by syntactic rules or interpretive semantic rules. Within generative semantics, we take as our primitive data intuitive judgements as to the degree to which a given sentence can have a given meaning in a given context. Actually, that's too brief to be entirely clear. Coded into the expression "have a given meaning in a given context" are such matters as: (1) When you use a given sentence to mean a given thing, are you being sincere or not, polite or not, formal or not, joking or not, etc. ? (2) Given a sentence and a fixed context what can the sentence mean in the context? If it can have more than one meaning, is one 'stronger' than another? Or more likely? Or more 'normal'? (3) Are certain sentences with certain readings limited to given types of discourses, e.g., answers to questions, astounded responses, stories, meek inquiries, etc. ? (4) What assumptions is a speaker making when he uses a given sentence to convey a given meaning in a situation? (5) What is the literal meaning of the sentence in a given situation, and what is 'conversationally implied' by the sentence ? This will give you some idea of the kind of data we are after. What do you mean by context: the behavioral, psychological, situational, or verbal context ? - Any aspect of context that interacts with rules of grammar. Previous discourse is obviously important. Answers to questions, for instance, require a knowledge not merely of the meaning of the question asked, but also various aspects of superficial form. Take an example: (1) Did you give a present to someone? a. Yes, Zelda. b. Yes, to Zelda. (2) Did you give someone a present? a. Yes, Zelda. b. *Yes, to Zelda.

GEORGE LAKOFF

157

The questions in (1) and (2) have the same meaning, but permit different answers because of differences in their superficial forms. Conceptual contexts, that is, the assumptions made by speaker and hearer, also interact with rules of grammar. For example, as Robin Lakoff shows,12 sentences with past and future tenses differ from corresponding sentences with the periphrastic equivalents of past and future. For example, in certain deictic constructions, past and future tenses but not their periphrastic equivalent, indicate present time but carry an assumption on the part of the speaker that the object in question was or will be within the perceptual field of the participants in the conversation. Consider the following examples: (3) That was a chipmunk; see, there he is again climbing that tree. (4) That'll be a chipmunk; wait till it comes out of the bushes. In both cases, the speaker is committed to the present truth of proposition, namely, that the object referred to is a chipmunk, and the tenses used reflect past or expected future appearance in the perceptual field of the participants. The so-called periphrastic tenses, used to and be going to, do not work this way: (3') That used to be a chipmunk; see, there it is again climbing that tree. (4') That is going to be a chipmunk; wait till it comes out of the bushes. (3') and (4') are not paraphrases of (3) and (4), if they can be used appropriately at all. She further showed that sequence-of-tenses rules are subject to the same pragmatic constraints. In another article of R. Lakoff 13 it was demonstrated that indications of the speaker's attitude toward a state or event were tied to the occurrence of fei-passives. For instance, a newspaper trying to maintain a stance of objectivity would use (5a) not (5b) in a newspaper story: (5a) Fred Snurdley was arrested yesterday on a marijuana charge. (5b) Fred Snurdley got arrested yesterday on a marijuana charge. (5b) would indicate that it was a bad thing to happen and hence express sympathy for Snurdley. For the same reason, taped TV programs use (6a), not (6b): (6a) This program has been pre-recorded. (6b) This program has gotten pre-recorded. (6b) would suggest that it was a bad thing for the program to have been taped. Bolinger14 had noted that assumptions made by the speaker some times motivated the choice between some and any. Klima 15 had assumed that the distribution of some 12

R. Lakoff, "Tense and its Relation to Participants". R. Lakoff, "Passive Resistance". 14 D. Bolinger, "Linguistic Science and Linguistic Engineering", Word 16 (1960), 374-391. 15 E. Klima, "Negation in English", in J. A. Fodor and J. J. Katz (eds.), The Structure of Language (Englewood Cliffs, 1964), 246-323. 13

158

GEORGE LAKOFF

and any was a syntactic phenomenon. He attempted to account for the phenomenon within syntax by postulating a syntactic feature [ + affective] to account for the distribution of some and any. Such examples were ignored by transformational grammarians until Robin Lakoff, in accord with her general program, suggested that the Bolinger-Klima approaches to the question be unified. She found further evidence that extra-linguistic assumptions affected the distribution of some and any, and proposed that such assumptions needed to be represented somehow in the statement of syntactic rules. Consider, for example, how sentences (7) and (8) below differ from (9a) and (9b): (7) If you eat any of the candy, I'll smack you. (8) If you eat some of the spinach, I'll give you a dollar. (9a) If you eat some of the candy, I'll smack you. (9b) If you eat any of the spinach, I'll give you a dollar. [any here is unstressed.] Note first of all, a subtle difference in meaning, or intention, in the first pair. Although they are syntactically parallel, (7) functions as a threat, (8) as a promise. That is, in (7) the speaker is warning the addressee that, if he does not comply with instructions, something will happen to him that he won't like; in (8), on the other hand, the speaker is suggesting that, if the addressee complies with instructions, something will happen to him that he will like. The first is negative in tone, the second positive. But there is nothing overtly present in either of the sentences that distinguishes them in this way, merely the implicit assumptions and their consequences : (a) Getting a smack is not good. Hence, it is not good to eat any of the candy. (b) Getting a dollar is good. Hence, it is good to eat some of the spinach. In Klima's examples, the clearer cases involved sentences where superficial negative (or non-positive) environments (e.g. negatives and questions) conditioned the occurrence of any. In such cases, it was possible to assign some and any to sentences by purely syntactic rules. But (3) - (6) show that, if we are to state a general principle to account for all uses of some and any, it must look beyond superficially present syntactic phenomena: we must admit to the discussion the implicit assumptions of the participants and the consequences of these assumptions. If the consequence is positive, we will find some; if negative, any. The conditions under which (9a) and (9b) can be appropriately used follow this principle. For instance, (9a) and (9b) are perfectly appropriate in contexts where (9a) constitutes a promise and (9b) constitutes a threat. Such an interpretation requires assumptions that are somewhat odd given the world as we know it, namely, that the addressee enjoys being smacked and hates monetary rewards. But given such assumptions, (9a) and (9b) would be appropriate. (9b) could also be a promise rather than a threat in case the speaker had negative expectations of the conditions being fulfilled. In

GEORGE LAKOFF

159

such a case a negative assumption could trigger the occurrence of any.16 Further studies following the program initiated by Bolinger and Robin Lakoff which have provided further analyses of pragmatic factors tied to grammatical processes were published in the meantime. 17 The importance of this work is that it shows that linguistic rules cannot simply be taken as having the function of distinguishing grammatical from ungrammatical sentences; grammar must also specify the conditions under which sentences can be appropriately used. A great many types of participants' assumptions in a conversation interact with rules of grammar. Some of the more startling ones are things that used to be considered purely part of speaking behavior, the placing of interjections, like uh, oh, ah, etc., in a sentence. James, 1 8 following up on work by Robin Lakoff, 1 9 has shown that even interjections such as these cannot be excluded from the study of grammar. He shows that they cannot just be randomly inserted into sentences. Rather their use is rulegoverned and specified by syntactic, semantic, and pragmatic conditions. She has shown that in order to determine (a) whether a given 'extralinguistic' particle is usable in a particular environment and (b) how it affects the meaning of the sentence, one must take into account such syntactic phenomena as Ross' constraints on movement transformations. She also shows that such particles interact differently with different types of idioms and negative polarity items, and that the nature of the interaction varies with the particle. The following are but a handful of James' examples: (10) (11) (12) (13) (14) (15) (16) (17)

John threw the ball, oh, up. *John threw his dinner, oh, up. I saw, uh, 12 people at the party. I saw, oh, 12 people at the party. *For some stupid reason, ah! not many people came to the party. I believe that Bill ate, oh, five cookies. *I believe the claim that Bill ate, oh, five cookies. *I regret that Bill ate, oh, five cookies.

The difference between (10) and (11) is due to the fact that throw up is in the first case composed of two semantically distinct units, a verb + directional particle, while in the second case it constitutes an indivisible idiom (it is equivalent to vomit). In (10), one might have several choices of direction in which to throw the ball, before settling on ' u p ' : the use of oh indicates this casting about among possibilities. But once one 16

Further examples along this line are given by A. Borkin, "Polarity Items in Questions", CLS 1 (1971), 53-62, and by L. Horn, "On the Semantic Properties of Logical Operators in English" (UCLA Diss., 1972). 17 Cf. the publications of R. Lakoff, mentioned in note 5; further G. Lakoff, "Presuppositions and Relative Grammaticality", A. Rogers, "Three Kinds of Physical Perception Verbs", CLS 7 (1971), 206-222; G. Green, cf. note 8; J. Lawler, "Generic to a Fault", CLS 8 (1972), 247-258; P. Postal, On Raising (Cambridge, Mass., forthcoming). 18 D. James, "Some Aspects of the Syntax and Semantics of Interjections", CLS 8 (1972), 162-172; D. James, "The Syntax and Semantics of Interjections" (U. of Michigan Diss., forthcoming). 19 R. Lakoff, "The Logic of Politeness: or, Minding your P's and Q's".

160

GEORGE LAKOFF

was aware of the general type of action described in (11), there is only one way to end the sentence: there are no other choices within the idiomatic framework. So oh is not possible here. The next pair illustrates a semantic difference between oh and uh: they are not mere meaningless particles, as many grammarians have stated, used merely to purchase time in speaking. (12) would be used if the speaker wanted to state an exact number (and knew this number), but it had slipped his mind for an instant. Oh (in (13)) would be used to mean, 'approximately', if the speaker were not certain of the exact number, or didn't care enough to give it. Thus, if I enter the room where the party is and find fourteen people present, I can accuse the speaker of (12) of misleading me, or being inaccurate at least, but not the speaker of (13). So these sentences are semantically distinct. The next sentence is of interest in that it shows that the occurrence of these particles is subject to syntactic or perhaps semantic constraints: they are not interjected randomly into sentences at will. When an adverbial phrase presupposing the truth of the main clause of a sentence is preposed, it cannot be followed by a particle, then the main clause. Finally the last three examples illustrate the interaction of strictly syntactic movement constraints and the presence of interjections. Interjections may occur inside a sentence embedded after believe, as in (15), but not in environments apparently similar, like (16) and (17). The reason, according to James, is that the embedded sentences in (16) and (17) form islands, in Ross' sense, but this is not the case in (15). Interjections reflecting the speaker's feelings obey typical island constraints. This shows that these interjections are governed by the same sorts of strict syntactic constraints that control movement rules, sequence-of-tenses changes, and other unquestionable syntactic rules. This rules out any possible claim to the effect that these parts of speech are in any sense 'performance phenomena', if by this we mean they are either (i) random, meaningless, and non-rule-governed, or (ii) part of general cognitive processing mechanisms and therefore outside the domain of linguistic rules, both language-particular and universal. Of course, one could always try to redefine performance to keep interjections and hesitation phenomena within its domain. What James' work shows is that any attempt to do so will make many principles of grammar, both universal and languageparticular, part of such a redefined 'performance', which would be a considerable change in the original sense of the term. Social context also interacts with grammar. For example, there are certain constructions in English which express polite requests and which express rude requests. Compare (18) and (19): (18) Can you take out the garbage? (polite) (19) You can take out the garbage, (rude) There are also idiomatic expressions that are rude and that yield grammatically illformed sentences when put in polite constructions: (20) You can take your methodology and shove it. (21) *Can you take your methodology and shove it?

GEORGE LAKOFF

161

You wrote that linguistic study is rooted in the study of human thought and culture. The domain of linguistics, so defined, is very large.... - That's right. I don't think one can describe, much less explain, linguistic rules unless one studies the uses to which language can be put. Transformational grammar tried and failed. One thing that one might ask is whether there is anything that does not enter into rules of grammar. For example, there are certain concepts from the study of social interaction that are part of grammar, e.g., relative social status, politeness, formality, etc. Even such an abstract notion as free goods enters into rules of grammar. Free goods are things (including information) that everyone in a group has a right to. What counts as free goods will, of course, vary from subculture to subculture. For example, in the counterculture in Berkeley, food is free goods; at a counterculture restaurant someone at the next table may ask you for a bite of your sandwich and refusal identifies you as being either out of the subculture, selfish or impolite. In other American subcultures the prices of household articles are free goods to friends; in others they are not. Someone from a subculture of the former sort might, upon walking into your house for the first time, say "Hey, that's a nice rug. What did it cost?" In that subculture, such a question would count as a compliment. In subcultures where prices are not free goods, such a question would be completely out of place. If someone wanted to know what your rug cost (say, because he wanted to buy one himself), he would have to say something like "May I ask you what that rug cost?" The form "May I ask you... ?" is used when the item asked for is not free goods, but when the speaker has no reason to believe that he is not welcome to it anyway. Note, incidentally, that this is not simply a matter of asking permission - one may not substitute "be allowed" for "may" in such questions. "Am I allowed to ask you what that rug cost?" or "Please give me permission to ask you what that rug costs?" would be totally inappropriate. Any adequate grammar of English would have to point out that may-questions of the above sort involve the notion of free goods. Now let's get back to the question of whether there are any concepts that do not enter into rules of grammar. The notion of political equality as opposed to social equality seems not to play a role. There seems to be no rule of grammar that I have heard of in any language that applies just, say, when both (or neither) of the speakers have the right to vote in national elections, or when one does and the other doesn't (unless, of course, there is a corresponding social equality or inequality). The sort of concepts discussed in political science as opposed to sociology seem not to play a role in rules of grammar. Has linguistics then to become sociolinguistics ? - It has been for a long time, except perhaps for the decade from 1957 to 1967, during which transformational grammar was dominant. Traditional grammars have always paid a great deal of attention to things like politeness, formality, status, etc., especially in languages like French and Russian, where the use of second-person

162

GEORGE LAKOFF

pronouns requires a fair amount of knowledge about social relations, and in languages like Japanese, where there are honorific particles. Anthropological linguists within this century have been especially careful to take note of such things. The idea may seem new to my generation, but only because we were brought up on transformational grammar. Is there no boundary between semantics and pragmatics, in your opinion ? - I think Richard Montague was on the right track when he suggested that the apparatus of model-theoretical semantics could be adapted to handle various so-called pragmatic phenomena. Given the fundamental notion of model-theoretical semantics, namely, satisfaction in a model, and the derived notion of logical consequence, one can deal with a great many phenomena that had previously been called 'pragmatic'. In my paper "Pragmatics in Natural Logic", 20 I suggest a way in which indexicals, speech acts, and conversational implicatures can be handled using just modeltheoretical semantics and transderivational syntax. What do you mean to be the scope of what you called'natural logic'' ? - Natural logic is the study of reasoning in natural language. As such it differs from classical logic in many respects. First, its scope is much broader. Classical logic concerned itself with concepts like and, or, if-then, not, every, and some. More recently, logicians have attempted to deal with a handful of other concepts, such as logical necessity, obligation, belief, knowledge, tenses, certain adverbs, many, few, etc. mostly in isolation as minor extensions of classical logic. A complete natural logic would have to deal with all of these concepts together, plus hundreds and perhaps thousands more - depending on how many primitive concepts natural language can be reduced to. In short, it is the full study of the conceptual resources of natural language. What is the importance of ordinary language philosophy in this respect ? - Even a cursory scanning of the linguistic literature of the past three or four years will show that ordinary language philosophy has had an enormous influence on linguistics. Strawson's pioneering work on presuppositions 21 has influenced a great many people and led to a considerable number of important studies. The work on speech acts by Austin and Searle and Grice's theory of conversational implicature 22 have also greatly expanded the range of linguistic studies. Incidentally, just about none of the linguistic studies that have come out of ordinary language philosophy have accepted the premises or claims of the ordinary language philosophers. What have 20

G. Lakoff, "Pragmatics in Natural Logic", in A. Rogers et al. (eds.), Performatives, lions, and Implicatures (Center for Applied Linguistics, Washington D.C., to appear). P. Strawson, Introduction to Logical Theory (London, 1951). 22 Cf. note 9. 21

Presupposi

GEORGE LAKÒFF

163

been taken over are the empirical observations, for example, Austin's observations about performatives. Model-theoretical semantics in the tradition of Tarski, Carnap, Kripke, and Montague 23 has also had a profound effect on linguistics in recent years. Chomsky and Katz, having been brought up more in a proof-theoretical than model-theoretical tradition, attempted to characterize meaning in purely combinatorial terms. That is, they assumed that whatever there was to be said about meaning could be said in terms of combinations of a finite stock of elements. This made it impossible for them even to begin to come to grips with problems of reference and coreference. Kripke's possible world semantics enabled generative semantics to begin dealing effectively with these problems. Moreover, purely combinatorial semantics cannot deal with fuzzy concepts at all. There is no way for combinatorial semantics to provide meanings for hedges like sort of, rather, pretty, etc., which map fuzzy concepts into new fuzzy concepts. Model-theoretical semantics, on the other hand, provides a natural way for dealing with both reference and fuzziness. Is the sentence, in generative semantics, the primitive unit of language, as in transformational grammar ? - No. In the theory of generative semantics as I have formalized it so far, the abstract objects generated are not sentences but quadruples of the form (S, LS, C, CM) where S is a sentence, LS is a logical structure associated with S by a derivation, C is a finite set of logical structures (characterizing the conceptual context of the utterance), and CM is a sequence of logical structures, representing the conveyed meanings of the sentence in the infinite class of possible situations in which the logical structures of C are true. But even this is inadequate. One must take into account much more than conceptual contexts (that is, assumptions of speaker and hearer). Rules of grammar also require that one take into account the stylistic type of discourse one is in. For example, there is a grammatical construction in English which can only be used in stories and not in conversation; and within a story, it can only be used to describe a setting. Consider sentence like: (22) Noon found Harry standing in front of the Blue Parrot Saloon. Compare (22) with: (23) At noon, Harry was standing in front of the Blue Parrot Saloon. They both have the same cognitive content, but (22) can only be used in a story while describing a setting. It is completely inappropriate in anything like an ordinary conversation. Somehow this fact must be represented in a grammar of English, 23

Cf. among others, S. Kripke, "Semantic Considerations on Modal Logic", Acta Phibsophica Fennica 16 (1963), 67-96; R. Montague, "Pragmatics and Intensional Logic", in D. Davidson and G. Harman (eds.), Semantics of Natural Language (Dordrecht, 1971), 142-168.

164

GEORGE LAKOFF

which means that one cannot view a grammar as merely generating sentences without any indication of the types of stylized discourses which they are restricted to. Is the logical structure of a sentence the same as its underlying grammatical structure ? - That is the basic assumption behind generative semantics. Of course, it requires more elaboration, since a sentence may have one logical structure in one context and another in another context. That is why we speak of generating quadruples of the form (S, LS, C, CM). What we are claiming is that sentences are not just paired with logical structures in isolation; what logical structures they are paired with, and the degree to which they are paired with them, depends upon context and upon constraints on possible conveyed meanings. Logical structures ought, incidentally, to be distinguished from semantic representations of the Katzian variety. One thing that I think my work on hedges and on conversational postulates shows is that there does not exist any such thing as a semantic representation, that is, a single combinatorial structure representing all elements of the meaning of the sentence. Literal meaning must be distinguished from conveyed meaning. And certain model-theoretical aspects of meaning cannot be represented combinatorily. What do you mean by the 'presupposition' of a sentence ? - The term 'presupposition' has been used (confusingly) to cover two very different concepts. A logical presupposition is a relation holding between two logical structures. We can further define an extended sense of logical presupposition as a relation holding between two surface sentences just in case the logical structure of one logically presupposes the logical structure of the other. A pragmatic presupposition is a relation holding between an individual and a proposition. This is the sort of presupposition that linguists usually talk about; perhaps there would be less confusion if we used the term 'assumption' or 'presumption' instead. The two notions are, of course, related, since speakers usually presume the truth of the logical presuppositions of sentences when they utter them sincerely.24 I assume that the study of logical presuppositions is part of the study of natural logic. I take pragmatic presuppositions as being handled by transderivational syntactic rules, which will take into account logical, syntactic, lexical, and phonological facts of the language in question. It should be noted that it is not at all obvious how, if at all, one is to draw the line between logical and pragmatic presuppositions. The problem of trying to sort out one kind from another is a little like the problem of telling when one should deal with lexical meaning by meaning postulates or by lexical decomposition. There are some clear cases, but mostly there is no evidence one way 24

For a discussion, see L. Karttunen, "Some Observations on Factivity", "Presuppositions of Complex Sentences"; "Remarks on Presuppositions". Further see R. Stalnaker, "Pragmatics", in D. Davidson and G. Harman (eds.), Semantics of Natural Language (Dordrecht, 1971), 380-397.

GEORGE LAKOFF

165

or the other. This situation may suggest that we are looking at things wrong. If generative semantics has an Achilles heel, it is here. How are the notions of reference and coreferentiality to be formalized in linguistics? - The same way they are formalized in formal semantics, namely, using the notion of a denotation at a point of reference. Recall that logical structures are taken to be model-theoretically interpreted. That means that there will be a denotation function assigning to each variable in logical structure at each point of reference a member of a universe of discourse. Two instances of the same variable will, of course, always be assigned the same referent at a given point of reference. Hence, two instances of the same variable will be coreferential. Reference and coreference for nominals in surface and intermediate structures will be defined using the notion of 'corresponding node' as defined in global grammar. To find out what a surface nominal refers to at a given point of reference, look at its corresponding node in logical structure and see what it refers to at that point of reference. Why was the introduction of transderivational rules in your theory necessary for grasping the contextual meaning ? - Before I answer your question directly, let me point out a couple of things. First, transderivational rules are not only necessary for dealing with contextual constraints in grammar; they are needed on totally different grounds, for example, cases where a derivation may be blocked because it could lead to ambiguity . 2 5 Note that ambiguity is not a property of a single derivation; we get ambiguity when there is more than one derivation for a given sentence. To block one derivation because of potential ambiguity we must know what the other derivation with the same surface string is. Rules that take into account the presence or absence of a morphological contrast are also transderivational in nature, since morphological contrasts do not usually occur in a single derivation. Moreover, rules of analogy are also fundamentally transderivational rules that have nothing whatever to do with context; they are needed on independent grounds. Next, it is impossible to build into a single derivation a specification of all of the contexts in which a sentence can be appropriately used. The reason is that there can be an infinite number of such contexts, and although they cannot be listed in a single logical structure (or deep structure or semantic representation or whatever), they can be finitely characterized by transderivational rules. For example, suppose we have a sentence, S, with logical structure P. Given a derivation relating S and P, we can ask with respect to what contexts it is well-formed. Suppose that it is well-formed with respect to every context X such that X u {P} | - Q , for some Q. That is, the set of logical structures X taken together with P entails Q. To state the general principle in finite terms we need only be able to specify finitely the form of Q, make reference 25

G. Lakoff, "Some Thoughts on Transderivational Rules", in B. Kachru et al. (eds.), Papers in Linguistics in Honor of Henry and Renée Kahane.

166

GEORGE LAKOFF

to the entailment relation, and use a quantifier over contexts X. Thus we can characterize by finite means an infinite class of contexts with respect to which a derivation linking P and S is well-formed.26 It should be clear that the rules that are needed to handle such cases must be transderivational. Assuming that S can have a literal reading (that is, CM = P), the grammar will generate an infinite class of quadruples of the form (S, P, X, P), where X u { P} ||- Q. A transderivational rule can generate an infinite class of quadruples of this form. But there is no way of listing in a single finite logical structure or semantic representation the infinite class of contexts that X can vary over. Is one of the purposes of generative semantics to discover meaning universals ? - The term 'meaning universals' has many senses, but the answer is yes in all cases. First, we assume that natural logic is universal, that is, it is intended to characterize the meanings of all the primitive concepts that occur in human language. In short, it is intended to characterize rational thought itself. Second, since we take rules of grammar as associating surface sentences with their logical structures, contexts for appropriate use, and conveyed meanings, universals of grammar are for us largely universals concerning the rules by which sentences are associated with their meanings. As should be clear, the terms 'universal of grammar' and 'meaning universal' encompass much more in generative semantics than they do in transformational grammar, since the subject matter of generative semantics is so much larger. Moreover, we assume that there is a relatively small and limited stock of rules of grammar that are possible in natural languages; that is, we assume that there are a great many substantive universals, rather than just universals of form, as Chomsky assumes for transformational grammar. In general, I find transformational grammarians relatively conservative on the issue of universals. Do you consider generative semantics only a new step in the development of transformational grammar, or is generative semantics a new theory dependent on a great number of inspiring theories ? - I guess that if you considered generative grammar a new theory as opposed to just an extension of Harris' transformational grammar, then you would have to consider generative semantics a new theory. The subject matter is much broader than the subject matter of generative grammar, the questions it seeks to answer are very different, the types of mechanisms it uses are different - so it would seem to be a new theory. Of course, historically it started out as an extension of generative grammar. This is clear in the first paper I wrote on the subject.27 But it has developed quite a bit since then. Just as Harris looked upon Chomsky as simply extending his theory (which is what 26

Examples of cases of this sort are given in G. Lakoff, "The Role of Deduction in Grammar", in Ch. Fillmore and T. Langendoen (eds.), Studies in Linguistic Semantics. 27 G. Lakoff, "Towards Generative Semantics" (M.I.T. Mechanical Translation Group, 1963, unpublished).

GEORGE LAKOFF

167

Chomsky started out doing), so Chomsky probably looks at us in the same light. But in both cases the differences are substantial. I think that what carried generative semantics furthest away from generative grammar was the work of Robin LakofF and Charles Fillmore on the role of context in grammar. I would say that their contributions in this direction, more than any others, split off generative semantics as a separate field. Since then, Ross' contributions to the study of fuzzy grammar and my work on fuzzy logic have made the division all that much wider. Our views of the role of context in grammar and the role of fuzziness in grammar and meaning distinguish us very sharply from generative grammarians. Can you say that the most important difference between Chomsky's theory of grammar and yours concerns the relation between syntax and semantics ? - No. That is one important difference, and historically it was at one time, around 1968, the most important difference, but now there are a number of differences of equal importance. First, the role of context, especially social context, implicatures, etc., in grammar is at least as important as the role of literal meaning. Second, the role of model theory has become very important. Thirdly, fuzziness, both in grammar and logic, is a matter of prime importance for distinguishing the theories. What attracted me initially to transformational grammar was that it seemed to give some insight into meaning. What is attracting our current students to generative semantics is not only that but that it seems to give one some insight into social interaction and the use of language. Do you agree with the terms 'autonomous syntax' versus 'semantic syntax' for characterizing Chomsky's theory as opposed to your theory? - I think it would be appropriate to call Chomsky's theory of syntax 'autonomous'. But I don't think that 'semantic syntax' quite gets at the heart of what we're doing nor does the term 'generative semantics'; but we started using it in 1963 and now we're stuck with it. Anyway, we are not just involved in the study of grammar and meaning, but in the study of the much broader relationship between language, thought, and culture. What's the kind of empirical evidence for making a choice between the two alternative theories: transformational grammar and generative semantics ? - That's a difficult question. In order to compare two theories at all on empirical grounds, one has to find enough common assumptions to make a comparison plausible. This is particularly difficult since the two theories make such different assumptions and cover such different subject matter. Basically, there are two approaches you can take, and both are exemplified in the literature of the field in the past five years. The first approach uses the fact that the subject matters of the two fields are very different, though they overlap in certain areas of classical syntax and the study of certain aspects of literal meaning. The idea is to show that you can state

168

GEORGE LAKOFF

some general linguistic principle if you take the wider subject matter into account, but that you cannot state any such general principle if you fail to. It is like arguing against someone who says that the subject matter of linguistics is the study of the first halves of sentences. You argue against him (if you really feel you have to) by showing that general principles emerge by studying whole sentences but not by limiting oneself to halves of sentences. I used this type of argumentation with respect to contrastive stress, relative pronouns, and anaphoric expressions 28 and others, with respect to polarity items. 29 Robin Lakoff's "Language and Woman's Place" ends with a particularly effective argument of this sort. 30 The second approach is to try to find some area in which the two theories cover the same subject matter and make the same assumptions - and what is most important of all, where all parties agree on how to interpret the crucial facts. This is particularly difficult, since the assumptions made by theories are so very different and since adherents of the two theories can very often choose to interpret the facts differently. If you take this approach you can never find any empirical evidence that will absolutely once and for all decide between the two theories to everybody's satisfaction. All you can do is provide evidence relative to certain assumptions and to a certain interpretation of the data. Then you can convince people who agree with your assumptions and interpret the data the same way you do. Many of the disputes between generative grammarians and generative semanticists are of this sort. Neither ever convinces the other, but people in the audience who are willing to accept one or the other's assumptions and interpretation of given data will be convinced. Most of the arguments about lexical decomposition fall into this second range. From 1963 to 1969, when it seemed there was a pretty much agreed upon set of assumptions shared by everyone in generative linguistics, it seemed that the lexicon was the place where we could most likely show that transformational grammar was wrong. If lexical items had to be syntactically decomposed into semantic elements, then there could be no independent syntax in the old sense. That is why so much of the argumentation centered on lexical decomposition. It was one area where there seemed at that time to be enough shared assumptions so that one could find empirical evidence against transformational grammar. As it turned out, all that one could frn-a was evidence relative to certain assumptions. For example, I and many others had always assumed that if anything was part of syntax, as opposed to semantics if they could be separated, then number agreement was. In other words, I assume that a violation of number agreement would lead one to a syntactically ill-formed sentence. For example: (24) I are tall. I had taken it for granted that if anything was ill-formed syntactically, sentences like 28

G. Lakoff, "Presuppositions and Relative Grammaticality"; "Linguistics and Natural Logic". R. Lakoff, "Some Reasons Why There Can't Be Any Some-any Rule"; A. Borkin, "Polarity Items in Questions"; L. Horn, "On the Semantic Properties of Logical Operators in English". 30 R. Lakoff, "Language and Woman's Place". 29

GEORGE LAKOFF

169

(24) were. One could imagine why Perlmutter's observation that number agreement in Spanish had to take semantics into account was of interest to me. It become the basis of an argument to the effect that one had to decompose a lexical item like parents into its component parts in order to account for number agreement by a single general rule. 31 If one accepted number agreement as a rule of syntax, that is, if one took (24) to be syntactically ill-formed, then Perlmutter's example showed that generative semantics was right and transformational grammar was wrong. However, at a conference at the University of Texas in 1969, Chomsky was asked about this case, and he replied that he saw no reason to think that (24) was syntactically illformed as opposed to being semantically anomalous. He simply chose to interpret the data differently. What this comes down to is that if you think that number agreement is a rule of syntax and that (24) is syntactically ill-formed, then rationally you should believe in generative semantics. If you don't assume that number agreement is a syntactic phenomenon, then Perlmutter's facts prove nothing. Since Chomsky can interpret the data to suit his theory, no absolute arguments along these lines are possible. However, it is still worthwhile to look for relative arguments, since they do force generative grammarians to ever crazier positions. Is the existence of grammatically ill-formed sentences an argument against your viewpoint ? - No. There are various types of shallow and surface constraints that do not in any direct way involve semantics. The claim that syntax is not autonomous is not the claim that there are no autonomous rules of grammar. Rather it is the claim that not all rules of grammar are autonomous. As it happens, most of them aren't. This point is often confused. Autonomous syntax claims that all rules of syntax are independent of meaning, context, and use. To refute the claim, you only have to find one that isn't. One needs in linguistics a logic where truth can be a matter of degree... - Yes. Natural language concepts are fuzzy; therefore natural logic must be a fuzzy logic - that is, a many-valued logic, perhaps even a continuous-valued logic. What I have called 'hedges' show this very clear. Expressions like sort of, pretty much, rather, strictly speaking, loosely speaking, technically, regular, par excellence, etc., affect truth values in a way that can only be described adequately if one assumes a fuzzy logic. Sort of, for example, raises intermediate values and lowers extreme values. There is even a hedge that requires one to have fuzzy presuppositional logic. The hedge 'to the extent that it makes sense to say that...' raises intermediate nonsense values and lowers extreme nonsense values.32 31

For the argument, see G. Lakoff, "On Generative Semantics" (1969), distributed by Indiana Linguistics Club, Boomington, Ind. 32 For details, see G. Lakoff, "Fuzzy Grammar and the Competence/Performance Terminology Game".

170

GEORGE LAKOFF

Language is not only rooted in thought but also in culture. Does linguistics have to have a specific sociological theory ? - Certain theories of social organization may be consistent with the linguistic data and others inconsistent. Linguistic data may very well, at some time in the future, enable one to decide among alternative theories of social interaction - at least for those phenomena that are reflected in linguistic structure. A part of the culture from which language arises is the ideological environment. Do you need in linguistics also a theory of the current ideologies ? Can one say that there is a link between politics and linguistics ? - Not on the theoretical level, so far as I have been able to tell. There is as yet no evidence that political concepts are reflected in linguistic structure. Teaching linguistics these days is not without some indirect - very indirect - political consequences. In linguistics, as in politics, much of the relevant data to support or refute many claims are available to the average person. In linguistics, it is in your mind and all you have to do is train yourself to recognize it. In politics, it is all around you, in the newspapers and on TV. Again you just have to be trained to recognize it. Just about any beginning linguistics student, with some careful thought, can in an afternoon think up enough crucial examples to show the inadequacy of our most sophisticated current theories. Similarly any citizen of average intelligence can pick out many of the lies that his government tells him. The thought processes are not all that different, though the subject matter is. Any beginning linguistics student will discover with a little thought that men of great stature in the academic establishment, even very bright ones like Chomsky, can be wrong on just about every issue. It makes one wonder about the 'experts' who are running our governments. If what you are getting at is the question of whether there is any link between Chomsky's politics and his linguistics, I would have to say no. Someone with William Buckley's political views could hold Chomsky's linguistic views without inconsistency. Nor is there anything politically revolutionary about the content of transformational grammar - if anything, it is reactionary today. I have seen some pretty stupid things written about the relationship between linguistics and politics. For example, I read in one report of the French student uprising in 1968 that linguistics became an issue: structuralism was identified with institutional rigidity and transformationalism with change. No one who knows anything about the actual content of structuralist and transformationalist theories could believe any such thing.

Transformationalism as a rationalism is usually opposed to structuralism as an empiricism or a behaviorism... - That is the typical line you get from reading most books on transformational grammar. I think it's false. Chomsky's theory about the organization of language, that is, that there are deep structures, transformations, etc., is consistent with strict

GEORGE LAKOFF

171

empiricism; while structuralist linguistic theories about the organization of language (i.e., that there are phonemes, etc.) are consistent with rationalism. Chomsky's argument for rationalism is based on the existence of complex linguistic universals. Chomsky always cites examples of putative universals from transformational grammar, but the fact is that just about every other theory of grammar that has ever been seriously proposed has, either implicitly or explicitly, incorporated claims for extremely complex and sophisticated linguistic universals. This is true of structural linguistics, stratificational grammar, tagmemics, Montague grammar, generative semantics, etc. In fact, contrary to what Chomsky suggests, the most extensive studies of complex linguistic universals have been carried out within the framework of structural linguistics. The classic works of the European structuralists Trubetzkoy and Jakobson in phonology and of Joseph Greenberg in American structuralist syntax have been the foundation for all of the more recent (and less extensive) studies of universals done in the tradition of transformational grammar. Chomsky's claims in favor of rationalism over behaviorism do not rest upon his theory of transformational grammar being right and structuralism being wrong. One can make exactly the same argument using the structuralist universals instead, since the universals discovered in structural linguistics are more than complex enough for the purpose of the argument. One should also be aware of the limitations of Chomsky's arguments for innateness. Chomsky has claimed that people possess innately not merely general learning mechanisms, but a specifically linguistic innate faculty. His argument is of this form: there are complex linguistic universals that everyone learns uniformly. There are at present no general learning theories that can account for this. It is hard to imagine what any such theories could be like. Therefore, it is plausible to assume that there can be no such theories. The argument is fallacious. Nothing follows from a lack of imagination. What Chomsky has shown is that either there is a specifically linguistic innate faculty or there is a general learning theory (not yet formulated) from which the acquisition of linguistic universals follows. The former may well turn out to be true, but in my opinion the latter would be a much more interesting conclusion. If I were a psychologist, I would be much more interested in seeing if there were connections between linguistic mechanisms and other cognitive mechanisms than in simply making the assumption with the least possible interest, namely, that there are none. Chomsky has characterized structural linguistics as being fundamentally behavioristic and concerned solely with taxonomy. This is a misleading view of a broad, diverse, and interesting field, which happened not to be very good at dealing with syntax, and which showed little if any interest in formalized theories. Chomsky's teacher, Zellig Harris, did happen to be an extreme case of a behavioristically-oriented taxonomist. Bloomfield and Hockett, in their theorizing moods, also fit the mold, though one can argue that they did not always adhere to their theories in their linguistic analyses. Though these were prominent structural linguists, they were by no means typical of the wide range of European and American structuralists, either in their interests or in their commitment to behaviorism. Distinguished structuralists like Boas,

172

GEORGE LAKOFF

Sapir, Jakobson, Pike, Weinreich, Bolinger and Greenberg never had much, if any, commitment to behaviorism. Their interests and their linguistic theories ranged far beyond mere taxonomy to such areas as linguistic universals, the relation between language and culture, dialectal variation, cross-linguistic interference, ritual language, poetics, and much much more. When transformational grammar eclipsed structural linguistics, it also eclipsed many of these concerns, much to the detriment of the field. Chomsky seems to have shared Harris' commitment to behaviorism to some extent even after developing his own theory of transformational grammar. In Chomsky's earliest book, "The Logical Structure of Linguistic Theory" (which was distributed in mimeoed form but was never published), Chomsky spends a good deal of time discussing the possible development of "operational procedures" for the validation of grammars. He cites the "pair test" favorably as "a thoroughly nonsemantic operational device" and ends Chapter I by suggesting a program of "formulating behavioral criteria to replace intuitive judgments". 33 Though Chomsky rejected Harris' ideal of using behavioral criteria for discovering grammars, he, at that point in his career, did not reject the idea of using behavioral criteria for validating grammars. How to define Chomsky's dependence upon his predecessors? Is it possible that he is more empiricist than he would like to be ? - I would say that, of contemporary linguists, Chomsky is among the more empiricist linguists. He is very much wedded to that period of linguistics represented by Harris and Bloomfield, in the sense that he is still principally interested in accounting for distributions of formatives in surface structures without regard to meaning. This was one of the principal characteristics of empiricist linguistics. He has given up working with corpuses, but from our perspective his goals are not all that different from Harris' goals. It is ironic that his own work has served to move people away from his views of what linguistics is all about. People got interested in transformational grammar not because it gave them a better way of accounting for the distribution of elements in surface structures, but because it gave them a way to approach the study of meaning. Transformational analysis seemed to be a way to get closer to logical structure - that was what was attractive about it for most people. Had early transformational analyses not been semantically revealing, no one would have cared about transformational grammar at all. Nowadays students are interested in generative semantics because it is a way for them to investigate the nature of human thought and social interaction. Somehow, the study of the distribution of surface elements as an end in itself couldn't be more boring to most people, including me. Chomsky was extraordinarily dependent on his teachers for his intellectual development. Most of his early linguistic analyses are taken directly from Harris, as is the idea 33

N. Chomsky, "The Logical Structure of Linguistic Theory" (Cambridge, Mass., 1955), I-36e and 1-59.

GEORGE LAKOFF

173

of transformations. The idea of evaluation metrics was taken over directly from Nelson Goodman. As much as Chomsky has rebelled against the empiricist views of Harris and Goodman, he has, I believe, still essentially retained their views of the relationship between meaning and the study of natural language. Is transformational grammar essentially a 'Cartesian linguistics'' ? There seems not to have been any such thing as 'Cartesian linguistics'. Chomsky claims in Cartesian Linguistics34 that Cartesian rationalism gave birth to a linguistic theory like transformational grammar in its essential respects. He bases his claims on the Grammaire Générale et Raisonnée by Antoine Arnauld, a disciple of Descartes', and Claude Lancelot, a language teacher, published in 1660. The Grammaire Générale followed a series of other grammars by Lancelot, the most extensive being his Latin grammar. Chomsky never checked out his Latin grammar (an English translation of which was in Widener Library) but Robin Lakoff did, and published her findings in Language.35 She discovered that in the introduction Lancelot credited all of his interesting findings to Sanctius, a Spanish grammarian of the previous century whose work antedated Descartes by half a century. Checking into Sanctius, she found that Lancelot was not being modest. He had indeed taken all of his interesting ideas from Sanctius. In short, what Chomsky called 'Cartesian linguistics' had nothing whatever to do with Descartes, but came directly from an earlier Spanish tradition. Equally embarrassing for Chomsky's claims is the fact that the theories of Sanctius and the Port Royal grammarians differ from the theory of transformational grammar in a crucial way. They do not acknowledge the existence of a syntactic deep structure in Chomsky's sense, but assume throughout that syntax is based on meaning and thought. Chomsky has steadfastly opposed this position from his earliest works straight through to his most recent writings. The publication o/Syntactic Structures in 1957 had been called a revolution in linguistics because grammar became then generative and transformational. What does it mean for you that grammar has to be generative ? - Transformational grammar was started by Harris, not Chomsky, and most of the actual linguistic analyses that you find in Syntactic Structures were present in Harris' work; the linguistic content is not that much different from what you find in Harris. So far as generative grammar is concerned, 'generative' simply means 'complete and precise'. The idea is that a grammarian formulates formal rules that characterize just what is and what is not in the language. The reason that such rules are formulated as operations rather than as static statements is that Chomsky got the idea from recursive function theory. I have called grammars of this sort production grammars, taking the term from Post productions, which are abstract directional operations on strings. 34

N. Chomsky, Cartesian Linguistics (New York, 1966). R. Lakoff, "Review oî Grammaire Générale et Raisonnée", Language 45 (1969), 343-364. Also, H. AarsIefF, "The History of Linguistics and Professor Chomsky", Language 46 (1970), 570-585.

35

174

GEORGE LAKOFF

Following ideas of McCawley, I have suggested replacing production grammars with what I have called well-formedness grammars. Well-formedness principles are static in nature, rather then being directional operations. Well-formedness grammars can do everything transformational grammars can do and more, since they also provide a natural way for formalizing global, transderivational, fuzzy grammars. In generative grammars it was assumed that a grammar generated sentences. In generative semantics it is assumed that a grammar generates quadruples of the form (S, LS, C, CM), where S and LS are linked by a derivation.36 What is the right sense of 'generative'1 in generative semantics ? - 'Complete and precise'. As I mentioned, 'generative semantics' is not a very accurate descriptive term for what we are doing. When I first used the term back in 1963, it was an amalgam of 'generative grammar' and 'interpretive semantics', since I was claiming then that there should be no line drawn between syntax and linguistic semantics. In those days, I even assumed that transformational grammars were basically correct. I didn't give up entirely on transformational rules until 1969. Is Chomsky's conception of generativity in language not linked with his more general and sometimes ideological idea of creativity ? - Chomsky has spoken of 'the creative aspect of language use\ by which he means that people can make up new sentences in new situations. This is a completely new and very strange use of the word 'creative', since it has nothing whatever to do with creativity in the ordinary sense of the word. There is nothing in transformational grammar that accounts for human creativity or that even pretends to do so. All that transformational grammar does is provide a recursive mechanism for generating sentences. There is nothing 'creative' about this. It is like constructing a computer program to do arithmetic. The program could perform an infinity of arithmetical operations, but no one would say that it accounted for mathematical creativity. Before concluding, some few more technical questions. Chomsky argues against generative semantics that it is only a notational variant of his extended standard theory... - That is a very strange thing for him to say, since he also thinks generative semantics is wrong. If generative semantics were only a notational variant of the extended standard theory, then by claiming that generative semantics is wrong, Chomsky would be claiming that his own theory was also wrong. There is an air of contradiction here, to say the least. Generative semantics is very very far from being a notational variant of the extended standard theory - and I think it takes a considerable amount of chutzpah for Chomsky to even suggest that it is. If you ask whether two theories are notational variants, the very first question you ask is whether they cover the same subject matter. As 33

Cf. p. 163 of this book.

GEORGE LAKOFF

175

was pointed out above, they don't. Just consider the principal areas of reasearch in generative semantics these days: Natural logic (that is, the study of human reasoning) including logic of fuzzy concepts, and the model-theoretical interpretation of logical structures; Pragmatics, including the appropriate use of language in context, especially social context, performatives, speech acts, implicatures, conveyed meaning in general, deixis, discourse types and styles of speech; Fuzzy grammar, including degrees of category membership and degrees of rule application. To my knowledge none of these is even part of the subject matter of the extended standard theory. So far as I have been able to tell, Chomsky and other lexicalists don't even consider these things to be part of the study of the structure of language; they seem to have arbitrarily defined them as being outside linguistics, they seem to have swept them under the rug of 'performance'. Since 1969, these have been the principal areas of study in generative semantics. They are not part of the subject matter of the extended standard theory. Since generative semantics and the extended standard theory don't even come close to having the same subject matter, they cannot conceivably be notational variants. In fact, it is hard to imagine how anyone who has kept up with developments in generative semantics since 1969 could even make such a claim. It has been suggested to me (by Arlene Berman, personal communication) that there might be one way of interpreting such a claim. Suppose we limit ourselves to just that subject matter where generative semantics and the extended standard theory overlap. Could it not then be the case that, only with respect to that subject matter, generative semantics and the extended standard theory are notational variants? This would be like suggesting that transformational grammar and phrase structure grammar are notational variants, if you just ignore all transformational phenomena. The question misses the point of bothering to do generative semantics, namely to account for that subject matter. But in addition, in this case, I don't even think the question makes any sense. The subject matter of the extended standard theory is the distribution of morphemes and the relation between surface sentences and literal meanings (without any model-theoretical interpretation); its assumption is that this is a coherent subject matter. But one of the principal claims of generative semantics is that this is not a coherent subject matter; instead, the distribution of morphemes depends upon various aspects of context, conveyed meaning, and the model-theoretical interpretation of logical structures. If one cuts out context, conveyed meaning, and model-theoretical interpretation from generative semantics, one does not get a coherent subtheory. These arguments concern the non-similarity of the scope of the two alternatives. But even the conceptual apparatus of the two theories is very different, at least apparently... - The theoretical mechanisms are very different, indeed. The extended standard theory has phrase structure rules, transformations, and surface (and deep) interpretive rules - which are not yet specified as to what their outputs are or what their form is. Generative semantics has correspondence rules, global rules, transderivational

176

GEORGE LAKOFF

rules, and principles of model-theoretical interpretation for logical structures (which are taken to be universally given). There is nothing like a one-to-one correspondence between the mechanisms of the extended standard theory and generative semantics. Moreover the grammatical categories are very different and cannot be matched up in anything like a one-to-one fashion. The extended standard theory has primitive categories like N, A, V, N, A, V, N, A, V, AUX, M(odal), DET(erminer), Sentence adverb, COMP, etc. It also uses syntactic features; in fact, it requires an infinite number in order to avoid having to use global rules. 37 For us, the primitive grammatical categories are the categories of logical structures: S, PRED(icate), ARG(ument). There are secondary (nonprimitive) categories which are defined in terms of primitive categories and global properties of derivations; membership in them is taken to be a matter of degree. 38 Because secondary categories are fuzzy, there is no possible one-to-one mapping between either the primitive or secondary categories of generative semantics and the categories of the extended standard theory. If the formal mechanisms are different and the categories are different and incapable of being set in a one-toone correspondence, then how could the theories possibly be notational variants ? For two linguistic theories to be notational variants, at least the following would have to be the case: (a) they would have to have the same subject matter; (b) their formal mechanism would have be able to be set in a one-to-one correspondence; (c) their grammatical elements would have to be set in a one-to-one correspondence; (d) there would have to be a one-to-one correspondence between their linguistic analysis of particular sentences, for all natural languages - in other words, they would have to make the same claims about every sentence of every natural language. We have seen that (a), (b), and (c) do not hold. It should be obvious that (d) also does not hold. In fact, the theories are so very different that I cannot even imagine one sentence of any language for which the theories provide the same analyses, or for which they even come close. Certainly the burden of proof is on someone who wants to claim that the theories are notational variants. He would have to show, at the very least, that (a)-(d) hold. No one who has made such a claim has tried to undertake such a demonstration. All that those who make such claims have done is to draw little diagrams with boxes and arrows to represent generative semantics and the extended standard theory, and to show that one can relabel the boxes to suit one's fancy. At this level of vagueness any theory can be made to have the same little-boxstructure as any other theory. But until you look at the details of the proposals and at the very least satisfy (a)-(d) above, drawing all the little boxes in the world won't prove a thing.

37

See G. Lakoff, "The Arbitrary Nature of Transformational Grammar", Language 48 (1972), 76-87. 38 See G. Lakoff, "Fuzzy Grammar and the Competence/Performance Terminology Game".

GEORGE LAKOFF

177

Another general remark made by Chomsky about generative semantics is that this theory is too powerful... - First, if you put this claim of Chomsky's together with his notational vaiiant claim, he turns out to be claiming that his own theory is too powerful, since it would be in one-to-one correspondence with a theory that was too powerful. Luckily for Chomsky, the theories are not notational variants. Secondly, with respect to weak generative capacity (which is about as uninteresting a subject as I could possibly imagine for a linguist) Peters and Ritchie have shown that transformational grammars can be made to mimic any Turing machines; in other words, it is impossible for there to be a theory which is more powerful than transformational grammar so far as weak generative capacity is concerned. The 'too-powerful' issue is usually directed at global rules. The only proposal to handle the same phenomena with nonglobal rules has been the Baker-Brame proposal, which would permit grammars to have an infinite number of grammatical categories. 39 That is hardly a proposal that reduces descriptive power. Moreover, the same people who claim that global rules are "too powerful" are more than willing to admit surface interpretation rules with no constraints whatsoever on them. Moreover, it usually forgotten that only global rules of a very restricted sort have been proposed. Just as transformational grammar admitted not every conceivable type of mapping from trees into trees but rather only a limited number of types of operations, so only a small number of types of global rules have been proposed. The real problem with global rules is not that they are too powerful, but that they are too weak. They appropriately handle a certain range of phenomena, but they are entirely inadequate to deal with most natural language phenomena, just as all other types of rules proposed so far have been. Anyone who believes that "you can do anything with global rules" ought to sit down and try sometime. The problem with all current theories is that they are just too weak to deal with most linguistic phenomena. Anyone who wants to convince himself of this should read the collected works of Robin Lakoff, Charles Fillmore and Dwight Bolinger. Can I ask you the explanation of one of your statements that I consider as central for your conception of language, namely, that " the form of language cannot be studied independently of its function"... - As I mentioned above, recent results indicate that the syntactic form of sentences is not independent of the meanings that they convey in context. In trying to account even for the distribution of morphemes, one must take into account not only the literal meaning of the sentence but also what you are communicating indirectly and how you are doing it; the function of the utterance in terms of communicative interaction cannot be ignored. One must consider both the expressive and communicative functions of language at the same time. 33

C. Baker and M. Brame, "Global Rules: A Rejoinder", Language 48 (1972), 51-75.

178

GEORGE LAKOFF

Modern linguistics was always operating with strong dichotomies. Is it your explicit purpose to transcend all these boundaries and distinctions ? - Yes. In fact, that is one of the most interesting things coming out of generative semantics. We have found that one cannot just set up artificial boundaries and rule out of the study of language such things as human reasoning, context, social interaction, deixis, fuzziness, sarcasm, discourse types, fragments, variation among speakers, etc. Each time we have set up an artificial boundary, we have found some phenomenon that shows that it has to be removed. That is not to say that there are no bounds on the study of linguistics. I only suggest that at this point in history the boundaries are disappearing daily, and one should not be too surprised if the domain of the field continues to expand. University of California at Berkeley Berkeley, Calif., U. S. A. October 6, 1972

SYDNEY M. LAMB

Sydney Lamb, you wrote a paper under the title "Epilegomena to a Theory of Language"-1 Is that an explicit allusion to Hjelmslev's Prolegomena to a Theory of Language ? 2 - Yes, it is, definitely. That article was a review article on Hjelmslev's Prolegomena. At about the same time, I wrote an article called "Prolegomena to a Theory of Phonology", 3 which was another allusion to Hjelmslev's work. I was and I still am a great admirer of Hjelmslev.

Can you say something about the substantial influence of Hjelmslev on your work ? - Well, he has been perhaps the greatest single influence on my work. I studied Hjelmslev with Frank Whitfield, the person who translated Hjelmslev's Prolegomena. When I was doing my graduate work in linguistics at Berkeley, I took Whitfield's course there in Russian morphology and it turned out to be largely a course in Hjelmslev, so I was exposed to him at an early stage of my career. And I was learning, like other students in American universities at the time, various versions of neoBloomfieldian structural linguistics. But at the same time, unlike most linguists, I was also exposed to Hjelmslev. Then, several years later, I undertook to write a review of the Prolegomena and I studied it over in detail a second time and I found that I was further influenced by him; there were things I hadn't fully appreciated earlier, and there were other things I had appreciated but had forgotten. This was mainly in the fall of 1964. At the same time 1 had begun to work out the relational network notation which is now associated with the stratificational grammar. It was a happy circumstance that I was doing both of these things at the same time, because I kept being impressed by Hjelmslev's view that the linguistic system is nothing but a system of relationships. I had always found that idea very attractive but I hadn't fully grasped its importance and significance until 1964. Until that time I had been working with

1

S. Lamb, "Epilegomena to a Theory of Language", Romance Philology 19 (1966), 531-573. L. Hjelmslev, Prolegomena to a Theory of Language (translated by J. Whitfield) (Madison, Wisconsin, 1961). 3 S. Lamb, "Prolegomena to a Theory of Phonology", Language 42 (1966), 536-573.

2

180

SYDNEY M. LAMB

relationships among linguistic elements, similarly to what Hockett did,4 and I was viewing the linguistic system as a system consisting of various kinds of linguistic elements together with the relationships between them. But as I worked through the details of this - under the influence of Hjelmslev - 1 discovered that every one of these linguistic elements was nothing but the intersection of various relationships. Is it true that Hjelmslev, say, in the early sixties, was unknown in United States ? - 1 would even say that he is still quite unknown. People have heard of him of course, I mean everybody knows the name Hjelmslev. But I find that very few people have known very much about his work. Whitfield of course was very influential in making it known in this country and there have also been a few articles by others which have had some minor influence in making him known. I don't know whether my "Epilegomena" has helped or not. So, I would say everyone knows Hjelmslev did something important, but I think relatively few American linguists know much about just what he did. In his theory of stratification of language, Hjelmslev distinguishes four strata based on the opposition expression/content and substance\form. Is that very similar to your opinion ? - Well, I would consider it a good starting-point. In my Epilegomena, I argued that the distinction between expression and content was not enough, that Hjelmslev's content-form is a conflation of what ought to be recognized as three distinct strata, which could be called morphemic, lexemic, and sememic. I now think, however, that the boundary between sememic and lexemic is more important than that between lexemic and morphemic, so (depending on how one defines the scope of the stratum) one might say that Hjelmslev's content form needs to be split into just two strata, the sememic (or content proper, or the conceptual system) and the grammatical (comprising lexemic and morphemic structure). We still must of course distinguish between form and substance with respect to both content and expression. Could I say that you consider glossematics as a stratificational view, in your sense, of linguistic structure ? - Yes, or you could say that stratificational linguistics is an extension of glossematics. Are there in European structuralism, say, in the Saussurean tradition, other sources of inspiration for your work ? - Saussure is certainly an inspiration for me. But of course he was also for Hjelmslev and most of my Saussurean influence has come through Hjelmslev. I have also been influenced by Halliday - as you know he is a contemporary of mine. We have been interacting with each other off and on over the past eight years. He was also influenced both by Hjelmslev and by Firth. So indirectly I have been influenced by Firth through Halliday. 4

C. Hockett "Linguistic Elements and their Relations", Language 37 (1961), 29-53.

SYDNEY M. LAMB

181

What is your opinion about the originality of Firth in the history of contemporary linguistics ? - I ' m really not qualified to make statements about that, but I know that he was influenced very much by Malinowski, perhaps more by Malinowski than by Hjelmslev. His view of language is less immanent than Hjelmslev's. Hjelmslev always stressed the closedness of the linguistic structure and of the linguistic universe; Firth developed a more functional theory of language... - Yes, Firth is more functional and this he gets from Malinowski. i am very much in agreement with that point of view, which is now championed strongly by Halliday. But 1 would also hesitate to accept the notion that Hjelmslev's view of language was closed, because he offers just a breathtakingly broad view at the end of the Prolegomena, in which language relates to practically everything. 5 Who were the most important of your teachers within the American tradition ofstructural linguistics ? - I was taught American structural linguistics mainly by Murray Emeneau at Berkeley. He is known as a Sanskritist and Dravidianist, but he was a very good structuralist too, who was influenced by Bloomfield perhaps more than anyone else. I also had Mary Haas as one of my teachers, and she in turn was a student of Edward Sapir. She taught Sapirian linguistics, so 1 learned Sapirian linguistics from Mary Haas and Bloomfieldian linguistics from Emeneau. And then, although I didn't have him for a teacher, I was very strongly influenced by Hockett through his writings, especially "Problems in Morphemic Analysis" 0 and "Two Models of Grammatical Description. 7 He at that time was influential in my thinking concerning the morpheme and its relation to the phoneme. What is your most important criticism of Bloomfield and the neo-Bloomfieldians ? - Well, we can first perhaps recognize a very important distinction, that would separate two kinds of linguistic theories from each other, namely the single level views of linguistic structure as opposed to the multi-level or stratified views. Among the stratified views, we find, besides that which is known as stratificational grammar, glossematics, the system-structure grammar of Halliday and, to some extent, tagmemics and the work of the generative semanticists, but this last is a kind of mixed system. On the other hand you find the single level systems such as Bloomfield and the original 1957 Chomsky theory. For Bloomfield the morpheme was a combination of phonemes; he then goes on to show that morphemes can occur in different forms or shapes, but he never quite reconciles this conception with the other view that a morpheme is a combination of phonemes - this point is discussed by Hockett in his article "Linguistic 5 6 7

L. Hjelmslev, Prolegomena, pp. 101-121. C. Hockett, "Problems of Morphemic Analysis", Language 23 (1947), 321-343. C. Hockett, "Two Models of Grammatical Description", Word 10 (1954), 210-234.

182

SYDNEY M. LAMB

Elements and their Relations".8 Later linguists, followers of Bloomfield, to varying degrees broke away from that notion, but they never broke away very far, so that you get the prevailing view that the morpheme is a class of allomorphs and each allomorph a combination of phonemes. Thus the distinction recognized between morphemes and phonemes was primarily on the basis of the class/member relationship. That I do not consider to be a separation of levels in the way we separate different strata in stratificational grammar, so that we could call it perhaps a quasi-stratification. It is a step toward a recognition of really distinct levels, but it is not moving far enough. On the other hand if you take a Hjelmslevian system, you get a really clear separation of expression-form from content-form and it is this kind of separation that we know between the strata. So my chief criticism of Bloomfieldian linguistics would be that it is mono-stratal or unstratified. I would also criticize neo-Bloomfieldians for their procedural orientation. Linguistic theory at that time consisted very largely, in this country, of trying to specify rigorous procedures for linguistic analysis and I think this attempt was just a mistake. It has been previously criticized by Chomsky and I agree with Chomsky on this point. In fact, I was preparing to write a paper on this topic when I discovered that Chomsky had already done it in Syntactic Structures.9 And what about the fact that for Bloomfield semantics is not a branch of linguistics ? - I would disagree with him at that point too. Bloomfield didn't see that the same kinds of methods which can be applied in analyzing the lower levels of linguistic structure can also be applied at the conceptual level. I would also criticize the Bloomfieldians and the neo-Bloomfieldians on grounds of their preoccupation with substance and with classification of linguistic data as opposed to trying to discover what the abstract linguistic system is which lies behind the data. Hjelmslev is more interested in finding the form of the linguistic structure which lies behind the data, whereas the Bloomfieldian tradition was much more concerned with classifying the data. Can you explain somewhat the interesting analogy you made between Bloomfield and Chomsky, in connection with your 'typology' of linguistic theories? - I would say that Chomsky's is fundamentally of the mono-stratal or single level type, despite his talk about deep structure. In the original version of Syntactic Structures, we find almost a pure mono-stratal view of linguistic structure, because one takes essentially a surface structure and then applies a transformation upon it to form another surface structure. As far as I know, it was only about 1962 that Chomsky began to recognize deep structure as somehow on a different level from surface structure.10 The transformations then took on a different aspect. 1 heard him give a lecture in 1962 in which he was explicitly recognizing a deep structure as somehow 8 9 10

Cf. note 4. N. Chomsky, Syntactic Structures (The Hague, 1957). N. Chomsky, Current Issues in Linguistic Theory (The Hague, 1964).

SYDNEY M. LAMB

183

on a separate level. I say somehow, because it was not on a really separate level in the sense that we, in stratificational grammar, speak of a separate level. This is essentially the view one still finds in Aspects of the Theory of Syntax.11 The so-called deep structure is made up very largely of elements from surface structure and one still uses transformations - this is a form of what I call mutation rule - as the means for getting from these so-called deep structures to surface structures. Now, a mutation rule by its nature can only be properly applied in a mono-stratal system. It takes a representation and it converts certain parts of that representation into another form; it converts one representation into another representation of essentially the same kind. So, there is no real break between the levels of deep structure and surface structure corresponding to the real break that we find in a stratificational framework between, say, the sememic stratum and the lexemic stratum. We didn't yet speak about Zellig Harris... - Right. He was of much more influence in the Eastern part of the United States than in the West. So in the fifties Harris's views, as expressed in his book Methods in Structural Linguistics12 were very influential in the Eastern part of the U. S., but I would say that in Berkeley perhaps we were ahead of the rest of the country in rejecting that approach. We didn't recognize it as a general advance over the ideas of Sapir and Bloomfield, although there were some excellent ideas here and there. As nobody has as yet spoken about Sapir, I would like to know what the opinion about him was like in the fifties and the sixties ? Has he been considered as a more vague philosopher of language or has he had a great influence on linguistics ? - I would say his influence was very great in Berkeley. One doesn't become aware of this if one looks at the literature. Mary Haas was working on American Indian languages, which of course was one of Sapir's great areas of endeavor. Students read his book and he influenced their thinking through the teaching of Mary Haas. Was there, at that moment, a clear distinction between Sapir's views and Bloomfield's, in other words, were they seen as opposite conceptions of language ? - Not at all. I don't believe we considered that Sapir and Bloomfield were in great conflict except perhaps for the issue of mentalism versus mechanism. I must say that I never could get excited over that issue - it didn't seem to me that it was worth all of the fuss. In fact it looked like something which was left over from an earlier period, viz., the thirties. In fact I would say that where I disagree with them I disagree with both of them in that they rely heavily on a process or mutation type of description where we would have a separation between different levels with the realization as the relation between them. Both Sapir and Bloomfield and also of course the so-called II 12

N. Chomsky, Aspects of the Theory of Syntax (Cambridge, Mass., 1965). Z. Harris, Methods in Structural Linguistics (Chicago, 1951).

184

SYDNEY M. LAMB

generative approach use a process type of description in which, for example, you take one phonological entity and then you have the rewriting of it into another phonological entity. One is making the distinction between morphophonemes and phonemes on a process basis instead of putting them on different levels with realization as the relationship between them. It would be interesting to sketch the framework of your linguistic theory, called stratificational grammar. Where has the name come from ? - This name I started to use in 1961, or rather I started to use the term 'stratum', because of the fact that the term 'level' which I had used up to that point was too ambiguous. The term 'level' was being used by linguists in many different ways conflicting with each other. For example, there are different syntactic levels at the same stratum, say, the level of the sentence as opposed to the level of the clause as opposed to the level of the phrase; and these can all be called 'levels' but they are all at the same realizational level or stratum. Then you also had the 'levels' that were talked about by the neo-Bloomfieldian linguists, what they called the 'morphemic and phonemic levels'. These were levels of activity being performed by the linguistic analyst, because to them the level of phonemics was a kind of procedure being followed by a linguist at a particular stage in his analysis of the language. So, I found that it was very hard for me to make clear to other people what I was talking about if I continued to use term 'level' and so I started to use the term 'stratum'. This was in 1961, and at the same time but completely independently Hockett had decided to do the same thing. 13 As we were talking about the same thing and came up with that same term, we congratulated each other afterwards. If you had to summarize your whole linguistic theory, what would be most central ? - The most important aspect of it is that, like Hjelmslev, I am interested in the structure which lies behind the linguistic data, rather than in the linguistic data themselves. Secondly and almost equally important is that I view the structure of language as a network of relationships. This structure has no items at all. In terms of Hockett's dichotomy between item and arrangement as opposed to item and process, I am opposed to both his views. I am opposed to the item and process view not only because I don't consider that linguistic structure in itself contains any processes, but also because I don't consider that the linguistic structure contains items. It's purely a network of relationships, i.e., relationships of relationships to relationships. Then the third point, and I guess almost as important, is that I have a stratified approach. That means that linguistic structure cannot properly be understood by looking at it as if it consisted of just one level with some kind of processes. Rather one has to recognize and distinguish several separate structural levels within this network of relationships. 13

Cf. note 4.

SYDNEY M. LAMB

185

Your linguistic theory is perhaps the most radical structuralism in the whole panorama of contemporary linguistics. Could we dwell on that for a moment ? - Yes. But is linguistics nothing but theoretical constructs or is it also an 'empirical science' ? - I would say it is an empirical science but that doesn't detract from the idea that the model is a theoretical construct. The kind of construction in linguistics is not a purely abstract scheme like, say, a mathematical theory, without regard to where it might correspond to reality somewhere in the world. It is strictly empirical in that we would not accept any hypothesis about linguistic structure as appropriate unless it could be shown through various kinds of testing that it does relate to the real world. 14 The particular phase of the world I am interested in, besides the linguistic data, is the human mind. The special aim, at least in current work in stratificational grammar, is to get closer to an understanding of the mind and how it works. This abstract system that lies behind the linguistic data is supposed to correspond in some way to the system in the mind of the speaker of the language, which for him lies behind the linguistic data that he is able to produce and understand. In other words stratificational linguistics studies the human information system. Can you agree with Chomsky's statement that linguistics is a subbranch of psychology ? - I would not agree with that, not at all. The current emphasis in stratificational grammar, to get at an understanding of the human information system, could be viewed as consistent with cognitive psychology. But that is only one branch of linguistics. If you take the kind of linguistics Halliday 15 is working with, which I consider equally important, that belongs more in sociology than in psychology. If you look at historical-comparative linguistics, one is finding information relating to prehistory, so this could be viewed as a branch of anthropology. Then if you extend the linguistic analytical techniques into the conceptual structure, you get into the area known as cognitive anthropology, because here one is dealing with a person's knowledge of his culture. Linguistics can be shown to be related to various other social sciences and even, perhaps more important, to philosophy. One can also find very close relationships to information science and computer science. The interesting thing about language is that it is related to so many phases of the human experience, so that linguistics overlaps with various other fields. It would be a mistake to assign it as a subbranch to any one of these fields. It would make just as much sense to assign any one of these fields as a subbranch of linguistics. In fact, I consider the view of Hjelmslev at which he arrives at the end of his Prolegomena quite attractive, where he indicates that all science (not only social science) will eventually find its foundations in linguistics.

14 15

Cf. L. Hjelmslev, Prolegomena, pp. 13-15. Cf. the dialogue with M.A.K. Halliday in this book.

186

SYDNEY M. LAMB

Can you give some characteristics of what you call the human information system ? What do you mean by that ? - This is just a name for the mental system which any human being possesses, whereby he processes information. He receives information from the outside through the sense organs, he processes such incoming information in the type of process known as perception. He stores it and performs processes such as comparing or putting out information. One of the major means of putting out information is through the use of language; but information can also be put out in any kind of motor activity such as walking, dancing, etc.... All of these activities can be studied from an information processing point of view, often adopted in computer science. When I speak of the human information system it is perhaps more a point of view that anything else. In adopting the viewpoint one is attempting to understand human activity in terms of the processing of information. You asked me actually for characteristics. We have evidently to divide the system into various modalities. We have different sense modalities for perception of incoming information: visual perception, auditory perception, etc.... We have various modalities for outgoing information. What I have been working on is the modality of language. But we might apply the findings and techniques of stratificational linguistics to an examination of other areas besides language. And it appears that it can be applied. For example, the system for visual perception seems to have a structure that is very similar to that of language in some respects. So my view of the human information system is that the whole thing is a network of relationships, which can be divided into various sub-networks, one of which is language. Language then can be divided into different stratal systems. The highest stratal system is of considerable interest here because of the central position that it occupies: it can be called the conceptual system, or, in terms of stratificational linguistics, the sememic stratum. The problem at this moment is whether this is the highest point of just the linguistic structure or whether it can also be viewed as the highest level for various other modalities as well. Certainly it is connected with the system for visual perception and other kinds of perception and to various motor areas. For example, if you take the meaning of a word like red, viewed in terms of the human information system, we may call it the concept "red"; that concept is an element in the sememic system. By the way, it is not an item in the usual sense but just an intersection of various relationships. It is connected to the lexical item red and it is connected to various other concepts within the sememic system, such as sub-types of "red" and the supertype "color". But it is also directly connected to a point in the visual system where we have the visual image of "red"; so that part of the meaning of "red" to a human being is the visual image of what "red" looks like. Now one finds a hierarchic connection between red and sub-types of red within the visual system as well as in the conceptual system. And then the question arises, in setting up a model of the system, whether to duplicate that hierarchy that evidently must be present in the visual perception system. Is that to be duplicated in the conceptional system or are

SYDNEY M. LAMB

187

they one and the same hierarchy ? Is this another case of stratification, separation of levels, or not? If not, then the visual system is very directly connected to language. Even if they are to be separated there still arises the question: is this conceptual system as closely related to the visual system as it is to the lexical and grammatical system? Going on with the characterization of the information system, one could think of it in sort of loose terms as like a clump of trees, where each tree is one of these modalities, and the branches of the different trees interconnect with one another. For the language tree the lower end would be the expression and the higher end would be the content or what we can call the network of concepts. The analogy with a tree is helpful within any of the modalities; as you go higher, i.e., more abstract, you find larger and larger inventories. For example, the number ot morphemes in a language is quite large in comparison with the number of phonemes, and the number of lexemes is even greater. There are perhaps just a few thousand morphemes in a typical language, but there are tens of thousands of lexemes (or lexical items); and the number of concepts which these lexical items represent is even greater, perhaps hundreds of thousands. This is similar to the structure of the tree. You start from a very few branches at the lower level of the tree and each of these branches out so that if you get up to the upper limits of the tree where the actual leaves are found, they are of course very numerous. Now a primary feature of the human information system is that it is a network of interrelationships. It can be divided into sub-networks each of which is roughly analogous to one of these trees in the clump. Then within some of them, e.g. language, one can further subdivide into stratal systems. In the current view I have of it at least three such stratal systems can be distinguished: the phonemic, the grammatical and the sememic or 'conceptual'. Are you stressing the analogy of language and other human activities... - Well, it is not so much the activities, but the systems which underly the activities. Does language occupy a privileged position among the information systems ? - Yes, I can see language as being of particular importance. We might be able to say that the language tree is at the center of the clump. I am not ready to commit myself to that point of view, but language is important in just that respect that the sememic system, which was arrived at on purely linguistic basis, purely as a result of studying the linguistic structure, nevertheless seems to be a central coordinating point for other modalities, other than language. It may be that one and the same conceptual system could be at the highest level of, say, visual perception, extra-linguistic auditory perception, etc. Our conceptual system permits us to think about anything in the range of human experience. Are you working on what Saussure called a 'semiology', a general science of signs ? - Yes, 1 would be doing that. This is another area in which I was very much influenced

188

SYDNEY M. LAMB

by Hjelmslev. And I would also say, that seems to come as a natural result of studying linguistic structure. It doesn't emerge if one is tied to the studying of linguistic data as in the general Bloomfieldian tradition, but it does when you look at the structure behind the data, and the relationships. In America, during the thirties, people studied phonological structure and after they figured that out to some extent they went on to morphological structure and after they had studied that to a certain extent they went on to syntactic structure, and then in the late sixties and the seventies, people have been looking at semantic structure. Once into semantic structure, you begin quite naturally to see relationships to these other areas: cultural anthropology, visual perception, etc. It is just a natural extension as we continue to see where this type of analysis leads. On the other hand, Pike had already begun this extension years ago. 16 How to make inference from the linguistic data to the linguistic structure ? And is that the way that has to be followed by linguists: from the data to the structure? - Here we have to separate the practical procedure from the question of scientific validity. As far as the theory of what we consider scientifically valid is concerned, it doesn't matter how we arrive at our formulation. So we can use, for instance, intuition or guesswork to arrive at a hypothesis, provided we test our hypothesis. But with respect to practical procedure I would emphasize that we have to examine the linguistic data very closely. Theoretically it doesn't matter, but in practice you find that those people who don't pay very close attention to the linguistic data are not likely to arrive at hypotheses that will check out. So I would not want it to be thought, because I am concerned with the formal structure, that I am advocating that people neglect the study of substance. I myself have been very closely concerned with the study of the raw linguistic data. Those who have criticized Hjelmslev on this point have been mistaken. That he made the separation between form and substance didn't mean that he was advocating that one not pay close attention to the substance. The opposition of empiricism versus rationalism is the most important issue in the epistemological discussion within American linguistics... - I have never been able to see such a conflict between empiricism and rationalism. I think it would be a mistake to adopt a purely rationalistic point of view. But I also think that to adopt an empirical point of view does not mean that one is rejecting rationalism. My general approach is that instead of choosing either one or the other, we should have both. We must be concerned with the raw data but it doesn't mean we have to get into the kind of trap that the neo-Bloomfieldians got into, where they seldom went beyond classification of the data. You start from a corpus ? - Right. I see no objection to starting from a corpus. I think it is a valuable exercise 16 K. Pike, Language in Relation to a Unified Theory of the Structure of Human Behavior, revised edition (The Hague, 1967).

SYDNEY M. LAMB

189

to start with a corpus, but this doesn't mean that the description one turns up with is just a description of that corpus. And in fact it didn't mean that to the neo-Bloomfieldian linguists. They were trying to arrive at description that would account for the corpus together with anything else that had the same kind of structure as the data in the corpus. It's very valuable to work with a corpus, but one must view such material as evidence or manifestation of some system. It is evidence for that system and it is the system which lies behind it that we are trying to get at through the study of these data. The neo-Bloomfieldians would never accept to make inferences from linguistic data to the system of the human mind. - That is true, they were never willing to make that step. This is the point at which I follow Hjelmslev as opposed to the neo-Bloomfieldians, though Hjelmslev didn't mention the mind in this connection. You define language as a network of relationships, which is a very radical point of view. What about the metaphor 'network' ? - Well, technically a 'network' is a type of structure like a tree, except that where a tree can only branch in one direction, a network branches in both directions - so you can have paths that first branch and then come together. If we draw a diagram of a particular linguistic structure, parts of it look something like a net which would be used for fishing, although, 1 must say, a good fishing net is a very perfect form of network in terms of mathematic structure. This we don't find. But anyway the system consists of lines and nodes, each node being a point at which lines intersect. And there is a basic distinction between two kinds of nodes: the OR and the AND. Can you draw some details on the structure of these networks ? - Let us begin with the microscopic point of view and focus on the nodes themselves. The type of node which has been in the literature so far is actually not of the finest level of structure. I have been working on the analysis of the basic OR and A N D nodes into their internal structures, but none of these has appeared in print yet because I haven't come up yet with a fully tested hypothesis about the internal structure. So, let me talk just about the type of nodes which have appeared in literature, although I want to emphasize I don't regard these as at an ultimate level of analysis. In dealing with the dichotomy between expression and content, I also recognize a kind of bidirectionality in the linguistic structure, and in drawing diagrams I put the expression at the bottom and content at the top - in other terms the concepts are at the top and articulations are at the bottom, in the case of spoken language. The network of the linguistic structure then is what intervenes between the concepts and the articulations. In keeping with that bi-directionality we can have a branching in the downward direction, i.e., towards expression, or in the upward direction, towards content; so we distinguish between the downward OR and the upward OR. The downward

190

SYDNEY M. LAMB

O R involves the choice oi alternative expressions for the same concept and the upward OR involves a choice of alternative concepts for one expression; there is also the downward AND as opposed to the upward AND. For the downward AND, the type of thing we find most often is right in the sign relationship. For instance dog manifests a single unit at one level leading down to a combination of three units d/o/g/ (Figure 1). This type of A N D by the way, has ordering in that the d/o/g/ have to occur sequentially in that order. There is a distinction between the ordered AND and the unordered AND. Unordered A N D involves simultaneous components such as in the lower levels of phonology. We might have a p in some language which at one level is most economically treated as one unit. At this level it would therefore be represented as one line in the network. But at the lower level it would connect to the phonological components/closure and labial. We get simultaneous components at higher levels of the linguistic structure too. To illustrate the OR relationship, we have a choice between two realizations of, for example, good, because in the comparative we don't say *gooder, we say better. So one uses bet instead of good when the comparative is occurring. These are quite different at one level but are just alternative realizations of what is the same thing at the deeper level. This is a downward OR because you have two different expressions for one concept (Figure 2). Upward OR would be found in the case of big as referring to "size" as opposed to big as referring to "importance", in "He is a big man in his town". Big can also be used to illustrate the downward OR, because when we have the concept of "big" referring to size, this can alternatively be represented as large. So, Figure 3 shows that we have both a downward OR relating to big and large and an upward OR relating the two kinds of big. The types of nodes are summarized in Figure 4. dog

Fig. 1.

GOOD

Fig. 2.

BIG

BIG

Fig. 3.

Now every stratum has a syntax, which is also built of lines and nodes. The upward direction within the syntax leads to different syntactic functions, so for a NounPhrase in English, we will have an upward OR that will relate Noun-Phrases to the different functions they have, like subject and object. And lines in the tactic plane connect downward to alternative linguistic units that can have the same function. So

SYDNEY M. LAMB

Types o f Nodes

Downward

Upward d

e

Ordered AND

b

c

a / b c

Unordered AND

h

ì

? / h-i

Ordered OR

n o m / n+o

r p+q / r

t u s / t,u

V,W

Unordered OR

Fig. 4.

X /

X

192

SYDNEY M. LAMB

at one point we have a downward OR connecting, say, to personal pronouns on the one hand or to Noun-Phrases on the other hand {Figure 5). There are also A N D relationships within the syntactic structures. A downward AND would be used where there is a combination of syntactic units which together perform a particular function; so for Noun-Phrase you would connect downwards in an A N D relationship to, say, Determiner followed by Adjective followed by Noun. The upward A N D would be used in a syntactic structure where a given unit has two Junctions simultaneously. This would occur for example in the semotactics where a given element, say "John" as in "John kissed Helen", would be both agent of the process of kissing and the topic in the topic-comment construction. So, it simultaneously functions as topic and agent - it will have an upward A N D linking it to the two functions. Or, in "Harry kicked himself", there is a third line, as "Harry" is also the target of the process. Subj.

Obj.

Fig. 5.

The term 'network' is originally, I believe, a term out of computational linguistics. - Actually, it got into computational linguistics from mathematics. It has been a relatively neglected topic in mathematics, one which should receive more attention. The use you make of the notion of'syntactic' is very different from that in traditional grammar. - Yes, that is probably the next thing I need to explain. Part of what I was criticizing earlier in other linguistic theories is their mono-stratal or single-level conception.

SYDNEY M. LAMB

193

One aspect of that is that they only have one syntax. From my point of view we see that a language has several syntaxes, in particular every stratum has its own. This idea is not original with me. It goes back to the neo-Bloomfieldian days where in some branches, including that of Hockett and that of Bloch, it was recognized that there is a 'syntax' of the phonemic level in addition to the traditional syntax. At this point Hockett decided to use the term 'tactics', which has the same Greek root as the term 'syntax', referring to arrangements; he said that we could use the term 'tactics' to refer to that part of the structure which is concerned with arrangements at whatever level. To be more specific we can use the term 'phonotactics' for the syntax of phonemes and 'morphotactics' for the syntax of morphemes. In stratificational linguistics, after I realized that it doesn't suffice to recognize only one stratum of content, I recognized another level above the morphemic, viz. the sememic, and it appeared that that level too has a syntax; so we called it 'semotactics'. Later as it became apparent that these levels were also not sufficient, there was another added in 1963, the lexemic; this lexemic level intervenes between the morphemic and sememic level. So this gives us phonotactics, morphotactics, lexotactics and semotactics. Recently, however, it has become apparent that the lower portions of morphotactics and lexotactics merge, somewhat like branches of a river. Could you say that the Chomskyan transformational syntax is only one syntax within this whole of linguistic levels? - This level, as in Aspects of the Theory of Syntax,11 in some respects resembles the semotactic level of stratificational grammar, in some respects the lexotactics and in some other respects even the morphotactics. This is because Chomsky is still mixing levels. But the generative semanticists have a tactics which comes rather close to the semotactics of stratificational grammar. That may give you some rough notion of what the semotactics deals with. As for lexotactics, to give you an intuitive feeling of what it is, this is the level at which we say that a clause consists of a subject followed by a predicate. There are different kinds of predicates but for English, at least, they consist of a 'Finiteness' element, which consists either of a past tense, or a modal auxiliary or a third person singular marker, followed by a Verbal unit. There are different kinds of Verbal units; they connect with prepositional phrases optionally, and some of them consist of a transitive verb followed by an Object; others consist of the verb to be followed by either a Nominal or an Adjective (Figure 6). And so forth. That kind of structure, it seems to me, is inescapable. Although it is not valid by itself, a very suggestive kind of evidence is the fact that it has been recognized throughout the centuries by grammarians. It must have some intuitive value. If you look at transformational grammar either of the traditional kind or of the generative semanticists (which I consider as a branch of transformational grammar though some of them might disagree with me), nowhere in their grammar is defined a level 11

Cf. note 11.

194

SYDNEY M. LAMB Clause

Fig. 6

at which this subject-predicate structure is recognized, as far as I know. At the level of morphotactics we are dealing primarily with word-structure. Words consist typically of roots together with prefixes and suffixes. In phonotactics we are dealing primarily with syllable structures; I say primarily because phonotactics also goes beyond the level of syllables. It goes to larger units: stress groups, tone-groups, etc. There is here another basic distinction between the stratified approach and the mutational approach such as transformational grammar. Let us consider a typical generative semantic theory, one which generates semantic combinations at the level of what I would call semotactics. Then to get from that level to a lower one, one applies transformations. Now within our framework the way we get from the semotactic representations to a lower level is not by application of transformations but by

SYDNEY M. LAMB

195

the use of lexotactics and of morphotactics. That is, the surface structure representations are not arrived at as a result of transformations but rather they are generated by the syntax of surface structure. At this point there is a nice analogy. Take the commonly occurring situation in which one goes to a cafeteria. One takes a tray, goes to the serving area in the kitchen and the first type of food one encounters is the dessert, let us say, then one encounters the salads and then the main course and then the soup and then the drinks. Then one gets into the dining-hall. One has to eat all this in a different order. Does one perform a transformation in which one transforms that structure, i.e., dessert (1), salad (2), main course (3), soup (4), drink (5), to 4 + 3 + 2 + 1 + 5 ? Is that what one does, or is it the case that what one does in the dininghall is simply to apply the syntax which governs the order in which courses of the meal are eaten, which everyone knows ? I would say it is the latter. If one adopts the position that is the former, one gets into the following difficulty: that if you go into a different cafeteria where the order of items is a bit different, you must learn a new 'transformation'. Or what about a person who has a broken leg and isn't able to walk through the cafeteria line and has to send a friend for him? How is such a person to proceed when he doesn't know the order of the 'deep structure' ? He isn't able to eat! Nonsense, of course, he is able to eat, because he knows the correct 'surface structure' even before one has told him what the deep structure is. In the same way, instead of having to perform transformations when we talk, what we do is to use the syntax of surface structure. So, the order of elements occurring in the surface structure is given by the syntax of the surface structure. So we see that as a corollary of the view that each stratum has its own syntax, there are no transformations in our linguistic structure.

Are the formal qualities of these different syntaxes similar? - Yes, in terms of just the formal properties, they are similar, they are all made up of lines and nodes of the same kind. Do you consider this similarity as a consequence of the fact that these syntaxes are the result of a theoretical construction ? - I have to say that I regard this as a finding rather than an assumption I started out with. My main attempt all along has been to discover the relationships existing among linguistic units. And I found that those existing in lexotactics are of the same kind as those in phonotactics, etc. I regard this finding as in harmony with the reality of the situation; that is, in the actual mental structures involved we will some day find the same kind of neural configurations. This similarity would not be found by a linguistc school which didn't get its structure abstracted from the raw data. If you are close to the raw data, then you will be struck by the differences, as in transformational grammar where one finds that phonological structure looks rather different from the structure of other levels.

196

SYDNEY M. LAMB

Are the relations between the different units on each level of a logical kind? Do you notice for instance the relations of equivalence, alternation, opposition, etc. ? - Each of these relations has its correlate in our relational network. And they are found at all levels of the linguistic structure. One of the most provocative elements of your theory is that language is a network of relationships and not of entities. That is a very radical conception... - Yes it is, and I have found that people have a great deal of difficulty in understanding this. Let me therefore try to explain this point by taking a typical example of a morpheme. We could think of a morpheme such as dog, and ask ourselves: what is this morpheme ? Let us not be satisfied with any answer that falls short of saying completely what it is. To characterize it we can give a complete list of its properties as a morpheme of the language. In the first place it has a phonological form which we can represent as d followed by o followed by g. In the second place it has a certain grammatical function which can be summarized as 'noun'. In the third place it has a certain meaning or several meanings. These meanings we will characterize by connecting dog to one or more concepts (and for the moment we leave aside the question of the nature of a concept). We connect it to whatever concepts it needs to be connected to, within the conceptual structure. We can indicate its grammatical properties by connecting this element dog to the noun position within the syntax. The expression can also be indicated by connections, as shown in Figure 1. Now I have completely characterized that morpheme in terms of connections which it has to these various locations, conceptual, grammatical and expression. The totality of the properties of dog is represented by those connections - therefore dog as an element has no existence in addition to that. There would be no information added by having a symbol or a label "dog"; in other words dog emerges as a line or a node which has connections to specific points in the network. In drawing a network it is of practical value to put a label next to the line; it makes it easier to read, but we recognize that those labels are not a part of the structure. You might say, it's easy with a simple morpheme like dog but what about a more complicated case ? Let's say well in English. Well is ambiguous in that it is a noun which means "a hole in the ground with water at the bottom of it" or it can be an adjective opposed to sick, or it can be an adverb corresponding to good or it can be a conjunction as in "Well, I don't know". Even with such a more complicated case we can arrive at the same result, because what we have corresponding to this four-way ambiguity (or quadriguity) is an upward OR-node with four lines connecting to four different nodes. This type of node may be called a tactic connector. It is a point at which you get an intersection between the realizational lines, which connect expressions to concepts, and the tactic lines. There have to be points at which the tactics interconnects with the realizational lines. Such points are tactic connectors. Notice that as I just described, each of the meanings of well is connected with a grammatical function. There will be therefore four tactic connectors, one connecting to each gram-

SYDNEY M. LAMB

197

matical function. Downward from that upward OR we have the same situation as for dog; we will connect to a downward A N D which in turn connects to three phonological segments w, e, I (Figure 7). So well requires just a slightly more complicated network. And similarly for every morpheme in the language. Then you look further, at the syntactic structure, whether that also can be characterized in terms of relationships; it can, because the totality of syntactic structure can be represented in terms of lines and nodes.

C O N C E P T S

E X P R E S S I O N

Fig. 7

One might ask, "If you get the same configuration of lines and nodes for one morpheme as you do for another, are you then failing to distinguish between them ?" It is quite interesting that if you look just at the formal structure representing one morpheme and then at the formal structure representing another morpheme, it may happen that these formal structures, looked at locally, are identical. On the other hand if you look at where the lines are connected, you find that they connect to different points. Suppose we have two elements with the same phonological form but different grammatical functions and different meanings. This is the case with well. The fact that it has different meanings and grammatical functions is shown in the diagram.

198

SYDNEY M. LAMB

And in general if we ask the question: how do we distinguish any element from any other element, the answer is: "by virtue of the different points to which those two elements are connected". Now if we found a case where two elements are connected, for all the connections, to the same points, then that would simply be a redundant representation of one and the same element. Such a redundant representation possibly occurs in the neural network in the brain, but in the linguistic description they would be one and the same. We are able to distinguish any linguistic element from any other linguistic element by virtue of the different connections it has, and therefore we have represented what it is not by virtue of what it consists of, but by what it is connected to. This conclusion is just a formal working out, as we have a formalizable notation, of what Hjelmslev was saying some thirty years ago when he stated that a totality does not consist of things but of relationships. Almost all linguists agree that phonological and syntactic (in the classical sense) units are structured. But what about the structure of the dictionary? As far as the lexical units are concerned, there you inevitably have to do with substances. - It might appear so, at first glance. The lexical item corresponds roughly to what we call the 'lexeme'. By the way, the term 'lexeme' did not originate with me. I mention that because some people criticize me for introducing new terminology; but I have wherever possible used terms that have been used before. 'Lexeme' was introduced by Benjamin Whorf in the thirties18 and used subsequently by Morris Swadesh19 and Harold Conklin 20 and others. I started to use it under the influence of Conklin. Similarly 'sememe' is not my term. It was first used by the Swedish linguist Noreen, back in 1908,21 and later by Bloomfield.22 'Morphotactics' and 'Phonotactics' are not my terms either. All 1 did was generalize and extend that terminology by adding 'lexotactics' and 'semotactics'; and I added terms for elementary components: 'phonon', 'morphon', 'lexon' and 'semon'. 'Lexeme' then, has been used by others and by me as a technical designation for the lexical item. Every lexeme has its connection to the grammatical tactics. And it connects downwards to expression in some cases as a simple connection; e.g., the lexeme "dog" coincides with the morpheme "dog". Others are more complicated; e.g., "German-shepherd" connects to the combination of morphemes "German" and "shepherd". And then any lexeme connects upwards to the sememic or conceptual system. Now in the case where it relates to several concepts, we have an upward OR node which allows one line to go 18

B. Whorf, "Language: Plan and Conception of Arrangement", in J. B. Carroll (ed.), Language, Thought, and Reality: Selected Papers of Benjamin L. Whorf (New York, 1956). 19 M. Swadesh, "Chitimacha", in H. Hoijer et al., Linguistic Structures of Native America (New York, 1946). 20 H. C. Conklin, "Lexicographical Treatment of Folk Taxonomies", International Journal of American Linguistics 28, n. 2 part 4 (1962). 21 A. Noreen, Vart Sprak (Lund, 1903-1918). Portions translated by H. W. Pollack, Einführung in die wissenschaftliche Betrachtung der Sprache (Halle, 1923). 22 L. Bloomfield, Language (New York, 1933).

SYDNEY M. LAMB

199

to each of the different concepts which the lexeme can represent. So this part of the structure, as you can see, is still purely relational. There is another aspect to the structure of the usual entry in a dictionary and that is the definition, the specification of what the meaning is, and that involves conceptual relationships or semological structure. Now we have found that that type of structure can also be represented as a network of relationships. In other words a concept is nothing else than a location in a network of relationships, this network being the conceptual framework or what we call the sememic stratum. Let's for example take the concept of "carnivore": a carnivore is a type of mammal and it eats meat. But one could go further and distinguish certain sub-types such as the feline animals and the canine animals. All of that information can be specified in terms of the same formal structure of lines and nodes {Figure 8). From the concept "carnivore" one (Properties o f Mammals)

Fig. 8

200

SYDNEY M. LAMB

has an upward A N D with one line connecting to the concept "mammal". In other words we go from a subtype upwards to a class of which it is a subtype. From there in turn, upward lines go to properties of mammals. This means that in describing the system we don't need a separate list of the properties which carnivores share with other mammals. This is automatically taken care of by virtue of the connection of "carnivore" to "mammal". All properties of mammals are automatically properties of carnivores, because of this line. But we also must have one or more lines distinguishing carnivores from other mammals; in this particular case it is a line going up to the fact that carnivores eat meat. And if we analyze that part of the structure, the specification that carnivores eat meat, we find that it also can be represented as a part of the relational network. The further characterization of 'carnivore" is into subtypes of carnivore; here you have lines going downwards to subtypes such as "feline animals" and "canine animals", and each of these is itself a concept and is distinguished from the general category "carnivore" by its distinguishing properties which are represented by further connections within the network. All this illustrates that conceptual relationships can also be represented in terms of the relational network notation. And in general we can completely characterize a concept by connections to other points in the conceptual network, plus, in some cases, networks of other modalities. Can there occur any changes in the structure of the concepts ? - As a result of experience, yes, changes occur daily in most individuals. What is the nature of concepts, are they mental entities ? Must we say that a taxonomy of concepts is possible by the experience of the external world? - I would be satisfied with calling every concept a mental entity; it is a mental entity which for the most part has been arrived at as a result of the person's experience in the world. The structure of the information system is what it is in part because of heredity. It has been built as it has because of the DNA. But a very important part of the structure is there because of the person's experience during his life. It would be a mistake to overemphasize either the hereditary or the experiential aspect. One would fail to understand this system unless one gives proper credit to both of these aspects. Take the case of the concept "carnivore", for instance. I would say that no human being raised in isolation from a society in which carnivores were experienced and talked about would ever arrive at the concept "carnivore". One arrives at the concept, with its properties, because one has learnt it. On the other hand it would have been impossible for an individual to arrive at the concept unless he had certain structures given him as a result of his heredity. But those same basic structures are so fundamental that they could just as well have made it possible for him to learn other concepts that he might have had to learn if he had been raised on another planet. One of the current issues for discussion is that the conceptual structures are 'innate'... - I think that most such discussion is being carried on in an impoverished conceptual

SYDNEY M. LAMB

201

framework. One can only carry on such a discussion meaningfully with reference to a fairly good model of what the human information system is like. For instance, I don't think the Chomskyan kind of model is adequate for talking in meaningful terms about such issues. Another point is that one tends too often to take such issues as involving a choice between two alternatives. Actually, instead of being an OR type relationship, it is an A N D type relationship. The question is not whether it is innate or experiential, it is both. Certainly a large amount of the conceptual structure is innate, say, for example, the concept "love", but on the other hand concepts like "carnivore" are not innate.

What is innate for Chomsky is not in the first place the substantial content of the concept but the syntactic relations between 'concepts'. - That's true. But concepts of the type I have been talking about are syntactic relations in the conceptual syntax. But even when we talk about Chomsky's kinds of syntactic relations I wouldn't adopt one point of view or the other, because we need more evidence. Yet, the evidence that we do have suggests that some kinds of very basic semotactic structures might be innate, such as the topic-comment construction and the agent relationship. These relations are found all over, possibly in all languages now spoken in the world. But that may be explainable on the basis that all languages now spoken on earth are genetically related. But to go from that to the conclusion that semotactic relations in general are innate would be a mistake. By the way, I should add that the semotactic structure of any language is specific to that language. It is not the case that there is a single universal semotactics that is equally available to all languages.

We already spoke about the similarity of the formal properties of structures on each linguistic level. There is a lot of discussion on Hjelmslev's statement concerning the isomorphism of the different linguistic levels. Your point is very clear here. - You don't find a true isomorphism in the technical sense, i.e., as the term would be used in mathematics. But I would say that, looked at locally, one finds the same kinds of formal structures. People who have looked at the linguistic data usually have not become accustomed to our degree of abstraction. Consequently, when considering such a statement, they may think I am asserting that there is a similarity between the data of phonology and the data of semology. I'm not asserting anything of that kind, but rather that one finds similar configurations of relationships. But there is the important distinction that one is very much bound by time at the phonemic level, as one has there linear sequences of elements. At the semological level one is not bound by time or linearity. One reason we have the linguistic code is to linearize non-linear structures so that they can be expressed in time. Looking at that aspect, we find a great deal of difference, but on the other hand if you look at the microstructures of the network you find a great deal of similarity.

202

SYDNEY M. LAMB

Directionality in the model of language has been strongly argued by Wallace Chafe.23 Could you explain your point of view concerning that very important point of linguistic theory ? - I will be happy to disagree with Wallace Chafe. People usually put the issue as whether the direction is from expression to content or from content to expression. I am not thinking just of Chafe at this point, but rather of the dispute between two schools of transformational grammar. The question is: do we have an interpretive semantics, with rules going from the deep structure to semantic interpretations, or do we have a generative semantics with rules going from semantic representations downward. The way I would look at this can be illustrated by the analogy with a highway-system. One school of transformational grammar is asserting that the highway goes from New York to Chicago whereas the other school insists that the highway goes from Chicago to New York. Our view in relational network grammar is that the highway extends between New York and Chicago and cars go either direction. The linguistic system is like the highway system and the linguistic description like the highway map. As a highway-system it is static, there is neither motion nor directionality and cars can go on the highway in either direction, except for one-way roads. One of the chief properties of the linguistic system is that it is a non-process type of system, there is no motion in the linguistic system itself. This doesn't mean that we are not interested in the processes, we are interested in all the actual (as opposed to fictional) processes. The primary processes are decoding from expression to content, or what happens when one understands, and that of encoding from content to expression. Corresponding to cars which move along in the highway system, we have impulses which move along the lines of the network. Processes here are impulses moving through a network which does not in itself consist of processes. There is an analogy here to something that comes up in computer science. When one is writing a computer program to perform some action, there are two basic strategies that one can adopt because one is always dealing with certain information. One can put the information in a static form and then have a generalized program which makes use of the information needed. Or one can build the information right into the instructions. The latter is done in transformational grammar, whereas the former is done in relational network grammar. Our processes for encoding and decoding are considered separately from the linguistic information itself. If you take Chafe's version of directionality, which agrees with the other generative semanticists, producing a sentence is done at the semantic level and then everything is sort of interpreting or realization from here on down to phonology. Of course, Chafe could hardly take it otherwise, since the only part in his grammar where he has a syntax is up at the top level. But when one discovers that every level of the structure has its own syntax, then directionality no longer seems inescapable. In fact it would seem that the lexotactics plays a very important role in the formation of sentences 23

W. L. Chafe, "Directionality and Paraphrase", Language 47 (1971), 1-26. Cf. the dialogue with Wallace Chafe in this book.

SYDNEY M. LAMB

203

along with the semotactics. In other words it is not our view that in the encoding process one first generates a string with the complete semantic representation at the semantic level and then encodes it at the lexemic level and then at the morphemic, etc.; but rather, these different levels are interacting with each other throughout the whole process. This particular point also relates to the notational question. Using a mutational notation, one is virtually forced to that point of view which Chafe and all of the different branches of transformational grammar have. They have to have a total representation at each level and then they operate upon it with these ordered rules successively going down toward the phonetic representation. In fact we know that people don't speak like that. A person utters the first few words of a sentence - in other words he encodes them all the way down to phonology and even he puts them out - before he knows how the sentence is going to end. So, he cannot have formulated his sentence completely at some high level. By keeping our process separate from our information, we don't have this difficulty. And it is in the information structure that the strict separation of levels must be maintained. But when you put it into operation, we find that the different levels are interacting with each other, so that in the process of producing a sentence the lexotactics is playing a very important role. We can put it this way, that the grammatical structure is helping the individual to formulate his thought. People often suppose that the thought is somehow formed independently and is then just encoded into the grammar; but actually the thought is formed only with the aid of the grammatical structure. You cannot say, as Martinet does,2i that linearity is a global property of language. Linearity is, according to you, the property of only one level of language. - Right. In fact perhaps not even of one level of the linguistic structure itself. It is a property of the outputs of the phonemic system, rather than the phonemic system itself, which is also a network. Here too I want to make the distinction between the system and the manifestations of the system. Whereas the manifestations are linear, the phonetic system is not linear. But for Martinet, linearity is an argument for directionality too. Language is defined first of all as phonologically oriented linearity. You have the opposite then in Chafe where language is semantically oriented. - Well, I both agree and disagree with both of them. I think language is semantically and phonologically oriented but with neither having priority over the other. About directionality, one further point. It has to do with notation. I think some linguists take the point of view that notation is relatively unimportant, that what really counts are the thoughts and ideas, and the means of expressing them is relatively insignificant. I disagree with that point of view. I think that notation is very important, because try as one will to avoid it, one thinks in terms of the notation that one uses. This is 24

Cf. pp. 236-237 of this book.

204

SYDNEY M. LAMB

true in any science. One cannot think without having the information stored in some way. Information has to be in some kind of a medium and any scientist uses his notation as the primary medium in which he does his thinking. It is no accident that those who use notations which have a direction also think in terms of directionality in linguistic structure. Generative semanticists have rules which go in one direction. There is an arrow in the middle of the rule - there is something on the left of the arrow and there is something on the right - and it is going in a particular direction. In interpretive semantics, one also has arrows, but they are going in the opposite direction. It is no accident. In the notation which we have, devised in such a way that it will be as close as possible to the actual structure we are trying to represent, there are no arrows and we don't see any direction in the system. Another important topic in linguistic theory is the question concerning the linguistic units and the relation between the different kinds of linguistic units. Are the linguistic units of a different kind at the different linguistic strata ? - First, I'd better clarify what I mean by 'linguistic unit' in this connection, because I previously said that there are no units as such, since one is dealing entirely with relationships. I would nevertheless think it appropriate to speak about linguistic units or elements, provided that we make it clear what we are talking about, that any linguistic element is actually a point of intersection of relationships. We still need to talk about such points of intersection; therefore it is convenient to use a term like 'element'. Perhaps I could clarify the position in this way. The inventory or the set of elements at a given stratum, say the morphemic, is discrete from the set of entities which we find at any other stratum. For example, the set of phonemes is disjunct from the set of entities which one finds at the morphemic stratum. This distinction is in keeping with several other points that have come up earlier; for example, with such a structure the use of mutation rules is incompatible. Instead, what we have is a relationship known as realization, whereby elements of one stratum are realized by elements of a lower stratum. In the case of mutation rules, for instance transformations for going from a deep structure to a surface structure, one is mutating or changing one or a few symbols at a time, but without changing the set of symbols; one is still operating with symbols from the same set. On the contrary, just as we don't find any feature which is in some context a semantic feature, and in another context a phonological feature, similarly, we don't find that kind of thing between any two strata in a stratificational system. Is the notion of'deep structure'' a valuable notion for linguistic theory or is it a notion leading only to more confusion ? - It is an interesting notion. Only the terminology might possibly be confusing in that we talk about a higher stratum instead of a deeper structure. So the analogy is going in the opposite direction. There is another source of confusion in that I have used

SYDNEY M. LAMB

205

the term surface information in my Outline of Stratificational Grammar25 to refer to something entirely different from surface structure. I think the use of the term 'surface' in both contexts has been unfortunate. You say that effective information is not identical to surface information ? - Yes. Now here I am talking about the description of some portion of the linguistic structure. It might be described for instance by means of a relational network diagram. Any such diagram is conveying information about the structure - this we call effective information. Now if you have two descriptions of the same part of the structure, they might make use of a different number of lines and nodes. The one making use of fewer lines and nodes will be the simpler description. Two such descriptions are different in surface information, for surface information can be measured in terms of the lines and nodes in the diagram. One tries of course to achieve a description which conveys the maximum effective information with a minimum of surface information. Here we are talking simply about the linguist's description of a portion of the linguistic structure; this could be of course at any level of the structure of the language, at any one of the strata. The other distinction, that between deep structure and surface structure, is quite different, because here one is talking about different portions of the linguistic structure. The question of 'deep structure' came up earlier in our conversation and I mentioned that Chomsky's 'deep structure' in some ways is like our semotactic level and in some ways is like our lexotactic level, and in part it even has things which in our system would be at the morphotactic level. The basic idea of distinguishing a deep structure from a surface structure is very attractive, because this is a kind of strataI distinction. This is an approach towards stratification but I would say that Chomsky has not gone far enough, because his deep structure is not as separate from his surface structure as it should be; he mixes surface elements into his deep structure. This can be viewed as a consequence of the fact that he is using mutation rules as his notation. Do you consider the traditional syntactic categories Noun-Phrase and Verb-Phrase as categories of the surface structure ? - Yes. Corresponding to the Noun and Noun-Phrase at the surface level, you have what we would call different kinds of things, animate beings, etc. at the sememic or conceptual level. To have a better view of your theory, it would be interesting to work out a specific example of traditional grammatical difficulty. I think here of the question of ambiguity. What can you do, in your framework, with problems of linguistic ambiguity ? - For any case of ambiguity in language, one will find an upward OR node somewhere in the system. Now we can distinguish different kinds of ambiguity on the basis of 25

S. Lamb, Outline of Stratificational University Press, 1966).

Grammar

(Berkeley, 1962; revised version, Georgetown

206

SYDNEY M. LAMB

where the upward OR node is situated. We can for example distinguish between 'polysemy' or 'homonvmy' and 'homophony'. Polysemy is the situation in which one lexeme can refer to either of two or three different concepts, as in the case of big which I mentioned earlier, or table which can be a table as an article of furniture or a table in a book. This can be called homonymy if we look at it from the point of view of the two concepts which have the 'same name' at the grammatical stratum. Homophony, on the other hand, is the situation in which two or more morphemes have the same phonological representation, e.g. well. One also has cases of syntactic or tactic ambiguity and such ambiguities can occur at any one of the tactic levels. And again one is dealing with the upward OR. A Prepositional Phrase for example can be ambiguous as to its function in lexotactics, and for each of its tactic functions there would be a different line coming from an upward OR. In connection with ambiguity, the problem that arises is disambiguation, i.e., how do we account for the ability of the speaker of a language to make a choice when presented with such alternatives. This is a problem that arises of course in the operation which we call decoding. It is in this connection that one of the reasons for having a separate syntax on every stratum becomes apparent, because what the speaker evidently does in the usual case of ambiguity is to disambiguate on the basis of context. How does he bring the context into consideration? It is through the use of his tactic patterns. Some cases of ambiguity are resolved by the lexotactics, some at the semotactic level. If we get "well" occurring in a context such as "they dug a deep well", the lexotactics at that point is prepared to accept a noun and only a noun. What it does then is to screen out the other possibilities. In terms of the diagram which we had above (Figure 7) when I was talking about "well", it is connected by means of an upward OR to four different tactic connectors. In the decoding operation impulses will go upward from that element "well" to each of the four tactic connectors. But an impulse will be allowed to go upward beyond that connector only if an impulse is also coming at that time from the lexotactics. Now in the context which I just mentioned, "They dug a deep well", at that point lexotactics will send an impulse down from the Noun connection, so that from that connector it will be possible to send an impulse upward to the appropriate concept. But for the other connectors there will not be the appropriate impulse from the tactics and therefore each of those impulses is blocked. In other words, the lexotactics is acting as a filter. It filters out all of those possibilities which do not fit the grammatical context at that point and allows to go through those which do. Other cases of ambiguity, which are perhaps more interesting, are those in which the lexotactics would allow more than one possibility to get through. This often happens with Prepositional Phrases, as in the sentence "John found a book on York street"; we don't know whether "on York street" modifies "book" or whether that's where he found the book. In this case the ambiguity is not able to be resolved even at the semotactic level, because in fact both possibilities make sense. So, this is the case of the fully ambiguous sentence where two different decodings are possible. It's in a

SYDNEY M. LAMB

207

case like this that a person who hears the sentence might ask a question - if it is in a conversation he'll get a clarification. But in another type of situation, it might have been "John found a bracelet on York Street". In this case, from the lexotactic point of view, the structure is just the same. In other words, both possibilities would get through. But at the semotactic level only one possibility is allowed here, essentially because "bracelet" is not a kind of discourse. In the interpretation of "John found a book on York Street" in which the book is about York Street, that meaning of "on" is allowed by the semotactics only where it is accompanying a kind of discourse, as in "John gave a speech on yoga" or "John writes on mysticism". This is a fact of the semotactics, that this concept for "on" goes with certain kinds of discourse or with discourse in general. So in this case the semotactics is blocking the notion of "a bracelet concerning York Street", because that is semotactically ill-formed. It was allowed to pass through the filter of the lexotactics, but it gets blocked by the filter of semotactics. In general, this is how disambiguation is accomplished. It is done by a tactic pattern, and that can happen at any level, because any syntactic pattern disambiguates those situations appropriate to its own level. It happens automatically as a result of allowing impulses to move through the network according to general rules. This is an interesting treatment. Generally, ambiguity is only seen as a semantic phenomenon. You make it possible to distinguish different types of ambiguity. Now, perhaps as the conclusion of this sketch of your stratificational theory: what are the most recent developments of the model, and what kind of research are you doing today ? - I should begin by giving just a little sketch of the past few years. The reason that I do this is that there have been a number of changes occurring over the past years which have made it difficult for people to follow the literature. In particular, we have had varying hypotheses as to the number of strata in the linguistic structure. In about 1960 or 1961, it became apparent that we needed to recognize an additional level above the traditional morphemic stratum. This would give us a four-stratum system. In 1965 I published an article in which there were five-stratal systems 26 and in 1966 there was the Outline of Stratificational Grammar27 in which there were six strata, i.e., below the phonemic you had the hypophonemic and above the sememic there was the hypersememic; it shouldn't be thought, however, that the four other names, phonemic, sememic, lexemic and morphemic, had the same meaning as they do in the current view. Since 1966 a variety of hypotheses have been attempted. And just in the spring of 1972,1 realized that phenomena of the sememic and the hypersememic or conceptual strata can be combined; one was really dealing here with two levels of the same stratal system. The situation gets rather complicated at this point, because what is going on down in the area of the morphophonemics and phonemics is rather 26

S. Lamb, "Kinship Terminology and Linguistic Structure", in E. Hammel, Formal Analysis, American Anthropologist 67 (1965), 37-64. 27 Cf. note 25.

Semantic

208

SYDNEY M. LAMB

technical. I wrote an article 28 in which I put forward the view that the hypophonemic and phonemic systems could be combined into one, but that view turns out to have been a mistake. Problems involving the interconnections between tactic patterns and realizational patterns have occupied me during the last few months and along with them, the possibility of merging what previously were thought to be independent tactic patterns. The latest such meiger is a joining of the lower levels of lexotactics into morphotactics. You are also concerned nowadays with the generalization of the linguistic model to other structures of the human behavior. - Well, its not structures in human behavior that I am looking at. That's what Pike did. 29 I am looking at the structure of the system which accounts for the behavior the human information system. This is one of my chief inteiests today. In preliminary exploiation I found that the kind of formal structure in the system underlying phonology and grammar can account also for the system underlying semantic relationships. But even beyond that, the findings relating to visual perception that for instance Hubel and Wiesel have come up with in their work with cats at Harvard seems to be accountable for by this same type of network structure. Every area of cognition that 1 have looked at appears to be analyzable, at least at first glance, in terms of the same kind of relational network. I would like to explore now these possibilities. I also am very concerned with doing more description of actual linguistic material, because we have had an important deficiency so far in the stratificational school - that there hasn't been very much linguistic description done within this framework. The main reason has been that for the last several years we have been in doubt as to what the overall form of the linguistic structure was. That was what has been occupying me. I was not prepared to do any detailed description, since I did not know what the form of the description should be. But now that 1 have resolved these problems, I would like to do some more actual description of linguistic material. That leads us to more general questions. Do you see any difference between the task of linguistic theory and the task of grammar - grammar in the traditional sense of the scientific description of language ? - I would agree pretty much with Hjelmslev on this point - as I understand him who would say, if we paraphrase a little bit, that a grammar of a language is a theory of the texts of the language and a linguistic theory a more general thing, namely the theory of grammars. Actually, this is going a little bit beyond Hjelmslev, because I think, in Hjelmslev's characterization, linguistic theory would have been the theory of the texts of all languages. I would put another level in there. A grammar in one sense

28

S. Lamb, "Some Types of Ordering", in V. Becker Makkai (ed.), Phonological Theory: Evolution and Current Practise (New York, 1972). 29 Cf. note 16.

SYDNEY M. LAMB

209

is a theory in that it is a theory of the texts of a particular language. But linguistic theory is a more abstract level, because it should account for all grammars. Is the construction of the stratificational model a part of linguistic theory ? - Yes, it is. Of course, you can engage in this kind of work only by reference to specific languages. Actually there are several dichotomies involved. The one that I just mentioned, the distinction between the theory of a specific language and the theory of language in general, is a very important distinction. For instance, in the stratificational theory, when we talk about the form of the network and these various tactic levels, etc., we are trying to arrive at the kind of thing that has sometimes been called formal universals, i.e. the universal properties that any language has. So that when 1 say that every stratum has a tactic pattern, with certain relationships to a sign pattern, these are supposed to be properties of all languages. I would like to quote one of your statements in connection with formal universals: "The description of an individual language can be properly constructed only in the context of a general linguistic theory. That theory should specify the properties which all languages have in common". - This is in keeping with the point I was just making. Then, another point is the dist inction between description and explanation. Chomsky makes the distinction between an explanatory account and an observational one. 1 think that is an important distinction, but I would have to disagree with Chomsky if he would assert that a transformational grammar of the language offers an explanatory account, because I think that to qualify as explanatory one's account has to have some fairly definite relation to reality. The kind of reality that I think is appropriate in this connection is psychological reality, so we could make a distinction between, on the one hand, a linguistic description which perhaps accurately presents or classifies the data - which has also been called a taxonomic description - and, on the other hand, a psychologically plausible model of the information that the speaker has in his mind. The latter could perhaps appropriately be called an explanatory account. But if one is going to propose that one's account of the language has psychological plausibility, then one must be able to show that this organization of linguistic structure is compatible with the most essential fact that we have about people who know a language, and that fact is that they are able to speak. It is very hard to see how a transformational grammar viewed as a competence model could actually be used by a speaker of the language for speaking. Chomsky tries to deal with this difficulty by relying on the difference between competence and performance. But one has to really stretch one's imagination to accept the notion of competence as applying to something irrelevant to performance. Because competence really means competence to perform. The competence that the speaker has is surely what he uses in some way when he speaks. So any model of the language which cannot be put into operation for actually producing utterances in real time certainly fails the test of psychological plausibility. And therefore I would say that

210

SYDNEY M. LAMB

the transformational grammar cannot be viewed as more than a descriptive account of the language; in other words, I would consider it to be another form of taxonomic description. On the topic of taxonomic linguistics, I would disagree with Chomsky on two points. First, 1 would say that transformational grammar is a form of taxonomic linguistics. Secondly, 1 do not share the view that taxonomic linguistics is undesirable. One gets from Chomsky's writing almost the feeling that it is immoral. I disagree with him. One needs to have taxonomic linguistics as the foundation for going ahead in the area of cognitive linguistics. That is, one has to have a large body of well described and classified data to use as the basis for constructing psychologically realistic models. Therefore I think that taxonomic linguistics, including the transformational version of taxonomic linguistics, is a valuable pursuit. But it becomes somewhat less valuable when people are thrown off by the illusion that transformational grammar has some kind of psychological validity, i.e., when people suppose that perhaps it is more than a form of taxonomic linguistics. How can a taxonomic semological system of language be constructed? It is perhaps easy to make taxonomies of phonological features, but how to make a taxonomy of meaning relationships ? - 1 must emphasize that I am in favor of taxonomy as a first step, a necessary step toward what I consider to be the more valuable goal of cognitive linguistics. But most of the work that has been done in semantics is on associative semantic relations such as polysemy, synonymy, and hyponymy and on combinatory relations: 1 think it would not be inappropriate to call that taxonomic semantics. Some might make a distinction, I suppose, between taxonomic and generative semantics, but the point I am making is that any kind of semantics which is not psychologically plausible is no more than taxonomic in spite of whatever claims are made to the contrary. And it is not psychologically plausible unless it can be shown that the information organized in the proposed form is usable for actually engaging in speaking and understanding. Anything else, even though it takes the form of rules rather than a more transparent form of taxonomy, is merely a way of organizing the information, and therefore (perhaps in a broader sense of the term taxonomy) is just another taxonomy. Is intuition or introspection a good criterion for stating the grammaticality or wellformedness of the linguistic phenomena ? - Sometimes the native speaker's intuition is all you have to go on if you ask if a particular sentence is grammatical or not. In other cases texts can furnish useful evidence. But this question is complicated, because we are really dealing with a continuum. There is no boundary between grammatical and ungrammatical. It is very hard to get reliable evidence from the informant. In many cases the native speaker himself is in doubt. The further difficulty is that of knowing exactly what is that native speaker's intuition. The communication about it may not be clear. But these difficulties are not so important as they might seem. They would be important if one's

SYDNEY M. LAMB

211

purposes made it necessary to draw a sharp boundary between grammatical and ungrammatical. But I would not consider stratificational grammar to be essentially generative in the sense of specifying the set of all possible grammatical sentences of a language. That kind of specification would be unrealistic. The reality that we have to deal with, if we want a psychologically realistic model, is this situation where we find 'borderline' sentences. How can we deal with this? By recognizing that the linguistic structure itself, i.e. the network, undergoes changes while it is being used. In other words, new connections get formed all the time, especially on the sememic level. And in fact, the usual result of understanding some sentence containing new information is that one or more new connections are formed at the sememic stratum. In other words, the distinction between the synchronic and the diachronic aspects of language, which was so important for Saussure, becomes less sharp within the current view in relational network grammar, because linguistic change would involve changes in the form of the network, and we find that such changes take place during the normal use of language. For instance, a new lexeme can come into being just through repeated occurrence of a given combination of morphemes for a given concept. Now the important point here is that we don't get a sharp distinction between the lack of a connection or a fragment of network on the one hand and its presence on the other. Rather, there is a continuum. Let us suppose that we are building a connection from point A to point B in the network; this connection gets built gradually, from repeated use. An example would be in the case in which a new idiom is coming into being. This phenomenon is directly related to the phenomenon of the 'borderline' sentences, where one is not quite sure whether they are grammatical or not. It is very important for psychological reality to recognize the fuzziness of this boundary. And therefore we don't want such a thing as a strictly generative grammar, since, as Hockett pointed out, 30 generative grammar presupposes a sharp boundary between the grammatical and the ungrammatical. Is there a way in stratificational grammar to formalize this kind of continuum? - In terms of a diagram, this would be represented by different degrees of faintness of the lines. Since it is a continuum, one has a special problem in the notation. We could have a faint line and as its gets used more and more it becomes heavier. This corresponds, in terms of neurophysiology, to a synapse which is gradually becoming established by being crossed repeatedly. Do you recognize the same fuzziness of the boundary in the classical dichotomy of language and speech flangue and parole, in Saussurean terms) ? Can you say here too that you don't have a gap but a continuum ? - Actually, I would hesitate to talk about langue and parole at all, because a number of different distinctions have been involved. 33

C. Hockett The State of the Art (The Hague, 1968).

212

SYDNEY M. LAMB

Or the Chomsky an opposition competence-performance? - Here too one has a number of different distinctions. There is on the one hand the distinction between the linguistic system and the products or outputs of that system, where the outputs have sometimes been called performance, as opposed to the system which would be the competence. But there is a completely separate dichotomy between the system and the use of the system. Are encoding and decoding a part of the use of the system or a part of the system itself? - Encoding and decoding involve the use of the system. They are processes, and they are distinguishable from the system itself. Are they in the domain of linguistics ? - They constitute part of the domain of linguistics. That is definite. Suppose we have been presented with the task of constructing a theory of the linguistic system which lies behind a particular language. There are a number of points at which we would have to choose between alternative ways of organizing the structure. Of the utmost importance in making such decisions is to choose that formulation which is compatible with the fact that this system is going to have to be used for encoding and decoding, because that is what speakers of a language do with their system. This type of consideration has been very important in stratificational grammar, but not in transformational grammar. Let us return for a moment to your conception of universals in language. Can you give a definition of what you call'universal' ? - I would hesitate to go beyond certain very obvious things. I would say that every language is a relational network and for every language this network can be divided into certain subnetworks, probably the same number in each language, which we could call sememic, grammatical, morphophonemic and phonemic. And within each of these stratal systems there is a certain internal structure which is doubtless universal. Of course, these are formal features. But if we get into the substantive types of things, I would be very cautious. We need more evidence. It does seem to me that in the conceptual area we would be on fairly safe ground if we would assert that there are certain concepts relating to emotions which are universal for human languages on this planet because of the fact that these basic emotions are related to the endocrine system, which apparently is about the same for all human beings. And in the phonological area we can say that there are certain properties which no doubt for terrestrial languages are universal, because the vocal organs are constructed in more or less the same way all over the world. It appears in the most recent trends that the autonomy of linguistics is transcended. I think here for example of Halliday's linguistic theory which is inspired by sociological viewpoints while other ones are inspired by psychological approaches. Do you see any

SYDNEY M. LAMB

213

contradiction in the fact that psychologically oriented linguists consider language as rooted in thought or in the mind, and sociologically oriented linguists consider language as rooted in culture ? - No, I see no conflict there. I would say both statements are true and we have just slightly different senses of the term "is rooted". I tend to be emphasizing the psychological viewpoint in my work, whereas Halliday emphasizes the sociological point of view, but I see no basic conflict. I would say that his work and mine are complementary. In fact one can look at the sociological situation and at cultural anthropology, and set up accounts of their structures in terms of the individual, because it is part of any individual's information system that he knows how to behave under different situations: he knows the different kinds of communicative function and their appropriate uses. A knowledge of the culture and of social relations is a part of the information system of any individual. Therefore, just this problem of trying to describe the human information system inevitably gets one involved in cultural anthropology and in sociology, so that one is treating here the same kind of facts which Halliday is dealing with but from a slightly different point of view. The information system of any individual includes all of his knowledge and that includes knowledge of the other individuals with whom he comes into contact; not only that, but he must have some internal model of the knowledge which they have. All of that kind of information, which one would have to get into in the sociology of communication, can also be treated from the point of view of the information system of the individual. In fact there is not really any boundary between psychology, sociology, and cultural anthropology.

Structuralism, and some interpreters of Hjelmslev in particular, have always stressed the autonomy of linguistics. - Yes, that's right and it might sound paradoxical, but I agree with that entirely. You stated very clearly that language should be described in its own terms. - Exactly. And in this I have been very much influenced by Hjelmslev. The important point he is making here is that when one is setting up the basic structures of one's model of language, one sets them up purely on the basis of the relationships observed among the linguistic data. One doesn't get fundamental constructs from psychology or any other theory, but from the linguistic data themselves. Then, once one has succeeded in building a model, then one has a basis for incorporating further, extralinguistic information. It is at that point that he comes into contact with the other fields of learning, such as psychology, sociology, and even physics. This is how I view this line of research I am engaged in and I think it is entirely in keeping with what Hjelmslev was proposing, in that the linguistic theory gets set up on a purely linguistic basis, but then, once set up, it can serve as an organizing device for bringing in the other material so that one ends up with linguistics very closely related to other

214

SYDNEY M. LAMÉ

sciences. It is at this point that linguistics becomes a sort of integrative focus for different areas of science. The end of Hjelmslev's Prolegomena contains an enormously enlarging view on language and man. It becomes a kind of humanism. - Yes. As I have been saying, we have the conceptual system, which seems to have the same kind of structure and formal relationships that one finds in language. This conceptual system is, for any individual, what all of his experience of the outside world gets filtered through. Try as we might, we cannot directly experience reality, because our own conceptual system intervenes. This is true of any scientist. So if we think of ourselves as scientists, it is essential to understand something about the conceptual structure with which we are operating. We have to get some notion of the properties of this filter, so that we can see to what extent our perceptions are properties of our own conceptual system instead of properties of the reality outside. I think this is one of the points which - if Hjelmslev didn't state it in just these terms at least is consistent with his theory. It is perhaps useful to clarify your conception of formalization in linguistics. What is according to you the role of notational systems ? - I believe that notation is of the utmost importance. It has a key role for any scientist, whether the scientist realizes it or not. That is a corollary of the Whorfian hypothesis - that a person's thinking is influenced by his language. As an extension of that hypothesis we could say that any scientist's thinking is influenced by his notational system. This is true even if the scientist tries not to be influenced by his notation, even if he tries to be thinking independently. His thinking cannot be independent from some information medium and generally that is his notation. Consider these rules with arrows, for example. The linguists, from Chomsky on, have felt that influence, I think, in unfortunate ways. Now, what can you do about this type of situation? What one should do is to be aware of the problem and then to devise a notational system which is as free as possible from extraneous properties. The notational system should have in it only what is essential for conveying the relationships that one is dealing with and nothing else. That means for example that any notational system which is borrowed from some other discipline is immediately suspect, because it has various properties that come from its historical background which may or may not be relevant to what one wants to describe. Take, for example, the mutational notation of transformational grammar. Where does it come from? From logic. Where does the logical notation come from ? It is a kind of refinement of ordinary writing. Where does writing come from? It is an attempt to represent phonological sequences, sequences of phonemes. And what are they? They are linear, because speech is linear. And so one has a linear notation, and the rewriting rules of transformational grammar go from left to right, just like English written discourse. But are those two properties, the linearity and the left to right direction, appropriate

SYDNEY M. LAMB

215

to linguistic structure ? Neither one of them is. They are appropriate to the manifestations of linguistic structure, i.e. speech, but they are not appropriate to the structure which underlies those manifestations. What properties do we need to talk about the structure which underlies1 We know that we are talking here about relationships and that they are multi-dimensional, not linear, and that the basic dichotomy is between AND and OR. It is inescapable. There is another basic distinction between expression and content. And so, starting on the basis of these observations I devised a notation which was new, not derived from any other science, in order to represent only those relationships which were essential. I, just like anyone else, cannot free myself from the notation with which I work. But I believe that this notation is at least appropriate to the thing I am talking about. That's the reason why you prefer 'networks' to 'trees' ? - The tree can be considered as a simplified form of network. The tree is appropriate for certain - not all of them - of the outputs of a tactics. But the tactics itself has a network structure. In other words, it has a branching in both directions. The tree is a good way to represent the manifestation or output of the tactic pattern, not the tactic pattern itself. Do you agree with the point of view that a two-dimensional graphic notation is superior to mathematic or algebraic notation ? - Yes, of course. At first glance, the algebraic notation appears to be more sophisticated, because it seems to be more formal. But I think that's an illusion. It is a sort of accident in the history of mathematics that more formal work has been done in algebra than in graph theory. But there is no reason graphic networks cannot be formalized to just as high a degree of refinement as algebraic notation. I have had some difficulty with your statement that a stratificational approach is a non-process description. I have in mind the Hjelmslevian dichotomy of system and process; both are aspects of language. In particular, I think here of the syntagmatic axis as a purely linguistic axis. - Yes, that's true. We have here two different meanings of the term 'process'. What Hjelmslev was getting at was the distinction between the OR and the AND relation; he refers to it in terms of 'either-or' and 'both-and'. It is the 'both-and' relationship which is at the basis of his 'process', which gives rise to the definition of the syntagmatic axis. Of course, in following Hjelmslev in the recognition of the basic distinction between OR and AND, I have also followed the implications of this and we have the paradigmatic types of structures as opposed to the syntagmatic ones. This corresponds also to 'system' and 'structure' in Halliday's terminology. The confusion here arises because the term 'process' is being used in different ways. I would accept the term 'process' in linguistic structure if we mean by it what Whitfield meant by that term in translating Hjelmslev. But I don't accept 'process' in the sense of mutation, as in Bloomfield's or Sapir's morphophonemic rules, in which we have

216

SYDNEY M. LAMB

one morphophoneme being replaced by another, or the kind of processes involved when we talk about deletion in transformational grammar. In a stratificational grammar we don't delete things, instead we omit things, and this is a very important distinction. If we are going to cook a turkey dinner, according to the transformational framework, what we do is we go to the grocery store and we get a turkey and a leg of lamb and a ribroast and a bunch of whatever vegetables, and so on. And we take them home, we roast the turkey and we roast the leg of lamb and we roast the ribroast, etc. And then we serve the turkey, we throw the leg of lamb in the garbage can, we throw the ribroast in the garbage can.... That corresponds to deleting. In stratificational grammar, we go to the store, we buy a turkey, we roast the turkey and we serve the turkey. In other words, if we take a sentence like "John knows how to swim, but Harry doesn't", we don't delete what might be following "doesn't", we omit it. Deleting is a process: you first put a thing there and then you get rid of it. Similarly, throughout a transformational description, one has a representation and then one performs an operation in order to change it into a different representation, and so forth. In the encoding process within stratificational grammar, one doesn't have a representation in that sense at all. The various operations work in such a way that the appropriate output comes out at the bottom. Nothing has been produced until these operations have been performed which have decided its correct order and its correct form; you produce things only at the end of the encoding process. Is your conception of language static or dynamic ? - I ' m not saying that there is no movement in language. On the contrary, new connections continually get formed within the network, and sometimes old connections get blocked; this is something which is involved primarily with linguistic change. The other kind of motion that takes place is the movement of impulses through the network. But I differ from the mutationists in insisting that the linguistic structure itself consists of relations rather than mutations. Can you say that language is generated ? - I would consider it dangerous mainly because the term 'generate' is ambiguous. If you take the term 'generative' as used by Chomsky himself within Chapter One of Aspects of the Theory of Syntax,31 you can find it used in about three or four different meanings. And as soon as you get into other authors, it takes on still more meanings. I have concluded that one begets confusion by using that term. Do you admit that the speaker has the faculty to produce an infinite number of sentences ? - Yes, indeed. What is the meaning of''infinite'' here? - I would phrase it this way: the speaker has the capacity to produce any one of an 31

Cf. note 11.

SYDNEY M. LAMB

217

infinite number of sentences, because he wouldn't live long enough to produce an infinite number of sentences. This is so because of certain properties of the linguistic system, including the fact that every tactic pattern has recursivity. In terms of the relational network tactics, recursiveness is shown quite directly by loops within the tactic network, which make it possible to go through the same structure more than once. You also have the property that new connections can be formed right within a moment. Just as a result of the exercise of thought, a new connection can be formed which makes it possible to formulate some new sentence which has never been formulated before. Another central notion in some current theories is the one of 'creativity'. It is used by Chomsky, but also by Chafe who says that the semantic structure of language is the creative part of the language. Some people claim that this is a typical 'ideological' notion. - What I have just been talking about is one aspect of creativity, i.e., the ability of people to create new sentences. They are able to do that by virtue of their tactic patterns at the various levels. But this is relatively less interesting than another form of creativity: our ability to create new idioms, which are new lexemes. This is the kind of creativity which, as I understand, von Humboldt was talking about. As far as I know, Chomsky's formulation is unable to deal with the ability to create new idioms. There is also the ability that people have to create new concepts, new collocations in which you take two ideas that have been separate and discover that they can be put together; this involves building a connection between two different points in the conceptual network. On the basis of certain general network structures that one finds in semotactics, one can see how an individual with those structures would be able to construct connections between certain points. A very simple example would be: if you are given the fact that all men are mortal and Socrates is a man, then on the basis of that, you are able through the quite general rules to build a connection such as: "Socrates is mortal" (Figure 9). That seems so simple a case of deduction that you would hardly call it creative. But a similar kind of process in a less obvious case we would call an act of creativity. In transformational grammar, the sentence is seen as the privileged unit. What is in stratificational grammar the status of the sentence ? - The sentence is primarily a lexotactic unit, but with some lack of complete confidence, I would say that we also have the sentence at the semotactic level. But an important point to make here is that at the semotactic level we also have units larger than sentences. That could be my next question: does discourse, as a unit larger than the sentence, contain some linguistic information? - Yes, I would say it definitely does.

218

SYDNEY M. LAMn Mortal (Properties of men)

"All men are mortal"

"Socrates is a man"

Socrates Fig. 9

Did you develop a kind of device to deal with discourse ? - The semotactics isn't really worked out enough to say much about that. Gleason has worked on this considerably more than I have, in a slightly different notation, but within the framework of stratificational grammar. 32 He has found a considerable amount of structure for instance in the narrative. Gleason has drawn diagrams of this kind of structure and it seems from what we have done so far that we can translate those diagrams into our kind of diagrams. Could it be very important to apply stratificational grammar in the area of poetics ? - Yes, this would be a worthwhile area to explore. You are one of the few people working very actively on computational linguistics and mechanolinguistics. - I ' m not working in that area now. But 1 have in the past. Yes. I recall having read somewhere that, according to you, computerizability is a criterion for theoretical validity. Do you still agree with this statement ? - 1 would say yes, but I would like to clarify that, because I am sure to be misunderstood. If it were the case that computers were defective in some relevant way, then I would have to reject the notion. But in fact it appears that the computer is flexible enough to act as a hypothesis-testing device for linguistic formulations. It does have the capacity to test grammar in one respect: given a grammar it should produce certain outputs and it should refrain from producing certain other outputs. Now, 32

H. A. Gleason, Jr., "Contrastive Analysis in Discourse Structure", Georgetown University Monograph Series on Languages and Linguistics 21 (1968), 39-63; reprinted in A. Makkai and D . G. Lockwood (eds.), Headings in Stratificational Linguistics (U. of Alabama Press, 1973).

SYDNEY M. LAMB

219

that kind of testing for any grammar larger than a very small size is just impossible for a human being. It is impossible to keep in mind everything that must be kept in mind. The human being's mind plays tricks on him: he makes allowances, he makes assumptions, for certain formulations in the grammar which haven't been completely specified. The computer doesn't do that. Therefore, the computer is a valuable hypothesis tester. Of course, the computer is deficient in some other respects: it doesn't deal with the appropriateness of a given text to a certain situation. There is also the psychological area. If there is no possibility of writing an economical computer program for producing sentences using a given grammatical framework, as will be the case with transformational grammar, then one is led to suspect that perhaps such a grammar lacks psychological validity. Do you believe in the possibility for the computer to simulate natural language ? - I guess it depends on what you mean by 'simulate'. One student of mine actually has written a computer system which makes it possible for a person to specify a problem to the computer in ordinary English within a limited range of problem areas. And the computer is able to take that information, decode it, construct an internal problem-description, write its own computer program and execute it. 33 To some extent language can be simulated by a computer. One can even do considerably more than has been done so far, but there are limitations. As a conclusion, I would like to ask you: in what sense can stratificational grammar be seen as an original and useful alternative within the whole of contemporary linguistics ? - You are asking me about three things. As to its usefulness, J hope I have given some idea of what I think about that in answers to the earlier questions. Now, can it be seen as original and as an alternative? Well, as a cognitive linguistic theory it is not an alternative, since it is the only one in existence. And I would not consider it an alternative to transformational grammar or to other systems for classifying linguistic data, since it is really in a complementary relationship rather than in rivalry with them. Cognitive linguistics needs taxonomic linguistics, just as the baker needs the miller. But if some miller goes around declaring that he is a baker, and if some people believe him even though he has never produced a loaf of bread, then perhaps a real baker might well be seen as an alternative. Nor is it an alternative to Halliday's sociologically-oriented theory, since we have a relationship of complementarity here also. Now as to the originality of stratificational theorv, let me just say that it is only a working out of some ideas that were already present in the thinking of Saussure, Bloomfield, Hjelmslev, Hockett, Pike, Harris, Chomsky, and Halliday. Yale University New Haven, Conn., U. S. A. November 6, 1972 33

G. Heidorn, "Natural Language Inputs to a Simulation Programming System" (Yale University Diss., 1972).

A N D R E MARTINET

André Martinet, one might consider your work a prototype of European structuralism. How do you feel about this assertion ? - I think of myself more as the representative of a functionalist than of a structuralist school. Linguistic structuralism in Europe was and still is represented by at least three fairly different trends although they have some aspects in common: those who are more or less connected with the Prague school tradition, those who have been directly affected by Hjelmslev's teaching, and finally, the Firthians in Britain. Within the first group, that of the so-called Prague linguists, we find two divergent trends : one represented by Jakobson, and the other by Trubetzkoy and myself. It was in the Phonetics Congress, in Ghent, in 1938, a few months after Trubetzkoy's death, that the opposition between those two trends materialized. Jakobson and I presented two papers which made our differences quite clear. 1 Mine was the first presentation, on an international stage, of the ideas I later developed in the field of diachronic phonology; Jakobson's was the first public profession of general binarism. To be sure, he had been entertaining binaristic views ever since Prague phonology was launched. But he refrained from publishing them as long as Trubetzkoy was alive. It is obvious that the first phonological pronouncements by the Prague Circle were largely influenced by binarism, for instance when obvious binary oppositions were favored, the others being cast aside as disjunct. Outside of Prague the normal reaction to such a procedure was one of amazement: in French, for example, the opposition between p and b, was called correlative and made a fuss about, whereas the opposition between p and t was pooh-poohed and put on a par with that between a consonant and a vowel, as, for example, in cap ~ cahot (/kap/ ~/kao/). Trubetzkoy was soon aware of the linguistic public's unfavorable reaction to that binaristic bias, and modified his initial point of view accordingly, as we may observe if we compare his 1933 article in the Revue de Psychologie normale et pathologique, and his 1935 paper, in the same journal. 2 Jakobson's reaction was to extend the preferential binaristic treatment to all proportional oppositions. Even after Jakobson settled in America, he was for a long 1

Proceedings of the Third International Congress of Phonetic Science (Ghent, 1939), 30-34 and 34-41. "La phonologie actuelle", Journal de psychologie normale et pathologique 30 (1933): 1-4; "Essai d'une théorie des oppositions phonologiques", Journal de psychologie normale et pathologique 33 (1936): 1-2. 2

222

ANDRE MARTINET

time still considered a European. In 1946 and later, the opposition between Europeans in America and Americans proper endured. The former were grouped around the journal Word; they soon established contacts with the Sapirians and the Pike group, but with the vast post-Bloomfieldian group the exchanges were less fruitful. Do you think of Bloomfieldian linguists as structuralists ? - Beyond any doubt. In spite of fundamental differences between the Europeans and the Bloomfieldians, they have many things in common. More or less consciously they study language in itself as, as least to start with, an object distinct from other objects of the perceptible world. For me, structuralism, and even Bloomfieldian structuralism, amounts to putting into practice the Saussurean program: the study of language in itself and for itself. That is Saussure's great contribution and not, as some philologists imagine, the langue-parole dichotomy. Wasn't there any direct influence of Saussure on American structuralism? Wasn't there any mention of Saussure in 1946, when you arrived in America ? - If there was an influence, it was a very indirect one. Saussure was never mentioned among Bloomfieldians. Wells' study on Saussure was so to speak an order by Jakobson. Wells accepted to analyze Saussure's Cours and he did it excellently.3 This was, however, exceptional. It can almost be said that there was no desire to know Europeans. The Americans had been hurt in their pride by the European supremacy in certain domains. The world being split in two, by the war, they took advantage of the opportunity to ignore for years the existence of European thinking and to assert their independence from Europe. This was clearly Trager's position when he was in Paris, in 1945; when some of them, like Hockett or Harris, decided to look at what Europeans had been doing, it was too late for a real cross-fertilization. What Harris finally did was to try to find a Bloomfieldian solution for some problems Europeans had stated and solved before him.... The Sapirians were closer to European thought. Bloomfield himself had been a student in Europe and he started his career as a comparatist in the line of the German linguistic school. In my doctoral dissertation on the consonantal gemination in Germanic, 4 I criticized Bloomfield for sticking too closely to the narrowest neogrammarian tradition, the one least open to the suggestion of such scholars as Schuchardt and Meillet. It is known that the first edition of Language was inspired by Wundt, but between the first version and the second, in 1933,5 there was a radical change in focus, a decisive americanization of his thought, which in a certain sense ran parallel to the evolution of European structuralist thought. The decisive point was that sound was integrated into language. Since Saussure, that has been the one decisive step taken in linguistics. Saussure was constantly on the verge of integrating 3 4 5

R. S. Wells, "De Saussure's System of Linguistics", Word 3 (1947): 1-31. La gemination consonantique d'origine expressive dans les langues germaniques (Copenhagen, 1937). L. Bloomfield, Language (New York, 1933).

ANDRE MARTINET

223

sound into language, but he never did it because he remained, to some extent, a prisoner of the psychologism and idealism of his time. The intermediate step initiated by Baudouin de Courtenay, who tried to integrate sound into language by attributing a psychological nature to phonological units, was certainly helpful. The fundamental characteristic of American structuralism is that language was studied for the first time as perfectly distinct from the human mind. Isn't the insistence on systematicity and structure in language a common feature of the different structuralist tendencies ? - Of course it is. There also, the discovery of the phoneme played a decisive role. The full consciousness of the existence of a certain number of discrete units paved the way toward a quantified analysis of linguistic units. This revolutionary step accounts for the recurring attempts to sidetrack phonology proper. That one aspect of human behavior, namely language, should evince discrete units, things that are what they are because they are not something else, is still hard to swallow; that we finally could emerge from subjectivity, from the necessity of treating all phenomena connected with the human mind in philosophical terms, is still an object of scandal. The acceptance of discrete distinctive units and its consequences was what constituted the fundamental bond between linguistics in Europe and in America. But of course, they immediately diverge. For the disciples of Saussure, there is an essential difference between units which have no meaning and units which are doubly articulated. It was a mistake not to stress that difference in my article on structuralism published in Anthropology Today.6 In America, the rejection of psychologism made people refuse to reckon with the co-existence of meaning in certain units, so that linguists operated with a unilinear gradation from sound to sentence. In Europe, we posit an essential difference between the phoneme on the one hand and the morpheme or the moneme on the other. In other words, the Saussurean sign is what distinguishes European and American structuralism. What do you understand by structure, the key concept of structural axiomatics ? - On this point, I am afraid I differ from my more vocal colleagues, whether they be structuralists or transformationalists. They assume and assert that linguistic structure doesn't exist outside of the minds of scholars who try to account for facts; in other words, that structure is invented. In my opinion, there is something in the object we can call 'structure'. Structure can be discovered; one does not invent it. 7 Of course, our divergences may be said to be of a terminological nature; and the term postulate could bridge the gap between discover and invent. Etymologically the structure of a building is what makes it stand; structure is the sum total of the pressures the different materials exert upon each other. Structure is in the stone, in the cement, in the steel frame. The identification of a structure raises a problem because you cannot see it. 6 7

"Structural Linguistics", Anthropology Today (1953), 574-586. Cf. A . Martinet, "Structure and Language", Yale Studies 36-37 (1966), 10-18.

224

ANDRE MARTINET

The structure is somewhere in some material which contributes to it, but which is not the structure itself. People with an idealistic turn of mind are apt to deny the material existence of what is not immediately perceptible. Personally, I would say that I know of the existence of a structure in the building, because the structure is what prevents it from falling down. If others prefer postulating that existence, let them do so. The next step is to give an analogical representation of the structure, in other words, a corresponding model. Now, what about structure in language? Isn't it a very daring metaphor to use the word structure here ? In Latin we encounter the term structure as applied to language : verborum quasi structura. Thus, we have a metaphor, but with a noble tradition! Linguistic structure exists in the linguistic process, in the muscles and in the minds of its users. The fact that we cannot touch it, see it, or sense it does not mean that it does not exist. Man has been cunning enough to find the means of identifying those things he cannot perceive with his senses; science enables him to go beyond perceptible data. The discussions around the term structure which might seem to amount to a mere semantic bickering are, nevertheless, very important because insistence on the unreal character of linguistic structures justifies a certain aversion for observation, which aversion leads, in my opinion, to idle discussions and sterility. Hjelmslev - Maybe we could postulate,

would probably have said that linguistic theory invents structure... he would have said that, at the start. But if we had talked it over, he and I, possibly have come to an agreement if we had found the suitable word for example.

Is the whole of European structuralism to be considered Saussurean ? Certainly, except for Great Britain where Saussure's influence was indirect and diluted. In the rest of Europe, including the Scandinavian peninsula, structuralism has been Saussurean. Saussure has always been invoked. Trubetzkoy for instance, referred to the Saussurean dichotomy langue-parole when he tried to justify the phonological point of view; in doing so, he was wrong, because phonological relevance versus its opposite does not coincide with langue versus parole. He just thought that if people were addressed in their own language, they would perhaps understand better. As for Hjelmslev, he accepted only one master - Saussure. While reading the Cours and the Sources Manuscrites, 8 one recognizes a genius' conception of language, although it is sometimes so contradictory and hesitant. - If Saussure did not write the Cours himself, it is largely because his thinking was still in progress. He probably felt he did not have the time or the strength to carry out his program: the study of language in itself and for itself. Godei 's outstanding contribution was to show the progression of Saussure's thinking from one course to the 8

R. Godei, Les sources manuscrites du Cours de linguistique générale de F. de Saussure (Genève, 1957).

ANDRE MARTINET

225

next. It is now possible to draw a curve which might allow extrapolations, so to speak. It gives depth to Saussure's teaching. In Engler's book, 9 we listen to Saussure thinking aloud. As it stands, we find in Saussure's teaching justifications for many different approaches. Hjelmslev had a right to say that he found in Saussure the rejection of substance, and the functionalists like myself are also justified when they say that Saussure did not reject substance. "The train leaving Geneva..." is never the same ; but there would be no train without an engine and cars. It is not the same train, but a formed substance is always there. Thus you don't admit that Hjelmslev''s is the most correct interpretation of Saussure? - I don't. There are two Hjelmslevs, at least the one before 1951 and the one after. After a very serious illness, in the early fifties, he reconsidered some of his fundamental tenets - there are crucial differences between the Danish version and the English versions of the Prolegomena.10 In the English version, formed substance is considered to be a linguistic fact - under those conditions there is no fundamental difference between functionalism and glossematics. Obviously Hjelmslev's construct remains quite original, but.... - I had long talks with Hjelmslev about the question of substance; to my objection as to how one could identify forms without reference to substance, he answered that he did not care - they will always be identified, one way or another. We all work that way - when we establish our theory, we just leave out certain elements of reality. Hjelmslev did not want to consider the problem of identification. His theory as such was admirable, but as soon as he tried to apply his theory to a particular problem of a particular language, everything got confused. To be sure, he is not responsible for Togeby's application of the theory to French, 11 but in all the applications he attempted, we find some traces of juggling not so very different from Togeby's. If we consider what happens in contemporary linguistics, we often get the impression that a strong theory entails weaknesses in its application and vice versa. - Of course. A theory always looks nicer when you do not let facts embarrass you. But in the end, you have to pass from theory to fact in order to verify the theory in all its aspects. It is assumed that a theory remains valid until the number of facts that cannot be integrated in the theory becomes so considerable that it has to be revised. A linguistic theory, like phonology for example, may not be perfect in all its aspects when applied to a language: do we have, in a given case, a variant or a phoneme? Do we have one phoneme or a succession of two phonemes ? As long as the facts that are not accounted for by the theory may be considered marginal, it is acceptable. But 9

Cours de linguistique générale, éd. critique par R. Engler (Wiesbaden, 1967-1968). Omkring sprogteoriens grundlaeggelse (Copenhagen, 1943) ; Prolegomena to a Theory of Language, Supplement to International Journal of American Linguistics 19 (1953): 1 (second revised edition [Madison, Wisconsin, 1963]). 11 K. Togeby, La structure immanente de la langue française (Copenhagen, 1951).

10

226

ANDRE MARTINET

as soon as the shortcomings outnumber the fundamentals, its very basis has to be reconsidered. You do not agree with opposing the formalizing, Hjelmslevian interpretation to your own work, do you ? - My approach and Hjelmslev's have many points in common: first and foremost, the Saussurean basis common to both of us. When Hjelmslev, in 1935, on the occasion of the London Phonetics Congress, decided to launch a new linguistic theory 1 ' 2 he expected me to join. But from the very beginning I had serious reservations : I am basically too much of a substantialist. It is true that my assent was expected because of certain things I had written. My remarks on a so-called aspirated h in French had made an impression on him. 13 He was fascinated by the notion of something that did not correspond any longer to a physical segment but still manifested itself in its neighbor's behavior. Under the term latency it was to play an important role in his theoretical thinking. As soon as you confront us both with the Americans, you immediately sense what we have in common. Look at their inability to understand an opposition that disappears, a neutralization, a syncretism - all of which play an important role with Hjelmslev as well as with the Prague people. A Bloomfieldian is perfectly incapable of understanding the notion of something that has linguistic value in some positions and none in others. His conception of the phoneme lies in Jones' tradition: a family of sounds; when you put sounds together, you get a phoneme; when you put phonemes together, you get a morpheme, and so forth. One sticks to surface reality because one dares not venture into the unknown. When 1 pin down a phoneme by opposition, I consider that I am moving on the most empirical level of reality - in Anglo-Saxon tradition this is already reaching for the sky. Their empiricism is terribly demanding... - The most dramatic fact about this empiricism is that it was bound to entail a violent reaction, transformationalism, which overstresses the opposite aspect. The middle road, which avoids those side-steps, is the Saussurean road. Anyway, adopting a clear position regarding the phonic substance was necessary to found structural linguistics. We talked at length about structural linguistics. But there are also the Analogue'' structuralisms. All human sciences now call themselves ''structural'. - I do not pretend to be well informed about structuralism when it is not linguistic. I only think that for historical reasons, 'philosophical' structuralism... ../anthropological'. Let us remember Lévi-Strauss. - 1 say 'philosophical' in order to cover the various domains of human thought. 12

Proceedings of the Second International Congress of Phonetic Sciences (London, 1935), 49-57. "Remarques sur le système phonologique du français", Bulletin de la Société de Linguistique 34 (1933), 191-202. 13

ANDRE MARTINET

221

The type of structuralism currently in fashion is derived from Jakobson via LéviStrauss. It is structuralism as interpreted or 'felt' by Jakobson - I am stressing 'felt'. But isn't Jakobson's structuralism the reflection of his temperament, more than anything else - the insistence of a universalist who wants to reduce everything to identities at once, even before having found différences? It is not fortuitous that linguistic universals were introduced by Jakobson. The Aginskys' paper, 14 which was the first to mention linguistic universals, had been ordered by Jakobson. It took some time for the seed to bear fruit, but, ever since, what has been produced is in the same vein. From the moment that everybody began to speak about universals I saw the shadow of Jakobson. They all yielded: even Hockett contributed to Greenberg's book. 1 5 Greenberg himself, who often opposed Jakobson when he and 1 were coeditors of Word, also swallowed the bait. When Lévi-Strauss uses linguistic models to build his anthropology, does he make the wrong choice ? - I should say that his usages are metaphorical. When Lévi-Strauss says that there is communication in anthropology as there is in linguistics, that the means of communication in linguistics is words, whereas it is women in anthropology, I find this cute, as does everyone else, but it is a metaphorical use. The problem of interdisciplinarity is precisely one of metaphorical uses. People forget that you may well get your inspiration from another branch of knowledge, but that, before you make use of a concept, you have to see what it amounts to in the framework of your own relevance. The overt anti-historicism of contemporary structuralism shows that what was done in structural linguistics with regard to diachrony - from the beginning it had been stressed that structuralism was not contrary to history - was not understood. As early as the late twenties, the incompatibility of structure and evolution had been rejected in Prague. But, for some personal reason, Jakobson chose to ignore it for a time. A clearer example of 'metaphorical structuralism'' is perhaps Jacques Lacan's which invokes the Saussurean sign... - I prefer not to dwell on that subject. 16 It is probably always useful to look at what is done in other disciplines. However, one should never operate with terms, but with basic notions. The main contribution of structural linguistics to the study of human thought is the notion of relevance. In a scientific study it is necessary to find a principle which enables one to select in the observed reality the facts which are relevant to the object of study. That is the foundation of science. In the physical sciences, man did not have to make a choice before distinguishing between physics and chemistry; it seems that nature chose for the scholars. But with regard to linguistics, since the advent of structuralism, it is well known that one has to choose a standpoint, linguistic 14 15 16

B. W. and E. G. Aginsky, "The Importance of Language Universals", Word 4 (1948), 168-172. J. Greenberg, Universals of Language (Cambridge, Mass., 1963), Cf. G. Mounin, Introduction à la sémiologie (Paris, 1970), 181-188.

228

ANDRE MARTINET

relevance. From then on, the existence of a scientific object is recognized, distinct from the rest of the world. The moment has come to define the object of linguistics. Do we have to use the langueparole dichotomy, langue being for Saussure the object of linguistics ? - Saussure, for all his genius, was a man of his time. He could not escape being influenced by the surrounding psychologism, and also the teachings of sociologists like Durkheim. For him, langue is a sociological reality with an existence of its own, independent of the individuals who practice it. It is clear that the notion was never thought out. Truly amazing, for instance, is the statement that the sentence does not belong to langue, the sentence, which is, so to speak, the locus of syntax. Saussure, in fact, is wavering. Just re-read the Cours; it is amusing to notice that those who have little understanding of Saussure's message are the most eager supporters of the langue-parole dichotomy. Actually, the 'circuit of speech' is the basic linguistic fact; at each point of that circuit we find linguistic reality. It amounts to a transfer of information from one person to another and back. There is a loss, but it is precisely that loss that shows the nature of linguistic structure. Language functions at the price of a loss - the poet may try to go beyond language, to express more than language; but language, organized as language, with its double articulation, supposes from the start a limitation of what can be communicated - only what can be said is said. Discrete units are at work here. A language segments the world in a certain way; our representation of the world is at least warped by our language; we are emprisoned in language, and it is in order to escape from their prisons that poets write poems. Our need for communication is limited by language and still, without language, we would be lost in the mist. Living in the real world means accepting frontiers, accepting the fact that communication is shaped by its limitations. Linguistic communication never can be total communication. Since we communicate by means of speech, every linguistic fact must be present in the circuit of speech. One could argue that two persons of the same linguistic community must have in their minds some traces of analogous experiences which could be suggested without being transmitted by the circuit proper. To learn a language under conditions peculiar to a community may entail certain types of experience that are common to all persons speaking the same language; those elements of experience could be considered part of the language, without actually corresponding to some element in the circuit of communication. But I do not think that this is the right picture, because linguistic facts proper are only such because they have been verified by language itself, by their passage in the process of speech. Some facts escape that verification; the ones I call connotations, which become language facts only if shared by different people. Connotations can be shared, i.e., communicated by stylistic devices, which are new, unexpected, literary or poetic contexts. What does it mean that linguistic facts have to be verified by language itself? - Let us say that a word gets its meaning, its denotation, not from reference to

ANDRÉ MARTINET

229

external reality, but from its various uses in the various contexts. In other words, the denotational meaning of the word, the only meaning common to the whole linguistic community, is the meaning which is finally established through the uses of the word in question in certain contexts. Reference to the world outside of language can only create connotations. And there, the post-Bloomfieldian definition of meaning is adequate - the meaning of the word is the set of the contexts where the word is used. Of all the learner's experience of the fact that we heard the word at a certain time in a certain situation, only that is kept which could be honored by all the contexts in which we heard the word later on. It can then be said that there is nothing in language that has not gone through the process of speech. The distinction between langue and parole has thus become superfluous. Langue is the whole process. This certainly is an extension of the Saussurean notion of langue... - No doubt. I actually start from Saussure's assertion that language is all that counts in speech. The rest consists of fringes, margins, connotations. One could conceive of a linguistics integrating these margins, but it is preferable not to do so at this point. What has to be understood is the fundamental fact of the functioning of language, the existence of the process of speech, the verification of all linguistic facts through linguistic contexts. This should account for the objectivity of linguistics as a science. - Right. I would like to draw your attention now to a criticism formulated by some transformationalists - that the objectivity of structural linguistics would be based upon an illusion, since only a minute part of the linguistic capacity of the speakers is dealt with. That linguistic capacity is immensely richer, since it enables the speaker of the language to create an infinity of sentences. - But what is this infinity? That seems to me a problem of elementary mathematics. Generative transformationalists are people who never operate with the results of a complete analysis. With such an analysis, one comes upon minimal units. I understand quite well how in America, linguists could be induced to drop the notion of minimal units. The notion of morpheme was doomed when people insisted on making it coincide in all cases with a segment. At Ann Arbor in 1959, after a meeting of the Linguistic Society, I went around asking some ten colleagues how they would segment some phrases. That inquiry revealed there was no agreement at all about the way to segment an utterance into morphemes. That could explain why segmentation was ultimately largely pooh-poohed. The solution, of course, lies in presenting the minimal unit corresponding to some modification in the utterance, even if the exact place where the modification occurs cannot be precisely localized. The minimum unit thus obtained, I call moneme. But as soon as you admit the existence of minimal units, where then can creativity be hidden? If one

230

ANDRE MARTINET

conceives of a language as composed of minimal units which can be combined according to habits prevailing in that language, where then is creativity? Is the number of sentences produced by the speaker infinite ? - It is not infinite. The number of monemes in a language is not stabilized, but it is a definite number, even if one cannot overcome the difficulties in determining it exactly. Then there is a definite number of restrictions to the compatibilities of moneme classes. Supposing that the average length of an utterance is fixed, one can compute the number of possible combinations. Starting from discrete numbers, one cannot but obtain a finite number. Of course it will be an extremely large number, the more so since I am not excluding unusual combinations of the colorless green ideas type. Therefore it is not difficult to understand why an ape, sitting before a typewriter, does not have the slightest chance of rewriting the work of Shakespeare. Where, in all this, lies the mystery of creativity ? You consider creativity in linguistics a psychologizing notion, do you ? - A child that has never heard the word 'uncoverable' can construct it; simply because there exists a derivational moneme -able which can be used in the combination. Indeed Saussure invented 'undecorable'. If that is a creation, there is nothing mysterious to it. Show me a creativity that could not be explained in terms of monemes and compatibilities. If you can show me that beyond monemes and compatibilities there is something else, I am ready to investigate the problem. Of course, there are linguistic behaviors that can lead to the unexpected. But in that case, people take some liberties with the language and with tradition. They go beyond what could be described as their competence. What is objectionable is toying with a term without ever actually defining it. Being scientific is not juggling with nicely sounding words but trying to give precise definitions of useful words. You have dealt mainly with phonology... - Allow me to interrupt you. I have dealt with other linguistic items than phonemes. It is so much simpler to brand someone as a phonologist and disregard what he does in other fields. When 1 published the Elements of General Linguistics, in I960, 17 some people hated me for it because it was becoming more difficult to say: "he's just a phonologist". When I say 'phonologist', I simply means that you are not a semanticist. - I am interested in language facts and as soon as one approaches semantics, contact is at once established between the extralinguistic and the linguistic world. I am a phonologist and proud of it. Phonology was indispensable so that linguistics could become a science. Before we could think of general linguistics as a science we had to 11

A. Martinet, Eléments de linguistique genérale (Paris, 1960).

ANDRE MARTINET

231

become aware of the existence of a kernel of linguistic facts conceivable without reference to non-linguistic reality and which exist in virtue of their mutual oppositions. Those, you find in phonology and grammar. If you are interested in the meaning of words, you will soon discover that you have become a gardener, a chemist, or a philosopher. Should we drop the notion of meaning as was done by Bloomfield? - No, but I shall simply state that it is difficult to proceed as scientifically on the semantic level as in phonology or syntax. It has to be done though, but, owing to my training, I am accustomed to handle very central and structured aspects of language and 1 have not so far felt tempted to deal with semantics. What others have achieved on this level does not convince me that we are in a position to do much better than our predecessors. 1 feel that even the best among transformational scholars have not got beyond what the better dictionaries have to offer. How can we define the linguistic notion of meaning ? - My approach to this notion is an indirect one. 1 distinguish denotation and connotation, denotation being the meaning that is shared by the whole community, established in reference to linguistic units, and not to the outside world. This use of 'denotation' and 'connotation' does not coincide with that of Hjelmslev or Barthes.18 I relegate 'connotation' to the margins, to the realm of thought, not of language. Some problems at least are thereby somewhat clarified. But I am afraid it does not lead to a scientific treatment of meaning as a whole. Some linguists believe that a syntactic difference is at the same time a semantic difference. The relation between syntax and semantics would then be vital. - It is essential, in order not to mix things up, to distinguish between syntactic types and lexical types of combination. Between syntactic patterns, on the one hand, which make possible the classification of monemes, and, on the other hand, differences in meaning within those classes of monemes. Of course, some lexical combinations are normal, frequent, because they correspond to realities outside language and thus often impose themselves upon human beings. Other lexical combinations are exceptional, and may never materialize simply because they correspond to exceptional situations. But we need only begin a sentence with "I dreamt that...", and any lexical combination becomes acceptable and is linguistically valid and real. In doing so, we are still using ordinary syntactic patterns. Syntactic patterns are permanently present in all lexical contexts whereas modifying a couple of elements in an utterance is enough to cause the lexical combinations to become abnormal although linguistically possible. As far as I know, transformationalists hesitated a long time before taking sides on this question. If we want to see things clearly, we should never shy from 18

Cf. J. Molino, "La connotation", La linguistique 7 (1971): 1, 5-30.

232

ANDRE MARTINET

analyzing linguistic facts, nor from pointing out where our analysis, if carried out, would lead to arbitrariness. It is obvious that there are border-line cases in all languages which have to be solved in an arbitrary way. Every methodology, every explanatory theory will yield marginal facts if this theory is correct and accounts for the reality it has to describe. The methods that are best adapted to facts will yield data which will not fit in, but that is no reason for rejecting a syntactic theory. This is exactly what those people do who maintain that there is no difference between syntax and semantics. Do you believe in a discourse analysis ? Such a discourse analysis would be essentially semantic... - With this type of research we venture outside the field of linguistics proper; we come to the world itself. My reservations as to semantics are caused by the fact that with this kind ot research we arrive at the world as it is, as we perceive it. It is fascinating, but we cannot do everything. That would lead us to a science of literature in the broad sense of the term.... Is there perhaps a science of political discourse? - Why not? If the people who conducted such a research were trained in linguistics, I would be delighted. What we linguists can contribute to our culture is precisely this: giving the people who are interested in human problems the feeling of what language is, what its limitations are with regard to the world. But it one considers that one is doing linguistics from the moment one studies literature, then there no longer is any linguistics. Do you think the notion of discourse is operational in linguistics ? - I once wrote 19 that there is nothing beyond the sentence which is not already within the sentence; the Spaniards are quite right in using the term oracion in reference to the sentence as well as to the discourse. I still believe that from the linguistic point of view, we have to stop at that point. Of course, there are connective elements between the sentences which linguistics cannot discard altogether. But even if you have a sentence where some elements are referential to something that has been said in another sentence, in another part of the work, those elements have to be defined linguistically and not in relation to whatever they represent elsewhere, in other terms, in relation to their function in the context of the sentence where they are found. Pronouns are the favorite domain of those who cannot be satisfied with linguistics as an analysis of linguistic structure itself. Are you referring to Benveniste ?20 - Benveniste approaches such problems from a metaphysical standpoint. There 19 20

"Réflexions sur la phrase", Language and Society (Copenhagen, 1961), 113-118. Cf. E. Benveniste, Problèmes de linguistique générale (Paris, 1966).

ANDRE MARTINET

233

would be no language, no humanity, without the "I". We are no longer in the domain of linguistics but in that of... ...ideology? - Right. Those children who speak their language without using the pronoun "1" until they are four or five years old and refer to themselves as Johnny or Billy show how speakers can get along without "I". Jespersen mentions a German philosopher who did not celebrate his son's birthday, but the anniversary of the day when he had said ich for the first time. I don't think a true linguist would ever do the same. Pronouns are no doubt often used for establishing some relation between one part of the discourse and another. But linguistically, pronouns are to be conceived as units that permit one to evoke, in a certain context, something that is not present in that context. There is a case when the pronoun does not refer to anything outside of the sentence where it appears. That is when he, for instance, refers to something that has been seen, perceived, but not mentioned linguistically, so that it functions linguistically only within the context where it appears. Your linguistics has been criticized as taxonomic... - That is just a case of giving your dog a bad name. If you market a new product, the best policy is probably to convince your prospective customers that your predecessors' wares were equally bad. Postal's review of my Elements of General Linguistics,26 a review which I do no want to qualify, was meant to show that Martinet's teaching was, after all, no different from the Bloomfieldian abomination. When I was in America, I was naturally considered by Bloomfieldians as an opponent, and when transformational grammar was launched, some people ventured to say that they found in it some echoes of Martinet's previous criticisms. This, of course, could not pass unchallenged. Of course I am not a taxonomist; 1 have been teaching and writing for years, that, in phonology, the position of the units is regularly relevant; in syntax, we must always check whether position is relevant or not. In "The man on the bench is my friend", "man" and "on", "the" and "man" are in contact, but it is obvious that the contact of "man" and "on" does not entail anything and has syntactically no relevance at all; as to the respective positions of "the" and "man", they are determined by tradition and have no informational value whatsoever, whereas the position of the subject before the verb is relevant and informational. I am for distribution when it is relevant, but 1 reject it when it means putting together elements which belong to different levels. There are other means besides distribution of securing syntactic relevance. Postal also blames you for your corpus based approach to linguistics. - Here again I am being held responsible for what other people did. It is clear that 21 P. M. Postal, "Review of Elements of General Linguistics by A. Martinet", Foundations of Language 2 (1966), 151-186.

234

ANDRE MARTINET

post-Bloomfieldian practice with regard to the corpus was untenable. The identification of the corpus as such with the language was ridiculous. The corpus was sacred.... Again the same empirical mentality... - The corpus was sacred. They never wondered whether the corpus was homogeneous or not. Whatever was there had to be submitted to the same treatment: normal speech, interjections, foreign words, nay, coughing. A young linguist in New York once attributed aspirated s and i's to Judeo-Spanish; actually they belonged to two English words used in the corpus by the bilingual informant. Yet abusus non tollit usum. A corpus is always useful, often inevitable, and linguists must learn how to deal with it. Replacing corpus by competence amounts to restricting a branch of scientific enterprise to a part of mankind, to saying, for instance, that in order to study English you have to be an English native speaker. Linguistics is a science only insofar as any human being with a suitable training can approach the study of language. It goes without saying that sanctifying the corpus is ridiculous. A corpus is a means and not an end. It is quite obvious that any one studying a language he knows well is justified in making some use of his own linguistic experience. But it will always be safe to check with informants who have no reason to be biased. The real thing is always more credible than what has been made ad hoc. The justification of those who made the corpus sacrosanct was, of course, that they wanted to devise a method that everyone could apply. It is a fact that observation can be warped by the desire to support one's theory. Therefore, with that American habit of aiming at mass production, they devised a method that any individual could apply. In Europe, we have a somewhat more aristocratic approach to research. We assume, maybe wrongly, that scholars will try not to let their prejudices interfere and taint the result. Actually, operating with an open corpus and checking one's hypotheses with informants is the method I would recommend. Do you admit that the role played by the inductive phase is your work is more important than for example in constructivistic theories like glossematics ? - It is hard to tell. It is true that I am more tempted than many to apply my theories to facts. It is also true that I don't feel as free to formalize as some others, largely because my aim is to study languages and not to widen the field of linguistics. But my approach is largely, although often implicitly, deductive. 1 start from a definition of language as an instrument of communication, with a double articulation and a vocal nature. All of this is, of course, based upon observation. Experience points to the existence of certain characteristics in all objects usually referred to by the term language, and I include these in my definition. From now on, 1 shall call language whatever fits that definition and refuse to call language an object that does not; I shall try to determine what is implied by each one of the three phrases 'instrument of communication', 'double articulation' and 'vocal nature'. These, plus their implications, are my universals, if you want to put it that way. It should be clear that, with

ANDRE MARTINET

235

such a definition, I try to deviate as little as possible from general usage because experience has shown that, in scientific usage, it is dangerous to deviate too much from general usage. And, indeed, I know that if 1 use 'language' according to my definition, most people, including linguists, will agree with me. I am not propounding a hypothesis about languages; I am not hypothetico-deductive, but empirico-deductive. In some natural sciences, in nuclear physics, for example, there may be no direct access to perceptible objects. One must present a hypothesis and, then, verify whether observable phenomena make sense within the framework of the hypothesis. But when dealing with language, an object we know quite well, since we approach it in a rather direct way, there is no need to hypothesize.... It is true that in comparative linguistics we have to resort to ad hoc hypotheses in order to explain certain facts; in a way, we could speak in such a case of a hypothetico-inductive approach, since it must be verified whether facts confirm the hypothesis. In contemporary descriptive linguistics, the object we propose to analyze can be observed, the characteristics of what we decide to call a language must be established, and then we must deduce the implications of those characteristics. We have come to the heart of our dialogue. Let us take one by one the three characteristics of your definition of language. Language is an instrument for communication, that is your functional view of language. In what way is language instrumental and in what way is it instrumental for communication ? - When we hear the word 'instrument', we immediately think of a hammer, a corkscrew.... Of course, language is more complex, and I should confess that my use of the term instrument is somewhat metaphorical. Still, language is something which is instrumental for communication; of course, it is instrumental for other purposes as well, but it is a language only if it is instrumental for communication. Of course there are languages that can survive for liturgical purposes and, at that point, communication ceases to exist. But everyone will agree that this is a residual use. So language is instrumental for communication. Our definition does not exclude in the least the possibility that language may also be something else. It merely implies that I refuse to identify an object as a language if it is not, in essence, instrumental for communication. An instrument that would serve only to help people think would never be a language. Let's define communication as the transfer of experience from one person to another. This, of course, does not imply that the transfer is complete. What is the difference between information and communication ? - 1 use the term information in the sense generally accepted today. A phoneme, for example, is a contribution to information without directly contributing to the communication of experience. If information is defined as the elimination of uncertainty, then the problem of communication does not coincide with the problem of information. The terms are not on the same level; one doesn't exclude the other; they are different. I want to stress that one of the important functions of language is the func-

236

ANDRE MARTINET

tion of expression, which must be perfectly distinguished from the function of communication. Expression is production without impartation. We have pure expression when a man talks to himself. The examination of the strictly expressive uses of language shows that the function of communication is essential since we act, in such cases, as if we wanted to communicate with ourselves. In that case indeed, one speaks as well as if we wanted to communicate. In other words, the communicative uses are those that are decisive for the functioning of language and its evolution. You emphasize the communicative function of language, Chomsky its expressive function. - I am just following Saussure in that respect. Chomsky is not. Let us talk about the other elements of my definition. Many linguists seem tacitly to agree with me that language has a double articulation. At any rate among Europeans who operate with the two faces of the linguistic sign. And, of course, glossematists would take exception to such a formulation, since they are looking for parallelisms in language. People free from isomorphic prejudices soon recognize that there are, on the one hand, units with one face and, on the other hand, units with two faces. Hjelmslev was not ready to see that because he favored the parallelism between the two planes. He wanted to find figures of content in language, as he had found figures of expression. Because of his intellectual preferences, he progressed in a direction that made it difficult for him to tackle linguistic reality. If Hjelmslev's and Togeby's failure in those matters is so patent, it is because they were looking for symmetry where there is none. Some people just cannot accept the idea that the same linguistic object is liable to contain, concurrently, simple and complex units. A simple anecdotal question. Did you discuss the problem of the isomorphism between the planes of expression and content with Hjelmslev ? - No, 1 did not. It took me a long time, if not to understand, at least to get used to operating with isomorphic planes. When I presented the double articulation theory for the first time, 22 it was in a miscellany presented to Hjelmslev. Accordingly, I was not tempted to insist on our differences, and the opposition of double articulation to isomorphism was not stressed as it should have been. This was done later in my Oxford lectures, 23 where I dwelt upon the differences and showed how one and the same utterance should be analyzed from the point of view of double articulation and from the standpoint of the isomorphism of the two planes. Where does that term 'articulation' come from ? - I just used the term in the sense in which it is commonly used although it is not perfectly clear what people mean when they say that human language is articulated. 22 "L a double articulation linguistique", Travaux de la Société Linguistique de Copenhague 5 (1949), 30-37. 23 A Functional View of Language (Oxford, 1961), 39-43.

ANDRE MARTINET

237

I assume that the articulation of sounds by the organs of speech is no part of double articulation. - No, certainly not. As far as possible, we should avoid giving the same term two different meanings. I have done it in the case of 'function'. I speak of the communicative function of language, and also of the function of a moneme. It is just a case of ordinary polysemy. Yet some critics objected to it. But 'articulation' has no double meaning for me. Some do not approve of my applying 'second articulation' to the phonematic articulation and 'first articulation' to what comes in the second place in linguistic analyses. What are the reasons for this order ? - My starting point is the transmission of an inarticulate experience which has to be articulated linguistically, first into monemes; then the signifiant of each moneme is articulated into phonemes. Historically the structural approach started with phonology and, from there, we came to syntax. Therefore people want phonology to be the first articulation. If, on the other hand, we view language as a means of conveying experience, experience is thought of as an unanalyzed whole and the first segmentation will be established on the level of this unanalyzed whole. We could loosely say that, prior to being communicated linguistically, a headache is articulated into monemes. We might conceive of a tool for communication that would not partake of the first articulation, if this tool were not a tool meant to convey any message. Think of traffic lights. It is because languages must be able to convey everything that one cannot be content with a tool where each communicative act would be an unanalyzable grunt. From the point of view of the general structure of language and with regard to the communicative function it is obvious that the first articulation is the articulation of experience into monemes. Your formulation of the theory of double articulation is remarkably clear. It is a real contribution... - I think it had to be formulated. Why didn't Saussure ever reach that point? - Saussure didn't quite arrive at the phoneme as a linguistic unit. At many points, you think he is about to take the first step, that he will finally tell us that the phoneme is an element of langue. But it never materializes. For the people of Saussure's generation, sounds were so definitely outside of language that such an integration was unthinkable. At what point and why should the concept o/linguistic economy24 be introduced? - I will come to that in a moment. Let us first take the third characteristic of my definition of language - its vocal character. This is the one feature many linguists are 24

Economie des changements phonétiques. Traité de phonologie diachronique (Bern, 1955).

238

ANDRE MARTINET

reluctant to accept. Hjelmslev, among others, always insisted on the necessity of disregarding phonic substance. In fact, he wanted to eliminate linearity from linguistics. Thereby he made a useful contribution inasmuch as he took a position against the most objectionable feature of distributionalism. When approaching syntax, it is preferable to forget about the linearity of speech and to reintroduce it only when order is relevant. Still it is the linearity of discourse which poses the syntactic problem, as we find it. Let's suppose language operates with two dimensions, that it is pictorial, for example. In that case, we will not be bound to perceive the elements successively. If linguistic communication took place in the form of a picture there would perhaps be a 'syntax' but no longer a syntactic problem. Given four elements occurring together in the experience of whoever wants to communicate, these elements could be arranged in the picture according to their actual place in reality. It really is the vocal character that imposes the linearity of discourse. It entails the auditory perception of successive elements and creates the conditions of syntax, as we know it. If language were pictorial, we would have all possible positions and not only the positions before and after. What is the relation of writing to language ? - The recent evolution of mankind has brought together two skills which were at the start completely autonomous: graphic or pictorial representation and vocal language. People who refuse to accept that language is basically vocal are, in fact, at bottom, afraid of being faced with a linguistic nucleus distinct from the rest of human experience. The relations that are present today between writing and vocal language represent a gradual adaptation of the pictorial to the linguistic form of the first and then of the second articulation: of the first, when writing becomes ideographic, and the drawing corresponds to a moneme; of the second, when writing becomes alphabetic and the drawing corresponds to a phoneme. Historically, this amounts to making pictures serve the needs of language. That writing may acquire a certain autonomy from spoken language is an interesting fact, but it must not incite us to confuse things that are distinct. Let's come back to your notion of linguistic economy. - It is derived from my experience as a phonologist. Quite naturally, I finally extended it to other domains of language. It is a really original notion in phonology only. It is true that Zipf applied it to language as a whole. I probably haven't sufficiently stressed my debt to him in my writings. But Zipf was in no way a phonologist. I was extremely interested by his book Human Behavior and the Principle of Least Effort,25 which I reviewed in Word;26 but at that time my conception of phonological economy was already established. It had taken shape, not without pains, toward the end of the 1930s. The first public presentation of that concept appeared in an article published 25 26

G. K. Zipf, Human Behavior and the Principle of Least Effort (Cambridge, Mass., 1949). Word 5 (1949), 280-282.

ANDRE MARTINET

239

in an issue of Travaux de Linguistique de Prague, in memory of Trubetzkoy. 27 This article contains my later book Economie des changements phonétiques in a nutshell. My predecessors did not speak of economy but of harmony. They believed there was a tendency for all systems to become more and more harmonious. That is the way Trubetzkoy, among others, put it. It proved very difficult to bypass this notion. It took me a month to compose that article, and today I am aware of the fact that no one could really understand it at the time. Communication with my readers could only be established when I published a series of illustrations in Word and after I had had time to compose a book, Economie.

What is particular to the notion of economy when compared to that of harmony which you just mentioned? - That is precisely functionalism. I didn't realize at once that I was a functionalist. Trubetzkoy and Biihler provided me with the notion of relevance. In the case of language, relevance is communicative, which means that linguistic elements do not have to be analyzed and classified according to their physical nature but according to the role they play in communication. When I began to study the economy ot phonological systems it became apparent to me that phonology is concerned with functions. It deals precisely with those physical realities that have a function in the communicative process.

But what are the connotations of the term economy ? - 'Economy' is sometimes taken in the sense of parsimony. In that sense, it applies to cases when there is less effort for an equivalent result or an equivalent effort for a better result. But in the framework of my own theory, it has pretty much the same value as in 'political economy', that is to say, it implies that the functioning of language is determined by the communicative needs of man within this framework. Speakers tend - not automatically, since there are restrictions of all kinds - to correlate the energy spent with the communicative goals they are aiming at. Of course there are many intervening factors. Play, tor example : people can make a playful use of language and, obviously, that part of the energy that is spent for playful purposes is lost for communication. There is also the pressure of tradition which increases the expenditure of energy, e.g., when the child or the adult is not allowed to adapt the form of the signifier to the informative value of the term. Many generations may continue to use a term that is not economical in the sense that it is heavier or more complex than its great frequency would require. I am thinking of the extreme case of the American phrase elevator operator for what the French call liftier, a phrase that sounds comical even for those who use it. 27

"Rôle de la correlation dans la phonologie diachronique", Travaux du Cercle de Linguistique de Prague 8 (1939), 273-288.

240

ANDRE MARTINET

You were talking about language universali in relation to the three characteristics cf your definition of language... - Personally 1 never use the term universals, precisely because it is used by people who, following Jakobson, see only identities and never differences. Transformationalism, obviously, tends to justify the point of view of those who want to ignore differences. It might be inaccurate to say that transformationalism resulted from research connected with mechanical translation but the latter certainly played an important role in its spread. It you want to bridge the gap between different linguistic structures, you just transform them so as to bring them nearer to each other until you reach the experience which was communicated and this you call 'deep structure'. It is a fascinating term that entrances people. Nobody actually knows what deep structure is, as is clearly demonstrated by the divergence between the Chomskyan and Fillmorean points ot view: Chomsky considering the subject as an essential element of deep structure, whereas Fillmore denies its existence. Fillmore was right to question the existence of the subject in deep structure. But his own construction is just as arbitrary as the one he rejects. I am convinced, for my part, that it is the linguist's duty to look for differences and to avoid the formulation of panchronic laws, unless they can be deduced from the definition we give of 'language'. If certain coincidences are frequently observed, they must be brought to attention; but it is just silly to call them laws. Take for example a panchronic law, presented by Jakobson, according to which a language cannot combine a distinctive place of accent and phonological length.... Yet the two features happen to co-exist in Franco-Provençal dialects: bére is opposed to bere and bólla to boia.28 And there goes your panchronic law.... Let's tackle a few methodological questions. What is the criterion of linguistic method? Chomskyans say it is introspection, linguistic intuition.... What is the alternative? The observation of the facts? - Observation supported by experience. We must, of course, establish a theory, but it must, at the same time, be founded on experience resulting from observation and be the framework for observation. The framework we establish on the basis of experience is relevance. One and the same object could be considered from different points of view, i.e., choosing each time a different type of relevance. In the case of language, one could imagine other types of relevance than the communicative one; a purely esthetic relevance, for example, that whereby we could rate the value of different languages for opera singers. But, of course, that is not the type of relevance that linguists will choose. Experience having revealed that language is fundamentally shaped by communication, communicative relevance will be chosen, and the theory of linguistic description which will be established on that basis will be the framework for observation. Why should we disregard experience in establishing our axioms? Basing axioms on experience is what is done in all sciences, except maybe when access 28

La description phonologique, avec application au parler franco-provençal d'Hautevilie (Genève, 1956), § 5-37 and 7-5.

(Savoie)

ANDRE MARTINET

241

to the object is impossible and when, consequently, the starting point has to be a hypothesis which seems intuitive to those who propound it. In the last analysis, a hypothesis, unless it is wild, cannot but be inspired by experience. Dont you think that a speaker is always more or less conscious of the relevance of his own linguistic system ? - 1 have never carried out a study of the relation between the reactions of the speaker and relevance. It seems that some speakers are not particularly sensitive to relevance. I observed real differences with regard to the reactions in phonology, mainly, it is true, in the case of a second language. There are people who when they learn a second language learn at the same time a new phonology; others don't, even though they may acquire a fairly satisfactory pronunciation. I knew a person who learned the signifiers of each sign of the second language as indivisible realities. Her realization of the signifiers was quite acceptable when she had practiced long enough, but at the start, the pronunciation of each word was distorted in accordance with the system of the first language. I think that in matters of relevance we should not have too much confidence in the speaker. Chomsky asserts that the speaker is the judge of grammatical y and possesses a certain 'interiorized, linguistic theory. - My experience is that speakers are quite incapable of passing such judgments. Just one example. An American professor once sent me an article in which he tried to paraphrase French compound words in the purest transformationalist tradition: aide-maçon, for example. Having a good knowledge of French, he didn't hesitate to use his 'competence'. I compared my own paraphrase of aide-maçon (the first I thought of) with that of my colleague Georges Mounin and that of my wife. All three were different: aide-maçon could be "a helper who is a mason", "a mason who is a helper", or "someone who helps (verb) a mason (object)". No wonder I remain somewhat sceptical about the value of competence in linguistic research. I have another reason to be sceptical. One of Harris' students, toward the end of the 1950s, came to Paris with a list of correct, incorrect, or non-existing French sentences established by Harris, and she had to ask three native professors of French whether the sentences were good, bad, or acceptable. Agreement among the three professors was the exception rather than the rule. Some logicians think that the structural linguistic model isn't sufficiently formalized. What role could formalization have in linguistics? Those logicians actually prefer Hjelmslev's theory since it is more formalized. - The degree of formalization one has to expect in a science depends a great deal on the object of that science. Some scientific objects practically cannot stand any formalization. My experience with Hjelmslev's theory, whose development 1 followed closely, showed that the degree of formalization after which he strove goes beyond what is

242

ANDRE MARTINET

useful and acceptable, with regard to languages. Maybe I have been too tempted to resist formalization, perhaps as a reaction against what I consider to be excesses. A recent attempt, however, convinced me that a certain degree of formalization is useful. Until this year, I had always presented my views with regard to syntax in a very explicit way, avoiding a formalization that would be too elaborate. I do formalize. I would not be a phonologist, a structuralist, if I did not formalize to some extent. But every one of my attempts at formalization is accompanied by constant checking against linguistic reality. Operating as I do with 'margins' means that I have formalized and that some elements of linguistic reality have resisted integration. Prestructuralists like Meillet did something analogous, leaving some marginal facts aside when they reconstructed Indo-european: the only vowels are e and o, since a exists only in marginal elements of the vocabulary. This year, somewhat pressed by my audience, 1 presented a visualization of syntactic relations. To visualize and to formalize are probably not quite the same, but visualization brings about formalization. I tried to determine how many graphic conventions are needed in order to present the syntax of any language. I applied it mostly to French texts, sometimes to English, German, and Latin sentences, but it really was an attempt to make general syntax more explicit. This visualization forced me to take a stand, in matters where 1 had so far left some doors open. Now, I had to make a choice since I did not want to add new symbols if I could help. Still, I did not stick to the initial number of conventions, for example, when I invented a new symbol for indicating that the expression of a function is required by the lexical context. If, for example, put implies putting something somewhere, both the something and the somewhere have to be expressed. The apposition also raised a specific problem since it is a determination which would bear the same type of relation to the rest of the utterance as the moneme it determines. Whence the combination into one symbol of the arrow as the mark of determination and the bar as indicative of coordination. The same symbol can be used to distinguish wA/c/z-relatives from ¿/iaf-relatives. Operating with a limited number of symbols forces me to give a discrete character to the recognized syntactic schemes. I do specify the nature of a determination by adding the preposition or the case indication in the midst of the airow relating the complement to the predicate or some other nucleus. But here again, 1 operate with a limited number of grammatical functions. We can certainly achieve positive results by setting up a vizualizing system which entails some formalization. In the last analysis, formalization is an effort to generalize, to group facts which at first sight seem to be very different, into one and the same category. The first step toward formalization in linguistics coincided with the appearance of phonology. Hjelmslev, who made a considerable effort toward formalization, was tempted to deny it. It remains true, however, that glossematics started as a 'phonematics', in opposition to phonology. Now, defining oneself by opposition to something presupposes a common basis. It is important to remembei that glossematics resulted from the activity of a Committee appointed by the Linguistic Circle of

ANDRE MARTINET

243

Copenhagen for a thorough investigation and evaluation of the publications of the Prague Circle. Do you consider formalization to entail a 'reduction' of the linguistic phenomenon? Isn't reductionism too considerable a danger ? - Yes, if that means eliminating differences between languages. In my effort to reduce the number of different types of syntactic relations, I am also at times tempted to lump together things that are different. Yet, I try never to do it unless there remains some element of the context which preserves the difference. This brings us to the problem of interdisciplinarity. How can logicians, mathematicians, and linguists be brought to work together ? - Don't foiget physicists, biologists.... I am a member of an interdisciplinary group organized by the mathematician André Lichnerowicz. The research carried out in that group shows that what is good for mathematics is not necessarily good for linguistics. It is interesting to notice that it pays for us linguists to look at things from the same angle as the physicists, who for the most part do not use the hypothetico-deductive methods of mathematicians. I am thinking of Jakobson, who recently wrote an article on the possible relation between biology and linguistics...29 - I am afraid biologists are trying to use our terms in a metaphorical way. In interdisciplinary matters we should not ape each other, and in the case of linguistics, ape mathematicians or logicians. That is what people like Chomsky have done, simply because they were mathematicians themselves at the start. It is understandable that someone who, at the start, is a mathematician is tempted to transfer his scientific experience and methods into another domain. But professional linguists should resist the temptation to follow suit. Physicists are wiser; their discipline has a long scientific standing. The mathematician's job is to apply his mathematical tool to every problem. Chomsky never wondered whether the subject-predicate group is really necessarily present in all languages - he probably does not care - for him this is just a hypothesis and therein he remains a mathematician. There is also the reverse movement, the influence of linguistics on other disciplines. Linguistics is considered to be the pilot science of human sciences. The possibility of a linguistic poetics is often brought up... - I think linguistic training is essential if one is dealing with poetics. A real linguistic training will make possible a more scientific approach to the problem of the natuie of poetic forms. Traditionally people thought that poetry had to be treated in poetic terms. Nowadays they are getting used to the idea of linguists investigating poetry as a linguistic form. 29

R. Jakobson, "La linguistique" in Tendances principales de la recherche dans les sciences sociales et humaines. Partie 1; Sciences sociales (The Hague, 1970), 504-556.

244

ANDRE MARTINET

Classical phonology probably long neglected the phonological particularity of poetry... - Yes, but there are too few people who have taken the trouble of studying phonology. People should be warned that 'transformational phonology' isn't phonology, but morphology. Phonology is by many considered an unattractive chapter; and therefore neglected. It is a great pity because nothing can be achieved in the field of style and poetic form without a thorough grounding in phonology. What is the relation between semiology and linguistics ? We already have the famous problem of the relationship between semiology and linguistics in Saussure. - I generally agree with Saussure's teaching in that domain. The existence of linguistic signs makes of language one of the objects of semiology. But Saussure suggests that linguistics is a part of semiology, and 1 would rather say that linguistics is the discipline that has revealed the possibility of a semiology; that will enable us to construct a semiology ; human language itself being among semiological systems the richest and most varied, it is of interest to classify semiological facts in reference to linguistic structure. Semiology should first and foremost concentrate on systems of communication. It would be very dangerous to base it on the examination of what things mean or suggest, because it would be impossible to delimit its object. Semiology is basically the study of sign system, excluding indices, those elements of human or non-human reality from which information can be derived but which were not intended as communication. There is a great tradition of semiology in France. I am thinking of Prieto30 and Mounin31... - We should indeed follow in their footsteps. Semiology calls upon the notion of code. A code, at the start, is a way of adapting an existing semiological system to a particular channel. Suppose you want to communicate from one ship to another; your voice alone would not carry so far. Before the telegraph was invented, people made use of a code of flags to relay language by making a particular flag correspond to a word or letter. Language is not a code, but an autonomous system, whereas the code is a way of adapting an autonomous system or another code to a new channel. During this dialogue, you have often stressed the immanence of the linguistic universe, the fact that there is no direct relation between language and reality. Vd like to ask again: In what way is this relation to be understood, what is the relation between language and experience, since you consider language to be the transmission of experience ? - When I stress the existence of language as an object distinct from the rest of the world, 1 am stressing the possibility of a scientific approach to language. We are in a position to extract from speech what is linguistically relevant. Among the linguistic 30 31

Cf. L. J. Prieto, Messages et signaux (Paris, 1966). Cf. G. Mounin, Introduction à la sémiologie (Paris, 1970).

ANDRE MARTINET

245

items that our methods enable us to isolate, we have some discrete, distinctive units, the phonemes, whose function has no relation whatever to non-linguistic reality. Furthermore, we have a number of grammatical items, some of which are monemes, others syntactic patterns, all of which do partake of meaning, but in such a way as to suggest that the meaning is a purely internal linguistic affair, since it seems to make little sense except in reference to other items in the utterance. This is largely an illusion, the same illusion as gave rise to the distinction between meaningful 'semantemes' and meaningless 'morphemes'. But what is true is that their appearance is dependent on the existence of other monemes and that they belong to limited inventories. It is therefore comparatively easy to define them, both syntagmatically and paradigmatically. As regards the remaining units, the ones which are designated as lexical monemes, their signifier can readily be identified since it is made up of discrete distinctive units. But what can we do about their signifié ? Rightly or wrongly, grammatical items are felt by language users to have a purely linguistic relevance; they are meant to secure the functioning of the language without reference to the outside world. But with lexical items, the situation is just the reverse: their duty is precisely to that outside world which they are meant to embody. They will have to yield to the needs of speakers who, if need be, will use them for new situations and in new contexts, whence polysemy, i.e., rambling meanings which make it impossible synchronically to distinguish between polysemy and homonymy. How are we to know how many units correspond to the signifier tablet If we choose our domain with care it is possible to arrive at such an inventory; but if it had to be done for the whole of a language, we would be embarrassed and tempted to operate in an arbitrary manner. Let's take again the word table. We can say that we will call table every object that presents some sort of flat surface, made of hard material, whether this flat surface is used for engraving inscriptions or for serving meals and presenting objects. But that is quite arbitrary. There are doubtless many English speakers who use the terms table of the law and table of the dining room, in their respective contexts without having ever thought of any relation between them. In other words, they operate with them as with two homonyms. Language is transmission of experience. How does one describe the relation between experience and language ? - Experience is a nonsegmented or partially segmented reality - the deaf-and-dumb are perfectly capable of perceiving the existence of objects independently of the existence of terms that would enable them to communicate their experience with regard to those objects. This implies that the acquisition of concepts takes place independently of language, were it only through the repetition of some actions whereby the deaf-anddumb can create concepts independently of the purely physical resemblances existing in reality; the repetition of the motion necessary to get hold of something may determine in him the appearance of the concept of 'prehension'. Our experience may have been, up to a certain point, previously analyzed before we had to communicate it

246

ANDRE MARTINET

linguistically. But it is clear that the use of a language imposes a certain analysis which does not necessarily coincide with that made prior to that use. It seems that when the deaf-and-dumb learn to speak, they have to learn to group the facts according to a culture that is transmitted by language. The existence of a language will thus impose a particular analysis of experience. This becomes apparent when we compare the expression of the same non-linguistic reality in two different languages. There is the classical example of those Neo-Caledonians who have no word for the dead or death. They have a word for damaged or bad shape which designates the dead or death from the physical point of view. From a social point of view, the dead are designated as "the strangers", i.e., those who do not fit in the social classes. In Europe we do not distinguish, as those islanders do, between physical death and the elimination of a being from the framework of socially organized humanity. All of this implies that experience is largely amorphous before people attempt to communicate by means of a definite language. Maybe there is already a certain articulation imposed by life and society, but before communicating one's experience, one has to choose a language, and experience will take shape according to the chosen language; it will be segmented and organized according to that language. There linguistic reality and non-linguistic reality meet. In order to be expressed, prelinguistic experience has to be integrated in the linguistic pattern that is characteristic of the chosen language. And again we find ourselves confronted with the difference between languages. N o doubt, the world sets or at least suggests certain distinctions. The necessities of reproduction imply the existence of distinct animal species. Nature establishes an opposition, before any language intervenes, between the camel and the bull. There is a physical, non-linguistic reality which separates the two species. But, on the other hand, we know that some languages are content with very few distinctions within the camel species, whereas others need hundreds. As a conclusion, could you evaluate what's happening in France today in linguistics? - In France as everywhere else, but maybe for different reasons, we notice, in some quarters, an infatuation for generative transformational grammar. Transformational grammar was conceived of as a vindication of traditional idealism as against the soulless mechanistic approach propagated by your servant. It is no chance matter if Chomsky insisted on his debt to Cartesian linguistics. It is among French language specialists that transformational grammar has found most of its followers. Francists, as they are called, were indeed much more deeply rooted in tradition than were specialists of other languages. They are, as a rule, pure unilinguals with little understanding of language differences. Ever since I came back from the United States, I have observed how difficult it is for unilingual Parisians, who are prone to identify the thing and the word, to understand linguistics as 1 teach it; whereas speakers of provincial dialects, students of Russian origin, bilingual Africans, all those who at an early stage were in contact with foreigners have no problems. In 1968, resistance to my teaching came from students of French, never from

247

ANDRE MARTINET

Anglicists, Germanists or Slavicists.... It would be interesting to observe whether there exists in America a similar dichotomy opposing students with an anthropological type of training, daily confronted with a variety of languages, to transformationalists who are mainly concerned with English. In some French universities, general linguistics is taught by French language specialists, in others it is taught by general linguists. Many secondary school teachers, who are eager to keep abreast of the latest developments, attend special classes in linguistics and are exposed to different approaches. Some of them try to synthesize what they have read or heard. Now, a synthesis is always difficult, even within structuralism. I do not see, for example, how Hjelmslev's and my own teaching could be synthesized. My American experience shows that cooperation between Bloomfieldians and Europeans has always been difficult, if not impossible. But when it comes to combining the generativist and the functionalist points of view, the attempt is doomed. I think, however, that practitioners of both sides may meet in the end and that functionalism, with its greater respect for facts, will eventually carry the day, under its present form or another. Let us hope that linguistics will soon cease to be fashionable so that we may again find ourselves among people really interested in language and languages. Université

René Descartes,

Paris

Sceaux, France July 22, 1972

JAMES McCAWLEY

James McCawley, the publication of Syntactic Structures1 has been seen by many people as a revolution in the history of modern linguistics. I would like to ask you: what does it mean for you now that grammar has to be generative ? - Let me say first something about my interpretation of the impact of Syntactic Structures. One of the most important consequences of the publication of Syntactic Structures and other work by Chomsky from that period is that linguists were forced for the first time to admit that a linguistic description could be wrong in a nontrivial way. While I do not accept Chomsky's notion of 'evaluation measure', I think it produced considerable progress in linguistics by forcing linguists to consider alternative analyses within the same framework and to conceive of the choice between alternative analyses as being not merely a matter of the whim of the investigator or the satisfaction of some desire for elegance or symmetry, but to be made on a deeper and more systematic basis. Chomsky's introduction of the 'asterisk' in Syntactic Structures also had a great impact: people are no longer concerned with just the tabulation of grammatical examples but also with stating explicitly what is not grammatical. I maintain that it makes no sense to speak of the grammaticality or ungrammatically of strings of words-, what I am doing is thus not generative grammar in Chomsky's sense, that is, giving rules that specify which strings of words are good and which ones not good in a certain language. Rather, I talk about good and bad objects of another sort and write rules that distinguish the good from the bad; these objects are entire derivations, that is, a semantic structure and a surface structure, with whatever intermediate stages relate them and information about context, speaker's intention, etc. However, while I talk about grammaticality of a different type of object than Chomsky does, I am still doing generative grammar in a more general sense, that is, giving rules that are intended to distinguish 'admissible' objects of some type from inadmissible ones, with the point of view that failure to predict correctly that something is inadmissible is just as much of a fault as failure to predict correctly that something is admissible. Explicit consideration of what is impossible or anomalous could be of much significance in other fields than linguistics. One thing 1

N. Chomsky, Syntactic Structures (The Hague, 1957).

250

JAMES MCCAWLEY

that I felt was missing at a highly stimulating conference on cognitive psychology at Pennsylvania State University that I have just come back from was precisely that: explicit attention to what is not possible. For example, Bill Brewer gave an excellent paper arguing that most of the supposed instances of 'conditioning' in adult humans involve higher mental processes and do not fit into the simple stimulus-response paradigm of behaviorism; however, neither he nor the behaviorists he was attacking have raised the fundamental question of what kinds of stimulus and response pairings can be conditioned and what kinds cannot. It would be strong support for Brewer's theory if he could demonstrate that certain pairings cannot be conditioned and show that that followed from his 'cognitive theory' of conditioning. Is generativity a methodological notion: that a grammar has to be exhaustive and precise? Or is the generativity of language rather a faculty of the human mind? - The question of whether one takes as his goal an 'exhaustive and precise' description of some class of phenomena is independent of the question of what one takes to be the nature of the phenomena he is describing. I take a highly mentalistic position regarding language, the position that the real phenomenon to be accounted for is some aspect of the human mind that is reflected in linguistic behavior. But there is no logical necessity to take a mentalistic conception of language in order to do what can properly be called generative linguistics. You can find all of the logically possible combinations of transformational grammar or not, generative grammar or not, mentalism or not. There are mentalistic structuralists like Jakobson, there are behavioristic transformationalists like Zellig Harris,... What does it mean that grammar has to be transformational ? - Let me start by mentioning one significant difference between the way George Lakoff uses the word 'transformational' and the way I do. Lakoff uses the term in a relatively restricted sense which would emphasize the differences between what Chomsky is doing and what he and I are doing, a sense according to which very few people other than Chomsky are transformational grammarians. I tend to use 'transformational' in a broader sense which covers both what Chomsky does and what generative semanticists do. A grammar, according to me, would be transformational if it involves the notion of a derivation, consisting of some kind of underlying structure and steps leading from it to a surface structure, where all steps of the derivation are represented as trees, and involves a system of rules that specify how the different stages of a derivation are related to each other. There is a lot of difference among linguists who are 'transformational grammarians' by this definition, as regards what claims they make about what a derivation contains and the nature of the rules defining possible derivations. Chomsky seems to shy away from the notion of tree he tries to reduces the tree to something else, to a set of strings. Lakoff and I, on the other hand, take the notion of tree as basic, using the word of course in a mathematical rather than a typographical sense: a tree is a set of nodes with three rela-

JAMES MCCAWLEY

251

tionships, one of 'domination', one of left-to-right ordering, and one (between nodes of the tree and an 'alphabet') of 'labeling', subject to various postulates.2 Our closest analogue to Chomsky's 'base rules' is our rules specifying what semantic structures are possible, and we take these to be a set of 'node admissibility conditions' - a set of conditions on what a node with a given label may dominate; a tree is 'admissible' if each of its nodes meets one or other of the conditions - supplemented by some overall constraints on what may be a semantic structure (e.g., that every 'variable' must be commanded by exactly one quantifier that binds it), whereas Chomsky defines a set of deep structure trees indirectly, by means of a system of rewriting rules and an algorithm for constructing a tree out of the strings of a rewriting-rule derivation. Our conception of how different stages of a derivation are related is the derivational constraint. For us, a movement transformation is a statement that a certain pair of consecutive stages of a derivation may differ to the extent that some constituent which is in one place in the 'earlier' stage is in such-and-such place in the 'later' stage; Chomsky has instead a direction to move a certain piece of the 'earlier' structure to such-and-such place. This is more than just a reformulation of Chomsky's conception of transformation,since it generalizes in a different way, namely to Lakoff's notion of global rule,3 a specification of how non-consecutive stages in a derivation may or may not differ. Chomsky also has a different conception from us of what may appear in the various representations: he allows arbitrary syntactic 'features' throughout derivations, whereas we do without syntactic features. I would say, incidentally, that the only respect in which Chomsky's Aspects4 has had a really pernicious effect is that it has lead to great promiscuity in the use of syntactic features, which in no case that 1 am aware of have contributed to any insight. Can you say that Chomsky's conception of ''transformation' is too empiricist? - I think there is some justification for applying the word empiricist to Chomsky's conception of language. His position on the relation of theory and data is quite empiricist and indeed seems to show a lot of influence of the philosophy of science of Nelson Goodman. 5 His notion of 'evaluation measure' seems to be a direct development out of Goodman's ideas on the choice among alternative theories. I find Chomsky not mentalistic enough to suit me, and principally because he gives too much priority to the kinds of facts that a real empiricist would be happy with: grammaticality judgments, as opposed to judgments about what a sentence could mean and under what circumstances it would be appropriate. The data he gives the most weight to are those which supposedly are most accessible to observation. Actually, though, I question that grammaticality judgments, independent of meaning 2

Cf. J. McCawley, "Concerning the Base Component of a Transformational Grammar", Foundation!. of Language 4 (1968), 243-269. 3 Cf. G. Lakoff, "Global Rules", Language 46 (1970), 627-639. 4 N. Chomsky, Aspects of the Theory of Syntax (Cambridge, Mass., 1965). 5 N. Goodman, The Structure of Appearance (Cambridge, Mass., 1951).

252

JAMES MCCAWLEY

and context, are accessible to observation. When an informant tells you that suchand-such a sentence is grammatical or not, what he has generally done is to attempt to envision a situation in which someone might say it and to report that it is grammatical if he has succeeded in thinking of such a situation; he may report as ungrammatical a sentence which he could perfectly well use if given appropriate circumstances, as a result of his failure to think of such circumstances. To take an obvious case which Jerry Morgan discussed recently,6 certain combinations of words are extremely strange if presented in isolation but are perfectly normal as answers to certain questions. "Spiro conjectures Ex-Lax" would generally be felt to be unintelligible if presented out of context but is a perfectly normal answer to the question "Does anyone know what Mrs. Nixon frosts her cakes with?" In general, responses that people give to questions about grammaticality are really responses to a different question: can you think of a way of using it? or perhaps, can you think of objections to using it, given what you think the linguist will accept as an objection ? There are cases where an informant will tell you that something is good which in fact he would never accept as normal, but he thinks it is not bad in a way which the linguist would care about. Probably the notion of grammaticality of a string of words doesn't play a role in any kind of linguistic behavior. You might consider defining 'grammatical' as, say, 'appropriate under at least some circumstances'. But that notion of grammaticality would be of no significance for linguistic competence. The fact that there are circumstances under which you could say something has no bearing on what happens when you combine it with other material in a complex sentence or put it in the middle of a discourse. Is mentalism a philosophical option or a methodological hypothesis confirmed by empirical evidence ? - I use 'mentalism' both in connection with what a scientist is trying to describe and what he will accept as data relevant to the description. I would classify someone as a mentalist if he accepts some notion of mind as what he is investigating, as opposed to holding that the notion of mind is not worthy of serious consideration and that one should only consider physical phenomena; also if he takes as appropriate to his domain data which would be mental in the sense that they would be products of the mind, reports of a person's thoughts or judgments. Is it necessary to believe in innatism in order to do good linguistics ? - The conclusion that linguistic competence has an innate basis is probably correct, but it doesn't play an influential role in the actual linguistics that people do. 7 Tf you examine the corpus of Chomsky's writings, you can find numerous arguments 6

J. Morgan, "Sentence Fragments", in Braj Kachru et al. (eds.), Papers in Linguistics in Honor of Henry and Renée Kahane (Urbana and Chicago, 1973). 7 See in this connection the letter by George Lakoff in The New York Review of Books 19 (1973), Febr. 8.

JAMES MCCAWLEY

253

that the form of grammar is innately present in the mind; but the actual linguistic analyses which Chomsky does do not involve the innateness hypothesis. They involve conclusions as to what the form of a grammar is, but whether this form is innate or not plays no role in the analysis. One can perfectly well reject the claim of innateness and still do the kind of linguistics that Chomsky is doing. There is, however, one linguistic theory in which innateness is intimately connected with the analyses, namely Stampe's theory of phonology.8 You don't need as a linguist the conclusion that the structure of the human mind is a neurophysiological structure ? - You don't need it, given the present state of our knowledge about the relationship of the brain to the mind. The only conclusions a linguist can need are those which can play a role in his argumentation, and at present we are not in a position to put together linguistic and neurophysiological premises in arguments. But of course that may change as more is learned about how the brain works. What does Chomsky's hypothesis about linguistic creativity mean for you ? - He is obviously correct in saying that human language is essentially creative. But I think it was more important to stress that idea fifteen years ago when Chomsky was combating the descriptivist orthodoxy in the United States, which involved total disregard for the creative aspects of language. There are an infinite number of conceivable theories of language which are consistent with the proposition that language is creative. It was important in 1957 to discredit theories which did not allow for creativity; a more important task now is to choose among the infinite range of theories which allow for it. The statement that language is creative is now generally accepted and is a common ground for the arguments among competing theories. Does creativity mean only that the speaker has the faculty to produce an infinite number of sentences ? - No, since you can produce an infinite number of sentences in all sorts of trivial ways. A person who could only produce the sentences "I am happy", "I am very happy", "I am very, very happy", etc. would be able to produce an infinite number of sentences, but his 'language' could hardly be called creative. The class of sentences that a speaker with normal competence can produce is not just infinite but interestingly infinite; it covers a range of sentences that you could not sum up in some neat little formula like "I am very happy". However, infiniteness, be it trivial infiniteness or interesting infiniteness, is not what makes language creative; rather it is that a language can be used as an all purpose instrument, to express meanings appropriate to essentially any situation, to convey essentially any information that one might want to convey. A language can accomodate itself to all possible subject matters, to any possible intention of the speaker. 8

See the discussion below.

254

JAMES MCCAWLEY

Does mentalism imply that linguistics has to become a part of psychology ? - It would be reasonable to say that linguistics is a part of psychology, but perhaps better to say that there is no dividing line between linguistics and psychology, that linguistics and psychology are areas without clear boundaries, in a single coherent domain. Can one say that some generative semanticists are more oriented towards psychology and others towards sociology ? Do you see a fundamental opposition between a psycholinguistically and a sociolinguistically oriented semantics ? - It would be misleading to speak of my work as oriented towards psychology. The work that I am doing is not what would normally be called 'psycholinguistics', since I do not measure reaction times, tabulate errors in production and comprehension, etc., but rely on more traditionally 'linguistic' data. I am as willing to erase the border between linguistics and sociology as that between linguistics and psychology. A real description of linguistic competence has to take into account the relation of utterances to the situations in which they are used. Let's turn to another matter and talk about the genealogy of generative transformational grammar. How do you characterize the development of American linguistics up to the generative semantics movement ? - Generative semantics is of course a direct development out of Chomsky's Aspects of the Theory of Syntax,9 involving first Postal, then Lakoff and Ross, and only later myself. People like Postal, Lakoff, and Ross explored in depth the consequences of the theory of Aspects, doing a lot of analyses of facts of English and coming up with deep structures that kept getting progressively deeper as the research progressed - deeper not just in the sense of further removed from the surface but also in the sense of closer to semantic structure - until the distance between deep structure and semantic structure approached zero. It occurred independently to a number of people in 1967 that the distance would eventually reach zero, and we then started operating within a framework that did not involve a level of deep structure as distinct from semantic structure. This involves a departure from the Chomskyan tradition: taking semantic representation to be in the province of the linguistics of now, rather than to be part of a program for research in the distant future. Since 1967, there has been a lot of influence on generative semanticists by various traditions outside of transformational grammar, or outside of linguistics entirely; for example, Lakoff and I have got quite a lot out of readings in logic. Chomsky's ideas of grammar were heavily influenced by Harris in the late forties. Chomsky did not become a mentalist until after he had already done a lot of linguistics; his early work was very much within the behaviorist tradition. 9

Cf. note 4.

JAMES MCCAWLEY

255

Are you thinking of Chomsky's unpublished book "The Logical Structure of Linguistic Theory" ?10 - Yes. For example, one chapter of that book is devoted to the foundations of phonemic theory and consists largely of a demonstration that many difficulties that arise in applying existing definitions of 'phoneme' can be avoided by applying an evaluation procedure to a class of alternative phonemicizations. The conception that Chomsky has even today of the place of meaning in linguistics is straight out of Harris : the conception that syntax is completely independent of semantics and that one can determine syntactic well-formedness independently of meaning. Actually, given that Chomsky and Hockett have very different conceptions of what is 'semantic' (Hockett takes the position that word boundaries are semantic), Chomsky and Hockett equally much take a position that syntax is autonomous vis-à-vis semantics. Does generative semantics find any stimulation in trends of European linguistics ? - I personally haven't derived much from it, though this may be simply because of my lack of knowledge of European trends. One thing that has been a source of stimulation to some generative semanticists is the analysis ot word order and stress from the point of view of new and old information by Prague-school linguists. Can we start now with more specific questions on the theory of grammar ? Can one say that the most fundamental difference between generative semantics and traditional transformational grammar is the conception of the relationship between syntax and semantics ? - That's certainly a major difference. We believe that a single system of rules is responsible for both determining what is 'grammatical' and for associating sentences to their meanings, whereas Chomsky treats syntax as an autonomous part of grammar and speaks about grammaticality as independent of meaning. Incidentally, Chomsky's conception of the factual domain of syntax has changed within recent years : he now calls 'grammatical' many sentences which had previously been called ungrammatical and says that they are semantically uninterpretable or semantically anomalous. For example, he indicates in "Some Empirical Issues in the Theory of Transformational Grammar" 11 that sentences which have faulty concord are grammatical but bad because of semantic uninterpretability. Can generative semantics adequately be called semantic syntax ? - For Chomsky's approach, 'autonomous syntax' is an appropriate term. For ours, I feel less happy atiout the term 'semantic syntax', since it is far from obvious what that term would mean. We have given up the traditional dividing line between syntax and semantics. 10

N. Chomsky, "The Logical Structure of Linguistic Theory" (Cambridge, Mass. [unpublished, microfilm], 1955). 11 In P. S. Peters (ed.), Goals of Linguistic Theory (Englewood Cliffs, 1972) and in N. Chomsky, Studies on Semantics in Generative Grammar (The Hague, 1972).

256

JAMES MCCAWLEY

Is 'generative semantics' a more adequate term for the kind of linguistics that you are doing ? - I am in the uncomfortable position of not having a name that I am happy with for what ] am doing. 'Generative semantics' has the unfortunate property that it uses the word 'semantics' and thus suggests a syntax/semantics opposition. I would be happier if 1 had a term which avoided the distinction between 'semantics' and 'syntax', though 1 deplore the barbaric term 'semantax' which some of my students have been known to use. I am also unhappy with the term 'generative semantics' because it suggests that the principal distinguishing mark of generative semantics is what part of the grammar is 'generative'. The term actually is not too inappropriate in that the closest analogue we have to Chomsky's base component is the rules that specify what is a possible semantic structure, and it is reasonable to use the term 'generative' for rules that specify what is possible at some particular stage of derivations, opposing that to 'interpretive' rules, which specify how different stages of a derivation are related. But the rules specifying what semantic structures are possible are not the only 'generative' rules in generative semantics: we also admit output constraints, which, according to this terminology, are 'generative' as opposed to 'interpretive'. The term 'generative semantics' is also misleading in that it leaves out any suggestion of our conception of 'interpretive' rules, which is as different from Chomsky's as our conception of 'generative' rules is from his. Would you say that each syntactic phenomenon in a language is at the same time a semantic one ? - It is counterintuitive to call certain phenomena semantic, for example, the restriction in English that a personal pronoun cannot be preceded by a full NP or a particle in surface structure (* "I gave Harry it"; *"He tried out it"). But there isn't anything approaching a division between a syntactic system of rules and a semantic one. The rules which determine what is a possible surface structure overlap to a very large extent with the rules which relate surface structures to the meanings which they can express. What's a possible surface structure is to a large extent a reflection of what is a possible semantic structure and how semantic structuies are related to surface structures. The question can be put in another form: must each semantic feature be syntactically embodied? - The word 'feature' has both a popular and a technical sense; 1 take it that you mean it in the popular sense. It is not necessary that all semantic phenomena have syntactic reflections, but I would maintain that any semantic phenomenon could in principle be reflected in grammar somehow. Let's talk about the nature of the underlying structure of language. This underlying structure is a semantic one ? - Yes, if by 'underlying' you mean the deepest end of derivations. However, all

JAMES MCCAWLEY

257

of the intermediate stages of the derivation can reasonably be called 'underlying' and have psychological reality, though they are not strictly speaking 'semantic'. Do we have to identify this underlying structure with what is traditionally called deep structure ? - It satisfies certain characteristics of 'deep structure', as that term was used in Chomsky's Aspects, but not others. It is the 'input' to the transformational rules and there is a system of rules specifying what is well formed at that stage; but it is not the level of derivation at which it is determined what lexical items are appropriate. I have argued explicitly12 that various transformations apply prior to lexical insertion, and 1 am reasonably certain that there is no level at which lexical insertion takes place, that is, that lexical insertions can be interspersed among other steps of a derivation. What are your arguments for the claim that lexical insertion is spread over the whole derivation ? - I don't claim that it is spread over the whole derivation. Rather I hold that lexical insertion takes place in the cycle. This implies that lexical insertion could not follow postcyclic rules such as Extraposition or Relative-pronoun-movement. One argument for having lexical insertions interspersed among 'ordinary' transformations is that certain transformations are part of the mechanisms that relate lexical items to their contribution to the meaning of a sentence. For example, "malinger" means "pretend to be sick", and a sentence containing the word "malinger" must thus have a semantic structure containing "x pretend that x is sick", in which "sick" has a subject that does not appear in surface structure. If I am correct in identifying the deletion of this embedded subject with the ordinary transformation of Equi-NP-deletion, then you have the following alternatives: either there is no separation between syntax and semantics such as Chomsky's notion of deep structure presupposes, or Equi-NP-deletion is not a transformation at all but a semantic interpretation rule and the line between syntax and semantics is pushed much closer to surface structure. Interpretive semanticists generally have not hesitated to take the latter horn of the dilemma. 13 There are several arguments for post-transformational lexical insertion based on syntactic peculiarities of specific words. For example, "apologize" allows Equi-NP- Deletion, but the 'controller' of the deletion is the subject of 'apologize' and not its indirect object. This deviation from the normal pattern of what the 'controller' is can be explained by deriving "x apologize to y for z" from "x request y [y forgive x for [x do

12

J. McCawley, "Lexical Insertion in a Transformational Grammar without Deep Structure", Papers from the Fourth Regional Meeting, Chicago Linguistic Society (1968), 71-80. Henceforth cited as CLS. 13 See Chapter 5 of R. S. Jackendoff, Semantic Interpretation in Generative Grammar (Cambridge, Mass., 1972).

258

JAMES MCCAWLEY

z]]" and saying that the subject of "x do z" is deleted under identity with the indirect object of "forgive", that is, the Equi-NP-deletion that "apologize" manifests really takes place in a clause that does not appear in surface structure and has its controller in the expected position. Has each lexical item of the dictionary an underlying semantic structure ? - A few lexical items are probably semantically primitive; the negative element "not", for example. But I don't have a clear feeling as to which items are semantically primitive. Words which denote biological species, like "dog" or "banana", probably don't have a real lexical decomposition. They are like proper names in that it is a historical accident that the species came into being and there is no list of characteristics that can be said to 'define' the species: for any such list there is nothing in principle to prevent two separate species with just those characteristics from having arisen. A person who uses a biological species word generally has an intuitive conception of 'species' and applies the word to those individual plants or animals that he believes to belong to that species. The one respect in which some decomposition of species words may be necessary is their relationship to taxonomies. I have no feeling as to whether the proposition "Dogs are animals" is analytic - i.e., true because of what the words mean - or merely a piece of knowledge. What about the possibility of decomposition of verbs ? - I have proposed lexical decompositions for many verbs, for example, "malinger", which I mentioned earlier. The existence of an underlying complement in the semantic structure of a verb can often be shown by the possibility of an adverb which doesn't modify any of the surface clauses and would make sense only if it modifies some subordinate clause of the semantic representation. A beautiful example of this was discovered several years ago by Masaru Kajita. In the sentence "Yesterday John lent me his bicycle until tomorrow", "until tomorrow" doesn't give the time that the lending takes place (which was yesterday) but rather the time that I am to possess John's bicycle. This provides evidence that "I possess John's bicycle" is part of the semantic structure of the sentence, since otherwise there is nothing for "until tomorrow" to modify. What are the value and pertinence of the classic syntactic categories ? - The distinction between noun, verb, and adjective isn't quite as arbitrary as generative semanticists have generally made it out to be. Lakoff and Bach14 have given excellent arguments that verb, adjective, and noun all originate in the same underlying position, namely predicate position, and can be considered as belonging to a single underlying category. However, the distinction between what is morphologically a verb or a noun or an adjective is only to a certain extent idiosyncratic 14

Cf. E. Bach, "Nouns and N o u n Phrases", in E. Bach and R. T. Harms (eds.), Universal in Linguistic Theory (London-New York, 1968), 91-124; G. Lakoff, Syntactic Irregularity (New York, 1970).

JAMES MCCAWLEY

259

and is far from totally arbitrary. Lakoff and Bach have pointed out that there are lots of examples of synonymy between words of different categories, for example, the verb "fear" and the adjective "afraid", or the noun "fool" and the adjective "stupid". Thus there is some overlap between these categories as regards what can be expressed within them. However, there also appear to be universal restrictions on how meanings can be distributed among these categories. Not many people have done serious work on this area; the only one I can think of offhand is Robert Dixon, who has worked on what kinds of notions can be expressed by adjectives (though his work unfortunately remains unpublished). Dixon concluded that there are certain classes of notions which it is natural for a language to express with adjectives if it has such a category - which many languages do not - and that there is an implicational hierarchy among these classes of notions, e.g., if a language has adjectives at all it will have adjectives that express value ("good") and size ("large"), but it need not have adjectives expressing physical properties ("sharp") or human propensities ("clever"). What is expressed by adjectives is then only to a certain extent idiosyncratic: each language allows its category of adjectives, if any, to penetrate to a certain depth in this hierarchy and systematically excludes adjectives which would express meanings either lower in the hierarchy or outside the hierarchy entirely. Regarding the notion of noun, I think you can provide a universal characterization which is related to Quine's remark that "To be is to be a value of a bound variable". Nouns are words which are used to indicate the domain of a bound variable. Whereas logicians generally treat nouns, verbs, and adjectives as if theie were no significant difference between them - e.g., represent "Some men are bald" as "Exist (Man x & Bald x)" with "Man" and "Bald" in apparently interchangeable roles - in real languages their roles are not interchangeable - e.g., one says "Some men are bald" and not "Some bald-things are men". In ordinary language one only takes certain concepts as defining domains of bound variables, and those concepts are expressed by nouns. Is the argument-predicate opposition fundamental for the semantic underlying structure ? - Yes. In semantic structure there are three categories: proposition, predicate and argument. A proposition consists of a predicate and one or more arguments. I am using 'proposition' in a sense which includes 'propositional function'. Many people15 distinguish between 'sentence' and 'proposition', with all occurrences of "It's raining" being the same sentence but expressing different propositions depending on the time and place that the occurrence of the sentence has reference to. I take semantic structures to involve propositions as opposed to sentences, that is, to involve referential indices that mark the entities that the (occurrence of) the sentence purports to refer to. I have argued in "A Program for Logic"16 and "Semantic Representation" 17 15 16 17

E.g. P. F. Strawson, Introduction to Logical Theory (London, 1951). In D. Davidson and G. Herman (eds.), Semantic of Natural Language (Dordrecht, 1972), 498-544. In P. Garvin (ed.), Cognition: a Multiple View (New York, 1970), 227-247.

260

JAMES MCCAWLEY

that the same representation of content is appropriate for a linguist's purposes as for a logician's, in the sense that while both linguistic considerations and logical considerations demand a representation of content in terms of immediate constituent structure, they do not impose conflicting restrictions on what that constituent structure can be. The logical considerations that ] have in mind are the formulation of rules of inference in their fullest generality, and the construction of rules relating the truth values of complex propositions to the truth values of their constituents. Do you agree with Lakoffs term natural logic18 as the adequate logic for natural languages ? - It is highly appropriate. It exactly fits a program for what linguists and logicians ought to be doing for a better understanding of language: not just to investigate the traditional logical topics, like "not", "and", "or", "all", and "some", but to do logic as it applies to all elements of meaning. The traditional distinction that is drawn between logical operators, logical constants, and everything else really is arbitrary. The various rules of inference amount to meaning postulates for "and", "or", etc. There is no particular reason why only those items whould be provided with meaning postulates; one could also give meaning postulates for "possible", "necessary", and further away, for "to order", "to request", "to intend", etc. It is the logician's job to consider whatever relationships of inferability, extensional satisfiability, etc. actually hold, and all elements of meaning belong in the domain of logic. The work that has been going on recently in various areas of modal logic, in the logic of questions and imperatives, etc., is in that direction. My only objection is that people in those areas take too narrow a view of what they are doing; they work on one problem at a time and rarely attempt to integrate the results of studies of different problems. Can you give us some indication of logics that are appropriate as 'natural logic' ? - Lakoff has received much stimulation from the treatment of modal logic in terms of possible worlds in works by Saul Kripke 19 and David Lewis.20 The notion of possible world is useful in describing various linguistic phenomena, for example, the use of pronouns and definite articles. A pronoun is not simply equivalent to a repetition of a 'full' noun phrase: you have to take into account the world that is being referred to at the particular point of the discourse. For example, in the sentence "John wants to catch a fish and eat it", "it" refers back to "a fish". To make sense of that you need to describe the verb "want" as making reference to a hypothetical world. The person who has the want is in the real world, but the want is a description of a hypothetical situation. The possibility of using pronouns is contingent on the 18

G. Lakoff, "Linguistics and Natural Logic", in D. Davidson and G. Harman (eds.), Semantics of Natural Language (Dordrecht, 1972), 545-665. 19 S. Kripke, "Semantical Considerations on Modal Logics", Acta Philosophica Fennica (1963), 67-96. 20 D. Lewis, "Counterpart theory and quantified modal logic", Journal of Philosophy 65 (1968), 113-126.

JAMES MCCAWLEY

261

pronoun and the antecedent being part of the same world, be it the real or the hypothetical world. The pronoun can occur in a separate sentence that refers to the same imaginary world; for example, you could perfectly well have a discourse with the sequence of sentences "John wants to catch a fish. I'm sure he wants to steam it with black bean sauce". Such a discourse is intelligible if the two sentences are taken as referring to the same imaginary world. Is it possible to make an exhaustive tabulation of possible worlds ? - 1 don't think any kind of tabulation of possible worlds makes any sense. The notion of possible world is of extremely wide applicability. A possible world, relative to some logical system, is simply a set of assignments of truth values to the propositions that are formulated in that system and assignments of referents to the various referring expressions that can be formulated within it. The linguistic uses of 'possible world' are phenomena in which one talks about alternatives to the real world as it is now, as in talking about the real world at different points in time or about conditions for the fulfillment of a wish or a conjecture. Counterfactual conditionals also involve alternatives to the real world: they refer to a world that you could get to by tracing backwards in time to a certain point, assuming that something else happened then, and tracing forwards again to an alternative to the present. Truth in language is a matter of degree. How can one deal with this difficulty in 'natural logic' ? - What 1 have said so far does not commit me to two truth values. I am willing to accept a logic with multiple truth values, as has been proposed in Lakoff's "Hedges: A Study in Meaning Criteria and the Logic of Fuzzy Concepts", 21 though multiple truth values such as Lakoff proposes - real numbers in the interval from 0 to 1 can be reduced to two truth values plus extra predicates, as Dana Scott has shown.22 In addition to truth values that are 'intermediate' between truth and falsehood, there is also good reason to accept a truth value that might be called 'tilt', to be assigned to propositions which have presuppositions that fail to be true. Natural language is always seen by logicians in its referential function. Is this treatment adequate ? Does language have nothing but a referential function ? - The non-referential aspects of language can be incorporated into a scheme which technically is referential. When a proposition expresses a person's attitude towards a proposition (e.g., "Sam believes that the world is flat"), its truth value is not a function of two individuals but of an individual and a proposition. One can talk about truthvalue assignment involving something predicated of propositions as long as one has a system of semantic representation which distinguishes between distinct propositions, even distinct propositions which are 'logically equivalent'. If p and q are distinct but 21 22

In CLS 8 (1972), 183-228. See the appendix to Lakoff's paper, mentioned in note 21.

262

JAMES MCCAWLEY

logically equivalent propositions, then "John believes that p" and "John believes that q" can have distinct truth values; however, that is no cause for worry provided your system of semantic representation correctly individuates propositions. If I understand correctly, 'natural logic' formalizes also the intention of the speaker. - Speaker's intentions are certainly in the domain of grammar. They are relevant to distinguishing different kinds of speech acts, for example. There are certain sentences which are bizarre if interpreted as conveying information but are quite normal if interpreted as a reminder, for example, "You have only two eyes". Do some units of the underlying semantic structure have to be 'fuzzy' ? - Yes. I think Lakoff established that point very well in his "Hedges" paper. 23 What are the advantages and disadvantages of formalization for linguistic theory and grammar ? - The biggest advantage to formalization is that it forces one to be explicit, to make up his mind about details. It is very valuable for a person to be as explicit as possible and to formulate things precisely. The principal disadvantage to formalization is that one very often doesn't have means of formalization at his disposal which are really appropriate for representing what he wants to say, and if he persists in using the formalism that he has, he wastes a large part of his time fitting square pegs into round holes. A lot of work in transformational grammar, particularly in the early sixties, had very little point, due to the fact that so much attention was devoted to questions of applying an existing formalism, without regard for substantive questions about language and its structure. A formalism is valuable only if one at every step takes a critical attitude towards the formalism and is prepared to reject or modify the formalism whenever it ceases to be servant and becomes master. An example, in the field of syntax, of the failure offormalism could clarify a lot. - The use of rewriting rules as the base component of a grammar created several pseudo-problems. Since rewriting rules operate on strings and transformations operate on trees, it was necessary to supplement the rewriting rules by something that would yield trees; what was proposed was an algorithm for constructing a tree out of a rewriting rule derivation. However, this algorithm fails in the important case where a line...ACB...follows...AB...in the derivation. Instead of questioning the appropriateness of rewriting rules, early transformational grammarians proposed simply to exclude rules which could give rise to derivations in which the algorithms failed, i.e., rules of the form A -*• AC or B CB, totally without regard for whether there were syntactic structures that demanded such a rule. Such structures actually are sufficiently common that they have a name: 'endocentric'. 23

Cf. note 21.

JAMES MCCAWLEY

263

Also, the format generally used for writing transformations is highly inappropriate, since it requires specification of all of the left-to-right relationships among the items mentioned but ignores the top-to-bottom relationships. There are rules in which the left-to-right relationship is irrelevant, for example, the rule converting some into any in the presence of 'negative' elements; what is relevant there is whether one particular node dominates another, independently of what is on the left or right. It is possible to formalize such rules using this linear format only with great ingenuity in combining curly brackets and parentheses. Let us return to some very specific points of your conception of grammar. In connection with the central notion of grammaticality, I would like to ask you if grammatical mistakes occur ? - A mistake occurs when one does something that does not conform with his intentions. A mistake can perfectly well involve a 'grammatical' phenomenon such as the choice of the right form of the verb. However, the notion of mistake doesn't have anything to do with grammaticality of a string of words: a mistake is a mistake regardless of whether the result is 'grammatical'. I like to avoid the term 'grammatical'. Generally I am talking about a particular sentence with a certain meaning in mind that it is supposed to express, and with assumptions about the circumstances in which it would be uttered. When I ask if it is grammatical, what I generally mean is, could you use it under the circumstances in question with the meaning in question ? would it sound normal if you heard it so used? My notion of grammaticality is not of grammaticality of the sentence per se but of the sentence relative to its use. Does this require a view of linguistic competence which conflicts with Chomsky's ? - The principal thing that I find wrong with Chomsky's terms 'competence' and 'performance' is that he used 'performance' to cover everything other than 'competence'. He has a reasonably homogeneous conception of competence but a total wastebasket for what he calls performance. What he takes in under performance includes a lot of things that are not linguistic performance but non-linguistic competence, for example, logical competence, that is, knowledge of what is a valid argument. Also, 'performance' is grammatically ambiguous between an object nominalization (meaning 'what one produces') and an action nominalization (meaning 'act of producing' or 'way of producing'), and Chomsky uses it in both of those senses. On the rare occasions when 1 use the word 'performance', 1 use it to refer to the strategies that a person employs in using his linguistic competence in producing and understanding sentences. These strategies, of course, interact with everything else that goes on in one's brain. The principal differences between Chomsky's notion and mine of competence and performance is that 1 attempt to distinguish the many things that Chomsky's term 'performance' takes in, and that I do not attach as much importance as Chomsky does to determining whether a particular fact is 'a fact of competence'. Indeed, any 'linguistic fact' involves an interaction between 'competence' and many

264

JAMES MCCAWLEY

other things. I find the sentence "He played K.515 with the Budapest Quartet" ambiguous between a sense in which he belonged to the Budapest Quartet and a sense in which he was an extra violist who joined the quartet; "He played K. 428 with the Budapest Quartet" unambiguously has the former sense, and "He played K.317 with the Budapest Quartet" is anomalous. I am not thereby committed to saying that Kochel's catalog of Mozart's works is pait of my linguistic competence; rather, my judgments involve an interaction between my linguistic competence and my knowledge, and they are as much facts 'about' the one as 'about' the other. Is introspection a criterion for grammaticality, in your sense ? - I am happy with introspective data, and indeed, most of what I have done is based on introspective data. But I do not object to non-introspective data either. What can one say, against this background, about the difference between language and dialect? - I can hardly improve on Robert Hall's definition: "A language is a dialect with an army and a navy." However, I think too much attention has been devoted to defining terms like 'language' and 'dialect' and too little to the question of how communication takes place between speakers of distinct but related varieties of language. Charles-James Bailey makes the excellent point, in unpublished work, that a person's linguistic competence involves not merely what he does in his own speech but also his knowledge of existing variants of his language. Linguists have overemphasized speaking and given inadequate attention to listening, and I applaud Bailey for attempting to redress the balance. I would say, for example, that even though a male speaker of Japanese may use only 'male' forms of speech and a female speaker only 'female' forms, both 'male' and 'female' forms of speech are part of the linguistic competence of both male and female speakers. For example, use of 'female' forms by a male speaker would, it anything, impede rather than facilitate communication with a woman. Is linguistic change a valuable object of study within linguistics ? - Certainly. Kiparsky has observed24 that you can get great insight into things by observing how they function in language change and thus that diachronic linguistics, apart from its inherent interest, can contribute significantly to synchronic linguistics. He has a beautiful analogy, in which he compares linguistic change to a tiger moving through the jungle: if the tiger stands still you can't see him, because his stripes blend in with the foliage, but if he moves, you can see him. There has been a lot of progress in recent years in studying linguistic change. I should mention the work of David Stampe,25 who has been working on a theory of phonology which is intimately 24

P. Kiparsky, "Linguistic Universals and Linguistic Change", in E. Bach and R. T. Harms (eds.), Universals in Linguistic Theory (London-New York, 1968), 171-204. 25 D. Stampe, "On the Natural History of Diphthongs", CLS 8 (1972), 578-590.

JAMES MCCAWLEY

265

connected with both language acquisition and historical change. Stampe argues that children are born with a large innate inventory of phonological processes, which for Stampe amount to strategies for avoiding difficult sounds and difficult combinations of sounds. Early language acquisition consists not so much in learning rules, as Morris Halle appears to hold, as in learning to suppress these innate processes. Does Stampe find a parallelism between historical changes in the language and language acquisition in an individual? - The most typical language changes correspond to those innate processes which are relatively difficult to suppress and which a child often retains until a late stage for example, replacement ol liquids by semivowels. Stampe is careful to distinguish between those substitutions which are not suppressed until relatively late and those substitutions which are always suppressed early and never come up in real language change - for example, the consonant harmony process involved in the frequently observed phenomenon of young children saying "gog" for "dog". Some people have argued that there is a distinction between the competence of the speaker and the competence of the hearer. Can one really distinguish between these two competences ? - There is linguistic knowledge which a person only uses in understanding and not in production, as in the examples discussed earlier of communication among speakers of different dialects - which often involves knowledge of the 'foreign' dialect and not just reinterpretation of it in terms of one's own dialect - and knowledge by men of 'female' forms of speech. But aside from these areas, I think that supposed examples of differences between 'speaker competence' and 'hearer competence' in fact show no such thing. I am thinking of the statements that one frequently encounters to the effect that 'I can understand all sorts of French sentences, but I can say hardly anything in French'. To see that such statements do not really point to a difference between 'speaker competence' and 'hearer competence', consider the phenomenon of Danny Kaye, who does acts in which he produces gibberish that sounds like French, German, Japanese, etc. If a person whose knowledge of French is imperfect supplements his knowledge with gibberish a la Danny Kaye, the difference between what he can produce and what he can understand vanishes: by acting like Danny Kaye, he could produce all of the French sentences that he understands, albeit at not so fast a tempo, plus of course an infinite range of French non-sentences that are consistent with his limited knowledge of French. By the same token, he understands not only real French sentences but the kind of phony French sentences that he might produce if he played Danny Kaye. The appearance that his 'hearer competence' is broader than his 'speaker competence' comes about only through the error of thinking of his 'hearer competence' only in terms of his ability to understand sentences that are known in advance to be good French and ignoring his ability to distinguish between

266

JAMES MCCAWLEY

real French and pseudo-French. Real French utterances, of course, have no priveleged place in his 'competence'. You have already mentioned global and transderivational rules. Why do generative semanticists find them necessary ? - Let me first clarify the terms. A global rule has to do with a single derivation. It states a relationship between non-consecutive stages in a derivation. A transderivational rule is a relationship between some stages in a derivation and something outside of the derivation - e.g., some stage of another derivation, or some logical inference from the semantic structure of the derivation in question. The best argument I know of for a global rule is the argument for a global concord rule in Ancient Greek given by George Lakoff and by Avery Andrews.26 The concord rule has to be global, since the stage of derivations which determines which noun an adjective agrees with is different from the stage which determines the case that the adjective is put into. The adjective agrees with the noun phrase that it had for its subject at the end of the application of the cycle to the clause that the adjective was the predicate of; however, the adjective is not put into the case which the NP is in at that stage of the derivation, but the case which the NP is in at the end of the derivation, after rules may have applied moving the NP into a higher clause and into a position far removed from the adjective. Paul Postal in "On Coreferential Complement Subject Deletion"27 presents a highly suspicious treatment of Equi-NP-deletion. In order to account for some funny restrictions on Equi-NP-deletion which are shared by pronominalization, he proposed that in the cycle the NP to be deleted is marked with a feature [ + Doom], and that unless that NP is pronominalized in the course of the derivation, in which case it is ultimately deleted, the derivation is inadmissible. Postal has since recognized that this proposal was merely an ad hoc way of stating a transderivational rule: Equi-NP-deletion is applicable only if in an alternative derivation in which the deletion is not carried out, the NP is pronominalized. I should point out that a lot of discussion between generative semanticists and interpretive semanticists revolves about global rules, not just those proposed by generative semanticists, but also those proposed by interpretive semanticists such as Chomsky under the name 'semantic interpretation rules'. Interpretive semanticists assume in effect that a deep structure is a first approximation to semantic structure and that semantic interpretation rules make successive corrections in that approximate semantic structure, on the basis of conditions relating to the corresponding nodes in surface structure or some other late stage of derivations. But that amounts to a global rule that mentions two consecutive stages of a derivation that leads from semantic structure to surface structure and says that those two stages differ in suchand-such a way if a third, very late, stage has such-and-such property. 28

G. Lakoff, "Global Rules", Language 46 (1970), 627-639; A. Andrews, "Predicate Modifiers in Ancient Greek", Linguistic Inquiry 2 (1971), 127-151. 27 In Linguistic Inquiry 1 (1970), 439-500.

JAMES MCCAWLEY

267

If there is that much similarity between interpretive and generative semantics, then can you say, as Chomsky has, that generative semantics ii only a notational variant of the extended standard theory ? - Chomsky has been able to call the two theories notational variants only by describing each of them in such broad terms as to ignore the differences; for example, he ignores the fact that generative semanticists have quite specific claims about the nature of semantic structure, whereas interpretive semanticists do not. One thing which Chomsky recognizes as a real difference is the ordering of lexical insertions relative to other rules. There he claims not that the two theories are notational variants but that generative semantics is wrong, though I would say that in this case the content of his claim is not that there is anything wrong with the derivations that generative semanticists have proposed but that he chooses to call all prelexical rules 'semantic interpretation rules' rather than 'transformations'. One issue which comparison of the two theories raises is whether globality can be restricted to semantic interpretation rules: one might attempt to justify a distinction between a semantic interpretation component and a syntactic component by saying that the rules of the syntactic component are local and those of the semantic interpretation component global. However, that would make at least some agreement rules semantic interpretation rules, which is somewhat counterintuitive. Regarding the possibility of translating between descriptions in terms of global rules and descriptions in the 'extended standard theory', it is instructive to read Baker and Brame's "Global Rules: A Rejoinder" and Lakoff's reply to it. 28 Baker and Brame attack Lakoff's arguments for global rules and propose for each of his cases a reanalysis that doesn't involve global rules. However, in each case what they propose involves a grossly extended version of some device of or suggested by the 'standard theory', such as syntactic features. Lakoff objected quite correctly that Baker and Brame were 'reducing' global rules to something more 'powerful': a theory having an infinite number of syntactic features and no constraints on what features are allowed or how they are used. Lakoff's approach actually restricts the class of possible rules to a much greater extent than does the theory with unrestricted use of features that Baker and Brame adopt. However, I don't find the question of whether a skeletal outline of one theory can be 'interpreted within' another theory a particularly interesting question. What is interesting is the question of how each theory can be modified or supplemented so as to allow it to provide an explanation, as opposed to just a description, of the various linguistic phenomena, and 'notational variants' often are not equivalent from that point of view. How can one explain then Chomsky's statement that generative semantics is too powerful? - It is probably force of habit, a throwback to the days when 'weak generative 88

C. L. Baker and M. Brame, "Global Rules: A Rejoinder", Language 48 (1972), 51-75; G. Lakoff, "The Arbitrary Basis of Transformational Grammar", Language 48 (1972), 76-87.

268

JAMBS MCCAWLEY

capacity' was something that linguists worried seriously about. In the present situation, theories divide into two types: those which are 'too powerful' and those which have been demonstrated not to work. When various syntactic-semantic phenomena have been discovered which cannot be described adequately within existing theories, something else is needed. When Lakoff proposed that you need global rules, that did not carry with it a proposal that every imaginable global rule is a possible rule. You can raise the question of what global rules are needed and set about restricting the class. You can say the same thing about all other descriptive devices, such as transformations, syntactic features, and, of course, semantic interpretation rules. One category which, to my knowledge, no attempt has been made to restrict in any way is semantic interpretation rules. When Chomsky claims that generative semantics is 'too powerful', he is raising an objection which his own theory is even more vulnerable to. The question of what class of global rules is appropriate for linguistic description has been taken much more seriously by generative semanticists than the question of limiting the class of semantic interpretation rules has been by interpretive semanticists. It of course makes no sense to compare just their transformations with our (local and global) derivational constraints; the entire apparatus for relating meaning to surface form must be compared if the comparison is to be at all meaningful. Is the question of the ordering of rules also a point of difference between the two theories ? - It is not a fundamental difference between the two theories. There is just as much difference within each of the two theories on this point as there is between them. Ordering has been attacked both within generative semantics by Postal and Lakoff 29 and within the 'standard theory' by Koutsoudas. 30 Probably there are more generative semanticists than interpretive semanticists who have strong doubts about ordering, but it is a point of controversy within each theory. I do not have a firm feeling as to whether ordered rules are needed in syntax, though I agree with Postal and Koutsoudas that most ordering arguments that have been given in syntax will not stand up. But this is independent of the other issues that we have been talking about. You have done a lot of work on languages as different as English and Japanese. What is your position on the question of linguistic universals ? - Languages do not 'differ from each other without limit and in unpredictable ways', to quote the famous phrase of Martin Joos. 31 The variation among languages is not totally wild. In describing a particular language it is important to identify what are the idiosyncracies of the language and what are the characteristics of it that are either universally or nearly universally shared. One area that is probably universal is 29 30 31

See remarks on various points in P. Postal, On Raising (Cambridge, Mass., in press). A. Koutsoudas, "The Strict Order Fallacy", Language 48 (1972), 88-96. M. Joos, Readings in Linguistics I (Chicago-London, 1957), p. 96.

JAMES MCCAWLEY

269

the general rules of semantic well-formedness, for example, the rule that a proposition consists of a predicate and a sequence of arguments. Those rules are surely universal as far as the constituent structure is concerned, though it is an open question whether the linear order of elements in semantic structure is universal; it is conceivable that some languages demand semantic representations in which each predicate precedes its arguments and others demand semantic representations in which each predicate follows its arguments. On the other hand, Bach and Ross have proposed that all languages have exactly the same semantic structures, including the ordering; according to Bach32 all languages have predicate-first order in semantic structure, and according to Ross 33 all have predicate-last order. The person who has made the strongest proposals regarding the role of language universals in grammar is Bach, who proposed that not only is there a universal base component but there is also a universal list of possible transformations, and that languages differ syntactically only as regards which transformations they select from this universal list, what extra restrictions or relaxations they impose on them, and at what stage of the derivation they apply - for example, Bach would say that both Japanese and German have a rule moving verbs to the end of the clause, but it applies precyclically in Japanese and postcyclically in German. I find Bach's hypothesis very plausible. Ross has been doing interesting work on the question of what variants a transformation can have. For example, the 'basic' form of the Reflexivization rule is probably that if a NP is coreferential with the subject of its clause it is turned into a reflexive pronoun. Various languages have less restricted variants of this rule; for example, English has relaxed the constraint that the antecedent be subject and requires only that the antecedent and reflexive be 'clause-mates', with the antecedent preceding the reflexive and not part of a larger NP (as in "I asked John about himself"). Japanese retains the requirement that the antecedent be subject but relaxes the 'clause-mate' requirement and allows a NP to be reflexivized by the subject of a higher clause. Ross's results are implicational universals; for example, if a transformation can apply 'into a subordinate clause' it can also apply between clause-mates. If you put Bach's and Ross's proposals together, you get a list of skeletal transformations plus ways in which those rules can be restricted or generalized. A long time ago Chomsky drew a distinction between formal and substantive universals. Are Ross's and Bach's universals formal universals? - Bach's universals are substantive universals. A formal universal, according to Chomsky, is a restriction on the form of rules or of grammars. For example, when Chomsky says that a transformation consists of a 'structural description' and 'structural change', with the structural description being a sequence of syntactic category names and variables and the structural change being a set of elementary operations 32

E. Bach, "Questions", Linguistic Inquiry 2 (1971), 153-166. J. Ross, "Primacy and the Order of Constituents", Paper read at the annual meeting of the New England Linguistic Society, 1972, unpublished.

33

270

JAMES MCCAWLEY

(adjunction, deletion,...) on the terms of the structural description, he is proposing a formal universal. Bach's list of possible transformations is a restriction not on the form but on the content of transformations and is thus a substantive universal. Ross is doing something between formal and substantive universals: he is specifying the relationship between a list of 'basic' rules and the full range of 'possible' rules. May I ask you now about your own experience in doing linguistics on Japanese, a language very different from our occidental languages? What seems to be universal when you are doing contrastive semantics ? - I haven't worked on non-Western languages to nearly sufficient an extent to learn much about syntactic and semantic universals. My work on Japanese goes much less 'deep' than my work on English. There are a lot of rules which appear in both languages: both have an Equi-NP-deletion rule, a reflexivization rule, and a Subjectraising rule, for example. The conditions on these rules vary quite a bit. In Japanese, as I mentioned earlier, you can reflexivize quite freely into subordinate clauses, whereas Subject-raising is much more restricted than in English.34 This is of course the kind of interlanguage difference that Bach's proposals predict. I cannot say much of significance about semantic structure. What I have done on Japanese is consistent with the kinds of semantic structures that I have posited for English, but I haven't studied Japanese in enough depth to really establish that there aren't differences. In Japanese, at any rate, one can give the same kinds of arguments as in English for quantifiers originating outside of their clauses; presumably quantifiers appear in the same positions in semantic structure in the two languages except perhaps for order: since quantifiers are predicates, they would be first in the clause in English and last in the clause in Japanese, given my current conclusions about constituent order. There are some interesting differences as regards what elements of meaning can be combined in a single word. For example, English has many adjectives referring to emotional reactions ("angry", "annoyed", "sad",...) and corresponding causative verbs ("anger, "annoy", "sadden",...) but no corresponding inchoative verbs - e.g., one does not say *"John angered" to mean "John became angry". Japanese, on the other hand, has inchoative verbs for all of these notions (e.g., okoruy 'get angry') and corresponding causative verbs, but generally not corresponding adjectives (e.g., the closest one can come to saying "is angry" is literally "has got angry"). Are universals always universals of the underlying logico-semantic structure ? Chomsky said recently that humans don't only express meaning but that they do it in a specifically structured and partly independent manner: through language... - There are a lot of universals which have to do only with surface structure. The work 34

See S. Kuno, "Evidence for Subject Raising in Japanese", Papers in Japanese Linguistics 1 (1972), 24-51, for a partial explanation of this difference.

JAMES MCCAWLEY

271

of Greenberg35 shows that there are many implicational universals about surface word order; for example, languages with verb first normally have postposed modifiers and languages with verb last normally have preposed modifiers. Ferguson36 recently has been investigating languages which have changed into verb-last surface order, especially Bengali and Amharic, and has found that this change in word order has been followed by many other changes, including changes in verb morphology, which make the language similar to Japanese and Korean, which have been verbfinal from time immemorial. For example, Bengali and Amharic have both developed a verb form which is unmarked for tense and is used both in complement constructions and in coordinate constructions, exactly like the -te form of Japanese. There appear to be certain surface patterns in language which are 'natural' and which linguistic evolution works towards. Obviously, universals of surface structure have to be incorporated into linguistic theory, though I have no proposal to offer as regards how. Generative semantics has brought a lot of philosophical notions into linguistic theory. What do you mean by the presupposition' of a sentence ? That notion seems to be used to cover a very broad area. - I and other generative semanticists are guilty of using the term 'presupposition* in a variety of senses and not always distinguishing between those senses. Sometimes the term is used to cover what Keenan 37 calls pragmatic presuppositions, that is, the kind of 'presupposition' that provides the basis for choosing the appropriate pronoun, for choosing between 'polite' and 'plain' forms, etc. Pragmatic presuppositions are not logical presuppositions, since they do not satisfy the criterion that failure of a logical presupposition makes the sentence fail to have a truth value. For example, "Richard Nixon regrets being a woman" has no truth value, since its logical presupposition that Richard Nixon is a woman is false; but "Richard Nixon admires herself" is true, though odd because of the falsehood of its pragmatic presupposition that Nixon is a woman. One might try to take in these two kinds of presupposition under a single definition: a belief which the speaker must be committed to in order to use the sentence sincerely. Besides the cases discussed so far, this would also take in presuppositions of sentences that have no truth value; for example, the question "Who is buried in Grant's tomb?" would presuppose that someone is buried in Grant's tomb; it would take in still other kinds of presupposition, for example the 'presuppositions' that the speaker does not know who is buried in Grant's tomb and that he believes the addressee does know. However, that definition is too broad, since any statement commits you to the truth of that statement, and it is wrong to say the 35

J. Greenberg, "Some Universals of Grammar with Particular Reference to the Order of Meaningful Elements", in J. Greenberg (ed.), Universals of Language (Cambridge, Mass., 1963), 73-113. 36 C. Ferguson, "A Sample Strategy in Language Universals", Working Papers in Language Universals 6 (1971), 1-22. 37 E. Keenan, "Two Kinds of Presupposition in Natural Language", in C. Fillmore and D. T. Langendoen (eds.), Studies in Linguistic Semantics (New York, 1971), 44-52.

272

JAMES MCCAWLEY

sentence "The police in this city are corrupt" presupposes that the police in this city are corrupt. There is a very strong relation between presupposition and reference as operational notions in linguistic theory... - That is true particularly with regard to pragmatic presuppositions. Knowledge about the things that one refers to is relevant to the choice of the pronouns one uses in referring to them. That observation is a fact of life, and a grammarian must incorporate it somehow into his grammar. Has the notion of reference always been used in the same sense in generative semantics ? - Our ideas have gradually progressed from fuzziness to relative clarity. A lot of confusion about notions like reference and coreferentiality is related to statements of identity: statements that two expressions both refer to the same individual. In such a case you still have to recognize distinct referential indices and to distinguish between coreference, taken literally, and what Postal 38 has referred to as 'stipulated' or 'presupposed' coreference, which has to do with formal identity between indices. In semantic structure, instead of just using two expressions that happen to single out the same object, you may presuppose that you are talking about a single thing in several places in the sentence or discourse. This is 'stipulated coreference' and is distinct from cases where you can infer that two expressions refer to the same thing or where you assert that they refer to the same thing. Only 'stipulated' coreference is directly relevant in linguistics. One complicating factor is that 'asserted coreference' at one point in a discourse may become 'stipulated coreference' at a later point, that is, after you have made an identity statement involving two indices, you can generally ignore the difference between them. Postal in unpublished work has discussed discourses such as "At 10 o'clock a gorilla escaped from the zoo. At 11 o'clock a gorilla ate the Polish ambassador. The police are convinced that the two gorillas are the same and are hunting for it", with it referring to the single gorilla that the police believe in; of course, the discourse could then continue, "But I am convinced that they were two different gorillas". Presuppositions are made just for stretches of discourse, and the various parts of a discourse provide presuppositions for subsequent parts. This is illustrated by the difference in normalness between "Sylvia is a woman and is pregnant" and "Sylvia is pregnant and is a woman": it is all right to presuppose what has already been asserted but not to assert what has already been presupposed. An identity statement provides a presupposition for subsequent discourse, and that gives you justification for using a single index where previously you had two, though sometimes that identification is appropriate only for parts of the discourse, for example, the parts reporting the beliefs, etc. of the police, in the above example. There are other difficult problems which arise in this area, such as the old problem 38

Cf. P. Postal, Crossover Phenomena (New York, 1971).

JAMES MCCAWLEY

273

of the morning star and the evening star, where you have to distinguish between objects and the phenomena which manifest them. A linguist must distinguish between the planet Venus, its manifestation as a point of light in the morning sky, and its manifestation as a point of light in the evening sky. The necessity of this distinction is shown by Jakobson's observation that if a young man walking with his girl friend in the early evening pointed up at the sky and said "Isn't the morning star beautiful?", she would have every reason to think he was out of his mind. Similarly, when one says "The morning star is more beautiful than the evening star", he is not making the contradictory statement that Venus is more beautiful than itself but is comparing the morning sky with the evening sky. The terms "morning star" and "evening star" thus refer not to the physical object but to specific manifestations of it. How do you stand on the issue of whether presuppositions have to be a part of semantic representation ? - I don't have a clear position. This involves the question of to what extent presuppositions are predictable from other things. Here one must distinguish between logical and pragmatic presuppositions. Pragmatic presuppositions are not predictable from anything else. On the other hand, they aren't strictly speaking a part of the semantic representation itself: they are more a reflection of the context of the utterance (including the speaker's beliefs) than part of what one is conveying by it; this view is ably defended in unpublished work by Osten Dahl of the University of Goteborg. Certain logical presuppositions are predictable from other things, for example, the existence presupposition that goes along with a definite description operator. On the other hand, I have no clear feeling as to whether the presupposition that the complement of the verb "realize" is true follows from any rule or should be simply an additional piece of semantic representation. Jerry Morgan 39 has pointed out that presuppositions are not presuppositions of the entire discourse nor even of the entire sentence necessarily, but just parts of sentences and discourses, for example the complements of certain verbs. The conclusions that Morgan has drawn about presuppositions tie in very closely with what I said earlier about possible worlds. Within a 'world-creating' context, you can use a factive verb which has as its complement not something that is presupposed to be true in the real world but something that is presupposed true in the counter-factual world under discussion. For example, the counterfactual if-then construction in the sentence "If Nixon had appointed Eldridge Cleaver Secretary of the Treasury, he would regret that had given a Negro a Cabinet post" sets up an imaginary world in which Nixon appointed Cleaver Secretary of the Treasury. The whole sentence has the presupposition that Nixon did not appoint Cleaver Secretary of the Treasury, and the factive verb "regret" carries with it a presupposition that Nixon gave a Negro a 39 J. Morgan, "On the Treatment of Presuppositions in Transformational Grammar", CLS 5 (1969), 169-177.

274

JAMES MCCAWLEY

Cabinet post relative to the imaginary world in which Nixon did appoint Cleaver Secretary of the Treasury. Do you see any possibility offormalizing in the semantic representation the contextuality of a sentence or a proposition? Is the distinction between linguistic and pragmatic context valuable ? - Regarding the linguistic context, it would simply be the derivations of the prior sentences of the discourse. These derivations involve performative verbs whose subject and indirect object indicate who addresses each sentence to whom. Regarding the situational context, I don't have any ideas about appropriate means of formalization, except that I really wonder whether any kind of formalization is appropriate beyond that which is inherent in the notion of logical structure. The relevant characteristics of the context would be simply propositions which the participants in the discourse believe. The various rules of grammar which pay attention to context would be simply interacting with the beliefs that the various participants have. To the extent that you need a formalization of context, it would be simply a set of propositions, with each proposition represented in logical form. What features of context are relevant to what linguistic phenomena varies from language to language. In Japanese, for example, there is a copula deletion rule which only women apply; in Dyirbal 40 demonstratives have different forms not only according to whether the referent is near or far (like English "this" and "that") but whether it is uphill or downhill, upstream or downstream. I have no feeling as to whether one should describe the language in terms of one gigantic set consisting of all of the things that each speaker happens to believe, with in principle any of those beliefs being linguistically relevant, or alternatively, to single out some very much reduced subset consisting of propositions that can serve as pragmatic presuppositions in the particular language.

In connection with this notion of linguistic context, I would like to ask you if you see any possibility of a linguistics of discourse ? - Yes, in the sense that to make a linguistic description of the sentence, you have to describe it in relation to its linguistic, and extralinguistic, context. There are, for example, deletion phenomena that are contingent upon having an identical item in an earlier sentence of the discourse. At least to that extent, an analysis of discourse is a reasonable endeavor, and generative semantics is engaged in analysis of discourse. Is there anything more to the analysis of discourse than that? Does it make sense to impose more structure on discourse than just to say that it is a sequence of utterances ? I suspect that there is in fact more, but in my present lack of understanding as to how discourse works, I cannot say what.

40

See R. Dixon, The Dyirbal Language of Northern Queensland (Cambridge, 1972).

JAMES MCCAWLEY

275

Some people, for example, make a typology of discourses, distinguishing between narrative, poetic, etc. discourse... - That's certainly worth doing. A failing of transformational grammar has been not distinguishing between different kinds of discourse. There are certain kinds of expressions which are perfectly normal in newspaper writing but not in ordinary speech. A newspaper article may begin "Senator Percy announced today that he would vote to disband the Postal Service. The Illinois Republican stated that ..."; but in conversation one could not say "Have you heard that Senator Percy is going to vote to disband the Postal Service? The Illinois Republican said that...". The whole question is how to introduce the type of discourse in the semantic representation of a sentence. - What we have here isn't a question of semantic representation but of different modes of expression. The various styles differ as to what factual information can be used as pragmatic presuppositions, though exactly the same meanings can be expressed in the different styles. In connection with pragmatic context, work is now done on 'conversational postulates'. How is that to be incorporated into grammar ? - Conversational postulates have to be paid attention to, if for no other reason, simply because linguists are involved with logic and conversational postulates are relevant to logic. David Gordon and George Lakoff 41 propose an extension of Grice's work 42 to describe what Jerrold Sadock 43 calls indirect speech acts: questions that are used as requests ("whimperatives") or as statements ("queclaratives"), etc. Lakoff and Gordon propose that what a sentence can be used for depends on its semantic interpretation plus interaction with rules of conversation, and that a "whimperative" such as "Would you please open the door?" has the same semantic structure as a question, but its semantic structure plus the conversational postulates imply that it can be used to make a request. One major thing that bothers me about Gordon and Lakoff's approach is that a lot of whimperatives don't have a 'literal meaning', that is, they can't be used as questions (for example, "Would you be a good little boy and hand me the newspaper?"). I don't have a clear idea of what the right treatment of "Would you open the door?" is, though I tend to agree with Sadock that syntactically and semantically it is some combination of interrogative and imperative; the specific proposal by Sadock that it is a conjunction of an interrogative and an imperative doesn't stand up at all, however.

41

D. Gordon and G. Lakoff, "Conversational Postulates", CLS 7 (1971), 63-84. P. Grice, "The Logic of Conservation", unpublished manuscript, 1968. 43 J. Sadock, "Whimperatives", in J. Sadock and A. Vanek (eds.), Studies Presented to Robert B. Lees by His Students (Edmonton, 1970), 223-238; "Queclaratives", CLS 7 (1971), 223-231; "Speech Act Idioms", CLS 8 (1972), 329-339. 42

276

JAMES MCCAWLEY

Do you consider the introduction of the philosophical theory of speech acts into linguistic theory as dangerous and premature ? - I find it beneficial and long overdue. I am committed to Ross's performative analysis44 (i.e., the claim that every sentence has an underlying structure whose highest verb is 'performative', that is, specifies the speech act being performed), since only the performative analysis provides a natural means of representing the difference in meaning between imperatives, questions, exclamations, and the many different kinds of declarative sentence and of making clear the syntactic function of the adverbial expressions in "Frankly, I'm disappointed in you", "If I'm not being too inquisitive, why was your father arrested?", etc. To carry out Ross's performative analyses properly, one has to get pretty deeply into details of speech acts. Do the uses of language and the intentions of the speaker have to be put into the linguistic competence ? - Treating competence as a system of rules for relating surface forms of sentences to semantic representations and information about contexts and intentions doesn't put use into competence; rather language use and competence are impinging on each other. An analogy with physiology may be helpful. It doesn't make any sense to say that the analysis of food is put into physiology when one talks about the physiology of the digestive tract in relation to the food that it digests; it makes equally little sense to say that use is put into competence when you talk about linguistic competence in relation to the uses of utterances. As a kind of conclusion to our conversation, two more general questions of great importance for philosophers. Do you agree with the statement that the form of language cannot be studied independently of its function ? - Yes. You can make some kind of generalizations about what is possible in a language without reference to what language is used for, but once you get into any interesting questions about language, even questions so prosaic as specifying what passive clauses are possible, you are forced to bring in what the sentence is used for. The rule given in Chomsky's Syntactic structures is only a first approximation to an adequate treatment of passives, and the later refinements of that rule still leave many mysteries unsolved. To make any sense of the enormous difference in acceptability among the passives which Chomsky's rule implies are grammatical, you have to take up the question of why people use passives.45 Practically any interesting problem about syntax leads one into questions of language use. What then has to be seen as the primary function of language ? Communication ? Expression?... - If you use 'communication' to mean transfer of information, then there is a lot 44

J. Ross, "On Declarative Sentences", in R. A. Jacobs and P. S. Rosenbaum (eds.), Readings in English Transformational Grammar (Waltham, Mass., 1970), 222-272. 43 See Robin Lakoff, "Passive Resistance", CLS 1 (1971), 149-162.

JAMES MCCAWLEY

277

more than that in language. The notion of semantic representation in generative semantics, which includes the whole gamut of speech acts, is of course much broader than communication in this narrow sense. I'm not sure that it makes sense to say that language has a primary function. I wouldn't say that any of the functions of the tongue is primary, for example. Lakoff wrote recently that linguistic theory is rooted in linguistic thought and culture. I am sure you agree. - Certainly. Language is intimately tied with the mind and with everything which is in the mind. To solve linguistic questions you have to pay close attention to what's in the rest of the mind. And that involves culture to a large extent, especially since the acquisition of language is intimately connected with the acquisition of the culture : a child learns words as he learns their uses and acquires knowledge of the things the words refer to. University of Chicago Chicago, Illinois October 27, 1972

SEBASTIAN K. SAUMJAN

Sebastian Saumjan, applicative grammar is a generative grammar. What does it mean to you that grammar has to be generative ? - Noam Chomsky uses the term 'generative grammar' in the context of such terms as 'a generative grammar of the English language', 'a generative grammar of the Russian language', 'a generative grammar of the Japanese language', etc. For Chomsky a generative grammar is always a generative grammar of some particular concrete language. Besides the concept 'generative grammar' in the Chomskyan sense, a fundamentally new concept may be introduced into theoretical linguistics, a concept which I call 'abstract generative grammar'. Why is abstract generative grammar necessary? Because to study natural languages it seems necessary to construct an ideal language, i.e., an artificial symbolic language which simulates the universal properties of natural languages. I call this ideal language the genotype language. The genotype language may be considered the semiotic basis of natural languages which I call phenotype languages. What are the aims of this 'abstract generative grammar' ? - The aims of abstract generative grammar include: the construction of the genotype language; the investigation of the formal properties of the genotype language; the study of the laws which govern the functioning of the genotype language; the investigation of the transformations of the genotype language into concrete natural languages and the construction of intermediary ideal languages which may serve as transitional stages from the genotype language to concrete natural languages; a semiotic typology of natural languages from the point of view of transformations of the genotype language and the projection of the typological features of natural languages in space and time; an explanation of the transformations of the genotype language from the point of view of semiotic laws and a prediction of the possible semiotic types of natural languages. The aims of the abstract generative grammar as set forth above call of course for a more detailed explanation. Can you dwell upon the most important epistemological characteristics of the genotype language ? - The genotype language exists objectively but it is not directly observable. That is

280

SEBASTIAN K. SAUMJAN

why we must construct it as a hypothesis. It is precisely in this sense that the term 'construction' is to be understood. Since our construction of the genotype language is of a hypothetic nature, other hypotheses are principally possible. The important thing, however, is that the genotype language is an objectively existing phenomenon. The question as to which of the constructions of the genotype language is closer to the objectively existing genotype language must be solved by the hypothetico-deductive method. There is a certain similarity between the problem of constructing the genotype language and the problem of reconstructing the parent language in comparative grammar. Just as comparative grammar considers related languages as different transformations of one and the same parent language lying at the base of the given group of related languages, so abstract generative grammar considers natural languages as different transformations of one and the same ideal language. There are, however, important differences. First of all, in the former case we have to do with genetic kinship whereas in the latter, with general semiotic affinity. Secondly, no parent language, unless it is attested with written records, is reconstructed as a functioning system, whereas the genotype language does possess this property. Thirdly, parent languages are arrived at as a result of the inductive process of comparing actual forms of existing natural languages, whereas the genotype language is constructed deductively on the basis of a very general notion about what the universal features of natural languages might be. We may study the degree of deviation of genetically unrelated languages from the genotype language. The genotype language has to be seen as the result of a mathematical calculus... - The genotype language is described in mathematical form. The genotype language is postulated as a calculus of semions. Semion - from the Greek arijisTov, sign - is its elementary semiotic unit. I propose this notion as the fundamental concept of semiotics comparable in its significance to the gene concept in biology. As in biology invariance manifests itself in a definite set of genes, so also in semiotics invariance manifests itself in a definite set of semions. This justifies the use in semiotics of the terms 'genotype' and 'phenotype' borrowed from biology. From the mathematical point of view any semion may be regarded as a function and the semion calculus as a calculus of functions of a special kind. In introducing into theoretical linguistics the concept 'abstract generative grammar' and constructing the genotype language, I have derived my inspiration from the works of logicians who pratice the construction of ideal languages, in particular, from the works of Rudolf Carnap devoted to logical syntax and semantics.1 The ideal languages constructed by Carnap are defined by two kinds of rules: rules of formation and rules of transformation. By the same token, the genotype language is also defined by rules of formation and rules of transformation. It goes without saying that the aims of constructing the genotype 1

Cf. R. Carnap, Meaning and Necessity: A Study in Semantics and Model Logic (Chicago, 1958); The Logical Syntax of Language (Paterson N. J., 1959).

SEBASTIAN K. SAUMJAN

281

language are very different from the aims of constructing ideal languages as practised by logicians. The genotype language is constructed with a view to empirical study of natural languages. As for logicians they construct their ideal languages with a view to studying the laws of various logical systems. But at the same time I would like to stress that the methods pertaining to the construction of ideal languages form common ground for the collaboration of linguists and logicians, because the necessity of constructing ideal languages is no less imperative for linguists than for logicians. How can such a largely hypothetico-deductive method be used when an empirical science has to build up its procedures? - It seems to me that the construction of the genotype language does not make linguistics a non-empirical science. Linguistics has always been and will always remain an empirical science. But the fundamental point is, what should be considered to be empirical research. I understand empirical research in the spirit of the hypothetico-deductive method. A linguist who makes empirical research on the lines of the hypothetico-deductive method need not be at all concerned with non-essential empirical constraints as, for instance, the linear order of linguistic units, or certain syntactic categories. A theoretically-minded linguist should make use of the liberty from non-essential empirical constraints and he should freely invent ideal languages which simulate the essential empirical features of natural languages. What is really 'empirical' is not given to direct observation; it is veiled from us and must be discovered by way of theoretical reflection. As a result we construct an ideal language which should be considered a hypothesis about essential empirical features of natural languages. It goes without saying that all empirical features of natural languages should be given due attention, but the point is that we should draw a distinction between essential and non-essential features of natural languages to find out how nonessential empirical features are derived from essential empirical features. In order to do this we introduce special rules, so-called rules of correspondence, which transform the genotype language into intermediary ideal languages and the latter ones into particular concrete natural languages. Why is the abstract generative grammar called applicative grammar ? - This is because the sole operation in abstract generative grammar is the application operation. Applicative grammar may be considered a special kind of formal system of the type which H. B. Curry calls applicative systems.2 The application operation may be defined thus: if X is a one-place function and Y is its argument, (XY) is the value of the function X. We say that the function X is applied to its argument Y. I also use another terminology. I call X the operator, Y the operand, and (XY) the result of the application of the operator X to its operand Y. Applicative grammar operates on semions. From the point of view of the distinction between operators and 2

H. B. Curry and R. Feys, Combinatory Logic (Amsterdam, 1958); H. B. Curry, Foundations of Mathematical Logic (New York, 1963).

282

SEBASTIAN K. SAUMJAN

operands there are two kinds of semions: first, semions which can be used only as operands, and secondly, semions which can be used either as operands or operators. - What are in this respect language universals ? - From the point of view of applicative grammar, language universals are universal linguistic categories. To define the fundamental linguistic categories we resort to a mental experiment. If we disregard all that may be considered secondary in natural languages for the process of communication between the speaker and the hearer, we may reduce the fundamental linguistic categories to two primitive categories: the names of objects, or terms, and the names of situations, or sentences. We shall illustrate by concrete examples how the principal linguistic categories can be reduced to the above meDtioned primitive categories. What for example is the adjective ? The adjective is a category which, applied to a term, generates another term: if we apply the word "hairy" to the word "dog", we shall get "hairy dog" which is also a term. If we apply the word "big" to the obtained term we shall get a new term, "big hairy dog", etc. Using the mathematical terms 'function', 'argument', 'value', we may regard the adjective as a function which, if applied to a term as its argument, will again yield a term as its value. When we speak about application we mean the application of a function to its argument. Let us consider some examples. What is the predicate ? The predicate is a function which, if applied to a term as its argument, generates a sentence as its value. For example, if the word "shines" is applied to the word "sun", the sentence "(The) sun shines" is obtained. What is the adverb ? There are two types of adverbs: first, the function, which, when applied to the predicate as its argument, generates a new predicate as its value ("walks quickly"); and secondly, the function, which, when applied to the adjective as argument, generates again an adjective as its value ("very beautiful"). The above considerations may serve as a starting point for the construction of a formal system. Our formal system consists of two parts: the episemion calculus and the semion calculus. These are crucial notions of your linguistic theory. Can you explain somewhat further ? - Episemions are linguistic categories as such, i.e., linguistic classes viewed as abstract object. Semions are linguistic objects belonging to a linguistic class, i.e., to a linguistic category. We shall denote terms by the symbols T, T 1 , T 2 ..., and the linguistic category to which they belong by the symbol a; then the notation aT will mean that the semion T, interpreted as a term, belongs to the episemion a, interpreted as a category or a class of terms. In the more general form the notation eX will mean that the semion X belongs to the episemion e.

SEBASTIAN K. SAUMJAN

283

The calculus of episemions has two principal initial objects a and (3, called elementary episemions, and one auxiliary initial object A. All the other episemions are defined by the following rules: first, the elementary episemions a, p are episemions; secondly, if p and q are episemions, A pq is also an episemion. The second rule may be represented as a scheme called a tree diagram: P

q A pq

Here the horizontal line means entailment, i.e., p and q entails A pq. The formula of the episemion A pq is read as follows: the class of semions transforming the semions of p into the semions of q. The process of generating episemions breaks down into steps. Any transition from the episemion p and q to episemion Apq is called a step in the generation of episemions. At the first step we get the episemions of the first degree of derivation: Aaa, App, Aap, Apa. At the second step we get episemions of the second degree of derivation, e.g., AAaPa, AAapAap, AaAap. At every step of episemion generation the episemions are derived from the episemions obtained at all the previous steps. Any episemion obtained at the n t h step is called an episemion of the n t h degree of derivation. Let us now give an empirical interpretation of some episemions. The interpretation of each episemion given below is only partial because it will be given only for English. In the scheme that I would like to draw, I mention in the first column the episemion, and in the second column its interpretation. a

P Aaa AaAaa AaP AaAaP AaAaAap APa App ApApp AAaaAaa AaAAaaAaa AAapa AAapAap AaAAapAaP

Common and proper nouns Sentences Adjectives or affixes forming nouns from nouns Prepositions, the possessive case affix and derivational affixes transforming nouns into attributes or other nouns Finite forms of intransitive verbs Finite forms of verbs with one object, verbal copulas Finite forms of transitive verbs with two objects The conjunction "that" Negation or question Subordinating conjunctions connecting two clauses Pre-adjectival adverb, affixes forming adjectives from adjectives Prepositions, transforming nouns into determiners of adjectives Affixes transforming verbs into abstract nouns Adverbs modifying intransitive verbs, affixes forming intransitive verbs from intransitive verbs Affixes forming adverbs from nouns.

284

SEBASTIAN K. SAUMJAN

The first two steps of the construction of aformal language have been done: the postulates are given and the terms of the system are interpreted. But a further step, I believe, would be the generation of the calculus of entities... - Let us move on to this semion calculus. In the semion calculus initial objects are postulated which are called elementary semions. The following rules of generating semions are postulated: first, elementary semions are semions; and secondly, if X is a semion belonging to the episemion Apq, and Y is a semion belonging to the episemion p, then XY is a semion belonging to episemion q. The second rule may be represented as the following tree diagram: A pq X pY q XY ' We shall call X the operator (or function), Y, the operand (or argument), and (XY), the result of the application of X to Y. The second rule will be called the rule of the application of semions. The generation of semions proceeds in steps. A step in semion generation is any transition from semion X to semion XY. At the first step we obtain semions consisting of two elementary semions, at the second step, semions consisting of three or four elementary semions, etc. A semion generated at the n t h stage is called a semion of the n th degree of derivation. The totality of all the semions which can be generated is precisely what we call the genotype language. The rules of semion generation considered above may be called the syntax of the genotype language. Besides the rules of semion generation there are also rules of semion transformation. The rules of semion transformation define the relations of meaning inclusion between semions. A pair of semions for which the relation of meaning inclusion holds in both directions is said to be in the relation of meaning equivalence or synonymy. The rules of semion transformation may be called the semantics of the genotype language. The semantics of the genotype language is the principal part of the applicative grammar because it is semantics that makes it possible to reveal how complex semantic structures may be deduced from a few primitive semantic structures. To construct the rules of semion transformation we use functions of a very general kind which Curry calls combinators. You explicitly state that applicative grammar, being generative, is at the same time 'structural'. In the past decade, 'generative grammar' has almost always been opposed to 'structural linguistics'. Your grammar is both ''structural' and 'generative'. What do you understand by 'linguistic structure'? - Structure may be defined as a network of relations between objects of a given set of objects. We may represent a structure by means of a graph, especially by means of a tree. Every node of a tree corresponds to an object of a given set of objects, and the lines connecting the nodes of a tree correspond to the relations between the objects. If we have a set of linguistic objects interconnected by linguistic relations, then we have a linguistic structure which we may represent by a tree or by a formula corre-

SEBASTIAN K. SAUMJAN

285

sponding to the tree. Depending on the kind of linguistic objects corresponding to the nodes of a tree we may distinguish two types of synchronic linguistic structures: static linguistic structures and dynamic linguistic structures. Take the following tree: A

/

B

\

C

If we consider this tree as an immediate constituent tree, we have to interpret it as follows: the object A consists of the objects B and C. If we consider this tree as a dependency tree, we have to interpret it thus: the objects B and C depend on the object A, or objects B and C are subordinate to the object A. If we consider this tree as an applicative tree, we have to interpret it thus: the object B changes the object C into A. In this case B is called an operator, C is called an operand, and A is called the result of the application of the operator B to the operand C. The application of the operator to the operand, or, for short, the application, is a two-place operation whose first argument is the operator and whose second argument is the operand. An applicative tree represents a relation between linguistic objects as a process of transformation of an element of one linguistic class into an element of the same linguistic class or into an element of another linguistic class. For instance, what is the relation 'adjective + noun'? This is a process of transformation of a noun into a noun. What is the relation 'noun + verb'? This is a process of transforming a noun into a sentence. What is the relation 'preposition + noun'? This is a process of transforming a noun into an adverb. The application operation should be understood as a process transforming one linguistic class into another. Thus, applicative trees may be used as a tool for the description of a language as a dynamic system of linguistic classes. The above considerations show that immediate constituent and dependency trees represent static linguistic structures and applicative trees represent dynamic linguistic structures. The description of a sentence by means of an applicative tree involves a new dimension, which may be called the dimension of syntactic time. Every stage of a syntactic process consists of an application of an operator to an operand. Let us turn to further aspects of the problem. We may consider the structure of a sentence to be stratified into a series of structures; accordingly, we may consider the applicative tree representing the structure of a sentence to be stratified into a series of applicative trees representing a series of structures. In this case we may speak of a semantic description of a sentence. From the point of view of applicative grammar the semantic description of a sentence is a series of ordered applicative trees of which each represents a structure of the sentence at a given stage in semantic process. A semantic description of a sentence involves a new dimension, which may be called the dimension of semantic time: we have an initial structure, a terminal structure and a finite number of intermediate structures. Having an initial structure,

286

SEBASTIAN K. SAUMJAN

we may generate from it several terminal structures .We have a finite set of intermediate structures between the initial structure and each terminal structure. A semantic process resulting in derivations of a finite set of terminal structures from an initial structure may be also represented by a tree called the tree of semantic derivations. In this case every node of the tree will correspond to some linguistic structure at a given stage of semantic process, and the lines between nodes will correspond to a semantic time relation between two given structures one of which is immediately derived from the other. The root node of the tree will correspond to the initial structure, the terminal nodes will correspond to the terminal structures, and the intermediate nodes will correspond to the intermediate structure. Our tree will correspond to a structure which represents the process of deriving one dynamic syntactic from the other. I call this structure the dynamic synchronic semantic structure. A dynamic synchronic semantic structure is a network of semantic time relations between applicative trees. How do you distinguish between a dynamic semantic structure and a diachronic one ? - Let us turn to diachrony. From the point of view of diachrony the dynamic synchronic semantic structure of a sentence is not a network of time relations between applicative trees but a single event which from the point of view of diachrony may be called a diachronic event. The description of a diachronic process involves one more dimension which may be called the dimension of diachronic time. The diachronic process may be represented by trees of a special kind, called the trees of diachronic process, or, for short, diachronic trees. The root node of a diachronic tree will correspond to the initial sentence which underwent various changes in the course of the diachronic process; the terminal nodes will correspond to the sentences which are final results of the diachronic process, and the intermediate nodes will correspond to the sentences which appeared at the intermediate stages of the diachronic process. In my book Principles of Structural Linguistics3 I proposed to restrict the term 'structure', as regards synchrony, to the network of linguistic relations, pertaining to the dynamics of synchrony, and I proposed the term 'quasi-structure' to name the network of linguistic relations pertaining to the statics of synchrony. Now I would not like to quarrel about terms. The essential fact connected with the differentiation of structure and quasi-structure is this: we must take a strict distinction between statics and dynamics in synchrony and consider dynamics to be a fundamental aspect of synchrony. Whether or not one adopts the term 'quasi-structure' is not essential. It is essential, however, to distinguish two aspects of synchrony: static and dynamic. If one prefers the term 'structure' in a broad sense, in this case, it is essential to draw a distinction between two kinds of structure in synchrony, static structures and dynamic structures, and acknowledge the fact that the dynamic structures are the fundamental ones. 3

S. K. Saumjan, Principles of Structural Linguistics, English translation by J. Miller (The Hague, 1971).

SEBASTIAN K. SAUMJAN

287

Traditional grammar distinguished between phonology, syntax and semantics. This fully remains that way in generative transformational grammar. Do these distinctions fit into your linguistic theory ? - I think that the traditional distinction of generative grammar into three components - syntax, semantics and phonology - should be reconsidered. To decide what components are essential to generative grammar it is useful to start from a strict differentiation between signifié and signifiant or, in Hjelmslev's terminology, between the content plane of natural languages separate from the expression plane. I call the simulation of the content plane the conceptual component of generative grammar. The conceptual component of a generative grammar should have two sets of rules: rules of formation and rules of transformation. I call the first set of rules the syntax of a generative grammar and I call the second set of rules the semantics of a generative grammar. The complex units of the conceptual component are always combinations of operators and operands. The combinations of operators and operands must have the following standard order: an operator always precedes an operand. This standard order has nothing to do with the order of units in concrete natural languages. It can be viewed as a typological standard for studying various regular deviations in the order of units in natural languages. To simulate the conceptual plane of a generative grammar it is necessary to construct an abstract universal language. The genotype language is nothing but an abstract language simulating the content plane of concrete natural languages. To construct concrete generative grammars of natural languages we must introduce for every concrete natural language a set of rules transforming the genotype language into the conceptual plane of the given natural language. The next step in constructing a generative grammar of a given natural language consists in introducing a special set of linearization rules. This set of rules can be called the linearization component of a generative grammar. As a result of linear transformations of phenotype semion combinations, for each natural language we get specifically ordered linear structures belonging to the content plane. The next stage is to introduce a mechanism into the generative grammar to map the structures of the content plane into the units of the expression plane, i.e., into phonological objects. As is well known, the elementary units of phonology are distinctive features. It is with the help of the universal set of distinctive features that the diverse content plane structures are coded. As to the phoneme, it is reasonable, within the framework of generative grammar, to consider it simply an abbreviation for a bundle of distinctive features. The elementary unit of morphology is the morph. A morph is the minimal unit which is the result of mapping an elementary semion or a bundle of elementary semions onto a set of phonological objects. Thus, a morph consists of a signifié and a signifiant. The signifié of a morph is a simple or a complex semion, i.e., a simple or a complex object, i.e., a unit of the expression plane, which is further indivisible from the point of view of its correlation with the content plane. A class of morphs with an

288

SEBASTIAN K. SAUMJAN

identical signifié and regularly varying signifiants is called a morpheme. Morphs representing one morpheme are called allomorphs of this morpheme. In applicative grammar morphological rules are rules for mapping the objects of the content plane onto the objects of the expression plane. The last component of a generative grammar is the phonological component. The goal of the phonological component is an explanation and prediction of the differences between allomorphs with the help of a set of assumptions about a hypothetical abstract phonological system and hypothetical sound processes which transform the abstract phonological system into the concrete phonological system. Can you summarize your picture of the components of a generative grammar indicating the relationships between them ? - In our conception we should distinguish two levels in the generative process. At the first level the generative mechanism generates the genotype language which is an ideal language simulating the content plane of the linguistic units. At the second level the generative mechanism transforms the genotype language into various conceptual components of natural languages which by means of morphological and phonological components are transformed into linguistics units of the concrete natural languages. Our conception of the components of concrete generative grammar can be represented by the following diagram I would like to draw for a better understanding of my linguistic theory.

Genotype language: (1) syntax (formation rules), (2) semantics (transformation rules)

Phenotype languages:

1 2 3 4

languagei

languages

conceptual componenti linearization componenti morphological componenti phonological componenti

conceptual component2 linearization component2 morphological component2 phonological components

...

languagen conceptual componentn linearization componentn morphological componentn phonological componentn

Central as well in generative transformational grammar of the Chomsky type as in applicative grammar stands the notion of transformation. Can you tell me the specific-

SEBASTIAN K. SAUMJAN

289

ness of your conception of transformation in order to clarify the difference between your language conception and the Chomskyan linguistic theory? - In applicative grammar the notion of transformation is connected with the semantic process in the genotype language which is an ideal artificial language. In transformational grammar the notion of transformation is connected with the conversion of deep structures into surface structures in concrete generative grammars. This fundamental difference between the notion of transformation has formal consequences: the transformations in transformational grammar do not form a formal system in the mathematical sense. Of course, there have been attempts to construct the transformational component of transformational grammar as a formal system in the mathematical sense. But all these attempts have led to systems which are far from linguistic reality: these systems have no empirical interest. On the contrary, in applicative grammar the transformations form a formal system in the strict mathematic sense, and this mathematical system is called formal semantics. What has been the principal source of inspiration in building up your concept of transformation ? - For formal semantics I have adapted the system of combinators which are part of combinatory logic as developed by Curry and Feys.4 In fact the notion of transformation rule in transformational grammar and the notion of the transformation rule in applicative grammar are so fundamentally different that it seems necessary to introduce a new term for the transformation rules in applicative grammar. In applicative grammar I propose to speak of semantic derivation rules instead of transformation rules. For the sake of clarity I shall give an example of semantic derivation rules. Semantic derivation rules are effected with the help of combinators, which are a very general kind of operators on functions. Combinators are applied to one-, two-, ..., n-place predicates of the basic genotype language to derive predicate structures of the derived genotype language. For example, the combinator B, a compositor of functions, is used according to the formula: F (GX)

BFGX

(1)

Here the compositor B is used to obtain one complex function BFG from two simple functions F and G. If F is interpreted as the predicate of the matrix sentence and G as the predicate of the embedded sentence, then BFG may be interpreted as a complex predicate of a derived sentence; that is, on the phenotype level, if F corresponds to the predicate "see" in the sentence "I see", and G corresponds to the predicate "runs" in the sentence "The boy runs", then BFG corresponds to the complex predicate "see run" in the sentence "I see the boy run". The combinator L is a confluentor of functions with identical arguments: FX(GX) ->• LFGX 4

Cf. note 2.

(2)

290

SEBASTIAN K. SAUMJAN

In (2) a complex function LFGX is obtained from two simple functions F and G. The two identical arguments are fused into one X. On the phenotype level the confluentor L may be interpreted as an operator which generates infinitival causative constructions from an underlying complex sentence. That is, if F corresponds to the predicate "persuade" in the matrix sentence "I persuaded John", G corresponds to the predicate "come" in the embedded sentence "John comes" and X corresponds to the object of the matrix sentence and the subject of the embedded sentence "John", then LFG corresponds to the complex predicate "persuaded to come" in the derived sentence "I persuaded John to come". I cannot dwell on the whole set of combinators used in the semantic derivation rules of the applicative grammar but I wish to emphasize that they form a unified system which adequately simulates the semantic processes in natural languages. Mathematics is of enormous importance for the construction of your linguistic model. Can you circumscribe the exact role of mathematics in the study of language ? - The importance of mathematics in linguistics is in providing a variety of methods for arranging hypotheses into a system; mathematics opens up new possibilities for the construction of various systems of linguistic hypotheses. A hypothetico-deductive linguistic system built with the aid of mathematics plays a fundamental role in the following respect. It makes explicit the explanatory and predictive power of the theory by making us represent all the data at our disposal in an explicit form. Only by means of mathematical systems can we detect lacunas in our knowledge of linguistic reality. Without using mathematics it is difficult to delineate the boundaries between what we know and what we do not know. Only mathematics gives the possibility of formulating a problem in a precise form and drawing precisely the necessary consequences from the initial hypothesis. We should, however, point out the following fundamental difficulty in applying mathematics to the study of language: as is well known, mathematics can be applied to the study of a well-defined object, but can we say that a natural language is a well-defined object in the mathematical sense? In other words: is a natural language a mathematical object? There cannot be the slightest doubt that our answer to this question must be in the negative. A natural language is an immensely involved system which is a mixture of the rational and irrational, and this system defies direct mathematical description. In the light of this answer it is interesting to consider the view of Noam Chomsky who writes: "The fundamental aim in the linguistic analysis of a language L is to separate the grammatical sequences which are the sentences of L from the ungrammatical sequences which are not sentences of L, and to study the structure of grammatical sequences. The grammar of L will thus be a device that generates all of the grammatical sequences of L and none of the ungrammatical ones. One way to test the adequacy of a grammar proposed for L is to determine whether or not the sequences it generates are actually grammatical, i.e. acceptable to a native speaker...." 5 5

N. Chomsky, Syntactic Structures (The Hague, 1957), p. 13.

SEBASTIAN K. §AUMJAN

291

In Chomsky's view the grammar of a natural language is a device that generates all of the grammatical sequences of this language. And as we know, the concept 'generation' has been taken by Chomsky from mathematical logic. Accordingly, to generate the sentences of a language means to specify the set of the sentences of this language by means of mathematical rules. The question now arises: how can we generate a non-mathematical object by mathematical rules ? - The difficulty can be solved if we use the method of idealization. We cannot generate the sentences of a natural language, but we can generate the sentences of an ideal language simulating natural languages. The ideal language is a theoretical construct, an artificial language. The distinction of two kinds of languages - natural languages and artificial language simulating natural languages - involves splitting the concept 'grammar'. We must now have two grammars: first, a grammar generating the sentences of an artificial language simulating natural languages; and secondly, a grammar which is a set of rules of correspondence between the sentences of the artificial language and the sentences of natural languages. The first grammar will be called a genotype grammar, and the second grammar, a phenotype grammar. The genotype grammar generates an artificial language simulating natural languages which will be called a genotype language. The phenotype grammar describes by means of rules of correspondence the relation of the genotype language to natural languages which will be called phenotype languages. A fundamental difference between a genotype and a phenotype grammar is this: a genotype grammar specifies a language which is well-defined in the mathematical sense, but a phenotype grammar specifies a language or languages which are not well-defined in the mathematical sense. Hence the empirical character of the rules of correspondence of the phenotype grammar, which means that only a subset of all the formulae of the genotype language can find an interpretation in natural languages. I call the explicit distinction of genotype and phenotype languages and of genotype and phenotype grammars the two-level principle in the theory of generative grammars. Another but related question in this field is: what is, according to you, the role of mathematical logic in generative grammar ? - In my opinion mathematical logic is important for generative grammar in two respects. First, mathematical logic is a tool for constructing formal systems in general and, in particular, a tool for constructing formal systems in linguistics. Generative grammars are nothing but formal linguistic systems. As for me, I consider the new branch of mathematical logic which is called combinatory logic particularly important for constructing formal linguistic systems. I have already discussed the importance of combinators for formulating semantic rules in generative grammar. In general, combinatory logic seems to me to be a valuable mathematic tool for the linguist. Secondly, in constructing a generative grammar we must take as a starting point the

292

SEBASTIAN K. SAUMJAN

fundamental notions of mathematical logic. For instance, we must conceive of the structure of a sentence at the deeper level of natural languages as consisting of predicates and arguments. We should consider every sentence as consisting of an n-place predicate and n arguments. In the light of this consideration we should give up the distinction between subjects and objects at the deepest levels of structure of natural languages. The distinction between subjects and objects belongs to the surface level of natural languages. Thus, mathematical logic should serve as a source of primitive semantic concepts for natural languages. These semantic concepts must underlie the traditional syntactical notions of linguistics, as has already been said in connection with such traditional syntactical notions as subject and object, i.e., the semantic structure 'predicate + arguments' must underlie the syntactic structure 'subject + predicate + objects'. Let's return to some philosophical background problems concerning the status of your linguistic theory. Applicative grammar has been built up by hypothetico-deductive method. What are the general characteristics of the procedures of this method? - The essence of the hypothetico-deductive method may be represented by the following diagram: D

P The diagram is a circle in which the arrows show the direction of the cognition process. The point P represents an initial problem, the point T represents a theory which is advanced as a system of hypotheses, the point D represents a set of deductive consequences from the theory, the point E represents the verification of the theory leading to the elimination of errors. The arrow from E to P shows that after E we return to the same problem, at a new level of the cognition process. The above scheme shows that the hypothetico-deductive method is a cyclic cognition process which consists of four stages: the first stage of the problem which arises as a result of the contradiction between the facts of reality and the existing theoretical

SEBASTIAN K. SAUMJAN

293

notions; the advancing of a theory as a system of hypotheses to solve the problem; the deduction of all possible consequences from the advanced theory; the verification of the theory by confronting all its consequences with new facts of reality and eliminating from it all the errors revealed as a result of the verifications. What is the function of a natural language ? - Every natural language has two functions: a communicative function and a function of expressing thought. These functions are complimentary. Communication implies expression of thought, and expression of thought implies communication. Does cybernetics have a constitutive role in stating a linguistic theory ? - A s is well known the main task of cybernetics is the study of the process of converting information from one form into other. This conversion may be realized by means of various logical machines. For instance computers may be considered such logical machines. A s regards natural languages, they convert information by means of translation processes. The system of translation processes may be formalized and represented as a logical machine. In applicative grammar this logical machine is nothing but the formal system which I call formal semantics. For formal semantics the fundamental relationship is translatability. Thus, generative grammar may be regarded as the linguistic branch of cybernetics if we base that part of generative grammar which I call formal semantics on the translatability relationship. A lot of trends in contemporary linguistics are explicitly tributary to information theory. Is that also the matter with applicative grammar ? - A s is well known, information theory studies phenomena of a stochastic nature. In that phenomena of a stochastic nature play an important part in natural languages, information theory is of considerable importance for linguistics. Nevertheless it is necessary to point out that language structure is a network of relations of a non-stochastic nature, which can be described by the formal apparatus of mathematical logic and other non-stochastic branches of mathematics. In this connection we may say that, in comparison with mathematical logic and algebra, information theory is of secondary importance for linguistics. The very philosophical problem concerning the relationship between language and the world can be concretized in linguistic theory as the question about the value of the referential or denotative function. What is your opinion here ? - I think that linguistics must study meaning, but not denotation. Denotation is the province of philosophical or logical semantics. What is meaning from the linguistic point of view ? The meaning of a linguistic expression X may be defined as a class of sentences X i , X2,...,X n into which the expression X can be translated. This definition of meaning is based on the relation of translatability which must be considered basic for linguistic semantics.

294

SEBASTIAN K. SAUMJAN

If we adopt this definition of the meaning of a linguistic expression, we may fruitfully compare the notion of meaning in linguistics with the notion of commodity value in economics. Meaning does not exist in a pure form but is always contained in a linguistic shell. By the same token, value does not exist in a pure form but is always contained in a commodity shell. Just as the value of a commodity may be discovered only if we take the given commodity in its relation with other commodities which have an equivalent value, so the meaning of a given linguistic expression may be discovered only in its relation with other expressions which have an equivalent meaning. The above method of defining meaning and value is called in mathematical logic definition by abstraction. What is the specific nature of linguistics in particular, and of human sciences in general, in the whole range of the sciences ? - The specific nature of the human sciences, and in particular of linguistics in science as a whole, lies in my opinion only in the specificity of their subject matter: they are sciences about man. As regards the method of study, all sciences must use the same fundamental method: the hypothetico-deductive method. Of course, every science has also its particular methods which are determined by its subject matter, but from the epistemological point of view differences in specific methods between sciences are of secondary importance. Is the opposition of mentalism versus empiricism, as the paradigmatic epistemological opposition in a lot of current trends, for example in American linguistics, valuable ? - In recent American linguistic literature an important place is given to the discussion of the fundamental differences between two approaches to language study - the approach of descriptive linguistics which regards advancing hypotheses as an unscientific procedure and uses only operational procedures, and the approach of generative grammar which uses hypotheses as a main tool of research and considers operational procedures to be of secondary importance. I think that the problem mentalism versus empiricism is significant only in the context of the recent development of American linguistics. As regards European structuralism - Saussure, Hjelmslev, Trubetzkoy, Jakobson and others - there never was an opposition between mentalism and empiricism. European structuralism was always 'mentalistic' in the American sense of the word. From the point of view of American descriptive linguistics, mentalism is a revolution in the science of language. From the point of view of European structuralism, mentalism is a thing which should be taken for granted in linguistic research. In this connection I would like especially to mention Hjelmslev who in his Prolegomena to a Linguistic Theory6 set up as a primary task of linguistic research the creation of an abstract linguistic deductive system which would be called linguistic algebra. 0 L. Hjelmslev, Prolegomena to a Linguistic Theory, translated by Francis J. Whitfield (Madison, Wisconsin, 1961).

SEBASTIAN K. SAUMJAN

295

/ would like to conclude with a more historical and anecdotal question: what are the sources of applicative grammar or how do you sketch its genealogy ? - I would like to point out my indebtedness to two current trends in the study of language. One trend originates with logicians: I have in mind the works of Husserl, Lesniewsky, Adjukiewicz, Bar-Hillel, Lambeck, Curry, Feys and others. The other trend originates in linguistics under the name of structuralism. I take the term 'structuralism' in a broad sense to include such works as those of Roman Jakobson and those of Chomsky and his colleagues. Applicative grammar synthesizes some of the fundamental ideas of these trends in logic and linguistics. Academy of Sciences of the U.S.S.R. and Moscow University Moscow, U.S.S.R. December 25, 1972

SYNTHESIS

This panorama of the most representative trends in contemporary linguistics may have demonstrated the diversity of the conceptual apparatus and the disparity of the methodologies. Equally different are the underlying philosophic presuppositions about what language actually is and what is its function for the individual, society and culture. This diversity may cause surprise, especially since linguistics is most entitled among all human sciences to claim a scientific label: the requirement of a strong and critical methodology is present in the entire history of twentieth century linguistics, and all important linguists of our time - among whom our ten contributors can certainly be reckoned - reflect, parallel to their descriptive and technical work, the bases of their scientific activities. The realization that the scientific character of linguistics must be warranted has been constantly present from Saussure and Bloomfield onwards in all linguists of all types of persuasion, in Europe as well as in the United States. Notwithstanding this factor, which should promote a methodological uniformity, the diversity of points of view and of fields of interest is the most important characteristic of this volume of conversations. Not that the explicitly thematized contrasts and contradictions always stand up on closer examination. The purpose in the conversations and in the following discussion with Jacques Bouveresse was exactly to shade and undermine these contrasts: structuralism versus generativism, rationalism versus empiricism, universalism versus relativism, formal versus linguistic semantics, and semantics versus pragmatics are all polarities of which the terms can no longer, after some investigation, be presented in all their clarity. The confrontation realized in this book probably brings to light the fact that other and implicit basic options which are worth being thematized and studied in their philosophic import are at work. It is, moreover, the purpose of the discussion now following to discover the true lines of force and the real diversity in the panorama of contemporary linguistics. It is not the fact that the philosopher can claim the right to delineate the linguistic field a priori and to press a methodology, any more than he is authorized to interfere in technical questions and detailed research. On the pertinence of language philosophy for linguistics and vice versa, linguists as well as philosophers disagree. With Vendler, we accept the tripartite division into philosophy of language, philosophy of linguistics, and linguistic philosophy, with the specification that these three disciplines delimit each other osmotically. To avoid all misunderstanding, it has to be mentioned that none

298

SYNTHESIS

of these philosophical subdisciplines precedes linguistics or gives immediate content to linguistic research (as Katz and Fodor believed when developing their semantic theory); they are either a posteriori, parallel, or separated from the linguistic approach to language. The philosophy of linguistics or methodology of linguistics subdivision of the philosophy of (human) sciences is a posteriori to linguistics as it classifies and hierarchizes the conceptual apparatus used in it, and combines the often fragmentary theoretical indications in the linguistic work; this requires rigorous attention to the technical work in linguistics and a good sense of synthesis. Linguistic philosophy is parallel to linguistics because they both start as the study of the structure and functioning of natural and formal languages; it is here that a mutual fertilization has taken place between generative linguistics and Anglo-Saxon philosophy of language for some time, and that the problem of language knowledge, language acquisition, the relation of language to perception, etc. is posed. Linguistic philosophy is, therefore, the most central and fertile area where the philosopher and the linguist meet, although here also the autonomy of linguistics cannot be affected. Philosophers - Wittgensteinian as well as continentally inspired - often emphasize the uniqueness of the philosophy of language which is separated from the linguistic approach and can at most serve as a standard to which the exhaustiveness, the totality of the scientific approach to language, can be tested. This does not alter the fact that all linguists who have contributed some renewal are constantly inspired by effective presuppositions of a language philosophy, even of an ideological nature. In the preceding group of conversations, sufficient examples can be found where linguists explicitly assume these presuppositions. Yet, it is with a strong realization that we cannot or should not want to raise a priori frames that Jacques Bouveresse and I have taken up a discussion of linguistics. We are of the opinion that the autonomy of both disciplines should not be exalted or underestimated; what matters is not the delineation ofthebordersinadefinitiveandprincipledway,buttoinvestigate with a pragmatic and realistic attitude how one's own preoccupations and approaches can be fertilized. Language philosophers as well as linguists, whatever color their activities may have, have linguistic phenomenality as their object, although from a differentpoint ofview. The discussion now following is concerned with philosophy of linguistics as well as with linguistic philosophy and philosophy of language. Without too many qualms one field is visited after another, but always reflecting the problems and suggestions made in the ten conversations with linguists. The topics were ordered as follows. In the first part, the question is explicitly the status of linguistics as a science: how should a linguistic description, explanation, method, be defined? Can interdisciplinary research be justified? What is the place of linguistics in the framework of the human sciences? What is the relation of linguistics to socio- and psycholinguistics ? What does it mean that linguistics is a hypothetico-deductive science ? This first part of the discussion is, therefore, clearly methodologically and even for a large part epistemologically oriented. In the second part the rationalism-empiricism debate is under discussion, with all the questions it involves: the complicity of structuralism-behaviorism-empiricism, Cartesian and Chomskyan 'mentalism\ the role of the environment in language acquisition and

SYNTHESIS

299

language use, universalism versus relativism, language universals, linguistic creativity, the innateness hypothesis, the introspection criterion, and linguistic intuition. It seems to me that, especially within this thematization, the explicit stands in contemporary linguistic theory do not always faithfully reflect the actual philosophic presuppositions. It is probably here that the philosopher can exercise a beneficial cathartic function. In the third part it is observed that linguistics since Saussure constructs the object of its knowledge by means of dichotomies of which the most radical and constant is that of language and speech, competence and performance. These dichotomies, which play an important methodological role, are criticized in the most recent developments of structural as well as generative linguistics as reduction mechanisms which artificially delineate the domain of linguistics. A series of questions and suggestions are immediately associated with this: does it make sense to adopt a 'semantic competence' next to a 'syntactic competence'; can performance constitute a theory; how is linguistic competence connected to other (behavioral) competences; can the notion of competence be extended? It becomes evident, especially within this area of discussion, how the methodological requirement of delineating an object of knowledge has a temporary character and how dichotomies can be redefined according to the stage of empirical research. In the fourth part, attention is paid to the renewed interest in the meaning of language. Foremost is the currently very interesting question on the relation between grammar and semantics, or of syntax and semantics. In light of the flourishing of linguistic semantics, two central philosophical theories of meaning are presented and examined as to their compatibility or incompatibility with the linguistic point of view: the Wittgensteinian functionally colored theory of meaning and the referential theory of meaning (having as a direct opposite the Fregean distinction between Sinn and Bedeutung). In the fifth part of my discussion with Jacques Bouveresse, the area of linguistic semantics is further considered, and the relation of semantics to pragmatics is more especially thematized. It is, moreover, by introducing the notions of presupposition and of reference, and by elaborating a linguistics of speech acts, that the traditional domains of pragmatics and semantics are transcended. The methodological status of pragmatico-semantics is until now very unclear, and the specifications made from the philosophic side are meant as constructive criticism of the underlying epistemology of the different semantic trends which are currently productive. It is in this regard that the relation between natural and formal languages is discussed and that the heuristic role of the most recent logical systems for semantics is raised. In conclusion, the sixth part evokes a more philosophic problem: how do the linguist and the philosopher see the role of experience in speech and in language acquisition ? How should the connection of language to the (real and possible) world be thematized? How is the speaker present in the language fragment ? These fields of discussion are, of course, not exhaustive and the treatment of the points mentioned here is informal and suggestive rather than systematic and explicit. Nevertheless, we think that these topics are justifiedly in the center of interest for everybody who is currently preoccupied with language.

JACQUES BOUVERESSE

I. ON LINGUISTIC METHODOLOGY

Jacques Bouveresse, what is your total impression after reading these dialogues with linguists ? I myself have the impression of enormous diversity, not only with regard to the results, but especially with regard to the conceptual apparatus used by these different linguists. What is your feeling ? - As a philosopher, I am naturally impressed first by the abundance and the diversity of the results obtained and by the technicality of the means employed - to such an extent that I am not at all sure, in the end, whether my intervention in this debate can really be justified, and I ask myself if it is not totally presumptuous on my part and rather incongruous to agree to answer your questions. The first thing that appears from the conversations you have had with the linguists may actually be the impossibility of talking about language in 'philosophical' terms, I mean by that, in vague, general, and definitive terms. I could just as well say, of speaking about language as an amateur, the philosopher - at least as he is usually conceived in France - being in fact somebody who has made enlightened amateurism a sort of specialty, ranging in each case from perfectly honorable to perfectly despicable. It is clear that what one considered not long ago to be questions of the philosophy of language or of general philosophy are actually entangled more or less directly with questions in linguistics or at least with linguistic theory or methodology. That is at any rate what Chomsky and his disciples think they have demonstrated for a certain number of problems which the specialists used to consider with condescension as problems of speculative metaphysics. Correlatively, however, one must recognize the merit of transformational linguistics for having underlined that, behind what the professionals like to treat as questions of technique or of pure scientific methodology, in most cases philosophic options are probably concealed or even - let us not hesitate to use the word - fundamental metaphysics. If Chomsky has succeeded in irritating or even exasperating a certain number of people in recent years, it is evidently because he questions severely the habits of traditional and comfortable thinking: he shows the philosophers of the classical school that they ought to be much more competent in linguistics than they generally are, to the linguists that they are unconsciously much more philosophic than they pretend to be, and generally, to the specialists of the different scientific

302

JACQUES BOUVERESSE

disciplines that they are much more committed, from an ethical and political point of view, than they would like to believe. The second element which is quite evident in this series of dialogues is, in fact, the plurality of trends. One could think for a while that the imperial authority exercised by Chomsky and the linguists of his school would soon lead to dogmatism and scholasticism, as usually happens not only in philosophy but also in the other sciences. There is, without a doubt, a form of dogmatism which is a characteristic trait of the school, not only with the disciples and the adherents, but also with the founder himself. It is clear, however, at least if one judges according to the conversations I read, that the genuinely Chomskyan trend has now become only an important trend among the others. Linguistics is a discipline which does not stop developing and where events proceed very fast - certainly too fast for the philosopher. In particular, whereas Chomsky contributed considerably to reinforcing the feeling linguistics has had since Saussure of its field's autonomy and its specific features, one can notice today that the need to reattach it closely to other disciplines, such as logic, psychology, sociology, anthropology, is expressed everywhere. The diversities do not only manifest themselves with respect to the conceptual apparatus used by these different linguists but also with respect to what one proposes to be the proper domain of linguistics. In other words, the divergences affect, if you prefer, the proposals for a definition of what language is. The first task of the methodology and the epistemology of linguistics would thus be to define, on the basis of the mass of raw data to which reference is made in the different linguistic theories, what constitutes, properly speaking, the study object of linguistics. At first sight, an opposition which stands out immediately is that between structuralism and generativism. There are linguists who call themselves structuralists, others who are openly generativists - although Saumjan, for example, is actually situated in both, as he refers to himself as a 'dynamic or generative structuralist'. Do you believe in an opposition between structure and genesis, between structure and generation of language ? - First a general remark. You are completely right when you say that the most spectacular differences are not concerned with methodology, but with the object of linguistics itself. What the linguists do not agree upon is, first of all, the definition itself of what they are supposed to study: language. Martinet, for example, believes that one can reasonably apply the term 'language' only to systems which function as communicative tools, possess the double articulation, and are of a vocal nature. 1 Chomsky considers language primarily as an infinite potential to produce sentences, making an abstraction of the communicative function, 2 A language philosopher like Searle, from his viewpoint, sees in language a system which essentially makes possible the accomplishments of very diverse speech acts according to constitutive rules, with communication - understood in the narrow sense as an exchange of information 1 2

Cf. pp. 235-238 of this book. Cf. pp. 52-53 of this book.

JACQUES BOUVERESSE

303

being only one of these acts among many other possible ones. It is clear that, in a certain sense, Martinet, who considers the communicative function, in its large sense, an essential part of language, and Chomsky, who treats it as a rather accidental characteristic of the linguistic system, do not study the same object when they study language. So that one could say, on the one hand, that they disagree totally, but, on the other hand, that there is no real disagreement because, when they talk about language, they talk principally about the same empirical entity, but not about the same scientific object. Besides, this situation is common in the sciences, and it is actually a condition for progress rather than an insurmountable handicap. After all, in a certain sense, one still does not know very well what exactly is covered by mathematics, what is discussed by it, what its object is. What is dangerous from the philosophic point of view, is to pretend to have brought something to light as the 'essence' of language. Concerning the opposition you invoke between structuralism and generativism, it is evident that it has been considerably exaggerated, essentially for polemic reasons. In fact, the divergence essentially involves how one views the elements to be copresent within the structure. The cardinal problem generative linguistics has tried to solve is basically that of finding out whether the mechanisms of sentence production must be considered a part of language or not. For Saussure, they do not really pertain to what he calls langue (language) but to parole (speech). Ryle is quite close to Saussure in that respect: he considers language first and foremost as a treasure, a stock, or a reserve of words, constructions, intonations, etc. These elements are the atoms of language, whereas the sentences are the units of speech or of discourse. Sentences are the entities we construct in our linguistic activity on the basis of the elements of language that have been acquired by learning. They are not, like words, things we have, but things we make each moment. 3 On the contrary, the principle aim of a generative grammar is to integrate the sentence in language, to see to it that one would no longer be obliged to consider sentence production as an act of invention or creation which intervenes in a specific way in every instance of discourse. Whoever has language at his disposal actually possesses a creative potential defined once and for all by rigorous and systematic rules. Altogether, one can say, in a way, that the sentences are part of our linguistic capital: we possess them, virtually at least, by the simple fact that we possess language. With regard to the distinction between the structural point of view in the narrow sense and the generative point of view, it may be remarked that all linguists today are structuralists in a relatively trivial sense: they are all preoccupied with the study of 'structures'. The structural conception of language is simply that which has replaced, for everybody since Saussure, the atomistic conception. The opposition is thus not exactly between structuralism and generativism, but rather between different forms of structuralism, between a static conception and a dynamic conception of the linguistic 3

Cf. "Use, Usage and Meaning", Proceedings of the Aristotelian Society, supp. vol. 35 (1961); reproduced in Colin Lyas (ed.), Philosophy and Linguistics (London-Basingstoke, 1971), 54-60.

304

JACQUES BOUVERESSE

structure. This is evidently unrelated to the opposition expressed by synchronicdiachronic. The static and dynamic structures of Saumjan are synchronic structures. In the first case, however, the structure is essentially conceived as a network of relations between objects. In the second case, the fundamental notion is that of operation or application: the relation between linguistic objects is described as a process of the transformation of one element into another. But all these terms - genesis, dynamism, process, transformation, - must, of course, be understood in the mathematical or logical sense, and not at all in a psychological or temporal sense, even if one proposes, as Saumjan does, to call syntactic time and semantic time the new dimensions that intervene in the structural description of the applicative type. When it is said that a generative grammar 'produces' all the grammatical sentences of a language, one must think, for example, of what is meant when it is said that an axiom 'produces' theorems, or yet when it is said that a group is 'produced' by a generator or a system of generators. It is extremely regrettable that an entire philosophical mythology has come to graft itself to the concept of generativity, simply because people ignore or lose sight of the technical, logical, meaning of terms such as 'generate' or 'enumerate'. There are people who tend to consider grammar as a sort of occult power, gifted with an infinite capacity to produce, as a sort of autonomous creative power which really produces sentences. It is true that Chomsky himself has sometimes encouraged this mistake - which he himself certainly does not commit. It is possible, in fact, to define the notion of productivity of a generative grammar in terms that are not proper to linguistic theory but are part of the general theory of formal systems. In short, it is a concept which belongs to the mathematico-logical sciences, and there is absolutely nothing mysterious about it. It is true that the notion of generation and production has an ideological connotation. One finds in Chomsky the reference to Humboldt... - Yes, for example when he says that 'generate', in the sense in which he uses it, is an appropriate equivalent for the Humboldtian term erzeugen. Chomsky is certainly a little responsible for the confusion that has sometimes arisen, implicitly or explicitly, between two very different concepts of creativity or productivity, a rigorous logicomathematical concept, and an organistic and romantic concept which is totally unrelated to the first. Humboldt insists on the fact that the linguistic subject, in a way, creates and recreates the language while using it, transforming it constantly. That has nothing to do with the 'rule-governed creativity' which, in Chomsky, belongs to the theory of competence and not to the theory of language use, which presupposes a clearly established distinction between synchrony and diachrony and which refers necessarily to a given state of language. As to Humboldt's form of language, it seems that this is essentially linked to the fact that every language possesses a characteristic individuality, determines a certain correspondent view of the world and, on the other hand, as an organism gifted with a specific constitution, obeys certain internal laws of maturation and transformation which condition the historical evolution of lan-

JACQUES BOUVERESSE

305

guages. This 'form of language' has, therefore, nothing in common with the generative grammar of a language. Granted that all contemporary linguistics is structural, in the sense you just mentioned, how should one then characterize the static and dynamic approaches of the linguistic phenomenon ? - If Saussurean structuralism may be qualified as 'static', it is essentially with reference to this apparent assumption that language is deposed in totality, like a sort of gigantic interior treasure, in the minds of the speaking subjects. In fact, the language, which constitutes the study object of linguistics, is, properly speaking, no one's property, according to Saussure: it is essentially social and independent of the individual person. For Saussure, language is a virtually abstract entity which must be considered to be present in some form in its totality in the background of all the individual applications of language, of all the instances of speech. Chomsky, on the contrary, has tried to give a precise idea of what it might mean for a speaker to 'possess' or 'master' a language, and the essential element, according to him, in the mastery of a language is the fact that the speaker has at his disposal a system of rules, rules which can in principle be explicitly formulated, even if the speaker is usually unable to formulate them explicitly, and which, because they are recursive, give an account of the speaker's ability to make infinite usage of finite means. The recursiveness is exactly the formal device which makes possible the representation of the potentially infinite nature of language, and at the same time the preservation of the finite characteristic which the mastery of language must necessarily have to make its acquisition by linguistic subjects possible. In fact, one could say indifferently that the non-generative point of view does not account for what Chomsky calls the infinity of language or that it must necessarily postulate a kind of infinite linguistic competence, an infinite capacity for linguistic invention and innovation. This is exactly what can be avoided by the notion of 'rule-governed creativity'. Knowing a language is not knowing in advance an infinite set of sentences in one way or another: it implies a disposal of the ability to construct them according to rules determined in advance. The 'dynamism' to which you allude is essentially contained in the notion of recursive processes of production. There have been people, though, who thought they could refer to dynamism in a much stronger sense, as if generative grammar were a theory of the actual production of the sentences in a discourse - which it is not at all, as you know. Chomsky has said and repeated it. I do not know whether I have expressed myself clearly with regard to Saussure. I wanted to say that what may appear to be static in his conception is essentially the idea of a depot or a treasure, of a kind of reserve, out of which one 'draws'pre-existing combinations in a certain way - although Saussure rather believes that one 'invents' them each time in a certain manner. Another opposition that seems to be clear in the dialogues is that there are linguists

306

JACQUES BOUVERESSE

who aré oriented psycholinguistically and others who are sociolinguistically oriented. One can find the double tendency in Saussure. He considers language as a sort of interior treasure, that is to say, as a certain property of the mind, and on the other hand, he also considers language as the founder of society. This double orientation is not present in Chomsky. Did you have the same reflection when reading these texts ? - Quite, yes. I believe it can be explained very simply though. The attitude one has in that respect depends above all on which function of language one has decided to give priority. That Chomsky is rather psycholinguistically oriented comes simply from the fact that he attributes a central position to the function one might call ideational: that is, the function of free expression of thoughts. Inversely, the fact that other linguists are sociolinguistically oriented derives from the fact that they attribute an essential importance to the interpersonal function of language, to language as social behavior. In the extreme, the first will say that linguistics is a branch of psychology, the others that it is rather a branch of sociology, anthropology, or of the theory of social institutions. Chomsky believes that language reflects in a privileged way the essential properties of the human mind. There are people who would prefer to say that it reflects and determines at the same time something similar to what Wittgenstein calls 'forms of life', collective modes of existence. Wittgenstein is rather clearly situated in the second category. He does not think one gains much if one lets psychological considerations systematically interfere with descriptions of language and of language use. He insists, on the contrary, enormously on language as a social institution and on language use as intersubjective behavior. Once more, what is strange from the philosophic point of view is to pretend that the 'essence' of language must be looked for in one camp or the other. Michael Halliday says the same: that there is no incompatibility whatsoever.1 There are also linguists who emphasize strongly the autonomy of linguistics with regard to psychology and to sociology. I think, for example, of Martinet and also a little of Chomsky, because, even if the latter is psycholinguistically oriented, it does not mean that he considers himself a psycholinguist. - Yes, absolutely. Chomsky is not a psycholinguist because he is not a theoretician of performance. He does pretend, however, to practice in a way theoretical psychology, rational or rationalist psychology. Note that, in order to say more about this question, one would have to know exactly what Chomsky and certain of his disciples mean when they say that language serves especially to express ideas which are in the mind. It is an extremely traditional and common conception, of which Locke may be considered one of the exemplary representatives - although Chomsky might think in another way about Locke. In The Philosophy of Language,5 Katz has provided a particularly explicit formulation of this relatively ingenuous theory: linguistic communication consists of the production of an external acoustic phenomenon, observable in public, 4 5

Cf. p. 85 of this book. J. J. Katz, The Philosophy of Language (New York, 1966), p. 98.

JACQUES BOUVERESSE

307

which 'encodes' a speaker's private internal representations and of which the decoding by other speakers results in a private internal experience of the same representations or of nearly identical representations. On the other hand, Katz has also made a totally different suggestion. In his Linguistic Philosophy,6 he criticizes the tendency that we have to suppose that the meaning of a word or an expression is a psychological entity corresponding to it, a mental image, an idea, or a thought. He makes reference to Frege, and he stresses that the objective semantic contents of expressions are ideal entities which have the status of Gedanken and, more generally, of Fregean Sinne, and which are very different from the subjective mental representations. Katz has visibly felt the need to react against semantic psychologism: he wants to de-psychologize linguistic semantics, just as Frege and Husserl tried to de-psychologize logic. But that does not prevent transformational grammar from preserving an essential link with psychology because what distinguishes it from the so-called 'taxonomic-behavioristic linguistics' is mainly the fact that it formulates hypotheses about the psychological mechanisms underlying language use. It is true that the 'underlying mental structure' of which it speaks is not mental in the traditional meaning of the term, because the transformationalists do not make any hypothesis about the manner in which this structure is realized and admit that we might in the course of time be able to describe it totally in neurophysiological terms. In Chomsky's language, the term 'mental' is not opposed, as is usually the case, to 'physical'; it is rather opposed, one could say, to 'superficial' or 'observable', and it is used in a very broad sense to characterize the kind of reality and the manner of existence of inobservables, such as deep structures, which one has to postulate in linguistics in order to explain the surface phenomena. In physics, the inobservable theoretical entities have a presumed physical existence; in linguistics, one will say that they have a 'mental' reality. It is not excluded, however, that it will, in the end, also be physical in the usual meaning of the term. Could it be said that the linguists who strongly emphasize the autonomy of linguistics have an 'abstract' and reducing conception of the object of linguistics, language? - I do not know whether here you attribute to the word 'abstract' the pejorative connotation it sometimes has in methodological discussions. There are people who accuse Chomsky, for example, of never talking about language as it is, but only about a certain 'abstract' representation which he has constructed on the basis of the real language. All linguists who try to work scientifically have, of necessity, an abstract point of view. No science has a natural and totally independent object. What Saussure says about the study of language, namely that "it is the point of view which creates the object", is, in fact, always true to a certain degree in any empirical science. That is, scientists always proceed basically in the same way: they roughly try to ask only questions they can solve. Consequently, they procure objects in relation to which the 6

J. J. Katz, The Underlying Reality of Language and its Philosophical Import (New York-Evanston, 1971), pp. 121-122.

308

JACQUES BOUVERESSE

means to study them really exist. Where one does not have any means to attack the problem, there is no real problem in science either. Think of the way in which things happened in semantics : today, there are still many people who contest that it could be the object of a science, a research object. I believe that the affirmation of the autonomy of linguistics was, in a certain era, simply the indispensable means of attaining a scientific linguistics, of studying scientifically certain well-defined aspects of language phenomena. I can see that, today, one thinks the moment to insist a little less on the autonomy and a little more on the dependence may have arrived, probably because one has the feeling that certain directions of research are close to exhaustion and one feels the need to try others. This leads us to another important point, one which is very crucial for the philosopher who is preoccupied with linguistics - namely, the status of linguistics as a science. As you know, linguistics has constituted itself as the pilot science of the human sciences. Do you think there is a certain value in looking at things this way ? - I do not know if one can say that linguistics has constituted itself as pilot science of the human sciences. It might be better to say that this role has been imposed on it a little in spite of it, because, with a few exceptions such as Jakobson, linguists have not looked very favorably on this matter. One must, of course, rejoice at the apparently spectacular progress the influence of linguistics has entailed in disciplines such as anthropology, psychoanalysis, or literary criticism. Except in some very specific cases, however (Lévi-Strauss, for example), it is difficult to consider the result positive in all respects. As a philosopher, I am, unfortunately, impressed especially by the extreme confusion which resulted from it. If one believed one could impose the concepts and methods of linguistics on such diverse fields, it is evidently because one considered that any system of signs whatsoever could be submitted, in the end, to a linguistic approach. There is, today, a characteristic and disastrous trend towards applying terms like 'language' or 'logic' to almost everything, to everything that somehow resembles a system of signs. It might not be difficult to understand the reason why. The linguistic model represents, in a certain sense, the ideal and universal type of intelligibility : it seems that one never understands a totality of phenomena so well as when one can treat it as a sort of language having its own logic. Venturing to use a bad play on words, I would say that one only understands fully that which 'speaks', which 'says' something - in short, that which behaves as a language. One has the feeling that one has really understood the things in depth if one has 'discovered' that the unconscious, the rules of marriage, or the genetic code are types of languages. Wittgenstein would say that we are simply impressed here by a pregnant analogy and that we are seduced by the prestige of a universal manner of representation. One must, as he stresses, avoid attributing to the thing itself what resides in the manner of representation as much as possible, and one must not consider the possibility of using a very general expression as a sufficient index to the existence of a very general state of things.

JACQUES BOUVERESSE

309

The majority of linguists whose texts we have here consider linguistics to be a hypothetico-deductive science. Do you have any comments on this? - Yes, I am totally convinced that linguistics must be a hypothetico-deductive science, but this is in a certain sense true of all the empirical sciences. They are hypotheticodeductive, not in the sense of a mathematico-logical science, but in the sense of Popper, meaning that they proceed by invention of conjectures from which one deduces consequences which can be tested empirically. In that respect, I think it should be remarked that the real opposition is probably not, as Chomsky affirms, between rationalism and empiricism, but rather between deductivism (in Popper's sense) and inductivism. Popper, for example, considers himself both a rationalist and an empiricist. The partisans of inductive logic can easily give the impression that, according to them, the least strong hypothesis must always be chosen - that which is closest to the facts, the most immediately 'suggested' by them. Nevertheless, they are not so naive as to believe that it is possible to formulate a set of rules, determined once and for all, which would make it possible to proceed automatically in any field from facts to hypotheses or theories. They do not believe in the existence of a kind ot 'inductive machinery' and do not at all ignore the irreducible specificity of the process which leads to the production of theories.7 A deductivist epistemologist like Popper, however, prefers to admit resolutely that the facts do not suggest anything by themselves, that we have to interrogate them and that we should not hesitate to ask them the boldest and most precise questions, those that, paradoxically, are most susceptible of receiving a negative answer in the end. In this respect, there is a very interesting discussion in the conversation you have had with Chomsky.8 He observes that, if we have a choice between two linguistic theories, T and T', equally adequate trom a descriptive point of view, but of which one is more simple and less restrictive than the other, we have to choose that which is more restrictive and more easily falsifiable. If T and T' both satisfy the condition of descriptive adequacy and if T restricts the range of possible grammars more than T', we have to prefer T over T', even if it is less simple. This is entirely in the sense of one of Popper's essential ideas: the theory with the richest content (thus the smallest logical probability) must always be chosen, that which is most easily testable; that means, for Popper, that theory which is liable to the most severe attempts of refutation. It may be remarked, in this respect, that the reason why Chomsky vigorously rejects the theory which he calls "resourceful empiricism" is that it does not have, according to him, any content - that is, it has no testable consequences because it is compatible with everything one could discover about language or about anything else, in short, because it is not an empirical hypothesis.9 In Peirce, the logic of abduction is the logic of the invention of hypotheses. It is a 7 Cf. for example R. Carnap, Philosophical Foundations of Physics (New York-London, 1966), pp. 33-34. 8 Cf. pp. 47-49 of this book. 9 Cf. Sidney Hook (ed.), Language and Philosophy, A Symposium (New York, 1969), pp. 158-159.

310

JACQUES BOUVERESSE

specific type of logic which is distinguished from both deduction and induction. There is an important idea for Chomsky in the Peircean conception of abduction: it is the idea that the human mind must possess a sort of natural, constitutive, aptitude for the invention of correct hypotheses and theories. Without this, it would never be able to construct an acceptable explanation of the external world. Very strict restrictions must weigh from the beginning on what may be accepted as an admissable hypothesis or theory. According to Chomsky, this is what happens in the acquisition of language-, the child learning a language constructs implicitly a certain 'theory' of its mother tongue on the basis of a very limited and imperfect experience with it, and an operation of this type is only conceivable because what is at stake is not at all the invention of an adequate grammar on the basis of very reduced and rather degenerated empirical evidence, but simply the selection of the 'best' grammar among a limited number of admissible grammars, having the form prescribed by an innate scheme of a possible grammar in general, and constituting possible grammars in the actual circumstance, i.e., being compatible with the facts observed. This presupposes the existence of an appropriate evaluation procedure, which is itself a part of the innate linguistic ability of human beings. It is certainly not possible to consider that the human mind somehow formulates innumerable conjectures and predictions spontaneously, haphazardly, and in all directions. Goodman recognizes that the traditional empiricist conception does not account for the distinction that must exist between the regularities that activate the mind and those that do not, but the inductive logic as he conceives it does not do it any better: he does not try to explain how the mind produces predictions but to describe how it distinguishes between valid and invalid projections.10 What activates the linguist's mind? - I think that, as in any other empirical science, the observable regularities set the linguist's mind in action, under the circumstances, linguistic regularities. This remains true in a certain way even if one admits that the 'observable' facts in a science are always facts interpreted in the light of theories. Do they pick up these regularities in a corpus, or... - It would be absurd to say that the task of linguistics, when it is applied to a particular language, is simply to describe and analyze a corpus. Linguistics does not describe corpuses, but languages. As far as I know, however, the corpus plays quite a fundamental role: after all, the linguists have nothing else at their disposal but a written corpus in the case of dead languages. The impression that the importance of the corpus is considerably reduced in transformational grammar is essentially caused by the fact that one is apparently less interested in the actual facts of language than in the knowledge the speaker has of his language - in his ideas about language. The observable facts 10

Cf. Fact, Fiction and Forecast, 2nd edition (Indianapolis, 1965), p. 87.

JACQUES BOUVERESSE

311

are less linguistic facts in the usual sense than the linguistic intuitions of the speaker. Chomsky is certainly right in stressing that, taking into account that all languages contain a potentially infinite number of sentences, a corpus of whatever size always amounts to something ridiculously small. On the other hand, one should not forget that the length and complexity of the sentences about which one can effectively test the linguistic reactions of the speakers aie rather tightly limited, so that what is important for the constitution and the confirmation of the theories is what I would call the 'possible corpus', the set of sentences that are or could be observed concretely in a language. I mean that a 'grammatical' sentence is not only a sentence which is recognized as such by a linguistic subject when he is asked the question, but also (in the case where semantic anomalies are not involved) a sentence which is or could be observed in principle in normal linguistic behavior. We know that normally a Greek author considers a sentence grammatical because he writes it... One of the main criticisms which has been addressed to the structuralists by the transformationalists was that classical structuralism was inductive linguistics. Did you understand while you were reading the conversations with the structuralist linguists that their epistemology was an inductive one? - As I said a moment ago, everybody proceeds deductively in a way, even if the interested parties themselves do not always realize it and try to give the impression that they do something else. There are people who think that linguistics, as all empirical sciences, must proceed by inductive generalization on the basis of independent data observed and described in advance, but the question is to know whether this has ever been done in any science, whether the inductive course as it has been conceived traditionally plays any role whatsoever in scientific discovery. The deductive approach consists for a great part in realizing that for certain theoretical presuppositions determine the selection and the description of the facts and that they must be recognized and formulated as much as possible at the outset. I do not know, however, if there are many contemporary empiricists who are willing to repudiate seriously such an elementary doctrine. Martinet might give the impression of having a typically inductivist attitude when he declares: "When dealing with language, an object we know quite well, since we approach it in a rather direct way, there is no need to hypothesize...." 11 But, on the other hand, words such as 'empiricist' or 'inductivist' are used with a certain malevolence and bad faith by transformationalists. They serve to designate a very general attitude which is characterized by rather diverse elements: almost exclusive interest for surface phenomena, priority given to classificatory procedures, concern for more constant and stricter verification, emphasis placed on the corpus, distrust of strong hypotheses in general, reservation with respect to formalization, etc. These elements are not necessarily present all at the same time nor are they all totally negative. It is regrettable that, in the eyes of 11

Cf. p. 235 of this book.

312

JACQUES BOUVERESSE

some, 'inductive approach' means almost the same as 'anti-theoretical approach'. It might be true for certain American structuralists. Anyway, it would be very absurd to maintain that structuralists such as Saussure, Hjelmslev, or Greimas approach linguistics 'inductively', in the rather pejorative sense of the word. Concerning Martinet, as he recalls himself, he was totally opposed to the hyper-empiricist and hyper-inductive methods of the Bloomfieldean structuralists. It is inadmissable in the controversy to try to deconsider an entire trend by constituting a caricaturized unitary doctrine on the basis of the most disputable elements one finds in the work of some representatives. What does it mean for you to explain a linguistic fact - that is, what is the status of linguistic explanation ? - That is a very delicate and important question. It could be approached by asking what Chomsky wants to say when he introduces generative grammar as an explanation of the speaker's grammatical competence. He quite often compares the explanatory power of a generative grammar with that of other scientific theories - physical theories, for example - and he seems to consider more or less that the predictions a grammatical theory of the considered language can make are not very different from the predictions which are presented in an astronomic theory, the theory of Newton, for example. Well, I think that this comparison is extremely misleading for the following reason: what would correspond, on the predictive level, to what happens in an astronomic theory would rather be a prediction of linguistic behavior, in my opinion, a kind of theory of linguistic performance. Just as the astronomer, taking into account a certain set of observational facts and certain laws (the Newtonian laws, for example), predicts that the position of a planet will be such and such at such a moment, one can envisage that some day it will be possible to predict that the linguistic performance of a given speaker will be such and such at a given moment and under given conditions. Of course, it must be specified immediately that this prediction would doubtlessly be only probabilistic, in the best case. One can imagine that the linguistic and anthropological sciences would finally be able to predict the speaker's linguistic behavior with some probability, taking into account a number of very disparate elements: his knowledge of the language, his individual history, in particular the history of his language acquisition, a description of the culture in which he lives, the linguistic and extra-linguistic context, etc. It goes without saying - this is a point on which Chomsky insists strongly - that this perspective is totally Utopian and rather absurd under present conditions. Still, one should contemplate considering it at a later date, because, after all, Chomsky himself does not exclude the possibility that the explanation of linguistic behavior could in the end be an integral physical and causal explanation, totally analogous to those given in the natural sciences. There is also a very strange remark in this respect in Language and Mind.12 Chomsky observes 12

N. Chomsky, Language and Mind (New York, 1968), p. 84.

JACQUES BOUVERESSE

313

that the explanation of the phenomena most characteristic for the human mind, when it has been found, will necessarily be a physical explanation because the meaning of the word 'physical' will have expanded appropriately. It is certainly right to say that all physical explanations of a new type modify our concept of what may be accepted as a physical explanation in some way. It is obvious, however, that this concept cannot be expanded in any way or direction whatsoever. One will certainly never consider as 'physical' a Cartesian explanation in the narrow sense of the word, an explanation which would allow the intervention of such things as ideas pertaining to a substance called soul or mind. One should rather say that when one has found a physical explanation in a scientifically acceptable meaning of the word 'physical', one could then consider that a real explanation had been found. Returning to our problem, the comparison with physical theory is misleading because the explanation in question in the case of generative grammar is rather an explanation in the Carnapian sense of the term 'explication'. It consists primarily of replacing a rather vague intuitive concept (explicandum) by a precise formal concept (explication). What is actually provided is a recursive and rigorously formal characterization of the set of grammatical sentences of a language. A scientific concept of grammaticality is substituted for a naive concept. As Black remarks, 13 one tries to make correspond, at the price of considerable idealization, a set of perfectly explicit and mechanizable rules to a set of embryonic, semi-explicit, and more or less (rather more than less) unconscious rules: "The verification route is from rules to rules, not from rules to hypothetical and idealized performance." Concerning the explanation in the strong sense, the explanation of what happens in the language at the performance level, I believe that one only obtains a purely negative type of prediction, because generative grammar clearly does not predict that such and such a sentence will be observed in the language. It rather predicts things of the following kind: such and such a nongrammatical sentence (in the strong sense) will not be observed in normal linguistic behavior. There is, of course, no objection in principle to the idea that in linguistics one must attempt to predict what normal language users are inclined to say about the phonological, syntactic, and semantic properties of the sentences in their language. Katz, for example, writes: "The intuitive judgments of fluent speakers constitute the empirical phenomena to be predicted by linguistic descriptions."14 One should, however, not lose sight of the fact that it is a very limited and rather special objective. As a very precise answer to your question, one could say that Chomskyan linguistics tries to explain essentially what Searle calls linguistic characterizations,15 At stake is the explanation of what language speakers want to say when they say that such and such a word combination is meaningless, that such and such a sentence is analytic, that two 13

Cf. R. Borger and F. Cioffi (eds.), Explanation in Behavioural Sciences (Cambridge, 1970), p. 454. J. J. Katz, The Underlying Reality of Language and Its Philosophical Import (New York-Evanston, 1971), p. 23. 15 J. Searle, Speech Acts, An Essay in the Philosophy of Language (Cambridge, 1969), pp. 4-11. 14

314

JACQUES BOUVERESSE

given sentences are synonymous, etc. And the linguistic explanation has been obtained when one has been able to bring in evidence systems of underlying rules which make possible and govern such characterizations. The most serious problem in this business is probably that of the existence itself of what Carnap calls the explicandum. There are people like Martinet who dispute that there would really be anything to explain in that field. They do not think it serious to consider that there exists a system of concordant linguistic judgments, a common linguistic competence.16 In short, according to them there is not the least empirical evidence susceptible to inspiring and justifying the constitution of a theory.... As for yourself, you seem to believe that there is an important difference between scientific explanation, that is, the explanation linguists provide when they study language, and the speaker's knowledge of his language, between the language theory of the linguist and that of the normal language user. - There is clearly a spectacular difference between an explicit knowledge formulated in rules, and an informal, 'tacit', and almost totally unconscious knowledge. In the mind of Chomsky and some of his disciples, however, the first is supposed to constitute a formal representation of the second. The difficulty here is to grasp what is meant when it is said that a generative grammar gives us an idea of the speaker's knowledge or mastery of the language. The Chomskyan concept of competence is actually ideal in two respects: it is ideal in the sense that what is called linguistic competence corresponds to a perfect knowledge of the language, which is actually non-existent; but it is also especially ideal in the sense that one has practically no idea of how this competence, which is only for a very small part constituted of the conscious possession of linguistic rules, is realized from a material point of view and is put to work in the production of discourse instances. The interest of a notion of competence is rather thin if one cannot indicate any research direction which would eventually be able to throw some light on the way in which this competence is materialized and put into practice. It is true that the transformationalists have repeatedly said that, as linguists, they did not have to give such indications and that the description they propose of the mechanisms underlying language use is a priori compatible with very different manners in which these mechanisms could be realized. It is precisely in these conditions, however, that it is hard to understand the insistence with which they maintain that they have provided the description of something which really deserves to be called the ideal speaker's competence. I have observed that, in fact, many well-intentioned people who accept the genuine linguistic methods and results of the Chomskyan school without reservation consider rather suspect the idea that a generative grammar constitutes a model for a speaker's grammatical competence, and entirely nonessential in the actual state of affairs. After all, one would probably not say that a formal axiomatic yielding as 11

Cf. p. 241 of this book.

JACQUES BOUVERESSE

315

theorems the whole set of the true arithmetic propositions (we know in fact since Godel that such an axiomatic cannot exist) would represent the arithmetic 'competence' of a mathematical subject. The difficulty in Chomsky's case is that he wants to accomplish two jobs at a time, which task is perhaps impossible: the first consists in describing a language as a potentially infinite set of well-formed grammatical sentences; the other consists in describing the speaker's knowledge of that language. Some people think that his real contribution concerns exclusively the first of the two tasks in question, that he has only revolutionized the descriptive methods of language. At least in the beginning, it was always believed that there was an isomorphism or even a certain identity between the explicit grammar of the linguist and the implicit and unconscious grammar of the speaker. Do you believe that there are more misunderstandings to be deplored in this respect than solutions obtained or to be hoped for? - I wish someone would tell me what the correspondence rules are that govern this isomorphism - what the transformation rules are which will make it possible one day to 'translate', as one calls it, the linguist's description into the description of a genuine linguistic mechanism. No independent characterization whatsoever, even a very approximate one, of the implicit and unconscious grammar to which you refer is indeed available; one does not even know whether the hypothetical mechanisms in question are of a 'mental' type in the traditional sense or whether they are, on the contrary, just mechanisms in the central nervous system. True, it is not really for a philosopher to discuss such a question. It would be rather for the psycholinguists and neurolinguists to determine whether the description of linguistic competence provided by transformational linguists provides a direction of effective research for the exploration of what I would call the actual competence, which materializes in concrete mechanisms which one can describe or hope to describe. Chomsky has suggested that this could be the case, indeed - that transformational grammar could actually contribute to guiding psycholinguistic and neurolinguistic research in the right direction. I then ask myself the question: what would happen if the theory of linguistic competence were to appear to be invalidated by the results obtained in the psycholinguistic field, if, for example, the assumed psychological reality of the transformations were to be questioned by certain experiments? I know there have been such experiments. Moreover, how could one even consider undertaking such experiments without implicitly admitting what the Chomskyan linguists dispute strongly in principle - namely, that generative grammar constitutes or can constitute a theory of the actual production of sentences? What would happen if one were finally obliged to admit that nothing in the psychological or physiological description of the mechanisms governing language use corresponds plausibly to the linguistic transformations? One could naturally feel that this does not at all diminish their interest and value for the linguistic description (this is my opinion), but, of course, if one wanted to keep saying that the formal model it constructs represents the

316

JACQUES BOUVERESSE

grammar internalized by the speaker, I would assume that there would be some problems.... What is, after all, the exact meaning of the word 'represent' in this respect ?

II. ON THE RATIONALISM VERSUS EMPIRICISM DEBATE

As a new 'mentalism', generative grammar is opposed at the same time to empiricism in philosophy, behaviorism in psychology, and structuralism in linguistics, and Chomsky identifies very easily, in a single constellation, empiricism, behaviorism, and structuralism, and structuralism with (keeping American structuralism especially in mind) Bloomfield and the people of his school. This presupposes a very rudimentary set of oppositions: mentalism versus, for example, empiricism. I would like to ask you - and we enter your fieldfully here - if you consider empiricism a unified doctrine, or are there, on the contrary, several empiricisms, and is it necessary to introduce precise distinctions in this field? - There are certainly several empiricisms. That of Quine, for example, is very different from that of Carnap. In fact, modern empiricism must be considered much more as a norm, a regulatory principle, than as a doctrine. In a way, all people working in the field of the empirical sciences are necessarily empiricists. Rather than talking about an opposition between rationalism and empiricism, it would probably be better, in certain respects, to talk about an opposition between different kinds of empiricism - empiricism like Popper's and Carnapian empiricism. In a way, Chomsky himself is a very strict empiricist. He constantly insists on the fact that the hypotheses he proposes are empirically testable hypotheses - which is, after all, the basic requisite of contemporary empiricism. The opposition between mentalism, as conceived by Chomsky, and empiricism is very disconcerting to a philosopher. I have already mentioned that problem before. The 'mentalist' terms in transformational grammar are merely theoretical terms which are not or are only partially interpreted; the 'mental' entities here are not, at least not necessarily, mental entities in the usual meaning of the word. The transformationalists have been careful to specify that their mentalism was not at all the kind fought so vigorously by Bloomfield. But the mentalism which aroused the suspicion of people like Carnap, Quine, and to some degree Wittgenstein is basically the same as that which was rejected by Bloomfield in the linguistic field. It concerns classical mentalism, which is expressed in terms of mental representation, of ideas in the mind in the psychological sense. Why are the philosophers I just mentioned hostile to this kind of mentalism? It is, at any rate, not only, as Chomsky seems to believe, because of theoretical pusillanimity. It is because they think that mentalistic terms, in their classical interpretation, have no or a very reduced explanatory power and that all they actually do is to reposit the same explanatory problem on another level. If the mentalist thesis is simply the thesis according to which one does not have

JACQUES BOUVERESSE

317

to be afraid, either in linguistics or in any other science, to adopt theoretical terms which are sufficiently abstract and removed from the immediate observation, I do not believe that Carnap or certainly Quine could actually find fault with it. Chomsky observes at some point in the conversation you had with him that there is a tendency in present discussions to interpret questions of terminology and notation abusively as a reflection of substantial disagreement.17 I wonder, however, whether there has been in recent times anyone who has contributed more than he to transforming differences of vocabulary into divergences of opinion on essential points and to creating or reinforcing Manichaean and simplistic oppositions by using a more or less adequate terminology. It is true that I discuss all this as a mere observer.... Carnap's and Quine's conception of language is basically dispositional. What exactly does that imply ? - Chomsky seems to assume that there is a strict distinction between a theoretical approach and a dispositional approach to language, but neither Quine nor Carnap view things that way. They do not make a decisive distinction between theoretical terms and dispositional terms. It is perfectly obvious in Quine's case. He considers a disposition as a sort of promissory note, which we hope we will be able to "redeem" eventually by providing an explicit characterization of the mechanism which comes into play. From this point of view, there is no basic difference between dispositional terms and theoretical constructs. The purely dispositional concepts do not have any explanatory value by themselves; they refer back to properties and mechanisms which are not yet identified, but which one hopes at some time will be, and which are supposed to account effectively for the observed facts. The difference between dispositional predicates and theoretical predicates for Carnap is schematically the following. The meaning of the dispositional terms is completely exhausted by the marking of a correlation of the stimulus-response type; that of the theoretical terms stays open, and we always have the possibility of reinforcing their interpretation by introducing supplementary postulates and new correspondence rules without, however, making it completely determined. Carnap specifies that we are free in many cases to choose the theoretical or the dispositional interpretation of a concept, and he observes that, in the case of abstract psychological concepts, the theoretical interpretation is generally more recommended. He thinks that, in the end, the theoretical terms of psychology might be interpreted by dint of theories of the central nervous system formulated in physiological terms.18 Judging from certain suggestions Chomsky makes, it appears to me that he considers things in a not so different way. What do you think about the Chomskyan interpretation of the Lockean doctrine - as he has always been vigorously opposed to Locke ? - That depends on what exactly one is talking about. Locke is a mentalistic empiricist 17

Cf. p. 47 of this book. Cf. R. Carnap, "The Methodological Character of Theoretical Concepts", Minnesota Studies in the Philosophy of Science 1 (1956), 38-76.

18

318

JACQUES BOUVERESSE

whereas Quine, for example, is a behavioristic empiricist. There is, as I have indicated, a totally Lockean element in Chomsky's theory of meaning: the idea that language serves essentially for encoding representations in the mind, for 'translating' thoughts into words. Moreover, this conception is common to Locke and Descartes, and it can be found in almost everyone, except precisely those people who have felt the need to combat mentalism. This conception is of the type Aristotle expresses when he declares that written words are symbols of spoken words and that spoken words are symbols of states of the soul. Chomsky has remained loyal to this way of looking at things, primarily because he wants to preserve the creative aspect of language use. His disagreement with Locke is primarily about the nature of the mechanisms governing the formation of ideas and about the problem of innateness. It seems to me that you do not consider it totally justified to make Descartes the prototype of a rationalist linguistic philosophy and Locke the prototype of an empiricist philosophy about language. - My reservations bear less on the choice of representatives than on the opposition itself. Actually, insofar as Locke and Descartes are concerned, the essential question is to what extent innateness must be considered one of the essential components of rationalism, but we will discuss that later, I suppose. Chomsky's reference to Descartes has always seemed rather strange to me. I have often wondered how Cartesianism can remain Cartesian after one has eliminated the essential element: dualism - the fact that man is constituted of two substances of a different nature, body and mind, one associated with the other and capable of acting on the other. For Descartes, there is really a spiritual substance which contains thoughts, representations, volitions, etc. I recall these banalities because Chomsky's philosophical eclecticism is really strange. He has the habit of setting apart in very different theories more or less isolated elements which interest him for one reason or another, without preoccupying himself excessively with the general philosophic context in which they appear. He is on the Cartesian side when the creative aspect of language use is stressed, but he does not exclude in principle that La Mettrie would finally be right, in other words that an integrally mechanistic explanation of linguistic behavior might some day be provided - exactly what Descartes considers completely impossible. Chomsky possesses a remarkable ability to reinterpret, more or less pertinently and exactly, traditional philosophical doctrines in the sense of his own theories. I am not sure, however, whether that adds much to what he does. Anyway, although I am a philosopher, I am personally much more interested in his contribution to linguistics proper than in what he says about the philosophers whose authority he invokes, or perhaps tries to restore. With regard to his 'Cartesianism', it is certainly so that Leibniz is in many respects a much more appropriate candidate than Descartes, because he thinks that what appears in our perceptions and knowledge at the conscious level is only the visible part of an iceberg, so to speak, of which the essential part remains submerged, and because he is often considered as one of the distant

JACQUES BOUVERESSE

319

precursors of the theory of the unconscious; Kant might also be a more adequate reference than Descartes in other respects. Yes. One finds in Chomsky dispersed allusions to Leibniz as well as to Kant. - You said it well: allusions. There is namely in Problems of Knowledge and Freedom19 an allusion to the famous remark of Kant according to which schematism is an art buried in probably inaccessible depths of the human mind. Chomsky means that the innate principles which make the acquisition of our system of knowledge and beliefs possible could, at the same time, impose such limitations on our ability of knowledge and comprehension, that we should probably not hope to obtain a scientific explanation of the way in which the knowledge and beliefs in question are acquired or used. But, let's say it in passing, since we are talking about Kant, of all the faculties of the mind, if there is one which has really interested Kant very little, it is certainly this very specific faculty which Chomsky calls the faculty of language.... I would like to ask you now to specify Quine''s position a little more, because one gets the impression that Quine is the most criticized empiricist philosopher, or at least the one Chomsky discusses most directly. How could one understand what is essential to Quine's empiricism? - I believe that he has explained himself quite clearly in this respect. Quine's empiricism is a behavioristic empiricism - that is, antimentalistic - but he is not a reductivist. His behaviorism consists essentially in his refusal to let a certain number of interconnected notions which he considers suspect interfere in order to explain linguistic behavior: those of idea in the mind, of meaning in the usual mentalistic sense of the term, of proposition, of propositional attitude, etc. "Semantics", he writes, "is vitiated by a pernicious mentalism as long as we regard a man's semantics as somehow determinate in his mind beyond what might be in his dispositions to overt behavior. It is the very facts about meaning, not the entities meant, that must be construed in terms of behavior." 20 Quine is hostile to the idea of occult internal states and processes: it must be possible to relate all the states and processes concerned in a theory (at least in principle) to observable characteristics; this relation may naturally be very complex and indirect. Chomsky says somewhere, I believe in a note in Aspects of the Theory of Syntax, that behaviorism is primarily a lack of interest in theory and explanation. This remark cannot at all be applied to Quine's behaviorism, however, because, even though his style may be quite 'Occamean', Quine does not show any timidity with regard to introducing highly abstract theoretical terms. His suspicion bears once more on a certain category of terms which pretend to be theoretical but which, according to him, have actually no explanatory power from a scientific point of view. One does 19 20

N . Chomsky, Problems of Knowledge and Freedom (New York, 1971), p. 17. "Ontological Relativity", in Ontological Relativity and other Essays (New York, 1969), p. 27.

320

JACQUES BOUVERESSE

not at all find in his work the kind of radical reductionism, the systematic refusal to admit irreducible theoretical entities, which Chomsky has sometimes presented as the characteristic trait of behaviorism. One can, in the end, have no sympathy whatsoever for Quine's conceptions, but one should still not simplify them outrageously. There is, in fact, an enormous difference between Quine's behaviorism and a Skinnerian type of behaviorism. One has the impression that Chomsky always conceives of behaviorism in the image of Skinner. The fact that Quine's behaviorism only remotely resembles what is usually called 'behaviorism' may be interpreted, if you want, with the meaning that Quine is not really a behaviorist but also as proof of the fact that behaviorism may not be so absurd and scandalous as is generally believed. Quine is certainly not, as the radical behaviorists, a superstitious worshipper of an explication provided in terms of correlations of the stimulus-response type. Look, for example, at what he said in his contribution to the symposium on linguistic rationalism and innateness: "It may well turn out that processes are involved that are very unlike the classical processes of reinforcement and extinction of responses. This would be no refutation of behaviorism, in a philosophically significant sense of the term; for I see no interest in restricting the term 'behaviorism' to a specific psychological schematism of conditioned response."21 For Quine„ observable facts are needed in the construction of a theory. What could these observable facts mean for a linguist? - Would it be something basically different from what is called observable facts in any scientific research? They are facts obtained partially by immediate observation and for another (essential) part by more or less technical experimentation. It is true that the theoretical point of view adopted determines for an essential part what the observable facts are. Just as the linguists do not agree on what constitutes the object of linguistics, they also disagree on what are the basic observational data of linguistic theory. For Chomsky, for example, "intuitive and introspective judgements are the primary data for the descriptive grammarian, hence also the linguistic theorist". 22 But this point of view is not at all shared by everybody. I do not grasp the import of your question too well.... I mean: What is at stake is to know what one observes in language. What one observes are surface structures, and one of the original impacts of Chomskyan linguistics is exactly the statement that the central part of language is not its surface structure but its deep structure and that there is a whole system of transformations which connects the two. What, then, in the Quinean framework, do you make of his notion of'deep structure'' which is not observable... - Oh, I see! I do not believe there is any real difficulty. The only real question is 21 22

In S. Hook (ed.) Language and Philosophy, p. 96. Cf. p. 40 of this book.

JACQUES BOUVERESSE

321

whether these deep structures are actually necessary for a correct linguistic description of the facts observed. A serious scientific mind would not refuse to acknowledge them because they are inobservable. I do not know what Quine's attitude is exactly with respect to the deep structure, but he does not have any reason to be hostile to it out of principle nor to challenge a priori that the linguistic deep structures should have a mental reality, if one interprets the term 'mental' in the sense which is acceptable to him. What happens in transformational grammar is totally consistent with what happens in any abstract theoretical construction, where one postulates inobservable or not directly observable entities in order to explain observable processes. I do not think that this procedure could raise objections in principle by itself. Katz talks about 'Democritean' linguistics, which consists of postulating underlying realities, totally removed from the immediate appearance, in order to account for what one observes, in opposition to a form of linguistics which is satisfied with registering, segmenting, and classifying surface structure phenomena. 23 If I did not have the impression that there has actually never been a linguistics of purely the second type, I would say that, with the appearance of the generative transformational grammar, linguistics had simply aligned itself to the methods of the other sciences. We have seen that the generative linguists declare themselves to be mentalists, but if one observes the whole doctrine from a distance, do you believe one could identify generativism with mentalism ? - A French philosopher is in a poor position to discuss this because it is obvious that the need to present themselves as mentalists must have been conceived bv the transformationalists primarily as a vigorous reaction against behaviorism - the dominant trend in a certain era; it seems that the weight of behaviorism on American linguistics has finally become totally intolerable. But what exactly do those linguists mean who declare themselves 'mentalists'? They are certainly not mentalists in the sense that they would allow the intervention of an irreducible non-physical factor, a 'mind' or a 'soul', in the genuine Cartesian sense of both words, in human behavior. Their conception of mentalism is, as Katz asserts,24 causal; it is, in principle, compatible with a totally mechanistic theory. That seems to imply that the underlying mental processes of which they speak are actually mere physical or physiological processes which are unknown or hardly known. Their disagreement with the behaviorists does, therefore, not bear on the existence or non-existence of spiritual, mental phenomena. As far as I can tell, what characterizes mentalism in opposition to behaviorism under these conditions is the decision to form systematically sufficiently rich hypotheses on the nature of the internal mechanisms, states and processes which play a role in the knowledge and use of language. A strict behaviorist, on the other hand, would be satisfied in principle with the establishment of functional correlations between observable data. It is obvious that the term 'mentalist' is used here is a rather peculiar sense. 23 24

J. J. Katz, The Underlying Reality of Language, pp. 1-4. J. J. Katz, "Mentalism in Linguistics", Language 40 (1964), 124-137.

322

JACQUES BOUVERESSE

Yes, indeed. But in the discussions of Chomskyan linguists, the argument for empirical evidence always reappears, also allowing us to say, it seems to me, that their position is intermediary between mentalism and empiricism. For them, too, the theory must correspond to the observable facts. - Oh, yes! Absolutely. They think they have demonstrated that a linguistics which would not be mentalistic in their sense would not allow accounting for the observable facts, in the very empirical sense of the word 'observable', in the most satisfactory way. There is not doubt that they are empirical researchers of a very precise sort. Chomsky has sometimes even been accused of being too empirical.25 Note that anyone is always too empirical for some and not enough for others, but it is clear that Chomsky, compared to Saumjan for example, may seem very empirical in certain respects! The Chomskyan linguists insist constantly on the tact that there is nothing metaphysical in their theories, that they are rigorously scientific theories, that is, that they must be open, in principle, if not to verification, at least to falsification by empirical evidence. Let us take up another notion which is also very much discussed and which is central to the linguistic theory of Chomsky and his successors - namely, the notion of linguistic creativity. You have probably noticed, in the conversation with Martinet, for example,26 that this notion has been severely criticized. Nevertheless, Chomsky remains very serene and continues to use it quietly. What is meant, in your opinion, when it is said that there is a nucleus of creativity in language ? Is this also a mathematical concept ? - That depends, of course, on which creativity one is talking about. In the dialogue you had with him Chomsky deplores the fact that two very different kinds of creativity have been sometimes confused - one which we have called 'mathematical' or 'formal' (the recursive property of grammars), and the other which he calls the "creativity of language use". I believe, however, that he himself has not always expressed himself very clearly in this respect, at least in his philosophical and polemic texts. It is true that two very different notions are at stake. A relatively simple automaton which has initial elements and recursive formation processes at its disposal may very well be gifted with the first type of 'creativity'. It is possible that, although its productions are potentially infinite, they are totally trivial.27 Moreover, it must be remarked that we will probably have little inclination to say that a machine which constructs an unlimited number of sentences by obeying recursive instructions is gifted with a creative power, in the true sense of the term. Chomsky specifies that those properties of language which consist in its being infinite and free of stimulus control do not exceed the limitations of a mechanical explanation by themselves. For him as for the Cartesians, it seems that it is primarily the coherent and adaptable character of normal linguistic behavior which distinguishes man from animals and from any automaton. What is 25 26 27

Cf. for example, by McCawley, p. 251 of this book. Cf. pp. 229-230 of this book. See the example given by McCawley, p. 253 of this book.

JACQUES BOUVERESSE

323

decisive is the fact that the speaker is capable of an infinite number of very diversified linguistic products which are obviously not formed haphazardly, but in relation to an infinite number of different situations. That is actually where the creativity of language use resides. This creativity, however, is not absolute, of course: due to the fact that linguistic behavior adapts to the circumstances, it 'depends' on them in a very strict sense; it 'responds' to something, even if it is not a response in the behavioristic sense of the term. On the other hand, to be really rigorous, it is the first creativity which is infinite and not the second. The infinite creativity is that which resides in the syntactic system - the system of formation rules - and it follows from the fact that this system does not impose any limits on the length and complexity of the sentences. Yet, the amount and the variety of the sentences which have been effectively used at one time or another remains constantly finite, although without a doubt very large. Chomsky suggests sometimes that infinity is an immediately recognizable property of language use, of linguistic behavior. At other times, however, he expresses himself as if it were merely a necessary characteristic of all grammars because it would be inconvenient to prescribe an upper limit to the length and the degree of complexity of the sentences. For what exactly does the description of the generative grammar, which is supposed to have been internalized by the speaker, account? Mainly for the possibility of producing and understanding an infinite number of sentences on request, most of which are new. Again, this holds no mysteries by itself. In return, what is highly mysterious is that this possibility should be used as it is, that we should actually use an infinite or a very large variety of sentences in an infinite number of different situations, in a coherent and appropriate way. Unfortunately, the grammarian does not have anything to tell us about that creativity. He restricts himself, like everybody else, to noticing it. Its study does not belong to the theory of competence but to that of performance. One can only say that creativity in the first sense is one of the conditions which render creativity in the second sense possible, that competence is combined in a form we do not know with other elements in order to produce performance. In the Chomskyan conception, the speaker has a potentially infinite number of ideas, and he has available, because of his competence, a potentially infinite number of sentences to express them when he feels the need to do so. The truly creative faculty of man is located on the level of the production of ideas. This is what needs explaining, if possible, in a way other than by the intervention of the Cartesian soul. In answer to your question, I can say that Chomsky is justified in remaining placid concerning the notion of linguistic creativity, because, as he says himself, it is finally a rather evident matter. It must be well understood, however, what generative grammar contributes in that respect and what it cannot contribute at all. As for Martinet, 1 am quite astonished about the placidity with which he, on the other hand, ignores the problem.... There is really a whole series of elements which we have to examine thoroughly here.

324

JACQUES BOUVERESSE

First of all, there is a question which I ask myself while I am listening to you: if I understand you well, the Cartesian soul is not the Chomskyan mind. - I suppose not; I mean that Chomsky's indications in this respect are not always very clear. In fact, as I have said before, the Chomskyan linguists deny that they are practicing a 'theological' mentalism. The philosophical perspective in which they use linguistic creativity is not at all that of Descartes. The Cartesians invoke this creativity in order to demonstrate that man is qualitatively different from any animal or machine, but Chomsky's question is rather the following: what sort of (physical) system must the human organism be, in order to be able to acquire and use language as it does? There is nothing in the properly linguistic description of language Chomsky provides which necessarily presupposes that it should be used by a conscious being. In fact, one could almost say that the question is the following: what sort of machine or (abstract) automaton is the being which is gifted with language ? Descartes thinks that it can in no way be a machine. For Chomsky, however, things are probably less clear than that. Maybe Sidney Hook is right when he writes: "I wonder whether lurking behind his animadversions against empiricism theie is a hostility to naturalism whose source is an unqualified Cartesian dualism that takes man and/or the human mind outside the order of nature." 28 For me, Chomsky is somebody who is motivated from one side by scientific preoccupations with a very clear empirical orientation - in principle, he never proposes philosophical theories, but only testable, scientific theories - and from the other side by ethical imperatives that exert a very strong pressure on him. Unfortunately, the two things do not always go together very well. I do not make this remark with any irony, because, in a way, it is the problem of all of us. One of the first postulates of scientific work, in human sciences as well as anywhere else, is precisely that no privileged position be a priori attributed to man in the totality of natural phenomena. For the sake of clarification I would like to ask you i f , in your opinion, a speaking automaton must possess this second kind of creativity or whether the mathematical creativity is sufficient for it to speak. - Chomsky shows us, for example, that a finite state machine, even if it can eventually produce an infinite number of different sentences, cannot produce all the grammatical sentences of a natural language, English, for example. A machine which is capable of enumerating the correct English sentences explicitly, however, is certainly not yet a machine which is able to speak English. An automaton appropriately programmed can produce an infinite set of sentences or even, if you want, utterances', but that does not mean that it could produce an infinite number of things, which we would call instances of speech, in the real sense of the word. Again, what is decisive is less the innovatory character than the pertinent and adjusted character of this production. Moreover, it must be specified that modern automata are capable of very or28

S. Hook (ed.), Language and Philosophy, p. 167.

JACQUES BOUVERESSE

325

ganized and adapted behavior, which Descartes could evidently not conceive, but there is a serious distance from there to being able to simulate human linguistic behavior exactly.... I do not know whether you have already seen samples of dialogues between a computer programmed for that purpose and a human being. One can obtain fragments of conversation which are not totally unnatural and lifeless. The machine can be deceiving on short stretches, but not on long ones. Yet, even if an automaton were capable of linguistic performances as coherent, diversified, and adapted as those of human beings, which is theoretically not impossible, there would be people who would say that it does not really speak because it does not express 'ideas', that what it says does not correspond to any 'intention', etc. Note that, like the ideas a speaker has about his language, the ideas he expresses in his language may very well be, according to the mentalist semanticists, unconscious and totally inaccessible to introspection. In their work one can find observations which suggest that the 'thoughts' or the 'ideas' in question could in the end be something like hypothetical states or processes in the central nervous system. If that is so, an automaton is evidently iess incapable of having 'ideas' than it seems to be the case if one adopts the classical mentalistic perspective. What do you think of Chomsky's affirmation that language use is free from control by internal or external stimuli ? - Sometimes he says only "stimuli", sometimes "detectable or identifiable stimuli", sometimes "independently identifiable" - which is not the same. If one accepts the idea that it should be possible for linguistic behavior, like all other observable phenomena, to be the object of a causal explanation, one should also admit, it seems to me, that it is conditioned in a strictly causal way, even if it is for the time being rather ridiculous or Utopian to attempt to describe it in such a way. What annoys me in the idea that linguistic behavior, or at least an essential part of that behavior, could be exempt from control by internal and external stimuli is that I wonder how one knows that. Do we not have the prototype of introspective illusion here ? Refer to what some philosophers have said about the 'feeling' or the 'impression' we have of our free determination. It would be rather ridiculous to describe the conversation we are having now by saying that we are producing relatively elaborate responses to stimuli. I am not sure, however, whether one elucidates things more clearly by saying that we are 'expressing ideas'. If you ask me exactly what idea I have expressed in my last sentence, I will probably not try to find out by introspection what happened in my mind while I talked. I would give you a paraphrase of what I said, perhaps more explicit. In a way, one never proceeds from words to ideas, but always from words to words. Although I do not have a priori very much sympathy for behaviorism, I think it is dangerous to suggest the existence of two mechanisms which function in a parallel way: the mental mechanism and the verbal mechanism, between which there exists a relationship defined by the so-called 'encoding' or 'decoding'. To consider language

326

JACQUES BOUVERESSE

as a sort of code for transcribing ideas does not account for what happens, so it seems, in most cases. Wittgenstein might have some interesting things to say here. I just wanted to ask you what Wittgenstein thinks of linguistic creativity... - Another delicate question. I think that, on the whole, Wittgenstein does not think anything special about it. You probably know that at the time of the Tractatus, he considered that one of the major interests of the "picture theory" of language was exactly the explanation of how we can understand a new sentence, a sentence we have never met before, without having it explained. He observed that one of the essential properties of a proposition is that it can communicate a new meaning to us. 29 Afterwards he abandoned the picture theory, and he has not expressly tried, as far as I know, to explain the so-called linguistic creativity in another way. One can find in his work a whole series of sometimes obscure philosophical reflections on the kind of 'creativity' contained in a recursive characterization: what is meant, for example, when it is said that an algebraic formula defines or produces an infinite series of numbers? But Wittgenstein never really tries to explain, to account for something. This is also what Chomsky mainly criticizes, although he also criticizes him for giving totally inadequate explanations. Wittgenstein's objectives are not at all explanatory but descriptive. In particular, he has never tried to produce a theory which would account for linguistic creativity; he has merely criticized systematically certain inadequate representations and improper ways of expression which give a misleading idea of what actually happens and which lead to errors or philosophical perplexities. Wittgenstein does not distrust theories because they are theories, but for the reason and to the extent that they are susceptible to leading to philosophical misapprehensions or sicknesses. One point must be emphasized here - almost always misjudged by the linguists - namely, that Wittgenstein is not an amateur linguist nor even a genuine language philosopher. He is very different from Austin, for example. How, then, did Wittgenstein describe the production of new ideas ? - As far as I can judge, Wittgenstein does not really believe in the importance of 'ideas' in language use. He has a real antipsychological orientation. He always insists on the observable aspects and, in particular, the social aspects of language use; he is, on the other hand, systematically defiant of the ideational aspect of language use. But if he insists particularly on certain aspects in opposition to others, it is not at all from a theoretical or an anti-theoretical conviction. It is because philosophers have a tendency to neglect these aspects constantly, and that is the beginning of their misfortune. I think Wittgenstein's attitude from the thirties onward can be characterized generally by stating that he considers psychological considerations as rather trivial and uninteresting; he does not think they can really clarify the nature and function of language, in particular because of a rather simple general idea which he 29

L. Wittgenstein, Tractatus Logico-philosophicus (1921) (London, 1922), 4.027.

JACQUES BOUVERESSE

327

formulates in the Blue Book - namely, that, in his opinion, everything one can interpolate between the symbol and the reality in order to explain the working of the symbol, is still essentially a symbol, everything one can propose as an interpretation of the symbol is still something which must be interpreted. But Wittgenstein is certainly not a behaviorist in the usual sense of the term. He does not at all deny the existence of mental processes, nor, of course, their intervention in linguistic behavior. You have written an article about Chomsky entitled "The Ghost in the Machine".30 Could you explain that title? - The expression "The Ghost in the Machine" is, as you know, used by Ryle to characterize the dualist Cartesian conception of the human mind. The article to which you allude is (among other things) mainly devoted to an examination of certain Cartesian ideas defended by Chomsky, principally in Cartesian Linguistics and in Language and Mind. As I have said before, what intrigues me most in Chomsky's philosophy is this kind of Cartesianism without the soul, the dualism without the duality, that mind which in the end might be a machine, and those mental processes which have probably nothing mental except the name.... There is another very discussed affirmation in Chomsky: the grammar interiorized by the speaker is unconscious. What is your reaction to that idea ? - My reaction would be less unfavorable if Chomsky had not elsewhere insisted so much on language mastery as a cognitive phenomenon and if he had not adhered so much to describing language disposal in terms of possessing certain (theoretical) knowledges. What exactly should the 'knowledge' we have of our language be? According to Chomsky, linguistic competence is at the same time (an important aspect of) the faculty to speak and understand language, and the knowledge of the fact that language is described by the rules of the grammar. In fact, the essential element of what we will call the 'competence' of a chess player is not his knowledge of the rules of the game, that is, of the admissible moves. In like manner the essential element of the language speaker's competence is not his 'knowledge' of the rules of grammar either, that is, of the admissible sentences. In both cases, it is rather the faculty to use in a certain way the possibilities defined by the rules. But let us return to our problem. For Chomsky, the knowledge (of the grammar) of language is really not a knowing how or a knowing that. It belongs to a third form of knowledge: the tacit or implicit knowledge. That notwithstanding, he constantly talks about 'principles', 'ideas', 'theories', etc., which we mainly possess unknowingly, especially when he discusses the problem of innateness. An implicit or unconscious knowing that is quite obviously at stake. I mean that an unconscious propositional competence, the unconscious knowledge of the truth of certain assertions about language, if one can talk about anything like that, is no less a prepositional competence, a knowing that. But what might the term 'unconscious' mean here ? 30

J. Bouveresse, "Le fantôme dans la machine", in La parole malheureuse (Paris, 1971), 397-470.

328

JACQUES BOUVERESSE

As one might expect, it has been suggested that the rules of grammar could in the end be located in something like the Freudian unconscious. Why not, after all? The Freudian unconscious is of a very hospitable disposition! The question is what is meant when one says that. I think, for my part, that, as Wittgenstein recommended, one should not use a concept without an appropriate antithesis. Now, when you talk about an 'unconscious' process, it means something different according to whether a process is referred to which could in principle also be conscious, or whether a process is involved which could in no way be conscious. Given that you have the unconscious desire to kill your father, you could also, if such should be the case, have that desire consciously, and you could also become conscious of the fact that you unconsciously want to do it. There are a large number of 'unconscious' processes, however, which are not at all unconscious in that sense - for example, a great deal of physiological phenomena which take place in the human body. One would have to know in what sense the speaker's possession of an adequate grammar may be considered as unconscious. Willfully taking a rough example and running the risk of passing for a totally unpolished behaviorist, I would say that we do not attribute to the cellular membrane a 'tacit' conceptual knowledge of the difference between some substances and others on the pretence that it does make such a difference, that it has a very selective permeability which explains its acceptance of some but not of others. The difficulty comes from the fact that Chomsky uses two very different languages simultaneously. He suggests on one side that the prescriptions of the grammar are simply registered in mechanisms, probably neurophysiological mechanisms, of which the functioning may naturally be partly accompanied by conscience; on the other hand, he presents them as systems of rules of which we have a certain form of 'knowledge'. It must be noted - I believe Chomsky says it somewhere himself - that the question is simply determining whether one accepts the application of the term 'knowledge' to a problematic case. In that event, however, the opposition between a cognitivistic approach to which Chomsky is visibly very attached for profound reasons - and a dispositionalistic or behavioristic approach loses a lot of its meaning and virulence. Could it be said that the phenomena which play a part in the unconscious linguistic mechanisms are accessible to conscience in different ways ? Could it be said that there are degrees of accessiblity to consciousness ? - I think one can do that, indeed. Moreover, the comparison with psychoanalysis is not totally absurd in this respect. Just as the subject can recognize an unconscious impulse as an impulse he had without knowing it, it may happen - and that certainly occurs - that he recognizes a rule which has been presented to him by the grammarian as a rule which he has always applied, but without knowing it. What is, in that case, the criterion that it is really that rule he applied? Wittgenstein would probably say that it is essentially the fact that he recognizes it as such, but let us drop this question. It is clear that transformational analysis is most probably not on the same level as immediate constituent analysis. I do not know whether it would

JACQUES BOUVERESSE

329

be very easy to make the speaker 'conscious' of the transformation rules he uses to proceed from the base to the surface structures. The problem is still much more difficult with regard to the 'knowledge' of language at its highest level of generalization and abstraction, to the universal grammar of which the speaker is supposed to know the principles innately. How can you make him aware of the fact that he had unknowingly available a sort of 'definition' of what a potential human language is, a possible human language in general? How can you make him aware, for example, that his language must necessarily involve transformations? 31 There is evidently no reason at all why the universal principles which make possible and determine the knowledge of any language could be 'known' themselves in an acceptable meaning of the term. The question itself may not make any sense, but, in that case, why does one cling so much to talking about knowledge ? Chomsky has always made a distinction between grammaticality and acceptability. One could say that the speaker is (up to a certain point) in a position to relate the intuitive acceptability or unacceptability of sentences to a concept of grammaticality and ungrammaticality which he possesses largely without his knowing it. Grammaticality in the narrow sense, however, is only a limited aspect of the linguistic phenomenon. One could wonder how the logico-semantic structure, for example, is accessible to consciousness. One could say that grammaticality, in the sense of syntactic correction, is quite accessible to consciousness, but there is still the problem of knowing what happens with the logico-semantic structure ... - You mean, I suppose, that there is also the question of how the system of rules which constitutes the semantic component of a transformational grammar could be 'recognized' itself. There is, I believe, a large preliminary difficulty which has often been indicated and which consists of knowing whether there exists a specific intuition of grammaticality in opposition to the intuition of meaning, whether the normal language speaker makes as neat a distinction, as is sometimes suggested, between syntactic correction and semantic correction. At the time of Syntactic Structures, Chomsky really believed in the autonomy of syntax, in the strict independence of syntactic considerations. Apparently, he now has more reservation. One could actually maintain that an informant, when he formulates a linguistic judgment on a sentence, always reacts to what you called the 'logical-semantic structure' of that sentence. I myself am inclined, with McCawley, 32 to feel that the predicates 'grammatical' and 'ungrammatical' are not applied to strings of words, but to things which are already essentially semantic structures, that the question of grammaticality is not that of "grammaticality of the sentence per se but of the sentence relative to its use". The question asked is probably always that of the potential meaning or use of the sentence, with ungrammaticality corresponding simply to a more radical impossibility 31

Cf. T. Nagel, "Linguistics and Epistemology", in S. Hook (ed.), Language and Philosophy, 171-

182. 32

Cf. p. 252 of this book.

330

JACQUES BOUVERESSE

of signifying. As William James says, "nonsense in grammatical form sounds half rational; sense with grammatical sequence upset sounds nonsensical - e.g., 'Elba the Napoleon English faith had banished broken to he Saint because Helena at.'" 3 3 In the conversation he had with you, Chomsky protests against the people who attribute to him the idea that syntax and semantics must be rigorously distinguished.34 He recalls that his position in this respect is anti-dogmatic and agnostic: whether there exists a strict distinction between syntax and semantics and also between semantics and the system of knowledge and beliefs (that is, between semantics and pragmatics) is an open question. But the problem might not be so much whether syntax is totally independent of semantics, whether there is a strict line of demarcation between the two: it might be that syntax is actually totally dependent on semantics (and vice versa), that neither of the two disciplines is really autonomous. It is true that the position adopted by Chomsky in Syntactic Structures leans on arguments which should not be disregarded. Probably everyone intuitively perceives a difference between the deviance of sentences such as "Caesar is and" and "Caesar is an integer" (examples of Carnap), or of expressions such as "green is or" and "a round square" (examples of Husserl). However, this does not imply, it seems to me, that the question asked in each case would not be a semantic question, that the four preceding expressions would not be first and foremost a-semantic. For Husserl, an expression like "green is or" is excluded by the pure logical grammar or by the 'pure morphology of meanings'; a sentence like "every triangle has four angles" is meaningful, but violates restrictions formulated at another level, namely that of the analytics of consequence or of noncontradiction.35 But where, for example, must Lewis Carroll's sentence be situated: "The slithy toves did gyre and gimble in the wabe", in one way syntactically well constructed, and in another syntactically (lexically) aberrant? On the side of Husserl's Unsinn, what is rejected at the level of the 'pure logical grammar', or of the Widersinn, what is rejected at the level of the analytics of consequence only? At any rate, the pure logical grammar is a semantic grammar. The 'pure morphology of meanings' is a formal discipline which concerns the forms of syntactically possible judgments in general. If it puts in parentheses semantic questions of formal and material consistency and of truth, however, it does not parenthesize that of meaning: its role is, on the contrary, to enumerate the ideal types of meaning. Are you defending Husserl's semanticism against the syntacticism for which Chomsky has sometimes been blamed? - Husserl's position as a whole raises, I think, much more redoubtable problems than Chomsky's, on which we cannot dwell here. There is a fundamental question 33

W. James, The Principles of Psychology (New York, 1890), I, 269. Cf. p. 30 of this book. 35 E. Husserl, Logical Investigations (Logische Untersuchungen, Halle, 1913), Fourth Logical Investigation. 34

JACQUES BOUVERESSE

331

which consists of the fact that a really explanatory theory ot language should succeed in somehow filling the ditch which exists between logico-semantic characteristics and syntactic properties, relations and classifications. This is the problem of what Strawson calls the "condition of perspicuousness". One may feel the need for a theory which would formulate general principles by which lexical formatives are assigned to grammatical categories, which would describe "intrinsic and general connections between types of element-meaning and potentialities of grammatical role in deep or base structure".36 Such a theory, which would link syntactic and semantic considerations, could bring us nearer, as Strawson observes, to a complete explanation of the linguistic competence of the ideal speaker-listener. Chomsky seems to have little sympathy for such a conception, however. He does not a priori exclude that the vocabulary of the grammatical description should in the end make reference to certain semantic concepts, according to a traditional conception. However, he does not consider it a priori evident nor established that it must necessarily do so. The question is evidently not whether one already has an adequate theory of the semantic component at his disposal. It is a preliminary question. Returning to our problem of to what exactly the notion of grammaticality must be applied, it is clear that, a predicate of expressions like the predicate 'grammatical' can be represented formally, in the technical sense of that expression, in an adequate system. But this formal characterization only provides an extensional definition; it does not analyze the predicate 'grammatical', it does not explain what the 'transcendent' category, as Quine calls it, of grammaticality is, what the expression 'grammatical in L' means for any language L. At this stage, the grammatical sentences are nothing but those for which there is a derivation in the system.... This discussion started about linguistic creativity. Let's come back to this notion. I would like to ask you to situate the concept of''rule governed creativity'. How must we characterize the difference between 'governed by rules' and'governed by mechanisms' ? - There is certainly a difference. The rules are in principle part of the description of the language used. Some people have indeed reproached Chomsky for considering as language universals the universal properties which the rules used in the description of any language must necessarily possess. The explanation by mechanisms, in the strict sense, is actually a causal explanation in terms of the laws of nature; it is basically a part of the theory of performance, and is dependent on psychology and physiology. Characteristic of transformational theory is the affirmation that a statement of the rules which serve in the definition of language must give us an idea of the linguistic mechanisms involved. On one side, we have an abstract 'mathematical' description of language; on the other, perhaps, an embryonic description of the psychological or neurophysiological mechanisms which are implied in language mastery. How can one hope to coordinate these two descriptions? How, in 36

P. Strawson, "Grammar and Philosophy", in Logico-linguistic Papers (London, 1971), p. 138.

332

JACQUES BOUVERESSE

particular, can one hope to 'translate' the knowledge of the rules the speaker is supposed to have in an element which can causally interfere in linguistic performance ? No one knows at the moment. Anyway, there is a priori a considerable difference between obeying laws, in the sense of natural sciences, and obeying rules, particularly 'constitutive' rather than 'regulative' rules, in Searle's terminology. What exactly is meant by 'following a rule' is a question which never stopped tormenting Wittgenstein. Do you agree that linguistic creativity would not exist if it merely obeyd mechanisms? - It would certainly no longer exist in the genuine Cartesian sense of the term, where creativity is supposed to be the manifestation of a spiritual power. Otherwise, it is a traditional philosophical question which I would rather not go into here, namely, to what extent creativity, innovation, or freedom in the strong sense are compatible with an integrally mechanistic explanation of the phenomena. It is no longer a real question oi the philosophy of language but a very old metaphysical question, that of determinism and freedom.... Chomsky automatically associates the innateness hypothesis to that of linguistic creativity. I would like to ask you if you see an inherent necessity to associate innateness and linguistic creativity automatically. - The dispute on linguistic innateness seems rather strange to me. Three very different spontaneous reactions are possible, I think, with regard to the innateness hypothesis. One may have the feeling that a rather trivial truth of common sense is involved, that a highly speculative hypothesis of a metaphysical nature is at stake, or, on the contrary, that it is a precise and interesting scientific hypothesis, an empirically testable hypothesis. My general feeling is that it is something quite evident which Chomsky tries with very relative success to present as a conjecture with real scientific content. What the discussions he has had with certain people in Sidney Hook's volume 37 clearly demonstrate is not that they are definitely hostile to the hypothesis itself, but rather that Chomsky has not persuaded them that it would really be an empirical hypothesis with an exact meaning. Indeed, you have to observe immediately that, if the innateness hypothesis were really empirically controllable, it could not a priori offend a consistent empiricist; on the contrary, it would considerably interest him. Initially Chomsky had actually associated, at least indirectly, the two ideas of linguistic creativity and innateness because he generally opposed rationalism and innateness to all those conceptions that are criticized for not or for insufficiently accounting for linguistic creativity: empiricism, behaviorism, structuralism, etc. Afterwards, he considerably modified his position in this respect, due partially perhaps to certain restatements by Quine, who showed that behaviorism as he understood it was not at all opposed to innateness. Now, see your conversation with him, 38 he clearly recognizes 37 38

S. Hook (ed.), Language and Philosophy. Cf. p. 30 of this book.

JACQUES BOUVERESSE

333

that all language theories, whether they are rationalistic, empiricistic, behavioristic, or other, need a specific hypothesis of innateness. There are, thus, two questions involved here - that of innateness in general, and that of the specific features of the rationalist innateness hypothesis. Let us start with the first question, if you like. What do you mean when you say that behind the innateness creeds there is a rather simple truth of common sense ? - This is precisely related to the second question, namely that everyone is necessarily a nativist in one way or another. Quine, for example, hostile to innateness interpreted mentalistically, to the hypothesis that there might be such things as 'ideas' or 'principles' deposited at the onset in the human mind, is not at all loath to admit, and admit largely, the existence of things like innate mechanisms, dispositions, or predispositions. Moreover, although the traditional empiricists speak a mentalistic language, their position is essentially the same as Quine's. Locke's criticism, for example, bears mainly on the existence ot innate contents of thought or representation, and not on the existence of innate mechanisms of thought or representation. He would not consider denying (but who could?) that there exist what one could call innate potentialities, faculties, or tendencies. Neither Hume nor Locke, in fact, considered the human mind as a tabula rasa in the strict sense, a sort of unwrought wax which could be modeled in any possible way. On the contrary, they are rationalists in a very precise sense: they believe in man's innate rational nature. At the beginning of Problems of Knowledge and Freedom,39 Chomsky quotes a very characteristic text where Hume insists on the important part of unacquired and almost unpeifectible knowledge which we have "from the original hand of nature". It is true that, according to Chomsky, an empiricist who comes to such conclusions actually abandons an essential aspect of what constitutes the specificity of features in an empiricist theory. Chomsky suggests that, considering what Quine understands by 'behaviorism', he has, in the final analysis, few reasons to consider himself an authentic behaviorist. He tends to forget, however, that the argument is also of value in the inverse direction and that, if 'reasonable' empiricists might no longer have any reason to present thenselves as empiricists, he, on the other hand, might no longer have any reason to present himself at all costs as a rationalist. Chomsky occasionally appeals to people like Jacques Monod or Konrad Lorenz, but I do not see how anybody could seriously dispute the existence of innate biological structures, in Monod's sense, or of innate behavioral schemes, in Lorenz' sense. The main difficulty in Chomsky's innateness hypothesis is the insistence with which he talks about innate 'concepts', 'principles', or 'knowledge'. Is the recourse to Descartes here justified? - I t obviously is not if the innateness in question is interpreted, as Monod suggests it should be, in biological terms. Chomsky probably does not really mean that the 9

N. Chomsky, Problems of Knowledge and Freedom, p. 14.

334

JACQUES BOUVERESSE

human mind has available from the outset a stock of innate 'ideas' or 'truths', in the sense Descartes and Leibniz understood it. In that respect his reference to these two authors is quite disputable. Even if it is taken into account that the Leibnizian version of innateness, for example, is largely dispositional, that he mainly considers as innate, inclinations, dispositions, habits, or faculties. I insist on the fact that, until the contrary is proven, the great classical rationalists really believed in the existence of a priori and necessary truths or principles, which are part of the original endowment of the human mind, even if they can only be progressively revealed in the course of human development and if their emergence implies previous contact with experience. What, then, justifies Chomskys reference to Monod? - Well, as I said before, I think one must choose; it is impossible to refer to Descartes and to Monod at the same time. One should know whether one is really talking about innate cognitive principles or, on the contrary, about innate mechanisms or cerebral structures. It is not permissible to reinterpret a traditional author completely in a totally different conceptual framework and at the same time pretend to have merely resumed a completely traditional conception.40 But, is there, in the end, an innateness problem in general? - I would not want you to get the impression that there is no such problem from what I said. There is, in particular, the problem Searle raised, namely that if the behaviorist admits the existence of innate mechanisms sufficiently rich to account for the power and complexity of the grammar interiorized by the speaker, then the crucial element of the learning theory consists of the mechanisms in question, and no longer of mechanisms of the stimulus-response type which would only serve to start the innate apparatus. 41 If you think about it, the more the mechanisms which are supposed to account for learning are simplified and impoverished, the more the innate apparatus, which allows these mechanisms to fulfill their function plausibly, must be enriched. The general problem posed here is the following: how can one hope to bridge the considerable gap existing between the poverty of experience and the richness of implicit theories which have been constructed on the basis of that experience? Chomsky considers that the procedures described by the empiricist and inductivist tradition are unable to accomplish this. When he observes, for example, that the principle of cyclical application of the phonological rules could not have been acquired by induction on the basis of the linguistic data a child has at its disposal, this argument should not be taken lightly. It should, however, also be said that the mechanisms of idea formation and of language acquisition, which have been admitted by the empiricists, or more generally 40 Cf. on this matter, D. E. Cooper, "Innateness: Old and New", The Philosophical Review 81 (1972), 465-483. 41 J. Searle, "Chomsky's Revolution in Linguistics", The New York Review of Books 18 (1972): 12, p. 21.

JACQUES BOUVERESSE

335

by the partisans of inductivism, are probably not so rudimentary as they seem at first sight, or, if you prefer, that their determination is not sufficiently strict to make it possible that one could a priori with certainty know what can and what cannot be learned in an inductive way. It is probably an illusion to believe that one has a clear idea of the maximum degree of the complexity and the richness of the results which can be obtained by an acquisition procedure conceived according to a classical empiricist scheme. As a final point, it must be stressed that the concept of an innate device which conditions the acquisition of knowledges is above all a residual concept. Innate is only that for which there are serious reasons to consider that it cannot have been acquired in any way. There are people, however, who, without taking a side on the theoretical problem, consider it more urgent and more promising to take an interest first of all in what has been learned and in the way it has been learned. This may be a difference of methodology. The nativists think it is indispensable to make sufficiently rich hypotheses from now on about the innate faculty of language; the anti-nativists, on the contrary, are of the opinion that the environmental variables must be studied first, that an explanation in that sense has to be pushed as far as possible with a recourse to innateness hypotheses only in the final analysis. As Campbell and Wales remark, "It is a matter of methodological emphasis whether one directs attention to the environmental variables or to the predispositions." They justify their preference for the first approach as follows: "... It seems to us (and to this extent we are in line with traditional approaches) that it is the environmental variables that should be made the primary object of study, since they are more accessible to investigation." 42 It might, of course, be a complete mistake to believe that the contribution of experience and of the environment to language acquisition is really more accessible to research than the specific contribution of the human mind.... With regard to this innateness hypothesis, is it so that what happens in the case of language is analogous to what happens in the case of other forms of behavior ? Do the innate dispositions exist in the same way for other forms of behavior as for linguistic usage or does language have a privileged position in this respect ? - This leads us to our second question - namely, how the rationalist innateness hypothesis is distinguished from other innateness hypotheses. If 1 understand it well, it is mainly through the eminent and very particular position which Chomskyan rationalism attributes to the faculty of language in the whole of human faculties. For Chomsky, man is first and foremost a linguistic animal and perhaps more exactly, as Searle remarks, 43 a syntactic animal. The syntactic component of a grammar, about which present linguistics has more precise things to say than about the semantic component, reflects highly specific properties of the human mind, but it is actually the total system of rules linking sound and meaning-the grammar-which represents, 42

R. Campbell and R. Wales, "The Study of Language Acquisition", in J. Lyons (ed.), New Horizons in Linguistics (Harmondsworth, 1970), p. 249. 43 J. Searle, "Chomsky's Revolution", p. 19.

336

JACQUES BOUVERESSE

in Chomsky's opinion, a very particular case of knowledge and of acquisition of knowledge. Everyone probably agrees with Chomsky that the knowledge of language results from the combined intervention of mental or cerebral structures given at the start, of maturation processes, and of an interaction with the environment. That being the case, there are different ways in which to account for the language invariants that can be observed or discovered. The rationalist innateness hypothesis consists of the consideration that they reflect directly properties characteristic of the innate faculty of language which is itself a characteristic faculty of the mind. A different solution would be to consider that language acquisition rests on more general principles which also condition other forms of learning: that the most specific aspects of language knowledge can themselves be explained in terms of 'learning strategies' with multiple visage which are applied to a sufficiently uniform environment. Chomsky does not believe in the possibility of explaining language universals that way, because there exist, according to him, very remarkable universal properties of language which are, so to speak, 'gratuitous', for which no extrinsic justification can be conceived and which, in particular, cannot be considered indispensable or even natural properties of any system whatever that is susceptible of fulfilling the linguistic function. Thus, for example, the fact that the formal operations in English grammar or in any other language are structure dependent and the fact that they are, in a certain sense, independent of meaning must be considered, according to Chomsky, formal linguistic universals or principles of universal grammar. It annoys some people that even if the absolute universality of such characteristics had been empirically established, the fact that every known human language possesses them would still not prove that they constitute essential properties of a language in general, that every possible human language must necessarily possess them, at least so that it could be learned as a first language. We have there what Putnam calls an argument of the type What else ? and Black a conclusion ab ignorantia rationis. I want to insist here on the main characteristic in Chomsky's attitude: he considers really interesting only the language properties which are directly relevant to the structure of the mind, and as essential properties of language only those properties which, apparently, cannot be explained by considerations of functional usefulness, or simplicity, etc. In short, he treats all the properties, which are conceivably imposed on language by the fact that it must be used as an instrument of communication, as rather accidental properties of language. He considers language radically different from any other system of communication. T o say that man is a linguistic animal is not just saying that he is an animal who has at his disposal communicative systems immeasurably more perfected than those of the other animals. Much more than that is said. It may be true in the end, but I cannot help finding Chomsky's way of viewing language strange at times. His constant insistence on the most gratuitous and disinterested aspects of the linguistic phenomenon is rather disconcerting: one sometimes gets the impression that, in his opinion, language serves mainly to produce an infinite number of sentences

JACQUES BOUVERESSE

337

ad majorem linguae gloriam " not in order that they may be used, but because they are good for something. Chomsky himself still talks about a certain role, though partial, of the environment in linguistic activity. - The environment necessarily plays a role which remains important in all theories, for language acquisition as well as for language use. As I have said, the differences here are rather of degree than of nature, whatever Chomsky may think. In the innateness theory of learning, the environmental data to which the speaker is 'exposed' serve to select among the a priori possible grammars a limited number of descriptively adequate grammars, corresponding to the data; a second selection among the 'candidates' is then made by dint of an evaluation procedure, which provides the optimal grammar. Characteristic of this theory is the consideration that the environment serves at the most to activate and specialize a pre-existing complex mechanism. Some people believe there is a contradiction in the attitude that admits at the same time innateness and a certain role of the environment in language learning and use... - There is no contradiction at all. That is what everybody does, in fact. It should not be necessary to specify this, if it were not that Chomsky has encouraged a tendency to reason according to the principle "all or nothing". Maybe there is something in your question that escapes me.... It is obvious that an empiricist nativist, even Quine, confides a larger role to the environment in language use than a rationalist nativist does... - Certainly. Previous to any really theoretical stand, some people are more particularly interested in the contextual and environmental constraints which language use obeys, and others in the apparently free, innovatory, and gratuitous aspects of this use. Both certainly exist, and we actually do not have any means to appraise the relative importance of one or the other exactly. The only important thing seems to me an examination of the practical consequences that spring from each of these two attitudes, of the concrete implications they may have concerning the research and the results obtained in each of the two directions. That was the purpose of my question. - One of the linguists you have interviewed - I believe it is McCawley - asserts that the repercussions of the innateness hypothesis on Chomsky's linguistic research are practically inconsequential, that one could perfectly well reject the innateness thesis and still continue to practice the kind of linguistics Chomsky does. 44 I am not sufficiently competent in linguistics to say whether this is the case, but, as a philosopher, 44

Cf. p. 253 of this b o o k .

338

JACQUES BOUVERESSE

I can say that I do not feel an urgent need to take sides in the innateness problem. It is true that I am much less interested in traditional philosophical problems than Chomsky.... Do you refuse to talk about a first and a second Quine with relation to the innateness problem ? - I myself have never had the impression that there were two or more Quines. Of course, it is quite possible that something essential has escaped me in that respect. It is true that Quine has been led to make his conceptions much more explicit and precise, notably as a result of Chomsky's attacks, but this does not mean, as the latter seems to think, that he has felt the need to modify them substantially. It is probably closer to the truth that Quine has never vindicated the rather caricatural conceptions which Chomsky attributed to him at the start and that the latter has considered an actually more detailed re-exposition of the same ideas, probably also more circumspect and with more shades of difference, to be an important evolution. What strikes me in Quine as well as in Chomsky is their loyalty to their main ideas. Generative grammar treats intuition and introspection as privileged criteria and procedures in linguistics. Does the philosopher have some methodological remarks to make here ? - It is really, I am afraid, a question for the language science specialist and not for the philosopher. Searle explains very well in Speech Acts45 why the 'verification' of linguistic characterizations does not need to be carried out by a statistical observation of the verbal behavior of the linguistic community, but simply operates by the testimony of the speaker's linguistic intuitions, for example of the linguist. The linguistic judgments of the normal language users, insofar as they exist, proceed from their more or less perfect mastery of that language, and mastering a language is, among other things, having interiorized the systematic rules which govern a certain form of behavior. The linguistic intuitions are the manifestations of the underlying rules, and the rules are the systematization of the intuitions. The intuitions exist only because there are systematic rules and the rules 'justify' the intuitions. In other words, the question whether there really exist sufficiently precise pre-systematic intuitions and the question whether there are really systems of underlying rules are, in a way, one and the same question. Of course, I believe that any linguist is obliged to start from intuitive linguistic characterizations in phonology as well as in syntax or semantics and that, in any case, he makes an essential usage of the personal linguistic intuitions. People like Martinet are visibly impressed by the considerable amount of idealization involved in the theoretical reconstruction of the potential of the speakers' linguistic intuition,46 but this in itself is not an objection. As Chomsky observes, all one can ask from such a theoretical reconstruction is that it agree with the intuition 45 46

J. Searle, Speech Acts, pp. 4-11. Cf. p. 241 of this book.

JACQUES BOUVERESSE

339

where it exists, that is, in the clear cases, and to decide in a rather conventional way in the other cases. Chomsky is related to Goodman here, who, in spite of appearances, has influenced him a great deal in certain respects. This means that it is actually useless to enumerate litigious cases, no matter how numerous, with regard to the question of grammaticality, for example, or that of semanticity, in order to raise doubts about the idea of linguistic competence itself. When I say that it is a question for experts, I mean that I myself do not know at all whether there are linguistic intuitions for the normal speaker, sufficiently developed to motivate the construction of a theory of competence. It is clear that, in the same way as the competence of which the linguist speaks is the ideal speaker's competence, that is, nobody's competence, so also the intuitions of which he speaks are the ideal speaker's intuitions. Linguists like Halliday think that the Chomskyan conception of competence implies a totally excessive idealization, that one must stick to the lowest degree of idealization compatible with the observed facts, and that the only interesting distinction is between the potential and the actual which realizes it, between what the speaker does and what he could do. 47 It seems that the distinction between competence and performance as it was conceived in the beginning by Chomsky is now being attacked from different sides, that one no longer needs, for example, to distinguish as clearly as he did between 'grammaticality' and 'acceptability'. I myself can understand very well that some linguists are not pleased with a definition of grammaticality which obliges considering as grammatical a sentence as unacceptable as "The rat that the cat that the dog that the cow tossed worried killed ate the malt". 48 Unfortunately, this is not the only problem posed by the notion of competence, but there is a last important point which I would like to mention - that a distinction must be made between the problem of the existence of immediate linguistic intuitions and the problem which has reference to the possibility of making them explicit. It is clear that, for the normal speaker, the possibility of formulating pretheoretical linguistic judgments which reflect certain aspects of his language mastery depends itself mainly on his mastery of the language or of a certain language level. There is certainly nothing less immediate and spontaneous than the technique and the concepts required for the reflexive analysis of linguistic 'impressions', for example, of the very special effect an unprecedented deviant sentence makes on you. Returning to a point which we have already discussed, it could even be said that there are also degrees of accessibility in introspection. That there are facts which escape introspection much more radically than others. - Let us rather say, if you want, that the ability to formulate intuitive linguistic characterizations is extremely variable according to the subjects and the fields considered. The questions of synonymy and that of analyticity, for example, are certainly not on the same level, in that respect, as that of grammaticality. I assume that 47 48

Cf. p. 85 of this book. In P. A. Reich, "The Finiteness of Natural Language", Language 45 (1969), 831-843.

340

JACQUES BOUVERESSE

everyone intuitively perceives a sentence such as "Will drink exquisite young wine the cadaver the" as totally ungrammatical, and a sentence such as "The exquisite cadaver will drink the young wine" as grammatically correct but abnormal in other respects. The existence of clear and reliable introspective data is, however, certainly less evident with regard to the attribution of degrees of grammaticality or with the affirmation whether or not a sentence is an exact paraphrase of another. Taking an example Chomsky himself has given: try to know by introspection whether a sentence such as "There is a non-enumerable infinity of irrational real numbers, but between any two real numbers, there is a rational number", or any rather complex tautological sentence, is normally felt to be analytic. Searle sees the doubt whether to classify a litigious sentence as 'analytic' or 'non-analytic' not as proof of the fact that we do not really have a concept of analyticity but as proof of the contrary: "We would not recognize borderline cases if we did not grasp the concept to begin with." 49 This is in a way incontestable; there are only marginal cases because there are clear cases, and there are only clear cases because there is an implicit concept. Any indecision results from the impossibility of applying a certain decision method and thus presupposes the existence of this method, the existence of 'criteria'. It is probably a decisive argument against Quine, but, in any case, it is rather difficult to speak strictly of an 'intuition' of analyticity, to the extent that what one might be tempted to call by that name probably presupposes a certain familiarity with the philosophical problem of analyticity. We must now pass to a very central and very important question of linguistic practice the question of language universals. It is an old, philosophical question, but it has an enormous practical scope. Some linguists are preoccupied with universal grammar and consider it the most important, most pressing task whereas others are rather preoccupied with the description of particular languages, etc. Chomsky has proposed the distinction between substantial universals and formal universals. How do you consider the relationship between the problem of language universals and the innateness hypothesis ? - I believe the relationship is quite evident. Innateness necessarily implies universality. Chomsky, for example, attributes to the speaker an innate knowledge of what the rationalist grammarians of the seventeenth and eighteenth centuries called a 'universal grammar'. This device is a constitutive property of the rational human nature or, if you wish, of the human mind. There is something which determines from the start in a rather strict way the form of any human language possible. Thus, there are universals - at least formal and perhaps also substantial, but that is already more delicate. Then I will shift my question a little and ask you whether it is necessary to have a rationalist conception of language in order to admit the existence of universal grammar 49

J. Searle, Speech Acts, p. 8.

JACQUES BOUVERESSE

341

or do you conceive that an empiricist approach might also, in certain conditions, lead to that idea ? In other words, do universals of language represent a reserved and privileged domain of rationalism ? - If innateness implies the existence of universal language properties, the latter does not by itself imply innateness or at least an innateness hypothesis of the rationalist type. There are universal language properties which can conceivably be explained by a genetic relationship of the different languages, by the mechanisms of evolution and adaptation, by functional necessities, even environmental constraints, etc. The fact that the sentences of any language never exceed a certain length in practice, or the "law of Zipf", constitute language universals in a way, but for Chomsky they are not really linguistic universals: 50 they can easily be explained by relatively trivial general considerations. As we have seen, the rationalist innateness hypothesis depends on the existence of very special universal properties. Chomsky thinks that he has demonstrated that there exist very general and specific abstract principles which determine the form and the interpretation of sentences, and he formulates the empirical hypothesis that these principles are language universals. This means that, in principle, this hypothesis might have to be specified, modified, or even abandoned with the progress of our empirical knowledge of the different existing languages. Again, to the extent that something is presented as an empirically testable conjecture, it possesses nothing scandalous for an empiricist. People who have reservations on this matter do not necessarily have them because they are a priori hostile to the idea of language universals, but perhaps because they wonder how Chomsky's hypothesis can be tested, how one can know that one has verified the existence of a presumably universal property in a given language, how one can be sure that a property whose presence has been observed in all the cases really results from the innate faculty of language and not from another source which we do not know, etc. Chomsky considers it totally impossible at the present time to express his opinion on the value of the explanations suggested by the empiricists because "only the vaguest of suggestions have been offered". 51 Some people probably think, however, that his hypothesis, in spite of the precise facts he invokes, also remains totally vague in its form, that there is no clearly defined and scientifically productive antithesis between his position and that of his adversaries. Some linguists and philosophers of language are incontestably very sceptical of language universals, and they do not consider the hypothesis which could exist about it worthy of interest... - This is a question about which a philosopher would not seem to have anything to say a priori. One may have the impression that the problem is just a problem of empirical linguistics - that it is for the linguists, and only for them, to say whether there actually exist language universals and which they are in the end. The arguments 50 51

N. Chomsky, Problems of Knowledge and Freedom, p. 42. N. Chomsky, Problems of Knowledge and Freedom, p. 26.

342

JACQUES BOUVERESSE

of the anti-universalists are always the same: show us a single universal whose existence is truly established, and we could then start to discuss seriously. To believe that the question may be decided upon only and completely by an empirical investigation of the diverse existing languages, however, is completely mistaken. The disagreement is a very deep theoretical disagreement. The universalists and the anti-universalists do not hold the same opinions, in fact, about what constitutes the task of linguistics and about what has to be considered an important linguistic problem. Some linguists consider that their work must be confined to describing different languages, one after the other and as precisely as possible. Others think that these descriptions must be mainly conceived in function of the contribution they can make to the constitution of a general theory of language. As Bach and Harms rightly state in the preface of their volume,52 the question of the existence of language universals is neither more nor less than the question of the possibility of a linguistic theory: "What is linguistic theory itself if not the attempt to discover what is common to all languages, what is essential in the notion 'natural language', what are the limits within which languages can vary, what are the (universal!) terms by means of which the variation can be described?" Considering the matter naively, one is tempted to say that the hypothesis that there exist language universals, formal and substantial, is a natural and reasonable one because, after all, all men are organisms constituted nearly identically and exposed to an environment which does not vary infinitely. It would be totally plausible that certain universal properties of language result from properties common to the species, in particular, from basic characteristics of the human central nervous system. On the other hand, this hypothesis is indispensable from a scientific point of view, at least as a regulatory principle, because one of the basic principles of scientific research is not to be impressed by the differences, but to postulate the maximum unity compatible with the diversity of the observed phenomena. The universalist hypothesis is not totally incompatible with the affirmation that "languages can differ from each other without limit and in unpredictable ways", if this is interpreted as having reference only to surface phenomena. The anti-universalists give proof of a strange bad faith when they ask that language universals be 'shown' to them, because it is evident that the universal properties under discussion cannot be exhibited in the sense they understand: one cannot let them touch them with their fingers, because they are not observable properties, but constructed properties, properties which only appear at a sufficiently deep level. People like Bach, who formulate very strong universalist hypotheses, are naturally led to introduce deep structures which differ much more from the surface structures than was generally conceived. The reservations concerning universals and the reservations concerning deep structure go together naturally. In a way, the fact that there exist a probably very limited number of deep structure forms and transformational mechanisms evidently supports the universalist thesis. In another way, however, one can also 52

E. Bach and R. T. Harms (eds.), Universals in Linguistic Theory (New York, 1968), p. VI.

JACQUES BOUVERESSE

343

say that it is the universality principle, the idea that there must be a common form of a relatively restrictive kind underlying the extreme empirical diversity of the languages, which leads naturally to the idea of a deep structure. There may, of course, be universals which have reference to surface structure, but I do not suppose that they pose most of the problems. Martinet accuses the people who talk about 'universals' of having only an eye for identities and of wanting to juggle the differences away, but there is no reason why this should necessarily be the case. In reality it is the recognition of identities which allows putting the real and significant differences into evidence in the end. One could, under these conditions, say that the divergences among the universalists begin when a certain content has to be given to universal grammar. Do you agree with the distinction at that level of the three following possibilities: one can say that what is universal is the totality of the language functions; or one can say that what is universal are the formal constraints imposed on a grammar; or, again, and this is the most radical position, a universal characteristic may be attributed to substantial contents, for example to semantic primitives, etc. - I believe, indeed, that these three types of universals have to be distinguished. They are evidently not of equal strength, and they are relatively independent of each other. For Halliday, for example, the nature of language is closely dependent on what is asked of language, of the functions it must fulfill: "The particular form taken by the grammatical system of language is closely related to the social and personal needs that language is required to serve."53 Naturally, on the most concrete level, the language functions are characteristic of a certain culture: they are linked with a certain way of thinking and of living, with certain habits and institutions. But the indefinite multiplicity of the possible language usages are organized around a limited number of abstract functions or of 'metafunctions', which are in fact inherent to all language use: the ideational function, the interpersonal function, and the textual function. These basic functions are determined extrinsically, if you like, but, for Halliday, they have an immediate linguistic pertinence: they are reflected more or less directly in the sentence structure. There are, then, linguistic universals which are, at the same time, functional and structural universals, but the fact that the ideational function, for example, should be universal, with the consequences this entails for the nature of language, does not imply by itself the universality of the ideational content: it remains possible and even probable that each language organizes and articulates the exterior and interior experience of the members of the linguistic group considered in a totally different way. The universal language properties in which Chomsky is interested are of a totally different kind. They are, as we have seen, intrinsic properties without a plausible functional justification, which directly reflect the innate properties of the human 63 M. Halliday, "Language Structure and Language Function", in J. Lyons (ed.), New Horizons in Liguinstics (Harmondsworth, 1970), p. 142.

344

JACQUES BOUVERESSE

mind. There again, however, I suppose it is possible to be convinced of the exactness of the transformational syntactic analysis and to admit that there exist formal syntactic universals - perhaps even substantial - while remaining a decided relativist in semantic matters. It is evidently the hypothesis of the universality of the semantic content which raises most of the problems, because of a classical idea that can be found in people like Humboldt, Cassirer, or Wittgenstein, and in its extreme forms, in Sapir and Whorf : the idea that every language contains a specific conceptual system, a particular kind of organization of the experience, that it determines a certain 'world view' or a certain 'metaphysics', and that, in the extreme, the different cultural and linguistic universes may be more or less impenetrable or incommensurable. According to what I have recently read, the linguists and anthropologists have now rather mitigated their attitudes with regard to the principle of linguistic relativity. Most of them do not accept the Sapir-Whorf hypothesis in its extreme form and are partisans of a moderate relativism, which does not exclude the universality of certain basic characteristics. There are even people like Max Black who doubt that the relativist thesis has a clear and well defined meaning,54 but this thesis continues to have, in its most radical form, a considerable seduction and influence on the best minds. Martinet, for example, thinks that the linguist must be ready to admit its consequences in the field of syntax as well as in that of semantics: "...We should never assume that what differentiates one language from another is essentially and basically a different choice of the formal means of expression, but rather the type of analysis of experience it represents and the type of relationship prevailing among the corresponding linguistic articuli. In other words, whatever has to be retained from the Whorfian hypothesis applies to syntax as well as to other aspects of linguistic structure."55 In this perspective, what should be thought of the Quinean thesis of the 'indétermination of radical translation1 ? - I myself interpret the Quinean demonstration of the indétermination of radical translation mainly as a parable meant to illustrate the fallacious character of certain notions which we have already discussed. First of all, Quine wants to destroy the traditional mentalistic myth of meaning, but the case studied by Quine is, of course, totally theoretical (at least, so I assume); and if it is true that his considerations are above all directed against a certain universalist conception of meaning considered as an autonomous translinguistic entity, it nevertheless cannot be said that they would actually go in the direction of relativism understood in the Sapir-Whorfian way. For Quine shows that when we have established a semantic correlation between a language which was totally unknown to us at the beginning and ours, we cannot 54

M. Black, "Linguistic Relativity: The Views of Benjamin Lee Whorf", The Philosophical Review 68 (1959), 228-238, and "Some Trouble with Whorfianism", in S. Hook (ed.), Language and Philosophy (New York, 1969), 30-35. 55 A. Martinet, "Elements of a Functional Syntax", Word 16 (1960), p. 10.

JACQUES BOUVERESSE

345

know whether the success of our analytical hypotheses of translation is due to sufficiently close real resemblances between the two languages or, on the contrary, to the particular system of analytical hypotheses adopted - the choice of which may have been determined largely by ethnocentric considerations. It follows that, if I understand Quine well, he has rather established that we do not have the means at our disposal which would make it possible to test the Sapir-Whorf hypothesis efficiently, precisely because the conceptual system of a linguistic community is too considerably underdetermined by the language (or rather the linguistic behavior) on the basis of which we try to reconstruct it.

III. O N DICHOTOMIES

Looking at the panorama of contemporary linguistics, one notices that linguistic theories, whatever they may be, have always had a need for dichotomies: the distinction between langue (language) and parole (speech) has become Chomsky''s distinction between competence and performance. Do you believe that one can also talk about'competence' for other behavioral systems, or must this notion of competence be reserved solely for the faculty of language ? - Each time an individual is engaged in what Searle calls a "form of rule-governed behavior", he manifests a certain competence of which the explicit or tacit knowledge of the rules is only a part. The mastery of the underlying rules confers on this form of behavior its regular and systematic character.lt is uncontestable that the intervention of the faculty of language introduces between man and the animal a difference which, as Chomsky says, is not merely a matter of degree of complexity but of quality of complexity. The fact that this difference is manifested in a privileged and spectacular way in language mastery, however, does not by itself imply that the faculty of language should be what distinguishes the human mind from animal intelligence. In addition to linguistic behavior, man is certainly capable of a quantity of behavioral forms governed by rules which are qualitatively different from the most complex forms of animal behavior. Characteristic of the Chomskyan conception of language is the consideration that the faculty of language is not only a distinctive attribute of the human species but that it also constitutes somehow the essence itself of the human mind, that a description of language mastery in its most abstract and its most specific aspects constitutes at the same time a description of the most fundamental constitutive properties of the mind. The infinity of language and the creative aspect of its use evidently play a crucial role in this matter, but it could be maintained, at least in theory, that rule-governed creativity and the potential infinity also belong to other forms of human behavior and that it is this which distinguishes man from the animal in a general way. The competence of a chess player, for example, makes possible a kind of creativity which is potentially unlimited and which is, in certain respects, as impressive as that manifested in language

346

JACQUES BOUVERESSE

use. The same is true in a large number of institutional forms of behavior. He who masters the rules of social etiquette, for example, has a certain 'competence' at his disposal of which he can make coherent and adapted usage in infinitely diverse situations which are more or less unique. Evidently, the resemblance which exists between the two cases I just mentioned and language should not be exaggerated. Yet, it is incontestable that the specific features of the human mind are probably manifested as clearly in mathematical activity, play, or social relations, as in language. The question remains, of course, whether all forms of behavior which are specifically human are not, in the end, tributary to the faculty of language. Nevertheless, in the present state of affairs, it is for Chomsky clearly the study of the linguistic competence and more specifically, as we have seen, the syntactic competence of man which constitutes the privileged access to a better knowledge of the structure and functioning of the human mind. A t the end of the dialogue you had with him, and in many other places in his work, he expresses the hope that the progress of linguistic theory will decisively establish the absolute specificity of human intelligence and the unique nature of man. This attitude has always seemed strange to me because I do not understand well what Chomsky hopes from the science of language in this matter. The qualitative difference which exists between human intelligence and animal intelligence seems to me sufficiently proven, by language and by many other things. Even if linguistic theory were to succeed in confirming it some time, in giving it a precise scientific content (for example, by effectively describing the mechanisms which are responsible for it), it would not for all that transform this qualitative difference into an ethical or axiological superiority. That, nevertheless, seems to be altogether Chomsky's problem. In short, if you allow me to use Pascal's language here, I would say that a (scientific) theory of language can never determine the 'infinite distance' between the bodies and the minds (for those who want to consider things this way), but merely the finite distance between bodies and other bodies, between certain biological systems and others. Chomsky considers indeed that linguistic competence is first of all syntactic competence, but can one talk about a semantic competence next to this syntactic competence? Would this make sense, in your opinion ? - Yes, 1 think one could talk about a semantic competence. It would, in the primitive version of transformational semantics, be that part of linguistic competence which allows the speaker to assign a semantic interpretation to an infinite set of sentences on the basis of their deep structure phrase markers and of what Katz and Fodor call a semantic theory of the language concerned, constituted by a dictionary and projection rules. In Katz and Fodor's presentation, this competence is purely interpretative, not generative. The generative semanticists now have a totally different conception of the relations existing between the syntactic component and the semantic component of a grammar, and it is certainly not possible, from their point of view,

JACQUES BOUVERESSE

347

to talk about a more or less separate syntactic component and a semantic component. Anyway, the important point is that the possibility of assigning structural descriptions and semantic interpretations to an infinite set of sentences, or, if one prefers to look at things this way, of assigning syntactic interpretations to an infinite set of semantic structures, is only a part, although essential, of the faculty to speak and understand a language. In order to actually use a language, one must have available quite a few other things next to the mastery of the syntax and semantics of that language as it may be defined by a complete transformational grammar (including the semantics in one form or another). As Wittgenstein says in his Tractatus, "the tacit conventions on which the understanding of everyday language depends are enormously complicated". 56 The knowledge of the grammar and of the semantics, in the true linguistic meaning of these two terms, represents only a small part of these tacit conventions. This is a point to which Wittgenstein constantly returned. To understand a language is always understanding much more than what the linguist calls a language; and to understand a language is understanding, or more exactly, being able to participate in, a certain "form of life". The whole question is exactly what, in the total competence of the language user, is a genuine linguistic competence and what is not - what linguistics must describe here and what it must leave to other disciplines such as psychology, sociology, anthropology, etc., to describe, if possible. It is, nevertheless, the crucial question for Chomsky to see a relationship between the different kinds of competence. This is probably the reason why criticisms have been formulated in recent years with regard to Chomskyan competence and why proposals have been made to replace this competence by an "enlarged competence", by what is sometimes called the "communicative competence" proposed by Dell Hymes in the United States and by Habermas57 and some other German philosophers, and which would be a more than linguistic competence. - Yes, I think it was perfectly normal and in a way inevitable that this enlargment should be proposed. This does not prevent me from believing that Chomsky's position, understood correctly, is pertectly legitimate in itself. It consists, in fact, of the isolation of a relatively important aspect of this competence which can be described in formal and rigorous terms - a sound scientific strategy. There is no intrinsic inconvenience in proceeding in this way. This evidently leaves the question open as to how the Chomskyan linguistic competence interferes with other competences, extralinguistic for an important part, in the production of the performance we observe. The speaker's most important linguistic ability (but it is not only 'linguistic' in the narrow sense), from the 'Cartesian' point of view which Chomsky shares, is the ability to produce and understand an infinite set of sentences which are not only grammatical, 56

L. Wittgenstein, Tractatus Logico-Philosophicus (1921), 4.002. J. Habermas, "Vorbereitende Bemerkungen zu einer Theorie der kommunikativen Kompetenz", in J. Habermas and N. Luhmann (eds.), Theorie der Gesellschaft oder Sozialtechnologie (Frankfurt, 1971), 101-141. 57

348

JACQUES BOUVERESSE

but which are also appropriate to the verbal and situational context. This very rema rkable faculty could be called communicative competence, although, from Chomsky 's point of view, it would rather have to be called expressive competence. A large part of this competence is composed once more of extralinguistic elements. It is particularly on those that Wittgenstein insists. He is of the opinion that a good deal of our philosophical difficulties - 1 insist on 'philosophical', I do not say 'linguistic' originate from the fact that we isolate imprudently the strictly linguistic aspects of language use, that we adhere too much to the "exterior form" of the utterances, that is, to what is suggested by the linguistic grammar and semantics, and not enough to the actual use, insofar as this is conditioned by very diverse factors which are largely extralinguistic. Another criticism which has been made against Chomsky's notion of competence refers to the fact that a distinction should probably be made between the speaker's competence and that of the listener. Chomsky's competence would only be a totally abstract representation which fails to take into account this important distinction. Do you think such a distinction would be valuable ? - Chomsky succeeds in neutralizing the asymmetry which exists between the speaker and the listener in a communicative situation, precisely because in his description he parenthesizes the context and the communicative function of language. Unfortunately, I do not know precisely the criticisms to which you allude, but it seems to me that the asymmetry is obvious from the moment that the respective positions of the partners in the linguistic exchange are taken into consideration. As Chafe explains in his dialogue, the essentially creative process of constructing well-formed sentences and of encoding which takes place on the side of the sender is not at all the same as the essentially interpretative process of decoding and reconstructing which takes place on the side of the receiver.58 I think that Chafe is right in underlining that certain sentence properties such as well-formedness or ambiguity have a directional aspect which is not secondary but essential. It is true that one can always procure a concept of linguistic competence sufficiently narrow for this aspect to find itself rejected in the performance theory.... Let us pass to the other term of the dichotomy, to that which raises the most difficulties - namely, what exactly performance is. One gets the feeling, at least in the early works of Chomsky, that performance is never defined, or that it is defined negatively in the sense that everything that is not competence is performance. On the other hand, Chomsky sometimes talks, for example in Aspects of the Theory of Syntax, of a performance theory. I would like to ask you now if you do not see a contradiction there: one does not define what performance is and, nevertheless, one talks about an alleged performance theory. 58

Cf. pp. 23-24 of this book.

JACQUES BOUVERESSE

349

- The notion of 'performance' has actually often been used to cover almost indistinctly all the untheorizable aspects of language mastery and use. Under these conditions it is, as you say, rather strange to talk about a 'theory' of performance. There has been an unfortunate tendency, in Chomskyan circles, to refer offhandedly to the study of performance everything one did not like or could not discuss in linguistics. As Halliday remarks, performance has often been treated as a sort of "ragbag including physiological side-effects, mental blocks, statistical properties of the system, subtle nuances of meaning and various other things all totally unrelated to each other". 59 McCawley is right in stressing60 that, in what falls under the concept of performance as used by Chomsky, there are a great number of things which do not belong to linguistic performance but represent diverse non-linguistic competences, logical competence, for example. In fact, in Chomsky, a theory of performance, if it were possible, would have to account for everything which had to be added and also, in a way, subtracted from ideal language mastery as represented by the possession of a generative grammar of the language concerned, in order to produce the linguistic performance we observe. The performance theory has, in fact, in Chomsky's mind, a clearly negative duty which consists of accounting on one hand for certain very general limitations which are imposed on the faculty of language as on all human faculties by the psychic and organic constitution of man, and on the other hand for the accidental difference which exists between the linguistic product we can observe and what this same product would be if it were the direct emanation and the exact reflection of the ideal linguistic competence. We have here two types of idealization which do not have the same strength, and two types of extrinsic, 'non-essential' factors which are not accidental or exterior to the description of competence in the same sense. For, after all, it could be rigorously maintained that the fact that an acceptable sentence can only allow a rather limited number of embeddings is actually a characteristic feature of linguistic competence and not something that refers essentially to the conditions of sentence production and reception - to performance. Yet, performance theory evidently includes an essential positive aspect also. It must account for the fact that the speaker does not only possess his language but uses it, and uses it as he does in situations where he does. In other words, it must account for all the elements that are combined with abstract competence to give the concrete linguistic product. Of those elements, once more, many are not linguistic; they are part of pragmatics and cause individual and collective factors of a psychic, social, cultural, or other nature to intervene. On the other hand, they sometimes correspond to forms of 'competence' or implicit knowledge: this is the case, for example, of the so-called 'communicative competence', the 'conversational postulates', etc. I do not know whether I have answered your question. I mean, more particularly, that the constitution of a total theory of performance (which might be a completely Utopian goal) would actually go through the constitution of theories of other diverse competences.... 59 60

Cf. p. 84 of this book. Cf. p. 263 of this book.

350

JACQUES BOUVERESSE

Yes. That's what I wanted to know. But can one say, then, that performance would be the object of psycholinguistics ? - It would be part of quite a few things other than linguistics - not only psycholinguistics, but also sociolinguistics, social and cultural anthropology, logic, etc. On the other hand, each time that one has found the means to theorize satisfactorily on a particular aspect of language use, it is very tempting and quite natural to want to integrate it with linguistics proper, with the description of language itself. This is what happens, for example, with the theory of 'speech acts' about which I assume we will have to talk again. Then, I will ask you a very brutal question in this respect: can a performance theory be conceived while ignoring competence altogether? Are these two totally independent theories, or must reference be made each time to the 'opposite term' when making the theory of one or the other ? - I have said before that 1 was rather sceptical about the interest of a competence theory which would not also in certain respects be a performance theory, which would not give any idea of the way in which this competence is retained and used. After all, as Lamb remarks, 61 "competence really means competence to perform". I could now add that I do not see very well how a performance theory could not also be somehow a competence theory, how one could theorize on what the speaker does without at the same time theorizing on what he can do. I have been extremely touched, when reading this series of dialogues, by the number of eminent linguists who think at present that the distinction drawn between competence and performance, after having played a positive and maybe indispensable role, has in the end become disastrous and could probably be abandoned without great inconvenience. Chomsky started from the idea that, when sentence use must be accounted for, we can say many interesting things about the role of grammatical information, and almost nothing on that of extra-grammatical information. But the fact that the only performance element which can now be described really satisfactorily is that which corresponds to grammatical competence does not imply that it should be the only nor a fortiori the most important. It is possible that in the end, as several linguists suggest, by insisting on the notion of linguistic competence, one would end by giving a strange and fairly deformed representation of what language actually is, and that one would disqualify or discourage a priori certain ventures which could be promising and for which the attempt can in no way be given up. It is regrettable, for example, that terms such as 'psycholinguistic' or 'sociolinguistic' are now being used by a certain number of linguists with a clearly pejorative connotation. Even if it is true that psycholinguistics, the results of which are certainly less spectacular and intellectually prestigious than those of transformational grammar, is still in a rather childish state, the state of affairs should really make no one happy. 61

Cf. p. 209 of this book.

JACQUES BOUVERESSE

351

What surprised me in the dialogues I read is exactly the vigorous reaction today in favor of a theory of language use. It seems that many people are now tired of studying language as a being or entity, if I may say, and that they are tempted to study language more as a doing, as action and behavior. In a way, the interest given to language as a system and as a structure, which has been dominant since Saussure and which generative grammar has carried to an extreme, is visibly decreasing. Some linguists and language philosophers realize that one has not really reached an understanding of the essence of language as long as it is not understood how language is used by those who possess it. This explains the attempts at rehabilitating the functional, instrumental, and various other considerations which were rather discredited by Chomsky. It was normal, in my opinion, that one should eventually get tired of considering language at rest, in vitro, autarchic and idle language, or language reserved only for noble and disinterested functions, and that linguists themselves should want to see it, so to speak, at work. If I was not afraid of preaching for my parish, I should say that with the decline of the 'aristocratic' conception of language, linguists tend to become or become again more and more Wittgensteinian - which is doubtlessly not surprising because I am convinced that Wittgenstein, as he states it, mostly says only what could not possibly not be conceded to him and to which one must therefore return sooner or later. In order not to risk that what I say be misunderstood, however, I should probably specify that it is in itself not at all incompatible to treat language as an autonomous system and as a means and an instrument serving well-determined goals. This is quite evident if one considers, for example, Lamb's case, who on one side professes a particularly radical form of structuralism, and on the other declares himself totally in favor of a functional approach of the kind proposed by people like Firth and Halliday. Lamb impresses me enormously by the serenity with which he rejects all the traditional or recent classical dichotomies: system/function, closedness/ dependence, immanence/extraneousness, competence/performance, rationalism/empiricism, etc. Concluding metaphorically, I would say that the philosopher, as I conceive him, does not have the same reasons as the linguist (or some linguists) for establishing a difference of nature or especially of dignity between 'intra muros' linguistics and 'peregrine' linguistics (annexed research, often not appreciated). Structural linguistics of Saussurean obedience, operating within the dichotomy of langue-parole, has the word as its privileged theoretical unit. This attitude can be opposed to Chomskyan linguistics which uses the distinction of competence-performance and which has the sentence as its unit. There are linguists - for example, Martinet and other structuralists -for whom the sentence itself already belongs to the field of parole. My question concerns the fact that the position of discourse with respect to the two antithetic terms is seldom mentioned when one works in the framework of these two dichotomies. What is, in your opinion, the position of discourse ? Should discourse be considered a part of langue or of parole, as relevant to competence or to performance ? - The problem of the exact position of discourse with respect to the two dichotomies

352

JACQUES BOUVERESSE

you mention constitutes, in my opinion, a direct incentive to reconsider them radically, to wonder whether it makes sense to classify language facts and linguistic research as a function of oppositions of this type. You said that the traditionally privileged linguistic units are the word for structuralist linguistics and the sentence for generative grammar. This is, in fact, rather striking. It is clear that Chomsky, for example, considers that he has finished his job, that he has given an adequate characterization of linguistic competence when he has accounted for the speaker's ability to produce and understand an infinite number of different sentences. The creativity in which he is interested is primarily sentential creativity, but it is clear that in certain respects, textual or discursive creativity, the faculty to produce an unlimited number of coherent sentence concatenations which are adapted to extremely diverse situations is much more spectacular and difficult to explain. Speaking is obviously not only being able to produce an infinite number of sentences on the basis of a finite number of words and constructions, it is also being able to produce an infinite number of texts on the basis of a practically finite number of sentences. For Chomsky, however, textual creativity is not part of linguistic competence in the narrow sense; it belongs to the level of what he calls "the creative aspect of language use" and belongs, consequently, to performance theory. I would like to open a more specifically philosophic digression here. It is very remarkable that the philosophers of analytic or linguistic inspiration have, in the end, behaved like the majority of linguists. Ordinary language philosophy has the word as its privileged unit. When the philosophers of this school talk about ordinary language use, it is mainly about the normal use of words and of expressions in sentences. For Ryle, for example, words have uses, that is possible roles in the constructions of sentences; but, properly speaking, sentences do not have a use. American-style analytic philosophy, that is, that which derives rather directly from logical neo-positivism, seems to have taken a greater interest, in its criticism of traditional philosophy, in sentences or propositions. It is very meaningful that, when they tried to realize their program of the elimination of metaphysics, the neo-positivists concentrated almost exclusively on the notion of metaphysical proposition, to the detriment of metaphysical text. One cannot find in their work any serious attempt to criticize philosophy, as it is expressed no longer in metaphysical assertions of the type "God exists" or "The exterior world does not exist", but in philosophical texts. We must now return, however, to the linguistic problem of the text as specific unit. It is evident that what has delayed the constitution of discourse linguistics is essentially the pre-eminence of syntactic considerations, and the ignorance and contempt in which semantics has been held for so long - for understandable reasons. It is, on the whole, normal that the syntactician does not feel any need to "venture", as Martinet would say, beyond the sentence. He may consider that his obligations and also his means stop at this level. If one takes only the syntactic point of view there are reasons to say, as Martinet does, that "there is nothing beyond the sentence

JACQUES BOUVERESSE

353

which is not already within the sentence".62 The concept of text is a semantic concept and, even more exactly, a pragmatic concept. As Halliday writes, "the basic unit of language in use is not a word or a sentence but a 'text'; and the 'textual' component in language is the set of options by means of which a speaker or writer is enabled to create texts - to use language in a way that is relevant to the context." 63 What Halliday calls the textual function is what allows a speaker to produce discourse fragments which are coherent and appropriate to the situation, and a hearer to distinguish a text from a set of sentences produced haphazardly. There is no reason to believe that the study of this function does not belong to linguistics proper, especially if one considers with Halliday that it has direct influence on sentence structure itself. There is also no reason to throw it back pure and simple to the performance theory and not to consider that the textual function sets a specific competence to work which could be called the 'textual' or 'communicative' competence, and which more particularly includes - among a mass of other things - the faculty to perceive logical relations between sentences. Among the linguists you have interviewed, some clearly feel the need and the urgency of a discourse linguistics. They are, of course, all those who are particularly preoccupied by the constitution of a theory of language in use and in action, of utterance or speech act linguistics. All this is rather directly related to the generative semantics reaction against the Chomskyan model. - Yes, certainly. Those linguists who react very favorably to the idea of discourse linguistics are also the linguists who are little disposed to maintain a strict distinction not only between syntax and semantics but also between semantics and pragmatics. The enormous difficulty a text theory faces a priori derives mainly from the number and the complexity of the semantic and also the pragmatic elements it would have to include. Martinet thinks that the study of discourse does not belong to the field of linguistics proper, but he believes almost the same about semantics itself.64 The fact that the elements of discourse theory available at present are extremely limited and that the results obtained in the fields of phonology and syntax are in any case much more convincing does not at all mean that, in the study of the textual function, we have to deal with an unessential aspect in which one had better be completely disinterested for the time being. What is true is that, as Martinet himself says, one cannot do everything at the same time. Our analysis of the notion of text here is evidently inchoative, but I would like to ask you if one has to define text as context - as context of the sentence, or how would you prefer to define text ? Are there limits to this notion ? Should a typology of texts be made ? - T do not know at all how the notion of text should be defined nor how the text, 62 63 64

Cf. p. 232 of this book. M. Halliday, "Language Structure and Language Function", pp. 160-161. Cf. p. 230 of this book.

354

JACQUES BOUVERESSE

as a unit of discourse, should be delineated, and I really do not have anything interesting to say on this matter. I just think that Halliday is right in stressing that the text is a specifically semantic unit and that it is quite different from a sort of supersentence. Again, the need for a text theory only means for me the need for a theory of effective language use, and it follows particularly from the fact that, in normal linguistic behavior, the speaker does not merely use isolated sentences but also organized sequences of sentences. Of course, it is not a question of 'length'; a text may very well be limited to (the utterance of) a single sentence; and every discourse act, however elementary it may be, operates the textual function of language, as Halliday conceives it. It is clear that the text is not at all identical to a fixed succession of sentences: the same arrangement of sentences may be used by different people, in different contexts, with different intentions, and consequently serve as support for different texts. Being interested in text is, as Halliday says, being interested "in what people actually do and mean and say, in real situations".65 That is why I do not know whether I would permit talking, as Chomsky does, about an ideal discourse, constituted by a succession of sentences used to support the succession of utterances which form the real discourse, and considered uniquely from the point of view of grammar, that is, of competence. In my opinion, a fixed arrangement of sentences does not at all constitute discourse, even 'ideal'; a sentence is not an 'ideal' utterance. The relationship of discourse to the language elements that are its vehicle cannot be discussed in terms of an opposition between real/ideal or concrete/abstract: it is defined by what is called (in necessarily vague terms) "to work", "utilization", "use", etc. In conclusion to this part of our conversation, we must take up the very central question which probably determines the whole future of philosophy of language as well as of linguistics - namely that which has reference to language functions. It has evidently been discussed for a very long time. With regard to structural linguistics, for example, Jakobson has built an entire theory of language functions. It is also discussed in psycholinguistics - I think here of Biihler's theory, which is very classical. One has the impression while reading these dialogues with linguists, however, that for most of them there is either a certain hypostasis of the communicative function - and this especially for the sociolinguistically oriented - or a hypostasis of the expressive function, in particular for the psycholinguistically oriented of whom Chomsky is certainly the most typical example. I would, then, like to ask you the following question: is there not, at the basis of this whole discussion - for example Chomsky versus Searle - an enormous misunderstanding ? - Yes, incontestably. I do not know whether I am a very meaningful example, but I myself have never perceived any real incompatibility between the results of transformational grammar and those of the theory of speech acts; and I have always accepted with equal favor what came from one side as well as from the other. In 68

Cf. p. 86 of this book.

JACQUES BOUVERESSB

355

fact, the publication of Speech Acts in 1969 represented for me almost as important an event as that of Syntactic Structures in 1957. The major interest of the theory of speech acts, from a philosophical point of view, is that it provides the means to reconcile a Chomskyan approach to language with an approach that can be called more 'Wittgensteinian'. It is situated somehow on the intersection of language theory and the theory of behavioral forms: it strives to treat language simultaneously as knowledge and as action, as knowing that and knowing how (to do). As it is conceived by Searle, it integrates without difficulty all the main acquisitions of transformational grammar and, according to what he says in the conversation you had with him, Chomsky seems to think that the inverse is partly true as well.66 This being said, the distance existing between Chomsky and a resolutely functional approach is enormous at first sight. I have already, I believe, quoted Halliday's statement: "The particular form taken by the grammatical system of language is closely related to the social and personal needs that language is required to serve."67 Halliday says in his text: "The grammatical functions, in the sense of roles, are derivable from the extrinsic functions of language."68 Yet, Chomsky considers that wanting to deduce the formal properties of grammars from functional necessities, in particular from communicative requirements, is a rather ridiculous enterprise. Two things must be observed here to start with. On one side, as I already had occasion to mention, Chomsky's position is largely determined by a theoretical and philosophical stand which consists of considering 'essential' and really meaningful only those language properties which do not have any evident relation to the functions language must fulfill and the needs it must satisfy. Such being the case, in the extreme, the statement that no fundamental language property proceeds clearly from functional constraints runs the risk of being rather tautological. On the other hand, Chomsky evidently does not state that no formal language properties brought to light by transformational grammar, for example, have a communicative function. He merely says that we should not hope to deduce the characteristics of language from communicative requirements, but that incontestably, once we have brought to light certain language characteristics independently, we can try to find out whether they have a communicative function. 69 So that, for someone like me, who views things in an abstract way, it is not a priori unthinkable that, started from opposed points, Chomsky's approach and the functional approach meet in certain cases. If I understand Chomsky well, however, he means that we can question the advantage of transformations from the viewpoint of language efficiency as a communicative tool, but that we would never have arrived at transformations if we had originally started from functional considerations of that kind. Strange as it may seem, there is a rather remarkable analogy between Chomsky and Wittgenstein on the problem of the possible relations between form and function 66 67 68 69

Cf. pp. 44-45 of this book. M. Halliday, "Language Structure and Language Function", p. 142. Cf. p. 94 of this book. Cf. p. 52 of this book.

356

JACQUES BOUVERESSE

or between structure and usage. Both think, in the end, that there are none, or at least that there are few interesting things to say about it. Chomsky thinks there is no evident and essential link between the function of linguistic expressions and their structural properties - that we should not try to proceed from the first to the second. Wittgenstein views things similarly, but inversely: for him, the formal structure of the expressions and the characteristics the linguist and the logician might attribute to them generally do not give us any idea, or only a very partial and misleading idea, of their real use, or rather of the extreme diversity of their functions and their utilizations. This is essentially what causes the philosopher's troubles. One can deplore, as Chomsky does, that Wittgenstein has so little sensitivity to the systematic aspects of language mastery and use, but it is no less regrettable, in my opinion, that Chomsky has so little interest in the fact that language is used, if one may say, to 'do things', such complicated and different things. Is it, after all, possible to speak of a primary or privileged language function ? - Although most linguists you have interviewed, including Chomsky, have marked orientations and preferences in practice, they state that in theory they would refuse to say that one of the language functions is 'essential' or first with regard to the others. The discussion about the exact role of the communicative function does not bear on the question whether communication is one of the important language functions, which nobody contests, but on the question whether all language functions and, what is more, its most characteristic syntactic and semantic properties should not be conceived mainly in relation to the communicative function - whether one can, at any time, when one is studying language, completely lose sight of the fact that it functions as a communicative instrument. In your conversation with Chomsky, he says rather prudently: "The function of language for expression of thought is not 'opposed' to its communicative function; rather, it is presupposed by the use of language for the special purposes of communication."70 This way of looking at things implies a conception of the communicative function which, although very widespread, is not any less peculiar: if one perceives the communicative phenomenon mainly under the aspect of an "exchange of ideas", one can say that the possibility of communicating presupposes the possibility of expressing ideas and that, consequently, the expressive function is in a way more basic than the communicative function. Husserl formulates the same conclusion in another language: if one considers how discourse is used in "the solitary psychic life", one realizes that what is essential to an expression, what causes an expression to be an expression, is the fact that it has meaning, that it expresses something, and not the fact that it has a manifesting function (kundgebende Funktion) and that it functions as an index in a communicative discourse.71 Some people maintain, on the contrary, that the main usage of language 70

Cf. p. 52 of this book. E. Husserl, Logical Investigations (Logische Untersuchungen) (Halle, 1913), First Logical Investigation, Chapter 1. 71

JACQUES BOUVERESSE

357

is its interpersonal use, and that its usage in solitary discourse is actually secondary and derived, or even degenerate. For either one, the absence of a listener in the soliloquy is actually rather subordinate; it does not basically affect the nature of the phenomenon, but for inverse reasons: for the first, because they think that what is essential in language use is the expression of ideas, whether or not it is for somebody else; for the second, because they believe that all language use is communicative to some degree, that the soliloquy is still basically a communicative act. Chomsky has frequently said or suggested that he does not consider the communicative function really essential to language. In your conversation with him, he alludes to the case of somebody who would spend years writing a book without worrying excessively whether anybody would read or understand it and observes that it would be absurd to suppose that he does not use language in a normal way. But I do not see how the (possible) absence of a receiver of the message prevents the consideration of this message basically as a communicative act, as well as an expressive act. In my opinion, a book, even without any reader, is always a complex discourse act which obviously has a communicative dimension. Taking the most disadvantageous example possible, I would certainly not say that the "expressionistic" poets do not, in the end, try to communicate something. And it generally does not seem to me that the absence of an explicit intention of communication would suffice to prevent the accomplishment of a communicative act. Martinet asserts in his interview that "an instrument that would serve only to help people think would never be a language."72 Yet, given that one thinks essentially in a language and mostly by dint of the same words one would use to express his thoughts, how, then, could an instrument which is mainly used for thinking not be basically of the kind of a language ? This observation is, nevertheless, not meant as a criticism of Martinet. I mean, on the contrary, that he is basically right in the sense that I myself cannot possibly conceive maybe I have a particularly poor imagination - how an instrument which would help efficiently in the formation and clarification of ideas could not be just as essentially an instrument which serves or has served to transmit them. On the whole, I think it is philosophically less dangerous to overestimate the importance of the communicative function than that of the expressive function. What is the origin of your reservations with regard to the conception that language primarily expresses 'thoughts' or 'ideas' ? - This conception is wrong in suggesting that the ideational process is somehow independent of and prior to the verbalization process, and it is largely erroneous to believe that things are so. Those who think that the essential function of language is the expression of ideas often talk as if the speaker should first form ideas and should then 'choose' the appropriate words to express them. Considering things this way is treating language merely as a kind of code and underestimating clearly the importance of the linguistic sign. Considered as really important in language here is the interior 72

Cf. p. 235 of this book.

358

JACQUES BOUVERESSE

discourse, the 'ideas'; if one had direct access to the ideas, one would not need signs. Wittgenstein has repeatedly criticized this point of view, but I will restrict myself to quoting here what Lamb says in his dialogue: "I think some linguists take the point of view that notation is relatively unimportant, that what really counts is the thoughts and ideas, and the means of expressing them is relatively insignificant. I disagree with that point of view. I think that notation is very important, because, try as one will to avoid it, one thinks in terms of the notation that one uses."73 This observation of Lamb's evidently concerns the linguists' language - the notation they use to express their conceptions. It can, however, easily be extended to language as a whole: language is not a notational system for independent thoughts; it is not only a vehicle, but also an instrument and a means of production for thought. As Lamb also says, "people often suppose that the thought is somehow formed independently and is then just encoded into the grammar; but actually the thought is formed only with the aid of the grammatical structure.'"74 Another major inconvenience of the excessive privilege sometimes granted to the expressive function is that it usually leads to a clear devaluation of the social aspect of language use. The 'essence' of language is expressed most purely in solitary discourse, free from the communicative constraints. As I have already stressed, the insistence on the expressive function of language is closely linked to the pathos of the developments on linguistic creativity. It is clear that in communication language use is subject to practical imperatives which may give the impression of considerably restraining its creative aspect proper. The idea that language is mainly a transmitter of the speaker's ideas, representations, and personal sensations inevitably leads to the famous question of the possibility of a private language which was raised by Wittgenstein and has been discussed frequently since then. Since, by definition, I am the only one who has direct access to my own ideas, whereas you only have direct access to my words and, only indirectly, through them to my ideas, what proof can we have that the words you and I use, in particular the words directly designating inner sensations and representations, have the same meaning ? With respect to the criticism on the Cartesian conception of the mind by people like Peirce and Wittgenstein (I would rather not call up Ryle's name here), it could be said that the conception according to which the essence of language consists of the expression of ideas, understood in the sense of representations of the mind, has this time a clearly 'Cartesian' resonance in the bad sense of the term. In the end, as paradoxical as it may seem, the primacy granted to the expressive function induces excessive privilege for the communicative function understood in the narrow sense of the exchange of ideas. Communicating is essentially expressing ideas for someone, formulating a question is communicating a state of ignorance and the wish to be informed, giving a command is communicating the wish to see a certain action accomplished, etc. Looking at things this way is, in the end, viewing all 73 74

Cf. p. 214 of this book. Cf. p. 214 of this book.

JACQUES BOUVERESSE

359

speech acts which may be accomplished with language elements on the model of the exchange of information. I think it would not be scandalous to say that one of the weaknesses of the Chomskyan conception is its exaggerated insistence on the communicative process in the narrow sense and its insufficient accounting for the extreme diversity of the speech acts through which the speakers come to be interrelated in the interpersonal linguistic practice. I wish you would make Wittgenstein''s doctrine on the multiplicity of language functions a little more explicit. Does Wittgenstein really not try to establish a hierarchy among them ? - No, Wittgenstein was very hostile to the idea that there could be a hierarchy among the language functions. He did not believe that it would be possible to attribute to language a privileged function or a unique function, a fixed function which would be the function of language in all its uses. He has blamed himself for having excessively privileged, in the Tractatus, one of the language functions: the function of representation of facts, language as a picture of reality; and he considered that the philosophers always have a conception of language which is too unitary, too general and eventually too simplistic. It is disconcerting in Wittgenstein that no general idea on language can finally be attributed to him, precisely because he thought that all general ideas on language and on language functioning are in themselves philosophically suspect. Wittgenstein is someone who always insists on the differences, because it is always the differences that are neglected, and it is always soon enough and rather tempting to look for analogies and formulate generalities. Wittgenstein always proceeds from examples, and he is not at all interested in questions of typology, hierarchization, and systematization. He refuses, for example, to define what a 'language game' is in a way other than by giving examples of language games; and he stresses that new language games appear constantly whereas some others fall into disuse, that the language games we practice are not the only possible ones and that one can easily invent fictitious language games which are more or less plausible or implausible. He himself excels in this exercise. The universe of language games is just as diversified and open as that of manners, customs, practices and institutions, forms of life and behavior. Although Wittgenstein had no general idea in principle nor any theories on language his approach is incontestably functional and instrumental: and this is exactly what prohibits him from talking about an 'essence' or finality of language: it is impossible to say a priori in what social and cultural context, in what way, for what purpose, and with what results language can be used by human beings. Although I do not at all agree, in general, with what Chomsky says about Wittgenstein, there is, nevertheless, a basic point on which their conceptions are diametrically opposed. Chomsky always tends to consider language as extraordinary and unique, to attribute to it an eminent and exceptional position. On the contrary, Wittgenstein constantly insists on the fact that language is very ordinary - something which is part of the "natural history" of man on the same grounds as many other less remarkable things.

360

JACQUES BOUVERESSE

Wittgenstein has, since then, not stopped denouncing the "sublime" conception of language and of logic which has been the inspiration for his oracular and definitive proposals in the Tractatus. "We are under the illusion that what is peculiar, profound, essential, in our investigation, resides in its trying to grasp the incomparable essence of language": 75 I do not know whether there is a sentence which could more appropriately characterize the philosophic inspiration which has animated Chomsky's works, both from the start and thenceforth increasingly more so.... The illusion which Wittgenstein ascribes first of all to philosophers, who always tend to consider language as something mysterious and elusive, is that the important things in language are not what everybody can see and observe - those things which take place at the level of linguistic signs and their use - but those that happen out of sight in an inaccessible and ethereal domain. Transformational grammar represents an investigative tendency for which Wittgenstein had incontestably little sympathy as a philosopher: it has striven to exploit in the extreme the possibility of treating language as a formal system or a calculus. As Halliday observes, "the great thing Chomsky achieved was that he was the first to show that natural language could be brought within the scope of formalization; that you could in fact study natural language as a formal system."76 As it happens, though, each time an attempt at formalization succeeds, it is difficult to appreciate exactly the nature and scope of the result obtained and to resist the temptation to believe that one has accomplished much more than one actually has. Although in principle it is not presumed to represent a theory of sentence production and comprehension, the Chomskyan model of linguistic competence irresistibly suggests the idea that language functioning takes place somehow in the form of the execution of a calculus or perhaps the successive or simultaneous execution of several calculi - I am drawing a very rough scheme here - a sort of calculus of well-formed expressions (corresponding to the syntactic component of the grammar), of meaning (corresponding to the semantic component), and of sound (corresponding to the phonological component). For Wittgenstein, this method of representation inevitably leads to the postulation of the existence of a mechanism which does not resemble any other - the mental mechanism in which, as it were, the total calculus of language is present each time and comes into play. From Wittgenstein's point of view, there exists a direct link between a certain conception of language as a system and as a calculus, and mentalism - not the perfectly harmless mentalism of transformational grammar but the most suspicious form of traditional philosophical mentalism. This is why one often finds in Wittgenstein's work recommendations like the following: "If you are puzzled about the nature of thought, belief, knowledge, and the like, substitute for the thought the expression of the thought, etc."77 Yet Wittgenstein certainly does not mean, as Katz seems to believe,78 75 76 77 78

L. Wittgenstein, Philosophical Investigations, § 97. Cf. p. 84 of this book. L. Wittgenstein, The Blue and Brown Books (Oxford, 1958), p. 42. J. J. Katz, The Underlying Reality of Language, pp. 5-17.

JACQUES BOUVERESSE

361

that a language theory can be restricted to the description of linguistic 'appearance' - that is, the most superficial and tangible phenomena of language use. On the whole, I do not see what might have kept Wittgenstein from recognizing the enormous contribution made by Chomskyan linguistics to the scientific description of an (essential) aspect of the linguistic phenomenon, but he definitely sympathized with those who think, as Strawson says, that "as theorists, we know nothing of human language unless we understand human speech"J9 Consequently, he would certainly have thought in the end, even after Chomsky, that "as theorists, we know very little about human language".

IV. ON THEORIES OF MEANING

I propose that in this part of our conversation we concentrate more on the status of semantics and on the notion of meaning in linguistics. This is a very particular and extremely important subject, especially if one considers the latest developments in generative grammar where, with the generative semanticists, the questions of pure semantics have again become predominant, arousing more and more interest. As an introduction to the debate, I first want to ask a question concerning Strawson, who, in an almost classical article,80 introduced the notion of "essential grammar" by opposing essential grammar to the syntactic Chomskyan grammar. Do you agree with Strawson in reproaching contemporary linguistics, considered as a whole, for not having been sufficiently sensitive - at least until recently - to problems of meaning ? - The importance of that article by Strawson to which you allude is mainly derived from his attempt to show that the question whether the universal vocabulary of grammatical description, with its rather traditional categories, should or should not be characterized in resolutely semantic terms, is not, as Chomsky seems to believe, a current issue in linguistic theory but a preliminary question, as I have pointed out: it is certain that the foundations of a general grammatical theory must be sought above and beyond what Chomskyan linguistics proposes and perhaps in the direction suggested by Strawson. If I understand him well, Strawson basically considers that, in spite of its theoretical pretensions, transformational grammar might still be too empirical and too intimidated, not sufficiently ambitious from the explanatory point of view nor sufficiently universalist - the level at which it is located is primarily that of the "variable" grammar. Undoubtedly, this situation originates from the clearly empirical orientation of Chomskyan linguists and, as Strawson remarks, a language philosopher would less suspect the direct use of an openly logico-semantic vocabulary and the undertaking of what Strawson calls "non-empirical linguistic" research. The position adopted by Chomsky in Syntactic Structures, maintaining that semantic considerations do not play any role in the analysis of syntactic structure, has always given me 79 80

P. Strawson, "Meaning and Truth", in Logico-linguistic Papers, p. 189. P. Strawson, "Grammar and Philosophy", in Logico-linguistic Papers, pp. 130-148.

362

JACQUES BOUVERESSE

the impression of representing a kind of theoretical curiosity which is, besides, very interesting and fruitful. In reference to the Chomskyan attempt to construct an entirely a-semantic theory of grammatical structure, Jakobson has said that this complicated venture actually served as excellent negative evidence to the benefit of current research on the hierarchy of grammatical meanings.81 In fact, if one thinks about it, the reinstatement of semantic considerations and research in transformational grammar was entirely predictable and inevitable. What allows you to say that ? - The fact that I am not a linguist but a philosopher and somewhat of a logician. In my opinion, what has happened in transformational linguistics between 1957, the year of Syntactic Structures, and 1963, the year of the official reintroduction of semantics by Katz and Fodor, strangely resembles what happened in logic in the years 1930-35. Considering Carnap's attitude at the time of Logische Syntax der Sprache in 1934, one notices that he divided the entire study of language into two parts to which he attributed unequal importance and dignity: the genuine scientific study - formal syntax - and the rest, which is quite summarily referred to psychology. His rather contemptuous treatment of the logico-linguistic problem of meaning is quite comparable to Bloomfield's. What prevents one taking the theory of meaning seriously is the formalistic and physicalistic stand which is considered the indispensable prerequisite for all scientific approaches to language. Of course, the logical problem of meaning does not occupy an unimportant place in Logische Syntax der Sprache, but Carnap treats it only for demonstrating that it can be accomplished in exclusively syntactic terms, and that, therefore, a logic of meaning with a different object and method than that of formal logic can be discarded. Afterwards, as you know, Carnap, together with Tarski, has been one of the pioneers in the construction of a logical semantics, and he has attempted to redefine a certain number of basic concepts of the theory of formal systems and of the theory of language in semantic terms - whereas he had originally thought that the definition of such concepts must be a rigorously syntactic one. Considering the way logicians proceed today, what is striking is a certain preponderance of semantic considerations on the one hand, and on the other the fact that few people are disposed to maintain that there is a very strict and basic distinction between syntax and semantics, the difference being primarily a difference of method. What happens now in linguistics reminds me of the discussions held between 1935 and 1940 between people like Carnap, who believed in the existence of a definite distinction between formal (logical) truth and factual truth as well as between syntax and semantics, and those like Tarski and Quine, who did not really believe in it. Of course, I do not at all suggest that the problem of linguistic semantics is the same as that of logical semantics. What I mean is that in linguistics as in logic it is normal that, after a syntacticist and formalist period, there be a reaction in the semantic direction. 81 R. Jakobson, "La notion de signification grammaticale selon Boas", in Essais de linguistique générale, French translation (Paris, 1963), p. 204.

JACQUES BOUVERESSE

363

As far as I ara concerned - although this comparison is not likely to please Chomsky - 1 think there is a striking analogy between Carnap's position at the time of Logische Syntax der Sprache and that of Chomsky at the time of Syntactic Structures. In both authors, one can find the same concern for preserving the dichotomy between syntax and semantics and for preserving the autonomy of the first, and one finds the same undeceived scepticism with regard to the possibility of treating the problem of meaning satisfactorily: '"Meaning"', Chomsky writes, "tends to be used as a catchall term to include every aspect of language that we know very little about". 82 In a way, it could be said that with the spectacular success obtained in the syntactic field, thanks to transformational syntax, the linguists have, as we say in France, "mangé leur pain blanc le premier" (had their best time first). Sooner or later it will have to be admitted that language, and perhaps even the syntax of language, is not really understood as long as it is not understood what it means for language utterances to have a meaning and to have the meaning they have. Whatever may be the difficulties and the disappointments to which one is exposed, one cannot indefinitely approach semantics as an ostrich would. As Barbara Hall Partee writes, "although the basic notions of semantics are about as little understood by generative linguists now as they were ten years ago, there has been a major shift in the general attitude of linguists toward semantics.... It is difficult to point to anything that might be called a substantive semantic result, but the most important step has been taken : we have stopped playing ostrich about semantics."83 A detail question: do the selection restrictions in transformational theory bear on the semantic component or on the syntactic component? Or should restrictions of both types be formulated? - I am, of course, totally incompetent to answer this question. In "The Role of Semantics in a Grammar", 84 McCawley observed that surprisingly little attention had been paid to whether an adequate language theory requires only syntactic selection restrictions or only semantic selection restrictions or both. He stated, on his side, that selection restrictions must be formulated at the level of the semantic component according to the solution adopted by Katz and Fodor, and not in the form of rules belonging to the base. McCawley is of the opinion that the 'selectional features' should be defined only in terms of semantic representation properties and that the base component must produce deep structures without any consideration of whether the elements that occur in it violate selection restrictions. I do not know in what stage the question is now and, again, I am in a very bad position to talk about the genuine linguistic implications the solution to that matter may have. Yet I am 82

N. Chomsky, Syntactic Structures, pp. 103-104. B. Hall Partee, "On the Requirement that Transformations Preserve Meaning", in C. F. Fillmore and D. T. Langendoen (eds.), Studies in Linguistic Semantics (New York, 1971), p. 1. 84 J. McCawley, "The Role of Semantics in a Grammar", in E. Bach and R. T. Harms (eds.), Universals in Linguistic Theory, 124-162. 83

364

JACQUES BOUVERESSE

tempted to say a priori that McCawley's position is right. It is for the semantic component, and not for the syntactic component of the grammar, to exclude sentences like "this stone is sad" or "this triangle is virtuous" (Carnap). If one wants the syntactic component of the grammar to reflect the speaker's syntactic competence proper in a plausible way, one cannot ask that it prohibit the construction of these sentences, which certainly seem to be totally correct from a syntactic point of view. Or, if you want: it is not because they are excluded by the rules of syntax that they are meaningless - it is because they are meaningless that one can think of formulating the syntactic rules so that they are excluded. The logical empiricists at one time considered that, in order to be really adequate, the syntax of a natural language should not only exclude a sentence like "Caesar is and", but also a sentence like "Caesar is a prime number". "The fact that natural languages allow the formation of meaningless sequences of words", wrote Carnap for example, "without violating the rules of grammar, indicates that grammatical syntax is, from a logical point of view, inadequate. If grammatical syntax corresponded exactly to logical syntax, pseudostatements could not arise."85 The logical neopositivists' attempt to construct an ideal scientific language was precisely an attempt to obtain a language possessing sufficiently strict formation rules to prohibit the formulation of any meaningless utterance (in particular, of any metaphysical utterance) and sufficiently liberal to allow science to say all it has to say. In other words - in their very particular philosophical perspective - their objective was the construction of a language in which the criteria of grammatical correction would coincide with those of meaningfulness. This implies the recognition that the rules of grammatical (syntactic) correction must be formulated as a function of the rules of logico-semantic correction and not the opposite. Concerning our specific problem, namely, the status of selection restrictions, I do not see how one could let the syntactic component express them without its risking the usurpation or at least the useless repetition of something which is the province of semantics. With respect to Strawson's desire to constitute "essential grammar" as opposed to "variable grammar", I might express some criticism. The semantic vocabulary he proposes as the content of this essential grammar does not seem structured or systematized to me. In other words, the importance of structure is not very thematized. Have you had the same impression ? - Yes, in a way. You might be right. To actually pass judgment, one would need concrete examples of what Strawson proposes to construct under the name "essential grammar" of a given type of language. In fact, he provides only some purely pragmatic suggestions. He is satisfied with giving a few examples of terms that may appear in the three vocabularies required for the constitution of an essential grammar: the ontological vocabulary, the semantic vocabulary, and the functional vocabulary. His concern is a clear distinction among these three vocabularies from a fourth, that of 85

R. Carnap, "The Elimination of Metaphysics through Logical Analysis of Language", transl. from German by A. Pap, in A. J. Ayer (ed.), Logical Positivism (New York, 1959), p. 68.

365

JACQUES BOUVERESSE

the diverse formal devices which may be used by a given form of language to satisfy the requirements that have been formulated at the level of essential grammar. The normal grammatical categories are part of this fourth vocabulary; they appear only when one proceeds from the essential grammar to the rather conventional forms in which it is embodied in this or that language, to the "variable grammar". Strawson feels that even if the traditional syntactic classifications can be explained in terms of formal criteria and if these criteria can be expressed without any recourse to semantic notions, "the point remains that the conventional categories reflect the interaction, in actual languages, of semantic and functional factors with actually found formal factors, and therefore have no place in the study of essential grammars". 86 The decisive point is that a really explanatory theory of language cannot stop at the level of variable grammar and cannot really separate grammatical considerations, in the narrow sense of the term, from functional, semantic, and ontological considerations. Husserl has an analogous attitude, of a different kind: he thinks the problem of the foundations of a grammar cannot be solved without simultaneously practising grammar, semantics, logic, and ontology - without making a link between the grammatical categories, the meaning categories, and the object categories. We have already talked about Husserl before and about the pure logical grammar. What is, in your opinion, the main purpose of the pure logical grammar ? The question seems interesting to me because Husserl remains rather unknown in linguistic circles ... - Yes, it is rather striking. The reason is probably that linguists have not read him. It is especially surprising, however, that Chomsky, to my knowledge, does not refer to him, because I wonder whether there is, among the great traditional philosophers, one to whom he would be closer and to whom he could refer with more justification: Chomsky's linguistics is, on the whole, more 'Husserlian' than 'Cartesian'. In Husserl, there is an entire series of elements which proceed unwaveringly in the direction of Chomsky's ideas. First, the fact that, for Husserl, the main language function consists of the expression of thought: "Thinking designates ... all experienced process in which the meaning to be expressed is constituted in a conscious form; and when the meaning is expressed, thinking designates the meaning of the expression, especially of the discourse at hand. This is called thinking, whether it is judging or wishing, wanting, asking, or conjecturing."87 Then, there is the very clear rationalist and universalist character of Husserl's theory of language which refers explicitly to the idea of universal grammar conceived by seventeenth and eighteenth century rationalism: "And I therefore side to a large extent with the old doctrine of a 'general and rational grammar', of a 'philosophical' grammar; in particular, with what, in it, under the form of an obscure and immature intention, aimed at the 'rational' element, in the authentic sense, of language, and especially its 'logical' element-the a priori 86

P. Strawson, "Grammar and Philosophy", p. 145. E. Husserl, Formale und Transzendentale Logik. Versuch einer Kritik (Halle-Saale, 1929), § 3. 87

der logischen

Vernunft

366

JACQUES BOUVERESSE

of the form of meaning."88 Finally and especially, there is the fact that the recursive property of the pure logical grammar and its potentially infinite generative capacity are clearly recognized by Husserl. The 'pure logico-grammatical morphology', as it is also called, formulates the laws of constructing an unlimited number of possible forms of meaning, by infinite composition and iteration, by virtue of some basic operations applied initially to a small number of primitive forms. 89 Of course, Husserl's pure logical grammar is inspired by a fairly obsolete theory of judgment from the point of view of contemporary logic and linguistics - but it is not the right place here to develop this point. Are Husserl's purely logical grammar and Strawson's essential grammar universal grammars ? - In "Grammar and Philosophy", Strawson manifests rather clear universalist tendencies, but his essential grammar is not a universal grammar: "The notion of an essential grammar is, evidently, a relative notion: an essential grammar is the essential grammar of a specified language type."90 The essential grammar evidently possesses a relative universality: it is universal with respect to all the 'accidental' or variable grammars that may correspond to it, but, as it is not any less the grammar of a given sort of language, we cannot prejudge the similarities and the differences which might exist between the diverse languages, no longer considered from the point of view of their variable grammar but from the point of view of their language type - their intrinsic grammar. Anyway, we are not there yet: in practice, Strawson notes, we are obliged to start by attempting to construct the essential grammar of relatively simple language types; and it is certainly impossible to tackle the problem posed by the reconstruction of the essential grammar of a natural language immediately. Husserl, as I have said, is a decided universalist. For him, the pure morphology of meanings "delimits itself like a sphere which, considered by itself, is primary and basic.... It lays bare an ideal scaffolding which all actually existing languages, obeying partly universal human motives and partly empirical motives which vary accidently, fill and dress with empirical materials, according to different modes, each in a way proper to itself."91 Yet Husserl naturally specifies that this "ideal scaffolding" is reflected only imperfectly in the empirical grammar of languages. This is why the pure morphology of meanings should not be guided blindly by linguistic grammar; it should be interested only in the 'purely grammatical' element - that is, the a priori element of grammar. On the other hand, the pure morphology of meanings does not contain the totality of

88

E. Husserl, Logical Investigations, Fourth Logical Investigation, § 14. E. Husserl, Logical Investigations, Fourth Logical Investigation, § 13; Formale und Transzendentale Logik, § 13. 90 P. Strawson, "Grammar and Philosophy", p. 146. 91 Cf. note 88. 89

JACQUES BOUVERESSE

367

the grammatical a priori "because, for example, the communicative relations between psychic subjects, so important for grammar, involves an a priori of its own". 92 Could HusserVs conception of the relation between logic and grammar be used in contemporary linguistic theory ? - For Husserl, it might be said in a way that there is no distinction between grammar and logic - or at least between grammar and one of the levels of logic - and, in another way, that there is a distinction. What is usually called formal logic corresponds to Husserl's logic of consequence or logic of non-contradiction. This discipline only intervenes in the second place, however, after another more basic, logical and formal discipline-the pure logico-grammatical morphology which is the theory of the possible a priori forms of meaning. Husserl says that "language has, not only physiological, psychological and historico-cultural foundations, but also its a priori foundations". 93 These are constituted by that part of the logic of meaning which is entirely a priori and independent of the particular logico-grammatical structure of the various existing languages. Once more, the a priori in question exists only in an impure state: not all the grammatical aspects of language are 'logical', but they constitute the empirical dressing of an essential and universal logical structure. On the other hand, as I have pointed out, the grammatical a priori itself is for Husserl not entirely logical, because there exists, for example, a specific communicative a priori. As a precise answer to your question, I would say that Husserl's conceptions (provided, of course, that they are translated in a language closer to that of the linguists and the logicians - less 'phenomenological', if you want) could regain interest today with the development of generative semantics, because the evolution we have been able to observe since 1965 has consisted, to a significant degree, of the fact that the deep structures have been considerably modified and standardized and have come to resemble semantic and logical representations more and more. In other words, the grammar tends to become more and more, in the work of certain linguists, a kind of theory of the possible forms of meaning - of the logical forms of sentences conceived according to a model which is similar to that of the predicate calculus - and of the relations between the surface forms of sentences and the corresponding logical forms. 94 Of course, as a philosopher, I am satisfied with registering a characteristic change of attitude. I am quite unable to say whether generative semantics is nothing more than a notational variant of the extended standard theory and whether the preference given to one of these two theories can, at the present time, be based on sufficient empirical arguments. It seems important to me that the problem of the relations between syntax, semantics, and logic is now really under discussion and is even discussed with a certain eagerness. 92

Cf. note 88. Cf. note 88. 94 Cf. for example G. Lakoff, "Linguistics and Natural Logic", in D. Davidson and G. Harman (eds.), Semantics of Natural Language (Dordrecht, 1972), 545-665.

93

368

JACQUES BOUVERESSE

We are now fully entering the very subtle and diversified discussion about the notion of linguistic meaning. I would first like to ask you if you see, in principle, a basic difference between formal or philosophical semantics and linguistic semantics. - 1 would first like to remark that one should not hypnotize oneself, as is often done, with the term 'formal', opposing logical or philosophical semantics, which is indeed formal, to a 'material' semantics which would be the semantic theory of natural languages. The semantic theory of Katz and Fodor, for example, is itself formal in a very strict and, if you want, a very trivial sense. The information contained in the dictionary on the semantic properties and relations of the lexical elements is represented in a formal way, and the projection rules are also formulated formally, of course, allowing, in principle, the assignment of an interpretation to any given sentence to occur exclusively by means of formal calculi, without any recourse to linguistic intuition. Katz and Fodor, moreover, emphasize that semantic theory must be formal at the risk of being empty. "The need to have a formal semantic derives from the necessity of avoiding vacuity; for a semantic theory is vacuous to the extent that the speaker's intuitions or insights about semantic relations are essentially relied on in order that the rules of the theory apply correctly." 95 It would be a mistake to believe that a semantic theory of that kind associates extra-linguistic entities, objects, or meanings to linguistic elements. It merely assigns to the lexical items formally specified lexical entries and to the sentences, interpretations, or "semantic readings", which are sequences of symbols of a certain type. It is evident as well that the difference between logical semantics and linguistic semantics is not connected to the difference in the complexity of the systems studied. The concrete examples of semantic systems examined by Carnap in Introduction to Semantics are very simple, but the problem of the definition of the basic concepts of the theory of meaning would remain exactly the same if, for example, they were to comprise as many designating terms as a natural language. The important point is rather, using L. J. Cohen's language, 96 that in the case of a Carnapian semantic system meaning is de jure while in the case of a natural language it is de facto. In the first case, the interpretation is stipulated by rules, and the expressions do not have a meaning other than that freely conferred on them by the rules of interpretation. In the second case, the meaning is given, and we must reconstruct the theory of meaning on the basis of the linguistic behavior of the normal language users. When he formulates the interpretation rules of an object language in a semantic metalanguage, the logician chooses the dictionary arbitrarily, in the case of a natural language, however, the dictionary is explicitly given, in part, and, for another part, it is implicit in the usage. The rules do not play the same role depending on whether a formal system or an ordinary language is at stake: they are definitional in the first case, descriptive and explanatory in the second. The rules of a formal language define that language; they 95

J. J. Katz and J. A. Fodor, "The Structure of a Semantic Theory", in Katz and Fodor (eds.), The Structure of Language (Englewood Cliffs, 1964), p. 501. 96 L. J. Cohen, The Diversity of Meaning (London, 1962), § 5.

JACQUES BOUVERESSE

369

'construct' it. Those of an ordinary language describe or explain what that language is 01 what the linguistic competence of its user is. With regard to semantics proper, in one case one proceeds from the rules to the meaning, in the other one reascends from the meaning to the underlying rules. A very debated question since 1963 is whether the linguistic semantic theory has a decisive advantage over the logical semantic systems when a number of problematic concepts of the theory of meaning have to be defined - those of analyticity, synonymy, entailment, etc. In your opinion, what could the answer to this question be ? - The major difficulty is, I believe, that which was formulated by Chomsky in his 1955 article against Bar-Hillel. If we had satisfactory operational criteria at our disposal for concepts such as synonymy, paraphrase, or analyticity, we would not need the formal procedure used by logicians to 'define' them. And if we do not have such criteria available, the formal characterization obtained by virtue of semantic rules or meaning postulates is arbitrary and deprived of any scientific interest. One should actually not be too optimistic about the explanatory power of devices such as Carnap's semantic systems. They mainly serve to define and relate a certain number of semantic terms such as "designate", "true", and "L-true", and this for a language or a language sample particularly specified. The definitions obtained are of the enumerative recursive type, and they do not explain, properly speaking, what the relation "designate in L" or the predicate "true in L" is for any language L. The study of semantic systems belongs to what Carnap calls pure semantics. This discipline does not have any factual content and is merely preoccupied with making explicit in a semantic metalanguage the analytic consequences of a number of initial stipulations concerning fixed linguistic systems. The extreme difficulty met by the definition of the main terms of the theory of meaning results from the fact that this definition is required to provide an analysis of the definiendum - and not only a formal or operational decision criterion, in the form of a nominative list, a recursive characterization, or a behavioral test. This difficulty, moreover, is not peculiar to semantics. It can be said, for example, that a (totally) ungrammatical sentence in a language is a sentence a normal user of that language will read with the same intonation as if he repeated a sequence of words without any link between them. This operational criterion, however, obviously does not explain what the grammaticality or the ungrammaticality of a sentence is. For Katz and Fodor, the abstract concepts of the theory of meaning should not only be defined implicitly with relation to artificially constructed semantic systems (but which may resemble more or less samples of natural languages, if such should prove to be the case), they must be genuinely analyzed in adequate semantic theories. Nevertheless, they tend to forget that their own theory also involves unanalyzed primitive notions, those of 'semantic marker', 'distinguisher', etc., which they consider self-evident, although one can rightfully call them to account for them. What objections can there be, in the end, to the solution Katz and Fodor pretend to have applied to problems such as the definition of synonymy and analyticity ?

370

JACQUES BOUVERESSE

- Katz and Fodor's mistake is a classical mistake. It consists in the belief that the dictionary entries provide the meaning or meanings of words, and nothing else.97 Yet the problem of analyticity is a problem specifically related to the respective roles played by the specification of meaning of the linguistic elements in question and by the factual information in the determination of the truth of an utterance. For Katz and Fodor, an analytic proposition is a proposition of which the truth merely results from the dictionary and the projection rules, but how could the objections formulated by Quine against the Carnapian notions of 'semantic rule' or 'meaning postulate' also fail to bear on those of 'semantic marker', 'projection rule', etc? In fact, if, as Quine thinks, nothing is gained by substituting for the problematic concept "true by virtue of the meaning" the concept "true by virtue of the semantic rules", nothing is gained either by substituting the concept "true by virtue of the semantic structure of the language considered as defined by a dictionary and projection rules". We will have solved the problem of synonymy and analyticity if we know precisely how Katz and Fodor's dictionary must be constructed and what exactly it must contain. Kant considers the judgment "All bodies are extended" as analytic, and the judgment "All bodies have weight" as synthetic. What motivated answer could the ideal dictionary of Katz and Fodor give to the question whether the predicate is actually contained in the subject in the first case, as Kant says, and not in the second? In other words, how should the adequate lexical entry for the word "body" be constructed without having just resolved the problem of analyticity for the two utterances in question? Or still, if you prefer, let us consider a problem which tortured Locke a great deal. The fact that gold should be malleable "is a very certain proposition, if malleableness be a part of the complex idea the word 'gold' stands for. But then here is nothing affirmed of gold, but that that sound stands for an idea in which malleableness is contained; and such a sort of truth and certainty as this it is to say, 'a centaur is four-footed.'" 98 But what are the properties that are and those that are not part of the complex of properties which we designate conventionally by the term "gold" ? If the property "yellow" is part of it, then the proposition "Gold is yellow" is analytic. How could one, without obviously begging the question, confide to a dictionary the task of distinguishing between the essential predicates and the accidental predicates of an object, or between what is relevant only to the determination of the meaning of a term and what constitutes factual information concerning the extra-linguistic entity it designates ? Evidently, no dictionary does it or can do it. Wilson states It magnificently: "We may gloss Quine's thesis about the analytic-synthetic distinction as: There is no sharp line between what properly belongs in a dictionary and what properly belongs in an encyclopedia."99

97

Cf. N. L. Wilson, "Linguistical Butter and Philosophical Parsnips", Journal of Philosophy 64 (1967), 55-67. 98 J. Locke, An Essay Concerning Human Understanding, Book IV, Chapter 6, Section 9. 99 N. L. Wilson, "Linguistical Butter", p. 63.

JACQUES BOUVERESSE

371

Do analyticity and synonymy really exist in natural languages ? - Although, for a while, I have been quite impressed by Quine's objections, I finally think that the correct answer to this question is that given by Searle. The search for an extensional criterion for concepts such as analyticity or synonymy must start from linguistic facts considered indisputable, of the kind: in French, oculiste 'oculist' means eye doctor. The adequacy of the criterion must be tested with regard to facts of this sort. Thus, "the claim that 'oculist' means eye doctor is not a claim that has to satisfy any criteria which philosophers might propose for synonymy, but rather any proposed criterion for synonymy has to be consistent with such facts as that 'eye doctor' is synonymous with 'oculist'". 100 Of course, this does not at all contradict the fact that there probably exist very few indisputable examples of synonymy or analyticity and a very large number of cases. Let us return to the more general question I asked you a while ago. To what extent can the syntactic and semantic study of artificial languages contribute to the understanding of natural languages ? - It seems to me that, if there is a possible contribution, Saumjan shows its best. Carnap wrote in The Logical Syntax of Language: "The syntactical property of a particular word-language, such as English, or of a particular class of word-languages, or of a particular sub-language of a word-language, is best represented and investigated by a comparison with a constructed language which serves as a system of reference."101 What Saumjan does consists mainly of using and specifying this idea. For him, when we want to study the properties of natural languages, we must first construct an artificial symbolic language, the genotype language, "which simulates the universal properties of natural languages".102 We must, then, examine the formal properties of the genotype language, the laws that determine its functioning and the transformations by which one proceeds from that language to the existing natural languages, possibly by interposing between the two types of languages ideal intermediary languages. What is important in Saumjan is that the genotype language does not at all intervene as a fiction or a descriptive artifice: it has an objective existence, although it obviously constitutes a hypothetical reality. Saumjan's "abstract generative grammar" is a theoretical construction which results from an extremely general a priori concept of what the universal properties of a natural language may be. Saumjan's method, which refers explicitly to Carnap, is on the whole much more deductive and much less empirical than Chomsky's. In fact, if one examines what the general attitude of Chomskyan linguists is with regard to Carnap, it can be observed that, for them, he is far too empiricist in one way and not enough in another. He is too empiricist from the philosophical point of view because he refers to the empiricist tradition and renovates it at the same time. He is not enough of an empiricist from 100 101 102

J. Searle, Speech Acts, p. 9. R. Carnap, The Logical Syntax of Language (Paterson, 1969), p. 8. Cf. p. 279 of this book.

372

JACQUES BOUVERESSE

the linguistic point of view because he is most often accused of having an exclusive interest in the study of abstract linguistic systems, and not enough in the syntactic and semantic properties of concrete empirical languages. Let us say, if you want, that, since he belongs to the tendency called "logical empiricism", he is at the same time too empiricist and too logical. There are two philosophical theories of meaning which have fascinated linguists. The first is that of Wittgenstein: "Meaning is use". But this theory is generally only mentioned to be rejected. I would like to ask you first if one can find in Wittgenstein a coherent and uniform theory of meaning ? - I think this can be answered very clearly: there is, in Wittgenstein, no unitary, elaborated theory of meaning, nor even, properly speaking, a theory of meaning at all. Wittgenstein has never attempted, contrary to what many linguists think, to construct a 'definition' or a 'theory' of meaning. In his work, one finds, for the most part, different suggestions concerning the way in which to ask questions, in problematic cases, about the meaning of a word or an expression. This is very evident if Wittgenstein's work as a whole is considered. The only moment when he had something which genuinely resembled a theory of meaning may be the period of Tractatus. This theory finally resembles an atomistic and pre-Saussurian conception of language. The thirties constitute an interesting transitional period in the evolution of Wittgenstein's ideas in that respect: he progressively abandons the "picture theory" and he makes other suggestions at the same time. This is the era when he says, for example, that the meaning of a word is its place in a grammatical system - in the calculus of language. He insists enormously, at that time, on language as a system of internal dependencies, and he reproaches himself severely for having neglected this essential aspect in the Tractatus. He observes, for example, that rather than saying what he says in the Tractatus, "The preposition is like a ruler laid against reality", we should say "A system of prepositions is like a ruler laid against reality". It is not the isolated proposition which attributes a certain color or length to an object that is compared to reality, it is the whole system of those propositions which express the color or the length of an object. This allows one to explain why, if an object has a given color or length, it cannot simultaneously have another one. This is a problem for which Wittgenstein could not find a solution in the perspective of Tractatus, because he assumed that the elementary propositions had to be independent. 103 In fact, in the thirties, one can see simultaneously in Wittgenstein several quite different 'theories' of meaning. Although he remains more or less loyal to certain aspects of the picture theory, he sometimes suggests a theory of meaning as use which, moreover, was already embryonic in Tractatus-, he insists, as I have said, on the fact that the meaning of a linguistic element is determined by its relative position in a system. He also proposes what has been called the 'verification principle' - that is, the idea 103

Cf. for example, Ludwig Wittgenstein und der Wiener Kreis, B. F. McGuinness (ed.) (Oxford, 1967), pp. 63-64.

JACQUES BOUVERESSE

373

that the meaning of a proposition is its method of verification. Actually, as far as can be judged, he has never presented this idea as a definition or a dogma, but merely as one of the possible ways to ask the question of meaning with regard to an utterance. An interesting way of asking questions about the meaning of a declarative sentence may be, in fact, to ask how one would verify it, but there are obviously others. This has certainly not escaped Wittgenstein. Concerning the famous equation "meaning = use", I must first say that, to my knowledge, nothing can be found in Wittgenstein which resembles that exactly. There are essentially two things in his work. On the one hand, there is the famous slogan "Don't ask for the meaning, ask for the use", which is something very different and which is primarily addressed to philosophers, not to linguists. On the other hand, there are remarks such as: "For a large class of cases - though not for all - in which we employ the word 'meaning', it can be defined thus: the meaning of the word is its use in the language. And the meaning of a name is sometimes explained by pointing to its bearer."104 In any case, it is far from the statement that "meaning is use", conceived as a theoretical identification of the two notions. "The big question of semantics", Katz writes, "is 'What is meaning?'" 105 Wittgenstein, however, does not ask this question at all, precisely because he feels that it has no clear meaning. For him, any statement of the kind "Meaning is ..." constitutes a false start. Only recommendations of the following form can be found in his work: "If you want to understand what the meaning is of such and such a form of expression, ask such and such a question." Did Wittgenstein perceive a real incompatibility between the different conceptions he has proposed at one time or another? - No, not exactly. He was hostile to the idea of a theory of meaning in general, but certainly even more to the idea that a particular theory of meaning could do more than account for an aspect of the phenomenon of meaning. He did not believe that one single theory could possibly do justice to all the aspects of the problem of meaning in their extreme diversity. This is why it would never have occurred to him to present his personal suggestions as tentative solutions to the problem of meaning as a whole. It is evident that, in the end, Wittgenstein did not consider the Tractatus a total and gigantic mistake; he certainly thought it had brought to light (but far too unilaterally) certain important aspects of the linguistic phenomenon - mainly the denotational and descriptive aspect, language as nomenclature and as image of reality. He would in no way have suggested afterwards that these aspects did not exist or that they were uninteresting. There are philosophers who, after having "discovered" that the meaning of a word in language is defined by its relation to other words, act almost as if the referential power of language was purely internal, as if an expression always only referred to other expressions. Wittgenstein did not reason that way, however; he did not go from one extreme to another. In the Tractatus, the contact of 104 105

L. Wittgenstein, Philosophical Investigations, Section 43. J. J. Katz, The Underlying Reality of Language, p. 84.

374

JACQUES BOUVERESSE

language with reality was assured by the kinds of "antennae" 106 the coordinations of nouns and objects represent. He later denounced this conception as rather rudimentary, but he remained persuaded, of course, that language came into contact with reality, that one had to leave language sometimes, even if, besides, he has maintained in that respect one of the basic theses of the Tractatus - the impossibility of describing in language the way in which language relates to the world. As he says in Philosophische Bemerkungen, "with language, I cannot get out of language". 107 This is also the reason why, in his opinion, there cannot really be a metalanguage. The survey you have just given of the different ideas Wittgenstein has had at one time or another on the problem of meaning is extremely valuable, especially since linguists are often very hostile to Wittgenstein, in particular because of the slogan "meaning is use", which they see as implying a confusion between language and language use, between the theory of language and the theory of 'parole' and performance. I would like to ask you - although you have already answered this question to some extent - what you think may be the interest of the Wittgensteinian conception of meaning from the view point of linguistic theory ? - When one is a philosopher and one is familiar with the way in which philosophers usually talk about meaning, it is easy to understand why Wittgenstein was led to insist so much on the notion of use and ordinary use of language. His recommendations are addressed, once more, mainly to philosophers; and what he suggests for them can be understood only if one opposes these recommmendations to something else, which they tend to do constantly. Wittgenstein did not have a theory of language, and he did not pretend to point out to linguists how to construct one. The task of a philosopher is to solve philosophic problems; it is not to construct, for example, an adequate semantic theory of the language concerned. Because, whatever Katz may have thought of it, it is not at all evident that such a theory would help the philosopher to solve these problems in any way. It is regrettable that the linguists do not want to understand that the philosopher's motivations and objectives are not the same as theirs. As for me, I have never quite understood the exact reasons for the extreme animosity often provoked among them by the way in which Wittgenstein tackles the problem of meaning. The idea that one must substitute the research of the use for the research of meaning is not at all a very particular or contestable theoretical conception. For me, it resembles more a simple truth of common sense: if one wants to know what an expression means, one must first see how it is used in linguistic practice, and I assume that is what everyone actually does; this is precisely the kind of truism philosophers tend to forget, however: they believe there exist much more direct and noble ways to ask questions about meaning. As for linguists, one sometimes has the impression that the dictionary of Katz and Fodor, for example, is something given a priori and which exists in itself - as if a dictionary must not first of all be constituted on the basis of a 106 107

L. Wittgenstein, Tractatus Logico-philosophicus, 1. 1515. L. Wittgenstein, Philosophische Bemerkungen (Oxford, 1964), p. 54.

JACQUES BOUVERESSE

375

certain practice of the language concerned, as if a dictionary was not mainly a codification of certain aspects of use. Chomsky 108 was shocked that Wittgenstein could have written: "I want you to remember that words have those meanings which we have given them; and we give them meanings by explanations.... Philosophers [my italics] very often talk about investigating, analyzing, the meaning of words. But let's not forget that a word hasn't got a meaning given to it, as it were, by a power independent of us, so that there could be a kind of scientific investigation into what the word really means. A word has the meaning someone has given to it." 109 Wittgenstein obviously does not mean that we give meanings to words by explicit and conventional decisions. He merely stresses that the meaning of a word is not given in itself, that it is not an independent object which we could study scientifically almost in the same way as a physical substance, for example. He wants to make us understand that the meaning of words is not fixed by a more eminent authority than that of the use we make of them, a common agreement, in the linguistic practice, and that it is exactly what we (the linguistic community) are inclined to recognize as their meaning. People like Chomsky, Katz, and Fodor, however, obviously are committed to a quite clear semantic realism. They believe in the objective existence of a rather independent semantic content of language which is probably common to all languages. Finally, what do you think of the general attitude of linguists with regard to Wittgenstein ? - I think it may be justified from a certain point of view, because all of Wittgenstein's research is situated within pragmatics. It belongs to the theory of performance and not to the theory of language. When Wittgenstein and the linguists talk about 'language', they do not talk about the same thing. What interests Wittgenstein is not language as the descriptive object of linguistics. His method is characterized by the refusal to privilege one aspect of the language phenomena, for example, the genuine linguistic aspect. In fact, he thinks that there is no reason why the aspects of the linguistic phenomenon which can be dealt with more or less scientifically - those with which linguistics is preoccupied - would be more 'important' from a philosophic point of view. It is certain, for example, that language resembles in certain ways a calculus operating according to strict rules, but the philosopher may have excellent reasons to stress, on the contrary, how much it differs in other respects from this seductive model. With his technique based essentially on providing examples,Wittgenstein never reaches the genuine linguistic level of the semantic and syntactic description of a language - because he does not believe that the solution of the philosophical problems is situated at that level. He does not believe that a semantic 108

N. Chomsky, Problems of Knowledge and Freedom, p. 23; and "Some Empirical Assumptions in Modern Philosophy of Language", in S. Morgenbesser, P. Suppes and M. White (eds.), Philosophy, Science and Method. Essays in Honor of E. Nagel (London, 1969), pp. 260-285 (see pp. 276-277). 109 L. Wittgenstein, The Blue and Brown Books, pp. 27-28.

376

JACQUES BOUVERESSE

or syntactic theory of language could seriously contribute to their solution. He may, of course, be partly wrong in that respect, but it strikes me that Chomsky, when talking about Wittgenstein, is extremely attentive to his 'solutions', and not at all to the specific problems he asks himself. There is a wide abyss between Chomsky and Wittgenstein. In fact, Chomsky has very little of the philosopher in the sense in which Wittgenstein conceives philosophy, although he has a lot, doubtless too much, of it in other respects.... Another 'theory' of meaning which fascinates linguists is that of Frege, with his distinction between Sinn and Bedeutung, between meaning and denotation. Where do you consider the place of Frege's theory in linguistic theory? - I am not sure I understand your question well. The distinction between meaning and denotation or, to designate it otherwise, without trying to be totally precise, between sense and reference or intension and extension is not exactly an operative distinction. It is rather a basic presupposition of any semantic theory which does not want to go down in total confusion on the first attempt. It is, of course, inevitable to make a clear distinction between three terms at the outset: the signifier, the signified, and the object, that is, in the case of a noun, between its meaning and the object it denotes; in the case of a declarative sentence, between its sense and its truth value (to talk with Frege); and in the case of a predicate, between the property it designates and the class of objects to which it is applicable. Frege is surely not the inventor of this distinction. The theory of the abstract proposition as objective meaning content of a declarative sentence, which represents by far the most problematic aspect of the distinction in question, has a long history which includes the Stoics, the logic and language philosophy of the Middle Ages, and, closer to us, people like Bolzano. From the moment the distinction is clearly recognized, in its Fregean from or a related form, semantics is divided into two parts, which Quine calls the theory of meaning and the theory of reference. The basic concepts of the theory of meaning are those of meaning or intension, of synonymy, analyticity, entailment, etc. Those of the theory of reference are concepts such as naming, denotation, truth, extension, etc. As Quine remarks, Tarski's semantics mainly represents a contribution to the theory of reference. Moreover, it is incontestable that the main concepts of the theory of reference now raise far fewer problems than those of the theory of meaning. Returning more specifically to the case of natural languages, it is clear that the semantic theory of the linguist belongs to the theory of meaning - not to the theory of reference - although the two are obviously not independent. A theory of meaning is at the same time necessarily a theory of reference for the expressions which can have one: the denotation of a noun, for example, as Church says, is a function of its sense. The determination of the sense of a declarative sentence is necessarily related to the determination of its truth conditions, even if the two determinations are not coincident. There is inevitably a part of the referential properties of language which is determined by the semantic theory of that language -by the dictionary, for example -

JACQUES BOUVERESSE

377

but a very important part is also determined by conventions of a different, less general nature, or by the context of the utterance (this is the case, for example, with regard to most of the proper nouns, the deictic terms, etc.). An important part of a theory of reference inevitably belongs to the theory of language use, of speech acts - not to the theory of language in a strict sense. Carnap makes an effort, on one hand, to rehabilitate the notion of intension and to show that the suspicion this notion meets in people like Quine is not justified; on the other hand, he tries to free it from its usual mentalistic and psychologistic implications and to show that operational tests of a behavioral kind can be conceived for the determination of the intension of terms in a language. He thus characterizes the relationship which exists between the theory of intension and that of extension: "The theory of intension of a given language L enables us to understand the sentences of L. On the other hand, we can apply the concepts of the theory of extension of L only if also we have, in addition to the theory of intension of L sufficient empirical knowledge of the relevant facts." 110 When the linguist studies an unknown language, he first knows the extension of certain terms and on this basis he reconstructs (with a certain margin of indétermination) their assumed intension. From a systematic point of view, however, it is rather the theory of intension which comes first. It can be said, on the whole, that understanding a language is not only knowing the theory of the intension of that language as it is given by the linguistic semantic description; it is also knowing how the data of the theory of intension are combined with other conventions and with the context to determine the concrete meaning of the utterance. This question will turn up again later, I think, when we discuss the relation between semantics and pragmatics, but I would like to ask you now whether you would affirm, against Wittgenstein, that the semanticity of languages includes a minimum of systematicity and that the systematicity can be formalized? Is linguistic semantics possible? - Yes, one should hope so. Moreover, I do not know whether Wittgenstein would really have been opposed to this idea. There again, it should not be forgotten that his point of view is above all philosophical. He would probably not have contested that one could, by placing oneself on a level of sufficient idealization and abstraction, constitute something like a semantic theory of the language considered, but he would have considered that a semantic theory of the Katz and Fodor type, for example, is of practically no help to philosophy: one could say it is far too superficial. It does not descend to the level of depth required for the solution of philosophical difficulties, that of use. For Wittgenstein, the essential distinctions do not necessarily appear at the level of the syntactic and semantic description which can be given by the linguist. There are quite a few important differences which are perceptible only at the level of the concrete linguistic practice, at the pragmatic level, if you want. They are not at all accidental differences, 'small' differences, but 'logical' or 'grammatical' 110

R. Carnap, Meaning and Necessity (Chicago, 1956), p. 234.

378

JACQUES BOUVERESSE

differences, as he calls them. Wittgenstein uses the term 'grammar' without taking into account the distinction between syntax, semantics, and pragmatics. He uses it in a very broad sense for designating all the essential aspects of the use of an expression, all the determinations and restrictions to which the use is a priori subjected, with the consequences they entail; and he does not separate the linguistic constituent from the non-linguistic constituent when it comes to recognizing certain particularities of the grammar, so understood. It is the 'grammar' which keeps us from saying that an object is greenish-red, that we experience someone else's pain, that there is a cardinal number larger than all the others, etc., these are 'grammatical' impossibilities; it is not just false, but it does not make any sense to say that. Wittgenstein, moreover, talks not only about the grammar of an expression but also about the grammar of a process, of a game, of color, etc. Wittgenstein's deep grammar doesn't have any similarity to Chomskyan deep structure grammar... - Oh! In that respect, I believe there is almost no relation! One may have the opposite impression because, when Wittgenstein introduces a distinction between "surface grammar" and "deep grammar" of an expression,111 it concerns, in a way, a distinction between the surface syntactical properties of the expression and something else. But this other thing is not another level of syntactic analysis. The distance which separates the surface grammar from the deep grammar, in Wittgenstein, is something which is measured by the gap that may exist between the idea of its use given by the exterior linguistic form of an expression, and what this use- really is. It is evident that the 'deep structures' ot Chomskyan grammar still belong to the category of these 'exterior linguistic forms', which Wittgenstein recommends philosophers to distrust. For him, most of the philosophic confusions and errors result from the fact that we are constantly tempted to attribute analogous functions to different words and expressions, by virtue of surface grammatical similarities between them. It might, therefore, seem that transformational grammar has something to say here, because it allows, among other things, assigning very different deep structures to sentences with identical surface structures and also assigning identical deep structures to sentences which are apparently quite remote from one another. Wittgenstein means above all that sentences with an analogous structure - even a deep structure - can nevertheless have very different functions or uses. The difference between surface grammar and deep grammar is not a difference between two structural levels but rather between form and function, between structure and use. This is the case, for example, for the two expressions "to say something" and "to mean something", of which the grammatical parallelism is, for Wittgenstein, the source of a philosophic confusion which is extremely difficult to disentangle. It is not evident what transformational analysis could contribute to such a case. In fact, Wittgenstein thinks that "it is impossible, 111

L. Wittgenstein, Philosophical Investigations, Section 664.

JACQUES BOUVERESSE

379

in most cases, to show an exact point where an analogy begins to mislead us", 112 and that, moreover, it is impossible to point out a priori how one can get out of a philosophic difficulty caused by a misleading grammatical analogy. This is why there is no general method in philosophy. It is particularly clear that neither the logical analysis as it was conceived at the time of the Tractatus, the search for the actual logical form hidden under the surface linguistic form, nor the syntactic and semantic description of the linguist can contribute very efficiently to the therapy of what Wittgenstein calls "grammatical" confusions or illusions. Anyway, they cannot properly constitute basic instruments for a philosophical technique. V. O N PRAGMATICO-SEMANTICS A N D N A T U R A L LOGIC

A longstanding trichotomy is that of syntax, semantics, and pragmatics. We have already talked about the relation between syntax and semantics, a problem for Chomsky himself but especially for his disciples. What is probably even more problematic is the relation between semantics and pragmatics. What do you think of Morris' distinction ? Is it operational and still valuable? - Perhaps one should first recall briefly what the tripartite division of semiotics, or the general theory of signs, into syntax, semantics, and pragmatics amounts to. Linguistic research belongs to pragmatics if it makes explicit reference to the language users. If the user is abstracted and only the expressions and their designata are studied, one is in the field of semantics proper. Finally, if the designata themselves are abstracted and only the linguistic expressions and their relations are studied, one is situated at the syntactic level. Pragmatics, as conceived by Carnap, can, in addition to the relation of the users to the expressions, involve the intervention of the designata of these expressions. There is, thus, no strict distinction in that direction. The big problem is whether there is one in the other direction, that is, whether one can treat the designata of the expressions (I continue to use Carnap's language) without taking explicitly into account the relations entertained by the users with their language. In Introduction to Semantics, Carnap writes: "Linguistics, in the widest sense, is that branch of science which contains all empirical investigation concerning languages. It is the descriptive, empirical part of semiotics (of spoken or written languages); hence it consists of pragmatics, descriptive semantics, and descriptive syntax. But these three parts are not on the same level; pragmatics is the basis for all linguistics."113 In the beginning of Meaning and Synonymy in Natural Languages, he goes further; he thinks that the analysis of the meaning of the expressions of a language takes place in two different forms: one which belongs to pragmatics, that is, the empirical study of the historically given natural languages, the other which is part of semantics, understood in the sense of pure semantics, that is, of the study of artifi112

L. Wittgenstein, The Blue and Brown Books, p. 28. Introduction to Semantics and Formalization of Logic, two-volume edition (Cambridge, Mass., 1961), p. 13.

113

380

JACQUES BOUVERESSE

cially constructed language systems defined by the choice of conventional rules, whereas descriptive semantics can be considered a part of pragmatics.114 And in On Some Concepts of Pragmatics, he observes: "There is an urgent need for a system of theoretical pragmatics, not only for psychology and linguistics, but also for analytic philosophy. Since pure semantics is sufficiently developed, the time seems ripe for attempts at constructing tentative outlines of pragmatical systems."115 The taskCarnap assigns to semantics is relatively restricted and well-defined: its goal is to define the predicate "true" for a given object language - that is, to assign truth conditions to all the propositions of that language. Carnap sometimes suggests that understanding a language L is knowing the semantics of that language in the sense just pointed out, that is, having mastered the (recursive) definition of "true in L". Yet, as I have mentioned, he has lealized that it was also much more than that. The sentences of a language are not only used to make affirmations, to say true or false things, but for a number of other purposes. Thence the necessity of what he calls "theoretical pragmatics", that is, a discipline which does for concepts such as belief, assertion, utterance, etc., what semantics does for concepts such as 'designates' or 'true'. Pragmatics has long been the most neglected of the three semiotic sections, but there are now people who feel the urgent need of a rigorously formal pragmatics which could, every allowance being made, be brought on the same level as logical syntax and semantics. What exactly did Carnap understand by "theoretical pragmatics" ? It is not immediately obvious how the object of this theoretical pragmatics could be formalized. - For Carnap, the question was simply defining in an appropriate metalanguage a certain number of pragmatic concepts corresponding to given linguistic systems. Such a discipline would be formal in the same sense as (pure) semantics is formal. Of course, Carnap did not propose to tackle the immediate and total problem of the pragmatics of a natural language at all. He thought one could start modestly with the definition of small groups of interlinked concepts corresponding to relatively simple linguistic systems. Until rather recently, the analysis of the most characteristic concepts of pragmatics was usually handed over to the philosophic research of ordinary language philosophy, but this is no longer the case. Carnap's wish has been granted, and very interesting results have been obtained. I am thinking, for example, of the works of Richard Montague. Naturally, it is quite well possible that linguists would be as sceptical with regard to the interest of research of that type as they have been to Carnap's work on semantics. The same principled reservations can be formulated in both cases. If it is admitted that semantics has its starting point in pragmatics, can semantics still 114

Cf. Meaning and Necessity, p. 233. "On Some Concepts of Pragmatics", Philosophical Studies 6 (1955), 89-91; reproduced after Meaning and Necessity, p. 250. 115

JACQUES BOUVERESSE

381

be considered a specific field? Is semantics still independent with respect to pragmatics ? - Yes, to the extent that one can really talk about the meaning of the expressions of a language without talking about the way they are used by the speakers in linguistic practice. If the theory of language is conceived, above all, as the description of the potential of illocutionary acts, which define a language as language, it is difficult to maintain a strict distinction between semantics and pragmatics. The semantic description of a sentence, for example, must inevitably be above all the description of the speech acts which it can realize in the appropriate circumstances. Yet, as Robert Stalnaker writes, "pragmatics is the study of linguistic acts and the contexts in which they are performed". 116 Does Searle's theory of speech acts actually belong to semantics or to pragmatics? As I have mentioned, the persistent reservations of certain linguists with respect to semantics mainly result from the difficulty of treating the problem of meaning without soon being obliged to do something which is not linguistics. "If you are interested in the meaning of words", Martinet says, "you will soon discover that you have become a gardener, a chemist, or a philosopher" 117 - note the honorable but rather strange company in which the philosopher is here! To me, the most interesting position adopted on this problem seems that of Oswald Ducrot, who considers the possibility of a properly linguistic semantics not as something self-evident, but as a working hypothesis. Observing that if we want to predict for each sentence of the language the infinite number of meanings it may have, taking into account the infinite number of possible contexts, we will have to represent explicit indications in the semantic description dealing almost de omni re scibili, he formulates the following hypothesis: "... The illocutional circumstances come into play to explain the real sense of a particular occurrence of an utterance only after a meaning has been attributed, independent of any context, to the utterance itself." 118 Inotherwords, he postulates that the whole of our knowledge which must be mentioned in the semantic description may be divided into two subsets: one, constituting the linguistic component which assigns to each utterance A, independent of the context, a certain meaning A'; the other, the rhetorical component, which would be responsible, given A' and a certain context X, for predicting the actual meaning of A in context X. This hypothesis, which is not at all self-evident, as Ducrot stresses once more (and in my opinion with justification), could be confirmed if one could attain a non-arbitrary distribution - that is, as a function of precise criteria - of a number of elements with an uncertain status, between the linguistic component and the rhetorical component. Ducrot's works constitute a beginning of the realization of this trend, and it is a very promising one.

116 R. C. Stalnaker, "Pragmatics", in D. Davidson and G. Harman (eds.), Semantics of Natural Language, p. 383. 117 Cf. p. 231 of this book. 118 O. Ducrot, Dire et ne pas dire. Principes de sémantique linguistique (Paris, 1972), pp. 111-112.

382

JACQUES BOUVERESSE

Is the questioning of the distinction between semantics and pragmatics not at the same time a questioning of the distinction between competence and performance ? - Yes, certainly. For Chomsky, the study of extra-linguistic competence which conditions the use of language belongs to the theory of performance. But if it can be said that the syntactic competence as he conceived it is clearly linguistic, it is difficult to say the same immediately about the interpretative or semantic competence. It can very well be maintained that the semantic competence of a normal language user is not the faculty to understand the meaning of an infinite number of sentences in isolation, but the faculty to understand the actual meaning of an infinite set of utterances of these sentences in infinitely many different contexts. There is a semantic competence, in the sense of Chomsky, Katz and Fodor, if there exists something similar to what Ducrot calls the linguistic component of the total semantic description. There is, of course, no reason not to admit that the rhetorical component itself corresponds to a certain form of 'competence', which is certainly constituted by the possession of a very diverse knowledge and faculty. Another way of instigating the more or less complete disappearance of the semanticpragmatic distinction is the introduction in linguistics of the notions of presupposition, reference, etc. I would like to ask you if you yourself have a clear and precise definition of the notion of presupposition. Looking at the use of this notion in linguistic theory, almost total confusion can be observed - everyone uses it in his own way. Would it be interesting, in your opinion, to stick to one single definition of the term ? - I obviously have no personal opinion about such a difficult question. You are right in saying that the notion of presupposition is used in a rather confused way. Keenan 119 has introduced some clarity by his proposal to distinguish two kinds of presuppositions clearly: the logical presupposition and the pragmatic presupposition. The notion of logical presupposition may be defined entirely by the basic concepts of logical semantics. It will be said that a sentence S logically presupposes a sentence S' if S, and also its negation ~ S, entail S' - that is, if the truth of S' is a necessary condition for the falsity as well as the truth of S. The logical presuppositions of a sentence are other sentences, and, in the case of natural languages, it is for the theory of Katz and Fodor to describe the logical presuppositions of a sentence, because it is supposed to provide the means of determining the entailment relations existing among the sentences of the language considered. The pragmatic presupposition is not defined by a relation between a sentence and another sentence, but by a relation between the utterance of a sentence and the context in which this utterance takes place: "An utterance of a sentence pragmatically presupposes that its context is appropriate." 120 Keenan specifies that neither of these notions of presupposition calls for the intervention of the speaker's beliefs. It is evident in the case of the logical 119

Cf. E. L. Keenan, "Two Kinds of Presupposition in Natural Language", in C. J. Fillmore and D. T. Langendoen (eds.), Studies in Linguistic Semantics, 44-52. 120 " f w o Kinds of Presupposition", p. 49.

JACQUES BOUVERESSE

383

presuppositions, but the pragmatic presupposition, which merely refers to the existence of an appropriate context, has no relation in principle to the speaker's state of mind (this means that this state of mind is not part of the context as it is defined here). The main difficulty concerning the term presupposition is, I believe, whether it is reasonable to use it, as is often done, to designate simultaneously something which results from the semantic structure of the language concerned, something which refers to the (objective) context of the utterance, and something referring to the particular intentions and dispositions of the speaker. I think that Richard Gardner is right when he observes that on the whole it might be better not to talk about the presuppositions of sentences or parts of sentences, but only about the presuppositions of types of illocutionary acts or of concrete illocutionary acts. 121 It obviously remains to be determined what happens when the presuppositions which condition the normal accomplishment of an illocutionary act are not satisfied - for example Austin's "felicity conditions". What is Ducrot's position with regard to that problem? - In Dire et ne pas dire, Ducrot criticizes the current conception according to which presuppositions refer to possible conditions or modalities of accomplishment of a given speech act. He proposes treating the presupposition itself as a particular speech act, just like the affirmation, the question, or the command: "...The choice of presuppositions seems to us a particular speech act (which we call presupposition act), a judicial act, and thus illocutionary, in the sense we have given that term: in accomplishing it, one transforms the interlocutor's possibilities of speech at the same time." 122 The presupposition determines the obligatory frame of the subsequent dialogue. Therefore, it is part of the organization of language itself. The theory of presuppositions is a particular branch of the general theory of speech acts: language must be conceived, among other things, as a store of presupposition acts. What Ducrot wants to accomplish "is to reconcile the specificity of presuppositions and their character as semantic constituents". 123 Supporting his theory is the fact that when we order someone to close a door, for example, when he is completely paralyzed or when the door is already closed, it is difficult to pretend that we have not accomplished any speech act whatsoever. Wilfully pronouncing an utterance whose normal presuppositions are not satisfied or not accepted by the interlocutor may have the purpose of embarassing or degrading him, or of provoking his polemic or aggressive reactions, etc. As Ducrot says, "the presupposition plays a primary role in the strategy of linguistic relations". 124 Ducrot's conception is, in a way, the normal result of the theory of speech acts, if 121

Cf. R. Gardner, "Presupposition in Philosophy and Linguistics", in C. J. Fillmore and D. T. Langendoen (eds.), Studies in Linguistic Semantics, 23-44. 122 O. Ducrot, Dire et ne pas dire, p. 91. 123 Ibid., p. 65. 124 Ibid., p. 95.

384

JACQUES BOUVERESSE

this is pursued consistently to the end. According to the partisans of this theory, we should not say that an expression has reference (predicts, asserts, etc.), but that an act of reference (predication, assertion, etc.) is accomplished by using this expression to accomplish a certain speech act. Similarly, we should probably not say that a sentence, or even an utterance, presupposes this or that, but rather that a certain act of presupposition is accomplished by using a sentence in a certain way in a certain context. The main difficulty is that, different from most other illocutionary acts, the presuppositional act is not 'marked' in a specific way in the sentence structure and that, consequently, its linguistic representation raises particular problems whose solution Ducrot outlines. Do you admit that there are lexical, sentential, and textual presuppositions ? That is another possible classification. One could try to define the different types of presupposition as a function of the linguistic units on which they bear. What do you think about that? - If I can venture an opinion, I would say that I do not see very well how, when one wants to talk about presupposition, one could reasonably consider going under the linguistic unit represented by a complete speech act. Of course, one can say that when one orders someone in a serious tone, "Close the door!", the fact that there is a door and only one which can be identified unambiguously as the object of the command is a presupposition of the definite description "the door", or maybe of the definite article "the"; the fact that the door in question is not yet closed is a presupposition of the verb "close". Yet it would be absurd to say, for example, that the expression "The present king of France" presupposes the existence of an entity - and only one - that would actually be king of France. Although this expression is a singular referential expression, it presupposes only the existence of an appropriate reference to the extent that it is actually used to accomplish a referential act within what Searle calls a propositional act - an act of reference which is common to the following: the affirmation, "The present king of France is bald"; the negation, "The present king of France is not bald"; and the question, "Is the present king of France bald?" However, in "X believes that De Gaulle is the present king of France", and "He is wrong because Dc Gaulle is not the present king of France" (I borrow these examples from Gardner), it cannot be said that the existence of a person and only one who would presently be king of France is presupposed, and that the absence of a descriptum for the definite description, "The present king of France," renders the two assertions incorrect and meaningless by themselves. More obviously difficult is the question whether one can talk specifically about lexical presuppositions. It could be said, for example, that "white" presupposes "extended", "colored", etc., that "mammal" presupposes "vertebrate", "animal", "living being", etc. But how can a distinction be made between what a word means and what it presupposes? How can one keep from saying in the end that a word 'presupposes' its meaning and, in the last analysis, the semantic structure of the

JACQUES BOUVERESSE

385

language considered? From saying that the accomplishment of an effective speech act 'presupposes' that the elements constituting the linguistic support be used with a meaning which is not totally removed from their normal meaning? The universe of presuppositions is an extremely anomalous one, and if one decides to make it sufficiently wide and uncontrollable, one would end up classifying almost any type of anomaly or linguistic deviance under the category of 'presupposition failure'. W h y not say, for example, that " D a s Nichts selbst nichtet" (nothing itself nothingeth) is a meaningless sentence because a verb presupposes a subject and, from the logical syntactic point of view, "das Nichts" (nothing) cannot be a subject; that "Caesar is a prime number" is meaningless because the attribution of the predicate "prime" to an entity 'presupposes' that the entity in question be an integer; that "This solid figure is a regular decahedron" is meaningless because the attribution of a certain geometric configuration to any kind of body 'presupposes' that this be a possible mathematical configuration, and so on? The interest of Ducrot's theory is the integration of the analysis of the implicit element contained in the presupposition, into the semantic description (the 'posed' and 'presupposed' are both part of the literal meaning of an utterance) while giving it an important specific role in linguistic communication. If you say "Peter suspects that Jack will come" or "Does Peter suspect that Jack will come ?" you place the dialogue in the hypothesis that Jack will come, and you implicitly determine what is relevant in the discussion and what is not. Just as promising to do something is not stating your intention to do it, but deliberately putting yourself under the obligation to do it, so is saying that Peter imagines that Jack will come not stating that Jack will not come, but putting your interlocutor under the obligation to continue the dialogue as if the fact that Jack will not come was an acknowledged fact (whereas it may very well not be acknowledged by him). The difference between "Peter thinks that Jack will come", "Peter suspects that Jack will come", and "Peter imagines that Jack will come" is composed of the fact that, different from the first, the last two utterances make it irrelevant whether Jack will come or not, as this question is supposedly resolved positively in the second and negatively in the third.

As you have talked about the problem of definite descriptions, we might mention the relationship between presupposition and reference... - There is obviously a relation between the problem of presupposition and that of reference - at least from a historical point of view. It is mainly the problem posed by the sentences involving improper definite descriptions (empty or ambiguous) which has allowed people like Frege and Strawson to isolate the notion of presupposition. The first presuppositions discovered and analyzed are the existential presuppositions of utterances which involve definite descriptions or more generally, singular designative expressions. Ducrot is probably right, however, when he says that "the study of definite descriptions has been, for the concept of presupposition, the best and the

386

JACQUES BOUVERESSE

worst of things". 125 That was probably what instigated the treatment of presuppositions in general in connection with the conditions of the 'normal' use of utterances, and not as an element belonging to the semantic content and constituting the object of a particular illocutionary act. In other words, the almost exclusive attention paid for a long time to the absolute or occasional referential presuppositions of an utterance has probably led philosophers and linguists to insist in an exaggerated way on certain extrinsic conditions that must be realized for a 'normal' utterance to be 'possible', and not enough on the genuine linguistic mechanism of the presupposition, on what exactly is done when somebody formulates an utterance involving certain presuppositions (verified or not, probable or improbable, acceptable or unacceptable). Another sign of the disappearance of the semantic-pragmatic distinction is that more and more emphasis is placed on the referential function of language. Chomsky obviously talks very little about it, and it is generally an aspect which has almost remained in total silence in structuralism. This is especially so, it seems to me, because it was always believed that, as Saussure stated, there was an enclosure of the universe of discourse, of language, and that reference did not have a place in this universe - that the linguist did not have to deal with the relation between language and the world. Today, however, it is more and more apparent that the referential function of language is emphasized. This is the case, for example, in generative semantics. Would you agree that this may be a sign of the disappearance of the distinction between semantics and pragmatics ? - Certainly. The theory of reference, particularly of contextually determined reference, belongs obviously to pragmatics (but, of course, not only to pragmatics). The referential function of language is certainly not an accidental and extrinsic function; it must be registered in language itself and it must have repercussions in the linguistic structure of expressions. You are completely right when you say that it has long been practically impossible to insist on the referential properties of language without the risk of being accused of returning to an atomistic, prescientific, and completely obsolete conception. It seemed that there was a choice between the structural conception - language as an autonomous system of internal relations and oppositions - and the theory of language as a reproduction of reality - if you want to take an extreme example, Wittgenstein's Tractatus. In my opinion, however, Wittgenstein was totally justified in preserving a certain esteem for his first work. Language is also a means to refer to objects and to describe states of affairs. Thank God, the disastrous situation to which we have just alluded is changing completely. All the phenomena which we have discussed at one time or another - the renewal of interest in semantic but also pragmatic research, the need many people feel for a linguistics of discourse, and no longer only of the word or the sentence, the opening 125

Dire et ne pas dire, pp. 221-222.

JACQUES BOUVERESSE

387

toward other disciplines such as psychology, sociology, and ethnology, the attempts to constitute a theory of reference and of presupposition in rather close conjunction with logic, etc. - are finally going in the same direction. What matters, once more, is to understand language no longer by itself, but as a potential of linguistic acts (whereas Chomsky understands the possession of language more as the disposal of a potential of linguistic characterizations). All these trends are interrelated, indeed. As you have already alluded frequently to the theory of speech acts, I would like to ask you if you consider a typology of speech acts possible and whether that typology is linguistic ? For example, does it depend on the linguistic categorization of verbs ? - It is certainly partly dependent on it, but for a rather trivial reason: the semantic description of a sentence must necessarily contain the indication of the type or the different types of illocutionary acts the accomplishment of which this sentence allows. It is quite evident in the case of sentences involving explicit performatives such as "I order..." or "I promise...". But if we take the case of a sentence whose exact illocutionary power is not clear, even taking into account the circumstances of the utterance, it is incontestable that he who understands it must have a precise idea of the different illocutionary acts it could accomplish; and this is exactly why he would hesitate to give his opinion in the case in question. Austin thought, as you know, that the theoretical possibility of eliminating ambiguity in all cases by using an explicit performative probably corresponds to a relatively late stage in the evolution of languages. The explicit verbal markers with illocutionary force represent only the most perfect and effective among a whole series of linguistic devices, which include verbal mood, word order, certain adverbs and particles, intonation, accentuation, etc. Thereupon, if it is conceded that in a sufficiently developed language the exact illocutionary type of a speech act could in principle always be specified by dint of a performative verbs, it is reasonable to suppose that an enumeration and a classification of that type of verbs in a language constitutes an important step in the direction of an inventory and a typology of the speech acts possible in that language. You know that Austin assumed that an enumeration of the explicit performative verbs - or more generally of the verbs that make explicit the illocutionary force of an utterance would produce, for a language like English, a list comprising between 1,000 and 10,000 elements. He proposed, obviously without great satisfaction, a preliminary division into five main classes: the verdictives, the exercitives, the commissives, the behabitives, and the expositives.126 Whatever objections one might raise against Austin's classification - he himself raises some - it is nevertheless admissible to think that such a typology should be theoretically possible and that it would probably have a universal value. It is not totally absurd to suppose that the indefinite multiplicity of "the things we can do with words", to speak with Austin, is regulated as a 126

J. L. Austin, How to Do Things with Words? (Oxford, 1962), Lecture XII.

388

JACQUES BOUVERESSE

function of a few main categories which, to join Chomsky this time, are part of the definition of what language is in general, belonging to the 'essence' of human language. Yet it is certainly impossible to construct a priori the detailed list of the elements which can occur under each of the headings in question. The constitutive rules which govern the accomplishment of illocutionary acts in a language are not only linguistic rules in the narrow sense; they arise from the existence of appropriate institutional practices - that is, they refer back to things like political, judicial, and economic systems, religions, ceremonies, myths, rites, etc. I am, of course, not suggesting that linguistic institutions have a priority over the others nor do I suggest the opposite. I am merely suggesting that linguistic behavior is always conditioned by institutions other than language as well. I will take up the same question again but in another form: do you believe in the predictability of speech acts ? - That is to say, to the fact that linguistic description, or maybe more precisely linguistic theory, should allow the prediction of .... Yes... - Certainly not. Because, for that, it would have to be possible to determine a priori not so much the totality of the things one can do with language, but the totality of the things one can generally do in which language is likely to be involved in one way or another. One would have to know in advance somehow all conceivable institutional behaviors which may have a linguistic aspect, all the 'language games', if you want. As I have pointed out, Wittgenstein excelled in the practice of 'speculative anthropology', that is, in the invention of very surprising fictitious forms of behavior and language use. The detailed prediction of speech acts, if it is a sensible venture, should in reality be accomplished at least as much by the anthropologist as by the linguist. I would like to talk to you about yet another unusual notion in linguistics: that of intention as opposed to that of convention. Neither structuralist linguistics nor generative grammar consents to formalize the speaker's intentions in the semantic representation. One could, then, wonder whether a technique of semantic representation which does not take these intentions into account may be considered sufficiently complete and general? - The notion of intentionality is not only suspect to the linguist but also to many philosophers, and maybe not entirely without justification. Do not forget either that generative grammar merely pretends to account for the speaker's faculty to produce and understand an infinite number of sentences. It does not at all try to solve the question why, that is, more specifically, with what intention the speaker produces these sentences. What Katz and Fodor call a semantic theory is a mechanism which associates semantic interpretations with syntactic structures by a purely formal procedure and without the intervention at any moment of the problem by what kind

JACQUES BOUVERESSE

389

of being and with what intention the sentences are formed and used. The only intention at issue on the level of the general theory of language is the communicative intention in the most vague sense of the term, the wish to exteriorize 'ideas'. For a partisan of the theory of speech acts like Searle, accomplishing a linguistic act consists generally of saying something with the intention of producing certain effects on a listener; and the possibility of obtaining these eifects is determined by the constitutive rules which cause the utterance of the relevant sentence in appropriate circumstances to have these very effects. Thence "semantic competence is largely a matter of knowing the relationships between semantic intentions, rules, and conditions specified by the rules". 127 The semantic competence of which the transformational linguist speaks represents a very small aspect of the speaker's real semantic competence at best, because it lacks both the reference to the speaker's intentions and the essential semantic rules, those which form a link between the grammatical sentences of a language and the speech acts they serve to accomplish. What is especially contestable in Chomsky, however, is probably his consideration "that sentences are abstract objects that are produced and understood independently of their role in communication". 128 He is otherwise right in saying 129 that we can consider the incorporation of grammatical description into a general theory of human behavior, including a good deal of the analysis of linguistic acts, but that we cannot consider requiring grammar to be at the same time a theory of language and a theory of linguistic behavior. If we consider how things actually happen, we observe cases where understanding the exact intention of a speaker amounts almost to understanding the meaning of the sentence he is uttering, as is given or should be given by the semantic description of this sentence, and, in the other extreme, cases where the function of an utterance and the effect it seeks to obtain in the listener do not correspond much to the meaning of the words used in it (this is the case, for example, with 'phatic communion' described by Malinowski, where the 'social' meaning of the utterance is almost totally independent of its linguistic meaning). Of course, the majority of the cases lie between both extremities. A sentence such as "Do not go there!" may be used as a command, a warning, advice, request, etc.... I think that Chomsky is justified when he says that, in these conditions, "semantic theory should entail that the semantic representation or logical form of this sentence makes it available for certain acts, not others". 130 It can certainly not be requested of the semantic component of the grammar that it allow the deduction of all the propositions of the form: "The utterance of sentence X in context Y counts as the accomplishment of speech act Z". Because in certain cases virtually infinite precisions would be needed of the type "Sentence X uttered with such an intonation, by such or such a person, addressed to such or such another person, in such or such linguistic 127 128 139 130

J. Searle, "Chomsky's Revolution in Linguistics", p. 23. Ibid. Cf. p. 45 of this book. Cf. p. 45 of this book.

390

JACQUES BOUVERESSE

and extralinguistic circumstances, etc...". The question here is simply that formulated by Ducrot: whether a properly linguistic and relatively systematic semantic description can exist or not. If one does not want to be placed in a position where one has to admit that any linguistic expression can serve as a vehicle, by virtue of appropriate conventions, of almost any message according to the circumstances, it is necessary that a sentence such as "What nice weather!" (example of Ducrot) could obtain a fixed linguistic meaning, before the fact, and independent of the fact - that it may be used in a totally normal way to 'say' things like "What abominable weather!" or "We do not have anything particular to say to each other", etc. Could it be said that the illocutionary force of an utterance, to the extent that it is not entirely determined by the linguistic meaning of the sentence which supports it, is determined by other conventions of one kind or another ? - I think Strawson justifiedly says that it is not generally true. 131 There are probably cases where the illocutionary force of an utterance, even though it would not be completely exhausted by its meaning (in Austin's sense), is not conferred on it by special conventions which would intervene in addition to those that contribute to giving it its meaning. To say that the exact strength of an ambiguous illocutionary utterance is normally determined by the circumstances of the utterance is not necessarily saying that the circumstances possess this 'disambiguation' force by virtue of conventions of one type or another which could be explicitly formulated. Strawson gives examples of such cases, but I would especially like to remark on that subject that even though the speaker has at his disposal a certain number of conventional linguistic piocedures which would allow him to make the type of illocutionary act he wants to accomplish perfectly explicit, we must consider as one of the essential communicative features not only the fact that he does not always feel the need to use this possibility, but also the fact that his illocutionary intention might not be absolutely clear to him at the moment : that the reaction of the interlocutor may act in return upon the exact determination of this intention; that one can even deliberately look for a certain 'illocutionary vagueness'; that it may, moreover, be imposed by social conventions, etc. In short, I mean that one should not lose sight of the concrete circumstances of the dialogue and one should particularly not forget that the speaker's problem is not always the exact explicitation of his intentions but, in many cases, their partial 'implicitation', the intimation of certain things without having really said them and retaining the option to deny having said them. This problem of the implicitation techniques which are operative in the most ordinary use of language has been, as you know, particularly studied by Ducrot. He makes a distinction between those that must be assigned to the linguistic component (like the presupposition (présupposé)) and those that must be assigned to the rhetorical component (as the implication (sous131 P. Strawson, "Intention and Convention in Speech Acts" (1964), in Logico-litiguistic pp. 149-169.

Papers,

JACQUES BOUVERESSE

391

entendu)), in the total semantic description. If I mention him again here, it is because he provides, in his analysis of the rhetorical component, some elements for an answer to your question. He tries, in fact, to formulate some general laws or conventions which, applied to pre-existing meanings, allow the accomplishment of a sort of 'calculus' of the non-literal, implied meaning. For him, the implication is discovered by a kind of discursive procedure, of reasoning, which operates not only on the basis of the sentence itself, but also on the basis of the event represented by the decision to use a sentence at a given moment and in fixed circumstances. Can it be said that all this also leads in the direction of the obliteration of the distinction between semantics and pragmatics ? - Yes and no. In a way, yes, if one considers that, for a linguist like Ducrot, the theory of presupposition, that is, of the act of presupposing, is part of semantic description. No, in another sense, if the fact is taken into account that the distinction of the linguistic component and of the rhetorical component has as its main goal allowing the isolation of a stable element in the meaning of the sentence, of a meaning invariant, on which contextual determiners can then operate. At stake is a sufficient delay in introducing certain variable semantic elements which could not be included without serious inconvenience in the description of the sentence, and they must be related to the utterance act itself: "The gradation of the two components makes it possible to parenthesize temporarily certain semantic elements which are undesirable for certain calculi and to insert them only after these calculi have been executed (it seems to us that the juxtaposition of the base component and the transformations in generative grammar often has the same function)." 132 This obviously means that the distinction between the semantic component and the rhetorical component has very little in common with the distinction between the semantic aspects proper and the pragmatic aspects of sentence use. Ducrot does not worry much more, in fact, about the distinction between semantic considerations and pragmatic considerations than Greimas about the distinction between syntactic categories and semantic categories. It is of no concern to know at what point the explicit reference to the user intervenes, because it is somehow present at the onset. The constitution of an independent linguistic component has an essentially operative justification. It seems to be highly desirable for obtaining a complete semantic description; and the criteria for the linguistic component, in contrast to the rhetorical one, are, on the one hand, the property of being able to be represented systematically, on the other hand, the fact that the meanings produced at this level allow the constitution of a rhetorical component which would satisfactorily fulfill its function that is, by mechanisms which would not be totally arbitrary and ad hoc. Finally, on this question, I would like to say that the question whether the semantic description should make reference to the speaker's intention or not is not clear to me. When one speaks about the speaker's 'intention', one never knows very well whether 132

O. Ducrot, Dire et ne pas dire, p. 136.

392

JACQUES BOUVERESSE

it is about what he meant, what he had in his mind (and which he may not have wanted or dared to say exactly or which he was unable to say), or what what he said means according to the semantic rules of the language concerned and the circumstances of the enunciation; one never knows very well either whether it is about what he has wanted to make known or understood, or about the perlocutionary effects which he possibly wanted to arouse in the listener. But it may be said that understanding a concrete utterance accomplished in fixed circumstances is mainly understanding the 'audience-directed intention' for which it is a vehicle; inversely, understanding the intention to which it corresponds in the speaker is nothing more than understanding the meaning a complete semantic description makes possible to attribute to it in the exact context. The sentences would mean nothing for those who use them if they would not express intentions. But one can only express an intention by means of a sentence; one can only mean something by it, because it means something which is partly determined by the semantic rules of language and partly by the context. I merely want to recall here a banality of the following kind: if we arrive at the recognition of the speaker's intention, this is not by intropathy, but because we interpret his words correctly. I am still returning to the same problem. Another sign of the disappearance of the distinction between semantics and pragmatics, which seems to me very important for the future of linguistics, is the introduction of what Grice calls the "conversational postulates", that is, there are postulates proper to conversation itself, to dialogue. - Yes, certainly. One can require that "conversational postulates" explain, for example, why sentences such as "Isn't the weather nice today?" and "Would you do that?" - sentences which have the linguistic form and the intonation of questions are actually used as an affirmation and a command, respectively.133 But the problem posed by the conversational postulates, or by any other system of conventions of that type, is that they must at the same time predict effects of very special contextual meanings and that they must do so according to sufficiently general principles, that they should not be maxims introduced to suit the purpose in each individual case. Anyway, as Chomsky notes,134 the possibility and the necessity of introducing things like conversational postulates for the explanation of meaning effects which the grammar cannot explain in any natural way does not condemn, but rather justifies, the idealization which is at the basis of the generative theory. I would like to ask you a last question in this field of the "semantics of the future": what do you think of the conception according to which it should be possible to construct all semantic theories on the basis of a certain number of unanalyzable entities, of semantic 'primitives' ? - This conception, even as a mere working hypothesis, raises enormous difficulties. 133 134

L. Wittgenstein, Philosophical Investigations, § 21. Cf. p. 45 of this book.

JACQUES BOUVERESSE

393

The basic idea of 'componential' semantics is that the semantic structure of a language can be reconstructed on the basis of a small number of elements, or atoms, of meaning, and of their possible combinations in various languages. I have already pointed out that a semantic theory which uses things like 'components', 'markers', or 'sememes' has probably no decisive advantage over a theory like Carnap's, which uses 'meaning postulates', for the solution of the problem of synonymy and analyticity. But many other questions can be asked. For semanticians of the Katz-Fodor type, the semantic markers are theoretical entities, postulated almost on the same grounds as atoms, electrons, or genes. Yet one may consider that the semantic content of a language could be fragmented endlessly in many ways and that there are no reasons to believe that it is possible to stop at indivisibles selected in an unarbitrary way and assumed to be common to all languages. It has not been remarked to what extent the Chomsky-Katz-Fodor conception comes close to the Leibnizian idea of a universal characteristics. Leibniz considered it theoretically possible to construct a sort of alphabet of human thoughts, which would assign biunivocally simple signs to elementary ideas and which would allow the representation, by appropriate combinations of these signs, of all the manners of composition of elementary ideas into complex ideas. The semanticists of the trend I have mentioned have a rather analogous idea: that it should be possible to express the semantic content of all languages in the end by means of a sort of universal semantic alphabet, comparable to the universal phonetic alphabet. In their conception, the 'semantic features' are prevented from being theoretical terms adopted rather arbitrarily by the linguist by the clearly ideational interpretations they give of them. The elementary semantic components represent the basic categories and principles by which the human mind structures and classifies the phenomena; they represent the internal mechanisms which operate in the processing of the physical and social environment. Hence arises the hypothesis that, as Bierwisch writes, "all semantic structures might finally be reduced to components representing the basic dispositions of the cognitive and perceptual structure of the human organism". 135 If this very strong hypothesis was verified, we would obtain a description of the innate constitutive properties of the human being, no longer considered as a syntactic animal but as a semantic animal. However, the least one can say is that it is far from being accepted by everybody, not only because it is probably not very plausible and that, in any case, its concrete verification is presently merely Utopian, but also because it corresponds to a general orientation whose basis is debatable and whose explanatory virtues are probably overrated. It is, without a doubt, not necessary to repeat here what has already been said on the problem of innateness, on the ideational and conceptualist theory of meaning, and on the overestimation of the cognitive aspect, in contrast to the communicative aspect, in the study of semantic structure. 135

M. Bierwisch, "Semantics", in J. Lyons (ed.), New Horizons in Linguistics (Harmondsworth, 1970), p. 181.

394

JACQUES BOUVERESSE

Another tendency one comes across in almost all the linguists who contributed to this book is a growing interest in logic. Talking about the logic adapted to natural languages, Lakoff uses the term 'natural logic'. I would like you to tell me, perhaps with respect to Wittgenstein's theories, which logic seems to you the most valuable in approaching natural languages. What is the status of 'natural logic' ? - Wittgenstein constantly talks, as you know, about the 'logic of language'. It is much more extended, much more diversified, much more complex and far less exact than what is usually called logic. In fact, where Wittgenstein talks about a 'grammar' or a 'logic' of language, many people would doubtlessly prefer to talk in many cases about a psychology or sociology of language. I have said that he also talked about the grammar or the logic of an entity or a process; and the 'logic' of a thing is something which is for him defined by the totality of what can meaningfully be said about this thing. He has rather soon come to think that the logical notation and the representation of the deep, latent grammatical form of the expressions of normal language by means of this notation generally give a totally inadequate picture of their important 'logical' characteristics, at least important for the philosopher. The transcription in logical language imposes on the linguistic expressions a totally disastrous excess of simplicity, uniformity, and piecision. Wittgenstein thought this true, after all, even in the case of mathematical language. With regard to the use of logical symbolism in mathematics, he writes: "The logical notation absorbs the structure." 136 Nevertheless, he would certainly not have said (he did not even say it at the time of the Tractatus) that ordinary language is 'illogical' or merely logically imperfect. For him, ordinary language really only remotely resembles the ideal languages of the logician; but in a way there is no language which would be more 'ideal' than ordinary language. Inversely, however, it can probably be said that the expression 'natural logic' is somewhat misleading. It suggests rather easily that the other logic is not 'natural', that it is somewhat artificial and arbitrary, that it is a rather gratuitous invention of the logician. This is, moreover, a point of view which the logicians themselves have often encouraged, directly or indirectly. In fact, formal logic represents a part, and an important part, of natural logic, of the logic of our language: that which has reference to the rules of deductive inference, in the narrow sense of the term. Let us not forget either that one has felt the need to propose so-called methods of 'natural deduction' which are supposed to correspond more exactly than the axiomatic presentations to the mechanisms actually governing ordinary reasoning. One can say, with Lakoff, 137 that the development of symbolic logic results essentially from the discovery of the fact that the regularities underlying the ordinary reasoning processes can only be explained by means of rules which stand in relation not to the surface grammatical forms of sentences but to the corresponding logical forms. What Lakoff proposes to 136

L. Wittgenstein, Bemerkungen iiber die Grundlagen der Mathematik (Oxford, 1964), p. 147. G. Lakoff, "Linguistics and Natural Logic", in D. Davidson and G. Harman (eds.), Semantics of Natural Language, p. 646.

137

JACQUES BOUVERESSE

395

call 'natural logic' results from a two-directional convergence: linguistics toward logic, and logic toward linguistics. On one side, there is the preoccupation of certain linguists, like Lakoff, to approximate the grammatical form of a sentence to its logical form - to show that the correspondence obviously existing in certain cases between the linguistic structure and the logical structure is not accidental. On the other side, for quite a while already but increasingly more so today, logicians feel the need to construct formalisms which allow the representation of very diverse and relatively neglected aspects of the logic of language. For example, the interest of professional logicians has for a long time been almost exclusively directed to the declarative sentence and to the logic of affirmation. Yet there are also attempts - which are not all recent - to constitute logics of command, of obligation, of questions, etc. Could it not be said that there is a problem resulting from the fact that 'natural' logic must largely use certain parts of logic which are considered peripheral and of a relatively secondary interest ? - Yes, this is certainly an important aspect of the problem about the relation between logic and linguistics. Some logicians are loath to extend the term 'logic' beyond the well-defined and relatively restricted domain represented by traditional symbolic logic. They consider disciplines such as modal, deontic, epistemic, and other logics as rather exterior appendices to the field of logic proper, and in the extreme as mere technical curiosities. Some logicians look rather condescendingly or even contemptuously on the works of people like von Wright or Prior. It is remarkable that the considerable progress accomplished in the last years in modal logic, or perhaps more exactly in the interpretation techniques of the modal calculi, greatly interests the linguists, in particular the theoreticians of generative semantics. A discipline which has long had an almost exclusive theoretical and esoteric interest can become concretely usable by others the day it becomes sufficiently refined and mature. It seems that the work of people like Hintikka, Kanger, Kripke, Dana Scott, etc., open up new perspectives for linguistic semantics; and one can only be happy about it. It is certain that any progress realized in the domain of what can generally be called 'intensional' logic - which is probably no longer the poor relative of logic will necessarily benefit, directly or indirectly, linguistic theory. As I have talked about the general attitude of the logicians, a word should probably also be said about that of the linguists. I think they have a tendency to ignore, or at any rate clearly underestimate, the resources offered by formal logic in matters of the analysis of ordinary discourse. It is certain that much more has actually been done in that field and that much more can probably be done than they generally suspect. The logical instrument is more powerful and refined than some of them seem to believe, although it is not as powerful as certain philosophers naively imagine. This being said, I think it must be resolutely admitted, as Weinreich does (who, moreover, proves himself a man with remarkable consideration for Reichenbach's attempt at

396

JACQUES BOUVERESSE

"analysis of conversational language"),138 that "the descriptive linguist has no interest in making language usage 'more logical' than it is - on the contrary, he should explain, if possible, why it is not, in effect, more logical".139

VI. O N PHILOSOPHY OF L A N G U A G E

We are slowly arriving at the end of our conversation. In conclusion, I would like to ask you some questions that can be posed only to a philosopher of language - but to which we have already alluded in passing. It may be assumed that the philosopher of language is very concerned with three problem areas: first, the area of questions which are concentrated around the relation of language to experience; then that of questions which refer to the referential function of language; and finally, that of the questions concerning the relation of language to the speaker. With regard to the problem of the relation of language to experience, the weight of behavioristic considerations still remains very strong without a doubt. How should the relation of language to experience be formulated and what are the consequences of this formulation with regard to the idea one can have of how language is learned? - Saying that language has no simple and direct relation to experience (for example, that it cannot have the kind of relation with experience implied by certain forms of empiricism or behaviorism) is not at all the same as saying that it does not have an essential relation to experience. It is absurd to imagine that learning to speak consists of learning to associate linguistic signs to elements already registered and classified in experience. In large measure, learning to speak is precisely learning to organize, structure, and conceptualize the data of immediate experience. The main problem posed by language learning is that, in a way, language is never related to experience, but only to experience already formed by language. It is characteristic of strict behaviorism to consider that the meaning of an utterance may be identified with "the situation in which the speaker utters it and the response which it calls forth in the hearer", as Bloomfield said. But as not all the elements of the situation are supposed to have a causal influence on the speaker's verbal productions, but only those that are relevant from the point of view of linguistic meaning, Chomsky is certainly justified when he says in his criticism of Skinner140 that the explanation of the notion of meaning in terms of a response to external stimuli is in a way totally circular, and thus empty. Yet, under these conditions, I really do not understand why Chomsky insists on linking Wittgenstein to the empiricist and behaviorist tradition, as Wittgenstein's observations proceed in exactly the same direction as his. To realize this, it suffices 138

Cf. H. Reichenbach, Elements of Symbolic Logic (London, 1947), Chapter VII. U. Weinreich, "On the Semantic Structure of Language", in J. H. Greenberg (ed.), Universal of Language (Cambridge, Mass., 1963), pp. 148-149. 140 N. Chomsky, "A Review of B. F. Skinner's Verbal Behavior", in J. J. Katz and J. A. Fodor (eds.), The Structure of Language, pp. 552-553. 139

JACQUES BOUVERESSE

397

to consider, for example, what he says about the ostensive definition and the role it can play in language acquisition. He is insistent on making us suspicious of the temptation to overestimate this role - to believe that one can learn a language mainly by ostensive definitions, that is, by the establishment of associative relations between words and the things in whose presence they are uttered. 141 No one has insisted more than Wittgenstein on the amount of understanding one should already possess to understand an ostensive definition, on all the intralinguistic and extralinguistic connections which must already exist so that a connection of the kind required by the so-called ostensive learning of the meaning of a word may be established. It is, therefore, surprising to see Chomsky attributing to him the idea that there are mainly two ways to learn the meaning of a word: the verbal definition and the ostensive definition, and that because the definition of one word by another must necessarily in the end be circular, "it is ostensive definitions that are crucial for language learning". 142 Wittgenstein certainly did not believe that language learning could rely essentially on ostensive definitions because he was, as I have said, particularly conscious of all that the mastery of an ostensive definition already presupposes as language mastery - of all that should already be taught to someone before and simultaneously as one teaches him a language by the technique of ostensive definition. In any case, Wittgenstein never proposed, properly speaking, a theory of language acquisition; and he would never have suggested that this acquisition could be explained entirely by the use of two or three well-defined procedures, such as the ostensive definition and the verbal definition. He evidently thought that things are much more complicated than that. He did not at all believe that the words 'derive' their meaning from experience. He considered that the justification by experience has very strict limitations, already determined by 'concept formation' (Begriffsbildung). I am very pleased that you stressed this point, because there are a number of traps in the tendency that attracts people to sociolinguistics. It is not always very clear that sociolinguists accept intermediary systems between direct (social) experience and language. - You are alluding, I suppose, to remarks such as Halliday made : "Language is a part of the social system, and there is no need to interpose a psychological level of interpretation." 143 This is actually very debatable. It can be explained in Halliday by the fact that he considers what the speaker can say as the realization of what he can mean, and what he can mean as the realization of what he can do. That is, the lexicogrammatical system is the realization of the semantic system, namely, of the 'meaning potential' incorporated in language, and this is the realization of the ''behavioral potential' conceived as a higher level semiotic system, a 'social semiotics'. In other words, Halliday chooses to proceed from language in the direction of sociology, and 141 142 143

See especially the whole beginning of the Philosophical Investigations. N. Chomsky, "Some Empirical Assumptions in Modern Philosophy of Language", p. 275. Cf. p. 85 of this book.

398

JACQUES BOUVERESSE

not of psychology or of a general theory of the perceptive, conceptual, and cognitive faculties of the human mind. He interprets the meaning potential of a language directly in terms of socio-behavioral potential; that is, he deliberately evades the detour by the conceptualist and cognitivist interpretation of the linguistic semantic system. However, it is probably not simply a rather optional intermediary level, as he believes it to be. For Chomsky, it is evidently a basic level: what the social subject can do, in the sense of Halliday, is determined for an essential part by what he knows, and by the way in which a human being in general knows and can know. By wanting to throw off the notion of competence entirely and avoid talking about "what happens in the mind" of the speaker, Halliday may have run the risk of throwing away the baby with the bath water. Are you willing to address the criticism you have made of a certain conception on the relation between language and reality to the Tractatus and Wittgenstein's picture theory of language ? - Yes, evidently. It is difficult, moreover, to criticize this theory more efficiently than Wittgenstein has done himself afterwards. Y e t there are, in fact, two exaggerations in the Tractatus: on one hand, an exaggeration of the importance of the representative function of language; on the other, an exaggeration of the 'resemblance' which exists between language and what it represents. With regard to language and the world, I was just thinking about certain tendencies that can be observed in contemporary linguistics, for example, in Lakoff and McCawley, who say that there is also a relation between language and the possible worlds, not only between language and the actual or real world. - Yes, it so happens that the recent developments in modal logic have made it possible to give a much more precise content to the traditional idea of 'possible world' or 'class of possible worlds'. That must apparently be beneficial for linguistics. If one thinks about it, however, one notices that, from the moment that one simply makes the distinction imposed between meaning and reference, and between the meaningfulness and the truth of a proposition, one is obliged to admit that language refers essentially to possible worlds and not to the real world. Language must describe possibilities rather than realities if one desires that a false proposition nevertheless be meaningful. It is not necessary to understand that language constantly refers to imagined, wanted, wished, desired things to be persuaded of that. This is true even when language is used for describing what happens in the real world, "what is the case" (way der Fall ist), as Wittgenstein says. When Wittgenstein says in the Tractatus that language is an image or a mirror of the world, he does not at all mean that it is a reproduction of the real world. Language reflects in its essence the essential, logical properties of the world, those that are common to all possible worlds; and it allows the representation, by means of contingent propositions, of that possible world which is realized, which is the actual world. Assigning truth values to all

JACQUES BOUVERESSE

399

the elementary propositions of language determines a possible world with the exception of its existence or non-existence. "A proposition", Wittgenstein writes, "constructs a world with the help of a logical scaffolding, so that one can actually see from the proposition how everything stands logically if it is true. One can draw inferences from a false proposition." 144 Structuralism has made us familiar with the fact that the speaker would be absent in language. Linguistics would be, in principle, 'anti-humanist'. How is the speaker present in the speech act? How is he present in language? And are these one and the same question ? - I am in a bad situation to answer this question. I have never understood where exactly the statement on the obliteration or the disappearance of the subject tried to lead. Certainly, there has been at a certain moment in France an extremely violent reaction against the idea of a metaphysical subjectivity of a Cartesian or Husserlian kind, and generally against any kind of philosophy of consciousness - mainly on the basis of data derived from linguistics, psychoanalysis, and anthropology. For my part, however, as I have never thought much of this kind of philosophy, I do not feel obliged now to think as derisively about it as is generally done. It is possible that, as Michel Foucault says, man is a recent invention, but the idea of the disappearance of man is an even more recent invention and without a doubt even more ephemeral. In fact, the philosophical criticism of the notion of subject and the constitution of 'no-subject-theories' are far from new. Wittgenstein, for example, professed a theory of that kind at the time of the Tractatus and in the decades of the thirties. If one really clings to saying that there is thought without a thinking subject, speech without a speaker, and finally, if one considers language from the point of view of the theory of speech acts, action without an actor, I see no inconvenience in it. If one does it to be able to say, as is generally the case, that it is the language itself, the unconscious, or things like that, which speak, I do not see very well where the benefit is. Substituting the question " What speaks?" for the question "Who speaks?" does not lead us far if we are obliged to adhere to answers of that kind. Is it not possible, in your opinion, to say that the subject is an intralinguistic function in Jakobson's and Benveniste's sense... ? - What do you mean by that exactly ? That there is a certain break in language itself. The enunciation is viewed as the presence of the pronouns T and ' You' in language. - In truth, I think especially of Benveniste, who states that language, by providing the speaker with the means to designate himself with the pronoun 'I', does something very different than what it does when it provides him with the means to designate other 144

L. Wittgenstein, Tractatus Logico-philosophicus,

4.023.

400

JACQUES BOUVERESSE

speakers or objects, something very specific and basic. The "transcendence of the ego" has, and has in a way exclusively, its ground in the linguistic possibility thanks to which he can "appropriate" the whole language by designating itself as he who speaks just as he speaks. For Benveniste, "It is in and by language that man constitutes himself as subject; because only language establishes in reality, in its reality, which is that of being, the concept of 'ego'." 145 Correlatively, "language is only possible because each speaker denotes himself as subject by referring to himself as I in his discourse." 146 In fact, for Benveniste, language itself is constitutive of subjectivity and creates the notion of person, because only the use of language allows the subject to become conscious of his permanence and his identity. But the 'nosubject-theorists ' could obviously object that language is less a generator of subjectivity than a generator of the illusion of subjectivity, that the idea of a subject is a 'grammatical' illusion which results from certain particularities (perhaps even universal) of our language. Is the Chomskyan 'mind' a subject? - No, not necessarily. Nothing in the Chomskyan linguistic description absolutely presupposes that language be used by a subject, in the sense it is usually understood. One could doubtlessly say, all in all, that it is used by a superior biological organism, gifted with an appropriate central nervous system. Behind the fact that there is an author of the utterance, there is probably nothing more than the material fact that all speech pronounced comes necessarily from someone's mouth, that someone is its producer in the neurophysiological sense of the term. On the other hand, with its insistence on the ideational and expressive aspect of language use, its tendency to treat communication as a (relatively) secondary phenomenon, and to ignore what Benveniste calls "the polarity of the persons", Chomsky's language philosophy has a clearly subjectivist, even solipsistic connotation. I would now like to take up two basic criticisms which have been formulated by the linguists with regard to analytic philosophy. The structuralists, in general, reproach it for not stressing the structure of language sufficiently, for having a tendency to underestimate the extremely organized and systematic character of language. - The notion of 'systematicity' has no absolute meaning. It is an essentially relative notion. When Carnap talks about the intrinsically irregular and non-systematic character of ordinary languages, he means, above all, that they really only remotely resemble the artificial languages constructed by the logician. He would certainly not have said that they do not possess any kind of systematicity and structure. I think the semantics of natural languages would have to be in a much better state than it presently is before Carnap's position could be considered totally obsolete. 145

E. Benveniste, "De la subjectivité dans le langage", in Problèmes de linguistique (Paris, 1966), p. 259. 146 Ibid., p. 260.

générale

JACQUES BOUVERESSE

401

Yes, but I was thinking especially of ordinary language philosophy... - Oh, sure.... I have already partly answered that objection. People like the second Wittgenstein, Austin, Ryle, etc. think that in order to solve philosophic problems, one must have very fine distinctions provided on request and point by point, and not a general theory of language. The position of people like Katz is, in my opinion, totally improper: on one side, he suggests that transformational linguistics is in a position to solve the problems analytic philosophers have failed to solve; on the other, he reproaches the philosophers for not having tried to do the work of linguists namely, constructing a systematic theory of the structure of the language considered. In short, he would want the philosophers to transform themselves into linguists, and he suggests that their problems would somehow be solved to boot. Yet, I see quite well what linguistics could have gained, for example, if Wittgenstein had decided to practice linguistics seriously; but I also see very well what philosophy would have lost. The criticism formulated by transformational grammarians against the method of the analytic philosophers mainly has reference to the fact that they have shut themselves up far too much in the description of surface phenomena, and that deep structures and also the generative aspect of language have been ignored by them. - It can certainly not be said that the analytic philosophers in general have been interested exclusively in surface linguistic phenomena. The philosophic method of people like Russell, the first Wittgenstein, and Carnap stems, on the contrary, directly from the recognition of the fact that the apparent, surface grammar of sentences does not necessarily correspond to their real, deep grammar. All the analytic philosophers, including those of Cambridge and Oxford, have actually concentrated their attention mainly on certain particular characteristics of language which (unfortunately) have no immediate character, which do not appear directly in the observable linguistic structure of expressions. As to the generative aspect, they have, indeed, almost ignored it. This is because they believed, rightly or wrongly, that philosophic problems can essentially be solved by the logico-linguistic analysis of isolated words, expressions, sentences, or reduced fragments of language, but not by a total description or theory of the language to which they belong. All this leads me to my last question. As the linguists adhere to the autonomy of linguistics, you adhere, I have the impression, strongly to the autonomy of the philosophy of language. - No, I do not really cling to the autonomy of the philosophy of language, but rather to the autonomy of philosophy. At stake is above all to know with what invariant properties (if there are any) the word 'philosophy' is used each time one talks about 'philosophy of language', 'philosophy of mathematics', 'philosophy of art', 'philosophy of religion', etc. In short, what matters is to know what generally characterizes the so-called 'philosophic' approach to a question, and in particular

402

JACQUES BOUVERHSSE

to know whether, as certain Chomskyan linguists believe, a problem of philosophy is a problem of theory, linguistic or otherwise. I can obviously give no categorical answer to this question. In fact, in the debate which is now taking place around the question about exactly what services linguistics can render to philosophy, there are two different problems. There is first the question whether linguistic progress has a direct impact on certain problems of traditional philosophy, for example those concerning the nature of the mind. Chomsky obviously thinks that the discipline entitled the 'philosophy of mind' may expect a lot from linguistics. This is actually a way of saying that the problems with which this discipline is preoccupied have ceased to be philosophic problems, that we have now the possibility of approaching them scientifically by formulating empirically testable hypotheses. There is, on the other hand, the question whether, assuming that one admits that the philosophic problems must be solved in all cases by the analysis of language, the methods and results of contemporary linguistics should be employed. Both questions are evidently independent, because there are many philosophers who are not at all convinced that the method of philosophy must be a linguistic method, specific or borrowed rather directly from linguistics proper. But, even if the basic presupposition of analytic or linguistic philosophy is accepted, it is not at all a priori evident that linguistics can substantially contribute to the solution of philosophic problems. It is even less evident a posteriori, in particular because of the absence of semantic results comparable to those obtained in the field of phonology and syntax. Chomsky gives a frankly positive answer to the first of the two questions I mentioned above, but he is clearly more reserved about the second. In particular, he does not agree with Vendler147 in considering that contemporary linguistics could provide a "new technique" for analytic philosophy, which would be truly effective, at least not in the present state of affairs. I am at least as sceptical as he is in that respect, but it must be clear from our conversation that I am not any less sceptical about the first. I believe that most of Chomsky's philosophic conclusions are at least as disputable as the philosophic conclusions of the traditional kind (I mean, the conclusions of professional philosophers), and that, in any case, he himself has not entirely avoided the philosophic illusions and confusions to which Wittgenstein refers. In short, I believe that Chomsky's works constitute at least as much an object for philosophic criticism as a positive contribution to the progress of philosophy. Returning to your question, I think that a relatively strict distinction in principle can be made between things like linguistics or linguistic theory, the philosophy or epistemology of linguistics, the philosophy of language, and linguistic philosophy.148 Yet if there is one thing which is clearly apparent from the discussions which have 147

Z. Vendler, Linguistics in Philosophy (Ithaca, N. Y., 1967); N. Chomsky, "Linguistics and Philosophy", in S. Hook (ed.), Language and Philosophy, pp. 51-94. 148 Cf. for example, Z. Vendler, "Linguistics and the a priori", in C. Lyas (ed.). Philosophy and Linguistics, pp. 247-248.

JACQUES BOUVERESSE

403

taken place in recent years between philosophers and linguists, it is, on one hand, the fact that neither philosophy nor even linguistic philosophy runs the risk of being absorbed by the three other disciplines I have mentioned, and, on the other hand, the fact that none of the disciplines in question, not even philosophy, can brag about being totally independent of the others. Centre National de la Recherche Scientifique

(CNRS)

Paris, France March 24 and April 18, 1973

BIO-BIBLIOGRAPHICAL NOTES

Jacques Bouveresse was born in 1940 in Epenoy (Doubs), France. He completed his studies in philosophy at the Ecole Normale Supérieure, and he obtained his degree in philosophy in 1965. He has instructed at the Sorbonne, the University of Paris I, and is presently a research fellow at the National Center of Scientific Research (CNRS). His works have mainly dealt with Anglo-Saxon analytic and linguistic philosophy and with Wittgenstein in particular. He is the author of La Parole malheureuse (1971) and of Wittgenstein: la rime et la raison (1973).

Wallace L. Chafe was born in Cambridge, Massachusetts, in 1927. He received his Ph.D. in Linguistics from Yale University in 1958 and was subsequently employed by the Bureau of American Ethnology of the Smithsonian Institution in Washington as a specialist in American Indian languages. Since 1962 he has been a member of the Department of Linguistics of the University of California at Berkeley. His major research has been in two areas: American Indian linguistics and linguistic theory. Major publications include Seneca Thanksgiving Rituals (1961), Seneca Morphology and Dictionary (1967), A Semantically Based Sketch of Onondaga (1970), and Meaning and the Structure of Language (1970). His current work stems from an interest in combining linguistic and psychological approaches to the study of semantics.

Avram Noam Chomsky was born in Philadelphia in 1928. He attended the University of Pennsylvania where he studied linguistics, mathematics, and philosophy. He took his Ph.D. at the University of Pennsylvania, although most of the research that led to the awarding of this degree was completed as a Fellow of the Society of Fellows at Harvard University between 1951 and 1955. Since 1955, he has taught at the Massachusetts Institute of Technology where he now holds the Ferrari P. Ward Chair of Modern Languages and Linguistics. In 1967 he delivered the Beckman Lectures at the University of California in Berkeley, and in 1969 the John Locke Lectures at Oxford University. Since 1965 he has been one of the leading critics of American foreign policy (American Power and the New Mandarins, 1969). His major publications in the field of linguistics are Syntactic Structures (1957), Current Issues in Lin-

408

BIO-BIBLIOGRAPHICAL NOTES

guis tic Theory (1964), Aspects of the Theory of Syntax (1965), Topics in the Theory of Generative Grammar (1966), Cartesian Linguistics: A Chapter in the History of Rationalist Thought (1966), Language and Mind (1968), Problems of Knowledge and Freedom (1971), Studies on Semantics in Generative Grammar (1972).

Algirdas Julien Greimas, director of studies in the Ecole Pratique des Hautes Etudes (Economic and Social Sciences), was born in 1917 and is a French linguist and semiotician of Lithuanian origin. Bachelor of Arts in 1939 (Grenoble), D. Litt. in 1948 (Sorbonne), he has been professor of French linguistics in the Universities of Alexandria (Egypt), Ankara and Istanbul (Turkey), and Poitiers (France). In 1965, he assumed the direction of general semantic studies in the Ecole Pratique des Hautes Etudes (Paris). His research, first limited to French (Dictionnaire de l'ancien français, 1968), has extended progressively to the general problems of semantics {Sémantique Structurale, 1966) and semiotics {Du Sens, 1970), seeking to apply semiotic methods to the fields of gestuality, spatiality, the analysis of poetic facts, ethnic literature, etc.

Michael A. K. Halliday was born in Leeds, England, in 1925. He took his B. A. at London University in Chinese language and literature; as a graduate student he studied linguistics, first at Peking University and then at Cambridge, where he received his Ph. D. in 1955. He has held appointments at the universities of Cambridge and Edinburgh, and at University College, London, as Director of the Communication Research Center. In 1965 he was appointed Professor of General Linguistics at the University of London, a post which he held until the end of 1970. He has lectured in many countries in Europe, Asia, and North America and has held Visiting Professorships at Indiana, Yale, Brown, and Nairobi. Currently (1972-73) he holds a fellowship at the Center for Advanced Study in the Behavioral Sciences at Stanford. Among his publications are Intonation and Grammar in British English (1967), Grammar, Society and the Noun (1967), A Course in Spoken English : Intonation (1970), Explorations in the Functions of Language (1973), Language and Social Man (forthcoming). His current work is concerned with various related aspects of language as social semiotic; it includes a functional-systemic description of Modern English, a sociolinguistic approach to early language development, a study of textual cohesion, and a general consideration of meaning in a sociological context.

Peter Hartmann was born in Berlin in 1923. He studied comparative and general linguistics and Indo-European and Oriental languages first at Humboldt University in East Berlin and later at the University of Münster, Westphalia. He received his Ph.D. in 1950 {Some Basic Traits of Japanese, 1952) and was appointed lecturer at

BIO-BIBLIOGRAPHICAL NOTES

409

Münster in 1953 on completion of the monograph Nominal Expressions in Scientific Sanskrit (1955). He has published in the field of general linguistics, especially in the field of general grammar, and exclusively in German: Word Class and Form of Predicates (1956), Typology of the Indo-European Language (1956), Problems of Linguistic Form (1957), Theory of Linguistics (1961), Theory of Grammar (1963). He was named full professor of linguistics at Münster University in 1956 and has been on the faculty of the University of Constance since 1969. He is a member of the German Scientific Council and of the Commission for Linguistics of the German Research Society and is the editor of the journals Folia Linguistica, Linguistic Reports, and co-editor of Foundations of Language. His principal interests are the theory of grammar, text linguistics, and the methodology of linguistics.

George Lakoff was born in New York. He received his Ph.D. in Linguistics from Indiana University in 1965 (published under the title Syntactic Irregularity in 1970), and he worked at the Computation Laboratory at Harvard University and later at the University of Michigan at Ann Arbor. After holding a fellowship at the Center for Advanced Study in the Behavioral Sciences at Stanford (1971-72), he now teaches linguistics at the University of California at Berkeley. He has published papers in various important volumes and in many leading journals on syntax and semantics; among his publications are "On Generative Semantics" (1971), "Linguistics and Natural Logic" (1972), and "Pragmatics in Natural Logic" (forthcoming). Currently his principal interests are natural logic, semantics and pragmatics, and the theory of grammar.

Sydney M. Lamb was graduated from Yale University with a B. A. in Economics and pursued graduate work in linguistics at the University of California, Berkeley. His dissertation was a grammar of Monachi, a California Indian language. After receiving his Ph.D. in 1958, he taught at Berkeley and directed a Machine Translation Project from 1958-64 with the support of the National Science Foundation. In 1964 he went to Yale University as Associate Professor of Linguistics, and he was named Professor of Linguistics in 1968. From 1966 to 1968 he was a member of the executive committee of the Linguistic Society of America. His research and publication interests have gone through three phases, from American Indian languages to computational linguistics to linguistic theory. His publications in the last field include Outline of Stratificational Grammar (1966) and several papers published from 1964 to 1973, some of which have now been reprinted in the collection Readings in Stratificational Linguistics (1973).

André Martinet was born in 1908 in a village in Savoy. He pursued higher studies

410

BIO-BIBLIOGRAPHICAL NOTES

in Paris. As a specialist in Germanic linguistics, he first studied for a year at the University of Berlin and dedicated his principal thesis to an examination of consonant gemination in Germanic. From that time he was in contact with the linguists of Prague and Copenhagen, and his complimentary thesis on Danish is the first complete phonological description of a language. Director of phonological studies at the Ecole des Hautes Etudes in Paris since 1938, from 1947 to 1955 he directed the Department of Linguistics at Columbia University in New York where he presided over the destiny of the journal Word. In 1955 his Economie des changements phonétiques appeared, establishing diachronic phonology. After returning to Paris, he instructed general linguistics at the Sorbonne and structural linguistics at the Ecole des Hautes Etudes. His Eléments de Linguistique Générale appeared in 1960 and is presently translated into a dozen languages. Two years later he published A Functional View of Language. Under his guidance the volume on language in L'Encyclopédie de la Pléiade and La Linguistique, Guide Alphabétique have been edited. He is the editor of the journal La Linguistique.

James D. McCawley was born in Glasgow, Scotland, in 1938 and has lived in the United States since 1944. He studied mathematics at the University of Chicago and linguistics at the Massachusetts Institute of Technology, receiving a Ph.D. in Linguistics in 1965. His dissertation was entitled The Accentual System of Modern Standard Japanese. Since 1964 he has been teaching linguistics at the University of Chicago; he has also served as a visiting faculty member for brief periods at the University of Illinois, Goteborg University (Sweden), the University of Michigan, the University of California in Santa Cruz, and Australian National University (Canberra). He has published papers in various journals and volumes on a variety of topics in syntax,, semantics, phonology, and the history of linguistics; a collection of his papers on syntax and semantics entitled Grammar and Meaning will be published shortly by Taishukan, Tokyo. His principal interests are semantic representation, the relationship between language and logic, and the typology of tonal and accentual systems.

Herman Parret was born in Belgium in 1938. He first studied Romance languages and literatures and later philosophy at Louvain University; he earned his Ph. D. in Philosophy in 1970. During that time he studied philosophy and linguistics at the University of Paris-Nanterre and at the Ecole Pratique des Hautes Etudes (Paris), and he spent long periods at the Universities of Freiburg-im-Breisgau, Oxford, and Berkeley, and was a research scholar in the Department of Linguistics of Massachusetts Institute of Technology. Currently he holds a research fellowship of the Belgian National Science Foundation. His major research has been in the methodology and epistemology of linguistics and in the philosophy of language. Among his

BIO-BIBLIOGRAPHICAL NOTES

411

publications are Language and Discourse (1971), Het denken van de grens. Vier opstellen over Derrida's grammatologie (1974), and Contextuality in Generative Linguistics (forthcoming). He is the editor of the volume History of Linguistic Thought and Contemporary Linguistics. His interest is focused on the debate between rationalism and empiricism in the history of linguistic theories, the epistemological presuppositions of contemporary linguistics, and the more general philosophical problem of the relationship between language and thought.

Sebastian Konstantinovic Saumjan was born in Georgia, U.S.S.R., in 1916. Since 1960 he has been the director of the Department of Structural Linguistics at the Institute of Russian Language of the Academy of Sciences in the U.S.S.R., where under his guidance new methods are applied for the simulation of natural languages. He is also a member of the Scientific Council of the Academy of Sciences of the U.S.S.R.; he teaches linguistics at the University of Moscow. His main fields of interest are phonology, the theory of generation of grammars, and semantics; his attention is also directed to problems of mathematical linguistics and cybernetics. His main publications are The History of the System of Differential Elements in Polish (1958), Problems of Theoretical Phonology (1962), A Model of Generative Application and Transformational Calculus in Russian (in cooperation with P. A. Soboleva, 1963), and Principles of Structural Linguistics (1966). This last work has been translated into English, Italian, and German.

BIBLIOGRAPHY

The bibliographic references provided in the footnotes of this b o o k were made partially by the editor and partially by the contributors concerned. This final bibliography contains all the works mentioned in the eleven dialogues or given in the footnotes. It is, of course, not exhaustive, not even for the works of the contributors to this volume (cf. also in this respect, the Bio-bibliographical Notes, pp. 405-411 of this book).

Aarsleff, H. 1970 "The History of Linguistics and Professor Chomsky", Language 46, 570-585. Aginsky, B. and E. 1948 "The Importance of Language Universals", Word A, 168-172. Alston, W. P. 1964 Philosophy of Language (Englewood Cliffs). Andrews, A. 1971 "Predicate Modifiers in Ancient Greek", Linguistic Inquiry 2, 127-151. Austin, J. L. 1962 How to do Things with Words? (Oxford). Bach, E. 1965 "On Some Recurrent Types of Transformations", in C. W. Kreidler (ed.), 16th Annual Round Table Meeting on Linguistics and Language Studies (= Georgetown U. Monograph Series on Languages and Linguistics 18). 1968 "Nouns and Noun Phrases", Universals in Linguistic Theory (London-New York), 91-124. 1971 "Questions", Linguistic Inquiry, 153-166. Bach, E. and R. T. Harms (eds.) 1968 Universals in Linguistic Theory (London-New York). Baker, C. and M. Brame 1972 "Global Rules: A Rejoinder", Language 48, 51-75. Benveniste, E. 1966 Problèmes de linguistique générale (Paris). Berlin, B. and P. Kay 1969 Basic Color Terms: their Universality and Evolution (Berkeley-Los Angeles). Bernstein, B. 1971 Class, Codes and Control I: Theoretical Studies Towards a Sociology of Language (London). Bever, T. G. 1970 "The Cognitive Basis for Linguistic Structures", in J. R. Hayes (ed.), Cognition and the Development of Language (New York-London), 279-352. Bierwisch, M. 1970 "Semantics", in J. Lyons (ed.), New Horizons in Linguistics (Harmondsworth).

414

BIBLIOGRAPHY

Black, M. 1959 "Linguistic Relativity: The Views of Benjamin Lee Whorf", The Philosophical Review 68, 228-238. 1969 "Some Trouble with Whorfianism", in S. Hook (ed.), Language and Philosophy (New York), 30-35. Bloomfield, L. 1933 Language (New York). Bolinger, D. 1960 "Linguistic Science and Linguistic Engineering", Word 16, 374-391. Borger, R. and F. Cioffi (eds.) 1970 Explanation in Behavioral Sciences (Cambridge). Borkin, A. 1971 "Polarity Items in Questions", CLS 7, 53-62. Bouveresse, J. 1971 La Parole malheureuse. De l'alchimie linguistique ä la grammaire philosophique (Paris). Campbell, R. and R. Wales 1970 "The Study of Language Acquisition", in J. Lyons (ed.), New Horizons in Linguistics (Harmondsworth), 142-160. Carnap, R. 1932 "The Elimination of Metaphysics Through Logical Analysis of Language", in A. J. Ayer (ed.), Logical Positivism (New York, 1959), pp. 68ff. 1956 "The Methodological Character of Theoretical Concepts", Minnesota Studies in the Philosophy of Science 1, 38-76. 1958 Meaning and Necessity: A Study in Semantics and Modal Logic (Chicago). 1959 The Logical Syntax of Language (Paterson, N. J.). 1961 Introduction to Semantics and Formalization of Logic (Cambridge, Mass.). 1966 Philosophical Foundations of Physics (New York-London). Cassirer, E. 1923 Philosophie der symbolischen Formen, I, Die Sprache (Berlin). Chafe, W. L. 1967 "Review of J. J. Katz, The Philosophy of Language", International Journal of American Linguistics 33, 248-254. 1967 "Language as Symbolization", Language 43, 57-91. 1970 Meaning and the Structure of Language (Chicago). 1971 "Directionality and Paraphrase", Language 47, 1-26. Chomsky, N. 1951 "Morphophonemics of Modern Hebrew" (Mimeographed, University of Pennsylvania, Philadelphia). 1955 "The Logical Structure of Linguistic Theory" (unpublished, microfilm, M.I.T., Cambridge, Mass.). 1957 Syntactic Structures (The Hague). 1959 "A Review of B. F. Skinner's Verbal Behavior", in J. A. Fodor and J. J. Katz (eds.), The Structure of Language (Englewood Cliffs), 547-578. 1964 Current Issues in Linguistic Theory (The Hague). 1965 Aspects of the Theory of Syntax (Cambridge, Mass.). 1966 Cartesian Linguistics (New York). 1968 Language and Mind (New York). 1969 "Some Empirical Assumptions in Modern Philosophy of Language", in S. Morgenbesser, P. Suppes and M. White (eds.), Philosophy, Science and Method (New York), 260-285. 1969 "Linguistics and Philosophy", in S. Hook (ed.), Language and Philosophy (New York), 51-94. 1971 Problems of Knowledge and Freedom (London). 1972 Language and Mind (enlarged version; New York). 1972 Studies on Semantics in Generative Grammar (The Hague). 1972 "Conditions on Transformations" (unpublished, M.I.T., Cambridge, Mass.).

BIBLIOGRAPHY

415

Cohen, L. J. 1962 The Diversity of Meaning (London). Cole, P. 1973 "Conversational Implicature and Syntactic Rules", in C. Bailey and R. Shuy (eds.), New Ways of Analyzing Variation in English (Georgetown). Conklin, H. 1962 "Lexicographical Treatment of Folk Taxonomies", International Journal of American Linguistics 28. Cooper, D. 1972 "Innateness : Old and New", The Philosophical Review 81, 465-483. Cordemoy, G. de 1666 Discours physique de la parole (2d ed., 1677; Engl, translation, 1668). Curry, H. and R. Feys 1958 Combinatory Logic (Amsterdam). 1963 Foundations of Mathematical Logic (New York). Dixon, R. 1972 The Dyirbal Language of Northern Queensland (Cambridge). Ducrot, O. 1972 Dire et ne pas dire. Principes de sémantique linguistique (Paris). Dumézil, G. 1949 L'Héritage indo-européen à Rome (Paris). 1953 La Saga de Hadingus: du mythe au roman (Paris). Edmonds, J. 1970 "Root and Structure Preserving Transformations" (unpublished, M.I.T., Cambridge, Mass.). Fann, K. T. 1970 Peirce's Theory of Abduction (The Hague). Ferguson C. 1971 "A Sample Strategy in Language Universals", Working Papers in Language Universels 6, 1-22 (Stanford). Fillmore, C. 1971 "Santa Cruz Lectures on Deixis" (unpublished). Firth, J. 1957 Papers in Linguistics 1934-1951 (London). Garner, R. 1971 "Presupposition in Philosophy and Linguistics", in Ch. Fillmore and D. T. Langendoen (eds.), Studies in Linguistic Semantics (New York), 23-44. Giglioli, P. P. (ed.) 1972 Language and Social Context (Harmondsworth). Gleason, H., Jr. 1968 "Contrastive Analysis in Discourse Structure", Georgetown University Monograph Series on Languages and Linguistics 21, 39-63. Godei, R. 1957 Les sources manuscrites du Cours de linguistique générale de F. de Saussure (Genève). Goodman, N. 1951 The Structure of Appearance (Cambridge, Mass.). 1965 Fact, Fiction and Forecast (Indianapolis). Gordon, D. and G. Lakoff 1971 "Conversational Postulates", CLS 7, 63-84. Green, G. 1972 "How to Get People to Do Things with Words", Georgetown Roundtable (Georgetown). 1972 Semantics and Syntactic Regularity (Cambridge). Greenberg, J. 1963 Essays in Linguistics (Chicago-London). Greimas, A. J. 1966 Sémantique structurale (Paris).

416

BIBLIOGRAPHY

1966 "Eléments pour une théorie de l'interprétation du récit mythique", Communications 8. 1970 Du Sens. Essais sémiotiques (Paris). 1970 "Conditions d'une sémiotique du monde naturel", in Du Sens. Essais sémiotiques, 49-92. Grice, H. P. 1968 "The Logic of Conversation" (unpublished, University of California, Berkeley). Habermas, J. 1971 "Vorbereitende Bemerkungen zu einer Theorie der kommunikativen Kompetenz", in J. Habermas and N. Luhmann (eds.), Theorie der Gesellschaft oder Sozialtechnologie (Frankfurt), 101-141. Halle, M. and N. Chomsky 1968 Sound Pattern of English (New York). Halliday, M. A. K. 1967 Intonation and Grammar in British English (The Hague). 1970 "Language Structure and Language Function", in J. Lyons (ed.), New Horizons in Linguistics (Harmondsworth). 1971 "Linguistic Function and Literary Style : An Inquiry into the Language of William Golding's The Inheritors", in S. Chatman (ed.), Literary Style: A Symposium (New York). Harris, Z. 1951 Methods in Structural Linguistics (Chicago). Hartmann, P. 1954 "Die Rolle der Sprache in Husserls Lehre von der Konstitution", Der Deutschunterricht 6, 29-55. 1958 Wesen und Wirkung der Sprache im Spiegel L. Weisgerbers (Heidelberg). 1961 Zur Theorie der Sprachwissenschaft (Assen). 1963 Theorie der Grammatik (The Hague). 1972 Zur Lage der Linguistik in der BRD (Frankfurt). Heidegger, M. 1959 Unterwegs zur Sprache (Pfullingen). Heidorn, G. 1972 "Natural Language Inputs to a Simulation Programming System" (Yale U. Diss.). Heringer, J. 1972 "Some Grammatical Correlates of Felicity Conditions and Presuppositions", Working Papers in Linguistics 11 (The Ohio State University Department of Linguistics). Hjelmslev, L. 1959 "Langue et Parole", Essais linguistiques, Travaux du Cercle linguistique de Copenhague, 12. 1963 Prolegomena to a Theory of Language (translated by J. Whitfield) (Madison). Hockett, C. 1947 "Problems of Morphemic Analysis", Language 23, 321-343. 1954 "Two Models of Grammatical Description", Word 10, 210-234. 1958 A Course in Modern Linguistics (New York). 1961 "Linguistic Elements and their Relations", Language 37, 29-53. 1968 The State of Art (The Hague). Hook, S. (ed.) 1969 Language and Philosophy. A Symposium (New York). Horn, L. 1972 "On the Semantic Properties of Logical Operators in English" (UCLA Diss.). Hudson, R. A. 1971 English Complex Sentences: An Introduction to Systemic Grammar (Amsterdam-London). Husserl, E. 1902 Logical Investigations (Logische Untersuchungen) (Halle, 1913). 1929 Formale und Transzendentale Logik. Versuch einer Kritik der logischen Vernunft (HalleSalle). Huttenlocher, J. 1968 "Constructing Spatial Images: a Strategy in Reasoning", Psychological Review 75, 550-560.

BIBLIOGRAPHY

417

Hymes, D. 1969 "Linguistic Theory and the Functions of Speech", International Days of Sociolinguistics (Rome). 1971 "Competence and Performance in Linguistic Theory", in R. Huxley and E. Ingram (eds.), Language Acquisition: Models and Methods (London-New York). 1972 "Review of J. Lyons, Noam Chomsky", Language 48, 416-427. Ivic, M. 1965 Trends in Linguistics (The Hague). Jackendoff, R. S. 1972 Semantic Interpretation in Generative Grammar (Cambridge, Mass.). Jakobson, R. 1963 "La notion de signification grammaticale selon Boas", Essais de linguistique générale (Paris), 197-208. 1970 "La linguistique", Tendances principales de la recherche dans les sciences sociales et humaines. Partie 1: Sciences sociales (The Hague). James, D. 1972 "Some Aspects of the Syntax and Semantics of Interjections", CLS 8, 162-172. Forthcoming "The Syntax and Semantics of Interjections" (U. of Michigan Diss.). James, W. 1890 The Principles of Psychology (New York). Joos, M. 1957 Readings in Linguistics (Chicago-London). Karttunen, L. 1971 "Some Observations on Factivity", Papers in Linguistics 4 : 1 . 1973 "Presuppositions of Complex Sentences", Linguistic Inquiry 4. Forthcoming "Remarks on Presuppositions", in A. Rogers et al. (eds.), Performatives, Presuppositions and Implicatures (Center for Applied Linguistics, Washington, D.C.). Katz, J. J. 1964 "Mentalism in Linguistics", Language 40, 124-137. 1966 The Philosophy of Language (New York). 1971 The Underlying Reality of Language and Its Philosophical Import (New York-Evanston). 1972 Semantic Theory (New York-Evanston). Katz, J. J. and J. A. Fodor 1963 "The Structure of a Semantic Theory", Language 39, 170-210. Katz, J. J. and P. M. Postal 1964 An Integrated Theory of Linguistic Description (Cambridge, Mass.). Keenan, E. 1971 "Two Kinds of Presupposition in Natural Language", in C. Fillmore and D. Langendoen (eds.), Studies in Linguistic Semantics (New York), 44-52. Kiparsky, P. 1968 "Linguistic Universals and Linguistic Change", in E. Bach and R. Harms (eds.), Universals in Linguistic Theory (London-New York), 171-204. Klima, E. 1964 "Negation in English", in J. A. Fodor and J. J. Katz (eds.), The Structure of Language (Englewood Cliffs), 246-323. Koutsoudas, A. 1972 "The Strict Order Fallacy", Language 48, 88-96. Kripke, S. 1963 "Semantic Considerations on Modal Logic", Acta Philosophica Fennica, 67-96. Kuno, S. 1972 "Evidence for Subject Raising in Japanese", Papers in Japanese Linguistics 1, 24-51. Labov, W. 1970 "The Study of Language in its Social Context", Studium Generale 23. Lacan, J. 1967 Ecrits (Paris).

418 Lakoff, 1963 1969 1970 1970 1971

BIBLIOGRAPHY

G. "Towards Generative Semantics" (M.I.T. Mechanical Translation Group, unpublished). "On Generative Semantics" (Indiana Linguistics Club, Bloomington). Syntactic Irregularity (New York). "Global Rules", Language 46, 627-639. "The Role of Deduction in Grammar", in C. Fillmore and T. Langendoen (eds.), Studies in Linguistic Semantics (New York), 63-72. 1971 "On Generative Semantics", in D. D. Steinberg and L. A. Jakobovits (eds.), Semantics: an Interdisciplinary Reader in Philosophy, Linguistics and Psychology (Cambridge), 121-296. 1971 "Presupposition and Relative Grammaticality", in D. D. Steinberg and L. A. Jakobovits (eds.), Semantics: An Interdisciplinary Reader in Philosophy, Linguistics and Psychology (Cambridge), 329-340. 1972 "The Arbitrary Nature of Transformational Grammar", Language 48, 76-87. 1972 "Hedges: A Study in Meaning Criteria and the Logic of Fuzzy Concepts", CLS 8, 183-228. 1972 "Linguistics and Natural Logic", in D. Davidson and G. Harman (eds.), Semantics of Natural Language (Dordrecht), 545-665. 1973 Letter to the New York Review of Books 19, February 8. 1973 "Fuzzy Grammar and the Competence/Performance Terminology Game", CLS 9, 271-290. 1973 "Some Thoughts on Transderivational Rules", in B. Kachru et al. (eds.), Papers in Linguistics in Honor of Henry and Renée Kahane (Urbana). Forthcoming "Pragmatics in Natural Logic", in A. Rogers et al. (eds.), Performatives, Presuppositions and Implicatures (Center for Applied Linguistics, Washington D. C.). Lakoff, R. 1969 "Review of Grammaire Générale et Raisonnée", Language 45, 343-364. 1969 "Some Reasons Where There Can't be Any Some-any Rule", Language 45, 608-615. 1970 "Tense and its Relation to Participants", Language 46, 838-844. 1971 "If's, And's, and But's about Conjunction", in C. Fillmore and D. T. Langendoen (eds.), Studies in Linguistic Semantics (New York), 115-150. 1971 "Passive Resistance", CLS 7, 149-162. 1972 "Language in Context", Language 48, 907-927. 1972 "The Pragmatics of Modality", CLS 8, 229-246. 1973 "Language and Woman's Place", Language and Society. 1973 "The Logic of Politeness: or, minding your P's and Q's", CLS 9, 292-305. 1973 "Questionable Answers and Answerable Questions" in B. Kachru et al. (eds.), Papers in Linguistics in Honor of Henry and Renée Kahane (Urbana). Lamb, S. 1965 "Kinship Terminology and Linguistic Structure", in E. Hammel (ed.), Formal Semantic Analysis ( = American Anthropologist 67), 37-64. 1966 Outline of Stratificational Grammar (revised version; Georgetown). 1966 "Epilegomena to a Theory of Language", Romance Philology 19. 1966 "Prolegomena to a Theory of Phonology", Language 42, 536-573. 1970 "Linguistic and Cognitive Networks", in P. Garvin (ed.), Cognition: A Multiple View (New York). 1972 "Some Types of Ordering", in V. Becker Makkai (ed.), Phonological Theory: Evolution and Current Practise (New York). Lawler, J. 1972 "Generic to a Fault", CLS 8, 247-258. Lévi-Strauss, C. 1958 "La structure des mythes", Anthropologie structurale (Paris), 227-256. 1960 "La structure et la forme, réflexions sur un ouvrage de Vlamidir Propp", Cahiers de VInstitut de Science économique appliquée 99, 3-36. 1962 La pensée sauvage (Paris). 1964-1972 Mythologiques (Paris). Lévi-Strauss, C. and R. Jakobson 1962 "Les Chats de Baudelaire", L'Homme 2, 5-21.

BIBLIOGRAPHY

419

Lewis, D. 1968 "Counterpart Theory and Quantified Modal Logic", Journal of Philosophy 65, 113-126. Locke, J. 1690 An Essay Concerning Human Understanding. Lockwood, D. 1972 Introduction to Stratificational Linguistics (New York). Malinowski, B. 1923 "The Problem of Meaning in Primitive Languages", Supplement I to C. K. Ogden and I. A. Richards, The Meaning of Meaning (London). Martinet, A. 1933 "Remarques sur le système phonologique du français", Bulletin de la Société de Linguistique 34, 191-202. 1937 Lagémination consonantique d'origine expressive dans les langues germaniques (Copenhague). 1939 "Role de la corrélation dans la phonologie diachronique", Travaux du Cercle de Linguistique de Prague 8, 273-288. 1949 "La double articulation du langage", Travaux de la Société de Linguistique de Copenhague 5, 30-37. 1953 "Structural Linguistics", Anthropology Today, 574-586. 1955 Economie des changements phonétiques. Traité de phonologie diachronique (Bern). 1956 La description phonologique, avec application au parler franco-provençal d'Hauteville ( Savoie) (Genève). 1960 Eléments de linguistique générale (Paris). 1960 "Elements of a Functional Syntax", Word 16, 10 ff. 1961 "Réflexions sur la phrase", Language and Society (Copenhagen), 113-118. 1961 A Functional View of Language (Oxford). 1966 "Structure and Language", Yale Studies 36-37, 10-18. McCawley, J. 1968 "Concerning the Base Component of a Transformational Grammar", Foundations of Language 4, 243-269. 1968 "The Role of Semantics in a Grammar", in E. Bach and R. T. Harms (eds), Universals in Linguistic Theory (London-New York), 124-169. 1968 "Lexical Insertion in a Transformational Grammar without Deep Structure", CLS 4, 71-80. 1970 "Semantic Representation", in P. Garvin (ed.), Cognition: A Multiple View (New York), 227-247. 1972 "A Program for Logic", in D. Davidson and G. Harman (eds.), Semantics of Natural Language (Dordrecht), 498-544. 1973 Selected Papers on Grammar and Meaning (Seminar Press). Miller, G. A. and N. Chomsky 1963 "Finitary Models of Language Users", in R. D. Luce, R. R. Bush and E. Galanter (eds.), Handbook of Mathematical Psychology (New York). Molino, J. 1971 "La connotation", La Linguistique 7, 5-30. Montague, R. 1971 "Pragmatics and Intensional Logic", in D. Davidson and G. Harman (eds.), Semantics of Natural Language (Dordrecht), 142-168. Morgan, J. 1969 "On the Treatment of Presuppositions in Transformational Grammar", CLS 5, 169-177. 1973 "Sentence Fragments", in B. Kachru et al. (eds.), Papers in Linguistics in Honor of Henry and Renée Kaharte (Urbana). Mounin, G. 1970 Introduction à la sémiologie (Paris). Nagel, T. 1969 "Linguistics and Epistemology", in S. Hook (ed.), Language and Philosophy (New York), 171-182. Noreen, A. 1903-1918 Vart Sprak (translated by H. W. Pollack) (Halle).

420

BIBLIOGRAPHY

Palmer, F. (ed.) 1968 Selected Papers of J. R. Firth 1952-59 (London). Parret, H. 1971 Language and Discourse (The Hague). Pàrtee, B. Hall 1971 "On the Requirement that Transformations Preserve Meaning", in C. Fillmore and D. T. Langendoen (eds.), Studies in Linguistic Semantics (New York), 1-22. Peirce, C. S. 1957 "The Logic of Abduction", in V. Thomas (ed.), Peirce's Essays in the Philosophy of Science (New York). Peters, P. S. (ed.) 1972 Goals of Linguistic Theory (Englewood Cliffs, N. J.). Pike, K. 1967 Language in Relation to a Unified Theory of the Structure of Human Behavior (revised ed.) (The Hague). Postal, P. 1966 "Review of A. Martinet, Eléments de linguistique", Foundations of Language 2, 151-186. 1970 "On Coreferential Complement Subject Deletion", Linguistic Inquiry 1, 439-500. 1971 Crossover Phenomena (New York). In Press On Raising (Cambridge, Mass.). Prieto, L. 1966 Messages et signaux (Paris). Propp, V. 1958 "Morphology of the Folktale, Part III", International Journal of American Linguistics 24: 4 (Publication Ten of the Indiana University Research Center in Anthropology, Folklore and Linguistics). Putman, H. 1967 "The Innateness Hypothesis and Explanatory Models in Linguistics", Synthese 17, 12-22. Quine, W. V. O. 1960 Word and Object (Cambridge, Mass.). 1969 "Ontological Relativity", Ontological Relativity and Other Essays (New York). Rastier, F. 1972 "Systématique des isotopies", in A. J. Greimas (ed.), Essais de sémiotique poétique (Paris), 80-106.

Reich, P. 1969 "The Finiteness of Natural Language", Language 45, 831-843. 1970 "Relational Networks", Canadian Journal of Linguistics 15. Reichenbach, H. 1947 Elements of Symbolic Logic (London). Rogers, A. 1971 "Three Kinds of Physical Perception Verbs", CLS 7, 206-222. Ross, J. 1967 "Constraints on Variables in Syntax" (unpublished, M.I.T., Cambridge, Mass.). 1970 "On Declarative Sentences", in R. A. Jacobs and P. S. Rosenbaum (eds.), Readings in English Transformational Grammar (Waltham, Mass.), 222-272. 1972 "The Category Squish: Endstation Hauptwort", CLS 8, 316-328. 1972 "Primacy and the Order of Constituents" (Paper read at the annual Meeting of the New England Society, unpublished). In press "A Fake NP Squish", in C. J. N. Bailey and R. Shuy (eds.), New Ways of Analyzing Variation in English (Georgetown). In preparation "Nouniness". Ryle, G. 1961 "Use, Usage and Meaning", reprinted in C. Lyas (ed.), Philosophy and Linguistics (LondonBasingstoke), 54-60.

BIBLIOGRAPHY

421

Sadock, J. 1970 "Whimperatives", in S. Sadock and A. Vaneck (eds.), Studies presented to Robert B. Lees by his students (Edmonton), 223-238. 1971 "Queclaratives", CLS 7, 223-231. 1972 "Speech Act Idioms", CLS 8, 329-339. Saumjan, S. 1971 Principles of Structural Linguistics (translated into English by J. Miller) (The Hague). Saussure, F. de 1879 Mémoire sur le système primitif des voyelles dans les langues indo-européennes (Leipzig). 1915 Cours de linguistique générale (published by C. Bally and A. Sechehaye) (Paris, 1968). Schegloff, E. 1971 "Notes on a Conversational Practice: Formulation Place", in D. Sudnow (ed.), Studies in Social Interaction (Glencoe, 111.). Schmidt, S. J. 1970 "Linguistik und Literaturwissenschaft", Linguistik und Didaktik 1. 1972 Zur Grundlegung der Literaturwissenschaft (München). Searle, J. 1969 Speech Acts. An Essay in the Philosophy of Language (Cambridge). 1972 "Chomsky's Revolution in Linguistics", The New York Review of Books 18: 12, 16-24. Simon, J. 1971 Philosophie und linguistische Theorie (Berlin-New York). Sinclair, J., I. J. Forsyth, R. M. Coultherd and M. Ashby 1972 The English Used by Teachers and Pupils (University of Birmingham). Stalnaker, R. 1971 "Pragmatics", in D. Davidson and G. Harman (eds.), Semantics of Natural Language (Dordrecht), 380-397. Stampe, D. 1972 "On the Natural History of Diphthongs", CLS 8, 578-590. Steinthal, H. 1855 Grammatik, Logik und Psychologie, ihre Prinzipien und ihre Verhältnisse zu einander (Berlin). 1881 Einleitung in die Psychologie und Sprachwissenschaft (Berlin). Strawson, P. 1951 Introduction to Logical Theory (London). 1964 "Intention and Convention in Speech Acts", Logico-Linguistic Papers, 149-169. 1969 "Meaning and Truth", Logico-Linguistic Papers, 170-189. 1969 "Grammar and Philosophy", Logico-Linguistic Papers, 130-148. 1971 Logico-Linguistic Papers (London). Swadesh, M. 1946 "Chitimacha", in H. Hoijer et al. (eds.), Linguistic Structures of Native America (New York). Togeby, K. 1951 La structure immanente de la langue française (Copenhague). Trubetzkoy, N. 1933 "La phonologie actuelle", Journal de psychologie normale et pathologique 30. 1936 "Essai d'une théorie des oppositions phonologiques", Journal de psychologie normale et pathologique 33. Turner, G. 1973 "Social Class and Children's Language of Control at Age Five and Age Seven", in B. Bernstein (ed.), Class, Codes and Control II: Applied Studies Towards a Sociology of Language (London). Vachek, J. 1966 The Linguistic School of Prague: An Introduction to its Theory and Practice (Bloomington). Van Dijk, T. A. 1972 Some Aspects of Text Grammars (The Hague). Vendler, Z. 1967 Linguistics and Philosophy (Ithaca).

422

BIBLIOGRAPHY

1971 "Linguistics and the A Priori", in C. Lyas (ed.), Philosophy and Linguistics (LondonBasingstoke), 245-268. Verbürg, P. A. 1964 "Vicissitudes of Paradigms" (Paper 25, Wenner-Gren Foundation for Anthropological Research). Weinreich, U. 1963 "On the Semantic Structure of Language", in J. Greenberg (ed.), Universals of Language (Cambridge, Mass.), 142-216. Wells, R. S. 1947 "De Saussure's System of Linguistics", Word 3, 1-31. Whorf, B. 1956 "Language: Plan and Conception of Arrangement", in J. B. Carroll (ed.), Language, Thought, and Reality: Selected Papers of Benjamin L. Whorf (New York). Wilson, N. L. 1967 "Linguistical Butter and Philosophical Parsnips", Journal of Philosophy 64, 55-67.f Wittgenstein, L. 1921 Tractatus Logico-Philosophicus (London, 1922). 1953 Philosophical Investigations (Oxford). 1958 The Blue and Brown Books (Oxford). 1964 Bemerkungen über die Grundlagen der Mathematik (Oxford). 1964 Philosophische Bemerkungen (Oxford). Zipf, G. 1949 Human Behavior and the Principle of Least Effort (Cambridge, Mass.). Zumthor, P. 1972 Poétique Médiévale (Paris).

INDEX OF NAMES

Adjukiewicz, D., 295 Aginsky, B.W. and E.G., 227 Alston, W.P., 21 Andrews, A., 266 Arnauld, A., 173 Austin, J. L., 162, 326, 387, 390, 401 Bach, E„ 42, 258, 269, 270, 342 Bach, E. and R. Harms, 342 Bailey, Ch. J., 264 Baker, C. and M. Brame, 177, 267 Bar-Hillel, Y„ 295, 369 Barthes, R., 231 Baudelaire, Ch., 62 Benveniste, E., 70, 232, 399, 400 Berman, A., 175 Bernstein, B., 83, 84, 89, 93, 110, 117, 118, 123 Bever, T. G., 46, 52 Bierwisch, M., 393 Black, M„ 313, 336, 344 Bloch, B„ 25, 193 Bloomfield, L., 25, 37, 55, 114, 116, 171, 172, 179, 181, 182, 183, 198, 215, 219, 222, 231, 233, 297, 316, 362, 396 Boas, F., 171 Bolinger, D„ 157, 158, 159, 172, 177 Bolzano, B., 376 Bouveresse, J., 297, 298, 299, 301-403, 407 Brewer, B., 250 Buhler, K., 5, 95, 97, 239, 354 Campbell, R. and R. Wales, 335 Carnap, R„ 163, 280, 313, 314, 316, 317, 330, 362, 363, 364, 368, 369, 370, 371, 377, 379, 380, 393, 400, 401 Cartesians, 28, 32, 173, 246, 327 Cassirer, E., 140, 141, 146, 344 Chafe, W., 1-25, 202-203, 217, 348, 407 Chomsky, N„ 1-4, 22, 24, 27-54, 57, 66, 70, 81, 84, 85, 97, 112, 114-116, 118, 127, 142, 151, 152, 153, 154, 155, 156, 163, 166, 167, 169, 170, 171, 172, 174, 177, 181, 182, 185, 193, 201, 205, 209, 210, 212, 214, 216, 217, 219,

236, 240, 241, 243, 246, 253, 254, 255, 256, 257, 269, 270, 276, 279, 288, 302, 303, 304, 305, 306, 312, 313, 316, 317, 318, 324, 326, 327, 328, 329, 334, 335, 336, 337, 338, 345, 346, 347, 348, 349, 354, 355, 356, 357, 359, 365, 369, 371, 375, 376, 387, 388, 389, 392, 393, 402,407 Church, T., 376 Cohen, L. J., 368 Cole, P., 154 Conklin, H„ 198 Cordemoy, G. de, 28, 32 Courtenay, B. de, 223 Cudworth, R., 34 Curry, H. B„ 281, 289, 295

249, 263, 290, 307, 319, 330, 339, 350, 360, 378, 396,

250, 266, 291, 309, 320, 331, 340, 351, 361, 379, 397,

251, 267, 295, 310, 322, 332, 341, 352, 362, 382, 398,

252, 268, 301, 311, 323, 333, 343, 353, 363, 386, 400,

Dahl, O., 155, 273 Descartes, R., 32, 34, 66, 78, 318, 319, 324, 325, 333, 334 Dixon, R., 259 Ducrot, O., 381-385, 390, 391 Dumézil, G., 64, 77 Dürkheim, E., 228 Dijk, T. van, 125, 126 Emenau, M., 181 Engler, R., 225 Ferguson, C., 271 Feys, R., 289, 295 Fillmore, Ch., 42, 154, 167, 177, 240 Firth, J. R., 81, 86, 87, 98, 101, 109, 114, 180, 181,221, 351 Fodor, J. A., 52, 298, 346, 362, 363, 368, 369, 370, 374, 375, 377, 382, 388, 393 Foucault, M., 399 Frege, G., 299, 307, 376, 385

424

INDEX OF NAMES

Gardner, R., 383 Gleason, H. A., 218 Glinz, H., 140 Godei, R„ 224 Godei, K„ 315 Golding, W., 113 Goodman, N„ 54, 173, 251, 310, 339 Gordon, D., 154, 275 Green, G., 154 Greenberg, J., 171, 172, 227, 271 Greimas, A. J., 55-79, 115, 124, 126, 312, 391, 408 Grice, H. P., 45, 154, 162, 275, 392

Haas, M., 181, 183 Habermas, J., 53, 347 Hall, R„ 264 Halle, M„ 54, 265 Halliday, M . A . K . , 25, 81-120, 180, 185, 212-213, 215,219, 306, 343, 349, 351, 353, 354, 355, 360, 397, 398, 408 Harris, Z„ 3, 36-37, 54, 116, 166, 171, 172, 173, 183, 219, 222, 241, 250, 254, 255 Hartmann, P., 121-150, 408 Heidegger, M., 149, 150 Herder, J. G„ 147 Heringer, J., 154 Hintikka, K. J. J., 395 Hjelmslev, L„ 20, 57-59, 64, 66, 86, 87, 89,91, 94, 101, 112, 114-115, 119, 140, 141, 150, 179, 180, 181, 182, 184, 185, 188, 189, 198, 201, 208, 213, 214, 215, 219, 224, 225, 226, 231, 236, 238, 241, 242, 247, 287, 294, 312 Hockett, Ch., 20, 149, 171, 180, 181, 184, 193, 219, 222, 227, 255 Hook, S„ 324, 332 Horn, L„ 155 Hudson, R. A., 88 Humboldt, W. von, 32, 140, 141, 146, 217, 304 344 Hume, D., 34, 333 Husserl, E„ 68-69, 129, 139, 140, 143, 147, 295, 307, 330, 356-357, 365, 366, 367, 399 Hymes, D., 53, 81, 84-85, 96, 97, 347

Kajita, M„ 258 Kant, E., 32, 150, 319, 370 Karttunen, L., 155 Katz, J. J„ 1, 31, 42, 45, 46, 163, 164, 298, 306, 307, 313, 321, 346-347, 360, 362, 363, 368, 369, 370, 374, 375, 377, 382, 388, 393, 401 Keenan, E„ 155, 271, 382 Kiparsky, P., 264 Koutsoudas, A., 268 Klima, E„ 157, 158 Kripke, S„ 163, 260, 395 Labov, W., 81, 84-85, 110, 154 Lacan, J., 66, 227 Lakoff, G., 14, 24, 38, 47, 108, 114, 151-178, 250, 251, 254, 258, 260, 262, 266, 267, 268, 275, 394, 395, 398, 409 Lakoff, R„ 154, 157, 158, 159, 167, 168, 173, 177 Lamb, S„ 86, 87, 88, 91, 98, 114, 115, 179-219, 350, 351, 258, 409 Lambeck, D., 295 La Mettrie, 318 Lancelot, Cl., 173 Leibniz, G„ 32, 318, 334, 393 Lesniewsky, A., 295 Lévi-Strauss, CI., 56, 61, 64, 76, 115, 119, 226, 227, 308 Lewis, D., 260 Lichnerowicz, A., 243 Locke, J., 22, 306, 317-318, 333 Lohmann, J., 150 Lorenz, K., 333

Ivic, M„ 142

Malinowski, B., 51, 92, 95, 181, 389 Mallarmé, S., 61 Martinet, A., 20, 115, 203, 221-247, 302, 303, 306, 311, 312, 314, 322, 323, 338, 343, 344, 351, 352, 353, 357, 381, 409 McCawley, J., 114, 155, 174, 249-277, 329, 337, 349, 363, 364, 398, 410 Meillet, A., 222, 242 Merleau-Ponty, M., 68-69 Miller, G. A., 52 Monod, J., 333, 334 Montague, R., 162, 163, 171, 380 Morgan, J., 154, 252, 273 Morris, Ch., 25, 53, 130 Mounin, G., 241, 244

Jackendoff, R., 42 Jakobson, R„ 62, 64, 171, 172, 221, 227, 240, 243, 250, 273, 294, 295, 308, 354, 362, 399 James, D., 159-160 James, W., 16, 330 Jones, D., 226 Joos, M., 268

Partee, B. Hall, 363 Pascal, B., 346 Peirce, Ch., 29, 309, 358 Perlmutter, D., 169 Peters, S., 47 Peters, S. and Ritchie, 177 Pike, K., 86, 114, 172, 188, 208, 219, 222

INDEX OF NAMES

Popper, K., 309, 316 Postal, P., 47, 48-49, 233-234, 254, 266, 268, 272 Prague School, 86, 95, 115-116, 221, 226, 227, 243, 255 Prieto, L. J., 244 Prior, A. N„ 395 Propp, V., 56, 64 Putnam, H., 50, 336 Quine, W. V. O., 33, 259, 316, 317, 318, 319, 320, 321, 331, 332, 333, 337, 338, 340, 344, 345, 362, 370, 371, 376, 377 Rastier, F., 61 Reichenbach, H., 77, 395 Ross, J. R„ 10, 42, 154, 159, 160, 167, 254, 269, 270, 276 Russell, B., 77, 401 Ryle, G., 303, 327, 352, 358, 401 Sadock, J., 154, 275 Sanctius, 32, 173 Sapir, E., 35, 172,181, 183, 215, 222, 344, 345 Saumjan, S„ 279-295, 302, 304, 322, 371, 411 Saussure, F. de, 20, 24, 35, 57-59, 63, 70, 74, 75, 76, 77, 78, 85, 91, 94, 101, 140, 141, 150, 180, 187, 211, 219, 222, 223, 224, 225, 228, 229, 230, 236, 237, 244, 294, 297, 302, 303, 305, 306, 307, 312, 351, 386 Schmidt, S. J., 124 Schuchardt, H., 222 Scott, D., 261, 395 Searle, J., 30, 44, 45, 52-53, 154, 162, 302, 313, 332, 334, 335, 338, 340, 345, 354, 355, 371, 381

425

Simon, J., 150 Sinclair, J., 102 Skinner, B. F., 320, 396 Stalnaker, R., 381 Stampe, D., 264-265 Steinthal, H., 141 Strawson, P., 162, 331, 361, 364, 365, 366, 385, 390 Swadesh, M., 198 Tarski, A., 163, 362, 376 Togeby, K„ 225, 236 Trager, G., 25, 222 Troubetzkoy, N„ 114, 171, 221, 224, 239, 294 Turner, G., 88, 90 Vendler, Z., 297, 402 Verburg, P., 129 Weinreich, U„ 172, 395 Weisgerber, L„ 140, 141, 146, 147 Whitfield, J., 179, 215 Whorf, B., 117-118, 198, 344, 345 Wittgenstein, L„ 27, 77, 298, 299, 308, 316, 326, 327, 328, 344, 347, 348, 351, 355, 356, 358, 359, 360, 361, 372, 373, 374, 375, 376, 377, 378, 386, 388, 394, 396, 398, 399, 401, 402 Wright, von, 395 Wundt, W., 5, 222 Zadeh, L., 155 Zipf, G„ 238, 341 Zumthor, P., 113

INDEX OF SUBJECTS

Abduction, 29, 309-310 acceptability, 41, 155-156, 329 acquisition of language, 100-101, 108-112, 310, 334-335 ambiguity, 165, 205-208 American-Indian languages, 2 American linguistics, 1-3, 5, 33, 35-38, 55, 179184, 188-189, 222-223, 233-234, 254-255, 294-295, 312 analyticity, 258, 339-340, 369-371 applicative grammar, 281-282, 284-285 appropriatness, 41, 159 arbitrariness 91 articulation, 236; double - , 73. 237 automated data analysis, 133-134, 192, 324 autonomy of linguistics, 81-83,212-213, 307-308, 351 behavioral potential, 85-86 behaviorism, 17, 34, 109, 170-172, 316, 327-328 biology, 243 change, linguistic, 264 codification, 23, 212 combination, 89 communication, 17, 51, 71, 97, 125, 177, 227, 235, 236, 244, 356-357, 358, 389; animal - , 18,72 competence, 3-4, 35,43, 57, 69, 98, 148, 153-155, 209, 212, 252, 263-264, 265-266, 304, 314-315, 331, 339, 345-351, 353, 382; semantic - , 4, 346-347; communicative 53, 85, 347-348, 349; hearer-speaker - , 265-266, 348 computational linguistics, 192, 218 concept, 20, 200-201; generic - , 8; subconceptualization, 9-10, 20 connotation, 45, 228-229 contextuality, 15-16, 39, 53, 74, 89, 104, 155, 156-160, 167, 177, 206, 229, 232, 274, 353-354 convention, 388-391 conversational implicatures, 162; conversational postulates, 45, 275, 392, 395-396 coreference, 165, 272-273

corpus, 188, 233-234, 310 creativity, 28-29, 117, 174, 217, 229-230, 253, 318, 322-323, 326-327, 331-332, 352-353; rulegoverned - , 29, 305 cybernetics, 293 deep structure, 5, 19, 38, 49, 66, 194-195, 204, 254, 256, 257, 320-321, 378 derivation, 4, 174, 250-251, 286 derivational constraints, 28, 49, 251 description, linguistic, 138, 153, 208, 209 dialect, 41, 264 directionality, 23-24, 202-204 discourse, 1, 14, 35, 39, 54, 57, 74-75, 100-101, 156, 163, 217-218, 232, 272, 274-275, 351-352, 386-387, 395 dispositional approach to language, 317 distinctive features, 287 domain of linguistics, 302-304 duality, 20 economy, linguistic, 237, 239-240 educational linguistics, 113 elementary structure of meaning, 67 empirical evidence, 167-168, 185, 281, 322 empiricism, 17, 33, 34, 170-172, 182-183, 188189, 226,234, 251, 255, 281, 294, 309, 316-318, 319 environment, 99, 109, 337-338 experience and language, 103, 117, 146-147, 200, 228, 244-246, 396-397 expression, 52, 71, 96-97, 144, 236, 356 faculty of language, 31, 58, 216, 319, 335; psychological - , 1 7 focus, 107-108 form; - and content, 58, 62, 75, 82; - and structure, 141; - and genesis, 141; - and function, 177, 276-277, 355-356, 378-379; - and substance, 225 formal languages, 138, 371-372 formalization, 84, 214, 241-242, 262-263

INDEX OF SUBJECTS functionalist!), 92-98, 105, 246-247 function, linguistic, 17, 51, 71, 92, 94, 144, 177, 235-237, 276-277, 293, 343, 354-357, 359; hierarchy of - , 96, 356-357 fuzziness, 167, 169, 175, 211 generative grammar, 36, 173-174, 195, 211, 246-247, 279-280, 284, 302-303 generative semantics, 47-49, 166-167, 174-177, 256, 266-268, 253-354 generativity, 27-28, 56, 60, 76, 216, 229, 250, 302-303, 304 genotype language, 279-280, 288, 291, 371 gesture, 17 global rules, 50, 175, 177, 251, 266-267 glossematics, 34,180, 234 grammar, essential, 361-362, 364-366; logical - , 330-331, 365-367 grammaticality, 40-41, 142, 155-156, 169, 210, 241, 249-251, 252, 263, 290, 329; - and grammaticity, 142 heuristics, 128, 135-136, 137-138 ideal language, 280 idealization, 40, 82, 349 ideational theory, 21 ideology, 170, 304-305 illocutionary acts, 381 immanence, linguistic, 74-75, 123, 181, 244-245 indexicals, 162 induction, 234, 311-312 inference, 188,189 information, 205, 235-236, 259 information theory, 293 information system, human, 186-187, 213 innateness, 30, 171, 200-201, 252-253, 332-338, 340 intention, 45, 52, 253, 262, 276, 357, 390; - and convention, 388-389 interdisciplinarity, 130, 132-133, 243-244 instrumentality, 83, 235, 359 intonation, 106-108 introspection, 16, 40, 210-211, 240-241, 264, 338-339 intuition, grammatical, 16-17, 40, 210-211, 240241, 338-339 isotopy, 59, 60-61 knowledge; experiential - , 6 ; conceptual, 6-8; - of language, 241, 314-315, 325, 327 language, private, 358 langue - parole, 24-25, 35,41, 57,98-99,211-212, 222, 224, 228-229, 237, 303, 345, 351-352, 361, 374

427

lexical insertion, 13, 38, 49, 257-258 lexical decomposition, 13, 168, 258 lexical units, 198, 231 lexicalist hypothesis, 49, 175 linearity, 201-204, 214, 238, 287 linguisticity, 132-133 linguistic theory, 208-209, 228-229, 342; goal of - , 39-40, 150-151, 294, 302-304; social function of - , 121-122 logic, concrete, 76, 77; mathematical - , 291-292; modal - , 260 manifestation, 22, 64, 66, 215 marxism, 71 mathematics, 290-291 meaning, 37, 140, 156, 172, 177, 229, 231, 293-294, 361-379, 372-374, 376-379; - potential, 85-87; contextual - , 165; conveyed - , 164; - relationships, 210; - universals, 166; - postulates, 164, 260 mentalism, 183, 200, 252, 307, 316, 321, 324, 360 mental image, 12 message, 19; - type, 105 metafunction, linguistic, 97 metaphor, 19 methodology, linguistic, 301-315 modality, 70, 186 model construction, 137 model theory, 163,167 motivation, 63 narrative structures, 64-65 natural logic, 50, 76,162,175, 260-262, 394 naturalness, 4 network, 87-88, 189-192, 196, 215, 284 nominality, 105 number agreement, 168-169 ordering of rules, 268 ordinary language philosophy, 162-163, 374-375, 401-402 participant role, 104 perception, 21-22, 68-69 performance, 3, 43, 153-154, 209, 263-264, 312, 345-346, 374-375; - theory, 348-349 performatives, 10-11, 163 phenomenology of language, 124, 129-130, 139-140 phenotype languages, 279, 288, 291 philosophy and linguistics, 53-54, 396-403 phonology, 221-222, 230-231, 244, 264-265, 287, 288 poetics, 61-63, 218, 243-244 politics and linguistics, 170, 232

428

INDEX OF SUBJECTS

pragmatics, 11, 124, 130, 159,162, 175, 273-275, 353, 375; theoretical-, 380-381 predicate, 259, 269, 282 presupposition, 11, 46, 71, 164-165, 271-274, 382-386; types o f - , 384-385 primitives, semantic, 392 process, 215 proposition, 39, 319, 352 psychology; linguistics, subpart of - , 5, 31-32, 69, 85, 185, 254, 306, 350 psychological reality, 209-210 psychosemantics, 5-6, 25 rationalism, 33, 170-171, 188, 306, 309, 335, 340 realization, 86-87, 89, 183-184, 204 reductionism, 78 reference, 46, 74, 75, 77, 96, 165, 261-262, 272-273, 293-294, 376, 385-387, 398-399 relevance, 227, 239, 241 rhetorical component of grammar, 381 selection restrictions, 363-364 semanticism, 3, 330 semantic units, 12 semantics, 37-38, 68, 90, 103, 127-130, 182, 353, 368-369, 393; logical 362-363, 368; - and pragmatics, 377-378, 379-380, 382, 386, 391 semiotics (semiology), 68, 72-74, 79, 124, 187; - and linguistics, 244 sentence as a linguistic unit, 39, 59, 101-102, 125-126, 163, 174, 217, 303, 353-354 sociogrammar, 81, 112, 397-398 sociology and linguistics, 69, 83-85, 122-123, 161-162, 170, 185, 213, 254, 306 sociosemantics, 81, 87, 102, 110 speech acts theory, 44-45, 70, 71, 162, 275-276, 387-388, 390 stratification of language, 180, 183, 189-192 stratificational grammar, 184,193,194,207,211, 216, 219 structuralism, 34, 35, 55-57, 150, 170-171, 180-182, 221-223, 246-247, 294-295, 302-303, 316, 399

structure, 55, 88, 94, 141, 179, 196, 201-212, 223-224, 284, 302-303; logical - , 152, 164, 196; - and usage, 356, 378-379 style, 163 subject, 399-400 symbolization, 13, 19-20 synchrony and diachrony, 100, 286 syntactization, 13-14 syntax, 90, 190, 192-194, 205, 287, 330, 352; autonomous - , 167; syntactical categories, 258-259; - and semantics, 30, 91-92, 128, 155, 167, 231-232, 255, 256, 287, 330 system and process, 100-101 taxonomy, 34, 171, 210, 233 text, 86, 353-354 text linguistics, 123-127, 139 theme and rheme, 93, 105 theory construction, 133-136, 152-153, 195-196, 225, 234-235, 280, 281, 292-293, 308, 320, 342 thought, 17-18, 172, 277, 357-359; - and language, 144-146 transderivational rules, 165-166, 175-176, 267 transformation, 4, 10, 20, 36-37, 62, 192-193, 216, 250-251, 280, 287, 288-290 transitivity, 93, 102-105 translatability, 125, 344-345 tree structure, 215 typology, 78, 127 uniqueness of man, 54 units, linguistic, 101, 204 universal, 22, 33-34, 42, 68, 106, 117-119, 130, 142-143, 166, 171, 209-210, 212, 227, 240, 268-271, 282, 287, 329, 340-345, 393 use of language, 29, 93, 99, 161, 167, 263, 276, 325, 326, 329-330, 349, 372, 374, 377 utterance, 39 variety, linguistic, 130-131 vocality, 235, 237-238 writing and language, 238