Foundations of Axiomatic Linguistics 9783110872019, 9783110112344

Table of contents :
Acknowledgements
Chapter I: Introduction
1. Foreword to the introduction
2. Descriptive linguistics – objects and method. The functional principle
3. Scientific linguistics. Epistemological considerations
4. The hypothetico-deductive method
Chapter II: Theory and strategy – C’est le point de vue qui crée l’objet (Saussure)
1. Preparing the raw material
2. The relation of theory to description, and of both to the speech-phenomena
3. The axiomatic approach
4. The three legs of the theory
Chapter III: Function, system, and systemology
1. The first axiom
2. The notion ‘self-containedness’
3. Earlier stages and development of the theory
4. Further on the theory of semiotic systems, and the sub-systems of language
Chapter IV: Semiotic foundations of axiomatic functionalism – Ontological considerations and the ‘linguistic sign’
1. The theory of indices
2. The ontology or ‘signum-theory’
Chapter V: Linguistic description I: General principles, and phonology
1. Introduction
2. Practical hints for successful descriptivism
3. Methodology and hypothetico-deductive logic
4. ‘Pseudo-composites and pseudo-words’ revisited
5. Productivity and pseudo-productivity. Analogy as motivation for the formation of new words
6. Productivity and non-productivity in so-called ‘derivations’
7. Further notes on ‘productivity’ and potential pitfalls
8. Some further remarks
9. The notion ‘simultaneity’
10. The phoneme as a simultaneous bundle of distinctive features
11. Phoneme tables as practical descriptive devices
12. Consonants, vowels, semi-vowels, and semi-consonants
13. The notion ‘position’, ‘nuclearity’, and ‘peripheralness’ in phonology
14. The notions ‘neutralization’ and ‘archiphoneme’
15. The description of vowel systems
16. Some further practical hints for establishing vowel-systems
17. The English vowel system
18. Some further examples of descriptive problems in phonology
19. Effective phonological description
20. Tone-languages of East and Southeast Asia
Chapter VI: Linguistic description II: Syntax and the sentential level
1. Introduction
2. Direct and indirect tactic and non-tactic relations
3. Functional dependency versus occurrence dependency
4. Correlations between functional dependency and occurrence dependency
5. Parallel determination versus diverse determination
6. The notion ‘semi-cluster’
7. The notion ‘plereme’
8. Syntactic description
9. Concord and congruity
10. Syntactic patterns in English
11. Some remarks on ‘sentence’ and ‘sentential entities’
Chapter VII: Further examples of syntactic analysis and additional remarks
Chapter VIII: Postulates for axiomatic linguistics. Revised version
Bibliography
Index to the definitions
General Index

Citation preview

Foundations of Axiomatic Linguistics

Trends in Linguistics Studies and Monographs 40

Editor

Werner Winter

Mouton de Gruyter Berlin · New York

Foundations of Axiomatic Linguistics

by

Jan W. F. Mulder

Mouton de Gruyter Berlin · New York

1989

Mouton de Gruyter (formerly Mouton, The Hague) is a Division of Walter de Gruyter & Co., Berlin.

Library of Congress in Publication

Data

Mulder, Jan W. F. Foundations of axiomatic linguistics. (Trends in linguistics. Studies and monographs; 40) Bibliography: p. Includes index. I. Linguistics. I. Title. II. Series. P1.21.M925 1988 410 88-13726 ISBN 0-89925-323-7

Deutsche Bibliothek Cataloging in Publication Data

Mulder, Jan W. F.: Foundations of axiomatic linguistics / by Jan W. F. Mulder. — Berlin ; New York : Mouton de Gruyter, 1989 (Trends in linguistics : Studies and monographs ; 40) ISBN 3-11-011234-5 NE: Trends in linguistics / Studies and monographs

© Printed on acid free paper © Copyright 1989 by Walter de Gruyter & Co., Berlin. All rights reserved, including those of translation into foreign languages. No part of this book may be reproduced in any form — by photoprint, microfilm or any means — nor transmitted nor translated into a machine language without written permission from the publishers. Typesetting: Asian Research Service, Hong Kong. — Printing: Gerike, G m b H , Berlin — Binding: Lüderitz und Bauer, Berlin: — Printed in Germany.

Contents Acknowledgements

ix

Chapter I:

Introduction 1 1. Foreword to the introduction 1 2. Descriptive linguistics — objects and method. The functional principle 2 3. Scientific linguistics. Epistemological considerations 7 4. The hypothetico-deductive method 36

Chapter II:

Theory and strategy — C'est le point de vue qui cr6e l'objet (Saussure) 1. Preparing the raw material 2. The relation of theory to description, and of both to the speech-phenomena 3. The axiomatic approach 4. The three legs of the theory

Chapter III:

Chapter IV:

Function, system, and systemology 1. The first axiom 2. The notion 'self-containedness' 3. Earlier stages and development of the theory 4. Further on the theory of semiotic systems, and the sub-systems of language Semiotic foundations of axiomatic functionalism — Ontological considerations and the 'linguistic sign' 1. The theory of indices 2. The ontology or 'signum-theory'

45 45 51 65 70 83 83 84 87 96

123 123 151

VI

Chapter V:

Linguistic description I: General principles, and phonology 169 1. Introduction 169 2. Practical hints for successful descriptivism 171 3. Methodology and hypothetico-deductive logic 179 4. 'Pseudo-composites and pseudo-words' revisited 183 5. Productivity and pseudo-productivity. Analogy as motivation for the formation of new words 193 6. Productivity and non-productivity in so-called 'derivations' 198 7. Further notes on 'productivity' and potential pitfalls 200 8. Some further remarks 207 9. The notion 'simultaneity' 209 10. The phoneme as a simultaneous bundle of distinctive features 214 11. Phoneme tables as practical descriptive devices 220 12. Consonants, vowels, semi-vowels, and semi-consonants 226 13. The notion 'position', 'nuclearity', and 'peripheralness' in phonology 228 14. The notions 'neutralization' and 'archiphoneme' 235 15. The description of vowel systems 240 16. Some further practical hints for establishing vowel-systems 246 17. The English vowel system 250 18. Some further examples of descriptive problems in phonology 252 19. Effective phonological description 256 20. Tone-languages of East and Southeast Asia 271

Vll

Chapter VI:

Chapter VII:

Linguistic description II: Syntax and the sentential level 1. Introduction 2. Direct and indirect tactic and non-tactic relations 3. Functional dependency versus occurrence dependency 4. Correlations between functional dependency and occurrence dependency 5. Parallel determination versus diverse determination 6. The notion 'semi-cluster' 7. The notion 'plereme' 8. Syntactic description 9. Concord and congruity 10. Syntactic patterns in English 11. Some remarks on 'sentence' and 'sentential entities' Further examples of syntactic analysis and additional remarks

275 275 276 288 291 293 299 301 305 306 309 365

385

Chapter VIII: Postulates for axiomatic linguistics. Revised version

435

Bibliography

459

Index to the definitions

463

General Index

467

Acknowledgments

The theory and methodology presented in this work did not, of course, just drop from the sky. Though the essentials of the theory may have had their actual origins in flashes of 'inspiration', the details and the ultimate formulation, not to mention the probable causes of the 'inspirations' themselves, are the products of environment. The theory in its present form is the product of a long development in which innumerable people and circumstances have played a part. It would therefore not be practicable to mention more than a few individuals and groups, namely those I can think of as having played a very direct role. If I take it chronologically, I must first acknowledge my debt to my early teacher in linguistics, E.M. Uhlenbeck, who instilled in his students not only a great enthusiasm for the subject, but who also encouraged a critical posture. In connection with this I have to express gratitude to my fellow students in the 1950s, with whom I had discussions often until deep in the night. One of them was D.G. Stuart, now Professor of Phonetics in Groningen, The Netherlands. Talks with him had a great influence on my phonological thinking, and through that on many other aspects. All this was in Leiden, The Netherlands. It was Stuart who advised me to go to Paris for a while to study Andr6 Martinet's 'functionalism'. Martinet's ideas of the late 1950s became very much my point of departure, and though I have diverged in many respects very widely from this, we have up to now always remained in contact, and I owe it also to Martinet in the first place that my associates and I have found a constant forum for our ideas in the International Society of Functional Linguistics and its journal 'La Linguistique', with both of which I have been actively associated. From 1961 to 1968 I was at Oxford where my, for that time perhaps, rather weird seeming ideas only slowly found some people, mainly students, willing to listen. One of those was M.A.L. Lamb, at present lecturer in Linguistics at the New University of Ulster,

χ.

Northern Ireland. To him I owe an important debt, and no doubt some of the things found in the present work originated in and after early conversations with him. Another, somewhat later, was S.G.J. Hervey, whose D.Phil, thesis on Semantics I supervised. He eventually developed the semantic part of the present theory, without which I would have been seriously handicapped in fully developing the theory outside the realms of methodology, ontology, and phonology. Some time after I had moved from Oxford to St. Andrews, Scotland, Hervey became my associate, moving up through the ranks from research assistent to Reader in Linguistics. Consequently, his influence on, and contributions to, the theory, especially, but not exclusively so, in respect of sharper formulation, must be greater than either of us can be consciously aware of. The reverse is, of course, true as well. Coming back to my Oxford days, I owe very special gratitude to the, in those days, very militantly hostile transformationalists and the equally, though more mildly, hostile attitude among adherents of the more typically British empirical tradition. Though being a serious handicap at the time, and for a long time thereafter, this turned out to be beneficial in the end. Those two prongs soon forced me into a deeper study of logic and the philosophy of science, in order to be able to defend my ideas more effectively. Thus my brand of functionalism became 'Axiomatic Functionalism'. I also owe, already in my Oxford days, gratitude to W.de Pater, now Professor of Philosophy at Leuven (Louvain), Belgium, who guided my first steps in formulating the first two axioms (the second one is now different) and some of the definitions. With de Pater I had numerous discussions on methodological issues. Philosophers in Oxford I further owe gratitude to are L.J. Cohen and W.C. Kneale, who were both involved in the research leading to my D.Phil, thesis 'Sets and relations in phonology — an axiomatic approach to the description of speech', Oxford 1966, published by the Clarendon Press in 1968 at the advice of L.R. Palmer, then Professor of Comparative Philology at Oxford. I owe gratitude to Palmer for his support and advice in those days. I should also mention A.H.C. Ward, Professor in Chinese, Toronto, Canada, who was on study leave in Oxford, and at a later date in St. Andrews. His influence was especially great on my grammatical thinking, which was in the beginning still very 'orthodox' functionalist. Ward was working on a syntax of Classical Chinese, and,

xi

attempting to use functionalist ideas, he was able to put his finger on many flaws and inconsistencies, and this has lead me to devise a completely new approach to syntax. I should also mention all my St. Andrews students and colleagues. Teaching leads to better formulation, and this in turn leads to a better understanding, not least by the author of the ideas in question. Not only the criticisms and objections by students, and indeed by associates, have been useful, but even more useful have been what in my opinion were wrong applications or even unproductive tinkerings with the theory. Many articles I have written, and much in the present work, especially in Chapter V, were in the first place prompted by this. As to the actual assistance in producing the manuscript for this work, I have to thank S.G.J. Hervey for reading and giving advice on the first three chapters, and the first half of Chapter IV. All such advice was carefully considered and usually acted upon. Very special thanks go to M.A.L. Lamb who went through the whole of the work with a fine tooth comb, commenting on almost every sentence, and even rewriting a few in his own razorsharp style. I did not agree with some of the observations he made, but even in those cases I always saw his point, and this forced me to reformulate the passages concerned. In respect of Chapter VI I owe gratitude to Sheena Gardner, who was at that time producing a Ph.D. thesis under my supervision on 'the sentence'. Some of her remarks have also lead me to make some minor changes to Chapter VIII. I also thank the University of St. Andrews, who gave me a grant towards the typing of an earlier version of the manuscript. Furthermore I owe gratitude to my wife Dora, and my sons Jan and Frans, who indirectly supported my work in various ways. I should also not forget to thank the 'Deutsche Forschungsgemeinschaft', whose generous grant enabled me to stay in Freiburg i. Breisgau, West Germany, for thirteen months (19851986), and I owe gratitude to the various friends and people interested in my ideas over there. Discussions with them have been very fruitful. To mention only a few, with whom I have spent perhaps more time than with others, there were Herbert Pilch, Professor of English, on whose recommendation I was able to come to Freiburg, Joachim Thurow, Lilo Moessner, Axel Koberne, Brigitte Gulden, Cynthia Dermarkar, and not to forget

xii

Heidi Poschbeck, the able and extremely helpful secretary of the 'Englisches Seminar'. Without the facilities available to me at Freiburg there would have been further delays in publishing this work. My stay in Freiburg had an important influence on my thinking and has led to small, but important, improvements, which are mainly incorporated in Chapters V and VI. In a similar vein I have to express my gratitude to the 'Institute of Language and Culture for Rural Development' of Mahidol University, in Thailand, and to its director Dr. Suriya Ratanakul, who invited me to teach for eleven months (1984-1985) at the beautiful campus at Salaya, about twenty kilometres outside Bangkok. I thank staff, as well as students for their overwhelming friendliness and cooperation. I cannot mention them all, but two whom I should mention are Mrs Pitsamai Yupho (teaching member of staff) and Miss Chutima Maneeruengdej (research student), with both of whom I had many discussions on matters of Thai phonology and syntax. It was not only the exposure to the Thai language, but also the circumstance of becoming indirectly involved in methodological problems concerning field-work, which influenced my thinking. This led me to reconsider some phonological descriptive techniques, and to rewrite parts of, and add a long section (Chapter V, 19) to, Chapter V. I also thank Dr. Pinit Ratanakul, Ms. Apinya, and other staff of the 'Research Center of Southeast Asian Cultures' for their friendship and assistance in numerous ways. Finally I thank Professor Werner Winter for judging the final manuscript and giving valuable advice. Freiburg, 1st June 1987.

Chapter I: Introduction 1. Foreword to the introduction The theory of axiomatic functionalism, the basics and an outline of which is presented in this volume, is a peculiar theory. The basis of the approach — and this includes theory as well as methodology — is vastly different from what most readers will be used to in linguistics in general, even though some of it may be superficially reminiscent of that used in Hjelmslev's theory of glossematics. I have to say this right from the start, and it should be constantly kept in mind, because when the discussion proceeds this fact may be in danger of being lost sight of. The reason for this is that, qua content, the approach is a functionalist one, and one will find all sorts of concepts mentioned that seem utterly familiar to anyone who has any knowledge of European linguistics in general, and Prague-school, and functionalist, and glossematic linguistics in particular. But in spite of the apparent familiarity of the concepts, each one of them is ontologically different from a concept, perhaps of the same name, or even of closely similar material content, employed elsewhere, for the simple reason that it has a different ontogenesis. The distinctive feature of Axiomatic Functionalism, and the prime (but not the only) reason for the axiomatic approach adopted, is that the ontogenesis of concepts is traceable step by step. I firmly believe that only in this way can those concepts be fully known and understood, and that only under those conditions' can one proceed to analyses and explanations that are truly and fully scientific. It was for this reason that already in the early sixties I started to deviate from the more orthodox Saussurean and Functionalist ways of thinking, and conceived of the approach that is now called 'axiomatic functionalism'. All this has implications for the presentation of the theory, and therefore for the form and content of this outline, as well. It

2

Introduction

means, in fact, that references to concepts in other linguistic theories have to be treated with extreme care, and that there are frequently even cases in which it may superficially seem that such reference is due, but that making direct references would be downright misleading. This does not, of course, mean that in building this kind of theory I am not thoroughly indebted to other linguists past and present. But it does mean that the exact nature and extent of the debt remains in most cases unknown and untraceable. If, therefore, references to and comparisons with other theories or approaches, other than in this introductory chapter, are less frequent than is common practice, I hope that it is accepted that there are good grounds for this. Discussion of different types of approach in this work especially in the first three chapters is necessary in order to set the scene, and to highlight the ways in which the axiomatic functionalist approach is different. In doing this the treatment is often rather global and generalizing. It is, therefore, inadequate as an outline or concise description of the approaches in question, but it is not meant to be one. The only reason for the critical discussions is to provide contrastive illustrative material in order to throw the different nature of the axiomatic functionalist approach into relief.

2. Descriptive linguistics - objects and methods. The functional principle One of the most thorny questions that a linguist is usually called upon to answer is: 'What is linguistics?'. The best answer to such an unfair question is actually the non-answer: 'Whatever linguists are prepared to call linguistics'. The usual answer, however, that is given is: 'The scientific study of language'. The advantage of the latter answer is that it has the effect of a gob-stopper. It may give the layman the impression that now he knows, but in fact he still does not know very much. In the first place there is no consensus of opinion as to what counts as scientific, but the layman is entitled to understand it in the vague sense of being recognized by sophisticated people as scholarly. Even less elucidating than the term scientific is the use of the term 'language' as an undefined term. But let us assume also here that it can be taken in the vague sense

Descriptive linguistics - objects and methods. The functional principle

3

of 'everything that has to do with speech or writing and the, specifically human, faculty of being able to communicate, and learning to communicate, by means of speech or writing'. As it is immediately obvious that this is much too wide, even the layman will understand 'having to do with' in the sense of 'having as its specific object', as there is very little in society and culture that has nothing to do with speech or writing, if only in the sense of presupposing or being dependent on them. But even so, the scope remains too wide, and, more importantly, it would still be impossible to get consensus of opinion about where the borderline between linguistics and some other disciplines would have to be drawn. Is stylistics, for instance, a subdiscipline of linguistics, or even the approach initiated by such linguists as Mathesius, and continued by Firbas and others, which sometimes goes under the name of 'studies in functional sentence perspective', and which in many respects is the continuation of some ancient interests already pursued by traditional grammarians? Or even traditional grammar itself. Is this linguistics? Perhaps it lacks scientific (in the vague sense) respectability, but lately it has become again respectable by the upsurge of transformational grammar, which latter, too, in many respects continues traditional grammarians' interests and modes of thinking. And what about transformational generative grammar itself? It is suggested, if not actually claimed by generativists, that structuralist linguistics is merely taxonomic, not explanatory, and therefore pre-scientific. The actual truth is, however, that generativism and structuralism are so basically different that they can scarcely be called the same discipline, which gives each of them the right to refuse to regard the other as linguistics as soon as one regards one's own approach as being linguistics proper. There is also linguistic philosophy of more than" one kind, logico- or mathematico-linguistics, machine-translation, socio-linguistics (a field rather than a discipline) and several kinds of psycho-linguistics. Furthermore, there are comparative and historical linguistics and related disciplines, which can claim to have a longer history than structuralism has, even though present day comparative linguistics, etc., has undergone much influence from more recent traditions. In short, there is not one approach or theory that embraces them all. Consequently we must either say that linguistics is not a discipline, or we must, arbitrarily to be sure, understand by that term, for idiosyncratic purposes, a

4

Introduction

severely limited field of study. I propose that the main criterion for the limitation of this field shall be that the whole field can be covered by one single unified principled approach. This field is then to be called 'linguistics' for our purpose, without, of course, denying to others the right to call any other such arbitrarily selected field 'linguistics', or even to refer by that term to any conglomeration of such fields. We may recognize that any approach falling within this field may be called 'linguistics', irrespective of whether the approach is in our view scientific, or partly so, or not. Such questions will be dealt with later, and they will be offered as a justification for our own approach, which is in fact — so we claim — not just a principled approach, but a proper scientific theory. But there may be more than one field that can be established by the above criterion. For instance, comparative and historical linguistics could constitute one field and synchronic descriptive linguistics another. There could be some overlap between those two fields, but one cannot conceive of a truly unified and yet manageably simple approach — rather than an eclectic one — which could have them both within its scope. Psycho-linguistics and socio-linguistics may have to constitute several fields each with some partial overlapping. I cannot see, to give further examples, the study of functional sentence perspective or 'discourse analysis' as being truly integrated with it, so these would probably have to constitute different fields as well. Generative linguistics need not a priori be excluded from among the synchronic descriptive approaches, and a choice between it and, say, some structuralist approach has to be made on some other, presumably epistemological, level, as has the choice between various structuralist approaches, or choosing to create one's own novel one. It is clear, therefore, that we have to add further criteria for choosing one's field, i.e. what will constitute 'linguistics' for our purpose. One common way of doing this is choosing between corelinguistics and other fields. By core-linguistics is usually meant 'synchronic descriptive linguistics'. Comparative or historical linguistics is excluded from it on the grounds that the latter is to some extent dependent on the former, but not, as a field, vice versa. The interdisciplinary approaches are, of course, automatically excluded. I shall refer then to the selected field by calling it 'descriptive linguistics', which is actually a misnomer,

Descriptive linguistics — objects and methods. The f unctional principle

as the other fields are also descriptive in one way or another. The term 'descriptive' has here to be taken to mean 'having the description of speech-phenomena as its object'. With speechphenomena is simply meant speech-events, usually protocolized (see below), as they can be observed by someone who is in full possession of the faculty of speech himself, and who, moreover, fully understands the messages conveyed. That is to say, the observer should be able to hear, as well as understand, sentenceutterances in the vehicle in question. For practical reasons, the linguist may wish to extend the range of the term 'speech', mutatis mutandis, to 'writing', and everything that has been said here, again mutatis mutandis, applies to the description of other semiotic systems (see below) as well. Though in the explanations and exemplifications I intend to concentrate on linguistics proper, the theory I am going to outline in this work is a semiotic theory which properly includes a linguistic theory. 'Descriptive linguistics' should be distinguished from 'linguistic description'. The former is the whole field of study. Within this field there are different approaches, methodologies and theories. Any of these, applied to speech-phenomena, yields a 'linguistic description'. In a more active sense, i.e. as an activity, one can call 'linguistic description' the application of a descriptively orientated linguistic theory of a kind (this includes approaches and methodologies that are perhaps not proper theories in our sense) to speechphenomena. The result of an act of linguistic description is a description of the speech-phenomena in question, i.e. a linguistic description. The scope of a linguistic description is an arbitrarily selected field of speech-phenomena. The selection of the field involves purely practical considerations. As these considerations must be satisfied, we may say that such selection is arbitrary but nevertheless appropriate in respect of one's aims. The scope of a linguistic theory is any potential description under that theory. The scope of the theory to be presented in this work is in fact the description of any semiotic system, including any of those semiotic systems commonly referred to as 'languages'. The objectives of the theory are to provide the means whereby semiotic systems, including 'languages', can be described in a fully scientifically justified and adequate manner, and in this way to

5

6

Introduction

contribute to the understanding of communication by semiotic conventions in general, and communication by 'language' in particular. In fact, a theory that satisfies these objectives cannot fail in itself to provide an understanding and insight into the most human of human faculties, i.e. that of speech. It will not give an insight into the psychological aspects, nor into sociological ones, but it will give an insight into speech in general as an observable phenomenon. There have been, and are, approaches that attempt to go beyond this limited aim, but in so far as they are successful — if they are really successful at all, which remains doubtful — they pursue their aims, at least partly, by methods that are not acceptable under the epistemological and scientific criteria that we set ourselves. This is not surprising. If we look around us, we see that in the more rigorously organised sciences there is a severe restriction imposed upon the types of aspect that are to be considered as falling within each of their respective scopes. The wider the scope in that respect, i.e. the more ambitious a particular discipline or theory within a discipline is, the more one has to sacrifice with regard to rigour. Rigour should not be confused with formalization, as formalization upon weak epistemological foundations can only amount to pseudo-rigour. It is perhaps possible in principle to have a thoroughly rigorous theory with an extremely complicated or multiple aim — one could, for instance, conceive of the possibility of having one integrated and unified theory covering what now is covered by physics, as well as chemistry, and perhaps even biology, but such a theory would be so complex and unwieldy as to be completely useless, except maybe in the hands of some universal genius. Though perhaps no hard and fast rules can be given for the minimum measure of simplicity that theorybuilders must comply with, simplicity is nevertheless an important factor in practice. Therefore, a theory that wishes to claim to be a truly scientific theory should be as nearly monistic as is practicable. In the first place, it should limit itself to the description of one type of phenomenon only, e.g. the speechphenomena rather than the intuitions of the producers of those phenomena, but not both. Even this is in practice too wide, and one has to limit oneself to only a few, ideally only one, aspect of those phenomena. For a functionalist this aspect is the functional aspect, i.e. the so-called 'functional principle' is applied and constitutes the filter through which the phenomena pass, as it

Scientific linguistics. Epistemological considerations

7

were, before they are considered worthy of treatment at all. The functional principle states that nothing is considered semiotic, unless, at some level at least (e.g. phonologically, grammatically, or otherwise), it is separately relevant to communication. This functional principle, introduced by the first axiom of the theory, pervades the whole theory and is relevant to every part of it. By limiting the scope of the theory it also determines what may be relevant and what not in the phenomena that constitute the objects of the descriptions.

3. Scientific linguistics. Epistemological considerations In choosing the type of theory one is going to accept or, if one does not find a theory that satisfies one's criteria for a good theory, the type of theory one is intending to create, one first has to set up or adopt criteria for philosophical or epistemological acceptability. Actually, if one wants a scientific theory, one becomes automatically involved in the philosophy of science, and especially in such questions as what 'science' actually is, or what it should be. The question 'what is science?' is no less thorny than the question 'what is linguistics?'. The fact that the English language — unlike, say, German, where there is the one word Wissenschaft — distinguishes between science and art is no great help, as any epistemological criteria we may set up will cut across such a division. There is no logical, nor factual, coincidence between object of investigation and methodology, except for a very weak one in that imponderable and unobservable objects (e.g. some of those posed by theologies) can only be dealt with in a speculative manner. But the converse does not hold. Even physics could conceivably have remained a speculative discipline. As the history of science shows, scholarly disciplines progressively tend to abandon speculation, as soon as this has become possible. It is true that this more often than not diminishes the 'ethos' of a subject, but so does writing in dry precise language, rather than in poetry or fine prose. It is also true that to admit some degree of speculation

8

Introduction

may add a great deal of interest to a subject, but, in my opinion, this does not enhance its status as a science. It may seem superfluous to say this, but I have heard a distinguished speaker in a linguistic conference advocating speculativism on precisely those grounds, without encountering more than a minority adverse reaction from the audience. At least some transformationalists now openly admit to be being speculative, and defend it. I shall later try to show that transformationalism, as soon as it operates with such notions as 'competence' and 'deep structure' is inherently speculative, unless these notions are purely operational, i.e. instrumentalist (see below), in which case the term 'competence' is a misnomer, and highly misleading. The use of the term 'competence' also indicates that transformational grammar is not merely descriptive in our sense, but includes within its scope also some areas which belong to the realm of psycho-linguistics. However, as a so-called grammar, apart from being 'a model of competence' also purports to be some kind of structural account of the language in question, transformational generative linguistics cannot be denied the status of belonging to core-linguistics as I have circumscribed it. Its scope is, however, wider than corelinguistics alone in respects too, which I shall refrain from dealing with. To come back to the subject of speculative approaches, already for the above indicated 'historical' reasons, one can deny that they are worthy of the epithet 'scientific' in a modern sense. It could, of course, have been the case that descriptive linguistics had to be speculative by the very nature of its subject-matter, like, for instance, theology, and some branches of psychology. If, however, it can be demonstrated that it can be scientific and nonspeculative — and for those areas for which this can be demonstrated — speculativism has to be rejected. Now I believe that it has been shown many times over since the beginning of this century that speech phenomena are capable of being dealt with in a scientific way, and the fact that transformationalism could get off the ground at all is therefore highly remarkable. The fact that it was a reaction against naive realist Bloomfieldianism and also against the rather sterile operationalism of Harris is not sufficient to explain this. There were almost certainly also some important socio-political factors involved . . . call it the spirit of the sixties, which has brought us many unexplainable twists of mind and

Scientific linguistics. Epistemological

considerations

9

forms of extremism. At any rate, from the point of view of proper science, it marks an important step back, a step towards increased speculativism, the method of the alchemists. I shall later try to demonstrate that transformational generative grammar is not purely speculative, but that it also has features that belong to inductivism, especially in the form of 'universalism', and also a strong instrumentalist or operationalist flavour, all being reasons why it has, in my opinion, to be rejected as being an epistemologically acceptable approach. Its self-proclaimed hypotheticodeductivism is — I shall try to demonstrate — not a hypotheticodeductivism in the Popperian sense at all; neither does the approach contain an axiomatic system in any normal sense of the term 'axiomatic'. The reasons why I have to dwell so long on speculativism is that owing to the success of the transformationalist movement a whole generation of linguists has grown up who know very little else than the transformational generative approach, and who have become accustomed to a way of thinking in which speculation and other scientifically unacceptable practices are the order of the day. They know no better than that this is entirely proper, and they have been so indoctrinated by slogans — the heritage of the early sixties, when transformationalists, just as members of other youthful groupings were particularly militant and sure in their opinions — that they feel certain that all other ways of thinking in linguistics are inferior, and not worth becoming acquainted with. From the philosophy of science, and from such books as Kuhn's The structure of scientific revolutions, they only extract, often via misinterpretations, whatever may seem to justify their own way of thinking and existence. As their discussions and debates are almost exclusively with linguists of the same persuasion, the degree of intellectual inbreeding is such that little from the outside can touch their beliefs, beliefs which are in this way constantly reinforced. We shall see later that not only the beliefs of transformationalists are untouchable, but that also everything they produce is virtually immune from criticism. This is yet another reason why transformationalism should be rejected, as immunity from criticism is contrary to the scientific spirit of conjecture and potential refutation (see Popper, 1965). Another approach, in many respects related to speculativism is what Popper has called instrumentalism. Indeed, instrumen-

10

Introduction

talism or operationalism turns speculative as soon as one attaches claims, or even merely hypotheses, of existence or reality to purely operational notions. To give an example, in order to account for the existence of the Universe one could pose 'God, as the Creator' as a purely operational notion, but as soon as one believes in the existence of God, and in the actual Creation by God, it has become speculative. For theology, speculation is a respectable method, as it is for any other activity in which one ponders the imponderable. Sciences or half-sciences that are ambitious enough to admit imponderables, inscrutinables, or unobservables among the objects they deal with have to admit speculation in their methodology, unless they are prepared to turn to operationalism. In so far as they admit speculation, they are, in my opinion, not sciences, and it is doubtful that they are truly scientific if they merely become instrumentalist. If one requires of a science that it be 'descriptive' in any real sense, one must reject instrumentalism or operationalism as well as speculation. Speculation is, in a sense, even preferable to instrumentalism, as it does create a link with some kind of imagined, and perhaps even possible reality. That is, it is likely to be intellectually, though not scientifically, more satisfying. This is probably why instrumentalism almost always leads to some degree of speculativism. Instrumentalism is an approach in which a calculus is set up which, in its end result, accounts for some state of affairs, but not every part of the calculus has necessarily any connection with real existence. It is typically the verifiable end result that governs the acceptability of terms in the calculus. If the proper end result is achieved by manipulating the terms, the terms are said to be empirically adequate, and the calculus is accepted as being a description, or even explanation, of the state of affairs in question. I am simplifying here, as the same terms may be used in more than one calculus, which, of course, complicates the issue. But the principle remains the same, and I am therefore for the sake of clarity presenting here, and in the following, a much simplified picture. A Chomskyan type grammar is a typical example of an instrumentalist calculus. Also Harris's operational approach is an example. For the rest, instrumentalism is scarce in linguistics, except in the rather mild form of using one or two purely operational concepts in order to make an otherwise fully empirical calculus work. But even this is dangerous, and is, in my opinion,

Scientific linguistics. Epistemological considerations

11

to be rejected. It is not only misleading, but it may even lead to absurdities. Take the notion 'morphophoneme' for instance. It is used in order to clear an already existing absurdity, i.e. the widely held view that a linguistic sign, qua form, can be divided into phonemes. This is absurd, because the form in question is a form in relation to meaning. But phonemes are not entities in relation to meaning.* Therefore it is necessary to introduce another concept, which is thought to be connected with meaning. This concept, known under various names, e.g. 'systematic phoneme' (in earlier transformational theory), 'morphon' (in stratiflcational linguistics), and 'morphophoneme' or even 'morphoneme' (by various schools), leads, however, itself to an even greater absurdity. It implies that, say, English 'book', which is of the phonological form/buk/as corresponding to any single member of a particular class of denotata, can be analyzed into /b/, /u/, and Ik/, each separately corresponding to a book as a single member of such a class, or part or aspect of a book. But let us consider here what descriptive or explanatory value we have to attach to a completely instrumentalist calculus. If the answer has to be: no value at all, then a calculus using instrumentalist terms has a diminished scientific value in as far as such terms are used and as far as the deductive consequences of the use of such terms reach. This could be much further than one might think. Let me first say that we have to distinguish between the introduction of only one term in a self-contained calculus, and the introduction of more than one term. Certainly, as in mathematics, one can introduce one unknown, because the operational value of an unknown can be derived by deduction from the other terms and the end-result together. Such a procedure is not necessarily harmful, and may even lead to future discovery. But in itself, it should not be confused with an actual descriptive concept, as it does leave uncertainty as to the exact nature of the reality covered by that term. An example of this is the concept of a 'black hole' in astronomy or astro-physics. But truly objectionable instrumentalism comes in when one operates with more than one unknown, let alone when all terms are purely operational. And if one does operate with one unknown, but this leads to absurdities as in the * I use here, of course, the terms 'form' and 'meaning' in a rather loose sense. Greater precision will be introduced later in this work.

12

Introduction

example of the notion 'morphophoneme', this unknown is, of course, not even a permissible term in the first place. It merely amounts to cancelling out an initial error by committing another one. This also signals the danger there always will be in introducing even only one unknown. It is quite possible that the necessity of, say, posing 'black holes' is due to an initial flaw in the theory. Even in physics, if the consequence of certain views is the logical possibility of, say, time travel, but if certain consequences of this lead to absurdities, these views ought to be seriously re-examined, as they constitute an intolerable form of instrumentalism. In a generative grammar a succession of re-write rules is introduced. The terms of these re-write rules are, in spite of being reminiscent of familiar traditional grammatical concepts, purely formal and void of empirical content. Transformationalists are at great pains to point this out. If one replaced say, NP by A, and, say, VP by B, and so on, nothing would change, provided that the terminal symbols are such that by transformations the surface structure of actual sentences is arrived at. As long as the result of this calculus, which takes the form of an algorithm, is all and only the sentences of the language in question, the calculus is considered empirically adequate, and so is any part of the calculus. As obviously there may be more than one calculus that is empirically adequate, one should be able to choose the best one of them. In practice, the main criterion for this is the relative measure of simplicity. Transformationalists must have been aware of the fact that there is no necessary link between the terms, the parts, as well as the progression of the calculus, at least not prior to the introduction of terminal symbols, and the actual sentences themselves, other than the calculus proper. It is, perhaps, for this reason that they have characterized the calculus as being a 'model of competence', thereby creating the impression that the algorithm is in some way isomorphic with the actual competence of 'the native speaker', 'The native speaker' is, of course, not an entity that exists at all, even transformationalists would not claim that it does, and also its 'competence' has only an ideal existence. It is, namely, the imaginary mental ability underlying the performance of an imaginary ideal speaker. We cannot therefore accuse transformationalists of speculativism — 'competence' is purely notional —, nor can we accuse them of not explaining or describing anything at all, because their 'grammars' are not purely operational,

Scientific linguistics. Epistemological

considerations

13

but are models of something, i.e. of 'competence'. The only way of escaping between the horns of this dilemma is, I think, by assuming that, absurdly enough, they are both speculative and purely operational at the same time, or rather, they are not, whichever of the two becomes inconvenient when criticised, or when transformationalists are puzzled themselves, if they are ever puzzled. This is another feature of transformational grammar, i.e. its inherent evasiveness and relative immunity from criticism. The immunity from criticism is not only due to a most ingenious elusiveness or vagueness of concepts, but it is an inbuilt capacity of instrumentalist (and speculative) approaches. An instrumentalist approach need never lay itself open to a charge of being lacking in adequacy. This is because, as I pointed out earlier, there is basically no restriction as to the terms that can be introduced in order to make the calculus work. Suppose one knows for a fact that a certain sum total of something is, say, ten, and one also knows that there is at least one sub-total of, say, four. The simplest way to arrive, then, at ten is to say either 4 + 6 = 10, or 6 + 4 = 10. But suppose that this, for one reason or another is not plausible, and that, for intuitive reasons, or even for reasons of convenience or consistency with some other assertion elsewhere, one has to posit at least one further term. One can then arrive at ten by saying: 1 + 3 + 6 = 10, 4 + 1 + 5 = 10, 4 + 2 + 1 + 3 = 10, or 1 + 3 + 2 + 4 = 10. All these are equally adequate, but only one of them can be in agreement with the (partly hidden) facts. If they are also equally plausible, the simplicity criterion is adopted, but this still leaves an equal choice between the first two. This is, of course, awkward, and some 'rule' has to be invented, perhaps not lacking in both ingenuity and plausibility, to make an unequivocal choice possible. This is yet another feature of transformationalism — and it adds no little to its attractiveness. The ingenuity and often unbridled fantasy with which rules, especially transformational rules, are invented, makes the whole procedure a most interesting and enjoyable intricate intellectual game which is bound to fascinate the participants. Whether this, however, should be called 'science' is, in my opinion, debatable. Let us now take a look at the nature of descriptive models and their role in science, and finish with an example from the transformational-generative approach itself. In many sciences one of the main descriptive activities is the creation of models. This is

14

Introduction

especially apparent in the structural (as opposed to causal) sciences. Models are structures that stand in a certain relation of isomorphism with observable phenomena. Actually, to say 'observable' phenomena ought to be considered pleonastic, as only observables are phenomena, but this is often forgotten. If models are analytical, i.e. complex, every sub-part of the model is to be a model itself. The degree of analyticity of a model stands in direct relation to its descriptive and thereby explanatory power. A monistic model is not descriptive at all, but amounts, at most, to a simple definition of a term. Models may range from scalemodels, to diagrams, to algebraic equations. As long as they stand in the above-mentioned relation of isomorphism with phenomena that constitute their purport, they are proper models, and as long as they themselves consist of models, they are analytical, and thereby descriptive, models. Explanation in the structural sciences is achieved by mapping phenomena onto a descriptive structure i.e. an analytical model. There is no reason to reserve the term 'explanation' for causal description of some kind, as has been done by transformationalists and some others. 'Description' in the structural sciences explains the phenomena by means of structures onto which they can be mapped. In this way it reveals a composition of the phenomena in question, and their interrelations. By doing this from diverse points of view correlations and relations between different aspects of the phenomena are established, and by this the phenomena become understood in their functional capacity. Explanation is making something understood, and this is what a structural description does. I have said that the components of an analytical model are models themselves. If those models are themselves analytical models, their components, in their turn, are models themselves, and so on, until one arrives at component models that are not analytical. It should be remembered that something is a model if and only if it stands in a relation of isomorphism with the relevant phenomena, i.e. if there are particular specific phenomena that can be mapped onto it. But in an instrumentalist approach it is not a requirement that terms in a model, say a 'grammar' in a Chomskyan sense, be themselves models, i.e. terms that stand in a direct mapping relation with phenomena. A Chomskyan type 'grammar' as, for instance, proposed in his Aspects of the theory of syntax, has a dual function. It is, like the type of 'grammar' proposed in

Scientific linguistics. Epistemological

considerations

15

his earlier Syntactic Structures, an account of the language in question, i.e. it is to have a structural descriptive function. As a language for Chomsky is the set of actual sentences, which as types stand for actual sentence utterances as tokens we can say that a Chomskyan 'grammar' ultimately among other things purports to be an account of the phenomena. Apart from this, a 'grammar' is a 'model of competence'. As Chomsky believed in distinguishing between 'descriptive' and 'explanatory' functions, the latter being more orientated towards 'causality', he needed to posit this double quality for his grammars, as otherwise his grammars could be accused of having no explanatory power. In order to make the two functions consistent with one another he had to concede that his phenomena are of an ideal, rather than of a real kind. The grammar was, then, an account of speech phenomena as they might be imagined to be produced by imaginary real speakers. Ideal performance is the reflection of competence. The fact that Chomsky's phenomena (our terminology, as elsewhere) are, in more than one respect, not actual phenomena, but they are, as it were, idealised phenomena, is not relevant to what follows. As a Chomskyan grammar with regard to its possibly being a model of competence is wholly dependent upon its being an account of ideal performance as well, we may investigate in how far it can possibly satisfy the latter function. If it cannot, it cannot satisfy the former function either, and Chomsky's claims with respect to the explanatory power of his grammars are wholly unsubstantiated. As was said earlier — and transformationalists agree with this — the non-terminal symbols in a 'grammar' do not stand in any relation with reality, except that they contribute to the final result, i.e. the sentential output. Also the rewrite rules fit into that category and transformations have no other formal justification than that via them the desired output can be achieved. Consequently, a Chomskyan grammar is not an analytical model, and can, at best, in its totality, be regarded as a definition of the language in question. But as definition is senseless unless it is used in further procedures, it fails in that respect as well. In fact any claim as to the descriptive power of such a grammar could be compared with a claim by someone that a certain sum-total, say one thousand, is properly described by saying that it is generated by ρ + q + r + s, i.e. by saying that p + q+ r + s= 1000. As one can

16

Introduction

attach any value to p, q, r, and s (i.e. link them by a rewrite-rule to terminal symbols), as long as one makes sure that the attached values together amount to one thousand, this formula does not tell us anything about how one thousand was arrived at, except, perhaps, that there are exactly four terms, but this too would have to be independently justified before it provides information of any value. But one does not need an example from mathematics to see the emptiness of the approach. One has only to contemplate and ask oneself, when presented with a Chomskyan type grammar, what actual (as opposed to imagined) insight one has gained into the language in question. Transformationalists may point to the plausibility of their rewrite rules. But plausibility is a matter of intuition and speculation. Though it may be used in the adequation of a model, i.e. in asking oneself if the final result of a calculus is a plausible one in respect of what we actually observe, it is not a precise enough criterion to be used during the calculus, if further deductions are still to be made. Plausibility or intuitive satisfactoriness is very much a product of our personality and upbringing. Influenced as most of us are by an Aristotelian-humanistic tradition, stipulating that, say, S >N Ρ + V P, may seem very plausible indeed, even after it has been stressed that Ν Ρ and V Ρ are empty symbols that might as well be replaced by A and B, or any other set of empty symbols. Without direct, both scientific and purely linguistic, justification for such a rewrite-rule, it is imprudent - to say the least - to use the criterion of plausibility in a calculus that purports to have any descriptive value. Besides, even mentioning it is already inconsistent with the assertion elsewhere that the terms of rewrite rules are purely formal. As is constantly the case, transformationalists are completely evasive in letting us know which one of two or several contrary statements, suggestions, or implications we have to hold on to. This is understandable, because holding on to only one may be satisfactory at specific points but it may lead to inconsistencies elsewhere. The only way to avoid this is to switch to and fro between interpretations. Let us turn to the Eskimos for another comparison. Whether or not my ethnological information is entirely correct is irrelevant to the principle involved or the force of the illustration. In order to account for the fact that children are born, the Eskimos have

Scientific linguistics. Epistemological

considerations

17

established a model which serves the dual purpose of being a model of competence, as well as a model describing the chain of events that lead to childbirth. The comparison is not perfect, as the second model is of a causal descriptive, rather than of a structural kind. This, however, only affects the issue at stake here to the effect that in the Eskimo model, unlike in the Chomskyan model, there is a direct natural link between the model itself, and the assumed competence it is said to be a model of. That is, the Eskimo model is, in this respect, an analytical model. The main purport of the model is, as I said, to be a model of competence. It purports to explain why Eskimo women are so competent in becoming pregnant and producing offspring. The causal descriptive part is of a generative kind. Indeed, to use Chomskyan terminology, the model can be said to be a 'grammar of childbirth'. Though it is not put in the form of rewrite rules, the principle with which we are concerned here is the same. In this case a succession of plausible events is generated. The Eskimo ladies walk, with their seal-skin boots, in the snow. The snow is populated by little spirits. These work their way up the boots, hide there for a while, and then enter the trousers. From there they work their way into the womb. There they might simply grow, until they are big enough and sufficiently developed to be born. However, up to this point the model is not fully adequate. It does not account for the fact that unless there is a man sleeping in the same igloo as the woman, no child is going to be born. Fortunately this is a speculative-instrumentalist model, and therefore new explanatory components can be brought in at will, as long as they are sufficiently plausible. This immunizes the model from failure, as is always the case with speculative or instrumentalist approaches. The husband is now brought in as a factor. His role is to feed the baby while it is in the womb. In the absence of a husband the spirit will not grow into a foetus, and ultimately baby, and it will disappear again at the time of menstruation. One sees that even the latter is explained, and the fact that it stops when conception is successful. All of this is completely plausible to the Eskimos, and is regarded by them as perfectly adequate as an explanation of childbirth. The model is as satisfactory and attractive to Eskimos as a transformational grammar is to transformationalists. If we reject the Eskimo model as not really describing, and therefore in fact not really explaining anything, we

18

Introduction

have in my opinion to reject Chomskyan models on the same grounds. If the term 'model of competence' can mean anything, it must at least imply being suggestive of some causal chain. In that respect, the Chomskyan model may well seem to many of us even more fantastic than the Eskimo one, if only because, unlike in the case of the Eskimo model, it is a priori implausible that it is in any way isomorphic with the actual processes involved in producing sentences. It is true that transformationalists do not maintain this, but then we may well ask what the relevance of the grammar as a model is, and what it could possibly mean to call it a 'model of competence'. I consequently reject the transformationalists' approach to linguistics on grounds of scientific irrelevancy. If it were not for the fact that the approach has, surprisingly enough, achieved such a powerful position on the international linguistic scene, we would be fully justified in ignoring it altogether. It marks a setback in linguistic, and indeed scientific, thinking, the equal of which has perhaps not been seen since the dawn of the Middle Ages.* With the rejection of transformationalism — which typifies both — I reject speculativism as well as instrumentalism or operationalism as valid methods in scientific research. After having separated non-scientific types of approach from the scientific ones we are still left, within the latter, with two basically different schools of thought. These two are, broadly speaking, the inductive and the hypothetico-deductive attitude. I have said here 'attitude' in preference to 'methodology', as especially inductivists do not always practise what they preach. This is, in my opinion, because strict induction, i.e. to derive organized and mutually consistent knowledge from the phenomena, is only possible for the simplest of universes. Suppose that one tries to describe a language by pure induction. Without an a priori criterion of relevance one would not even know where to start, and what aspect to deal with. Even the simplest of objects have in principle an infinite number of aspects. Some inductivists believe that objects and phenomena have a fairly limited number of inherently essential aspects. It is those aspects that they intend *

Even though the strength of their position is now rapidly declining, it is likely to be for many years that the spirit of generativism will still remain with us.

Scientific linguistics. Epistemological

considerations

19

to study to the exclusion of all others. I shall overlook the possibility that this already introduces either an arbitrary theoretical or philosophical criterion, or some speculative element, as inductivists may maintain that what is essential is immediately evident from the phenomena themselves. Let us assume that they may be right, and that they can proceed to making descriptive statements immediately from a sufficient number of observations, keeping in mind that also inductivists, as good scientists, require of descriptive statements that they be consistent with one another, and that this is an overriding consideration. It is, perhaps, easy to control such consistency between the first ten or so statements, but the more the body of statements expands, the more difficult it becomes to be certain of its overall consistency. It is as if one had a bag full of fleas one wished to count. Perhaps one can take a few out of the bag without difficulty, but the more one takes out, the more will escape before being caught and counted, until in the end the procedure becomes totally chaotic. Perhaps the analogy is not in all respects a proper one, but the fact remains that dealing in a direct way with an intricate body of phenomena is a well-nigh impossible undertaking. Therefore inductivists too allow themselves shortcuts, in the form of educated guesses which can well be called hypotheses. They also develop rules of thumb and so-called discovery procedures, which implies admitting some kind of aprioristic theoretical considerations. What I want to say is that in practice inductivists also have some kind of hypotheticodeductive attitude, and even some aprioristic kind of methodology which is in some respects akin to what one might call a theory. The reason why, in my opinion, inductivism has to be rejected is, however, not in the first place because inductivists tend to play down the role of hypotheses and the arbitrary (i.e. aprioristic) nature of their methodology. It is rather the fact that they do aim at a theory, but it is a theory that purports to be some kind of universal grammar. Such a theory is arrived at by generalization from particulars. Within a field of speech-phenomena, they select a so-called corpus of data. Preferably this is a sufficiently large corpus, i.e. large enough in practice to derive significant generalizations from. It is generally not stipulated what the requirements are for generalizations to be significant. Within this corpus one finds recurring forms and patterns. Forms and patterns are said to be recurring if they are not significantly different. Again there

20

Introduction

are no independent criteria of significance. For this one would have to have a theory in advance, but inductivists wish to arrive at a theory a posteriori. One may also ask how one could recognize patterns, e.g. how could one decide — to use an extremely simple example — by mere observation, without criteria a priori, whether, say, 'the very old man' is structured out of the components 'the' and 'very old man'; 'the very old' and 'man'; 'the', 'very old' and 'man'; or even 'the', 'very', 'old', and 'man'; or 'the very', 'old' and 'man'. It could, perhaps, be maintained that the last two analyses are ruled out by intuition, but is this observational intuition? Though observation is a matter of intuition, not all intuition is a matter of observation. As I said, this is an extremely simple case, but nevertheless different inductivists have come up with different analyses for cases of this kind, and — even more significantly — they were invariably obliged to devise certain a priori considerations in order to justify the choice of one analysis in favour of others. This is again an illustration as to the fact that the inductive method is not possible in practice. A record of patterns and forms, according to inductivists 'discovered' within a corpus, is called a linguistic description. When a sufficient number of corpora, ranging over a sufficient number of fields (i.e. natural languages), has been described in this way, one may discover similarities, and of course differences, between those descriptions. One may discover recurring forms, patterns, and categories across descriptions of different fields of phenomena. These forms, patterns, and categories can then be generalized, provided that such coincidences are abundant enough to arrive at significant generalizations, often of an either formally or informally statistical nature. There are here, again, no criteria for 'significance', and if there were they better had remained unmentioned, as they would be aprioristic, and this is not tolerable to true inductivists. Some forms, patterns and categories (let us call them 'features') may be found in all descriptions, i.e. they may pertain to all corpora hitherto described. Such features are said to be universal, Alas, inductivists frequently seem to confuse 'all' with 'any'. Only if they were found in all descriptions, using exactly the same methodology, of any corpus could the feature in question be said to be universal. But it is this very inductivist philosophy that prevents one from statements about 'any language1, except when one confuses it with 'all languages hitherto

Scientific linguistics. Epistemological considerations

21

described under a particular method'. There are only two ways by which one can arrive at 'any language', i.e. either by hypothesis, or by postulates, i.e. by an a priori theory. Though both are actually used by inductivists, they are foreign to the inductivist philosophy. Scholars like Bloomfield, the father of inductivism in linguistics, and some of his followers, e.g. Pike, openly admit some kind of postulates, but proclaim, or imply, that those are arrived at by proper induction; but in that case they are not true postulates, and the same consideration applies, i.e. they cannot pertain to 'any' language, except in a hypothetical fashion. But, of course, the said Bloomfieldian type postulates do contain theoretical a prioris, and we may, in fact, ask ourselves how far universals that derive from a priori definitions or premises can be truly called universals. Several theories — using the same terms with various different meanings — assert that in any language there are phonemes, words, syntagms, sentences, intonations, a double articulation, etc. But, we may ask, does it mean anything to say this? We may as well attach significance to saying that all bodies have molecules, all molecules have atoms, and all conglomerations of molecules have measure. If they did not, they would not be bodies or molecules. That is, all these features derive from definitions in a priori theories. All one can say about such definitions and theories is that they are appropriate to the objects within their scope, but this does not diminish the fact that, say, the specific gravity of 'silver', like that of any other element, is not a fact at all, but is due to a certain theory that is applied to the phenomenon 'silver'. Without the a priori arbitrary notion of 'specific gravity', plus the arbitrary convention of stipulating that the specific gravity of water provides the standard unit, no element would possess it, nor could any element fail to possess it under present scientific theory. It is exactly the same with the above-mentioned linguistic notions. Without a theory they would not exist, and given the theory that contains them, and given a definition of language that excludes semiotic systems that do not contain them, every language must contain them. Chomsky has called universals of the latter kind: 'formal universals', as opposed to 'substantive universals' (Chomsky 1965:27-30). This shows that transformationalism has not completely freed itself from its Bloomfieldian inductivist heritage. To recognize the existence of formal universals, and to attach any importance to them at all, is based on a typical inductivist misconception.

22

Introduction

Let us now turn to the so-called 'substantive' universals. These can - as will have become clear by now — at best be regarded as statistically significant generalizations, whatever 'significant' here may mean. There is, of course, nothing a priori objectionable about generalizing about languages, and, in this way, looking for similarities between them. However, two dangers lurk here, and inductivists — though they are not alone — have appeared to be largely unaware of these dangers. In fact, one of them has not only caught inductivists — and to some extent others as well unawares, but invariably constitutes the very essence of the inductivist method. It is the danger of incorporating the set of descriptive generalizations, which — we should remember this for inductivists are inductive generalizations, into one's body of theory. Whether the hypothesis that such generalizations are truly universals remains unchallenged or not is largely irrelevant because they can only be assumed to be universals ex hypothesis and this has to be tested. What is relevant is the fact that, when one applies a theory that includes inductive generalizations, even if they are assumed to be universals in new fields of phenomena, the effect will be that of a Procrustean bed. F o r instance, when one has the concept of a 'verb', or 'the passive', or any other such concepts, including even say such features as 'vocalic' and 'consonantal', as a universal in one's theory, and this theory is applied in the description of languages, one will be almost certain to 'discover' verbs, the passive, and the features 'vocalic' and 'consonantal', in Chinese, Hottentot, Blackfoot Indian, and whatever language one describes. This, in its turn, of course, strengthens and corroborates the idea that these are true universals, but such corroboration is clearly circular. The other danger is not unrelated to this. It is the temptation to rarefy the meanings one attaches to universalist notions to such a degree that it becomes very difficult indeed not to find the universals in question in any language. An extreme example of this is Fillmore's case-grammar, in which the familiar 'cases' of some European languages, and some others as well, are recognized as existing in any language. This type of universalism almost inevitably leads to the recognition of a distinction between a deep structure which is the same or virtually the same for all languages, and a surface structure, which accounts for the differences in usage, which are thought to be rather superficial. For the area of grammar proper, i.e. for morph-

Scientific linguistics. Epistemological considerations

23

ology and syntax, it also leads, again almost inescapably, to some form of conceptualism. That is, one recognizes a world of pure concepts, distinct from, but underlying, the world of language utterances. The structure of utterances is not in those utterances themselves, or in their direct descriptions, but in the concepts which underly them, or rather their representations. Description then consists of mapping the surface realizations via a translationmechanism, called 'transformational component', onto a representation of the relevant part of the conceptual world, called 'semantic representation'. Or, to put it in reverse, one recognizes a conceptual base common to all languages (note the multiple speculative element introduced) for which one establishes a semantic representation. Such a semantic representation has, of course, to take the form of sentences, not so much linguistic sentences, but rather sentences of another semiotic system. This semiotic system is either wholly or partly created for the purpose, or one uses wholly or partly existing systems, e.g. some system of formal logic, if necessary in an 'enriched' form. In either case the semantic representation is an arbitrary, though appropriate, representation of a largely speculative conceptual system. Such a system is regarded as being the base of any natural language. The sentences of the representation are now translated, in a fairly mechanical way, by the transformational component into actual sentences of the language in question, or rather into representations of them. This type of approach, with some variations, is adopted by those transformationalists that call themselves 'generative semanticists'. It is, perhaps, not difficult to see that their type of approach is the almost inescapable consequence of reflecting on certain features already inherent in Chomsky's Aspects of the theory of syntax. Much of the criticism I have launched above against Chomsky's instrumentalist-speculative approach is also relevant here. But what I am concerned with at present is the fact that the true nature of conceptualism together with transformationalism has been made clear in such a lucid fashion by generative semanticists that criticizing it has become much easier. It is considerably more difficult to get a firm grip on the orthodox Chomskyan approach, owing to the evasive nature of the way in which it is stated, especially by Chomsky himself. It being clear, in the generative semantic approaches, that the transformational component is no more than a formal and mechanical translation

24

Introduction

device, and it being equally clear that the semiotic system underlying the semantic representation is an arbitrary one, one could ask why we could not merely select a natural language, e.g. Arabic, English, or Chinese as the semiotic system for the base. A generative account of the sentences of, say, French, and their structure, could then be the structure of Arabic, English, or Chinese plus the necessary transformations for translating Arabic, English, or Chinese, into French. The answer to this is to be found in the first place, in an extension to the ancient, especially eighteenth century, philosphers' dream of discovering the Universal Language. That is, for rational thinking to be optimally successful there should be a semiotic system that can be the vehicle of pure concepts — a system that is free of any ambiguities, and does not contain any formal, nor substantive redundancies. Such a nonambiguous, maximally simple system, if it existed, would be able to handle any rational problem in a most efficacious manner. Generative semanticists are in the first place concerned with trying to discover this Universal Language which they call 'the semantic representation'. Their efforts are justified by the consideration that, if such a system were found, and if we could reasonably assume that the system that we established was either the only possible system of the required kind, or, if not the only possible one, we could assume that among the various possibilities this one was the very one that underlies all languages, we could do away with conventional descriptive linguistics altogether. All we would need in order to account for the variety of languages would be to set up for each of them a translation mechanism, i.e. a transformational component. This would, of course, be a major step forward, and the generative semanticists' aim is therefore a laudable one. To make the whole of linguistics redundant, and replace it by a series of completely mechanical transformational components, is even outclassed by a discovery of the Universal Language itself. Think of the effect that this would have on the whole of science and civilization. But this is true of, say, the alchemists' quest for the discovery of a way to make gold, or the search for an elixirium vitae, which would prolong life indefinitely, as well. The fact that success in the latter would make much of contemporary medicine redundant is not a sufficient reason for closing medical research departments in a hurry. Moreover, as long as the ultimate goal of generative semantics remains an object of

Scientific linguistics. Epistemological

considerations

25

pure speculation, we might as well make sure that at least some researchers carry on with the much less exciting, even if perhaps seemingly old-fashioned, work of trying to build, refine and improve theories for the actual description of languages. Until such time as generativists have come up with the final solution, and this is no longer based on pure speculation, we are entitled to claim not only scientific, but also practical validity for a purely descriptive method, and characterize the generative semanticists' methodologies as unscientific. This is not to say that their work may not turn out to be of value in areas that are outside our immediate spheres of interest, e.g. automatic translation, the development of systems of formal logic, etc., but even this has become more and more doubtful in recent years. To be sharply distinguished from the above-mentioned types of universalism is the establishment of restricted inventories from which all languages are said to select their own set of substantive entities. Here it is not claimed that a particular item is found in any given language, but that there is no language that does not draw all of the relevant part of this inventory from the universal set. A classic example of this is Roman Jakobson's universal inventory of distinctive features. Another example is Chomsky's set of conceptual features as suggested, though not exactly presented or established in to to, in his Aspects of the theory of syntax. The latter is clearly inspired by the Jakobsonian approach. In the case of Chomsky's so-called selection restriction features, in agreement with the methodology employed, there are probably instrumentalist or speculative elements, but let us assume that this is not the case. In that event we may perhaps not object to the actual type of list, but only to the rules for its application in deciding whether certain sentences are fully grammatical, or less grammatical. Certain types of grammaticality for Chomsky are, in my opinion, not of a linguistic kind, but they are based upon considerations as to what the real world is like. That is, we could imagine other worlds, as for instance that of Lewis Carroll's Alice in Wonderland in which, for our world fairly ungrammatical sentences would be perfectly grammatical, because there is something for them to refer to in that imaginary world. Are we then to say that the language of Alice in Wonderland is not the same as that used in say Charles Dickens's Great Expectations? But apart from that, we may not a priori deny the possibility of having

26

Introduction

a universal inventory of conceptual features which is comparable to Jakobson's universal list of distinctive features. We may even acknowledge the fact that having such lists is a prerequisite for the approach to be inherently generative, rather than that a description should be imagined to act merely as a processing device that converts external substance into potential sentenceutterances, which is in principle what one should require of a real description. Indeed, any description, even one without explicitly generative claims, if it is exhaustive also in respect to realization, can be 'activated' by turning it upside down, as it were, and can produce new data on that basis. Such a description can, therefore, be seen as a potentially generative device. But if we require a theory (as opposed to a 'description') to be inherently generative, we must have within the theory Jakobsonian type inventories of simple entities, both in phonology and in grammar. Only then can a theory uniquely determine descriptions under that theory (cf. Chomsky's Aspects, where such a device is indicated). The question we now have to pose is, in the first place, whether an inherently generative approach has advantages over other approaches. If it has, we should consider the further implications as to its desirability from a practical point of view, e.g. feasibility and relative measure of simplicity. Though the latter should be regarded as secondary, we may, for instance consider whether an exhaustive description of a natural language under Chomsky's Aspects model would be at all feasible in human terms and, if it were, whether it would not be so unwieldy and immense that it could, for this reason alone, already have little practical value as an account of the language in question. One could, perhaps, object that this is not the point, as in principle one could set up such a 'grammar' part by part, according to which area of the language we are focussing on at a given moment, and, in principle, at one time or another all relevant aspects of the language in question might come to be dealt with in that way. It could with reason be claimed that under most other approaches as well, for a description to be maximally exhaustive would imply an intolerable degree of unwieldiness. But in these approaches it is usually a matter of lesser or greater generality, i.e. greater or lesser descriptive detail. Granted that total exhaustiveness, especially in the area of semantics, would be less than feasible for

Scientific linguistics. Epistemological

considerations

27

descriptive approaches too, these approaches are generally able to present a clear picture with room for further details to be added when so desired. Not so with a generative device like the Aspects model. Here even every lexical detail has to be accounted for, right at the beginning, in the base. At every stage everything relevant to that stage has to be exhaustively treated, and it is not permissible to fill in certain details later. This state of affairs is inherent to the generative approach adopted by the transformationalists, but not to most conventional descriptive approaches. An exception to the latter is stratificational linguistics which suffers from a similar practical handicap as the generative models do, but then stratificational linguistics is in principle generative itself, even though perhaps not yet fully developed in that respect. But at the same time it is fully descriptive. I have to say this with certain reservations, as what we know as Stratificational Linguistics is mainly representational, i.e. it is a sophisticated, in fact generative, representation of a Hockett (neo-Bloomfieldian) type of analysis. How the description is arrived at is not shown. Stratificationalists seem to assume that, at least within the neo-Bloomfieldian world, the actual linguistic items and their structures are no longer a matter of dispute. Unless, of course, they assume that the ultimate result justifies the initial analyses, as is the case for transformationalists. In that case, stratificational linguistics too is instrumentalist in nature. But let us assume that this is not the case, and that there is an analytical procedure underlying it, even though this is hidden. Then for stratificational linguistics to be essentially generative, one can envisage the possibility of providing Jakobsonian-type inventories for both the hypophonemic and the hypersememic levels, in which case also stratificational linguistics might, in principle, be capable of specifying the class of possible languages, just as is claimed for transformational-generative linguistics. If there were not the above mentioned practical objections, nor further possible objections of an epistemological kind, which we shall be investigating presently, a stratificational description as a generative device would be preferable to a transformationalist one, as it is not open to the same charges of being inherently speculative or instrumentalist, at least not on such a vast scale. It is true that in stratificational linguistics there are some instrumentalist correctives, e.g. the notion 'morphon' (morphophoneme), and some stratificationalists, e.g. Lockwood, adhere

28

Introduction

to the speculative 'black box' philosophy (see below), but every level in their approach is empirically justified by direct mapping reference to the data as entities outside and independent of the description. Therefore stratificational linguistics is in addition not open to the same charge of circularity as transformational linguistics is. Let us now investigate whether there are epistemological grounds on which we might wish to reject the establishment of Jakobsonian-type inventories as constituting an integral part of a theory. But before we do that, we may as well ask ourselves whether there is any advantage at all in being able to generate all and only (in my opinion 'all' should be read 'any') the grammatical sentences of any language. For a descriptivist like myself the only possible use this could have is in the adequation, i.e. corroboration of the hypothesis of 'adequacy', of the description and indirectly of the theory. But for this, the potential generativity of a description itself, as opposed to an inherent generativity of the theory, is totally sufficient in principle, and infinitely more feasible in practice, if only because there is no need to go through the performance of setting the whole machinery in motion every time we want to test the adequacy of a descriptive statement. But, as we have seen, a transformational grammar does not contain descriptive statements. In fact, if the transformationalists' claims were justified, statements of the descriptivists' kind would become superfluous. There only remains the other claim, i.e. that their causal type of 'grammars' is explanatory in respect of the 'competence' of native speakers. We have seen that such a claim is speculative and therefore — we have maintained — it is of no scientific validity. But charges like this cannot be levelled against, for instance, a generativized stratificational approach, or any other approach that could be envisaged in which a generative procedure was step by step descriptive as well. But as also in such approaches one would need to have Jakobsonian-type inventories within one's theories, we must now investigate whether there are objections that can be levelled against this. If not, such a procedure may well be considered to have important advantages, as it would facilitate the comparison of languages, but I shall not go into that aspect of the problem here, as this is outside our immediate sphere of interest. Let us devote our attention to Roman Jakobson's universal inventory of distinctive features, as any general conclusion reached

Scientific linguistics. Epistemological considerations

29

from this will apply to any inventory of that kind, including conceptual 'alphabets'. As the present work is not a primer of linguistics, I may assume that the reader is acquainted with Jakobson's list, and I need not describe it. But we do have to consider its precise ontological status, which is that it is a hypothesis about languages. All languages are assumed to possess distinctive features that are accounted for in this inventory, and no others. This is a true hypothesis, as, in principle, it can be refuted, in which case the inventory has to be revised, or another distinctive feature or features will have to be added. The inventory is, therefore, in principle, an expandable one. I assume that Jakobsonians and other linguists who employ such inventories agree with this assessment, as, if they were to take such inventories to be final and not hypothetical, i.e. aprioristic, they would be open to similar charges as I have levelled against the other type of universalism, i.e. that their employment constitutes a Procrustean bed. Any statement about what the phenomena are actually like is inherently a descriptive statement. But if descriptive statements are proclaimed to be a priori true we get into a situation of describing and prejudging at the same time, which is, of course, absurd. Therefore, unless we reject a Jakobsonian-type approach out of hand, we must assume in such an approach that the inventories in question retain their descriptive character, and have a hypothetical status. Could having inventories of this kind within the theory be consistent with the approach adopted by axiomatic functionalists? In the axiomatic functionalist approach, as we shall see, a sharp distinction is drawn between theory and descriptions. The former is in the first place an instrument for the description of languages. It contains no other statements than axioms and definitions, and those statements have, to use Hjelmslev's words, no existential import. Descriptions, on the other hand, do contain descriptive statements, and those are statements about observables that are subject to potential falsification of their being in agreement with the facts to be observed. Though descriptive statements are not generally hypotheses themselves, they are arrived at via hypotheses. It is the theory that justifies their launching. That is, the theory determines what constitutes a well-formed and meaningful hypothesis. It is, of course, out of the question that a theory of that kind could contain hypotheses, because, if it did, descriptions in which such

30

Introduction

a theory was employed would be question-begging. We could, of course, regard the hypothetical character of a Jakobsonian-type inventory to be free of a priori existential import. That is, we could stipulate that the actual hypothesis has to be taken as stating that 'if there are languages, then they draw their distinctive feature inventory from this universal list'. This, however, would not really help, as the fact remains that the list is in principle flexible and expandable, which implies that it has existential import. A better way to escape between the horns of this dilemma would be saying that the list has a theoretical aspect, as well as a descriptive one. As an item of theory it would be axiomatic, but it would also have an outside status as a hypothesis. The consequence of this would, however, be grave, as this - we have seen this already from several angles — would make the theory useless as an instrument for description. It would not be a true theory, but a Procrustean bed. Inductivists, by the essence of their philosophy, could not object to this, but in our approach this would be intolerable. Yet there is nothing objectionable in having inventories of that kind as such, provided that they do not form part of what we call the theory. A theory of the kind we envisage is a both sufficient and necessary instrument for linguistic description. It is, by itself, not a generative device, as a necessary condition for having a generative device is the setting up of inventories of a both distinctive features and conceptual kind. But a generative device cannot be a descriptive device as well. Moreover, a generative device is tautologous or circular by its very nature. Being a Procrustean bed is its virtue not its vice. This is yet another reason why the two functions, if we want them both, have to be strictly separated. The question whether we want them both or not boils down to the question whether we are satisfied with the potential generative power of descriptions, which are capable of generating sentences of one particular field of speechphenomena, i.e. the field constituting the scope of the description, or a device that specifies all sentences in all possible languages. But if we do insist upon the latter, we still need to have a way to distinguish between different languages, and this we cannot do without feeding the generative apparatus, presumably at the level of its so-called terminal strings, with items from a very restricted inventory, i.e. an inventory connected with one specific language only. This is all that transformationalists have managed to do so

Scientific linguistics. Epistemological considerations

31

far, and this does not surpass what can be done by any descriptivist by merely turning a description upside down, as it were. In spite of all their claims, transformationalists have not brought us one step nearer to the promised land where we should find the final solution, and this at the cost of creating chaos, not just in linguistics, but in views on established scientific procedure as well. I would not have dwelt so long on transformational grammar and its speculative tendencies, if it had not had such a penetrating influence on some basically scientific contemporary approaches. Those transformationalists who do not openly admit to speculation, clothe their approach in a hypothetico-deductive garb, and this attitude has been adopted by some representatives of other linguistic schools as well. What I am referring to here is what one might call the 'black box philosophy'. A tagmemicist, W.A. Cook, quoting J.G. Kemeny, says for instance: 'Suppose you have a locked box, which cannot be opened without destroying its content. One can observe how the box works, and predict how it will act' (Cook 1969:2). Note already here the introduction of what is, at best, a half truth 'one can observe how the box works'. This should actually be interpreted as 'one can observe its output', but this is not the same as observing how it actually works. Cook still quoting Kem6ny continues: 'We can form a theory as to what kind of machinery is in the box, and we will maintain that theory as long as our predictions concerning the box are verified' (Cook 1969:2). The wording here too is highly, if conveniently, misleading. The term 'theory' does not mean at all what scientifically one understands by 'theory', no matter whether one understands it in an inductivist sense, or as an aprioristic decision device. The word 'theory' here has the ordinary lay-man's sense of 'bold speculation'. At the same time 'theory' is confused with description. The term, therefore, here simply means 'speculative description' or 'descriptive speculation', but this meaning may not be immediately apparent to the unwary. I am, of course, not assuming that Cook wilfully tries to mislead the reader, but only pointing out the devious method of argumentation that has become commonplace, in linguistic writings of the last three decades. Also the term 'hypothesis' is often used in the sense of 'speculation', a sense that has very little to do with scientific hypotheses. Cook goes on to say — thereby betraying even more his transformational source for this part of his philosophy - that

32

Introduction

the linguist 'constructs a formal grammar, a statement of competence, and maintains his theory as long as it continues to predict accurately the performance of native speakers' (Cook 1969:2 my italics). What Cook is advocating is actually to guess what the contents of the box are — he even uses the term 'guessing' in the accompanying scheme —, to make a description of this guess, and call this a scientific account of the real thing. More specifically he advocates following the philosophical practice of transformationalists. The hypothetico-deductive disguise comes in when one calls such a procedure 'launching a hypothesis', and one regards the outcome of one's predictions as the test-case for either refutation or corroboration of this 'hypothesis'. That is, another misleading element is introduced here, i.e. the procedure superficially has features in common with the hypothetico-deductive method, but a veil is drawn over the all-important respects in which the latter is intrinsically different. There is yet another, all too common, confusion involved i.e. failing to distinguish between necessary and sufficient conditions in an argument. It would be quite correct to claim that the mechanism that one imagines to be in the box should of necessity have the same output as the real mechanism actually has, if there is going to be any chance of one's imagination coinciding with reality. To have identity of output is, however, not a sufficient argument for concluding identity of mechanism. There is yet another thing that Cook has learned from the transformationalists, especially Chomsky himself. This is, always to leave a back door open through which one can escape, and so make oneself immune from criticism. Cook, thus, hastens to add that 'this formal grammar has only a relation of equivalence, not a relation of identity, with the fundamental grammar in the mind of the speaker (Cook 1969:2)'. In what the 'equivalence' lies is not stated, but the implications of this statement are elastic enough to include the weakest possible sense of 'equivalence' here, i.e. a mere simulation of output. Taking it in this sense would reduce the account to having a similar value (at least in pre-scientific times) as that of the story of the tower of Babel as a causative account of the diversity of languages. Also this is reminiscent of the procedures adopted by transformationalists, as I have pointed out repeatedly in the foregoing. The 'black box' parable was used by Chomsky in a discussion with Stuart Hampshire, broadcast by the BBC in May 1968. In defending the view that 'the best

Scientific linguistics. Epistemological considerations

33

evidence as to what is a universal property of language comes from the study of a single language', he said: 'Suppose you think yourself as a scientist who's trying to study a black box which has certain data as input and a grammar as output — well, any property of the output which goes beyond the organization in the input must, vaguely speaking, be attributed to the character of the device.' Nobody can have any quarrel with this, nor is this a specifically interesting observation, as it is so obvious that it is barely worth stating. Chomsky continued: 'We have to ask what in the device led the output to have this specific property.' We may well ask that, but how do we get the answer? Though the rest of the quotation is not of immediate relevance to what we are discussing on these pages, it is so surprising and at the same time revealing as to the way transformationalists tend to reason, that I may as well give the remainder. Chomsky went on: 'Anything that must be attributed to the device, by virtue of the study of this input/output relation, must be a universal of language - that is, if we assume that people are not genetically endowed in such a way that they can learn one rather than another language. And by posing the problem in that manner we can really get a great deal of very strong evidence about what must be a linguistic universal from the deep study of a single language. Of course we have to test that against comparative evidence. Such a hypothesis must stand the test that it is not refuted by some other language. In this sense comparative evidence is relevant but it is actually less illuminating, at the moment at least, than deep study of a single language.' We have seen above what this 'deep study' involves, i.e. setting up an instrumentalist or speculative device for the simulation of sentence production. To come back to the point, D.C. Lockwood, a stratificationalist, argues along similar lines as Cook does, invoking astronomy as an example of a science in which the 'black box' approach is employed. This is another favourite way of white-washing a speculative approach. Also transformationalists, in discussions, frequently point out that in the natural sciences too this sort of procedure is adopted. The argument is then that a method employed by the natural sciences is by definition scientific. But, in the first place, what is an acceptable scientific procedure for one science may not be so for another, just as an acceptable procedure for theology, or psychology, may not be so in physics and chemistry. Why should

34

Introduction

then the fact that a procedure is used in, say, astronomy or chemistry justify its adoption by linguists? Secondly, we must distinguish between the ordinary language use of the term 'science' or 'scientific' - science is what scientists do - and a more technically defined use of this term. It is quite possible that a lot of what scientists do is not 'science' in the latter sense, i.e. not scientific. According to reasoned criteria I have adopted and stated above, a science is unscientific to the extent that it indulges in unnecessary speculation. I say unnecessary speculation, as, for instance, evolution theory (e.g. Darwin's) could not even exist without a great deal of speculation. The aim there is to create a link (which remains speculative) between observables at different stages in history by providing the most plausible explanation. If, say, the physicist would do the same, it would be grossly unscientific. But we should be much less hard on even the most advanced of the natural sciences if they indulged in some mild form of instrumentalism. The reason for this is that natural scientists on the whole are at least equally interested in predictions and practical results as they are in the acquisition of knowledge and true understanding of the phenomena. Therefore, rather than wait for the means of observation, which are a prerequisite to observation itself, the latter being a prerequisite to description and explanation, they are often tempted — and, in view of their aims, perfectly entitled — to introduce instrumentalist concepts. I have discussed above why such concepts do not actually further our knowledge. But in the natural sciences, rather than being a hindrance to the acquisition of future knowledge, such instrumentalist concepts frequently lead to discoveries once the means for observation are there. There are at least three reasons why this is possible. In the first place these instrumentalist notions are used as sparingly as possible and more often than not involve the substitution of one single unknowable factor by an instrumentalist concept. In the second place, generally speaking, the assumed phenomena are not intrinsically unobservable, but are only so by the accident of our lack of technical skill. They may become observable at some time in the future. Least of all are they imponderables. Thirdly, scientists on the whole, make a conscious effort to develop the technology for the exploration of precisely those unknown factors that are arrived at by pure calculus, so that they may become empirical objects as soon as possible. Even

Scientific linguistics. Epistemological

considerations

35

though such instrumentalist notions may satisfy their practical aims, they do not satisfy their descriptive curiosity. They remain constantly aware of the element of uncertainty that is introduced by the deployment of instrumentalist notions. I do not have to explain in detail how different this is from the way some contemporary linguists, especially transformationalists, deal in instrumentalist notions and speculations. Instead of one single unknown factor within a multitude of observational certitudes, they pose a complicated fully instrumentalist 'grammar' which links another unobservable to the set of sentences which are observable. Even if the grammar constituted a substitution for an unknown factor in a calculus in which the other factors were known, it would already be unfair to compare it with methods used in the natural sciences. But what is done here is far more objectionable. It is as if one had χ + y = 12. One now guesses that y = 8, and then uses y = 8 as evidence for χ = 4. If we replace jc by 'competence', '+' by an appropriate relator, y by the grammar, and '12' by the language, we are not far off from what transformationalists actually do and call 'scientific'. I am not sure whether 'competence' is intrinsically unobservable or if it is accidentally so, but this is irrelevant unless not only ways in which it could be observed are explored, but one has also a fairly clear picture as to how it could be observed provided the technical means were there. It could, of course, be argued that in order to fulfil the aims of transformational-generative linguistics, one needs to have an instrumentalist or speculative device, and that, therefore, one has to abandon the desire of being truly scientific. It could be argued, that like the natural sciences the practical result is, in first instance at least, as important as acquiring real knowledge. The acquisition of knowledge is then put off to a time far in the future, in nevernever land. Given this aim, transformational-generative grammar is not only perfectly entitled to adopt this instrumentalist-speculative attitude, but it is this very attitude which guarantees success. The procedures are fully commensurate with the aim of producing the sentences of a language, and no other procedure is known to exist that could do that any better. Further research in transformational linguistics should then consist in making the production apparatus for producing sentences even more consistent, more efficacious, and increasingly simpler. But there are two things we may say with regard to this. In the first place, transformationalists

36

Introduction

should cease to imply that such a procedure leads to an analytical knowledge of languages and of the competence of native speakers, rather than to an ever deeper and more penetrating knowledge of generative grammars themselves. Secondly, is producing sentences really the aim of transformationalists? Is it the set of sentences they are after, or is it the grammars generating those sentences? If it is the grammars we may again ask what use they are. We have seen that they have no use as descriptive and explanatory devices, either with regard to actual languages, or with regard to the competence of native speakers. We are then left with their only use, i.e. the production of sentences, which is perhaps not the actual aim of transformationalists, nor is it easy to see what use in itself this could really have.

4. The hypothetico-deductive method After having rejected instrumentalism and speculativism on the grounds that adopting these methods diminishes the scientific value — sometimes as opposed to practical value - of one's approach, and after having rejected inductivism on other grounds, we are in practice left with what Karl Popper has called 'the hypothetico-deductive method' (Popper 1965). I should hasten to say that adopting the latter does not imply rejecting empiricism in a wider sense. In fact, as we shall see, in my version of the hypothetico-deductive method, in which a sharp distinction is maintained between theory and description, the antithesis between 'hocus-pocus' and 'God's truth', or, if you wish, between rationalism and empiricism becomes resolved, that is, as long as 'hocus-pocus' or 'rationalism' remains free from instrumentalist connotations. Popper expresses himself in the following way: 'The theory . . . stands directly opposed to all attempts to operate with the ideas of inductive logic. It might be described as the theory of the deductive method of testing, or as the view that a hypothesis can only be empirically tested — and only after it has been advanced . . . . From a new idea, put up tentatively, and not yet justified in any way — an anticipation, a hypothesis, a theoretical system, or what you will — conclusions are drawn by means of

The hypothetico-deductive

method

37

logical deduction . . . We may if we like distinguish four different lines along which the testing of a theory could be carried out. First there is the logical comparison of the conclusions among themselves, by which the internal consistency of the system is tested. Secondly, there is the investigation of the logical form of the theory, with the object of determining whether it has the character of an empirical or scientific theory, or whether it is, for example, tautological. Thirdly, there is the comparison with other theories, chiefly with the aim of determining whether the theory would constitute a scientific advance should it survive our various tests. And finally, there is the testing of the theory by way of empirical applications of the conclusions which can be derived from it (Popper 1965:30-33)'. This testing, according to Popper, is done by attempting to falsify singular empirical statements or predictions. Popper says: 'It should be noticed that a positive decision can only temporarily support the theory, for subsequent negative decisions may always overthrow it. So long as a theory withstands detailed and severe tests and is not superseded by another theory in the course of scientific progress, we may say that it has 'proved its mettle' or that it is 'corroborated'. Nothing resembling inductive logic appears in the procedure here outlined. I never assume that we can argue from the truth of singular statements to the truth of theories. I never assume that by force of 'verified' conclusions, theories can be established as 'true', or even as merely 'probable' (Popper 1965:33)'. Unfortunately, Popper — in accord with standard philosophy of science — fails to distinguish between theory and description, i.e. between purely formal theoretical statements and statements derived from hypotheses about the phenomena. In fact, in the above quotation, Popper's term 'theory' refers to four different things. The first two mentions refer to his own philosophical methodology. When he talks about the logical form of the theory, the third could refer to description and scientific theory indiscriminately. In his reference to the comparison with other theories it has to be taken as 'the theory proper', whilst when referring to empirical applications he must mean 'a description under the theory in question". In the remainder there is the same confusion. The term theory here refers to description, except in two cases. When he talks about superseding by another theory, it must mean "theory proper', and when he mentions that theories cannot be

3X

Introduction

'true', nor even ' p r o b a b l e ' , w e m a y as well t a k e it as r e f e r r i n g t o b o t h t h e o r i e s and deseriptions. Surprising

as

it

may

seem

that p h i l o s o p h e r s

have

remained

c a u g h t and e n t a n g l e d in this k i n d o f c o n f u s i o n , it is, p e r h a p s n o t all that surprising if w e c o n s i d e r t h a t t h e y have d r a w n m o s t

of

their illustrations in that respect f r o m the natural sciences. It is precisely

the d i f f e r e n c e b e t w e e n

t h e natural s c i e n c e s , and such

social s c i e n c e s as cultural a n t h r o p o l o g y , s o c i o l o g y , a n d linguistics, that the f o r m e r e x h i b i t u n i t y o f universe, w h e r e a s the latter are, when

it

c o m e s t o d e s c r i p t i v e activities, f a c e d w i t h an

infinite

n u m b e r o f parallel universes. T h o u g h also in t h e natural s c i e n c e s — I a m g e n e r a l i z i n g and s i m p l i f y i n g t h e issues i n v o l v e d - t h e scientist m a y o c c u p y h i m s e l f w i t h o n e s p e c i m e n at a t i m e , the e m p h a s i s is generally o n a c c o u n t i n g f o r relevant a s p e c t s o f the w h o l e o f a particular

universe.

Chemists,

for

instance,

are

not

so

much

interested in a p a r t i c u l a r l u m p o f silver, b u t rather in t h e c o m p o s i t i o n a l p r o p e r t i e s o f silver in general. P h y s i c i s t s are i n t e r e s t e d in falling o b j e c t s , rather than in a p a r t i c u l a r falling a p p l e , w h i l s t zoologists may f o r statistical

investigate large n u m b e r s o f single rats, p e r h a p s p u r p o s e s , b u t their aim is t o arrive at s t a t e m e n t s

a b o u t rats in general. A t a n y rate, t h o u g h in t h e a c t u a l p r o c e d u r e s there m a y

be s o m e o v e r l a p b e t w e e n natural sciences and social

s c i e n c e s , t h e f o r m e r are n o t g e n e r a l l y f a c e d w i t h t h e d e s c r i p t i o n o f o n e single universe t o w h i c h the w h o l e o f a p a r t i c u l a r t h e o r y is a p p l i e d in o r d e r t o arrive at the d e s c r i p t i o n , and t h e n h a v i n g t o repeat t h e w h o l e p r o c e d u r e f o r a n o t h e r t o t a l universe o f a similar k i n d . T h i s is. f o r i n s t a n c e , t h e s i t u a t i o n in linguistics, w h e r e o n e m a y m a k e d e s c r i p t i o n s o f single languages, o n e at a t i m e , in e a c h o f w h i c h d e s c r i p t i o n s t h e w h o l e t h e o r y is a p p l i e d . I a m , o f c o u r s e , n o t r e f e r r i n g t o t h o s e linguists w h o s e aim it is t o arrive at a 'general g r a m m a r ' . F o r t h e m , separate languages have a status n o t u n l i k e that o f ' s p e c i m e n s ' in the natural s c i e n c e s , w h e t h e r t h e y f o l l o w m a i n l y inductivist p r o c e d u r e s o r n o t . W h a t I a m here s p e c i f i c a l l y referring t o is w h a t w e have classified as core-linguistics, and a similar c l a s s i f i c a t i o n can be m a d e w i t h r e s p e c t t o o t h e r

social

sciences. F o r t h e s e reasons it is m u c h clearer in t h e social s c i e n c e s t h a t a d i s t i n c t i o n m u s t be d r a w n b e t w e e n s t a t e m e n t s a n d c o n c e p t s w i t h e x i s t e n t i a l i m p o r t and t h o s e w i t h o u t . I say ' m u c h clearer', as s u c h d i s t i n c t i o n s exist in t h e natural s c i e n c e s as w e l l , b u t t h e f a c t t h a t

The hypothetico-deductive

method

39

non-existential and existential concepts generally occur in pairs, rather than in one-many relations tends to blur the distinction, except with regard to the most abstract of the scientific notions. I shall explain this presently. Having no existential import means that the statements and concepts in question are independent of whether there is anything in reality to which they apply or not. As this proves difficult to grasp for many people, it deserves some further explanation. It is, of course, correct to say that the notion 'phoneme' would not have been established in the theory, if there were nothing in reality to which it could apply. That is, we require from theoretical notions in a theory that purports to be an instrument for description that they are appropriate to descriptions, just as the shape of a hammer is appropriate to hammering in nails. But this is as far as the connection between theoretical concepts or other tools and reality goes. Theoretical concepts and other tools are not only appropriate, but they are also arbitrary, in spite of their being appropriate. The fact that their existence and appearance are motivated by what is expected from them, does not diminish the fact that they could have been different, or not in existence at all, perhaps because other tools could make their function redundant. There is no necessity for a theory to contain the concept of a 'phoneme', nor, if the concept 'phoneme' is used at all, to define it as 'a simultaneous bundle of distinctive features'. This is what is meant if we say that there are concepts which, though appropriate and motivated by their task, are independent of reality, or, to put it in Hjelmslev's words, that have no existential import. Their motivation may be rooted in reality, but the concepts themselves are independent of it. It is these concepts which are theoretical, rather than descriptive concepts. Examples in the natural sciences are such notions as 'force', 'gravity', 'inertia', 'atom', 'cell', etc. The same goes for theoretical statements, such as axioms and definitions. They are arbitrary, but motivated by their being appropriate in view of the aim and scope of the theory. It is quite different with descriptive concepts and statements. Once we have accepted a particular theory, we are forced, at the moment of applying that theory in the description of phenomena, to make certain statements, and to establish certain concepts, within the limits, of course, of a certain freedom as to the degree of exhaustiveness, and of a certain freedom of organizing our statements in particular ways, provided

40

Introduction

that such different modes of organization are mutually equivalent in respect of the descriptive content. A description remains, of course, arbitrary to the extent that it depends on an arbitrary theory, but its further arbitrariness is severely limited by the requirement that every one of its statements must be in agreement with the phenomena as we observe them under the accepted point of view. As observation introduces its own type of uncertainty, there should be further safeguards to guarantee an appropriate measure of objectivity, but this will be discussed later. It is descriptive statements, and not theoretical statements, that are arrived at via hypotheses concerning the phenomena to be described, and it is to these hypotheses, and to them alone, that the Popperian method of testing applies. Apart from descriptive statements, descriptions of phenomena generally also contain certain stipulations for conventions to be adopted solely for the description in question, even though some of those conventions may be used in the same form in other descriptions as well. For a description of English, for instance, we may stipulate that we shall call a 'vowel' a phoneme that can only occur in nuclear position, a 'consonant' a phoneme that can only occur in peripheral positions, and a 'semi-vowel' a phoneme that can occur in either. Such stipulations may take the form of statements that we usually call definitions. But as there is an intrinsic difference between definitions in the theory, and the stipulations in question, the latter being of a taxonomic kind, I prefer to call them 'quasidefinitions' in order to emphasize the important distinction. We may use formally the same stipulation for descriptions of other languages where this is appropriate, but we must remain aware of the fact that such stipulations are not necessarily appropriate to the description of all languages. For many languages, for instance some Slavonic languages, it may well be more appropriate to add another category, e.g. 'semi-consonant', as there is a hierarchy of affinity towards nuclearity involved between say 11/ or /r/ and 'semi-vowels'. That is, in those cases where a 'semivowel' and a 'semi-consonant' are juxtaposed within the same phonological unit, it is always the former that stands in nuclear position, and the other in a peripheral position. Only if there is neither a juxtaposed 'vowel', nor a juxtaposed 'semi-vowel' can 'semi-consonants' occur in nuclear position. Similar stipulations can be made in respect to such terms as 'verb', 'transitive', 'loca-

The hypot hcl ico-dcduct

ive method

41

tive\ to mention only a few. All such terms are to be regarded as language and description specific, and the appropriateness of transposition to other descriptions has to be constantly examined. Moreover, one should always regard this as a transposition qua form of the stipulation alone, as - according to functionalists and other Saussurean type linguists — a language is a self-contained system, and, therefore, no two things ever have exactly the same identity in different languages. The identity of an item in a selfcontained system depends totally on its distinctive function within that system, i.e. on the full set of items to which it is opposed. But even if we leave this out of consideration, a transposition of such definition-like stipulations to the description of any language, i.e. making them into items of theory, as universalists do, can only be made possible at the cost of either rarefying the meaning of the terms in question to an intolerable extent, as I have pointed out earlier, or by turning to some form of conceptualism. The latter I have also criticized earlier. Thus, as I have argued, it is only descriptive statements that are hypothetical in the scientific sense, and it is with regard to them, and them alone, that the Popperian method of empirical testing can be applied. For such testing to be possible at all the hypotheses underlying descriptive statements must have certain properties. In the first place they must be fully interpretable. Their interpretability derives from the fact they do not contain any technical, i.e. non-primitive, terms that are not defined by the theory. It also derives from the fact that the hypothesis in question remains within the precise scope of the theory, as otherwise the hypothesis would be totally arbitrary. If we did not make this stipulation we could launch any hypothesis we pleased, provided it only contained non-technical terms, i.e. there would be an infinite number of non-theory bound alternatives to a set of hypotheses, which in itself would reduce the set of hypotheses to meaninglessness. What value could one possibly attach to statements that have genuine alternatives that contradict them? Secondly - and this goes without saying - the statement must refer, directly or indirectly, to observables. If not, how could it be empirically tested? I must here sound a warning, i.e. that one should not corrupt this requirement by saying that the statement must refer to direct or indirect observables. There is a branch of philosophy, in my opinion in that respect speculative philosophy.

42

Introduction

i.e. phenomenology, in which it is permitted to infer the true nature of a phenomenon by its manifestations. The phenomenon is then regarded as observable via its manifestations. This comes very close to the 'black b o x ' strategy, which I have criticised above. Thirdly, the conditions under which a hypothesis may be falsified must either be clear and known in advance, or they must be laid down by the theory or the philosophy behind the theory. The latter requirement prevents the approach from being immune to refutation. This is a very important requirement and helps to distinguish the hypothetico-deductive method from pseudohypothetico-deductive approaches. One may say that hypotheses in the Popperian sense, which sense we adopt, are in essence descriptive. Does this mean that the hypothetico-deductive method only applies to descriptions, not to theories? It is true that theories in our sense do not contain hypotheses. But a theory, or any statement in a theory, once it is launched, is a real thing in itself. It may not refer to reality, but it is itself there and observable, i.e. it has its own reality. Therefore it is itself subject to the possibility of description, i.e. we may make descriptive statements about a description. Such descriptive statements may therefore have hypothetical status. We may, for instance, hypothesise that the theory or a particular description is self-consistent, or that a statement within the theory or within a description is consistent with other statements. Or we m a y , for instance, hypothesise that theories, descriptions, or statements in descriptions are adequate with regard to their purport. We may even hypothesise that the theory or a description are as simple as they possibly could be in respect of their adequacy. All these are proper hypotheses, as they are interpretable, refer to observables, and the conditions for eventual refutation are known by simply knowing what is meant by self-consistency, adequacy, and simplicity. I shall return to this in Chapter II. S u f f i c e it to say for the moment that this approach recognizes three kinds of descriptive statement and consequently of hypotheses, i.e. statements about linguistic phenomena, statements about the theory, and statements about the description. As the latter two are about statements, not about linguistic phenomena, I shall call the corresponding hypotheses me fa-hypotheses in order to emphasise the distinction. Nevertheless meta-hypotheses are as much hypotheses as the hypotheses about linguistic phenomena.

The hypothetico-deductive

method

43

They only differ from the latter in that they are hypothetical statements about statements, rather than about natural languages. In Chapter V I shall come back to the notions of 'hypothesis' and 'descriptive statement', and develop a simple and practical kind of hypothetico-deductive logic. It will then be seen that the theory is the instrument by which refutation or corroboration of descriptive statements is possible. It is this process of refutation and corroboration which leads to descriptive statements. Descriptive statements themselves do not generally have the form of hypotheses. In earlier publications I have used the terms 'hypothesis' and 'descriptive statement' synonymously, but I see now that this was misleading. The hypothetical character of descriptive statements is in that they are the result of conjecture and refutation (cf. Popper 1965), and that their material adequacy can be challenged. They are not, however, themselves, presented for refutation. They are rather conclusions at the end of a chain of attempted refutations.

Chapter II: Theory and strategy C'est le point de vue qui cree l'objet (Saussure)

1. Preparing the raw material I have argued in the previous chapter that the most successful way to describe languages is by adopting the hypothetico-deductive approach. I have also argjied that this approach, in order to be maximally effective and consistent, presupposes that we draw a sharp distinction between a theory and descriptions under that theory. Description, according to this view, can be defined as the application of a descriptively, and therefore empirically, orientated theory to a selected field of phenomena. In the case of linguistics these phenomena are speech phenomena, but as the present theory is basically a semiotic theory there is no reason not to use it in the description of other semiotic phenomena, including writing conventions of languages, as well. To be descriptively orientated, for a theory, implies that it is free from Procrustean tendencies, which in its turn implies that nothing descriptive is already present in the theory itself. Such a theory is without existential import, whether in the form of linguistic universale or not. Also the hypothetico-deductive approach applies both to the theory and to descriptions themselves, but the hypotheses in question are outside and above the theory and descriptions. For this reason I have called them meta-hypotheses. We have to distinguish between statements in the theory, i.e. theoretical statements, and statements in the description, i.e. descriptive statements. The former make no existential claims, the latter do, at least to the extent that they are arrived at via existential hypotheses. Descriptive statements have further to be distinguished from 'protocols'. The term 'protocol' is derived from the German term Protokoll, which means 'record' or 'report'. Rather than using one of the two most common English trans-

46

Theory and strategy - C'est le point de vue qui tree l'objet (Saussurej

lations I prefer the germanism as a technical term, in order to stress its special status in the present strategy. According to Boch6nsky (1959:105): 'Protokollaussagen sind Aussagen, welche das Vorkommen von Phänomenen feststellen'. That is to say, protocols are direct statements of particular observational facts. A protocol is, for instance, what the nurse writes down or enters in a chart after she has examined a patient's pulse. The phenomenon she observes is actually the beating of the pulse. By using some kind of a priori theory, she translates this into a pulse-rate, and it is this that she enters, together with time, place, individual and situational circumstances, and so on. One sees that already at the protocol stage, part of the theory intervenes in the observation. The part of the theory that does intervene is especially its explicit or implicit relevance criterion. However, at the protocol stage it is not so much that only what is relevant is taken note of, but that everything which is a priori irrelevant under the theory to be employed is omitted. A protocol is, as I have said, a direct statement of observation, but at the same time it presents the phenomena in a useful form, free from unnecessary fringe features and also in as organized a way as possible. Though there is still a hypothetical factor involved, i.e. we may reexamine the phenomena if a particular observation is inconsistent with observations elsewhere — we are justified in doing so because it is a wellknown fact that our senses cannot be fully trusted — a protocol once made is regarded as a true record of the phenomena in question as soon as we use it in order to test descriptive statements. We are forced to do so, as otherwise no description would ever be possible. We may, in fact, say that every protocol carries with it the meta-hypothesis that it constitutes a faithful record of observation. Though we are aware of our observational inadequacies, we have to live with them, and we have implicitly to trust our senses. By making observations from as many angles as possible, and as often as feasible, we may overcome these inadequacies — at least we are, for practical reasons, forced to believe that we may. The protocol stage is a stage between the phenomena themselves and the descriptions, or even between observation and description. For all practical purposes we may, however, regard it as the observational stage itself. There is indeed no observation without, at least implicit, protocolization of a kind, if only protocolization of a purely perceptual kind. It is

Preparing the raw material

47

protocols, not the phenomena directly that scientific descriptions refer to. Phenomena are evasive and fluid. At the moment of observation they have already receded to the past, and are then no longer available phenomena. This is why one has to 'arrest' them as is done by protocolization. This is also why descriptive statements can only refer to protocols, not to phenomena directly. By protocolization, phenomena, which are 'events', become 'facts'. Facts are already to some extent theory-bound, and this is one sense in which Saussure's dictum: 'C'est le point de vue qui cre6 l'objet' has to be understood. But Saussure's statement also refers to the fact that descriptions are theory-bound to an even greater degree, and for Saussure the 'object' was not so much the phenomena, as the picture that emerges from the description, i.e. 'langue' as opposed to 'parole'. I shall not go into this here, especially because Saussure's term 'objet' serves to cover both the meanings of material object described and purported objective of the description (teleological object). Any reference in this work to phenomena or to facts has to be seen in this light, i.e. it has to be understood that protocolization, whether explicit or implicit, is involved, which in its turn implies that some kind of theoretical considerations have been applied. This, of course, introduces some degree of distortion. Rather than this being a drawback, it not only facilitates the description, but it also acts as a safeguard against distraction by an infinite number of irrelevant details. Apart from phenomena being presented in a simplified way we also tend to build up an idealized picture of the phenomena. That is, we ignore those features that are clearly abnormal or pathological, and present the phenomena in a form that is in agreement with their function. Under a functionalist theory this means that we regard the form characterized as being at the 'upper limit of distinctive realization' as the most representative type of utterance. Examples of this will be given when we deal with the upper limit calculus in phonology, and also in syntax when we deal with 'elliptical' realizations. But there is in fact a great multitude of realizational phenomena that fall under the heading of requiring to be idealized. If we were to make a record of live speech, we would find that the majority of utterances are not well-formed, either by being incomplete with reference to what one actually means to say, or by shifting or merging, hesitating or slipping of the tongue, and by a host of other malformations, such as speech-

48

Theory and strategy - C'est le point de vue qui cree l'objet (Saussure)

defects, and so on. It is commonly understood that distortion by idealization is a necessary condition for description being possible at all. This is clearly one of the reasons why, for instance, Chomsky found it necessary to draw a distinction between competence and performance, even though we may not go along with the implications inherent in the way he has presented the distinctions. Basically, however, I am not in disagreement with Chomsky at this point. The distinction between ideal and real is in fact, not very new. For instance, some cultural anthropologists in the 1930s referred to 'ideal culture', as opposed to 'real culture'. Also in the natural sciences one used similar procedures, if only by selecting, for instance in taxonomic zoology or botany, perfect specimens on which to base one's descriptions. I do not think it is necessary to argue this any further, but it should be kept in mind that some degree of distortion in description is not only inevitable, but even desirable. Without this an even more distorted picture would emerge from descriptions, owing to the fact that marginal and abnormal phenomena would be given equal importance to the kind of phenomena that are true to type, or — to use a more formal way of expressing it — that are in agreement with the intension of the class they belong to. This brings us to the distinction between intensions and extensions of classes, an important distinction in any descriptive science. Basically the distinction is that between 'class by definition' and 'class by ostension', or rather 'definition of a class' and 'the inventory of actual members of a class'. The intension of a class is the criterion by which it is a class, whilst the extension is the actual class. But as there can be no actual class — at best a haphazard conglomeration — without a class-forming criterion, it is clear that the intension of a class is logically prior to its extension. We may, however, ask, are there no natural classes, i.e. classes that have a particular intension by virtue of their extension, which in itself is due to natural or inescapable grouping? Let us consider the class of all people wearing spectacles and the class of all people who are linguists. The intensions of those two classes are 'wearing spectacles' and 'being a linguist' respectively. The extension of the sum of these two classes, which constitutes a separate class, is those people who satisfy the intension of either wearing spectacles, or being linguists, or both. The extension of the intersection, also constituting a separate class, is all linguists

Preparing the raw material

49

who wear spectacles. Clearly, the sum does not constitute a natural class, though the intersection might. Are we then to say that the intension of the non-natural class derives from the extensions of the original two classes? That is, the class intension would derive from extensions of classes which are formally independent, as the class in question could have been established without any thought of the original two classes? This would be rather awkward, to say the least. Besides, one could as well have had, say, the larger class first, and then arrive at the two classes by further sub-classification. Or one might have thought of the intersection-class first, even have considered it a natural class, and then by first sub-classification, and then widening of the ranges, have arrived at the original classes. Let us examine what we might call a natural class. It is in the first place a class of objects that have a conspicuous feature in common. It is that feature that constitutes the intension of the class. But even though conspicuous, one might not have noticed that feature, or one might not have regarded it as interesting enough to base one's classification on. That is, one might not have established that class at all, i.e. classification is a wilful, i.e. intentional,* act. This implies that the intension, at least logically, is prior to the extension of even a natural class. Even supposing that there were no intention or intension in our mind before we noticed a certain assembly of features or objects, the mere fact of regarding them as constituting a class implies that an intension has been given to them, perhaps, actually, in retrospect, but still either before, or at the same moment as, the assembly became formally established as a class. In other words, it is the intension that makes the class, even though it is the extension that constitutes it. Yet, the idea of a natural class is an important intuitive notion, especially with regard to protocolization, which in itself is an at least partly intuitive operation. If we were to give a more formal qualification to the notion 'natural class', we might say that it is a set of objects or features the homogeneity of which depends on external, i.e. non-theoretical, considerations. It may still be that the point of view inherent in the theory is a delimiting factor as to membership within the class, but the limits of the class as such depend on external factors. Typical examples of natural * Note that 'intention' and 'intension' are etymologically identical.

50

Theory and strategy - C'est le point de vue qui cree l'objet (Saussure)

classes are sets of all speech-phenomena within a field constituted by what we call a 'natural language', e.g. German, Japanese, or English. Grouping together all utterances of a certain type under the heading of, say, 'German' is based on purely practical considerations of convenience, feasibility and interest. The latter may shift, and so lead to a further limitation of the field, e.g. admitting only phenomena belonging to a certain dialect, or even ideolect, or register, to the set considered. Or it may lead to enlarging the field, and for instance also considering Low German, Swiss German, and other Alemanic fields, though with regard to descriptions this may well come into conflict with criteria of feasibility, unless one allows a greater degree of generalization, which either means less particularization, or more distortion, or both. Generally, the more varied the phenomena within a certain range, the more difficult it will be to be exhaustive, detailed and precise in one's descriptions. But the criterion of interest requires that the range be wide enough to satisfy one's particular aims to an acceptable degree. Let me give two fairly extreme examples. Though theoretically speaking a description of a field of speech-phenomena in which utterances of French and Chinese were indiscriminately classified together might be conceivable, it would not be a feasible undertaking, nor is it likely that any interest be served by such a description, even if practicable. At the other extreme, a description of the speech of one particular person is likely to be much more detailed and precise than the description of speech-phenomena within a large speechcommunity can be, but such a description will usually have a very limited interest for the linguist. It may even give him a false picture of what the speech of the community that particular person belongs to is like, but he may, of course, not be interested in the latter at all. Some inductivists have believed that the best account of a natural language would emerge if the speech of a representative number of speakers were described for each of them separately. One could then, by generalization and statistical averaging arrive at the speech of the community. However, even if such a procedure were feasible, there is no guarantee that the picture arrived at would not strike one as a caricature, rather than as a description of real speech such as one might encounter in such a community. In physical anthropology — a science in which one can resort to measurements, which can be statistically processed —

The relation of theory to description, and of both to the speech-phenomena

51

attempts have been made to arrive at a picture of 'ideal man'. The result was, however, the statue of a being which, were it to be encountered in the street, would immediately be recognized as highly abnormal. What I want to stress here is that the inductivists' ideals are not likely to be of any avail, even at the stage of preparing the raw material for the description.

2. The relation of theory to description, and of both to the speech-phenomena We can picture the relations between theory, descriptions and speech-phenomena in the approach we have adopted as follows: Fields of phenomena:

etc.

etc.

52

Theory and strategy - C'est le point de vue qui cree l'objet (Saussure)

I prefer to speak about 'fields', rather than 'classes' of phenomena, as the infinite number of actual and potential phenomena falling within the limitations of such a field are all members of the class. There is, however, no formal difference between the idea of a field and that of a class. One may envisage classes with an infinite actual or potential number of members, but whenever the potentiality rather than the actual occurrence is the most striking feature, as it is with speech-phenomena, to call such a class a 'field' seems to have certain intuitive advantages. I also make no distinction between the notions 'class' and 'set' in this work. A description of a field of speech-phenomena of course presupposes those phenomena. This is indicated, in the scheme, by an arrow pointing to the particular field. There is, in fact, no objection to calling this stage, i.e. the actual description, the inductive stage, but it is better to call it the empirical stage, as it is clear that descriptions depend at least as much on an a priori theory, without which no description in any scientifically acceptable sense is possible. It is the latter that inductivists deny, even though in practice they too deploy theoretical considerations of a kind in their descriptions. But the 'God's truth' requirement is satisfied in the present approach, at least to the extent that what we are doing is not mere 'hocus-pocus'. The dependence of any description on the theory too is represented in the scheme by an arrow, pointing to the theory. I should remind the reader that the theory is arbitrary, but should be appropriate, and that it has no existential import. This implies that the theory is independent not only of the phenomena, but also of descriptions of the phenomena. The latter are, however, dependent both on the theory and on the phenomena. As to the actual form or nature of the theory with respect to its being appropriate, descriptions serve, of course, as test-cases for the latter requirement to be satisfied. Though the theory as such is independent of descriptions, there is a teleological relation that goes from theory to descriptions. I shall come back to this point when we are discussing the evaluation of theories and descriptions. In the above scheme it is assumed that there is one single linguistic theory under which descriptions of any chosen field of speech-phenomena can be made. Unless this is the case, one has to do with more than one different approach, and perhaps even with different disciplines. Descriptions under different theories

The relation of theory to description, and of both to the speech-phenomena

53

are not inter-comparable, except on the most global and general, perhaps epistemological level. Descriptions made under approaches different from that of the describer are of no use to him in his actual description other than that he can extract further protocols from them, and that they can inspire him and make him aware of points that he may otherwise have overlooked. That is, such descriptions are only indirectly useful to the describer, but in this respect they may be very useful indeed. This is why axiomatic functionalists tend to insist on the necessity of not ignoring any of the work done by linguists of other persuasions. Especially the debt we owe to Bloomfieldians and Firthians, who have been particularly industrious in the description of great varieties of languages, is great. I should like to mention in passing that there is another important reason why we should not ignore other approaches, as transformationalists tend to do. Only by a deep and critical study of the greatest variety of approaches can we understand the how and why of our own epistemology, and avoid mistakes and traps into which those who tend to wear blinkers are likely to fall. Let us now look at a situation in which descriptions are made under t w o different theories, which I shall call theory I and theory II. Let us assume that the fields of speech-phenomena, which I shall label A , Β and C, and their protocolization are the same, which, in actual fact, is not likely to be the case. We then get the following picture:

54

Theory and strategy - C'est le point de vue qui cree l'objei

(Saussure)

Though descriptions A l , B1 and CI can be compared with one another, and may be used as the bases for comparing the fields A, B, and C, none of them can be compared with any of the descriptions A2, B2, and C2, in respect of A l , B l , and CI. The difference between the former and the latter set of descriptions is due to the difference in theory employed. The differences between A l , B l , and CI are, of course due to the difference in fields of phenomena described, and so are the differences between A2, B2, and C2. It is, of course, the aim of a linguist to have just one theory for the description of any language. Because of the facts considered above, it would be of no direct use to describe one language under theory I and another under theory II. But as there is no consensus of opinion as to which theory to choose, and for the great multitude of languages to be described a great multitude of linguists are needed, we shall have to live with a state of affairs in which plurality of approach is the rule, rather than the exception. Perhaps this is even a good thing, as it gives inducement to the description of roughly the same field of phenomena over and over again, but under different theoretical approaches. The only requirement for making this indirectly useful and leading to an enrichment of our knowledge of the fields concerned is to acquire an understanding of the approaches and theories involved. This illustrates again the fact to which I have been pointing earlier, i.e. that the duty of the linguist is not merely to develop a theory or produce descriptions under a particular theory (though this should probably be his primary concern), but also to make a thorough study and take due notice of other linguists' theories and descriptions. Given the facts that theories are arbitrary, a choice between alternative theories can only be based upon considerations of appropriateness. For this, the meta-hypotheses, referred to earlier, come into their own. What we should be looking for is the best possible theory, and, of course, the best possible descriptions under that theory. What is 'the best possible' depends upon our aims, i.e. it is largely a teleological matter. Let us assume that the latter has been settled, and that the theory in question aims at the possibility of having descriptions of any chosen field of speech-phenomena as such from the point of view of efficiency in communication. That is to say that we shall only consider those

The relation of theory to description,

and of both to the speech-phenomena

55

features in the p h e n o m e n a that may c o n t r i b u t e to the fact that c o m m u n i c a t i o n by speech is possible, and that potentially contribute to or whose absence may interfere with the efficiency of c o m m u n i c a t i o n . This means, a m o n g o t h e r things, that we leave at least for t h e m o m e n t , and as long as we employ this particular theory — non-communicative f u n c t i o n s of language out of consideration. According to functionalists, w h o have a d o p t e d this point of view, no feature can be relevant t o c o m m u n i c a t i o n unless it is functionally opposed to something else, or to its absence. That is to say that not only its replacement by something else, or its omission should be possible in fact, b u t also that this should have potential implications for the message t o be conveyed. This consideration I have called 'the functional principle'. It is inherent and overriding in all functionalist t y p e theories. It is, for instance, embodied in Andr6 Martinet's statement that ' f u n c t i o n is the criterion of linguistic reality' (Martinet 1961:5), and I have built it into the first axiom of my theory which states that 'all features in semiotic sets are functional'. In the absence of r e f u t a t i o n , we may regard the meta-hypothesis assuming the appropriateness of this statement t o be c o r r o b o r a t e d , but we have still to examine what the conditions are for regarding a statement in a theory as being appropriate. A good theory is a theory which is totally appropriate. That is to say that it only contains appropriate statements, and that the total b o d y of statements is complete and well-integrated enough to a system that is itself appropriate qua system. The appropriateness of a s t a t e m e n t is n o t a singular m a t t e r but also depends on its f u n c t i o n , usefulness, and integration within the whole of the theory. But let us start with descriptions first. Assuming that a description is in agreement with the theory e m p l o y e d , what are the conditions for regarding it as a good description, in fact as the best possible under the circumstances? In the first place, the statements of the description should be consistent a m o n g themselves. This does not only mean that there should not be statements, the c o n t e n t of which is in direct conflict with that of a n o t h e r statement or statements, but also that it should not be possible to infer conclusions f r o m certain sets of statements which are in conflict with conclusions drawn from o t h e r sets of statements. Consistency is a m a t t e r of the whole of the description. If there are any elements of the description that

56

Theory and strategy - C'est le point de vue qui cree l'objet (Saussure)

are inconsistent within the description, the description as a whole is not self-consistent, and cannot be regarded as a good description, whatever its other merits may seem to be. Of course, the consistency between statements need not affect all the statements of the description, and therefore a description may not be beyond repair if inconsistency is discovered. But it is correct to say that mutually inconsistent statements nullify each other, and, therefore, by leaving dead patches, as it were, inconsistency also affects the cohesion and integratedness of the description. Consequently, we cannot tolerate any inconsistency within a description, and have to remove it, if possible as soon as it is discovered. For inductivist descriptions it is, as I have already said, practically impossible to keep consistency, which is the most basic of all scientific requirements, under control. Unless one has a theory which almost automatically organizes the body of descriptive statements into a well-integrated and structured description, descriptions are likely to be fairly chaotic, and there is no way in which we can be reasonably sure of overall consistency. But in the variety of hypothetico-deductive approach which we might call the theory-based approach, or even the hypothetico-theoretical approach, the fact that the theory is the main source for the organization of descriptive statements does not exhaust the importance of the theory in respect of descriptions. It is essential to the approach that everything in the description has to be justified by the theory behind the description. It is not only the case that all technical terms used in the description must come from the theory, where they have been defined and so provided with a precise meaning, but also that the descriptions should not contain any statements that are not allowed by, or relevant under, or meaningful because of, the theory in question. Not only should descriptive statements not surpass or bypass the scope of the theory, but within that scope they should be confined within the limitations set by the theory. This does not mean that one is not allowed to make interesting ad hoc observations, while making statements about the phenomena, observations which may at times be rather far afield. But the ad hoc character of such observations, and the fact that they do not belong to the more formal part of the description in question, should be made abundantly clear in the presentation. Such observations should clearly be recognized as belonging to a side-track, and they need not, nor

The relation of theory

to description,

and of both to the speech-phenomena

57

in fact could they be, integrated with the rest of the description. Though, strictly speaking, they are not part of the actual description, and t h e r e f o r e meaningless in respect of it, it would be rather inelegant if we allowed them to be intuitively inconsistent with formal statements. If t h e y were inconsistent, this would not affect the overall consistency of the description, as this could only be due to the fact that t h e y were implicitly based on a t y p e of theory that is inconsistent with ours. But in that case one should, of course, refrain f r o m making such observations altogether. If we indulge in t o o m a n y side-tracks of that nature, descriptions may acquire a rather eclectic flavour. One could, in fact, characterize eclecticism as an approach in which t h e main road cannot be distinguished f r o m the side-tracks. But we may distinguish between t w o kinds of eclecticism. The first is the natural child of an inductivist a t t i t u d e , i.e. there is no explicit theory or point of view involved at all, and therefore any observation that seems interesting can be m a d e at any time. The o t h e r t y p e of eclecticism, probably more damaging than t h e previous one, is using different theories for different parts of the description, or even using criteria f r o m that theory which most serves the convenience of that particular m o m e n t , or which leads to the intuitively most interesting results. The latter type of eclecticism is widespread and is practised by those linguists or students of linguistics who claim not to adhere to a particular school, but accept everything that is 'good' in any theory. The understandable source of this a t t i t u d e is the despair of the student who is c o n f r o n t e d with a most confusing picture of a great m u l t i t u d e of divergent approaches. Nevertheless, unless one can pick one's way through these approaches, and select the best one, or — as I have done initiate one's o w n well-integrated approach, no really scientific work or progress will be possible. A n o t h e r source for such an a t t i t u d e may be t h e fact that there are several i n t r o d u c t o r y works f o r linguistics that are highly eclectic, including some that have b e c o m e fairly famous, and the use of which is widespread in linguistics courses in universities. T h o u g h such books have certain advantages as an initiation t o the subject over works that confine themselves t o a particular school, they tend t o condition s t u d e n t s to a non-critical a t t i t u d e in this respect. Hqually i m p o r t a n t as self-consistency for a good description is that it be adequate in respect of its aims. It may seem that

58

Theory and strategy — C'est le point de vue qui cree l'objet

(Saussure)

adequacy in descriptions is even more important than consistency, as a description is only a real description in so far as it is adequate. This may be so, but without consistency, any effort with respect to adequacy will be wasted, as the result will be void of relevance. The requirement of consistency is, therefore, logically prior to that of adequacy, even though consistency without adequacy is equally useless in practice. We can, in descriptions, distinguish between three aspects of adequacy, i.e. adequacy of range or scope, adequacy as to detail, and material adequacy respectively. Adequacy as to detail is flexible. If one purports to have a more global generalizing description one will go into less detail than if one wants greater precision. But once one has accepted a standard in this respect in advance, one should make sure that the statements one does make together add up to a satisfactory and interesting picture. The requirement of material adequacy is, in practice, slightly more complicated. In theory it is very simple, i.e. it means that the description and every single statement in it should 'fit the facts'. In practice, however, it is not always easy to decide which one of two or more conflicting potential statements is the more fitting. Take, for instance, in English, 'He fell under the bridge.'. Should this in first instance be analyzed into 'he', 'fell' and 'under the bridge', or into 'he' and 'fell under the bridge', the latter to be divided into 'fell' and 'under the bridge', or should it, perhaps be first divided into 'he fell' and 'under the bridge', and only then 'he fell' be divided into 'he' and 'fell'. One may even consider the possibility of 'to fall under' being a 'transitive' verb with 'the bridge' as its object. It is very difficult to get a direct clue from the facts as such as to which of these analyses is the most fitting one. Still it is from the facts that the answer must come, provided that all three solutions are equally possible under the theory employed, and there are no reasons to regard one or more of them as being inconsistent with statements elsewhere. But considerations of consistency will also come into the solution, albeit not so much direct matters of consistency between statements, as consistency of treatment, also in respect of showing up differences where there are differences in messages that must be due to organization or structure. This, in itself, is, of course a case of adequacy, but the contrary, i.e. treating as the same that which in all relevant respects is the same is solely a matter of consistency of treatment. To come back

The relation of theory to description,

and of both to the speech-phenomena

59

to o u r example, we know that an utterance of this form exhibits ambiguity, i.e. that there are at least t w o different syntaems that have that type of realization. Under the one interpretation, he was under the bridge, when he fell, and under the o t h e r he was somewhere else, perhaps on the road, or on the bridge, and his falling caused him t o land under the bridge. If there is no reason to regard, say, the t w o forms 'fell' or the t w o forms 'under', as h o m o n y m s , i.e. as different words with different meanings, and assuming that the difference in meaning is denotational, which implies that it comes under the scope of t h e present theory (as we shall see), t h e difference must be due t o a difference in structure, which must be accounted for. I cannot go into details here, but I shall c o m e back to it when we are dealing with syntax. I shall only say that f u r t h e r considerations of a similar kind have led me to analyse the syntagm under t h e second reading as 'he', 'fell' and 'under the bridge', whilst that under the first reading is analyzed into 'he fell' and 'under the bridge' in first instance. These analyses are the ones that satisfy considerations of material adequacy on all levels of description simultaneously, i.e. phonetics, phonology, grammar, as well as semantics. Some of these levels, i.e. phonetics and semantics, are nearer, as it were, to the protocols than others, and it is generally on these levels alone that direct reference t o the p h e n o m e n a can be made. But as descriptions on all levels should be mutually consistent, we can, by referring t o the more protocolizing levels, i.e. by keeping it consistent with statements on those levels, apply the criterion of material adequacy in an indirect manner. A word of warning is, perhaps, not unjustified here. When I have said 'the more protocolizing levels', I d o not mean t o imply that those levels provide us with the protocols, only that at those levels one can m a k e direct use of the protocols. In this respect, phonetics is in a different situation f r o m semantics. T h e former, t h o u g h a descriptive science in its own right, can be used by the linguist as a means of protocolization, i.e. of making the raw material ready for description. Semantics, however, as we shall see, is a descriptive branch of linguistics itself. It needs protocols itself, of which a description is made. It is, perhaps, tempting to regard phonology and grammar, which t w o together I shall later call 'systemology' (see section 4) as constituting a link between 'phonetics' and 'semantics', but this position is untenable in my

60

Theory and strategy - C'est le point de vue qui cree l'objet (Saussure)

opinion. What we could say is that 'phonology, grammar, and semantics' constitute a link between the soundform of sentence utterances, and the messages conveyed by and inherent in those utterances. As we shall see, it is appropriate to regard phonology and grammar on the one hand, i.e. systemology, and semantics on the other, as the two descriptively orientated components among the three components of linguistic theory. The third component is the ontology in earlier publications referred to as 'signtheory' or 'signum-theory'. Though the term 'sign-theory' is actually a misnomer, as long as one realizes that it stands for 'ontology' in this particular type of theory, there is nothing against retaining it. The main thing to keep in mind is that - as we shall see - the notion 'sign' itself does not figure in the signtheory proper, but the notion signum, a more general term covering both 'sign' and 'symbol' does. The latter distinction is not relevant within the ontology, nor even — strictly speaking — within the systemology. A further warning would be in order at this point. There may exist the temptation in applying the criterion of material adequacy, which is done by direct reference to sentence-utterances, to take points into consideration which have no relevance under the theory employed. Take, for instance, 'a book and a pen', in English, or 'good but expensive' in 'a good but expensive pen'. Most informants would claim that in 'good but expensive' the message refers in the first place to the quality 'good', but 'expensive' comes in rather as a secondary consideration. Some, but not all, may claim something similar for 'a book and a pen' as well. But even among those who are not certain about the latter, there are likely to be some who 'feel' — intuitively that is — that whatever the treatment of the 'but' constructions is going to be, the 'and' constructions should be fitted into the same type of pattern. Now, in the first place, 'feeling' is not a good enough criterion for assigning the same type of pattern to two different syntagms. In fact, intuition is never a good enough criterion for decisions. Moreover, in this particular instance there could well have been another 'feeling', equally inadmissible as a decision criterion, almost to the contrary of the original one with r.spect to both the 'and' and the 'but' construction, that such items as 'and' and 'but' function as relators between two terms, and that, therefore, the analysis should not be 'a book// and a pen', and 'good// but

The relation of theory to description, and of both to the speech-phenomena

61

expensive', but rather 'a book// and// a pen', and 'good// but// expensive' respectively. And as both hypotheses are equally plausible to those who try to remain objective, one may as well admit as another possibility 'a book and// a pen', and 'good but// expensive' respectively. But there is a difference in quality between the first two of these hypotheses, i.e. the first one of the two can be demonstrated to be of a non-denotational nature. That is, one can ask oneself if there could be a denotable, i.e. a real or imaginary item to which an utterance of 'a book and a pen', or of 'good but expensive' could refer, but not an utterance of 'a pen and a book', or of 'expensive but good'. If the answer is that as items there could not be, even though there is in most or all of the utterances a difference in shade of meaning which is a matter of connotation rather than denotation, we may not use this as confirmation of 'material adequacy', as connotation as such is not within the scope of the theory, except in a negative way as the complement of denotation. Even if there were merely doubt as to whether the difference in meaning is of a denotative or of a connotative kind, it would already be imprudent to use it as a criterion. Of course, the fact that there is a contrary argument, i.e. the 'relator' one, confirms the invalidity of the former in even stronger terms. This argument is, in fact, of a denotative kind, and it is, therefore, admissible as a criterion of material adequacy. But as we shall see when we deal with this problem in chapter VI, it too is not a valid criterion for establishing that particular pattern. The hypothesis for the pattern in question, i.e. 'a book// and// a pen', and 'good// but// expensive' will in fact be refuted on grounds of inconsistency. This points to the fact that the criterion of material adequacy can only be applied in a negative way, i.e. the existence of material adequacy does not confirm the correctness of a pattern, but the absence of material adequacy may refute the hypothesis, by refuting the meta-hypothesis of its adequacy. This is in complete agreement with the hypothetico-deductive method, which is essentially a negative procedure. Thus there are at least three things we ought to guard against when applying the criterion of material adequacy, i.e. (a) confusing intuitions about the facts with the actual facts, (b) using aspects of the facts that may be outside the scope of the theory, and (c) violating the essence of the hypothetico-deductive ap-

62

Theory and strategy - C'est le point de vue qui cree l'objet (Saussurej

proach by seeking confirmation rather than attempting refutation. I shall come back to these examples, and the solution to some of the problems, in chapter VII. I shall also exemplify the application of the criterion of material adequacy in chapter V, especially with regard to agreement with phonetic observations and statements. Thus we require of descriptions that they be not only justified by a theory, but also completely self-consistent, i.e. free of contradictions within the total body of statements, adequate in scope and in detail, and in agreement with the facts. The latter is a matter of the whole of the description. That is to say that in those cases where material adequacy cannot be directly assessed we may indirectly assess it by examining the consistency of a statement with another statement of which the material adequacy can be assessed by direct reference to the phenomena (or rather, of course, protocols of the phenomena, but it is not necessary to point this out all the time). But apart from a description being self-consistent and adequate, we also require it to be as simple as possible, as otherwise its usefulness may be diminished for practical reasons. As simplicity is largely a practical and a relative, or even subjective matter, it seems unprofitable to look for hard and fast rules to measure simplicity by. Such rules can probably not be given anyway, as simplicity, like 'elegance', which is connected with it, is in itself probably too complicated to formalize. Not only that, but there are psychological factors involved which are not open to observation, and therefore any attempt to give a precise specification of simplicity as such, would remain speculative. This does not mean that certain guidelines could not be given, e.g. that there should be no redundancy in the statements of the description, but it seems to me preferable to say no more than that a description should be reasonably simple. These are, then, the requirements for being a good description, i.e. a description that maximally fulfills its purport. The criterion of simplicity has been criticized (Suleiman 1983:461) as not providing a metahypothesis because of its elasticity. However, as long as there are cases — and these there are in abundancy, in which simplicity can be assessed and contrasted with gross complexity, even if there is some elasticity and therefore indecision when there is no conspicuous contrast, it can be validly used as a criterion. The critic is rather inconsistent anyway when he admits that 'it is

The relation of theory to description,

and of both to the speech-phenomena

63

nevertheless a very important evaluation-criterion'. How could it be if it were not a meta-hypothesis? (Suleiman 1983:509 (Fn 21)). We must now consider what will count as a good theory. As the purport of a theory differs from that of a description, the requirements for a good theory will differ from those for a good description. However, this will only affect the requirement of adequacy. We require of theories too that they are self-consistent, adequate, and reasonably simple. The principles of consistency and simplicity, however, are virtually universal. Therefore we do not have to stipulate them for theories and descriptions separately, except that, when there is a choice between sacrificing some simplicity (but this should always be within reason) in the theory in order to simplify descriptions or vice versa, it is good strategy to opt for the former. This is because, as we have seen, there is a one-many relation between theory and descriptions, and it is better to sacrifice simplicity only once, rather than doing this in every single description under that theory. This is all I need to say about simplicity, and also about consistency for the moment, though I should like to point out that self-consistency for a theory is even more important than it possibly could be for a description. This may seem paradoxical, but this too is due to the one-many relation (in fact it is a one-infinite relation) between theory and description. Inconsistency in a theory has potentially infinite consequences for descriptions under that theory, whereas inconsistency of descriptions — provided that its source is not in the theory itself — does not touch the theory, except that such descriptions cannot, of course, be of any use in the evaluation of that theory. But the latter only confirms what I have just said: It does not affect the theory or the principles it stands for. As to the requirement of adequacy for a theory, this is simply a requirement that it be potentially instrumental in the production of good — i.e. self-consistent, in all respects adequate, and reasonably simple — linguistic descriptions. Indeed the requirement of appropriateness, stressed by Hjelmslev, for linguistic theories should be taken as implying that all the conditions we have set out for a theory and for descriptions under that theory can be met. This in its turn implies that the point of view inherent in the theory in question is a feasible one, but as this is a point that seems in danger of being overlooked — I have discussed this with respect to both transformational linguistics and stratifica-

64

Theory and strategy - C'est le point de vue qui cree l'objet (Saussurej

tional linguistics - I prefer to mention it separately. We can, then, set up the following scheme, showing the relations between all these requirements:

The solid arrows, from tail to head, are to be read: 'implies', the broken arrow as: 'implies potentially'. The scheme indicates that the feasibility of the point of view depends itself on the possibility of satisfying, under such a point of view, all the requirements. I realize that the previous few pages may have caused difficulties in comprehension for those readers who have not been previously exposed to the thoughts that lie behind the axiomatic functionalist approach, and that the above discussion is almost too involved to be included in a work that purports to be a textbook. The views

The axiomatic approach

65

expressed here are, however, so fundamental to the whole of the approach that I could not afford to omit, or even to simplify them. But in order to follow the rest of the discussions, it is sufficient merely to grasp that what is laid down in the above scheme itself, i.e. to understand what is meant by the appropriateness of theories. It is also essential that the reader understand the separation of theory from descriptions, and its implications for the deployment of the hypothetico-deductive method, especially in respect of a distinction between hypotheses in the description and meta-hypotheses about both theory and descriptions. He should be aware of the fact that there is no conflict between a theory being thoroughly motivated and yet being arbitrary. The motivations for both form and content of a theory derive from the requirement that it be appropriate, but as, at least in principle, different theories could be equally appropriate, this does not diminish the fact that theories are arbitrary.

3. The axiomatic approach As the name suggests, the theory of axiomatic functionalism is both functionalist and axiomatically based. As I have indicated already, the main advantage of and reasons for having an axiomatically based theory are that the ontogenesis and mutual relations of all concepts employed in the approach can be traced, and therefore these concepts are fully explicit and understood. This provides, in addition, the advantages of being able to keep overall consistency under control and of preventing descriptions from becoming incoherent. The separation of theory from description — the former being an instrument in productions of the latter — leads to a comparability between descriptions, by imposing a certain uniformity of organization on descriptions under that theory. A deductive system is of necessity an axiomatic system if one wishes to keep it free from circularity. It is clear that unless in a system of statements there are statements that are not inferred from other statements, i.e. primitive statements, the system is circular. Such primitive statements are either axioms, or they are definitions. The difference between axioms and definitions is that the latter provide what some logicians call the

66

Theory and strategy - C'est le point de vue qui cree l'objet (Saussure)

'semantics' of the actual axiomatic system, which is the set of axioms and the potential inferences from them. There is also a distinction qua form, i.e. definitions have the form of an equation, whereas axioms generally do not. An axiomatic system that includes a set of definitions is sometimes called an interpreted axiomatic system, or a material axiomatization, as opposed to an axiomatic system pure and simple, i.e. left undefined, which is sometimes referred to as a formal axiomatic system. An axiomatic system per se does not constitute a theory, but remains a sort of algebra, i.e. a pure system of calculus, unless it is interpreted. As we are dealing here with a theory, not a mere algebra, I shall mean by 'axiomatic', 'interpreted axiomatic', unless it is made clear that I refer to axiomatic systems in general, or to purely formal systems in particular. The theory of axiomatic functionalism, being a theory, is, of course, an interpreted system. It should be noted that under this kind of view, say, Euclidean geometry constitutes a theory, unlike ordinary algebra, which is a mere system of calculus. Indeed, Euclidean geometry is very similar in type to axiomatic functionalism. One should therefore not confuse a mere axiomatic system with an axiomatic theory, albeit that the latter includes the former. One should also not confuse an axiomatically based theory like the present one with a mere axiomatization of an already existing theory. By axiomatization in the latter case is generally meant a formalization of a more informal and therefore less explicit theory by means of putting certain tenets in the form of explicitly stated axioms. This is what, for instance, Batög (Batög 1967) has attempted to do in respect of Harris's (Harris 1951) theory. If such an axiomatization is immediately successful in respect of consistency and other formal requirements, without there being any necessity of having to make important changes to the original theory, we may say that the theory was implicitly axiomatic, but that this has now been made explicit. But only if the axiomatic basis of such a theory is used in its further development, would I call it an axiomatic, rather than an axiomatized theory. Axiomatization of a theory may, of course, lead to the development of an axiomatic theory proper. Indeed, the present theory had its origin in an attempt to formalize two of Martinet's tenets, i.e. that of function as opposition (the 'functional principle'), and the 'double articulation of language', as well as an attempt to

The axiomatic approach

67

formalize Saussure's theory of the linguistic sign. These attempts date from between 1958 and 1960. Indeed the ideas first published in 1965 in my article 'Some operations with sets in language' (Mulder 1965) were already fully developed by 1960, together with a formalization of the notion 'linguistic sign' and a classificatory scheme for semiotic systems, incorporating a specification of 'human language', the latter two first presented in my Oxford D.Phil, thesis in 1966, and published in 1968 under the title 'Sets and relations in phonology. An axiomatic approach to the description of speech' (Mulder 1968). An axiomatic system in the sense we have adopted for that term in the present theory, then, consists of a set of axioms, and a set of definitions giving an interpretation to those axioms. In a workable theory, among the definitions there will be a fair number that go well beyond merely interpreting the axioms. They are the definitions that introduce notions and models, and technical terms for those notions and models to be used in applications of the theory. Such definitions provide, together with some form of the hypothetico-deductive method, the methodology to be employed. It is debatable whether the theory as such would be affected if definitions of this kind were either omitted, or replaced by other definitions, i.e. whether the methodology as such belongs to the theory itself. It is, perhaps, safe to say that there is some latitude in setting up the actual procedural part of the methodology that belongs to a particular theory. However, the scope of the methodology, in extent as well as in detail, is dictated by the scope of the theory. Moreover, without a methodology being provided, either explicitly or implicitly, the conditions under which the hypothesis of adequacy of the theory may be refuted or corroborated remain unknown, if only because the theory lacks potential applicability. It is clear that it is necessary, for a useful theory, that one knows how to apply it. Therefore, unless this is intrinsically clear from the axioms themselves and their interpretations, i.e. unless a methodology is already implied, a methodology has to be provided. I believe that for our purpose, i.e. that of being empirically orientated, we can ignore theories where it remains uncertain how they are to be applied. Therefore we must consider a methodology to constitute an integral part of the theory. I have already said that the scope of this methodology is without latitude, but we may recognize that certain

68

Theory and strategy — C'est le point de vue qui cree l'objet (Saussure)

formal deviations within that scope may not affect the theory as such, provided that the methodologies in question remain equivalent as to their empirical consequences. As long as different methodologies remain mere 'notational variants' (to borrow transformationalist terminology) the otherwise equivalent theories having these differences in methodology need not be regarded as different theories. Any difference between axiomatically based theories is then a matter of either different sets of axioms, or difference in the interpretation of those axioms, or of non-equivalence between methodologies, or a combination of these. In fact, methodologies might well be regarded as part of the interpretation of the axioms, as, to say the least, they provide a further clarification. As I have said earlier, a theory is an essentially arbitrary, albeit appropriate, construct, and this holds for the axioms, their interpretation, and the methodology, except insofar as the latter is necessitated and dictated by the scope of the axioms as interpreted. We may then ask: is it reasonable to expect others to accept a particular theory or parts of it? The answer to this must be that it is not. It may be reasonable for someone to accept a theory in whole or in part, but there cannot be anything unreasonable in refusing acceptance of any particular theory. Denying this would amount to denying the arbitrary nature of theories. On the other hand, it might be judged unreasonable to accept a particular theory if that theory can be demonstrated to be itself an unreasonable theory, i.e. a theory that is inconsistent, inadequate, or so complicated that its aims could not possibly be fulfilled in practice. Another reason could be that it is clearly based on weak epistemological foundations, e.g. that it is eclectic, or speculative, or instrumentalist, etc. Yet another reason for claiming that it is unreasonable to accept, or at least to take seriously, a particular theory might be that the theory in question can be shown to be trivial. It is, of course, quite possible — and judging from the general history of theory-formation even likely that one day it will become unreasonable to be an axiomatic functionalist, just as it is at present unreasonable to be a member of the 'Flat Earth Society'. All we can ask of a theory in this respect is that it should not be inherently unreasonable. But no theory can be said to be inherently reasonable. It is typically adherents of unreasonable theories that claim their theories to be inherently reasonable.

The axiomatic approach

69

There is, then, an important element of faith in the acceptance of any theory. This puts the constructor of a theory under certain obligations, as in the end he will be inviting others to accept his tenets. These tenets, non-reasonable (though not unreasonable) as they are, must, on these very grounds, be as few in number as is possible in proportion to the scope of the theory he wants to propound. That is, the axioms of a theory should be limited in number, and yet powerful enough for the richness of their consequences. Any addition to the number of axioms proposed (together with their interpretation, of course) introduces an added element of non-reasonableness. It also introduces an added element of uncertainty, as one cannot be sure that it is not precisely that axiom which may ultimately lead to a refutation of the meta-hypotheses of consistency, adequacy, and simplicity, and, therefore, appropriateness of the theory. We have to live with this kind of uncertainty in science, but we should keep it within limits. There is yet another constraint that has to be imposed upon the establishment of axioms within a theory. That is, that neither they, nor any part of them should be interderivable. This is not only because this would introduce an element of redundancy, but also because it would refute the assumption that the proposition in fact is an axiom. Axioms are wholly primitive propositions, and as such they are opposed to theorems, which are derived propositions or statements. Though it is, theoretically speaking, possible to have a hybrid type of proposition or statement, i.e. part axiomatic and part theorematic, in the axiomatic system I have envisaged there is no room for these. Having them would upset the neat syllogistic nature of deductions, and it might introduce some lack of clarity and explicitness. Though I cannot think of any objective reasons why this kind of hybrid should be ruled out of order, I subjectively believe that it does not have a place in a proper axiomatic system. It is of course not necessary to have them in any system, as the axiomatic parts of such hybrid statements can always be incorporated into axioms, or given separate axiomatic status. I have used the term 'postulates' for any primitive statement, i.e. a statement that is not derivable in a deductive manner. This term, therefore, includes 'axioms' as well as 'definitions'. It should not be confused with other conventional uses of this term by

70

Theory and strategy — C'est le point de vue qui eree l'objet

(Saussure)

philosophers and linguists. In this approach it is merely a coverterm of all the statements that together constitute the theory proper. The term does not include 'theorems' i.e. deductively derived statements. Such statements are already inherent or implied in the postulates, and do not, therefore, constitute a separate element of the theory. Although theorems may be stated in the development of a theory, none of them constitutes a necessary, though they may constitute a useful, element in the presentation. As the possible set of theorems is vast, and perhaps even infinite, only a small subsection of the set of potential theorems can be stated in any presentation anyway I shall not go here into comparisons with Bloomfield's wellknown set of postulates. Our use of the term should not be confused with Bloomfield's. It should, of course, also be clear that axioms have nothing to do with 'truth', let alone 'self-evident truth'. It is only inductivists who can believe in this. From our point of view this is sheer nonsense.

4. The three legs of the theory A theory that is concerned with the description of speechphenomena and is based upon the foundations outlined in the foregoing divides itself quite naturally into three mutually autonomous, though not completely independent, parts. By 'the theory', unless otherwise indicated, I mean, of course, the theory of axiomatic functionalism. The theory, then, can in fact be compared to a stool with three legs. The seat of the stool can be seen as being the overriding and all-pervading point of view of the theory, i.e. the functional principle. The functional principle is logically prior to anything else in the theory, as without the seat there would not be legs either, just three meaningless sticks of wood. Yet, as we shall see, the comparison is not without flaws, as one of the legs is logically prior to the two others, and the legs are, therefore not of equal standing. But let us now leave the analogy and consider why there should be these three sub-sections of the theory.

The three legs of the theory

71

Descriptive linguistics has as its primary activity the study of what one might call the internal deployment of linguistic objects. By linguistic objects' I mean any entity-like item we may establish in the description of a particular language, e.g. distinctive features, phonemes, self-contained groupings of phonemes, monemes or morphemes, pleremes or words, syntagms, clauses, sentences, and the like. The reason for this is possibly a historical one, i.e. this is how descriptive linguistics has developed. Even transformational grammar has retained this character to a certain extent. There does not seem to be any logical necessity why other descriptive aspects of the speech-phenomena should not have been adopted as the focal interest, but as we linguists are ourselves products of history and conditioning as to views of science in general and linguistics in particular, it is even difficult to indulge in fantasies of possible other ways of describing languages. That is, descriptive linguistics is, for perhaps historical reasons, essentially structuralist, i.e. a study of items and their arrangements. When I talk about the internal deployment of linguistic objects, I mean the use of these objects in the system as such, not directly with regard to the way in which they fulfil their communicative purport. That is, I mean the way they may be combined with linguistic objects of the same nature, perhaps forming other objects of greater complexity, or the converse of this, i.e. the analysis of objects into their components. I also mean the study of structural relations between objects and everything else that can be regarded as purely features of internal deployment. Though this constitutes an autonomous field of study, there is an indirect relation with another possible field of study, i.e. that of the external deployment of items in actual communication. The study of the internal deployment, covering what is traditionally understood by the terms 'phonology' and 'grammar', which two taken together I shall call 'systemology', should be such as to account for the fact that successful external deployment is at all possible. For functionalists this means in the first place adherence at all points to the functional principle. According to functionalists nothing can be significant in respect of communication unless it presents an element of choice, i.e. unless its presence and absence can be manipulated. Ultimately, having something else, or nothing at all, should affect messages conveyed in a fully conventional manner, otherwise the differences in question are not significant with a view to communication,

72

Theory and strategy - C'est le point de vue qui cree l'objet (Saussure)

and the different objects in question (not being functionally different) are said to be merely either contextual or free variants of one and the same thing. There are some complications, since what may be a variant at one stage of the analysis may correspond to a functional item at another, but I shall deal with this in subsequent chapters. I cannot deal with this here anyway, as it presupposes reference to yet another autonomous sub-section of the theory, i.e. the ontology, also called signum-theory or, less aptly, sign-theory. Though in the sub-section I have called 'systemology', no direct reference is made to the actual meaning of the established items in question, all the items established and described must have a particular differential distinctive function, as otherwise they cannot play any role in the conveyance of messages. That is, in that case they cannot play any role in the form or function of items that are to be described in respect of their external deployment. By external deployment, as I have already indicated, I mean deployment in actual communication, i.e. with regard to the conveyance of messages. The sub-theory that deals with the latter is called 'semantics'. The term 'semantics' as a sub-section of the theory has to be understood in this sense. I have to say this, as in other theories, and in most approaches to philosophy, 'semantics' refers sometimes to a much wider area and sometimes to a quite different area, i.e. the world of concepts as such, i.e. independent of their expression in specific languages. The term 'semantics' in our sense as a sub-section of the theory is an instrument in the description of the meaning-communicative aspect of any particular field of speech-phenomena, e.g. a natural language. A semantic description in that sense is language-specific, and no reference is made to a largely speculative world of concepts. At the basis of this kind of semantic theory lies what one might call the 'theory of indices', developed by Hervey and myself (Mulder and Hervey 1971, 1972). It creates the link of the 'theory of denotation', mainly Hervey's creation (Hervey 1970, 1979), with the ontology or signum-theory. Though systemology is traditionally the most central area of linguistic study, and has remained in structuralist approaches for a long time the only area, it is clear that linguistic approaches that set up systemologies from the point of view of communicative relevance cannot for ever remain satisfied with this state of affairs. Sooner or later they will feel the need of dealing

The three legs of the theory

73

with actual communication as well, as otherwise their systemologies will be found to remain rather sterile. It is thus that since the 1950s the study of semantics has become more and more popular. Most conspicuous were the efforts of transformationalists, first Chomsky's associates such as Katz, Postal and Fodor, to mention only a few, and since the early and middle 1960s the proponents of Case Grammar, and the generative semanticists. All these, in accordance with the universalist philosophy underlying the Chomskyan approach, have turned to a conceptualist type of semantics, in some cases intermingled with some form of analysis into components, as is done in phonology, and these approaches do not, therefore, constitute semantics in our sense, which is the establishment of the denotational range of particular signs belonging to particular languages (or other semiotic systems). Our approach is a denotational one and is fully integrated with and an essential part of the whole of the theory. I have above mentioned the efforts of the transformationalists. It is, perhaps, mainly to their credit that the study of meaning has come very much into focus in recent times. Also among nontransformationalists there have been many that have set up semantic theories of one kind or another, but I do not know of any approach other than axiomatic functionalism (with the exception of generative conceptualist approaches) that is fully integrated with a systemology, i.e. phonology and grammar. In all other cases I have found either eclecticist tendencies, or total separation. Space prevents me from going into details, which are not all that relevant to this discussion anyway. At this point I should only like to say that the idea of the denotation of a linguistic sign derives from the relation of correspondence between that sign and (in so far as this is wholly fixed and conventional) a particular class of denotable objects (in the widest possible sense). The class to which the sign corresponds consists of single members. The signs are language-specific, i.e. they are entities of one and the same semiotic system or natural language. To take examples from English, the sign 'table' corresponds to a class of single instances of table-ness in the real or imaginably real world, the sign 'tables' to a class of plurality of tables (i.e. groups of tables as single members of such a class), the sign 'and' corresponds to a class of relations of 'coordination' between objects (not between 'signs', but between objects denoted

74

Theory and strategy - C'est le point de vue qui cree l'objet (Saussure)

by signs), and so on. A sign is a signum with a wholly fixed conventional information-value. According to Hervey's axiom, 'Signa may be realized an unlimited number of times (in actual communication), each resulting utterance denoting a denotatum which may belong to a potentially infinite denotation class' (Mulder and Hervey 1980:203). The term signa can be read here as 'signs'. Hervey's axiom has a wider scope than what we are concerned with at present, recognizing the fact that, though signa that are not 'signs', even including nonce-words, and that therefore do not possess a wholly fixed conventional correspondence relation to a particular class of objects, may nevertheless occasionally denote objects. The act of denoting by an utterance (qua token) has to be distinguished from 'denotation', which is the correspondence relation between a sign (qua type) and the class of objects of which the denotable denoted by an utterance of that sign is a member. Unfortunately Hervey has used the term 'reference' for the meaning-bearing aspect of utterances, not for their meanings, otherwise we could call the act of denoting by an utterance 'reference', which would be in accordance with at least some conventional usages of that term. The term 'utterance' is defined by Hervey as 'member of a signum (as a class) such that it is a model for a single realization (in actual communication) of that signum' (Mulder and Hervey 1980:203). This definition implies that an utterance is a model for a (protocolized) event, i.e. that it refers to something that occurs only once, being wholly determined as to time and place. But this is the very reason why it is right at the base of any attempt to deal successfully with the external deployment of linguistics objects, i.e. with the actual use of linguistic items in actual speech-events. The establishment of the notion 'utterance' by Hervey, and its precise definition, constituted an important breakthrough in axiomatic functionalist theory-formation, and has opened the way for having a systemology and a semantics as part of one and the same integrated theory. We have now come to the third leg of the theory, i.e. the ontology. Every science implicitly or explicitly needs an ontology if statements about the 'objects' it deals with are to have any really meaningful content. If one does not know the precise nature and relative status of those objects, it is clear that no statement one makes about them can contribute in any significant way to one's knowledge of those objects. Even though statements may

The three legs of the theory

75

still give an indication as to the deployment or function of those objects, there is, if one does not understand the nature of the object, something lacking. Yet, in a science where one has objects of only one ontological status, e.g. physical objects, or living organisms, etc., one can leave it to philosophers to ponder about questions of the essence of being or existence. In linguistics, however, we are dealing with two essentially different categories, i.e. that of pure forms, and that of meaning-bearing entities. Some of those forms may potentially be attached to meaning, but for several reasons it is imperative also within the latter entities to be able to distinguish between the abstract form of an entity and the entity in toto. Otherwise one could not, for instance, consistently deal with such problems as homonymy, nor could one treat forms not attached to meaning and those attached to meaning as the same forms, in spite of the fact that they have the same types of property. A typical example is that we would be prevented from consistently regarding the form of, say, genitive V in English 'father's' as being phonemically the same as the final 's' in English 'as'. This would affect the adequacy of the theory concerned. But the mere fact that in the above we would have been forced to distinguish between potentially meaningbearing and not potentially meaning-bearing implies that even here there is some crude type of ontology, albeit a hopelessly inadequate one. If there were no ontology at all, the theory in question would be even cruder, and lead to the most weird absurdities, not just inadequacies. In that case there could not consistently be any a priori distinction between the two types of entity. The absurdity this has led to in the majority of approaches to linguistics is stating, for instance, that a sentence consists of one or more clauses, that these in their turn consist of one or more word-groups; word-groups consist of one or more words, and that words consist of one or more monemes or morphemes. There are already some inadequacies here, in particular where the sentential level (sentences and clauses) is treated as part of the same hierarchy. The other inadequacies can be resolved and are usually resolved by amusing acrobatics such as rankshifting, back-looping, level-skipping, and the like. But none of these are necessarily flaws of an ontological nature. More serious are the problems raised if one then continues by saying that morphemes — or 'monemes' in the case of functionalists —

76

Theory and strategy - C'est le point de vue qui cree I'objet (SaussureJ

in their turn consist of phonemes. One is tempted to compare this with claiming, say, for a living organism that there is no more involved than the ultimate components of the matter of which it consists, except that it is perhaps still debatable whether or not 'life' is an inherent property of matter, i.e. given that we arrive at the proper combination of matter and the right conditions for fusing it, 'life' will come into being. But in the case of a cluster of phonemes being a morpheme, the absurdity should be abundantly clear. In the first place, it is evident that exactly the same clusters of phonemes, say /kats/ may correspond to different entities at another level, say 'cats' (the plural of 'cat'), and 'cat's' (as in 'the cat's tail'). Or even the same phoneme-cluster, e.g. /and/ may correspond t o a meaningful 'and', as well as to a meaningless /and/ as in 'hand'. For these reasons alone it should already be considered absurd to say that a morpheme or moneme consists of phonemes, i.e. that a morpheme or moneme is one or more phonemes. But if we turn this upside down, the absurdity involved becomes even clearer. If the word 'hand' (consisting of one morpheme or moneme), which is an entity that includes in its identity the fact that it corresponds to a particular class of denotables were to consist of the phonemes /h/, /a/, /n/, and /d/, every single one of them, including /a/, /n/, and /d/, should bear some kind of correspondence to some aspect of the class of denotables. But the phonemes of 'and' would, in that case each correspond in some way to an aspect of another class of denotables, i.e. the class to which 'and' corresponds. The phoneme inventory should, therefore, include more than one /a/, more than one /n/, and so on. As similar considerations apply throughout the language, one can imagine the size and, indeed, unfeasibility of inventories that would have to be established if one decided to be consistent. It is absurdities of this kind that have led some, especially Bloomfieldian and transformational, linguists to establish — under various names — the notion 'morpho-phoneme', the phlogiston of linguistics. That is, they have introduced some ontological distinctions a posteriori. But the problem still remains that the concept of a morpho-phoneme is wholly instrumentalist, as I have pointed out in Chapter I. The second problem, that of the unwieldiness of the inventories is in practice solved by establishing morpho-phonemes only in those cases where the necessity — which itself arises from an original flaw in the theory —

The three legs of the theory

77

is most immediately apparent. For instance, in those cases where a phoneme under certain grammatically determined conditions alternates with another phoneme or group of phonemes, e.g.: j'mnaitj ~ /iunzYi/ ('unite, unity), the two items concerned are regarded as one and the same morpho-phoneme, whereas the items /ai/ and /i/ elsewhere are not so regarded. One sees that this shifts the position somewhat in that phoneme and morpho-phoneme are not necessarily standing in one-many relations, but potentially in many-many relations. It does not, however, remove the absurdity, i.e. that there are units below the morpheme or moneme (minimum meaningful entity) level that bear some relation of correspondence to classes of denotables. If one does not have a consistent and powerful ontology right from the start, if only of an implicit nature, the above kind of inadequacies and absurdities cannot satisfactorily be avoided. We shall see that the ontology of axiomatic functionalism, i.e. the signum-theory, not only keeps the systemology free from absurdities of this type, but it also partly predetermines the stratification of the system, as well as a distinction between realizational and structural stages of the analysis. With respect to semantics, owing to the fact that that sub-section is only concerned with the meaning-bearing entities, the ontology provides the necessary understanding of the exact nature of those entities, as well as in what way they differ from, but are via their formal aspect related to, mere abstract forms. This formal connection in itself further elucidates the nature of the meaning-bearing entities. Therefore, also with regard to semantics, having an adequate ontology is a prerequisite for understanding what kind of entities we are dealing with. Yet, the three sub-theories, i.e. signum-theory (ontology), systemology, and semantics, are autonomous in the sense that certain distinctions relevant to the one sub-theory are not relevant to the other, and that each of the sub-theories creates its own proper distinctions. Though the distinctions evolved in the ontology are relevant to both of the other sections, the additional distinctions arrived at in systemology or in semantics are not relevant to the ontology, and nor are any of the distinctions developed in the systemology relevant to semantics and vice versa. It is irrelevant for the ontology whether a particular item is a moneme, a plereme (roughly word, see below Chapter VI, Section 1), a syntagm, a clause, or a sentence, and nor is this directly

78

Theory and strategy - C'est le point de vue qui cree l'objet (Saussure)

relevant to semantics. It is also not relevant to the ontology whether a particular item is a sign rather than a symbol (a distinction developed in the theory of indices, which belongs to semantics) and nor is such a distinction relevant to the systemology. But the ontological distinction between phonological form and signum, and that between the realizational levels and the other levels, developed in the ontology, is relevant to both systemology and semantics. In systemology it even has repercussions as to the internal organization, as I have already said. There is, therefore, in spite of all three sections being autonomous, a stronger connection between ontology and systemology, and between the former and semantics, than there is between systemology and semantics. The latter two are only connected, and integrated within the theory, via the ontology. If we symbolize the stronger connection by a solid line and the weaker by a dotted one, we can symbolize the theory with regard to its three components as follows:

One could in fact have a well-formed theory that only consisted of an ontology and a systemology, or of an ontology and a semantics, but such a theory would leave something to be desired as to completeness and as to the satisfaction of the curiosity of the majority of linguists. Moreover, it is far more economical and elegant to have just one fairly simply ontology that constitutes a basis for both a systemology and a semantics, i.e. having one and the same integrated theory covering both descriptively orientated areas, than having tw.o largely unrelated theories, one dealing with the internal, and the other dealing with the external deployment of linguistic objects.

The three legs of the theory

79

There is one further consideration which it is important to make here. That is, one should not confuse the notion sub-theory with that of sub-discipline. A discipline, or a sub-discipline is an area of dealing with problems, largely delimited on the basis of practical considerations of interest and feasibility. A discipline, as an area, may, or may not, coincide with an area covered by a sub-theory, or part of it. Of course, parts of a sub-theory, if they are self-contained, can be regarded as sub-theories themselves, e.g. the theories of phonology and grammar are sub-theories of the sub-theory of systemology, and the theory of indices is a sub-theory of semantics, which is itself one of the primary subtheories of the linguistic theory. Phonology, in axiomatic functionalism, as we shall see, itself comprises three sub-sections, i.e. phonematics, phonotactics, and para-phonotactics (roughly 'prosody'). Grammar comprises plerematics or morphology, plerotactics or syntax, and the sentential level. Thus far we have been referring to the theory, but descriptions too may, for practical reasons, be divided into particular sections, each of which may be treated separately. Such divisions, being largely based on similar practical considerations as divisions into the sub-disciplines themselves, more often than not follow the latter type of division, and vice versa. In order to avoid a proliferation of terminology, largely the same or similar terms as are used for sub-theories are used for sub-disciplines and sub-descriptions as well; but the terms may not cover the same area in all cases. We may, for instance, have the following situation. Systemology as a sub-theory consists of phonology* and grammar* and the two realizational levels allophony, dealing with the phonetic *

N o t e that what I am saying here is with reference to the theory, not the description. When dealing with systemology from the point of view of description, i.e. 'systemology' as a 'system of systems', the terms 'phonology' and 'grammar' are used in a narrower sense (i.e. complex phonological system, and c o m p l e x grammatical system respectively). T o the terms as used here (i.e. with reference to 'theory'), in a description correspond the terms 'phonological system' and grammatical system', and as each of these primary subsystems contains, as a further subsystem, an inventory of simple entities as well, there are in 'a systemology' in total eight (not six) subsystems. We shall see all of this later. Please note that the term 'systemology' with reference t o 'description' is virtually equivalent to 'semiotic system', when we speak about 'simple semiotic system', 'unordered semiotic system', etc.

80

Theory and strategy - C'est le point de vue qui tree l'objet (Saussure)

realizations of phonological forms, and allomorphy, dealing with the phonological realizations of grammatical forms. Phonology is further divided into phonematics, roughly the analysis and composition of phonemes as to their distinctive features, phonotactics, i.e. the analysis of self-contained groups of phonemes, called phonotagms, and para-phonotactics, which includes prosody and other non-phonemic features, such as juncture, etc. Grammar is divided in a similar way into plerematics (roughly the analysis of 'words', also called 'morphology'), plerotactics or syntax, and paraplerotactics, which deals with the sentential aspect, including intonation. Semantics as a sub-theory can be regarded as consisting of the theory of indices and denotational semantics. Though Hervey (Hervey 1980, and elsewhere) has considered non-denotational areas of semantics, this does not form part of the axiomatic theory as such, though the area is, in a negative way, implicitly defined by the theory. The relations between subtheories and sub-disciplines or descriptive areas might, then, be as follows:

The three legs of the theory

υ

JS 2

ο •C

1-3 ο Λ ··" Χ> Χ cd 1) βο Ä e 3 Ο υ

C/5 C CC Π cd Ο _ 0C 3 =3 Λ ο W 'S CO * ζ cd C/3 ΪΛΛ2 — < ω « cd ι— Γ-1ι ι_ g o

° c —ι cd Ο

ΙΛ D ω C ω . 3 2 -> t! 5

υ 2 Ε

£°

81

Ο .2 Cd t; ο ί

ο

ΙΛ Ε 3 1 Μ ω>

ο C υ 5 Τ3 Ό ω « C u, Ό I« XJ Ο ω ι> ο Ο >> • δ Ό χ: C Ä -C 00 = ω ο Λ* «β Τ3 Ο η £1 cd ι_ C C Q, >> H ω « ω Ό υπ 5 ω cd d · ° 00 S •υτ—I c 5 U o . ä >g _ 00 c £ Ο ο 00 1> Λ C s-s Ε I υ 1) > 2 S υ ε * C Iυο υ c Π5 22 J3 ω ο τ3 ω •2 g c Λ · ο * > Ο ~ Cd £ ο 73 tg . . a 13 υ ° S ο ο ü cj Ο w! Μ 1 >> Cd Ε Ο ω Λ OJ >> ä 35 ä 2 U cd S Λ cd C Ä ο. cd c e c o ™ £Λ C υ 1) 3 >1) >> 3 >> i- S ο e u, Ο > 00 '*-· £ 3 c t jj Ο u κ -τ-. .£5 .s s Ν ο -ο ο C _>. u c cd f> χι ω cd X υ cd ο ccd Ο ^ cd y D ce C Ό C OJ JS d j: t ι ή, '3 1 ωS c '5 ^ C Ο .23 S ο ω Z> s « 2 Ό > ti Λ U C ω C Τ3 cfl ao cn en ΙΛ Ό ο C C _ c t/5 Ό rο C tSZ. ο C Ο "ο » .2 υ υ > α' Ο ,-, Ό O 4-· C C d 1) ω ·,£ Od Η -C° R « C £ Ε ^ λ ο u. C ο Ο >. "u Ο ο tΛ υ 5ό Λ .Η « cd c« 5ä Ο > T3 υ •ο •c 1/5 C 1/5 C Ο T3 « U 13 C c ω id ω ·- S Ο •a g- υ Τ3 α, ο «

Chapter III: Function, system, and systemology

1. The first axiom Basic to the theory are its six axioms. Of these the one that pervades the whole of the theory is Axiom A: 'All features in semiotic sets are functional'. As we are generally dealing with self-contained sets of features having a common purport, i.e. with 'systems', we may read 'semiotic sets' here as 'semiotic systems'. The reason why I have chosen to use 'sets', rather than 'systems' in the axiom is that I did not wish to rule out, a priori, the possibility of referring to any set of semiotic entities, selfcontained or not. The term 'functional' is defined as 'separately relevant to the purport of the whole of which it is a part'. A 'semiotic system' is a 'system of conventions for communication'. The repercussion of this axiom for the whole of the approach is that nothing is considered to belong to the semiotic system in question, unless it is functional. The definition of 'functional' implies that there is opposition or choice, i.e. the possibility that presence and absence of the item in question can be manipulated, as I have already said. Another consequence, inherent in the choice of the very general term 'feature', rather than the more specific term 'entity', is especially noticeable in the methodology. Let us take an example from semantics, where, quite naturally, one is dealing with features rather than entities. Some linguists, not adhering to the functionalist approach (to call it the oppositional or differential approach, whatever you wish), have come to the conclusion that such words as 'and' and 'or', have no meaning, at any rate no denotation, but only grammatical function. Under the present methodology one will, however, reason as follows:

84

Function, system, and systemology

(a) a syntagm such as 'a book and a pen' has a different denotation from 'a book or a pen'. (b) substitution of 'or' for 'and', and vice versa, amounts to a valid commutation i.e. the relational structure is not affected by the commutation. (c) we conclude that the difference in denotation resides in the difference between 'and' and 'or', which must, therefore, themselves have denotation. All this amounts to saying that 'and' and 'or' are distinctive in respect to one another. If an entity is distinctive in respect of at least one entity at the same level of analysis, it is functional within the system. If, and only if it is functional within the system, it belongs to the system (cf. Axiom A). But not only the distinctive function, i.e. its having semiotic identity, is confirmed by the above, but also the fact that it has denotation, i.e. its own inherent information-value. That is to say, it corresponds to a particular class of denotables, and can be used an infinite number of times in actual communication in order to denote a single member of that class (cf. Axiom F, which is Hervey's axiom for semantics). The above considerations (a), (b) and (c) are — in agreement with the hypothetico-deductive method — of course descriptive statements, the material adequacy of which may be refuted. One sees that, in the first place, only linguists who do not use this kind of functionalist methodology may come to the conclusion that, say, 'and' and 'or' are mere 'empty morphs', with only grammatical, but no semantic function. Secondly, the functional principle not only makes it possible to corroborate, by refutation of the contrary, that a particular entity has a denotation (which incidentally implies that it has distinctive function, but not vice versa), but it also lays down the criteria under which contrary assumptions may be refuted.

2. The notion 'self-containedness' When we talk about entities in linguistics, we generally mean selfcontained entities. We are not generally interested in establishing

The notion 'self-containedness'

85

entities that are not well-formed at the level we are considering at that particular moment. I say 'generally', because we should leave the possibility open for dealing with parts of forms in the course of our discussions. We may, for instance, in phonology, wish to deal with clusters of consonants, e.g. /st/, as in English 'stands', but we do not establish these as units unless in some way or another they do behave as a unit. Even less are we considering groups of words, such as 'the very big', let along 'the very', in, say, 'the very big box'. We may be considering 'big box' on its own, e.g. when we are dealing with the relation between 'adjective' and 'nominal', but even this we do not set up as a unit, because without 'the' or one of its commutants (i.e. entities that can stand in its place) it is not well-formed or self-contained. Therefore, any time we deal with entities, it is assumed that we mean self-contained entities, even though this may not have been specifically mentioned. Apart from being concerned with entities, we are also in many cases concerned with sets of entities, and also here we are generally only concerned with self-contained or complete sets of functional entities, i.e. with what we have called systems. This may also not be specifically mentioned in all cases. Even when we are dealing with infinite sets, the assumption of self-containedness is tacitly implied. Whether we talk about the set of vowel-phonemes of English, the set of Spanish words, the set of prepositions in German, or the set of sentences in Thai, we always mean the set of all the items that are covered by the intension of the set in question, whether infinite, as in the latter case, or not. The notion 'self-contained' is formally defined as 'representing all relative dependencies of its members, as members of the set in question'. It is the intension of the set (see Chapter II) that determines which entities are to be regarded as belonging to the set, and, in a negative way (by implicitly specifying the complement of the set) which entities do not belong. In a functionalist approach, the identity of any item depends on its being functional, i.e. in the case of self-contained sets or systems it depends on its being functionally distinct from any of the other members. Every other member, including every other potential member, contributes to the identity of the member in question. As this is true for any member of the set, to be self-contained is in practice synonymous with being complete, i.e. containing all members that come under the intension of the set.

86

Function, system, and systemology

Thus, as we have seen, systems (and also sub-systems) are, by definition, self-contained, which can, as I have said, be taken as meaning complete, and also items (linguistic objects) are selfcontained, which can in practice be equated with being wellformed. The mutual relative dependency criterion also applies to the latter, as a well-formed, complex entity in a functionalist description is an entity that contains all the constituents that are necessary in order to make the entity in question well-formed, i.e. it contains all the constituents on which the function of the constituent in question depends. Much of this will become clearer in the chapters dealing with linguistic description (Chapters V, VI and VII). For the moment I should like to stress that, for example, 'the big' in 'the big box' is not self-contained, as there is no relation between 'the' and 'big', other than via the relative functional dependency of each of them on 'box'. It may, however, not be directly apparent that also in 'big box' the self-containedness criterion is not satisfied. The reason for this is that, without 'the' or something of the sort, 'big box' has no function within the unit which can be demonstrated by attempting to use it as a constituent in a larger unit, e.g. '* big box is on the table'. There is something missing here. In order to make it well-formed '. . . big box' has to be supplemented by, say, 'the', 'a', 'one', 'that', 'my father's', etc. Consequently the group 'big box' is not self-contained. If it has no potential function, it is not functional, is therefore not an entity, and the self-containedness criterion cannot even be applied, let alone be satisfied. The fact that 'big box' by itself could be a constituent in some cases — e.g. in an Alice-in-Wonderland situation: 'big box go home' - is not relevant, as we are dealing with 'big box' in 'the big box', and this does not commute with 'big box'. That is, the two 'big box'es are functionally not equivalent. But in the case of the box', matters are different. It commutes with 'the big box', and the function of 'the' is completely determined by its relation to 'box'. No intervention of 'big' is necessary here. But though an entity 'the box' is self-contained and well-formed, within 'the big box' it is not a functional entity, because it is not a constituent. It would be absurd to regard it as being a constituent, because, in that case, there would be functional dependency between 'the . . . box' and 'big', which clearly is not the case. One sees that, just as in sets the notion of self-containedness can only apply if we

Earlier stages and development

of the theory

87

are dealing with actual members of those sets, in constructions it can only apply if we are dealing with immediate constituents of the construction. Self-containedness o f sets is a matter of dependency between members, and self-containedness of constructions is a matter of dependency between immediate constituents. I especially emphasize the word 'immediate', as the greatest mistakes are made - sometimes by axiomatic functionalists themselves — when they forget that in analysis one is always concerned with only one level at a time, i.e. only with immediate constituents and relations between those. I shall come back to this in chapter V , Section 8.

3. Earlier stages and development of the theory

The present theory had its origin in certain ideas I cherished, and which had already acquired a certain recognizable form by about 1960. In retrospect, I can say that the whole theory, with the exception of the theory o f denotation, developed by Hervey under my supervision, was at that time already there in an embryonic form. There was, in the first place, the insight that the distinctive function of an entity was wholly determined by the set o f oppositions into which that entity enters, and that it could be expressed in a Boolean form, i.e. the distinctive function of a in the set a,b,c,d equals ( a - b ) υ ( a - c ) υ (a-d) = a-(bncnd), which by de Morgan's theorem is equal to α π {~b υ -c υ -d). In plain language this amounts to saying that the intrinsic identity of a is determined by the fact that it is neither b nor c nor d. Conversely, the intrinsic identity of b in this set is that it is neither a nor c nor d and so on, and the relation 'is opposed t o ' is isomorphic with 'complementation' or 'negation' in set-theory. That is to say, ' - a ' can indiscriminately be read as 'opposed to a\ 'the complement of a' or 'not a\ and all relevant set-theoretical operations can be applied to the linguistic notions. Actually the basic principles in this were not new. What I was doing here was a mere formalization and reformulation of Trubetzkoyan Prague School, glossematic, and functionalist (Martinet) concepts. These ideas, first published in 1965 (Mulder 1965. 1968), together with

88

Function, system, and

systemology

Martinet's insight that distinctive function exclusively hinges on opposition, which in its turn implies commutability, soon led me to realize that one needed the notion 'position', not as a linear relative concept, but as a constant division within a structure, 'such that in every such division an entity (as an immediate constituent) can stand and alternate (i.e. commute) with other entities, or with zero'. The latter was at variance with ordinary functionalist practice. For instance, Martinet maintained that in French 'cahot' or 'chaos', phonemically /kao/, /o/ commutes with, say, /p/, as in /kap/ ('cap'). This was disputed by me, as I maintained that the 'positions' in which joj and /p/ stand are not equivalent. I was applying these ideas and some further consequences, and testing them at that time in a phonological description of the vernacular of Peking. They were basic to the phonological part of the theory, and also to the grammatical part, even though I perhaps did not realize the latter yet at that time. All this actually amounted to a recognition of the all-importance of the 'functional principle', later incorporated as the first axion of the theory. At that time, I had also set up schemes and developed ideas about the classification of semiotic systems and, fairly analogously to this, about the basic structure of natural languages. The schemes incorporated, but in some respects surpassed, Martinet's tenets about the 'double articulation' of language. For Martinet this amounted more or less to a simple duality, but under the influence of E.M. Uhlenbeck, my former teacher in linguistics, it acquired for me the form of a fourfold system. Uhlenbeck believed in the importance and special status of the concept 'word' as an entity standing at the top of morphology, so to speak, and the b o t t o m of syntax. That is to say, there should be a sharp division between morphology, i.e. the analysis of words, and syntax, the analysis of word-groups into words. I saw an analogous situation in phonology, in that one had to distinguish between the analysis of phonemes and the analysis of, say, syllables into phonemes. Also at these points my ideas were at variance with orthodox functionalists. Martinet, perhaps in agreement with French usage, called 'morphology' the study of variants or allomorphs. There was even a branch of this called 'morphologie de la phrase'. What he calls 'morphology', we call 'allomorphy' or 'allomorphics' in the present approach. But, more importantly, Martinet barely distinguished between morphology in our sense,

Earlier stages and development

oj the

theory

89

and syntax, using the term 'syntax' for the study and analysis of all meaning-bearing entities. His term syntax can, therefore, roughly be equated with our term 'grammar'. The latter ideas are now incorporated in the second axiom: 'semiotic systems contain simple, and may contain complex unordered, or complex ordered signa and figurae\ The 'or' here means 'either . . . or, or both'. The term 'signum' is defined as 'semiotic entity with both form and intrinsic information-value\ the term 'figura' (borrowed, but with a different interpretation, from Hjelmslev) is defined as a semiotic entity which only has form. This axiom originally, until about 1972, had the form: 'Semiotic systems may contain complex elements which can be articulated into elements which have both form and meaning, or elements which have only form' (Mulder 1968). The term 'articulation' was, and is still, interpreted as meaning 'ordered complexity', unlike Martinet's term 'articulation' which, for grammar at least, merely means 'complexity' (Martinet 1960). Though more clumsily expressed, the older form of the axiom clearly betrays its origin, which is in the recognition of the importance of the 'double articulation' concept, albeit that for me this concept differed from that of other functionalists. Also under the influence of Uhlenbeck, I was at that time already aware of the fact that the sentential level had to be treated as separate from the syntactic level, as otherwise both adequacy and consistency of statement were likely to suffer. But I had, at that time only very crude ideas as to how to integrate this with the rest of the linguistic theory. More advanced were my ideas about the area in phonology which is analogous to this. The latter is now called para-phonotactics. Though much still has to be done in these respects as to detail, working out the consequences of the premises and testing their adequacy in actual descriptions, they are now incorporated in the theory under Axioms C and D. Axiom C reads: cenological entities may have para-cenotactic (for language read para-phonotactic) and plerological entities may have para-syntactic features. This includes but surpasses the area of what is traditionally known as 'prosody', as we shall see. Axiom D merely states: All semiotic systems contain sentences, constituted by a base and para-syntactic features. The term 'sentence' is defined as plerological entity with such features that it cannot be a feature (constituent, or other

90

Function, system, and

systemology

feature) of another plerological entity. An alternative, but in practice equivalent definition is 'plerological entity such that it is a self-contained vehicle for conveying messages'. It is clear that, as Uhlenbeck used to point out to his students, a sentence is not a mere string of words or word-groups, but that other, mainly prosodic, features have to be taken into consideration when we are dealing with sentences as such. Plerological entities that are not sentences can only convey messages if and when belonging to, or constituting the base (word-content, by Uhlenbeck called 'phatic layer') of a sentence. Axioms B, C, and D together form the basis of what is now called the systemology, i.e. the whole of phonology and grammar. The consequences of Axiom Β in its original form were worked out in a detailed article by Mulder and Hervey: 'Language as a system of systems' (Mulder — Hervey 1975). On the basis of the original scheme, which dates from about 1960, but which was first published in my Sets and Relations in Phonology (1968), also in my D.Phil, thesis (1966), and on an unpublished anologous scheme for the structure of natural languages, we arrived at a calculus of the logically possible types of semiotic system, and at a definition of the notions 'language' and 'proper language'. The term 'language' is defined as 'semiotic system with double articulation'. 'Articulation' is defined as 'complex ordered system', i.e. cenotactics (in natural language: phonotactics, dealing with ordered combinations of phonemes) or syntax (roughly dealing with ordered combinations of words or word-complexes). The term 'proper language' is defined as 'semiotic system with a cenology (in natural language: phonology), containing both a cenematics (phonematics) and a cenotactics (phonotactics) and a grammar containing both a plerematics (morphology) and a plerotactics (syntax)'. All natural languages known to date are proper languages. Though not published until 1975, this article was actually written in 1973, i.e. it predates the 1974 Postulates for Axiomatic Functionalist Linguistics, presented to the 'Premier colloque international de linguistique fonctionelle' 1974 at Groningen, the Netherlands. In these postulates, the para-tactic levels (Axioms C and D) are added to the levels just mentioned. Recognizing the distinction, made in the ontology (signum-theory), between realizational levels (i.e. the levels of allophone and allomorph) and structural levels, we can set up the following

Earlier stages and development

of the theory

91

scheme for the divisions within the systemology (cf. Mulder 1975): Systemology allophony

κ

phonetics

phonematics

Η

«

phonology

phonotactics

paraphonotactics

plerematics (morphology)

plerotactics (syntax)

paraplerotactics (para-syntax, or the sentential level)

ι

As I have pointed out already, the nature of the division between phonology and grammar is developed within and derives from the ontology. Though phonetics does not belong to linguistics proper, and, therefore not to the systemology, it is an important auxiliary discipline which helps us to make statements of realization, and it is linked to phonology by the level of allophones which, by some, has been called an inter-level between phonetics and phonology. The level of allophones, or allophony, does, however — for reasons which will become clear when we are dealing with the signum-theory — fully belong to phonology, rather than to phonetics. Within this theory this field could, indeed, well be called 'functional phonetics'. This term was used by Martinet (Martinet 1949) to characterize the whole of phonology, but I believe it is more appropriately reserved for 'allophony'. An 'inter-level' between phonology and grammar is the realizational level called 'allomorphy'. This is what French functionalists tend to call morphologie, which should not be confused with our use of the term 'morphology' (as a sub-system, not a sub-discipline, see the scheme at the end of the previous chapter), which is merely an alternative term for 'plerematics'. In that scheme I have used the terms as applying to natural languages. For general semiotics, 'phonology' is to be replaced by 'cenology', and so on. The terms 'plerematics' and 'plerotactics' can be used as alternatives to, or instead of, the terms 'morphology' and 'grammar'. The same is true for 'para-plerotactics', which can also be

92

Function, system, and

systemology

called the 'sentential level'. I shall come back to the latter. For the moment it is only important to remember that the sentential level differs from the syntactic one in that such additional features as 'intonation' and also some other features such as juncture or pause, etc., have to be taken into consideration. It does have to be distinguished from the syntactic level, as there is a level (i.e. the syntactic one) at which, say, 'John is a fool' in the sentence 'John is a fool.', and 'John is a fool' in the syntagm 'if John is a fool' are equivalent, but at another level (i.e. the sentential one) they do not have the same status. As to the distinction between phonotactics and para-phonotactics, for instance, the syllable /ma/ in Pekingese Chinese can be analysed into the phonemes /m/ and /a/. At that level (the phonotactic one), all /ma/ are equal, but as /ma/ can occur with any of the four tones of Pekingese or with 'zero' tone, there is a level (i.e. the para-phonotactic one) at which there exist five different /ma/. Similar considerations apply, for instance, in English, where the same phoneme sequence, say, /diskaunt/ ('discount'), or /kontrast/ ('contrast') can occur with an accent on the first, as well as on the second syllable, and this may correspond to a difference in communication, i.e. it is functional. There are also purely contrastive features of accent, etc., which form an integral part of the unit in question, and there should therefore be a level at which the unit can be described with the inclusion of those features. Thus, there should be a level, other than the level dealing with ordered complexes of phonemes, at which this can be accounted for. This is the level of para-phonotactic entities (e.g. 'phonological' words and word-groups, which are purely phonological entities, but of different and greater complexity than phonotagms), just as there should be, as we have seen, a level of para-syntactic entities (i.e. clauses and sentences, as opposed to 'syntagms' proper). Para-phonotactic entities and para-syntactic entities are phonotactic entities or syntactic entities (syntagms) together with some further features, called para-phonotactic or para-syntactic (or 'sentential') features. A cover-term for the two is 'para-tactic features'. The arrows and lines in the present scheme point to sub-theories and further sub-theories, except in the case of the broken arrows pointing to allophony and allomorphy, where they merely indicate that phonetics and phonology, as auxiliary disciplines, contribute

Earlier stages and development

oj the theory

93

to the levels of allophony and allomorphy respectively. The latter two are called 'realizational' levels, precisely because 'allophones' are realizations of 'phonological forms', and 'allomorphs' are realizations of 'grammatical forms', which correspond, in fact, to signa. Realizations of the former are stated in terms of 'phonetic form', and realizations of the latter are stated in terms of 'phonological form'. Though 'phonology' is a sub-theory of systemology, with respect to 'allomorphy' it is an auxiliary discipline, merely serving the purpose of stating the realization of the allomorphs in question. It would be redundant to regard phonetics as ultimately being an auxiliary discipline to allomorphy as well, as the realization of phonological forms, if necessary, has already been taken care of at its proper level, i.e. that of allophony. It is very easy to see how this scheme, with regard to the structural levels phonematics, phonotactics, morphology and grammar, relates to the classification scheme for semiotic systems. The oldest form of this scheme was as follows: semiotic systems

complex

simple

unordered

ordered

one articulation

/

two articulations

/

into phonological elements

into grammatical elements into* signs

*

into symbols

In a g r e e m e n t w i t h m o r e recent ideas, e m b o d i e d in the revised p o s t u l a t e s ( C h a p t e r VIII), t h e word " i n t o " should be read as ' c o r r e s p o n d s t o " , as sign" and " s y m b o l " are not strictly speaking grammatical n o t i o n s .

94

Function, system, and

systemology

Simple semiotic systems are semiotic systems that consist of no more than an inventory of simple elements. No elements in the inventory can combine with other elements to constitute self-contained complex entities, thereby adding further entities to the inventory, entities that are composed of entities that were already there. A complex semiotic system, on the other hand does have such self-contained complex entities in its inventory. As was pointed out much later (Mulder — Hervey 1975) a description of a complex semiotic system involves more than listing alone. It also calls for a description of the way in which entities may or may not combine to constitute a more complex entity. Complex semiotic systems, of course, also contain an inventory of simple entities. It may, or may not be the case that such simple entities are capable of occurring on their own to convey messages. But there are two logically possible types of complexity. In both types, elements occur together and stand in a relation to one another such that they constitute a unit. The latter condition, i.e. that of constituting a new unit, is important. A mere juxtaposition of entities, i.e. without unit-forming, does not constitute a complex, but at best a conglomeration. There are, even in a complex with only two immediate constituents always at least two constructional relations. For instance, in a complex (a, b), there is the relation between a and b, and its converse, i.e. that between b and a. In a complex (a, b, c) there are, of course more than two relations. We shall later see that in a complex with more than two immediate constituents the relations between some of the constituents may not always be direct relations, but this is not relevant at this point. If it is the case that all the constructional relations within a given complex are equivalent, including each relation and its converse, we say that the relations are symmetrical or unordered. Symmetrical or unordered relations are called 'relations of simultaneity', and the complex in question is called a 'simultaneous bundle (of constituents)'. An example in daily life of a symmetrical relation is, say, 'being the cousin of someone'. If a is the cousin of b then b is the cousin of a, and if c is the cousin of a and b (if he is cousin of a, he must be of b, i.e. this relation is also transitive) then all are mutually cousins. Compare this with the relation 'being the father of someone', which is asymmetrical. At this point, i.e. when we are talking about complexes in terms of constituents, we are, of course only concerned

Earlier stages and development

of the theory

95

with constructional relations as such, not with, say, particular semantic relations, or relations of any other kind there may be between constituents. This is not, however, to say that constructional relations may not have a certain semantic import, but whether they have or not, or what the nature of the import may be, has no influence whatsoever on our decision to regard a certain relation as symmetrical or not. Examples in linguistics of complexes in which all relations are symmetrical are, for instance, the phoneme /p/ in English (a phoneme is by definition a bundle of entities in symmetrical relations), or the word 'cats'. The phoneme /p/ is a bundle of three distinctive features: labial, occlusive and unvoiced. In this case they are even physically simultaneous in realization, but this is not a necessary condition for simultaneity. For example, the pre-vocalic part of German Pfund ('pound'), realized as a medium [pf] is nevertheless, under this theory, a simultaneous, i.e. unordered, bundle of features: occlusive, fricative and labial. Crucial here, for a functionalist, is the fact that the order of the features cannot be manipulated, i.e. there is no choice as to changing the relations, whilst keeping the constituents (in this case 'distinctive' features) the same. In the case of the word 'cats' (words, or more precisely, 'pleremes', are by definition simultaneous bundles) too, there is no choice as to manipulating the relations. One may, perhaps, object: but what about 'cat's'? The latter is, however, as we shall see, not one, but two 'words', which may even not be in direct constructional relation with one another (e.g. in 'the old cat's tail' they are not), so there is here no valid commutation of relations. Another objection at this point may be: but what about say the difference between 'hits' as a plural noun and 'hits' as in 'he hits him'. Again, here, as we shall see (or may have seen already), there is no valid commutation (in fact there is no commutation at all), as '/s/' in the latter, though functional at the phonemic level, is not functional at the signum-level.

96

Function, system, and

systemology

4. Further on the theory of semiotic systems, and the sub-systems of language An example of a complex semiotic system in which all complexes are simultaneous bundles is the system of traffic signs (not traffic lights, which constitute a simple semiotic system). One may have, say, an arrow on a round blue shield. Though the tentative feature 'round' in the code is opposed to 'square', and 'blue' is opposed to 'red', the latter does not occur with squareness. That is, 'red and round' constitute a single feature, a feature that is opposed to 'blue and round'. The other feature is constituted by the arrow. As the relation between 'blue and round' and the arrow cannot be manipulated so as to change the message conveyed, the sign in question constitutes a simultaneous bundle of two features. As this is the only type of complexity to be found in the system of traffic signs, that system constitutes a complex unordered semiotic system. The complexity is of a grammatical kind, Cenologically, this is a simple system, as there is no functional complexity within its mere forms, colours, etc. Had it been the case that redness too could occur with squareness, but without this amounting to complexity at the grammatical level, the system would also from the cenological point of view have been an unordered complex system. If, however, shape and colour had each been features at the grammatical level, then it would still have been a simple system at the cenological level. Between cenology (phonology) and grammar there is so-called 'interlock', i.e. cenology (phonology) provides (generates) the forms of grammatical entities. There is in semiotic systems that contain these sub-systems another type of interlock, i.e. between phonematics and phonotactics and between morphology and grammar, in that the ultimate (i.e. 'molecular') entities of phonematics (i.e. the phonemes) correspond to the minimum (i.e. 'atomic') entities of phonotactics. The same is true for morphology versus syntax, that is, the output of the one, so to speak, is the input for the other. A third type of interlock is that between phonotactics and para-phonotactics, and between syntax and the sentential level. The entities of the tactic (phonotactic and syntactic) levels provide the base (roughly the phoneme-content or the word-content respectively, i.e. what Uhlenbeck on the

Further on the theory of semiotic systems, and the sub-systems of language

97

sentential level has called the 'phatic layer 1 ) for the entities at the para-tactic (para-phonotactic and sentential) levels. Let

us now

move

to

ordered

c o m p l e x i t y , and to

complex

ordered semiotic systems as a type. In the original scheme I have distinguished between ordered c o m p l e x i t y o f t w o kinds, i.e. o f the

cenological

(phonological)

and

of

the

plerological

(gram-

matical) kind. Though, o f course, it was recognized that, logically speaking, a similar division could be made in respect o f unordered and indeed simple systems, this was not regarded as interesting enough at the time, especially characterization

of

language

as the aim was no more than a

and

a redefinition

of

Martinet's

concept o f the double articulation. Language was characterized as having ordered c o m p l e x i t y o f both kinds, i.e. phonologically as well as grammatically. It was not until about 1973 that Hervey and I worked out the full consequences o f the scheme (Mulder - Hervey 1975). A typical semiotic system with only a cenological articulation, i.e. ordered c o m p l e x i t y code.

This

example

some

additional

delimitations preventing

of

some

o f a purely

formal kind, is the Morse

also gives us the

interesting semiotic possible

opportunity

observations

as t o

systems, if only misconceptions

of

the

making

unity

and

by anticipating and which otherwise,

es-

pecially with regard to the Morse code, may arise. T h e inventory of

simple

forms (cenological

c o d e contains only

entities) belonging to the Morse

t w o items, i.e. a dot and a dash, or their

equivalents, such as a long and a short sound, flash o f light, etc. Though a dot can occur on its o w n as the f o r m o f a sign denoting the letter 'e', and so can a dash, denoting the l e t t e r ' t ' , this does not, o f course, exhaust its possibilities. One or more dots, and one or more dashes can also occur in combinations, denoting particular letters, order

particular in

which

numerals,

and

some

other conventions.

The

dots and dashes occur within a self-contained

complex o f dots and dashes is relevant, i.e. different orders imply forms o f different signs. F o r instance denoting the letter 'a'. The sign denotes *1\ ' denotes 'oblique

stroke 1 ,

is

the form o f a sign

denotes 'n', *

' denotes '3', '

' denotes

T,

' denotes *7\

etc. Though it is trivially true that non-

mixed forms, e.g. '·•', which denotes 'i\ or ' —

1

which denotes l m \

are in actual fact simultaneous bundles (as there is no possibility o f choice in ordering), and also the simple V or

" may occur,

Function, system, and systemology

98

we may say that the Morse code is an ordered system. Under the definition of 'ordered system', i.e. a 'complex system with ordering relations between elements', it is sufficient to contain only two self-contained ordered complexes f o r a system to be considered an ordered one. There must be at least two such complexes, as the existence of functional choice of ordering can only be ascertained by demonstrating functional opposition between two different orderings of the same elements. It is important to note that I have said that certain signs of the Morse code may denote 'numerals', not 'numbers'. It is the numerals that may or may not denote numbers in actual messages, but as numerals are the objects denoted, i.e. they are objects in the world external to the Morse code as such, they do not belong to the Morse code, any more than real cows or cats belong to the English language. It is in this area that misconceptions may arise, and one may even think that the Morse code has more than one, and even three articulations. The reasoning, albeit absurd, could be as follows. Ordered complexes of dots and dashes signifying letters of the alphabet are themselves ordered and constitute the forms of 'words' which, in their turn, are ordered into complexes which convey messages. That is to say that ordered complexes constitute the forms of signs, which, in their turn, constitute the forms of signs at a further level, which signs, in their turn, constitute further ordered complexes. The picture emerging from this reasoning can be schematized as follows:

Ordered Combinations o f :

•

+

dots and dashes

letters

words

e.g.

e.g.

e.g.

-

the + ant + is + an + insect =

=·- = a

=

a + η + t = ant

the ant is an insect 11

cenological'entities ,v

ι ι ι

ι

nJ/

grammatical entities I ι

Further on the theory of semiotic systems, and the sub-systems of language

99

One sees that from the point of view of the ultimate message conveyed (which is in the English written language) the 'letters' are cenological entities, whereas from the point of vie w of the Morse code, in which semiotic system, say, ' . _ ' actually denotes the letter 'a', the ordered self-contained combinations of dots and dashes correspond to signs. And this is precisely the point. We are here dealing with two different semiotic systems. In fact the two systems are totally independent. The Morse code as such is indifferent as to whether the ultimate message is conveyed in English, in any other language, or indeed, other semiotic systems, codifications, and the like. The assumed articulation into letters and the one into words, does not belong to the Morse code, but to another semiotic system, in this case to the writing-conventions of English. To the latter, the fact that the 'letters' are ordered combinations of dots and dashes is irrelevant. At best, from that point of view, one could say that the written forms are multiply ordered, e.g. ' ( · + — ) + ( — + · ' ) + —' just as is often the case in the syntax of language, e.g. '(the + man) + (is + (in + (the + garden)))', but this is irrelevant to and has nothing to do with the Morse code as such. As a rule of thumb — and this is, in fact, theorematic under the present theory (cf. Mulder — Hervey 1975) — we may say that any semiotic system contains exactly one cenological system and exactly one plerological system, either of which may be simple, or complex unordered, or complex ordered, or any combination of these. The term 'cenological system' is defined as 'system of cenological entities', i.e. 'selfcontained set of formal entities'. The term plerological system' stands for 'system of plerological entities'. In actual fact, semiotic systems must of necessity also contain para-tactic features, if only in the form of junctures, features of discreteness, and the like. For the simplest of systems this may be trivial, but even for such a relatively primitive system as the Morse code, it is important that junctures between self-contained complexes, i.e. the beginning and the end of a complex, and those between the constituents of those complexes, are properly adhered to. These features properly belong to the Morse code, and generally take the form of intervals between complexes of the same duration as the realization of a dash, and between the constituents of a complex of the same duration as a dot. Trivially, also the beginnig and the end of a transmission are, of course, indicated. Within

100

Function, system, and

systemology

para-tactic features I distinguish between contrastive and distinctive para-tactic features. The former have the sole function of giving form over and above the form of tactic entities. They indicate groupings and provide unity to tactic complexes as such. It is this type of para-tactic feature that is, by necessity, part of any semiotic system. As to distinctive para-tactic features, I do not know of any semiotic system that contains them, except languages. Distinctive para-tactic features are para-tactic features that are in relation of commutation (functional opposition) with other such features, or with 'zero'. Of these, only distinctive parasywtactic (i.e. sentential) features seem to be part of all natural languages, but many natural languages, notably the so-called tone-languages (but to a limited extent even such languages as English, e.g. the difference between /köntrakT/, i.e. for the noun 'contract', or for the verb indicating 'instigating an agreement', and /konträkT/, i.e. for the verb 'contract' roughly meaning 'to shrink', or as in 'contracting a disease', in addition contain distinctive para-p/zo«otactic features (which, however, perhaps always function contrastively as well). An example of a semiotic system with an articulation into grammatical elements, i.e. a syntax, but of which the cenological system is a simple one, is the system of decimal number-writing. The cenological system is an inventory of ten entities (numerals), i.e. 1,2,3,4,5,6,7,8,9,0. Any one of these numerals is the form of a sign, utterances of which denote a numerical quantity, i.e. 'one' 'two', etc. As the forms of these signs are cenologically (graphologically) simple, the corresponding signs, i.e. the numerals in their capacity of having denotation — in which case we call them numbers' — must be simple as well. If they were complex, their constituents would be signs as well, but as there are by definition no signs without form this could only be the case if the forms of complex signs were, in some way or other, complex themselves. We have therefore at the grammatical level, to begin with, an inventory of simple signs. These signs have the status of 'pleremes' (roughly: words), as they are orderable. They can be used in ordered combinations, in this case forming new self-contained entities which too are numbers, i.e. utterances of those complex signs too denote numerical quantities. It should be noted that numerals can be used in many other ways then for the denotation of numerical quantities, but in the system of number-writing we

Further on the theory of semiotic systems, and the subsystems

of language

101

are only concerned with the latter deployment. If they are used in other ways they constitute the cenological inventory of other semiotic systems, or they do not belong to a semiotic system at all. The system of decimal number-writing is interesting in two important respects. In the first place it vividly and in a simple manner illustrates the idea of the nucleus of a syntagm with a sub-ordinative structure, this nucleus being the identity-element for the function of the peripheral entities, i.e. the entities subordinated to this nucleus. Secondly, the syntax of this semiotic system resembles, qua tactic structure, the phonotactics of languages, rather than the syntax of the latter, as we shall see. The identity element in any complex number is the entity in unit position. It is formally marked by being the last one in the sequence, or, if there are decimals, by being followed by a comma (in most European countries), or a full stop (in Britain). We can in fact regard a numeral followed by a comma (or a full stop, under other conventions) as representing an allomorph of the sign in question when standing in unit-position and followed by one or more decimals. As, say, the difference between 123 and 312 is functional, i.e. corresponds to a difference in communication, the self-contained complexes constituted by the simple signs are ordered grammatical complexes, i.e. they are syntagms. In the case of, say, 123 (one hundred and twenty three), the values of Ί ' and '2' are each determined by the relation in which they stand to the nuclear element '3'. In the case of, say, 812,16 (eight hundred and twelve and sixteen hundredths), the values of '8', Ί ' , Ί ' , and '6', are each determined by the relation in which they stand to the element in unit-position, i.e. '2'. In the case of number-writing the values of entities, deriving from their relation to the nucleus — i.e. from what we call their 'position' — can be mapped onto a simple ascending and descending scale. Any position indicates the ten-fold of the value indicated by entities in a position immediately to the right, and one tenth of that indicated by entities in a position immediately to the left. Without knowing in a sequence of '8', followed by ' Γ , '2', Ί ' and '6', which of these entities stands in the nuclear position we would have no clue as to the syntactic value of each of these constituents, and consequently we would not be able to judge the denotation of the whole.

102

Function, system, and

systemology

The system of decimal number-writing however also poses a problem with regard to immediate constituent analysis precisely for this reason, namely, that it remains arbitrary whether we regard the immediate constituents of, say, '812,16' to be its ultimate constituents, i.e. '8', ' Γ , '2' (allomorph '2,'), Ί ' and '6', or indeed whether we analyze it (8 1 ) ( 2 ) ( 1 6), 8(12,1)6, or even 8(1(2,1)6), and so on. But we cannot analyze it, say, (812)1(6), or (8)1(2(16)). This is because, in whatever way we may analyze it, the nucleus nucleorum, i.e. the ultimate nucleus of the construction is '2', so '2' should be, or be part of (in fact the nucleus of) the nuclear immediate constituent. This situation is, as we shall see, analogous to the normal situation in phonotactics. In phonotactics it has to be so, because we are only concerned with functional form, not with anything that has denotational import in respect of the messages conveyed. And the form of the relative position in phonotagms can be nothing else than relative peripheralness or nuclearity. We may say that, for instance, in /sTror]/ 'strong', the relative degree of peripheralness of 's', 'T', and 'r' can be mapped onto a descending scale, and T ' is as much more peripheral in respect to 'r', as 's' is in respect to 'T', and so on. But in syntax everything that does not directly have 'denotation' has at least 'denotational import'. That is, it contributes to the denotation of what it is a feature of. It is purely accidental from the point of view of syntax that the system of number-writing has a syntax that resembles the phonotactics of natural languages. It is only because of the nature of the decimal system of calculus, which determines the denotational range of the notation in terms of number-writing, that this situation prevails. Not all syntactic systems need to possess this kind of regularity, and indeed many, including those of languages, do not exhibit anything like it, and it is, in this case, entirely due to a strict regularity within its denotational range, i.e. within that part of the external world to which entities of the system of number-writing purport to refer. Two entirely different types of semiotic system with a simple cenematic system, but with an ordered grammatical system, i.e. a syntax, are the notational systems of arithmetic and ordinary algebra. As the inventories of simple elements, and consequently those of complex elements, differ, they constitute different semiotic systems. This is not to say that there are no structural

Further on the theory of semiotic systems, and the sub-systems of language

103

similarities between these two systems. It is even the case that their inventories overlap to a certain extent. The syntax of these two systems has much more in common with the syntax of natural languages than, say, the syntax of decimal number-writing has. This is probably due to the fact that their syntagms and sentences are modelled on certain syntagms and sentences of certain natural languages, but as this is a diachronic matter it is irrelevant. As systems they are independent of natural languages, even though they can be translated into natural language, which is to a certain extent done when one reads a formula aloud. Also notational systems for formal logic are of this kind. It is important to distinguish between notational systems qua semiotic systems, and the same systems extensionally speaking as devices for calculus. From the latter point of view they are, of course, not semiotic systems, but they do determine the denotational range of the semiotic systems in question. An interesting difference between the system of algebraic notation and notational systems for formal logic on the one hand, and the system of arithmetical notation on the other is that in the latter the denotations of the elements of the system are wholly fixed-conventional, whereas in the former they are not. Such symbols as a, b, c, etc. in algebra require separate definition for any separate operation before they can be denotationally interpreted. This has prompted the last division in the classification scheme for semiotic systems, i.e. that into an articulation into signs (wholly fixed-conventional), and into symbols (not wholly fixed-conventional). I shall come back to this point below, when discussing the theory of indices. Yet another example of a semiotic system with a syntax, and even a morphology as well, is the system of common musical notation. One might perhaps think that this is not a semiotic system, as it lacks a plerological system, but this is confusing the notation for music with music itself. If we analyse the latter we may find a cenological system — which in this case can truly be called a 'phonological' system — and even a true phonology including phonematics and phonotactics, but as no communication in our sense (i.e. conveyance of messages) is involved, there is no real plerology. However, the phonology of music itself constitutes the denotational range of the notation for music. It is not unlike the situation we have found to prevail in

104

Function, system, and systemology

the Morse code, i.e. that the graphology (cenology) of one system coincides with the grammar (plerology) of another, except that in the case of the Morse code the system interlocking with it can be a semiotic system (e.g. written English), whereas in the case of musical notation there is only one use, i.e. 'music', which, in the absence of a plerological system, is not a semiotic system. The graphological system of musical notations is not entirely a simple one. Each grapheme, i.e. 'note', is in fact a simultaneous bundle of placement on a particular line, or between particular lines, indicating pitch, and a formal feature indicating relative duration. These features, which trivially we may by abstraction consider as purely formal, correspond to monemes at the plerological level. They denote pitch and relative duration respectively, and as a simultaneous bundle of two of those monemes, i.e. as a plereme, they denote a 'sound' of a particular pitch and relative duration. It would be wrong to consider some of the phonological, or their corresponding grammatical, features of relative duration to be further analyzable. Even though in, say ' J) \ the ' >• ' part is opposed to 'zero', and to ' " ', etc., the ' J ' part is not functionally opposed to anything, and or ' J ) ' , etc. or their variants (as in ' ', etc.), are therefore not further analyzable.* This is due to the 'functional principle', which is the leading principle of functionalism. There is no feature, unless it is functionally opposed to its absence, or to another feature, and if part of something is not a feature, the other part cannot be either, as nothing can be separated from parts of itself. A different situation, however, prevails in the case of 'dotted' notes, e.g. ' J ', ' J ) ' , ' ', etc. In these cases, at the graphological level, commutation' of either part is possible. We have, therefore, at that level to recognize two features. As there is no ordering relation involved, ' J # ', etc., are graphemes (cenemes), and they are simultaneous bundles of two features. Their grammatical correspondents, however, are not further analysable into monemes. This may seem strange, as one might be tempted to say that the 'note' denotes a particular length, and the dot denotes 'half of that length again'. There are, however, two flaws in this argument. *

I have left abbreviatory conventions, such as > e t c -> f ° r J 1 ' e t c -> out of consideration for the sake of simplicity. It is not my intention to present here a semiotic description of musical notation, only to clarify certain points of descriptive methodology in general.

Further on the theory of semiotic systems, and the subsystems

of language

105

In the first place, the denotation is not 'half of that length again', but it is rather an instruction to give the note in question one and a half times the relative duration value of a similar, but not dotted, note, i.e. — and this reveals the second flaw — it indicates at best the relative duration-value of the note in question. If there is 'denotation' connected with the dot at all, it is not denotation in respect of anything in the world external to the notation in question, but it is rather an instruction for the interpretation of a sign employed in the notation. A dot used in this way is inseparably linked with the 'note' it follows, and together — but only together — they correspond to a particular relative duration of a sound in the external world. The inescapable conclusion is that 'note' and dot do not correspond to separate monemes, i.e. that 'dotted notes' are not analyzable at the grammatical level, notwithstanding the fact that they are at the graphological (cenological) level. The system of musical notation, then, has a cenematics, a morphology, and a syntax, but no cenotactics. In addition, of course, like any other semiotic system, it has both a cenological inventory of simple figurae, and a plerological inventory of simple signa. It may, perhaps, seem that the case of 'dotted' notes is analogous to, say, the plural formation in such languages as English, but such a view would be based on a wrong interpretation of the so-called 'singular'. The latter is not really 'singular' but rather 'neutral', the plural moneme adding something and thereby interacting with it. It would be absurd to assume that 'plural' is added to singular. We have, for instance 'two varieties of flower', 'a kind of disease', etc. I have gone into this in extenso, because it illuminates and exemplifies pitfalls that may to an even more dangerous degree be encountered in the phonological, as well as the grammatical, analysis of natural languages. There is, of course, much more to be said about musical notation as a semiotic system. There are other conventions, such as time-signature, which is perhaps strictly speaking redundant, the use of Italian words (and certain symbolizations which are equivalent to some of those), which can perhaps be seen as hints towards the realization, the use of vertical lines to make a division into bars, which may perhaps be regarded as para-tactic features, reduplication signs, 'sharps', and 'flats', 'clefs', notation for chords, etc. With regard to 'sharps' (#) and 'flats' ( Z?) similar observations can be made as the ones we have

106

Function, system, and

systemology

made in respect of 'dotted notes'. As to chords, they are simply simultaneous bundles of one or more features of relative duration, and more than one feature of pitch. Some descriptive complications arise from simplifying conventions other than the ones already referred to, e.g. that of indicating 'sharps' or 'flats' at the beginning of a piece or section, but such complications can be easily overcome, by regarding them as equivalent to writing them every time immediately preceding the relevant 'notes'. There are different notational conventions, largely additional to the ones already mentioned, such as the use of two sets of five lines, each with a different 'clef', for key-board music, or the occasional addition of 1, 2, 3, 4, and p, i, m, a, in music for guitar, giving advice (but no more than that) as to which fingers of the left and right hand respectively may be used. The former difficulties can partly be resolved by similar equivalence-considerations to those mentioned with respect to the use of 'sharps' and 'flats' at the beginning of a piece, or part of it, whereas the latter - especially as they are no more than advisory — can be regarded as hints towards the instrumental realization of a piece. They do not belong to the semiotic system as such, but at best to an auxiliary, but separate, semiotic system. With regard to the classification of semiotic systems as such it is irrelevant whether all or some of the forms used are 'iconic' or not. By 'iconic' I mean reminiscent of, or directly hinting at, objects or parts of objects which lie within the denotational range of the semiotic system in question. Examples of partly iconic notational devices are, for instance, used in guitar-music. In the first place, there is the system of so-called tablature, originally devised in the 16th century for the Spanish vihuela, but more recently adapted to the use of the modern six string guitar. It has six lines, representing the six strings, on which numerals are projected, indicating the frets on which the fingers of the left hand are to be placed, together with some other conventions. Assuming that the guitar has the correct tuning, these two formal features together, i.e. placing on a particular line, and the numeral in question ('o' indicates 'open' string, i.e. not depressed by a finger of the left hand), result in a 'note' of a particular pitch. Relative duration of the notes is indicated by horizontal spacing between the cyphers used, or by employment of the 'relative duration' part of ordinary musical notation as an auxiliary system.

Further on the theory of semiotic systems, and the sub-systems of language

107

It is not difficult to see that this too is a semiotic system with a syntax, in which — if this device is used — also the linear order of numbers corresponds to ordering relations in grammar. Another system of notation in guitar-music, exclusively used for indicating chords involves no syntax but it involves a cenematics and a morphology (plerematics). This system too is 'iconic'. It represents, in each of the figures used, frets as well as strings, on which numbers are placed to indicate which of the fingers of the left hand are to be placed where. Cenologically the figures are simultaneous bundles of three distinctive features, each corresponding to a moneme on the grammatical (plerological) level. The whole figure, thus, corresponds to a complex plereme at the grammatical level. These chords, though usually symbolized by a mere indication of their names — can be put in an ordered succession for musical accompaniment purposes, thus adding a syntax to the system. To be distinguished from semiotic systems are what can be called semioid i.e. semiotic-like (rather than 'semiotic'), classificatory devices, such as, for instance, systems for library classification. To this type also belong iconic indications, such as those used for 'gentlemen' or 'ladies' on toilet-doors, 'no smoking' signs in trains, etc. Even a little heart fretted out on the door of a little wooden hut in gardens of farms or old houses in some Continental countries conventionally indicating that the edifice in question is a toilet, may perhaps be considered as belonging to this type. Though the signs in question are indices (see below), and even conventional indices, they do not normally belong to semiotic systems. While providing information, they are not used for communication. By 'communication', I mean the wilful act of giving new information, not just putting a label on something for easier reference. The possibility of giving new information presupposes that there is an element of choice, i.e. that the entities used in the system are functional. This is not generally the case with classificatory devices. Were it the case that, say, the little man and the little woman on public conveniences could regularly alternate with one another, and perhaps even with their total omission, indicating that now the place is accessible to gentlemen, then to ladies, and on yet another occasion may exclusively be used for storing fire-extinguishing equipment, the system in question would be a semiotic system, formally of an

108

Function, system, and

systemology

iconic nature. As the convention stands at present, however, it is a mere device for classifying certain places as to be used by ladies, and others to be used by men. It is not different with systems of book-classification in libraries. These classify books as belonging to certain types, and the classificatory label is determined by the nature of the book in question. The fact that different labels might have been appropriate, owing to the hybrid nature of certain books, does not make any difference, and nor is it relevant that librarians may make mistakes, or different librarians might have classified certain books differently. The only thing important here is that the classification label does not provide information about the book that is not already there, in the book itself. Also national flags and symbols of that kind belong to this category. Though the term 'classificatory' is perhaps less appropriate here, such flags, coat of arms, banners and the like, represent, as it were, what they stand for, but they do not provide new information. They can, of course, be used for communicative purposes, for instance, indicating that someone important has died, that the family is at home, that there is joy about the birth of a princess, or about other things, including even that now it is time to attack an enemy. In that case the devices become part of, perhaps ephemeral, semiotic systems, by providing the formal elements of the systems. But this has nothing to do with their use as representative or classificatory devices. The same forms may be used for different purposes and in different systems. For instance, numerals may be used in number-writing (a semiotic system) and in systems of library classification (not semiotic systems). An interesting instance of this is the use of proper names. On the one hand, these are classificatory labels for the objects (for instance persons) so named. On the other hand, they are used in languages to refer to those persons, and in that capacity they acquire the status of semiotic entities. I shall come back to this, and other points already raised in this section when we are dealing with the theory of indices. We shall see that proper names used in language are symbols rather than signs, i.e. they are analogous to algebraic symbols rather than to ordinary 'words'. It is clear that — as Martinet has pointed out on several occasions — complexity is intimately linked with systemic economy. We should here distinguish between 'potential output' and 'ec-

Further on the theory of semiotic systems, and the sub-systems of language

109

onomy'. By the former I mean the total inventory of simple, as well as complex self-contained entities. This potential output depends on at least two factors: (a) the number of simple entities, (b) the possibility of combining these entities. If the latter includes the possibility of so-called 'recursivity' (e.g. utterances of 'the house that Jack built' type) the potential output of any such system, of course, attains infinity. However, recursivity can in practice play only a limited role in day-to-day communication, as its extensive use is bound to conflict with feasibility of transmission. I shall therefore ignore it at this point. We can then say that the economy of a system equals the relation between ultimate output and initial input (i.e. the sub-inventory of simple entities). Given a certain measure of economy, adding items to the inventory of simple entities would increase the output, but the relation, which can be expressed in the form of a mathematical formula, would remain the same. There are in actual fact some complicating factors, taboos, 'accidental gaps', etc., but I shall ignore these, as the aim of this discussions is not to predict potential output, but to show the economy of complexity in principle. For the same reason I shall also ignore factors that have directly to do with feasibility in actual communication, which is also — as Martinet has pointed out repeatedly — an important form of economy. I am only concerned here with purely systemic economy. In fact, the latter may clash with the former, even to the extent that what is systemically uneconomical may become economical, i.e. it may enhance possible success in communication. But let us return to systemic economy by degree and type of complexity. Suppose we have a system of simple entities (a, b, c), and the system allows for an unordered combination of these entities. The output of the system would then be: (a, b, c, ab or ba, ac or ca, be or cb, abc or acb or bac or bca or cab or cba) i.e. 7 entities. The output of a system under those conditions is always 2 n — 1, where η is the number of initial terms, i.e. (a, b, c) and the subtracted 1 represents 'zero'. In such a system ab and ba, etc., are equivalent, i.e. any combination of a and b, or of a, b and c, constitutes a simultaneous, (i.e. unordered) bundle of entities. If also the difference in order, i.e. the difference between ab and ba, abc and acb etc., had been functional, i.e. manipulative for communicative purposes, it would have been an ordered system, and, if we

110

Function, system, and

systemology

add the restriction that no term may occur twice in the same construction, the total potential output of entities would have amounted to 15. If one had an unordered system of eight simple entities the potential output is 255, whereas that of an ordered system of the same number of entities is 109600, i.e. the latter has a significantly higher degree of economy. If, as is the case in many systems, items can occur more than once in the same complex, the output becomes quite astronomical. In an ordered system of 32 entities, with no restrictions on repetition, only on length, which should not be more than 32 simple entities, the potential output is 150864685143835139943606176280390505254859399200, whereas in an ordered system with the restriction that items may occur only once in the sequence, the possible output would only be 715263772544066789526519831424, i.e. roughly 1 of the previous figure.

200000000000000000

Of course, only a small sub-section of the logical possibilities are used, as there are all sorts of restrictions and factors having to do with conflicting considerations of communicative, as opposed to systemic, economy, but the above considerations go a long way to explain why natural languages can be such efficient and rich systems. A practical restriction, though not a restriction in principle, is the fact that in normal communication only utterances of limited length can be tolerated. Nevertheless the economy of natural languages is considerable. There are, in natural languages two unordered systems, i.e. phonematics and morphology (plerematics), the output of which is the input for two ordered systems, i.e. phonotactics and syntax (plerotactics). Added to this we find the para-tactic levels of which especially the sentential one adds further potential to the already mighty apparatus. There is also the feature of recursivity, but this is, as I have said, in my opinion — though interesting — not so important a feature with regard to the actual (as opposed to systemic) output as some seem to assume. It is not surprising, therefore, that the majority of sentences uttered are completely novel, i.e. they have perhaps never been produced before. Chomsky has made quite an issue out of the notion 'completely novel sentence', claiming, or at least implying that only his type of approach accounts for this phenomenon. This has been rightly criticized by Uhlenbeck in a pungent article (Uhlenbeck 1973), almost to the point of ridicule.

Further on the theory of semiotic systems, and the sub-systems

of language

111

Of course, if one's horizon is so narrow as to only consider behaviouristic approaches, and within those only the most naive versions, the fact that speakers are capable of constantly producing completely novel types of utterance must come as a surprise. If one, however, considers the potential of natural languages towards complexity, and especially towards ordered complexity, one would rather be surprised if there were many repetitions of earlier produced sentences. Though functionalism, let alone axiomatic functionalism, does not claim to account for the actual competence of native speakers, if one wanted to go into speculation about this, it would be at least as good a basis for pondering these as yet imponderables, and perhaps even a better, and, epistemologically speaking, certainly a far sounder one. Though there is a relation of analogy between the sub-systems of language on the one hand, and the classification of certain types of semiotic system on the other, one should not confuse the former with the latter. Though languages are systems of systems (cf. Mulder — Hervey 1975) they are not systems of semiotic systems. As I have said, all semiotic systems incorporate at least, and at most, one cenological system and at least, and at most, one plerological system. This condition is, of course, not satisfied with respect to the sub-systems of language. Every language, in its totality, contains itself only exactly one of each, otherwise it could not be a semiotic system. I shall now reproduce from Mulder and Hervey 1975, the scheme for the logically possible types of semiotic system. In the reproduction of the earlier scheme I have omitted examples, but they are incorporated in the newer scheme. Question marks indicate logically possible types for which we have not been able to find any clear-cut examples. I have added ordinary music notation to the original scheme. Among these logically computed types, we find four that satisfy the definition of 'language' as 'semiotic system with double articulation', i.e. those four categories found in the bottom right hand corner of the scheme. Two of those are represented by question-marks, as no examples have been found. They are: (a) just syntax and cenotactics, (b) syntax and cenotactics, plus cenematics, (c) syntax and cenotactics, plus morphology (plerematics), examples of which are most alphabetically written natural languages, and (d) syntax and cenotactics, as well as

112

Function, system, and systemology

morphology (plerematics) ana cenematics (phonematics). Examples of the latter are to our knowledge all natural languages, and possibly also written Korean, in which there are at least traces of distinctive features within the writing system. It should be remembered that 'articulation' in the present theory means 'ordered system', whereas in Martinet's approach it is not completely clear whether it stands for ordering or mere duality. As Martinet in his phonology regards the ultimate elements of the articulation to be the phonemes, rather than the distinctive features, this 'second articulation' is, in fact, an ordered system. But in his grammar, which, in the absence of a clearcut distinction between plerematics and plerotactics, he calls syntax, the ultimate elements are the monemes (roughly: morphemes). As, however, ordering for Martinet is linear, rather than functional ordering, from his point of view the first articulation may also, to a certain extent, constitute an ordered system. From our point of view, however, it does not. Under our definition of simultaneity, i.e. symmetry of relations, the monemes are grammatical analogues of distinctive features, not of phonemes. The grammatical analogues of phonemes are the pleremes (roughly, but only very so, 'words'). Cenological inventory

cenematics

cenotactics

Both cenematics & cenotactics

Plerematic Inventory

E.g. gestures

?

Morse

?

Morphology

E.g. traffic signs

?

?

?

E.g. some forms of 'classical' written Chinese*

?

?

E.g. modern written Chinese. Also music notation

E.g. most alphabetically written natural languages

E.g. natural languages; perhaps also written Korean

Syntax

E.g. number writing

Both morphology & syntax

E.g. picturewriting of natural language

* The case of classical written Chinese has to be reviewed, since it was pointed out correctly be Yasuko Obana that on a limited scale functional ordering is employed in the form of Chinese characters.

Further on the theory of semio tic systems, and the sub-systems of language

113

I shall reproduce here, with some adaptations, partly of a terminological kind, and omission of reference to what are now called para-tactic features, a scheme for my view of the two articulations, in its original form first appearing in my Oxford D.Phil, thesis 1966, and published in Sets and Relations in Phonology (Mulder 1968). Language 1st articulation

2nd articulation phonematics

phonotactics

distinctive features phonemes

morphology

syntax

monemes phonemes

phonotagms

pleremes

pleremes

syntagms

The broken line under 2nd and 1st articulation (rather misleadingly represented by a solid line in the original scheme) indicates that phonematics and morphology do not belong to the articulations proper, but are connected with them by the interlocks referred to earlier. One sees that the phonemes are molecular with regard to phonematics, but atomic with regard to phonotactics, and so are the pleremes (roughly: words) with regard to morphology and syntax respectively. This type of 'interlock' is represented by the fact that phonemes and pleremes figure in two columns each. In agreement with the earlier scheme for the systemology we have to add to this the level of para-phonotactics and the sentential level (para-syntax or para-plerotactics). We have also to add the two inventories of simple entities, i.e. 'distinctive features' and 'monemes'. It was not so necessary to do the latter in the earlier scheme, as it referred to systemology as one of the subtheories, rather than its application in a description. Here, however, we are concerned with the latter. From that point of view, i.e. the systemology of a language, we are dealing with a system of systems. In order to make a comparison of language (or any other

114

Function, system, and systemology

complex semiotic system) with simple semiotic systems at all possible, we need to represent simple systems within the scheme for language (or any other complex semiotic system). It also preserves the isomorphism with the classification of semiotic systems. Another reason is, as we shall see in Chapter V, that some simple phonological entities do not play a role in phonematics at all, i.e. phonemes such as /1 / and /h/ in English, which are not phonematically complex. In grammar there are even those, including 'yes', and in some languages 'sentential markers' (see Chapter VI), which only play a role at the sentential level. At any rate, as a system of systems, language can be said to have eight types of item, to which the sub-systems relate, i.e. they provide the justification for the division into sub-systems. They are phonologically simple, phonematically complex, phonotactically complex, para-phonotactic, grammatically simple, morphologically complex, syntactically complex, and sentential entities. The systemology of a semiotic system is, in fact, that part of the description which deals with all this, i.e. with the inventories of items in terms of their internal deployment. It does not deal with the semantic conventions of the semiotic system in question. It has the status of a model, onto which types of phenomena can be mapped, and by which they are classified. There is one other thing I should like to add before I present the scheme. It has sometimes been suggested that there is a difference between the para-syntactic level and the sentential level, as the terms 'para-syntactic features' and 'sentential features' do not mean the same. However, the former imply the latter, though admittedly not vice versa. That is, para-syntactic features are properly included in the class of sentential features. As 'parasyntactic features' implies that we are dealing with the sentential level, and also — even though there may be non-para-syntactic features as well — there is no sentential level unless there are para-syntactic features, the terms 'sentential level' and 'parasyntactic level' are equivalent by mutual implication. The scheme should then be as follows:

Further on the theory of semiotic systems, and the sub-systems of language

α> 43 Ο οο -Μ C α> ε Ί)

cd C ti

< 35 > o>

ε ω < «>Λ 13 +'ο-> cd ε εcd — ι00
IM w ccd ε w Ο ·Β C/1 ο cfXd a υWn ~αο υe αε> οC/3 cd (ι .ε ο 0) ·£2 1-4 0X1 Ο ο ο Τ3 C Ο "ο CX £ n ο ρ β 'S e < υ c d < υ >, •x >> ΪΛ Χι (Λ ta

ε νVi 3 ε α> Ο Ο ε cbd c.5 43 Ρ, ο 5 . CO ia ο

υ •c•->d Ο C Ο 43 Οη

ο ο c ο 43 α.

υ cd ε < υ C Ο 43 α,

•a

Λ υ« >> 'ob ϊ; ο οο -ιΟ-> 3ο •— ιΙ tc-4d

4> ε ω C Ο 43 α, 00

ω > 2 2 .ε 3 vi cd "σ "4-1 > Μ Τ3 a 2 .ε "Κ 3 C A 3 «Scd

^synta^naticentities syntagms

morphological analysis

pleremes > monemes

para-phonotactic analysis

complex

pleremes

para-phonotactic > other para-phonotactic entities entities (ultimately simple ones)

analysis of para-phonotactic entities into base and para-phonotactic features phonotactic analysis

phonotagms > phonemes

phonematic analysis

phonemes > distinctive features

It is appropriate to use the sign borrowed from mathematics, and meaning 'is greater than or equal to', for analysis into components. Only the type of analysis indicated by ' > ' is 'componential analysis', ie. analysis into immediate constituents. Of course, immediate constituent analysis does not exhaust the analysis, as there are also the relations between immediate constituents to be considered, especially in case these are ordering relations. Nevertheless the sign ' > ' seems to me quite acceptable, as long as one bears this in mind. Analysis into base and para-

118

Function, system, and

systemology

tactic features is analysis of a quite different kind, i.e. it is simply analytical description in terms of properties rather than constituents. A requirement for being considered constituents of a complex is being of similar status or nature. We do not generally say that a red vase is composed of a vase and its colour, rather that redness is a feature of the vase in question. This is not to say that one cannot abstract from its colour, and describe the vase with no reference to its colour. This is in fact all we do when considering the base in separation from its para-tactic features. The base, or elements of the base, can then be subjected to tactic (cenotactic or syntactic) analysis. It should be noted, however, that in the case of even well-formed sentences a base does not necessarily correspond to a syntagm, and nor is the whole, or are parts of it, necessarily syntactically well-formed. Sentential well-formedness and syntactic well-formedness are separate issues. By well-formedness I simply mean self-containedness, which — I should like to reiterate — means that it represents all relative dependencies of its constituents, as constituents of the combination (complex) in question. An example of a perfectly well-formed sentence, the base of which does not correspond to a syntagm, and which even contains a syntactically non-wellformed clause is: 'No, I don't think he is.'. There are here two clauses, of which the first, i.e. 'No', does not stand in any syntactic relation to the second, and nor is the second clause well-formed as a syntagm. Of course, with a difference in intonation one could have two separate sentences: 'No I don't think he is.' but it can also be realized as two clauses. If we were to analyze the second clause into immediate constituents (which actually we should not do) we have Ί / d o n ' t think/he is'. About Ί', which is syntactically simple, the question of well-formedness does, of course, not arise. The syntagm 'don't think' is a normal wellformed 'extended predicative syntagm' (see Chapter VI), but 'he is' consists of a subject and an incomplete 'copulative predicative syntagm' (see Chapter VI), and it does not, therefore, satisfy the condition of syntactic well-formedness. The rest of the scheme is fairly straightforward. Recursivity is built in at the syntactic level of analysis. Syntagms may be multiply complex, in which case they are further analysable into syntagms, and so on. Syntagms that do not contain further syntagms are directly analyzed into pleremes. I should at this

Further on the theory of semiotic systems, and the sub-systems of language

119

point explain why I prefer the term 'plereme' to 'word'. One advantage of using the term is that it reminds one of the fact that it is the grammatical analogue of ceneme (phoneme), i.e. it preserves the symmetry of terminology, and thereby 'reveals' to a greater extent the structure of the theory. In that respect the present theory follows the example of Hjelmslev, albeit on a more moderate scale. Terminology is, however, not the only reason, nor is it the prime and initial reason for more or less abandoning the term 'word' when dealing with signs, rather than their realizations. As a compromise between Uhlenbeck's insistence on the importance of the notion 'word', and Martinet's equally strong and, in fact, valid arguments for abandoning the notion, I had originally coined, as a grammatical analogue to 'phoneme', a term 'grammeme', now called 'grammateme' in order to avoid such dilemmas as to whether the 'words' contained in Martinet's example (1960:101-2) 'la grande montagne blanche' (/la gräd mötan bläS/) are Ί ' , 'grä', '. . . a . . . d . . . mötan . . . δ', and 'blä', or simply'la', 'gräd', 'mötän', and 'bläs'. Two insights, in fact already theorematic within the theory, led me to regard 'word' as a term that could be retained alongside 'grammateme'. The first of these insights was the fact that the 'discontinuous', i.e. 'grammateme', view, though functionally a correct one, does not reveal the fact that the actual form of this particular allomorph of 'montagne', i.e. ' . . . a . . . d . . . mötan . . . δ', in respect to its ' . . . a . . . d . . . δ' part is governed by the identity of the signs they are attached to, whereas the 'continuous', i.e. 'word' view, did not account for the fact that these attachments were governed by the identity (as to 'gender') of the sign 'montagne'. The two views are, therefore, complementary, i.e. one needs to say that 'a' belongs to 'la', 'd' to 'gräd', '§' to 'bläS', and yet ' . . . a . . . d . . . δ' belong to ' . . . a . . . d . . . m ö t a n ' . . . δ'. The second insight, equally theorematic — and surprisingly enough overlooked in the first instance — was the fact that these considerations are only relevant at the allomorphic, not at the structural (sign) level. For syntactic analysis it does not make any difference whether one takes the 'grammateme' or the 'word' view. On both views there are four minimum syntactic entities, and the syntactic relations between those entities are the same in both cases. Therefore, if one wished to retain both terms, a third term had to be coined, serving the dual purpose of being a

120

Function, system, and systemology

term for a type of grammatical entity, and being a cover term for both 'word' and 'grammateme'. I call a plereme a 'grammateme' when I specifically refer to it as a set of allomorphs at least one of which has a discontinuous, i.e. interrupted, phonological form, whereas I call a plereme a 'word' when I specifically refer to it as a set of allomorphs all of which have a continuous, i.e. uninterrupted, phonological form. Sometimes, where it does not theoretically matter very much and I use ordinary language rather than technical jargon, I use the term 'word' instead of plereme. I apologize for the use of the term 'discontinuous'. It is a more or less standard term for this type of phenomenon in linguistics, though the term 'non-continuous' may to some seem more appropriate. I come back to these issues in Chapter VI. As I fear that there may be readers who are mainly interested in the syntax of English, and may rather skip over the introductory chapters, I thought it advisable to reiterate some of these things there. Before I leave the topic of semiotic systems and the sub-systems of language, I should like to say something more about written languages. What I have in mind are alphabetically written languages, such as written French, written English, and so on, but much of the following may be relevant to other systems of writing conventions as well. As to the relation between speech and writingsystems there are, in my opinion, many misconceptions, most of them due to a lack of sufficiently distinguishing between synchronically and diachronically based arguments. We can, I think, in fact, even from a purely synchronic point of view — i.e. without any reference to origin of writing-systems, which is admittedly in a representation of speech — look upon writingsystems in two different ways. In the first place we can look upon them as systems of conventions in their own right, conventions which are markedly different from speech-conventions. There is no one-to-one correspondence between phonemes in natural languages and letters (graphemes) in the corresponding writingsystems. Also, and connected with this, rules for the distribution, succession and deployment of graphemes are markedly different from that of phonemes in the corresponding speech-conventions. Consequently, what may be homonyms in the one system are not necessarily homonyms in the other. In short, graphological systems are different, and often (e.g. in English and French) very different from spoken conventions. Differences in the

Further on the theory of semiotic systems, and the sub-systems of language

121

plerological plane may be less conspicuous, but they too exist. In many cases there even tend to be lexical differences but this may not be 'wholly fixed-conventional'. Under this view, say, written German, written Dutch, or written English are qua systems totally independent from their corresponding spoken conventions, and if subjected to semiotic description they have to be dealt with without any formal reference to their spoken counter-parts. We can, however, look upon materially the same systems in another way. That is, writing systems can be employed as a notational system for speech, just as musical notation can be used as a notational system for music, or the Morse code can be used as a notational system for written languages. But in that case the forms (written words and other conventions) denote mere sounds, i.e. sounds of spoken words, intonations, and so on. As written systems, actually, on the whole — though some more so than others — are not very well adapted to that purpose, the reader (i.e. the receiver) has to be reasonably well acquainted with both the written and the spoken conventions before he can interpret the messages conveyed by the written text. These messages would not then be — I should like to reiterate this — the meanings of the written sentences, but the mere sound-symbolizations which have the sounds of a spoken language as their denotational range. Only if we make use of written texts in their own right do we interpret the messages conveyed directly, but in that case the vocal 'reading' of the text is redundant and irrelevant. We must thus distinguish between two uses of what is materially largely the same system. From a semiotic point of view, these two uses amount to two different semiotic systems, systems with different purport, which is reflected in a difference in denotational range. Actually, the second use is of a similar nature to phonemic or phonetic representation, both of which are much better vehicles for the direct representation of speech. In the case of a phonemic representation, the symbols represent phonemes, i.e. functional entities in a given language, whereas, for instance, the signs of the International Phonetic Alphabet denote classes of conspicuously similar sounds. The latter are not language-specific, and purport to provide a direct representation of any speech sound. Both in phonemic or phonetic transcription of a text and in the use of writing-conventions as a representation of speech, we have to deal — just as in the Morse code — with two semiotic systems

122

Function, system, and

systemology

at a time. The signs of the one system correspond to the forms of the other. What has added to the confusion about the relation between speech and writing is the fact that often the two functions are used simultaneously. We read texts, and — depending on to what degree we are also acquainted with the corresponding spoken language — we tend in our mind, or even in actual fact, to vocalize what we read, and grasp the meaning of the written and the corresponding spoken sentences simultaneously, just as a skilled marconist may do when he receives a message in the Morse code. This does not, however, affect the formal necessity of recognizing here two separate semiotic systems. As to the fact that writingsystems generally have their origins in speech, this is, of course, a purely diachronic matter, and irrelevant to the issue. Many semiotic systems, including the Morse code, algebra, mathematics and systems of formal logic have, not surprisingly, to a certain extent a linguistic origin, but this does not affect their status of being independent of language from a synchronic point of view.

Chapter IV: Semiotic foundations of axiomatic functionalism - Ontological considerations and the 'linguistic sign' 1. The theory of indices In the earliest scheme for semiotic systems (see Chapter III, 3) we find a division of 'grammatical elements' into 'signs' on the one hand and 'symbols' on the other. This division was appropriate to the scheme, since - as one should remember — its sole purpose at that time was to classify some more familiar types of semiotic system, and ultimately to arrive at a comparison of those with human language. Though not represented in the scheme — because at the time it seemed to be of limited interest - it was also recognized that there were symbols of two kinds, i.e. those to which certain conventions are attached, which are, however, not wholly fixed, and those to which no fixed conventions are attached at all, such as occasional number or colour codes, and nonce-words. An example of the latter, used at that time, but in a slightly different context (Mulder 1968) was 'whilk', phonemically /huilk/, as in 'a whilk or so ago'. I claimed that noncewords in a natural language generally were phonemically wellformed, but they were void of meaning, unless one attached a definition to them or they acquired at least some 'grammatical' meaning when used in a context. The latter point is, however, not relevant to this discussion. Soon after the publication of Sets and Relations (1968), I had coined the term 'signum' as a coverterm for both 'signs' and 'symbols'. By about 1970, G. Soutar, at that time still a student, suggested the term 'nonce-symbol' for wholly non-conventional symbols. Other symbols were then referred to as 'proper symbols'. The term 'proper symbol' was reminiscent of 'proper name' which does, as I have said, belong to that category. Thus we had the following classificatory scheme

124

Semiotic foundations of axiomatic functionalism

for discrete plerological entities used in semiotic systems:

signs

symbols

proper symbols

nonce symbols

In Hervey's ideas about denotational semantics, the notions 'information' and 'index', the latter loosely defined as 'anything from which information is extracted', played an important role. It was, of course, recognized that signa were a specific and interesting type of indices, but Hervey realized that there were other types of index as well. Taking these considerations as our point of departure Hervey and I built a new scheme as the framework of what we now call the 'theory of indices' (Mulder — Hervey 1971, 1972). This sub-theory constituted, on the one hand, a link between the theory of semiotic systems, and the signum theory, and, on the other, between the latter and Hervey's theory of denotational semantics. In fact, it was regarded at that time, as part of the signum theory, then called 'sign-theory', but now we regard it as mainly a sub-theory of semantics. There are some minor flaws in the original presentations of the theory of indices, i.e. no clear-cut distinction is made between 'utterance of the index' and the index itself. It is, however, difficult to see how this could have been avoided without making the presentation too complicated. As anything from which information is extracted is considered to be an index, the notion 'index' is, in actual fact, linked to a single event, i.e. it is bound to a particular time and a particular place. Therefore, initially one should regard indices as utterances, i.e. models for a single realization, rather than as signa, i.e. classes of both formally and referentially equivalent utterances. When we consider conventional

The theory of indices

125

indices, however, the signum-view becomes more interesting than the utterance view. Therefore, in the original classification scheme, which I shall represent below, one should regard the right-hand side, with the exception of the category 'noncesymbols', as referring to classes of indices, rather than to single members (utterances) of such a class. The same holds for the class 'signalling devices' in the left-hand side of the scheme. Here follows the scheme as originally presented (Mulder — Hervey 1971, 1972). In the discussion of the scheme, I shall bring in the necessary correctives to the original presentation, without mentioning them specifically. The newer presentation, therefore, supersedes to some extent the previous ones, including the one immediately below. indices

natural

conventional (signa)

proper symbols

nonce symbols

Almost anything in the world can at one time or another be an index for something, i.e. information can be extracted from it. We consider something as an index therefore, if and when it is actually used to extract information. We consider something as a class of potential indices if the item in question is specifically constructed, set up or developed for the specific purpose of providing information. To the latter category belong signalling

126

Semiotic foundations of axiomatic

functionalism

devices and signa. They are objects or items with certain conventions attached to their usage (though not in all cases to their interpretation) and those conventions are such that the main purport of those objects or items is to provide information. The difference between signalling devices and signa is, as we shall see, in the fact that the former are only conventional qua 'object' or 'entity' and in respect of their use, but the relation between object or entity and the information extracted is a natural, in fact 'causal', one. As to signa, they are conventional both in respect of their use, and in respect of the relation between utterances and objects denoted. We shall, in the following, adopt the convention where a stands in a relation to b, i.e. aRb. or (in equivalent notation) R(a,£), of calling the whole of aRb (or R(a,b)) a relation, and R in this relation the relator. The two should not be confused.* It should be kept in mind, however, that even the form of signa may be of a 'natural', i.e. non-artifical or non-conventional kind, e.g. 'sunrise' can be used by an army as a signal for 'commence the attack' (see below). In that case the index in question is likely to belong to the category 'noncesymbols', though, of course, most nonce-symbols are not of that kind. Almost anything in the world can be employed as the form of a nonce-symbol, including forms borrowed from other semiotic systems, but let me first explain the ideas inherent in the scheme. If we have an index a, from which we infer the information b, we can say that between a and b there is a relator R, such that R can be read as 'conveys the information that . . . ', i.e. from a it is inferred that b. In the case of conventional indices and also of signalling devices we may say that from an instance (which we shall call an 'utterance') of a it is possible to infer b. All this comes intuitively very close to admitting that a implies b. Therefore, we have, in the earlier publications (Mulder — Hervey 1971, 1972) for reasons of abbreviatory conventions, rewritten *

This does not prevent m e from using the term 'relation' in accordance with ordinary usage in other contexts. It would be clumsy and tiresome t o do otherwise. The distinction between 'relation' and 'relator' is only maintained on those relatively rare occasions when I specifically refer to algebraic formulae of the type is in some conventions of formal logical notation used to indicate 'implies', or 'is a sufficient condition for inferring .. I have used it throughout this work, and in other publications, in this sense, unless indicated otherwise. When used with regard to indices, however, it should be understood that the 'implication' is not as strict as to be completely equivalent to logical implication. We should also still remember that in the case of conventional indices, with the exception of nonce-symbols, 'index' has to be understood as referring to a class of instances (utterances), rather than to a single member of such a class, whilst also the arrow, in accordance with this, has a slightly different significance. In the case of nonce-symbols, this rather awkward situation could be resolved by recognizing that nonce-symbols constitute classes with only one member. In the case of symptomatic indices it will, however, remain a serious dilemma. Let us first consider what we mean by 'symptomatic indices'. We can say that a certain kind of spots is a symptom of measles, that limping is a symptom of an injured leg, that a white tongue is a symptom of an upset stomach, that high temperature is a symptom of something going on in the body which we call fever, and as the latter is a symptom of ill health, a high temperature too, perhaps indirectly, is a symptom of ill health. From all of those symptoms one can infer information, and this is because of a natural relation of effect and cause, or the converse of this. Such symptoms are therefore indices, and because the relator between them and the information inferred is a natural one, i.e. there are no conventions involved in contracting the relation by recognizing a relator R that is not inherent in the nature of both a and b, we say that they are natural indices. And because of this, in the actual construction (as opposed to 'observation' with a thermometer) of such an index, we call it a symptomatic index. Compare this with the situation in conventional indices (i.e. signa), e.g. the word 'house' in English. There is nothing in the formal entity 'house' which obliges it to designate a house, nor is there anything in houses which predetermines them to be designated by a word having the form /haus/. This is, of course, what Saussure meant by 'L'arbitraire du signe\ Compare the situation in the case of symptomatic indices also with, say, the whistling of a kettle, indicating the water has come to the boil. Here too the relation

128

Semiotic foundations of axiomatic fitnctionalism

between the index and the information inferred is a natural one, depending as it does solely on the laws of nature, i.e. on a cause and effect relation. But in this case it is 'man' who has used the laws of nature in order to construct a device specifically for making the housewife know that the water she has put on the fire is now boiling. It would still be correct to say that the whistling of the kettle is ultimately a symptom of water boiling in a kettle which is fitted with a certain acoustic device, just as a certain kind of spots on the face of a child indicates measles. Just as in the first case the shape and the structure of the kettle, as well as that of the whistle, intervenes between index and information inferred, so also in the case of a child having measles, the structure of the child's body and its physiological functions intervene between the spots and diagnosis. Just as in the case of the whistling kettle, it is not necessary to retrace the whole process, nor even to know its details, but this does not diminish the fact that it is by virtue of a cause and effect relation (or its converse) that the diagnosis can be made. There is nothing conventional involved in the relator itself, i.e. one can not manipulate it by, for instance, stipulating either specifically or by changing custom that the same spots in children henceforth will indicate, say, appendicitis, or the whistling of that kind of whistle on that kind of kettle will from now on indicate that the water has become contaminated by some kind of germ. Of course, there may be, under different cultural conditions, a different information-value attached to the child's spots, e.g. one may take that type of spots to be an indication that the child has been visited by the devil, or one may use the whistling of the kettle as a sign for 'now it is time for setting the house on fire', or 'shooting the host', or 'kissing the hostess', or whatever one can imagine. The latter would amount to no more than using the same form of index in a quite different occasional semiotic system, where it acquires the status of a nonce-symbol, and thus, as an index, has no more in common with the whistling as an index for boiling water than, say b [bij] as the name of a letter of the alphabet has with, say, '(to) be', or 'bee' (both [bij] in the English spoken language). In the case of the interpretation of a certain type of spots as the 'devil's visit', there are at least two possibilities. One of the possibilities is that the illness 'measles', in another type of culture than ours, is called —

The theory of indices

129

and indeed believed to be — nothing else than merely having been visited by the devil. In that case there is no real difference from the usual interpretation by doctors in our type of society, in spite of the fact that there may be an element of superstition (but what is 'superstition'?) and, indeed, ignorance in other interpretations. Heaven alone knows how much 'superstition' and ignorance our own doctors possess, and the differences between the two interpretations is, objectively speaking, only a matter of degree, and it definitely does not affect the status of both as natural and, indeed, symptomatic indices. Slightly (and only trivially) different from this is the following. A certain illness, the existence of which is evident from the child's having a temperature, and its being clearly uncomfortable, is recognized. Because there are spots, this illness is understood to be of a particular kind. Up to this point, there is no difference from the usual diagnosis of measles, whether one gives a name to the illness in question or not. The illness, i.e. measles, is now regarded as evidence, i.e. as a symptom, of the fact that the devil has visited. We see here how an analogy with the earlier mentioned case in which high temperature is a symptom of having a fever and the latter a symptom of ill health. In the same way we can say that, for superstitious people who have such a belief, spots, which indicate measles are indirectly a symptom of the devil's visit. Moreover, the relation between a devil's visit and measles, or indirectly, between the former and spots, is a natural relation of at any rate believed or supposed (rather than 'real') cause and effect, however wrongheaded other people may suppose these beliefs or superstitions to be. It cannot (unless one's beliefs are changed) be manipulated by, say, substituting the archangel Gabriel for the devil. Also the fact that in some cases the interpretation is an indirect one, i.e. via an intermediate interpretation, is irrelevant; as long as the relation is contracted by an uninterrupted chain of supposed cause and effect, the index in question belongs to the category of 'natural index'. It is irrelevant whether or not the recognition of measles intervenes between spots and the 'diabolistic' interpretation except trivially in that one can recognize different, but interlocking, stages of index simultaneously. But this is the normal situation in the case of natural indices. One can perhaps always break up a chain of cause and effect into different stages. A doctor could, for instance, trace, step by

130

Semiotic foundations of axiomatic

functionalism

step, the physiological and chemical changes in the body which cause spots in the case of the illness called 'measles', and an engineer or scientist could, for instance, trace the whistling of the kettle from boiling, via the forming and accumulation of steam via pressure, and the acoustic shape of the whistle, and via airvibrations and co-vibrating enclosures, to the whistling sound affecting our eardrums, cochleas, and ultimately brains. It will now have become clear that many other man-made devices, such as a red or amber light glowing on the dashboard of a car, indicating that (usually in the case of the red light) the battery is not charging, or (perhaps in the case of an amber light) there is insufficient oil-pressure, belong to the category of natural indices, and, more specifically, are signalling devices. Other examples of this are the telephone bell, or the door-bell ringing, a bluish light glowing on the dashboard, indicating that the lights are on full beam, indicator lights, temperature-meters, thermometers, and so on, almost ad infinitum. In the case of, say, the dashboard devices it is irrelevant whether the chain of cause and effect is properly understood (as by a good mechanic) or whether one is mechanically unsophisticated so as to have no clue as to the workings of the device, so long as one is able to make the proper final diagnosis, i.e. there is something seriously wrong which must be remedied. Even the latter is, in the case of signalling devices as opposed to symptomatic indices that are not specifically constructed for the purpose — irrelevant. That is, signalling devices are there for a specific purpose, i.e. they belong to an inventory, perhaps only consisting of the activation of the device and its non-activated state, of indices. Therefore, they retain their status of index, even if they are not interpreted at all. For instance, dashboard devices, when operating, do signal information, even if the driver ignores this, and drives on, perhaps to his regret. The whistling kettle does signal that the water has come to the boil, even if the housewife is stone-deaf and also does not notice the conspicuous amount of steam escaping. The neighbours might hear it, give an interpretation, and warn the lady in question, but whether they do this or not does not diminish the fact that there is a device that is there for the purpose of conveying a certain information available to an interpreter and that does, in fact, convey that information when in an active state. And this brings us to the reason why in the scheme a distinction has

The theory of indices

131

been made between symptomatic indices and signalling devices, even though the latter are no less 'symptomatic'. In fact, one may say that the term 'symptomatic indices' is actually a misnomer, and should be read 'merely symptomatic indices', or, even better, 'non-contrived symptomatic indices'. I retain the term, however, as it seems less clumsy than the alternatives, and as long as one understands it in the correct way - i.e. as opposed to 'signalling devices' — it does not really matter what term is used. The reason, then, why the two types of index are distinguished in the scheme is that in the case of signalling devices we can look upon the index as a class of potential signals, i.e. of instances of that index — which in accordance with the terminology appropriate in the case of natural languages I shall call 'utterances' (of that index) — whilst in the case of symptomatic indices this cannot consistently be done. Even though by habit doctors, nurses, and mothers may interpret a certain kind of spots as measles, and say, farmers may be in the habit of looking at the sky and interpreting heavy clouds as a symptom of impending rain (to give another example of a symptomatic index), such phenomena are not specifically there in order to provide this kind of information. Though for the habitual interpreter the indices in question may be regarded as constituting a natural class of similar phenomena, i.e. a type, they do not constitute a class by intension other than the fact that they are observed as formally similar. It is true that for the habitual interpreter they are also referentially similar, but this is due to the habits of the interpreters, rather than to any semiotic, as opposed to purely natural, properties of the index as such. In fact, we may say that symptomatic indices have no inherent signalling properties, and that they are only indices if and when they are actually interpreted. Signalling devices, on the other hand, do, as the name suggest, have signalling properties, as they belong to systems (of perhaps only two members, see above) of artifacts especially constructed in order to provide information. There is a convention between the installer or constructor of the device and the user of the device with regard to the purport of the device, i.e. the type of message that utterances may carry. Though once the device is set up the relator between device and messages conveyed is natural, i.e. non-conventional or non-arbitrary, it may, perhaps seem that the indices themselves are, at least qua form, conventional, i.e.

132

Semiotic foundations of axiomatic

functionalism

arbitary, and they have therefore to be regarded as conventions. It may, therefore, seem to be the case that we should regard systems of signalling devices as semiotic systems. Before investigating whether such a view could be correct, let me first correct what was already a flaw in the above argument. It is not the index that is an artifact, unless we wish to say, that, for instance, the index is the aluminium whistle, or rather the total device: kettle with water (and whistle), heating system, and all, rather than the sound this device makes. This, of course, would be absurd. We can at best say that the device is the 'emitter' of the index (or rather of 'instances' of that index) just as a speaker is the emitter of utterances. Actually, in the case of signalling devices the constructor is helping nature, or even creating or rearranging a bit of nature. The only difference from symptomatic indices is, therefore, the fact that in the case of signalling devices there is signalling purport, albeit by natural means. But if this were not the case, we still could not regard signalling devices as semiotic systems. Semiotic systems are defined as systems of conventions for communication. But 'communication' in the sense I attach to that term implies that (a) there is a conveyance of messages, not just information that one can read off from the workings of a device, and (b) that there is an element of choice, i.e. communication is a wilful act, in which there is freedom as to what message is to be conveyed or whether any message is to be conveyed at all. Even though someone might say that the motorcar conveys a message to the driver — which is already an odd thing to say —, condition (b) is, of course, not satisfied, and therefore systems of signalling devices are not semiotic systems, i.e. systems of conventions for communication, even though they are 'artificial systems for indication'. Because they are purposive systems which are not, as in the case of systems of nonce-symbols, based on purely occasional bursts of creativity, it is possible and feasible to regard the indices in question as classes of their instantiations. Coming back to the point, if we consider some other clearly symptomatic indices for which there are no conspicuous habits of interpretation, it will become even more clear that symptomatic indices cannot be consistently regarded as constituting classes of utterances (i.e. as instances of indices). Though we do not generally call the following examples 'symptoms', such

The theory of indices

133

cases as a wet raincoat hanging in the hall, indicating that the owner is home and has just arrived, rabbit tracks in the snow, indicating that rabbits have visited the garden, lipstick on a cigarette end, from which a detective can make certain 'deductions', a red face, indicating that the person is embarrassed, a frightened face, indicating fear, etc. clearly belong to the same cause and effect type of interpretables as, say, heavy clouds and measles do. They are not specifically created for the purpose, and more importantly, they are only indices if, and at the moment that, they are actually interpreted. It would not be feasible to draw a clearcut borderline between such 'accidental' indices and the ones that seem more habitually interpreted in a particular way, such as symptoms of illness, of impending weather conditions, etc. It should not be overlooked that spots are observed as a matter of habit by doctors and mothers, and weather conditions as a matter of habit only by farmers, boat-owners, and other out-door people. But also, say, in the case of rabbit-tracks there may be a habit of observation and consequent interpretation by certain groups of people, such as keen gardeners. Detectives may be in the habit of looking out for cigarette remains, and some school-children in the habit of looking to see whether father's raincoat is there when they come home, and, who knows, some people, e.g. interrogators in a totalitarian state, may be in the habit of watching and interpreting their victims' facial expressions. It should then, be clear, that symptomatic indices can only on occasion be regarded as a class of utterances, because basically they are only index if and when they are utterances (i.e. utterance = index and vice versa), though there may be repetition and thereby habit-forming. The most important point is, however, that, unlike signalling devices, they cannot be regarded as belonging to a 'system' of indications. They would belong to a 'system' if they were at least opposed to their absence (nonactivation), but this can, of course, not be the case, as absence would imply 'no utterance', i.e. 'no index'. As any system by definition must contain, at least, two members, of which one could be 'significant absence' (i.e. 'absence' as an index in its own right), symptomatic indices cannot be said to belong to a system of indices. Also, 'system' is defined (Mulder, Postulates) as 'self-contained set of features with a common purport', and in the case of symptomatic indices, indication is not their purport, but merely a contingent function.

134

Semiotic foundations of axiomatic

functionalism

With regard to the right-hand branches of the original scheme, some alterations and extensions have to be introduced, in accordance with points raised in the previous chapter. It was said in that section that 'classificatory devices', such as systems for library classification, signs for 'gentlemen' and 'ladies' on doors of public conveniences, etc. are to be distinguished from semiotic systems. Even the titles of books and chapter or section headings can be regarded as belonging to this type, and so do, say, different colours of towels in the bathroom for different members of the family, or even (though this is perhaps a borderline case) a pink outfit for a baby girl, and a blue one for a baby boy. It is clear that classificatory devices of one sort or another are all around us all day and every day. Proper names belong to this category, at least from the point of view of (i.e. in their capacity of) indicating certain persons. From the point of view of (i.e. in a capacity of) communicating about, or addressing persons, however, proper names belong to the category of proper symbols. I should like to stress again that the identity of entities depends on the capacity in which they are considered. In the next section, when we are dealing with the signum-theory, we shall, for instance, see that we must distinguish between, say, /pit/, a succession of phonemes, and '/pit/', the same sequence but in its capacity of being the phonological form of the word 'pit'. In the first capacity it is a phonological form, in the second it is an allomorph. But before considering the identity of an item within a system we must consider its identity as a member of a system, i.e. we must ascertain to what system it belongs. As we have seen, items of similar form can be used in different systems, and systems borrow forms from other systems. Borrowing changes the identity of those forms, a type of identity which we call systemic identity. The systemic identity of an index depends on its deployment at the 'utterance' level. Since for those indices that are considered as classes of utterances the identity of the index depends on the intension and extension of the class the utterances belong to, the indentity of the latter wholly determines the identity of the indices in question. Systemic identity takes logical precedence over other kinds of identity. That is, even two formally identical indices, when belonging to different systems, correspond to different indices. For instance, no two different languages can have 'pleremes' or

The theory of indices

135

'monemes', etc. in common. These are basic tenets of European structuralism. The same principle holds for any semiotic entity. No language can have, say, phonemes in common with another language. There are actually two reasons for this, reasons which are, however, equivalent by mutual implication. The first reason, already mentioned, is that the identity of an item depends on its belonging to a particular system. The second is that the identity of an item belongs to its 'function' within the system. This, in its turn, is determined by the set of oppositions into which it enters (the 'functional principle'), and also by the use to which it is put. If it were the case that in two systems all the functions were the same, which implies that the systems contain exactly the same items, the two systems would not be different systems. And, according to the 'functional principle', as long as there is one function, or one item, different in a system, all the other functions must be different. As Saussure has said: 'une langue est un syst^me ού tout se tient'. The assumption of the apparent logical possibility that systems could have complete sub-systems in common is also refuted, if we consider the interlock between sub-systems. It is clear that, via this interlock, even sub-systems that apparently contain the same items can only be said to be to some extent formally similar, but to no extent identical. Before returning to the discussion of a typology of indices, let me first give the new classificatory scheme for conventional indices. This replaces the right-hand branching of the earlier scheme. I introduce here a distinction between 'signa' i.e. conventional indices in semiotic systems, and 'labels', i.e. conventional indices as classificatory devices. As the further branching of 'signa' and 'labels' is largely the same, I shall, in the following, first deal with the former. Most of what I am going to say is relevant to 'labels' as well. Labels are distinguished from signa in that the former are characterized by having classificatory, and the latter by having communicative purport. Both have what one could call 'indicative purport'. As we have seen, among natural indices it is only signalling devices that have indicative purport. We shall see below that nonce-symbols and nonce-symbol-like 'labels' do not have inherent indicative purport, but have purport attached to them by occasional definition, which, at the same time endows them with occasional information value.

136

Semiotic foundations of axiomatic

functionalism

The theory of indices

137

Examples of 'signs', which are wholly-fixed-conventional, are 'numerals' as in number-writing and arithmetic. In natural languages most of the indices e.g. monemes, pleremes, syntagms, clauses and sentences are signs. Compare, say, 'numerals' with, for instance, a, b, c, x, y, z, n, i, etc. as used in algebra, or P, Q, R, S, etc. in systems of symbolic logic. The latter types of item are not wholly-fixed-conventional, but require separate definitions before they can be used actually to denote something specific. The same is true for 'proper names' such as John, Mary, etc., as used in language-utterances. Algebraic and logical symbols have in common with proper names the fact that, though requiring separate definition, there are some 'semantic' conventions attached to their usage. The symbols a, b, c, d, are not generally used in the same way as x, y, z, and n, i, etc. are not used in the same way as either. Using them in the same way would constitute a violation of the conventions for algebraic notation. The same goes for logic-symbols and proper names. The latter are really symbols, not signs, as a John, or a Mary, is not a member of a particular well-defined class of denotables, except for the former when it is written with a small 'j' and refers to the smallest utility room in the house when it serves a similar purpose. In that case it is not the same item, as it belongs to a different system, which happens to be a system of signs, not of symbols. The systems of algebra and symbolic logic of course contain signa with wholly fixed 'semantic' conventions as well, e.g. +, — , = , < , 00 , for the former, and V, A, ->, etc. for some of the latter. In the case of algebra this may seem to constitute a refutation of the identity principle earlier stated, i.e. that no two semiotic systems may have entities in common. It may seem that arithmetic and algebra have sets of symbols with functions in common, as some of them may seem to belong to the same sets of entities in direct opposition, and moreover they often have the same deployment at the 'utterance' level, i.e. when the symbols have been given a denotational interpretation. Quite apart from the fact that the former is hardly the case on closer scrutiny, it is safer to assume a different identity according to which of the two systems they are considered to belong to. One can, of course, after definition, replace all the symbols of algebra by signs of arithmetic, in which case an arithmetical notation could be considered as a defined, i.e. interpreted, algebraic notation, and the

138

Semiotic foundations of axiomatic

functionalism

two systems could then be regarded as, for all practical purposes, isomorphic. In that case, it remains to be seen whether this constituted ordinary arithmetical notation, as certain arithmetical conventions would have been surpassed (perhaps even bypassed) and thereby violated. But these considerations lead us to another typical feature of symbols. Once a symbol has been defined, and is deployed at the utterance-level, it assumes a status indistinguishable from that of an utterance of a sign, i.e. it denotes a single member of a particular class of denotables. As, however, denotation (to be distinguished from 'an act of denoting', which is a function of an utterance) is defined as a correspondence relation of an index with a particular class of denotables, the symbol in question cannot be said to have 'denotation'. Even if a symbol has not been defined, as an utterance it denotes something, e.g. in the case of a in algebra it denotes 'anything that can be denoted by a\ As anything that can on one occasion be denoted by a, can on some other occasion be denoted by b, etc., we have with regard to undefined usages of a, b, c, etc. at the utterance level cases of, what, for lack of a better term, one could call utterancesynonymy. In fact, one can even regard a, b, and c, or x, y, and z, in algebra to be synonymous symbols. The same goes for, say, John, Peter, and Paul, or Mary, Angela, and Dolores. This of course constitutes a dilemma, as synonyms must denote members of the same sets of denotables at the utterance level. This dilemma can be resolved by considering defined, i.e. interpreted algebra (in notation) and undefined algebra as different systems. This is not just a devious way of dealing with the problem, as in interpreted algebra numbers too are used, whilst in undefined algebra this is not the case. The above 'synonymy' interpretations only hold for undefined algebra, i.e. algebra in its undefined state. Another, perhaps much simpler, way of escaping from this dilemma — suggested to me by Hervey in a personal discussion — is to say that the extensions of the class of denotables of, say, John, Peter, etc., or a, b, c, etc., as in algebra, are different, but their conventionally determined intensions are the same. That is to say, one could recognize a discrepancy between intension (as determined by systemic conventions) and extension as being typical of proper symbols. But it would have to be investigated how this would affect the logic of intensions and extensions of classes, and whether this would not create an even greater dilemma

The theory of indices

139

within the logic of classes. I should like to reserve this solution as a possibility without, however, committing myself to it. One can, of course, have mixed systems of notation — in fact the above example of interpreted algebra is an example of such a mixture with borrowings from arithmetical notation. But as most of such mixtures occur at the utterance-level rather than affecting the semiotic system as such, there is little interest in regarding the utterances as corresponding to separate, perhaps occasional or idiolectic systems. Let us take two examples of mixtures that are not infrequent in utterances of both spoken and written languages. We can, in a discussion of, say, mathematical problems, insert 'spoken' symbols or even whole sentences, of ordinary algebra. The same is, of course, true in respect of borrowings from formal logic, and in respect of written algebraic insertions in written texts. Such insertions partake in the syntactic and para-syntactic (and to a certain extent also in the phonotactic and para-phonotactic) relations of the borrowing systems. The borrowing system in question can be analysed in the ordinary way, e.g. a sentence such as O n e can work out that (a—b)7 equals a 7 - 7 a 6 b + 21a 5 b 2 35a 4 b 3 , etc. by using the Pascal triangle'. In such an utterance, the algebraic formulae correspond to the 'subject' and part of the 'direct object' of the predicative 'equals' respectively. However, at least in the case of ( a - b ) 7 , a further linguistic analysis becomes difficult, if not impossible, unless one includes the structure of algebra in the structure of the language in question, a rather redundant procedure, as the system of algebra can be, or is already, analyzed as a system in its own right. Moreover, one would have to do this for any wholesale borrowing of foreign structures, including logical notation and a host of other things, but also — and this brings us to my second example — quotes from foreign languages. What I have said about the above utterances of English, with borrowings from algebra, holds in exactly the same way for, say, a sentence such as O n e of Saussure's main tenets is 'l'arbitraire du signe". If we were to extend the analysis to the internal structure of the quote 'l'arbitraire du signe', we would have to include the structure of French, and of any language in the world (as one can quote from any language) in the structure of English. This would amount to a really 'universal grammar', different from what some linguists are after in that

140

Semiotic foundations of axiomatic functionalism

it would include every structural detail of every possible language. But this would, of course, not be feasible, and nor would it be very useful. We may then conclude that quotes and borrowings from foreign languages as well as from non-linguistic semiotic systems have to be regarded as simple, i.e. unanalysable, from the point of view of the system which borrows them. It is not only that their form is alien, but also as signs or symbols they do not belong to the inventory, and therefore do not have a semantic interpretation within the borrowing system. That they do denote, when they are employed as part of a sentence-utterance of the borrowing system, depends on their denotation within the original system, or on what the borrowers believe to be their denotation within that system. Even in ordinary speech there are many occurrences which could better be regarded as items belonging to separate, and what could perhaps be considered as either auxiliary or relatively independent, sub-systems. Examples are designations such as 'the Virgin Mary', 'Pope John Paul', 'King Edward', 'Doctor Johnston'. Other examples include names of numbers, e.g. 'three thousand four hundred and twenty one'. Such items have their own grammatical structure, not corresponding to structures elsewhere to be found in the language. There are here two possibilities of dealing with those cases. If we regard them as not belonging to language we should regard them as simple, i.e. Gust as quotes and borrowings) as unanalyzable. We can also, however, (and this seems to me the better solution to adopt) regard them as fairly independent sub-systems of the language in question. They should, in that case, be analyzed within their own system, but their structure should be further incorporated into the representation of the whole structure in which they are constituents. See Chapters VI and VII for further discussion and examples. So far, the division of signa into signs and symbols, and the division of the former into nonce-symbols and proper symbols, is completely analogous to a division we can make under the heading of labels. The same is true for the division of signs into iconic and abstract, which is, in fact not a very important distinction, as it is purely based on considerations with regard to the form of the index in question. Furthermore, iconicity is frequently a matter of degree, and therefore not really decidable.

The theory of indices

141

Yet there are certain interesting things that may emerge from a discussion of the distinction. I shall come to the further division of nonce-symbols into natural phenomena, borrowings, and occasional creations, and a division of the latter into explicitly and contextually defined, later. It too is based on purely 'formal' considerations, but is added, as it may lead to certain important insights. Examples of nonce-symbol-like labels are occasional colour or number-codes, say, in a filing system. We can also have classification-systems, again, for instance, in filing, in which, say, the most general classification uses coloured cards, which are then further distinguished by number, and under the latter by adding a, b, c, etc. Though the whole of the system is arbitrary, once established there are different conventions for different symbols. Such a system would constitute a system of proper symbol-like classificatory indices. I assume here that in filing-systems there is some kind of tacit convention which gives colours precedence over numbers. Even if I am wrong here and this is not the case, (though I believe it is) numbers tend to take a precedence over letters, i.e. a 1 , a 2 , a 3 , etc. rather than l a , l b , l c , etc. would seem to violate certain tacit conventions. It may seem paradoxical that, after definition of symbols, fixed conventions suddenly come into existence, but we must not forget that all symbols after definition require, at least at the utterance-level, the same status as utterances of signs, i.e. they actually classify, or in the case of semiotic systems, they denote. We may in fact say that, at the utterance level, the distinction between sign and symbol or sign-like and symbol-like labels disappears. The gap between the two is bridged by attaching definitions to symbols before they are put to use. These definitions only hold for the duration of that particular use. The difference between proper symbols and nonce-symbols is then that in the case of the former there are already certain conventions with regard to their potential use. These conventions are of two kinds, the second implying the first. The first is a convention as to the denotable purport of the whole system, e.g. algebra purports to deal with denumerable quantities, proper names purport to indicate people, domestic animals, etc. The second kind of convention, not necessarily appertaining to all systems of proper symbols, is that of dividing their inventory into classes

142

Semiotic foundations of axiomatic

functionalism

according to specialization in usage. For instance, in algebra, a, b, c, etc. are used in a different way from χ, y, z, etc. and John, Peter, Paul, etc. are used differently from Mary, Ann, Lily, etc. A clear example of a system of proper symbol-like labels is the peculiar though conventional way in which numbers are often used in order to distinguish sections within books or articles. A complex index such as 137.11, in that case, does not signify 'one hundred and thirty-seven and eleven hundredths' but rather a general division numbered one, under which there is a less general division numbered three, under which there is a division numbered seven, under which there is the finest division numbered 11. This should be distinguished from certain systems of librarybook classification, for instance the Dewey system, which uses exactly such a classification, but as this is a wholly fixed conventional system, i.e. the indices inherently correspond to certain classes of book, the digits in question belong to the category of sign-like, rather than symbol-like classificatory indices. I should now like to say something about the division, within signs, between iconic or denotation-motivated, and abstract signs. In fact, it may seem that one can divide the latter further into systemically motivated and entirely non-motivated signs, but the truth is that perhaps practically all signs are systemically motivated at least insofar as their forms have to comply with what is permitted within the system. This holds for derivations and other complex signs as well as for the forms of simple signs. They are composed of elements already present in the system according to the rules laid down by the system. It should also be kept in mind that the rules, and indeed the forms themselves, are only motivated in the sense that they contribute to the efficiency of communication, i.e. one should say that all signs are systemically motivated, or say that none of them are, except in a trivial and necessary sense. Consequently, I call an abstract signum a signum whose ultimate constituents are abstract. By abstract I mean here that there is no plausible connection between the form of the sign in question and its correspondence with a particular class of denotables. Even in the case of so-called soundsymbolism (e.g. words beginning with /si. . . /, in English) any assumed connection remains vague and cannot be said to be an obviously plausible one. But there could be a borderline case, for instance in the case of clearly onomatopoeic forms. In Malay,

The theory of indices

143

for instance, there are two kinds of crisps for eating, i.e. 'krupuk' and 'kripik', the latter of different substance and harder, consequently making a noise of higher pitch than 'krupuk' does when biting it. In Javanese (information from Uhlenbeck), there is a whole range of words denoting the making of crackling sounds of different pitches, 'makropok', 'makrupuk', 'makrepek', 'makripik', and perhaps others. Onomatopoeic words of that kind may perhaps be regarded as denotationally motivated. I have not made a distinction between iconic and denotationally motivated signs or sign-like classificatory indices, as I consider this distinction — though perhaps interesting from an artistic or cultural point of view — unimportant in respect of communication or classification. It is, for those who can read the relevant indications, unimportant whether there is a little drawing of a man or a woman on a toilet door, or whether it carries the printed words 'gentlemen', 'ladies', or (in German areas) 'Herren', or 'Damen', or even Ή ' or 'D'. It should be noted that, say, 'gentlemen', though qua form borrowed from English, is no longer an item of the English written language, denoting pluralities of respectable male humans. Even as a classificatory device it is not a label for such a class of denotables. This will become immediately obvious when one opens the door and enters the place. There are not always gentlemen there. But even so, there is a certain connection with the denotation of the entity before borrowing. It is in this respect more appropriate a sign or rather label for a particular man's place than the written form 'ladies' would be, just as it would be inappropriate to put a drawing of an old man with a beard as a classificatory label on a children's toilet, or write the word 'synagogue' on a Protestant church or a mosque. In natural language the word 'blackbird' can be regarded as belonging to this category, though certain reservations have to be made here. From one point of view, there is no connection between the form of the word and its denotation, i.e. it is an abstract sign. One could, however, also consider that part of the form is reminiscent of the word 'black', without however corresponding to a moneme. The word 'blackbird' is denotationally appropriate, as blackbirds tend to be black. One may, however, have albino blackbirds and blackbirds painted green remain blackbirds. Moreover, there are many black birds that are not blackbirds. One cannot, therefore, equate the /blak . . . /part of the form with

144

Semiotic foundations of axiomatic

functionalism

/blak/ in its capacity of being the phonological form of the word 'black'. In fact we cannot, as I have said, assign sign-status to the 'black' part of the form at all, which implies that we cannot assign sign-status to the 'bird' part either (cf. Hervey - Mulder 1973). As I have said, nothing can be part of itself, i.e. a simple sign cannot include, as a part, a simple sign. So we may say that '/blak/' ('black . . . ') here is merely reminiscent of the word 'black' just as 'gentlemen' on a toilet door is reminiscent of the word 'gentlemen' as used in English, or a little drawing of a male person on a toilet door is reminiscent of featherless male bipeds. Regarding a sign to be denotationally motivated rather than completely abstract enables us to explain the obvious connection there is between the word-form and its denotation, and yet to recognize that the 'black . . . ' part of the form has no separate denotational status. In the particular case of 'blackbird' and other pseudo-composites (see Chapter V) that are nouns, this interpretation is made even easier by the fact that we may regard them first as denotationally motivated sign-like labels, which are then again used in speech or written text, just as may be the case for written classificatory labels. The sentence Ί am going to the "ladies" ' is not usually understood as referring to a visit to some feminist gathering. But there are some differences as well, and cases of the 'blackbird' type do constitute borderline cases for which either of the two interpretations hold. This is reflected in the fact that intuitively one would be inclined to write 'ladies' in the above example between inverted commas, whereas one would not give the same treatment to 'blackbird'. But this may, of course, be due to the fact that there is a conspicuous discrepancy between the denotational range of 'ladies' as a word in the English language, and the indicational field of 'ladies' as a label for certain establishments. Be that as it may, I recognize the fact that either interpretation — i.e. that of considering 'blackbird' as abstract, or considering it as denotationmotivated — is appropriate, but the latter interpretation adds explanation to the fact that, say, the form of the word 'butterfly' (the possible diachronic origin of 'flutterby' is irrelevant here) gives no clue as to its denotational range, nor does, say, 'squirrel', whereas such forms of words as 'blackbird', 'whitewash', etc. do, in spite of the fact that they cannot be analyzed into meaningful components. In those cases where forms are complex at the

The theory of indices

145

sign level, however, such an explanation is neither necessary, (in fact it would be redundant) nor profitable, as in that case all complex signs would have to be considered as being motivated in that way (cf. Saussure's 'grammatically motivated signs'). Crucial for the distinction I am advocating is to consider whether a form is systemically motivated in all respects, i.e. whether it merely complies with rules of the systemology, or whether the form of a particular sign bears an obvious relation to the denotation of that sign, in spite of the fact that signs cannot be consistently analyzed into meaning-bearing constituents. In the latter case it is not different in principle from such forms as 'makrupuk' etc. in Javanese, or drawings or written labels on toilet doors. I have said that nonce-symbols correspond to index utterances rather than to indices as classes of utterances, but this is not completely correct. Also nonce-symbols can be 'codified' or defined for a certain length of time, in which case the indices in question become classes. But it is correct to say that their use is an occasional one. Similar considerations obtain with regard to symptomatic indices. Of course, one cannot have a really conventional system of nonce-symbols, i.e. a set that is codified for relatively permanent use. In that case it would be a system of proper symbols. A necessary condition for a semiotic system to be a system of proper symbols is that regular conventions are attached to their use. This does not necessarily imply that different conventions are attached to sub-sets of symbols within the system as a whole. As the latter implies the former, however, the latter is a sufficient condition. The division, under 'nonce-symbols', into natural phenomena, forms borrowed from other systems, and occasional creations, is based on origin, i.e. it is strictly speaking a diachronic matter and therefore irrelevant in an approach like ours, which purports to arrive at a description of the speech-phenomena in a structuralist synchronic fashion. I have, however, made this division for didactic purposes. From experience with students I have learned that the notion 'nonce-symbols' tends to leave question marks at exactly those points. If a soldier waves a white flag, indicating that he wants to surrender, he is using a wholly fixed conventional way of indicating this, and the index in question is a sign. If a general arbitrarily decides that hoisting a white flag on the mast

146

Semiotic foundations

of axiomatic

functionalism

indicates 'attack' he is creating a nonce-symbol, as there are no such conventions. The nonce-symbol in question is an occasional creation. He might, however, also have decided to use a natural phenomenon, e.g. sun-rise, for the indication that it is time to attack. There is no difference in principle. Or he might have borrowed a form from another system, e.g. the word 'overlord', called over the radio. There is again no difference in principle. Occasional creations include such forms as 'brillig', 'slithy', 'Jabberwock', 'vorpal', 'chortled' (all in Lewis Carroll, Through the Looking Glass), 'whilk', as in 'a whilk or so ago' (example used in Sets and Relations, [Mulder 1968]), and 'whilk' defined as a 'white elk with a missing eye' (Mulder — Hervey 1973). The latter is an explicitly defined occasional creation. Examples of these are, of course, scarce as they do not easily find their way into literature, and if they do, they may easily become conventionalized and change their status to that of signs. All the other examples share the property of being contextually allusive. The Lewis Carroll examples occur in the famous poem 'Jabberwocky', a poem without defined meaning but full of contextual connotation. The form 'slithy' definitely exhibits sound-symbolism in addition to being contextually allusive and also in the other forms there may be traces of sound-symbolism, albeit that this is difficult to assess, as sound-symbolism is not a very precise phenomenon, and is largely subjective. The form 'whilk' as in 'a whilk or so ago', is clearly allusive as to some kind of unit of time, because of the context in which it is placed. One could, of course, extend this scheme much further. It is quite possible that nonce-symbols the forms of which are borrowed from other systems, can also be further divided into those that are explicitly defined, those that are contextually allusive, and those that exhibit sound-symbolism. It is even possible that we could establish a category of nonce-symbols that are allusive by virtue of the very fact that their origin is known to the interpreter. It is not unlikely that within the category of labels we could make similar divisions, though it is difficult to see how natural phenomena could be employed as conventional classificatory labels. But the above scheme does systematize a sufficient variety of conventional, i.e. arbitrary, indices to include the types most frequently encountered in ordinary life. Borderlines between some of these types are difficult to draw, especially

The theory of indices

147

because a particular index may belong to more than one category at the same time. We have already seen examples of this in the 'Jabberwocky' poem. Another distinction that people might wish to make is between indices of a system that is the product of historical evolution and change, and is still changing, and the indices of a system — often called a code — in which the rules have been explicitly laid down. Notorious examples of the former are ordinary natural languages. In such systems rules have come about, and are still coming about, by adaptation to communicative needs, contact with fellow-beings, intermingling of different speech-communities and so on. These rules affect the indices as much as the system as a whole. But even spoken and written languages can be, to some extent, codified, even though gradual changes are bound to happen, and the codification, i.e. the prescriptive rules for proper usage, will have to be revised. We may, perhaps, think of systems, wilfully created by man, for specific purposes, as being codified for now and for ever. But these too are subject to change. For instance, even in closely similar notational systems for formal logic there are different 'dialects' and systems are changing all the time. I do not, therefore, regard it as profitable to distinguish between 'codes' on the one hand, and systems in which there is only a 'tacit' agreement as to what belongs properly to the system, and what should be considered a violation of the rules. I find even less profitable another distinction sometimes made, i.e. between languages on the one hand — which are all-embracing and incorporating — and man-made systems for special uses. It is said that the latter can always be translated into language, but not vice-versa. But this is only to say that languages are infinitely richer than these other systems art, and that they are the vehicles for all possible intellectual thoughts. It is very possible, and even likely, but this is pure speculation, that the limits of intellectual thought and those of our command of language are exactly coincidental. If this is the case, then artifical semiotic systems that cannot be translated into language are most probably a priori ruled out in principle, as it is unlikely that we would create what lies beyond our powers of thought. Though of course this alone fully justifies regarding language as very special among semiotic systems, it is not a distinction on which we can base a comparative classification of semiotic systems.

148

Semiotic foundations of axiomatic functionalism

The theory of indices also introduces the notion 'informationvalue', which is akin to the notion 'denotation', except that the former is a more general term, and applies to any index, i.e. to natural as well as conventional indices, to labels as well as to signs, and to signs or sign-like labels as well as to symbols or symbollike labels. Denotation, however, as defined by the present theory, is a property of signs only. Indeed, signs can, tautologically, be defined as 'indices possessing the property of denotation'. We can, in fact, define 'denotation' as 'the wholly fixed-conventional information value of an index', i.e. we can equate 'wholly fixedconventional information value' with 'denotation', because this implies that the index is not only a signum, but a sign. This interpretation of 'denotation' is equivalent to Hervey's notion, which, freely paraphrased, is defined as 'the relation of correspondence with a particular self-contained class of denotables'. I have defined 'information-value' as 'the specific set of potential interpretations'. The term 'specific' is meant here to rule out farfetched, completely idiosyncratic, interpretations, as otherwise for any item this set would not only be infinite (there is no objection to this), but also unlimited. This means, in fact, that even for natural indices I expect that there are certain natural or cultural restrictions as to the interpretations that can be attached to them. The fact that these restrictions may be culturally determined (e.g. witch-doctors' interpretations) creates a dilemma, which can, however, be resolved if we consider that from the point of view of the cultural community such interpretations are, as I have said, a matter of effect and cause, i.e. they are based on an interpretation of nature, however misconceived. As I have said for indices, 'wholly fixed-conventional information-value' can be equated with 'denotation', but it would be wrong, even for signs, to equate 'denotation' merely with 'information value', as it may seem that even utterances of signs can provide information of a non-denotative kind. The sign 'cop' for a policeman seems regularly to indicate dislike but not necessarily so, as one can speak about 'a good cop' (Mulder and Hervey 1980:214). The sign 'fressen' (to eat, gobble) in German, can indicate a disapproval of table manners or eating habits, as it is rude to use it when referring to eating by humans, but an angry father might say to a child: 'Du sollst es fressen, sonst kriegst du auf den Hintern!' without implying that he

The theory of indices

149

wishes the child to 'gobble'. However, precisely because such a correspondence is not wholly fixed-conventional, that is to say implications are not included in the specific class of denotables concerned, we have to see 'connotation' as pertaining to the utterance level, rather than to the 'class of utterances', i.e. 'sign' level. We may, then, say that 'denotation' is the informationvalue of a sign as such. Any utterance of a sign 'denotes' something specific in the external world, but by this 'denoting' we only mean those aspects that derive from its being an utterance of a specific sign, i.e. 'to denote' only refers to that part of the information provided which is wholly fixed-conventional. Anything else belongs to the realm of 'connotation', a cover-term for anything not wholly conventional even though perhaps alluded to by context or imprecise conventions. Looking back at what we have said about the use of nonce-symbols of the allusive kind, we can now say that their information value only comes into its own at the utterance-stage, and that they do not denote, but only connote. Even if there are several realizations of the same nonce-symbol, every separate interpretation may well be — and is even likely to be — slightly different. I should like to put this more strongly and claim that every time one reads or listens to the same nonce-poem it is not unlikely that one extracts a different message from it. Therefore we cannot consistently regard 'connotation' (i.e. non-denotational features) as having any relevance to the 'sign', or even 'index as a class' level. I should like here to represent a scheme which may clarify what is involved in denotation and the act of denoting (cf. Mulder — Hervey 1972). As I have said, a sign is ultimately a self-contained class of utterances, any of them being an instance of that sign. Every single utterance denotes a denotable entity, relation, situation, or whatever, which is itself a member of a self-contained class of denotables, the intension of that class deriving from the fact that any of its members, one at a time, can be denoted by an utterance of that sign. This is what is meant by saying that the sign corresponds to that particular class. It is this relation of correspondence between a sign and a self-contained class of denotables that is called 'denotation'. If there is a relation of correspondence between more than one sign and the same class of denotables, those signs are synonyms in respect of one another.

150

Semiotic foundations of axiomatic

functionalism

It is not possible for the same sign to correspond to more than one class of denotables, unless these classes are sub-classes of a class to which that sign bears a relation of correspondence. In all other cases where it may seem that non-intersecting classes correspond to one sign, we recognize 'homonymy', i.e. difference in sign-identity. A sign does, of course, bear a relation of correspondence to every single member of its corresponding class or classes of denotables, if only because an utterance of that sign may denote a member of such a class.

Hervey has defined 'utterance' as ''model for a single realization (in actual communication) of a signum'. Mulder and Hervey 1980:203. The initial reason for this was that a signum cannot, of course, be a class of communication-events, as this would be absurd for ontological reasons. But we can regard it as the formal record of a communication event, such that the event is not only protocolized, but also considered as a member of such a signum. In fact, in the exceptional case of puns and in quite a lot of poetry or rarified prose at sentential level, this may result in recognizing two utterances corresponding to a single realization. In those cases it is no longer an event, but it has been 'arrested' and so has assumed the status of a 'model'. It allows us to deal with utterances that belong to the past, and we may even consider potential future utterances. Therefore 'utterance' cannot

The ontology

or 'signum-theory'

151

ontologically be conceived of as belonging to exactly the same level of abstraction as 'protocols' and 'facts', though it is nearest to this level. Remember that I have used the term 'fact' in the sense of 'protocolized event'. Remember too that protocolization is a function of observation, and is not necessarily explicit. We shall see in the next section that 'images', which are models for single realizations of a speech-sound, ontologically speaking can also be said to belong to the level that is nearest to that of protocols, albeit 'images' as such are ontologically different from 'utterances' as such, because in the case of utterances distinctive function and 'meaning' too contribute to their identity. In order to satisfy considerations of consistency, we have, therefore, to distinguish two ontologically different types of protocol, i.e. one in which we abstract from meaning and distinctive function — and are therefore able to consider even those formal entities which by themselves are incapable of being connected to meaning — and another one at which we consider entities in their capacity of having both 'form' and 'meaning'. In the former case we actually abstract from all linguistic considerations, except that the fact that there may be linguistic relevance is tacitly understood. Utterances belong to the second type, i.e. near to the prot o c o l i n g phase of signs, whereas images belong to phonetics, which, though being a protocolizing auxiliary to linguistics, also requires its own protocols. It is near to the protocolizing stage of phonetics proper that the notion 'image' corresponds. We may thus conceive of utterances and images as belonging as it were, to the third level of abstraction, i.e. (1) events, (2) facts, (3) utterances and images. It is at this level that we begin to conceive of the notions as being 'models'. The term is appropriate, as at this level for the first time, some formal considerations have come into play, i.e. those affecting member-to-class relations.

2. The ontology or 'signum-theory' The signum-theory was, as I have already said, among the earliest ideas from which the theory of axiomatic functionalism

152

Semiotic foundations of axiomatic

functionalism

evolved. In fact, the continued success in the development of axiomatic functionalism is probably largely due to the fact that at least since the early 1960s the necessary ontological distinctions were clearly laid out. Only the notion 'image' has been added since, and only minor details as to the formulation have been altered, mainly in order to simplify statements and explanations. The first name given to this part of the theory was 'sign-theory', but by about 1971 the theory of indices, which was originally conceived as actually belonging to the sign-theory, though it had also strong links with the theory of semiotic systems, had developed in such a way that the notion 'sign' became redundant in the original sign-theory. Until that time the term 'sign-theory' was quite appropriate, partly because it was 'signs', rather than 'symbols', on which the interest was focussed. After the notion 'sign' had been removed from the 'sign-theory', and I realized that a distinction between 'sign' and 'symbol' was not relevant to that part of the ontology, the name 'sign-theory' was changed to 'signum-theory'. Some adherents of axiomatic functionalism, amongst them Sändor Hervey prefer the term 'ontology' to 'signum-theory'. Personally, I prefer the latter term, but there is something to be said for, as well as against, both. It is true that the whole of the signum-theory is an ontology, and that it is this very ontology that dominates the systemology as well as the semantics. As to whether the name 'signum-theory' is less appropriate than 'ontology', it is true that the notion 'signum' represents only one of the several ontological distinctions drawn, but all the distinctions drawn within the general ontology are relevant to the notion 'signum', in that they represent stages in the definition of that term. Moreover, the sub-theory of semantics develops certain ontological distinctions for its own sake, and so does the sub-theory of systemology. And as a theory seen by itself, i.e. without reference to systemology and semantics, the signumtheory is without any doubt a theory clarifying the notion 'signum'. I believe, therefore, that the term 'signum-theory' has slightly better claims in respect of appropriateness. An added reason why I should not wish to reject the term 'signum-theory' is that it is a more linguistic term. The term 'ontology' is of a more general philosophical nature. Besides, 'signum-theory' is

The ontology or 'signum-theory'

153

closer to 'sign-theory' and so preserves a link with the past, not just the past of axiomatic functionalism, but, more importantly, with regard to all those linguists who have dealt with the notion 'linguistic sign', especially Saussure, to whom I owe a very special debt in this respect. The signum-theory in the way it was first presented in Mulder (1966, 1968, 1969, 1971, 1972) and Mulder and Hervey (1972), has been criticized for using set-theoretical operators with notions that seemed to represent single items, rather than classes (sets) of items. However, though at the time this was regrettably not explicitly stated, all these items did represent classes, so the use of these operators was appropriate. This criticism, however, had a very beneficial result as it induced me to introduce the notion 'image', and define 'phonetic form' as a self-contained class of images. By these means not only did the 'class' status of the items dealt with become clear, but also the introduction of this notion had the effect of linking this part of the theory directly to Hervey's notion 'utterance'. Until that time there was only a very complicated indirect relation of equivalence between Hervey's and my definitions of 'sign', and related notions, but now the link could become a directly theorematic one. The relation of the notion 'image' to that of 'phonetic form' was introduced in the following way under Def. 22 (Mulder 1980, Postulates 1980 version): 'Cenetic form (symbolization / ) ' for 'generalized model for a class of impressionistically similar phenomena that may correspond to one or more figurae\ Because of the generalization involved, a phonetic form is already a class of what one could call 'images', these being models of the unique form of a single realization. If we symbolize image as i, i.e. / = {*'}({ } indicating that it is a class), we may, in anticipation of what follows, define 'utterance' as (iRd)Rd (where the first d stands for distinctive function in a phonological sense, and the second d, being the further distinctive function of a phonologically distinctive entity, for the distinctive function of that entity at the signum level), i.e. as 'a model (image) for the specific form of a single realization in its capacity of having the particular distinctive function of a particular signum'. This is in agreement with Hervey's definition of 'utterance' as a model for a single realization of a 'signum,' and with his tenet that a signum, is a

154

Semiotic foundations

of axiomatic

functionalism

class of utterances.'* In natural language 'cenetic form' is called 'phonetic form', i.e. a phonetic form / is a self-contained class of images { i ] . A phonetic form is, for instance, the phonetic feature 'labiodental', or for instance the class of denotata corresponding to a 'letter' of the International Phonetic Alphabet, duly defined within phonetics. In general, all phonetic features that may be the realizations of figurae (i.e. mere linguistic forms, as opposed to signa) are phonetic forms. As far as phonology (as opposed to phonetics) is concerned, phonetic forms (like all kinds of cenetic-form) have the status of generalized protocols — or rather of the material terms used in such protocols — i.e. they represent statements of fact, notwithstanding that there exists a highly developed science (phonetics) that provides us with those protocols. But phonetics, as a descriptive science in its own right, needs its own protocols, and it is next to the level of the latter that we must place the notion 'image'. For reasons of simplicity I shall use the term 'phonetic form' throughout the following discussion, though in fact this term is only appropriate for natural languages. When applying this theory to other semiotic systems, the term phonetic form has, of course, to be replaced by 'cenetic form', or another appropriate term of one's own invention. At this juncture I should like to bring up an extremely important philosophical point — perhaps considered rather controversial by most realists and empiricists. It is ultimately connected with the establishment of the notions 'image' and 'utterance' as realization forms. It is also at the very roots of the philosophical considerations underlying my approach to linguistics. In my view, which I believe to be more 'realist' than that of the 'realists', the notion 'structure' does not refer to any reality outside the description or interpretation of observed events, which we perceive as having 'form' of some kind. But even 'form' itself does not have any intrinsic reality. It emerges from a relation

* Originally I have used such terms as 'realization-form' and 'substance-form' and equated these with 'phonetic (or cenetic) form', but I regard it now as more appropriate to equate 'realization form' or 'substance form' with 'image', and (in a wider sense at another level) with 'utterance'. I have taken the liberty of making minor alterations in what is otherwise a quote.

The ontology or'signum-theory'

155

between an observer, perceiver, interpreter, or what you will, and the object or event perceived, i.e. it is never completely objective. Yet, without reducing events and the like to form (which is quite automatic in observation) we are unable to deal with them, let alone to describe them in an orderly fashion in terms of structure. This is not, of course, to say that 'form' and 'structure' are fictitious and have nothing to do with the event-object as such. But the objective event or thing is only knowable via our observations, etc., which bring in an important subjective, and therefore arbitrary, element. It is therefore appropriate to distinguish between the speech-event as such, and the speech-event observed. It is the latter, already 'form', already a 'model' for the event, which is called 'utterance' and the 'soundaspect' of an utterance, or any part of that 'sound-aspect', is called 'image'. An obvious example of how subjective our observations of events may be is the observation of a rainbow. Where is it, what is it, and how is it, if we do not include an observer of a particular optical kind, in a particular place, at a particular time? That is, 'form' emerges from a relation between an otherwise unknowable object or event and the senses, perhaps supplemented by instruments, used at a particular time and in a particular place in the observation of that object or event. It is that 'form' which is the actual (but already abstract and subjective) object of our description. We can assign that form a 'structure', which itself emerges from a relation between that form and the theory we employ in describing it. But let us now return to the topic, after this - I hope — useful intermezzo. The signum-theory is basically very simple. It employs, apart from the all-pervading notion 'self-containedness' for classes, only one basic entity, i.e. 'image', from which by repeated separate operations of classification and adding distinctive function all the other levels are derived, and ultimately the level of 'signum'. That is to say, for arriving at the notion 'signum', all that is needed is the notion 'image', and two operators, i.e. 'memberto-class' and 'distinctive function'. Strictly speaking, only the latter of the two has ontological consequences but the former has categorial consequences nevertheless. It is clear that a linguist is not the same as the class of all linguists. As the notion 'selfcontainedness' is all-pervading in this connection, the notion

156

Semiotic foundations of axiomatic

functionalism

class in the following has in all cases to be understood as 'selfcontained class' or 'class of all. . . Let us then, start off with the notion 'image' i.e. i. By applying the member-to-class operator we get 'class of (phonetically similar) V i.e. { / } . We now rewrite for simplificatory purposes { / } as / , i.e. 'phonetic form', i.e. / = { i }. To / we apply the distinctive function operator. That is, we consider an / (e.g. a back open unrounded vowel [a], (as opposed to a more fronted [a]), which happens to have a distinctive function in the language in question), in its capacity of having a particular distinctive function d1, where i stands for a particular diacritic that can be used in order to distinguish between different distinctive functions d. We now write this relation between [a] and d as [aJRJ 1 , and this we call an allophone (in this case an allophone of a particular phoneme, whose identity we have, however, not yet established). The formula [a] RcP can be read as 'the phonetic form [α], having and in its capacity of having the particular distinctive function d \ ' or '[a] in relation to We now apply again the member-to-class operator to this formula. I shall symbolize this operator C . The distinctive function operator I shall symbolize D°. We then eet { [a], [a], [ae], etc. ] RcP, which is the same as saying { [a] R ) } , and { Rid^p 1 · n ) } f o r / W , { f 1 · n R ^ } ,p'lRd[, { p{- - " R d ' }

166

Semiotic foundations of axiomatic

functionalism

and { c^Rp* · · n j respectively. As I have said, the notation for these used in previous publications is f^d1, [ / j ' R d 1 , plRs1, { p) i ^ a n d ^ R f p } 1 . For further classification we can set up the following scheme: Ontological levels

level of phonetic entities

level of figura

level of signum

i (image)

iRd (allophonon) X

(iRd) Rd utterance Y

MO (phonetic form)

fRd (allophone)

(fRd) Rd (allomorphon) Ζ

p= {?-nRd] (phonological form)

pRrf (allomorph)

S={pi-nRd} (signum)

One sees that both 'image' and 'phonetic form' belong to the level of phonetic form, etc. Note that I do not say level of phonetics, phonology, and grammar. These are terms which refer to the systemology, not to the signum-theory. In the systemology one should not use such terms as phonological form and signum, but rather call those 'phonological entity' (i.e. distinctive feature, phoneme, phonotagm, or para-phonotactic entity) and 'grammatical entity' (i.e. moneme, plereme, syntagm, clause, or sentence) respectively. The terms 'allophone' and 'allomorph' are, nevertheless, retained in the systemology. There, however, they refer to entities of a particular kind, e.g. allophone of a phoneme, etc., which is not the case when the term is used for notions in the signum-theory.

The ontology or 'signum-theory'

167

There is little necessity for giving specific names to X and Ζ though the levels have, as levels, to be recognized. By way of a joke I have on occasions called Ζ an 'allophonomorph' and X could be called 'distinctive image'. But for Y there actually is a term, i.e. 'utterance', defined by Hervey as 'model for a single realization of a signum'. Level X, of course, represents 'a particular image having and in its capacity of having a particular distinctive function', whereas Y is X having and in its capacity of having a particular further (i.e. on the signum level) distinctive function. Level Ζ stands for 'allophone having and in its capacity of having a particular further distinctive function'. Level Y, i.e. utterance, of course, ultimately stands for 'image' — i.e. model for a single realization of a speech-sound — having and in its capacity of having a particular distinctive function, the resultant X (i.e. 'allophonon') having and in its capacity of having a further distinctive function. The terms 'allophonon' and 'allomorphon' (inspired by terminology used in stratificational linguistics) seem to have a certain elegance, and I therefore have replaced 'distinctive image' and 'allophonomorph' by those two. Actually, from the above scheme one can deduct some alternative and equivalent definitions for 'allophone', 'allomorphon', and 'allomorph' respectively, i.e. {/Rd}, class of allophonons, {(/RcORd}, class of utterances, and {(/RcORd}, class of allomorphons. The importance of these alternative definitions is in that they clarify the relation of 'phonological form', via 'allophone', to 'allophonon', and through this to 'image', and the relation of 'signum', via 'allomorph' and 'allomorphon', to 'utterance', and through this, via 'allophonon' ultimately to 'image'. One also sees that it is equally correct to say that a signum is a class of allomorphs, as saying that it is, though more indirectly so, a class of utterances.

Chapter V: Linguistic description I: General principles, and phonology 1. Introduction To know and understand even a difficult and intricate theory, and to be able to discuss — or even teach — its tenets is relatively easy. But to apply that theory adequately in the description of phenomena is immensely more difficult. The reason is that, when one is dealing with the theory on its own, one can afford to think or talk about things one at the time, and draw conclusions from premises more or less in a straight line. One can follow the argumentation by scanning, as it were, from top to bottom, but it is seldom necessary to be aware of every aspect of the theory, all at once. The act of description, i.e. of applying the theory, however, requires the person to be able to perceive everything at once. Not only does he have to see which parts of the theory are directly, and - more difficultly — indirectly, of relevance to the problem in hand, but he also has to see what consequences his decisions will lead to in the most remote potential areas of a description of the same material. For instance, a descriptive decision taken with regard to a particular syntactic structure may affect the setting up of other syntactic structures and ultimately lead to difficulties or even inconsistencies in the semantic interpretation of those structures in a quite different area of the description. Or a decision made at one point, while enhancing the immediate simplicity of treatment, may lead to gross and even unacceptable complexity elsewhere. All this, if one thinks about it, is so complicated that it is almost unthinkable that anyone could learn the skills involved. But one might equally wonder how it is possible that anyone will ever be able to learn to drive a car through crowded city traffic. The answer is perhaps that human

170

Linguistic description: General principles, and phonology

beings are capable by proper training of acquiring skills which allow them to perform the most complicated action in such a way that to outsiders and themselves it looks as if they purely act by instinct. Now, even in driving a car quite a bit of training is involved in gradually building up one's seemingly instinctive abilities. But for learning to become a skilled descriptivist, especially in the area of systemology, the time spent as a student in a university can hardly be sufficient. Many more years of experience in the description of languages are generally needed before one develops an instinctive fluency in dealing with problems. Alas not everyone will ever acquire such a fluency. Some, perhaps, simply have different kinds of talent. Many, however, if not most, are too impatient to acquire the proper experience, or are not self-critical enough to see that their skills as descriptivists are still under-developed. The latter ones are the most dangerous, as it is they who may ultimately cause the decline and downfall of the theory in question. When they come across a problem which they are unable to solve, or they notice inconsistencies in their descriptions, their inclination is to blame it on the theory, rather than on their own incompetence. They then start tinkering with the theory — but someone who is not even a good descriptivist can hardly be expected to be a good theory-builder. People who are incapable of foreseeing all the consequences of their descriptive decisions are bound to be even less capable of foreseeing the descriptive consequences of changes in, or additions to, the theory. This is how theories usually die. Not by old age, but by tinkering and surgery by insufficiently skilled adherents of the theory. Of course, theories can be improved, and should be ever improved, via an application in descriptions, but this can only be done by good theory builders, which implies that they are extremely good descriptivists. I may have given the impression that I regard the act of linguistic description by a skilled person a purely instinctive act. I should like to take some — but not all — of this back. Certainly, in most cases one hits at the better solutions in a flash rather than by painstaking calculus. Just as one reacts and hopefully makes the correct decision in a flash, when driving a car through city traffic. If this were not the case, complicated activities would proceed in too clumsy and time-consuming a manner to be of any practical usefulness at all. In the case of linguistic and other

Practical hin ts for successful descriptivism

171

intricate scientific description, nevertheless, one is always required to be able to make one's reasons explicit. This is often not easy. One frequently knows that something is either right or wrong long before one knows why, but even so the 'why' has to be investigated. This is not only because the scientist is under the constant obligation of allowing public scrutiny, but also because otherwise one cannot keep consistency, adequacy, and simplicity sufficiently under control, let alone judge the consistency, etc., of subsequent descriptive decisions. But the instinctive in a flash approach to description is, as I have said, in practice very necessary. And human beings are on the whole, inexplicable as this may seem, capable of developing skills in that respect. Unfortunately, for some equally inexplicable reasons, there are those who constantly hit on the wrong decisions to begin with. It is for this reason that I should like to give some partly informal, and partly formal, guidelines as to how to proceed when attempting to describe a language, or anything else complicated for that matter.

2. Practical hints for successful descriptivism The first golden rule — often forgotten in one's zeal to be original and bold — is to keep everything as simple as possible. If it is not necessary to set up a new structure or model to account for certain phenomena, i.e. if existing structures are already accounting sufficiently for it, one should resist the temptation of finding an original solution. Sometimes it is not just due to one's desire to do something special, but at least as must to a lack of vision. Let me take an example from Japanese syntax (e.g. Yasuko Obana, 1984, who has dealt with it correctly). One finds in Japanese such sentence utterances as watasi ga anata ga suki da Ί like you'. The words watasi, anata and suki da translate as T , 'you', and 'to like' respectively. The item ga is a kind of 'post-position'. Such post-positions in Japanese are in many respects similar in usage to 'prepositions' in European languages, but in other respects they are quite different, as they often have wider syntactic implications. Just like 'prepositions' — and the

172

Linguistic description: General principles, and phonology

same is true for 'conjunctions', and other so-called ' f u n c t i o n a l ' — in European languages, these 'post-positions' denote a relation between the things denoted by other items. There are, of course, exceptions, e.g. where one of the denotables is not expressed, or even left vague in the situation, but basically this is their semantic role. In 'the man in the garden', the item 'in' denotes a relation between that which is denoted by 'the man' and that which is denoted by 'the garden'. In this case, 'in' syntactically governs 'the garden', rather than 'man', therefore it is called a preposition. In Japanese it is the antecedent which is governed, therefore I call it a pos/position, but the principle is the same. In Tökyö e ikimasu 'going to Tokyo', the item e ('to, towards') denotes a relation between that which is denoted by an utterance of the placename Tökyö and that which is denoted by an utterance of ikimasu 'to go'. In the case of, say, watasi ga ikimasu ( Ί go') the item ga denotes a relation between that which is denoted by watasi, and that which is denoted by ikimasu. To give one more example, in 'John and Mary', the conjunction 'and' denotes a relation (call it a relation of 'conjunction', but do not confuse this with a syntactic relation; it is a denotatum) between what is denoted by 'John' and 'Mary' respectively. The important thing to remember is that the relation denoted is a relation between 'denotata', i.e. between 'objects' (in the widest possible sense, in this case 'persons'), not between linguistic entities, as most linguists, absurdly enough, seem to think (e.g. Quirk R., Greenbaum S., Leech, G., and Svartnik J.: A grammar of contemporary English, 3rd impression, London 1973.) However, it is not impossible, that quite separately, relations between linguistic items may be implied (though not directly denoted) as well. In the case of ga in Japanese, for instance, it would be very difficult to paraphrase the relation denoted by ga without pointing to the syntactic relation implied by the use of ga. The best one can say it seems is that ga denotes a relation between, say, what is denoted by watasi and that which is denoted by ikimasu, such that the denotatum of watasi can be called the 'subject' (in a traditional, vaguely semantic, not in any strict syntactic, sense, (cf. Yasuko Obana 1984)), of the denotatum of ikimasu. One sees clearly that 'subject' here refers to a denotatum, i.e. to a 'thing', not to a linguistic entity, but because the term 'subject' is used for linguistic entities, and for syntactic relations,

Practical hints for successful descriptivism

173

as well, some people may be confused between the two. This is why I cannot insist enough on the necessity to be constantly on one's guard against such confusions, in spite of inadequacies in terminology. The syntactic relation implied (unlike the 'thing' relations denoted) in the case of ga is not between watasi and ikimasu, but between watasi ga and ikimasu. The former is syntactically the subject, and the semantic import of the syntactic relation of subordination (symbolized > ) in watasi ga > ikimasu is in agreement, i.e. consistent (to put it no more strongly) with the denotation of ga. Let us now return to our initial example, i.e. watasi ga anata ga suki da and compare it with watasi ga ikimasu. It is only if one forgets that there is such a pattern as 'subject > predicative', exemplified by watasi ga ikimasu, and if one at the same time forgets that what is relevant in construction types is only immediate constituents, that one may regard watasi ga anata ga suki da as belonging to a different type, e.g. watasi ga > {anata ga > suki da). However, though there is nothing objectionable in representing it thus, nor even in calling anata ga a 'nested', or a 'secondary' subject (as Yasuko Obana has done), or calling watasi ga the primary subject, it would be a major failure not to recognize that qua immediate constituents (and this is the only thing that counts) the construction on that level is identical with that of watasi ga ikimasu, i.e. 'subject > predicative'. The only special thing is that the predicative in its turn (but this does not become relevant until the next step in the analysis) has the structure 'subject > predicative'. Here then, unless considerations of material adequacy were to demand otherwise (which they do not) it would be a major failure if one were even merely to consider alternative structures, where one has already a structure that can fully account for it. Were one to commit this mistake, it would be extremely easy to fall into another trap, especially because the translation of watasi ga anata ga suki da has 'you' as the 'object' and Τ as the subject. Also in Japanese there are constructions in which 'object' as well as 'subject' are distinguished, e.g. watasi ga mizu wo nomimasu Ί drink water'. Syntactically the 'subject' is here watasi ga, the 'object' is mizu wo (mizu means 'water') and the predicative is nomimasu 'to drink'. Also wo is one of those postpositions. It denotes a relation, such that the denotatum of

174

Linguistic description:

General principles, and

phonology

the item governed by wo can be said to be (in a traditional vaguely semantic, not in any structural sense) the 'direct object' of what is denoted by the 'predicative'. That is, it denotes a particular difficult to paraphrase relation (call it a relation of 'wo-ness') between what is denoted by mizu and what is denoted by nomimasu. I should actually say in all those cases 'denoted by an utterance of . . . but as this goes without saying, I take the liberty of refraining from this tedious way of expressing myself. In the case of watasi ga mizu wo nomimasu, ga and wo clearly show the difference in position, i.e. syntactic function. This may, however not imply a difference in semantic relation with the predicative, as the semantics of the relation between watasi or mizu and the predicative may be entirely due to the difference in meaning between ga and wo. In the case of watasi ga anata ga suki da, however, someone might suggest that there must be a difference in invisible and therefore type of 'syntactic' relation between the predicative and watasi ga and anata ga respectively. This would, however, amount to committing the mistake of regarding them as immediate constituents at the same level, i.e. failing to notice that it represents an ordinary intransitive construction, i.e. 'subject > predicative' in first instance. Such cases as watasi ga mizu wo nomimasu I have called 'parallel determination', represented as subject -

watasi ga —> mizu wo

nomimasu, i.e.

J

> predicative, object

whereas the other case, for instance in English 'John likes Mary1 I represent as subject

John -> likes, Mary

i.e.

-> predicative. object

This is called 'diverse determination'. The immediately apparent difference between the two is that in cases of parallel determination the position classes (i.e. the set of items capable of occurring in a particular position) of the peripheral constituents are different

Practical hin ts for successful

descriptivism

17 5

(though they may have some members in common), whereas in diverse determination they are basically the same. Note that I speak about position classes, not the actual members realized. In the case of parallel determination, the hypothesis that there is no diversity in determination (e.g. for all we know, the import of the relations of 'the' and 'old' versus 'man', in 'the old man', may be exactly the same) remains unrefuted, precisely because the position classes are different. I shall come back to this later. One might now (as I have suggested, because the position classes may seem to be the same) fall into the trap of regarding watasi ga anata ga suki da as a case of watasi ga ->

suki da,

anata ga i.e. as a case of diverse determination as well. One can, after all, have, with a corresponding difference in denotation, anata ga watasi ga suki da, just as one can have in English 'Mary likes John', which is denotationally different from 'John likes Mary'. Mizu wo watasi ga nomimasu, however, though perhaps unusual, would be denotationally the same as watasi ga mizu wo nomimasu. Not only would adopting such a view violate the requirement of simplicity as it would unnecessarily introduce a new structure, but it would lead also to further absurdities, which I cannot go into here, and to serious difficulties in describing both the semantics and the syntactic deployment of postpositions in Japanese in particular, and of postpositions and prepositions in general. In fact, I do not see how one could get out of any of the major difficulties short of denying ga to have any status at the signum-level at all. This has, in fact, been suggested by one critic. But one should not really have encountered any of these time and energy consuming difficulties. The initial mistake was to fail to see that the simplest solution, and at the same time the only consistent and fully adequate one, was to do nothing at all, as the pattern was already established in the description of the ordinary intransitive construction, i.e. 'subject > predicative'. Another suggestion, which I have heard proposed for this case is to regard watasi ga anata ga together as the subject. It should, however, be almost immediately clear that it would not

176

Linguistic description: General principles, and phonology

be possible, by any stretch of the imagination, to establish a meaningful direct relation between watasi ga and anata ga, and it could therefore never be a single constituent. Another frequent mistake is unnecessarily to deviate from certain time-honoured conventions. Not everything that linguists before us have done is wrong, and there are often good 'adequacy' reasons for particular conventions to have become well established, even though one does not always immediately see why. I shall take an example from phonology. Traditionally one speaks of voiced or unvoiced, stops and fricatives, not, say, of stopped or fricativized voiceds and unvoiceds, say in a language such as English. Consequently, if one sets up a phoneme-table for the majority of English consonants, it may perhaps in first instance look something like:

stop

fricative nasal

voiced

unvoiced

labial

b

Ρ

V

f

m

apical

d

t

a

θ

η

hissing

ζ

s

hushing

ζ

5

dorsal

g

k

voiced

unvoiced

(χ)*

1

* mostly in Scottish and Irish dialects.

We shall see that we may set up a far better table, but for this we have first to develop the notions 'hyperphoneme' and 'hyperfeature' (see below). For the sake of argument, however, I shall proceed for the time being in the more conventional fashion. The reason why Trubetzkoy-oriented linguists tend to set up phoneme tables in the above manner is presumably that 'voiced' and 'unvoiced' account for what phoneticians call 'secondary articulations', and it is to be expected that this has certain structural consequences as well. At any rate, whether this is the

177

Practical hints for successful descriptivism

case or not, the adequacy of the above-mentioned habit among functionalist linguists seems to have been corroborated over a long stretch of time. This is not to say that habits should never be changed, or that such a change may not bring important improvements, but I should like to insist that it is only prudent to think twice and hesitate for some time before one decides upon a change. There should not only be good reasons for such a change, but all the consequences should be thoroughly investigated. Now suppose someone is somewhat disturbed — and understandably so — by the gaps in the above representation. It can scarcely be called elegant that at the point of 'hissing' and 'hushing' there are neither 'stops', nor 'nasals', i.e. there are six gaps. And as two gaps is less inelegant than six, the describer may be tempted to change the presentation, and give priority to 'voiced' and 'unvoiced' over 'stop' and 'fricative' in the following manner:

voiced

unvoiced nasal

stop

fricative

stop

fricative

labial

b

V

Ρ

f

m

apical

d

5

t

θ

η

(χ)

Ί

hissing

ζ

s

hushing

ζ

§

dorsal

g

k

This 'looks' much better than the previous representation. The only immediately obvious flaw is that /x/ is represented as a 'fricative unvoiced', but as it is not opposed to any 'fricative voiced', this is in violation of the 'functional principle', which is the major principle underlying this type of presentation. However, one could of course leave the admittedly marginal /x/ out of consideration. But there are other violations of the functional

178

Linguistic description:

General principles, and

phonology

principle, not so directly visible, but nevertheless violations which in fact render the adequacy of this presentation practically nil. Let us first consider what is the actual purpose of this type of presentation. Broadly speaking, we can say that the tables serve to visualize the proportionality between the phonemes in terms of their distinctive features, i.e. b : p : : d : t : : g : k , b:v::p:f::d:5 ::t:0. Even b : f : : d : 3 , and so on. Also b and v:p and f::d and 3:t and Θ. Furthermore, according to the second scheme, d and 3 : n : : b and v:m, ρ and f:m::t and 0::n. But according to the first scheme this is not the case, at least not in the sense that the set of voiced apicals stands to n, as the set of voiced labials stands to m. There b and p:m::d and t:n. The two schemes are therefore not equivalent, i.e. they are not mere notational variants. Now we can also set up a table for the archiphonemes, which result in English from a neutralization between 'voiced' and 'unvoiced' in certain contexts. An archiphoneme represents in that context both voiced and unvoiced, that is to say that the proportionality of the archiphonemes, say, /B/ and /D/ is the same as that of '/b/ and / p / ' versus '/d/ and /t/'. As there are no archiphonemes in English neutralizing oppositions between fricatives and stops, only the first, not the second, table is consistent with this state of affairs. In fact, this is as expected, because, as everyone should know — if only from experience — 'secondary articulations' tend to be more prone to neutralization than 'primary articulations' are. Only if one no longer regards proportionality in terms of distinctive features as the prime function or even any of the functions of this type of phoneme-table, can one accept the second type of table. But in that case it is difficult to see what the purpose of such tables is at all, i.e. the adequacy of the presentations is in serious doubt. As I have said already, all this results in the first place from lack of caution when wanting to change time-honoured conventions, and from a lack of foresight with regard to the consequences of such a change. For reference I give here the scheme for the archiphonemes which result from the neutralization. It is easily seen that it is consistent with the first of the schemes, but not with the second.

Methodology and hypothetico-deductive

logic

stop

fricative

nasal

labial

Β

V

m

apical

D

D

η

hissing

Ζ

hushing

1

dorsal

G

X

179

1

As is customary, the capital letters represent archiphonemes. We could have written equally well /?/, /F/, /T/, etc. We can imagine the archiphoneme-table to have been written on transparent paper. When placed over the first scheme it shows the same proportionality and structure. I should like to raise here yet another point, and I am addressing myself now especially to friends of axiomatic functionalism. Those tend to be in love with the very beauty of logical reasoning itself, and therefore particularly prone to certain errors. It was, perhaps, the insistence on logic, which attracted them to the approach in the first place. But description involves the application of logic, not the pursuit of logic for its own sake. As soon as there is the slightest external hint of absurdity or inadequacy, one should immediately abandon one's calculus and try to find where something went wrong. Perhaps it was at the very beginning. Failing to do so may have the most ugly consequences. We shall see examples of this in the following sections. Even better to have a fair amount of cosy waffle (linguistic literature is full of this) than a perfect piece of beautiful logical nonsense.

3. Methodology and hypothetico-deductive logic Few questions have baffled epistemologists more than the question of what is to be regarded as the 'methodology' connected

180

Linguistic description:

General principles, and

phonology

with a particulary theory. It is clear that in a theory, e.g. a linguistic theory, one finds notions that lead to the establishment of models, and one even finds ready-made models, such as 'phoneme,' 'syntagm', 'signum', etc. which can be used in a description. So, it has been suggested that the methodology constitutes an integral part of the theory, or even that theory and methodology are basically one and the same thing. If one regards the methodology to be an integral part of the theory, rather than identical with the theory, one most likely regards the methodology to be seated in, or at least to be introduced by, certain definitions. Perhaps not by every one of the definitions, but at least by a substantial number of them, either directly or indirectly. But once one acknowledges that methodology may also be introduced indirectly, it is difficult if not impossible to point to any definition which is completely unconnected with the methodology, i.e. which is not relevant to the actual mode of description under that theory. But is not the same true of the axioms? It may be that the bulk of the definitions is more closely and directly relevant to actual description, but not only are the axioms also relevant, the definitions themselves would not have any status and meaningfulness without the axioms, which, as it were, constitute the soul of the theory. Without the definitions, the axioms would be sterile with regard to description, but without the axioms, the definitions would be mere ad hoc empty statements. So it seems that we have come round in a circle and have to conclude that theory and methodology are one. Or do we have to? One also speaks about the hypothetico-deductive method, the inductive method, about operationalism, instrumentalism, speculativism, etc. Though the theory I have in mind has been set up with a view to the hypothetico-deductive method, one cannot say that this method is inherent in it. True enough, inductivist brands of philosophy and epistemology would not allow for an a priori theory to be set up as an instrument for description, but once one had a theory like ours, it would be quite possible to proceed from there on in a completely inductivist manner. In fact this is what self-confessed axiomatic functionalists have done on occasion, and this is precisely why I am writing this section. The correct application of the theory of axiomatic functionalism in the description of languages or other semiotic systems hinges on it. What do I mean by proceeding in an inductivist fashion?

Methodology and hypothetico-deductive

logic

181

Perhaps I should say 'positivist', rather than 'inductivist', as the latter perhaps implies a philosophy in which theories derive from generalizations, and are not a priori constructs. One would be a positivist if one were to take, for instance, the definitions of phoneme 'a self-contained simultaneous bundle of distinctive features', or (equivalently but from another angle) 'a minimum syntagmatic entity in phonology', and were then to use these as parameters by which one could decide whether certain phenomena corresponded to phonemes, or whether they did not. One could even delude oneself into thinking that one is a proper hypotheticodeductivist by 'hypothesizing' that such and such a phenomenon corresponds to a phoneme, and, in the absence of refutation, regarding this as being corroborated. But this is exactly where real hypothetico-deductive philosophy differs, or ought to differ, from positivism, though I have not found this spelled out clearly elsewhere. Nevertheless, after having added to the Popperian philosophy a precise distinction between theory and description, we have to go back to Popper and see what exactly he has in mind when he speaks about a hypothesis. I think that for something to be regarded as a proper hypothesis in the operational sense, i.e. not in the sense of being merely hypothetical, but in the sense of being offered for direct refutation, a hypothesis should at least fulfil the following three conditions: 1. The statement should refer to observables. There might be some discussion as to what counts as an observable, but for linguistics it should mean that the sentence-utterance within which the phenomenon in question is protocolized should be clear as to both its formal properties and its meaning-bearing properties. 2. The statement should be meaningful under the theory employed. That is, it should not contain any non-primitive terms that are not defined by the theory, nor should it state anything that is either irrelevant under, or inconsistent with, the theory in question. 3. It should be in principle refutable, i.e. it should not A PRIORI be irrefutable. The latter is the actual essence of what is a hypothesis in the Popperian sense. Yet it is the least understood and the most violated.

182

Linguistic description:

General principles, and

phonology

Let us give some examples of violations. Suppose one launched the hypothesis that 'dark 1' and 'clear 1' are separate phonemes in English. The positivist would probably believe that this is a proper hypothesis. He would attempt to use the definitions for 'phoneme' and see whether they would fit both of them. In particular he would see if the two are different bundles of distinctive features, or if the two are functionally opposed to each other in a wider context. But would the fact that we did not find an instance which showed that their distinctive features make-up was different, or the fact we did not find an instance in which the two were functionally opposed in a wider context, constitute a refuting example? How could, by any stretch of the imagination, or by any logic, the fact of not finding an example constitute a refuting example? Only if we could scrutinize all relevant speech phenomena, past, present and future, which, of course, we cannot, would the conclusion that 'dark Γ and 'clear Γ are variants of the same phoneme, i.e. allomorphs of the same phoneme, in the absence of consistent refutation be warranted. If, as a self-professed hypothetico-deductivist, we were to fall into such a trap — and this happens all too easily — we would be doing what we are criticising the naive realist inductivist of doing. Examples of this kind can be easily multiplied ad infinitum. If we conjectured that there were two, or three, or twenty homonyms 'dog' in English, this would be a priori irrefutable. Therefore the assumption (no matter how true) that there are at least two homonyms '/her/' (i.e. 'hare' and 'hair') in English is not a valid hypothesis, because of its a priori irrefutability. In the same way the assumption that /b/ and /p/ are separate phonemes in English does not constitute a valid hypothesis. Other examples are to assume that there are umpteen homonyms 'de' in French, umpteen homonyms '-ist' (as in 'typist', 'lutenist', etc.) in English (provided that '-ist' is an instance of a moneme, rather than merely part of the form of an unanalysable signum) and scores of other examples. The intuitive reason is that to find no examples cannot constitute a refuting example. The formal, i.e. logical reason, intimately connected with this is that the theory contains such notions as 'phoneme', 'homonym' and many directly or indirectly connected notions, all of them duly defined in such a way that they can be used in order to find

'Pseudo-composites and pseudo-words' re visited

183

refuting examples — provided there are any to be found. But in that case the hypotheses in question must contain the opposite of what is stated in the definition. That is to say, we must assume that tentative /p/ and jbj are not different phonemes. Our definition of 'phoneme', readily refutes such a hypothesis. We must assume that 'dark Γ and 'clear Γ are not different phonemes, which hypothesis up to now has remained unrefuted. We must assume that there is only one noun, say, 'rabbit' in English, which perhaps remains unrefuted, but we must also assume that there is only one plereme (roughly: word) '/her/' ('hare' or 'hair') in English, which is readily refuted by applying one of the definitions of homonymy, and related definitions. Once refuted, of course, the hypothesis is that there are no more than two homonyms '/her/' in English, which either remains corroborated, or is refuted, In any case it constitutes a valid hypothesis. Of course, just as refutation of a hypothesis leads to further hypotheses, the hypotheses used as examples in this discussion may depend themselves upon the refutation of earlier hypotheses, e.g. before one hypothesizes that something is not a separate phoneme, at least the hypothesis that it is not a distinctive entity in phonology must have been refuted. That is, we would not bother to apply the definition of phoneme, if we have already corroborated that it is not a distinctive entity anyway, unless, of course, this happens to be done in one and the same operation. In the next section we shall look at an earlier publication, in fact a publication done well before I was consciously aware of the procedures just outlined. We shall see how it still stands up, or not, to the test, under the present much more severe requirements.

4. 'Pseudo-composites and pseudo-words' revisited Though this ('Pseudo-composites and pseudo-words — suffient and necessary criteria for morphological analysis') is a very early 'pilot' article, it may be enlightening to compare what is advocated there with what I have said in the previous section (Hervey and Mulder 1973, Mulder and Hervey 1980). Before we judge the article, it has to be kept in mind that it was written well before my

184

Linguistic description:

General principles, and

phonology

postulates, which did not acquire a more or less complete form until 1974. Moreover, the aim of the article was a limited one. It was never intended to provide a complete basis for morphological analysis. The aim was two-fold. Firstly, it was a reaction against Martinet's 'synthkme' idea. We thought this inconsistent, as it seemed to imply that words such as 'blackbird' consisted of two monemes, and were yet simple — or more precisely: they behaved as single monemes. We claimed that 'black' and 'bird' were mere formal elements, and could not be assigned independent 'sign' status. The other aim was to draw the line between morphology (roughly the analysis of words) and syntax (the analysis of syntagms). Thus items such as 'black-bird' were 'pseudo-composites' and items such as 'garage door', which we could demonstrate to be syntactic complexes, were therefore not words at all, but syntagms. In English this is perhaps readily accepted, but, say, in German or Dutch where one writes the items in question without space between them, they are generally considered words. Hence the term 'pseudo-words'. We see already that the title is slightly misleading, as some are criteria for syntactic analysis. It is indeed an article that has at least as much relevance to syntax as it has to morphology. There is another misleading feature in the title: students have in some cases taken it as meaning: criteria that are both sufficient and necessary. This was, however, never the intention. Some criteria were thought to be sufficient, others necessary, and some both. But it was never implied that the article was exhaustive with regard to the criteria given. Let us now see what the proposed procedures were. It is indicated that 'signs' are either simple or complex. This implies, but this is not mentioned in the article, that the item under investigation is a 'sign'. According to the principles stated in the previous section this should have implied that the hypothesis that it has no distinctive function at any level is refuted. That the hypothesis that it has no distinctive function at the signum-level is refuted, and that the hypothesis that it is not a sign is refuted. I will come back to some of these points later on. Complex signs, it is stated, are either morphologically complex, or they are syntactically complex. A morphological complex is a self-contained simultaneous bundle of two or more minimal signs, i.e. monemes, in a morphological relation. To say 'minimal signs' is actually

'Pseudo-composites and pseudo-words' revisited

185

redundant, as we shall see, as it is theorematic that unless the immediate constituents are minimal signs, the relations between the constituents are not merely relations of simultaneity. It is then said: The hypothesis that a given complex sign Ρ (and this is the type of descriptive hypothesis for which we wish to provide a theoretical framework in this article) is a simultaneous bundle of the set JC of monemes, implies the following: (a) The complex sign Ρ is a potential constituent. The equivalent definition of 'simultaneous bundle of distinctive features' (in grammar: monemes) is 'minimum syntagmatic element', and all minimum syntagmatic elements are potential constituents, though not vice versa. (b) The set JC consists of at least two signs. If this were not the case — and a condition is that each constituent must be identifiable as a sign with a sign in at least one other context (see the discussion below) — then the sign Ρ could not consistently be said to be complex. (c) The set χ contains only simple signs. Monemes are all simple signs (though not vice versa) and therefore any complex containing at least one constituent which is itself a complex sign, cannot be a 'simultanous bundle of monemes', by definition. This requirement, as we have said above, could be restated as a requirement that the immediate constituents of a 'morphological complex' should at the same time be its ultimate constituents. (d) The members of the set χ (of simple signs) all stand in a relation of simultaneity to one another in the complex sign P. The other alternative, which is that at least two members of the set χ are not simultaneous in P, implies that they must stand in syntactic positions in that complex, which, in turn, implies that the complex Ρ is a 'syntactic complex', and thereby directly refutes the hypothesis that Ρ is a 'morphological complex'. The above implications, which are conditions for valid hypotheses of the form the complex sign Ρ is a simultaneous bundle of the set χ of monemes are theorematic in Mulder's axiomatic linguistic theory, being deducible from the set of axioms and definitions of that theory, as outlined in Sets and Relations in Phonology (Hervey - Mulder 1973).

One should remember that the postulates in their 1974 form (in Mulder — Hervey 1980) had not yet been completed at the time of writing this. The postulates in their present form (Chapter VIII) constitute a revision of, and an improvement on, the 1974 postulates. Criteria (c) and (d) can only become relevant after it has been established that the sign in question is a complex sign, i.e. after

186

Linguistic description: General principles, and phonology

the hypothesis that it is a simple sign has been refuted. We shall see, however, that (c) is the odd one out, and should not have been listed as a criterion of the same kind, and nor is (d) a criterion belonging to this set. It is in fact a hypothesis for the next step, as we shall see. The way to attempt to refute the 'simple sign' hypothesis is by commutation. As it has to be assured that this is valid commutation, in which one of the requirements is that the context has to remain constant, certain criteria are applied to assure sign-identity. One of the criteria for sign-identity involves the notion 'denotation'. This has been misunderstood by Suleiman (1987), who claims that in the article semantic criteria are used in morphological analysis. In fact they are only used in order to guarantee identity of signs, and thereby valid commutation. As there are some flaws in the article (e.g. 'black bird' (as opposed to 'blackbird'), used as an example, is not a potential constituent) I shall use the example 'the black bird' which is a potential constituent. We have the following commutations: the a the the

black black white black

bird bird bird horse

As the items 'the', the items 'black', and the items 'bird' are, within each of the three sets, identical at the sign-level, and all four examples are potential constituents, and some other conditions are fulfilled, which I shall discuss later, this is valid commutation, and it shows that the examples in question refer to genuinely complex signs. If, however we try to do the same with 'blackbird', which is a potential constituent — i.e. the hypothesis that it is not has been refuted (see below, when we are dealing with that criterion), we find that the hypothesis that it is a simple sign remains unrefuted. Though it was not put in this way in the article, and it was at that time still not realized that this is the proper way of going about it, it is implicitly the case that we regard 'blackbird' a simple sign because we cannot establish that it is a complex sign. The procedure is in fact the application of criterion (b).

'Pseudo-composites and pseudo-words' revisited

187

In the article it is represented as follows: The attempted analysis of 'blackbird' breaks down on the identification of the tentative constituent *'black' in that complex with any other element having a form /blak/, or any contextual variant element, playing a semantic role equivalent to that of *'black'. The semantic role of the tentative constituent *'black' is demonstrably not that of indicating the 'possession of a certain colour-property' (N.B. an albino blackbird does not have this colour-property, yet it is correctly denoted by utterances of 'blackbird' according to the conventions of English, while a crow does have this colour-property and yet is not correctly denoted by utterances of 'blackbird' according to the same conventions). If the tentative element *'black' in ('blackbird') can be said to have any semantic role at all, this role approximates to the paraphrase 'membership in a certain sub-species of the thrush family'. In no other complex of English, to our knowledge, does an element with the phonological form /blak/, or with another form in complementary distribution, occur with a semantic role even remotely approximating to 'membership in a certain sub-species of the thrush family'. The tentative constituent *'black' is neither identifiable with 'black' (denoting a colour-property), nor with any other element *'black', say, an element *'black' in 'blackberry', as we shall see below. Nor can we find an allomorph with a different form but similar semantic role (i.e. a 'membership in a certain sub-species of the thrush family'). P: 'blackbird' *'black'

R

*'bird'

denotation of *'black' (membership of a subspecies of the thrush family)

which bears some relation to

denotation of *'bird' (membership in the class of birds)

denotation of *'bird' (membership in the class of birds)

which bears some relation to

denotation of *'black' (membership in a subspecies of the thrush family)

188

Linguistic description: General principles, and phonology

Ρ: '?'

*'black'

R

denotation of *'black' (membership in a subspecies of the thrush family)

which bears some relation to

denotation of

?

The element *'black' cannot be identified in any other complex and therefore not in 'blackbird' either. Certainly this element cannot be identified (with a view to denotational role) with the sign 'black' (having a certain colour-property), as in 'black bird' or 'black horse'. Nor can it be identified with such tentative constituents as *'black' in 'blackberry', because of the differences in the semantic role displayed by these tentative constituents. P: 'blackberry'

*'black'

R

denotation of *'black' (membership in a sub species of berries)

which bears some relation to

denotation of *'berry' (membership in the class of berries)

denotation of *'berry' (membership in the class of berries)

which bears some relation to

denotation of ""black' (membership in a sub-

*'berry'

species of berries)

As a matter of fact the unanalyzability of 'blackberry' seems to us even highly satisfactory from an intuitive point of view, since 'blackberry' would seem to be analogous with 'cranberry' (e.g. neither *'black' or *'cran' are productive). In the case of 'cranberry' an analysis into *'cran' and * 'berry' would, probably, not even be attempted in (synchronic) descriptions of English, and therefore it is as well that the attempted analysis of its analogue, 'blackberry', should prove to be unsuccessful (Hervey - Mulder 1973, Mulder - Hervey 1980:130).

'Pseudo-composites and pseudo-words' revisited

189

Let us now turn to criterion (a). This criterion has the sole function of making sure that we are dealing with real entities, not with «o«-entities from the point of view of grammar. For instance, without such a criterion one might consider for potential analysis 'the three', 'three black', 'black birds', 'birds do', 'fly very', 'not fly very', etc., none of them potential constituents in any given syntactic complex. At the time of publication, it was not seen that this also might rule out syntagms or words that are ipso facto not combinable with other syntagms or words on the syntactic level (e.g. in English 'yes', but even this is a potential constituent, vide: 'yes and no', 'no or yes', etc; but in principle non-combinable entities may exist, and they do in fact exist in some semiotic systems). All we meant to say by this requirement was that the complex sign Ρ should be grammatically self-contained, in order to rule out invalid commutations, i.e. commutations with non-constituents. In other words, all that this criterion does is to assure that we are dealing with a grammatical entity. The hypothesis has, of course, to be put in a negative way, in order to be potentially refutable. That is, the criterion in question, and this holds for criteria (b) and (d) as well, enables us to refute the hypotheses, which should state, in order to be effective, the negation of what is stated in the criteria. They are not — though in the article this impression was given — parameters which can be used to find things. Criterion (c), which strangely enough is singled out by Suleiman (Suleiman 1983) as being straightforward, and easy to apply, is not a valid criterion at all. The hypothesis that the set only contains simple signs, i.e. 'none of them is a complex sign', is a priori irrefutable. This is because the hypothesis of any sign that it is not simple, i.e. that it is complex, is irrefutable. May we then, perhaps, say that we have here not a criterion for refutation, but a proper hypothesis? Unfortunately, even this is not the case. There is no possible direct refutation of this, though it may theorematically follow from definitions and the refutation of proper hypotheses via criteria (a) and (b), applied to constituents of the sign in question. In fact (c) is theorematically correct if it has been established by the other criteria that the sign Ρ is a morphological complex. Neither (c) itself, however, nor its negation can be refuted. Nevertheless, this theorem can have its uses. It is just possible that because of lack of refutation to the contrary one

190

Linguistic description: General principles, and phonology

comes to the conclusion that the sign Ρ has no constituents that are themselves complex. A case that comes to mind is in the system of number names in English, say the number three hundred and forty two thousand four hundred and twenty three. I may be mistaken, but I do not think that it can be established that this is a syntactic rather than a morphological complex by any of the above procedures, i.e. we have to assume by theorematic implication that all immediate constituents are ultimate constituents. However, I believe that it can be established that such an assumption lacks adequacy in view of the message conveyed in sentence utterances containing this item in the base. In that case we have to review our conclusion by default. We see that adequacy has the final word, final, that is, in the most literal sense logically speaking. We must never substitute this for procedures based on the application of the theory. In any cases of /low-refutation, however, there is still an element of uncertainty left, and we may therefore overrule a decision on the grounds of its being inadequate, or perhaps even inconsistent. If a conclusion, however, derives from refutation (as opposed to «oH-refutation, i.e. corroboration), and — God forbid — we find that it is inconsistent, or inadequate, we have an extremely serious situation. Either our theory must be at fault, or something along the line must have gone wrong in its application. This has to be investigated, beginning with the latter. We now turn to criterion (d). It is easily seen that this is not a proper criterion either, as its negation is not a proper hypothesis. Namely the statement that the constituents of the complex sign Ρ do not stand in a relation of simultaneity, i.e. that they stand in a syntactic relation with one another is a priori irrefutable. This is because 'simultaneity' is assumed only in the absence of refutation to the contrary, the latter implying that it is syntactic. To quote from the article: 'What evidence is it possible to adduce in support of the statement that the relation between the simple signs a and b is a morphological relation of simultaneity in P? The answer most probably is that there is no direct, positive evidence for proving such a statement, but that such statements can be regarded as constituting valid hypotheses, as long as no evidence can be found to disprove them. The problem devolves then, not on finding evidence of a morphological relation, but on attempting to find, and finding or failing to find, evidence of syntactic relations' (Hervey — Mulder 1973).

'Pseudo-composites and pseudo-words' revisited

191

One sees that the so-called criterion (d) is actually not a criterion in principle enabling us to refute a certain hypothesis, but it is itself a proper operational hypothesis. But being the latter, it requires, of course a criterion or criteria in principle enabling us to refute such a hypothesis. That is, we need a criterion or criteria spelling out what constitutes 'syntactic complexity'. I shall not go here into the fairly longwinded, but by no means uninteresting, way this is explained in the article, but give instead some of the relevant definitions. We can define 'syntactic complex' as 'self-contained complex of syntactic entities'. 'Syntactic entities' are 'syntagmatic entities' in a grammatical system (cf. Chapter VIII, Def. 9b). 'Syntagmatic entities are entities capable of standing in ordering relations with other entities, or having an internal structure such that they are capable of containing — as constituents — entities capable of standing in ordering relations with other entities (cf. Chapter VIII, Def. 7b2)'. In other words, to be a syntagmatic entity implies being itself orderable, or having in its structure something that allows it to have orderable constituents, unless — as we shall see later — it is not combinable at any level below 'sentence' (see Chapters VI and VIII). From these definitions it is not difficult to derive two tests, two criteria, each of them, if they apply, sufficient to enable us to refute the hypothesis that something is a morphological complex. The first test is to see if the immediate constituents are ordered, i.e. whether the same immediate constituents could make up a different complex sign. Matters of valid commutation are extremely relevant here, and this includes matters of signidentity, as well as matters of identity of relations between constituents in the complex concerned. One of the cases where this applies positively is 'John hit Mary', as opposed to 'Mary hit John'. It is clear that the relation of 'John' versus 'hit', and also that the relation of 'Mary' versus 'hit', in the first example differs from that in the second. This test has been called the 'permutation-test'. It is the simplest way of determining syntactic relations, i.e. of refuting the hypothesis to the contrary, but unfortunately the cases that it can be applied in a clear and unambiguous way are not very numerous. That this test can be derived from the definition of 'syntagmatic entity' is clear, as entities that do stand in ordering relations are capable of standing

192

Linguistic description: General principles, and phonology

in ordering relations. Note that the permutation test, if positive, establishes at the same time that it is a case of 'diverse determination' (see above, and Chapters VI and VIII). That is, the permutation test is only possible in case of 'diverse determination', and the latter implies that there are at least two peripheral constituents. The second test — and this is the one that always can be applied — consists of trying to commute one, or more, of the constituents with an entity which independently has been shown to be a syntagm. In order to be valid commutation, the relations between the immediate constituents have to remain the same. For instance in 'the book' one can commute 'the' with 'my father's', which gives 'my father's book'. The hypothesis that the relation between 'the' and 'book' and that between 'my father's' and 'book' is the same remains unrefuted, so the syntactic structure is not affected. By 'structure' is always meant the relations between immediate constituents. Now to come back to the actual test, which can be called 'the commutation with syntagm test', one sees that this is in agreement with the second part of the definition of 'syntagmaiic entity', i.e. 'or having an internal structure such that they are capable of containing — as constituents — entities capable of standing in ordering relations with other entities'. It should be remembered that these are ordering-relations if and only if 'simultaneity' has been refuted. Each of these tests, when positive, and if complexity has already been established, refutes the hypothesis that the item in question is a morphological complex, because a multiply complex syntagmatic entity in grammar is ipso facto a syntagm, and its immediate constituents are ipso facto syntagmatic entities in grammar themselves. The latter is in agreement with the definition of 'syntactic complex'. Note the connection between this and (pseudo) criterion(c). Failing the possibility of applying both tests, there is always the possibility that the adequacy of an unrefuted (i.e. corroborated) hypothesis is itself refuted, in which case one has to reverse one's decision by default, as I said already. On the other hand, it is not logically possible — if the theory itself is consistent and adequate, and has been properly applied, and one's observations and all other things have been correct - that the adequacy of a decision deriving from the refutation of a hypothesis is refuted.

Productivity and pseudo-productivity

193

5. Productivity and pseudo-productivity. Analogy as motivation for the formation of new words If we look upon the article discussed in the previous section as an at least practical, albeit incomplete, piece of methodology, we must admit that there is one important omission. Namely, in the 'sign-identity' procedure as outlined in the article, there is no indication that for finding valid commutation it is a necessary, but not sufficient condition, that the hypothesis of sign-identity for the contexts in which the tentative commutations takes place remains unrefuted. That is, if it is refuted, the commutation is invalid, but if unrefuted it may still be invalid. What was omitted, because it was perhaps not fully realized at the time, was that the commutation process must be productive, i.e. the potential commutants should be all the members of the distribution class in question. Distribution classes should be established independently of the particular context in question, i.e. one cannot say that a distribution class consists of the commutants in that particular context, as this would be begging the question. If particular commutations are unlikely, but this is due to the real world, rather than to the structure of the language in question, e.g. cases of semantic incompatibility, or of meaninglessness, or redundancy of 'message', etc., this can be ignored. Let me first give some examples, and first turn to the difference between syntax and pseudo-syntax, as it is more readily accepted that in syntax (in contradistinction to 'morphology') one hundred percent productivity rules supreme. In fact, I shall argue that less than one hundred percent productivity is in all cases mere pseudo-productivity. If we have an item such as 'Jack-in-the-box', 'Jack-of-all-trades', or 'go-between', it is readily accepted that these are 'words', not syntagms. There is very little one can do with these by way of internal commutations. One can add a plural, but that is about all, though jocularly, one can have word-plays, such as 'Jack-ofno-trades'. One can even by close analogy with those existing forms coin a new form. Actually, this is what one does if one jocularly or perhaps even seriously, says such things as 'Jack-ofno-trades'. If this were to catch on, one would have a new word. Without the requirement of full productivity one might now say

194

Linguistic description:

General principles, and

phonology

that 'Jack-of-all-trades' is complex, i.e. consisting of the monemes 'Jack-of . . . trades' and 'all'. However, unless all the members of the distribution class to which 'all' and 'no' belong could occur in the same position, we cannot say that the occurrence of 'Jackof-no-trades' alongside 'Jack-of-all-trades' is more than accidental. Perhaps motivated by analogy, but nevertheless arbitrary in the Saussurean sense. It would have to be listed in the lexicon, but it would not be inherent in the structure of language. In other words it would not be productive. Only that which is productive belongs to the structure of language, or rather 'productive' and 'derivable from the structural rules' are synonymous. Of course, the fact that one cannot establish a meaningful grammatical relation between 'Jack of . . . trades' and 'all' makes it already a priori impossible to regard them as immediate constituents. It has been alleged that in axiomatic functionalism one cannot use the criterion of productivity, because it is not defined in the theory. Such an allegation is, however, based on a (perhaps double) misconception. In the first place, I do not see how 'productivity' could belong to the theory as it stands, any more than 'consistency', or 'adequacy', or 'simplicity', or 'logical argument', or 'conclusion', or 'observation' could belong to the theory as such. Secondly, as the act of linguistic description as far as the systemology is concerned, after listing the simple items, consists of giving the structural rules by which complex items can come into being, it goes without saying that accidental items, i.e. items that can not automatically be traced back to an application of those rules to simple items will be treated elsewhere — in fact in the lexicology, and as far as their creation is concerned, perhaps in a theory of motivation (especially the role of 'analogy') in word-formation. Such a theory would, of necessity, have a diachronic, or perhaps dynamic, flavour, and would be additional to, but not part of, a theory of the kind we are concerned with here. Thus, as far as systemological description in our sense is concerned, only that which is 'productive' can fall within our scope. I use the word 'non-productive' only in order to indicate to others who have overlooked this, or in order to warn others, that certain types of data are not amenable to structural analysis and can therefore at best be listed in an inventory, or be dealt with under additional (but not sub-) theories to the present one. The term 'productive' in this context can only mean that the

Productivity and pseudo-productivity

195

formation is not accidental but according to purely structural rules, and hence it falls within our scope of systemological description. Let us turn to other examples, in this case to a productive pattern in English, but where fossilization, and hence specialization, and consequently 'non-productivity' may occur. It is the pattern of nominal determining nominal (nominal > nominal). By both tests mentioned earlier it can be established that the items in question are sytagms (cf. Hervey — Mulder 1973), not morphological complexes. Examples are 'garage door', 'door bell', 'key box', 'door key', 'garage door key', 'garage door key box', 'garage door key box key' (bizarre, but possible), Dutch Hottentottensoldatententententoonstelling, and one of the possible English translations 'Hottentot soldiers' tents exhibition'. One could even have '(bottom door) > bell', i.e. 'bell of a bottom door', structurally opposed to 'bottom > (door bell)', i.e. 'bottom bell on or beside a door'. An item such as 'sledgehammer' is, of course, just as unanalysable as 'blackbird' or 'cranberry'. The form 'sledge' here has nothing to do with a 'sledge' as a vehicle used in the snow, but comes from Old English slug, related to Middle Dutch slegge, a large hammer, perhaps related to Modern Dutch slaan to hit, slag a 'blow', etc. But we could have, as the pattern is fully productive, an item 'sledge hammer', if there were, for instance, hammers that were generally used on sledges. And if they were only used on a particular type of sledge, we could equally well have 'hammer sledges'. Other examples of nominal > nominal are 'garden flower' and 'race horse'. There are, of course — precisely because this is productive, and because of recursivity, as we have seen in the rather bizarre earlier examples — theoretically speaking an infinite number of other possibilities. Note that we also have 'flower garden' and 'horse race', as well as — potentially — 'honey flower garden', 'rock garden flower', 'cart horse race', 'sulky race horse', 'flower garden shed', 'horse race manager', 'race horse manager', and so on. Nevertheless, the pattern in question may lead to some difficult problems. That is, when some of these items, though resulting from inherent structural productivity, catch on and are used fairly frequently in particular contexts or particular situations, they may become connotationally associated with those contexts

196

Linguistic description: General principles, and phonology

or situations, and hence become more or less conventionalized as a whole. For instance, 'race horse', 'cart horse', 'garden flower', etc. may become standardized and regarded as items in the lexicon, rather than as 'products' created in speaking as the need arises, just as constantly novel syntagms and sentences are produced without becoming standardized. Of course, there are cases in which the latter become standardized as well, in which case the following also applies to those. When such items become standardized, and acquire a particular connotation, there are still no real difficulties. If, say, 'rock garden' tends to refer to a stereotype, but may in principle refer to any garden involving a rock, or even strewn with rocks, we can at best assert that specialization qua denotation, i.e. in a 'wholly fixed' sense and hence fossilization of this originally complex entity, is likely to be on the cards at some time in the future. But even when this stage is reached we would still in principle have two entities, i.e. the specialized one and the potential one (because of productivity) which is not so specialized. An example of the latter would be 'fire engine'. This item is completely specialized and standardized and though it is reminiscent of the structure 'nominal > nominal' we have, in fact to regard it as being unanalysable, just as 'blackbird', 'blueberry', 'raspberry' and 'cranberry' are unanalysable. Notwithstanding that, however, given the fact that the structure of English is such that in principle any nominal can determine any nominal, one could, in principle, call any engine that has to do with fire a fire engine. The fact that the existence of a non-analysable 'quasi-homonym' makes it less likely that 'fire engine' in the sense 'fire > engine' will actually occur in speech is irrelevant. I should also like to say that, though I have presented it as if it were, it is not necessarily the case that pseudo-composites, i.e. unanalysable fossils, which are reminiscent of productive patterns, actually started life as such. In many, if not in most cases, such an item was probably invented, coined as it were, as a sign having a denotation that was not fully derivable from what might appear to be the components that motivated this particular coinage. That is, by - albeit sometimes superficial — analogy with existing forms, no matter whether fossilized, or productively arrived at, new forms may come into being. But the difference between this, and productive processes is that in

Productivity

and non-productivity

in so-called 'derivations'

197

the case of 'productivity' the items are not new, but already potentially there, whilst in the other case they are products of invention, and like all inventions accidental and arbitary, but nevertheless motivated, in this case mainly by analogy. I should like to stress again that the above considerations are not part of the theory as such, but are only indications as to where we should draw the borderline (note that 'borderline' is a fossil, but there is also a productive 'border line') between what is structural, and what is accidental, and therefore purely lexicological. We see that, where one could perhaps find apparent commutation between signs with identity of context, one is still not sure whether the commutation is a valid one. Failing to mention this was in my opinion, in retrospect, a serious omission in the article. We still have to ascertain that the complex exhibits what is loosely, in a kind of linguistic lingua franca, called 'productivity'. 1 have argued that 'productivity' should always imply 'full productivity', and that anything less, say, 'semi-productivity' is in fact 'pseudo-productivity'. For an item to be analyzable implies the following three conditions: (1)

(2)

(3)

There should be the possibility of commutation between any of the tentative components in identical contexts. This is the sign-identity criterion, fully dealt with in the article. The combination should be productive in a structural sense. This implies that the immediate constituents should, in principle, commute with any other member of the distribution class in question. Distribution-classes to be independently established. 'In principle' implies 'save conditions that are extra-linguistic'. There should be no 'wholly fixed' specialization in deployment or denotation. This actually implies both (1) and (2).

Now let us put this within the framework of a proper hypothetico-deductive method. It is clear that the hypothesis 'x is nonproductive' is a priori irrefutable, as we may at any time in the future find a member of the distribution class that does not commute with a particular item. We therefore have to assume that a tentative combination is productive, and hence a true combination — provided that non-productivity is not demonstrated.

198

Linguistic description:

General principles, and

phonology

But at the same time we see that refutation is not always a matter of merely looking and deciding. For instance, in order to demonstrate non-productivity we have first to refute that an assumed non-permitted member of a distribution-class does not in fact belong to another (independently established) distribution class or sub-class, that the assumed non-permittedness of that member is actually due to external, extra-linguistic conditions, which latter itself can only be refuted indirectly, i.e. by implication via other refutations, and it may involve work with informants. But especially at this point informants should, for obvious reasons, be listened to with great caution. The easiest way of looking at it is perhaps to say that simplicity is refuted by valid commutation, which latter not only implies sign-identity, but also productivity.

6. Productivity and non-productivity in so-called 'derivations' By 'derivations' one usually understands presumed morphological complexes, i.e. 'words' which are 'derived' from other words by affixation rather than inflection. Typically a derivation belongs to another category than the 'word' it is derived from, e.g. 'kindness' (a noun), derived from 'kind' (an adjective), 'worker' (a noun), derived from 'to work' (a verb), 'typist' (a noun), derived from 'to type' (a verb). But this is not always the case, e.g. 'boyh o o d ' (a noun), derived from 'boy' (a noun), 'lutenist' (a noun), derived from 'lute' (a noun), more uncertainly 'hatter' (a noun), derived from 'hat' (a noun), or from 'to hat' (a verb), itself — as is quite common in English — derived from 'hat' (a noun). Another typical feature of 'derivations' is that the forms in question are not closed, as 'inflected' forms tend to be, but have the same possibility of inflection or further derivation as other, e.g. 'simple', members of the category. It goes without saying that the above traditional view of 'derivation' has to be taken with a pinch of salt. In particular, among those 'derivations' we have to distinguish between real complexes, and pseudo-complexes, i.e. we have to apply the principles laid down in the previous few sections in order to

Productivity and non-productivity

in so-called 'derivations'

199

ascertain if the items in question are complex at all. I may be mistaken, and someone may provide a refuting example, but such items as 'kindness' and 'worker' are, as far as I can see, genuine complexes. The hypothesis of simplicity is refuted by valid commutation, including the corroboration of the hypothesis of productivity. As far as I can see, one can add '-ness' to several categories, e.g. 'table-ness', '/h/-ness', 'kindness', and I do not see any genuine, structural, intralinguistic restrictions, other than ridiculousness, redundancy, or other matters of irrelevantness and irrelevancy, including considerations such as cacophony. Of course, one will not find real monstrosities in the dictionary, but they are not structurally impossible - only subjectively ugly, and therefore unlikely to be encountered. Perhaps combinations with 'non-' are productive as well. As to '-er' derivations, it seems that those are in principle possible in the case of any main verb, e.g. 'a habitual kicker of empty tins', 'a teaser of little children', 'an eater of heaps of bean curd', 'a walker of great distances', 'a dreamer', 'a knitter'. Similar restrictions as to likelihood (orlikeliness, if you prefer) exist as in the other cases, e.g. 'assassinater' is possible, though unlikely, as 'assassin' is the common term. I wonder how adherents of the traditional 'derivation' view deal with 'assassin' versus 'assassinate'. Perhaps they will regard the latter as complex, but how, then, to deal with 'procrastinate', 'fornicate', and 'copulate'? And what about 'procurator' which is not the same as 'procurer', when there is no 'procurate', but only 'procure'. This is not to say that all '-ness' forms and all '-er' forms are of necessity complex. Just as in the case of pseudo-words, as we have seen, there is a strong possibility of specialization, and thereby fossilization, so also some of these forms may be pseudocomplexes. Though it is possible that, say, 'hatter' came into being via 'verbalization' of the noun 'hat', it may well have been coined by analogy with an, in this case productive, existing process. At any rate, its meaning is so specialized that there is no question as to its being unanalysable, and this may have been the case since it was used for the first time. It may also be the case that two realizationally similar items exist side by side, one unanalysable, and one of a productive complexity. An example is perhaps the unanalysable item 'goodness' as it is used in the context of health food, and so on, versus the complex 'goodness' which can be said of anything that is 'good'.

200

Linguistic description: General principles, and phonology

The traditional distinction between 'derivations' and other morphological complexes is, of course, not a very precise one. For instance, to which category would one assign present and past participles? Also, as we have seen, no distinction within 'derivations' is drawn between real complexes and pseudocomplexes. Nevertheless, there seems to be a difference in practice in many languages where there is a fairly rich morphology between items of which the morphological make-up plays a direct grammatical role, e.g. 'work' ~ 'worked', 'book' ~ 'books', French: 'donner' ~ 'donnerions', and so on, and items whose primary function seems to be to enrich the inventory of signs by coining new ones. The same seems to be true, in syntax, with regard to such syntactic complexes as 'garage door', 'laundry basket', etc., as opposed to other syntactic complexes, such as 'the big house', 'the little girl near the fountain', 'may not have been beaten', etc. It is, therefore, not surprising that within these inventoryenriching processes, specialization, and therefore fossilization, is likely to occur. These productive processes, derivative processes, if you wish, are thus an important source for the production of new signs in languages, but often the production of new signs by-passes the productive stage completely and new signs are produced that never went through a productive stage, but are reminiscent of those that did. This is what traditionally used to be referred to as 'word-creation by analogy'. Analogy is actually just one of the many, immensely complicated, types of motivation one finds in the creation of linguistic items. But it is a very important one, and moreover one that is easily, but wrongly, regarded as something synchronic, i.e. as something inherent in the structure of the language, rather than as something that has to be dealt with elsewhere, and probably not under the same theory.

7. Further notes on 'productivity' and potential pitfalls To show what may happen if one forgets that there is an essential difference between regular processes, and accidental ones, though

Further notes on 'productivity'

and potential pitfalls

201

motivated by analogy, let us look at such items as 'duck' ~ 'duckling', 'goose' ~ 'gosling', 'suckling', 'darling', 'goatling', 'codling', and 'sapling'. The items 'goatling' and 'codling' are listed by Munla (1982, see below). I have personally not come across them. This is definitely not a productive paradigm. I cannot even think of another example, let alone create one. But if one does not take productivity into consideration, one may easily (and some have done so) regard 'duckling' as being a complex of 'duck' and 'ling', the same for 'gosling', with 'gos' being an allomorph of 'goose', and what would stop one from regarding 'dar' in 'darling' as an allomorph of 'dear'? Once one takes the initial steps with 'duckling' and 'gosling', nothing stops one from asserting that 'sap' in 'sapling' is an allomorph of 'tree'. A bit more difficult is 'suckling', but one could at least express one's suspicion that 'suck' here is either an allomorph of 'sucker', or the denotation of the complex is that it signifies a 'ling' that 'sucks'. One may already begin to think that this is absurd, but these are direct logical consequences of disregarding the productivity issue. And there are further consequences that make it even more absurd. As one does not have 'pigling', but one does have 'piglet', one can scarcely escape the conclusion that 'ling' and 'let' are allomorphs. They certainly have intuitive semantic similarities as well. To the list we may as well add 'et' as in 'nymphet', perhaps not 'ette' as in 'launderette', but probably 'ette' as in 'majorette' and 'usherette'. All those have intuitive connotations ranging from small or young to frail or female. Once one has gone as far as this, nothing stops one from adding endings in 'y', and cognate forms to the list of the allomorphs in question. There are the nouns 'girlie', 'dogie', Scottish 'beastie', 'pussy' (where 'puss' would probably be an allomorph of 'cat'), and many others. All these are consequences, or at least possible consequences, that have to be taken seriously, once one has ignored the criterion of 'productivity'. The reader may perhaps think that much of this is rather far fetched, and no-one is likely to go as far as this. There are two things that can be said to this. Firstly, if one is not very strict on the 'productivity' issue, it is impossible to indicate a precise borderline between entities which have to be considered as possible candidates for analysis into constituents, and secondly, the reason why — in the absence of a 'productivity' criterion —

202

Linguistic description:

General principles, and

phonology

we might not go as far as we have done above is most probably either because the absurdity would stare us in the face, or because it would lead to insurmountable difficulties, or both. Neither, of course, is a scientifically respectable reason, although they do point to a serious flaw in the methodology adopted, i.e. in this case, not employing a strict 'productivity' criterion. At any rate, people have gone much further than the above where the absurdity was not so clear, and where the difficulties could be resolved, albeit by committing other errors, e.g. that of not properly applying the hypothetico-deductive method. Though, as I said, the absurdity was not directly visible, it was actually worse, because the absurdities became compounded. As an example I refer to a Ph. D. thesis by Munla (1982), which, in spite of the severe criticism I shall give below, in many respects does not lack interest and originality. I should first like to say that Munla is less to blame for what I am going to criticize than may appear, as he has tried faithfully to conform to the article 'Pseudocomposites' which did give insufficient guidance in these matters, and also because within the approach he tried to follow nothing else had been written or said on the subject elsewhere. Though at that time I was already aware of the importance of 'productivity', it was reading Munla's thesis that prompted me to think much more about it. At any rate, I did not see before then that 'productivity' should imply one hundred percent productivity, and that anything less is in fact 'pseudo-productivity'. Munla also deals with the 'ling' moneme, which he calls the 'young' moneme, 'being denotationally equivalent to the rough paraphrase 'baby x' ' (Munla 1982:246). Though he did not go as far as I have done in my examples, probably consciously or subconsciously inhibited by the fact that it would lead to an hilarious situation, he does list as an example of this moneme 'duck/wg', 'goatling\ 'codling', 'owlet\ 'piglet\ 'trout/ef', and 'gosling\ Though I do not wish to deal with this here again, it does indicate his stand with regard to 'productivity' and his consequential mode of reasoning. It also indicates something even more serious which goes all through this thesis, i.e. the semantic, in fact conceptualist, basis of his identity assessment. This is different from the way identity was dealt with in 'pseudocomposites'. It is also directly attributable to a faulty application of the hypothetico-deductive method. As the theory does not

Further notes on 'productivity'

and potential pitfalls

203

contain the notion 'different sign' (i.e. there is no definition covering this), in case two forms are different one cannot start by hypothesizing the opposite, i.e. 'sameness'. The reason is that the theory does not provide any means to refute such a hypothesis in principle, and it is, therefore, not a valid hypothesis. That is, if we have two or more different forms, and we are satisfied about their having sign-status, we must hypothesise that here we have two different signs with different denotations. The definitions of 'synonymy' enable us in principle to refute the latter, the definitions of 'allomorphy' enable us in principle to refute both. If, on the other hand, we have two signs of the same form, we must hypothesize that we are dealing with one and the same sign. This may or may not be refuted by applying the definitions of 'homonymy'. But unless there are such refutations we have to regard occurrences of different forms as different non-synonymous signs, and occurences of the same form as being occurrences of the same sign. The latter can, as we have seen, in some cases be overruled by refutation of the meta-hypothesis of 'adequacy' or 'consistency'. Munla deals with a vast number of 'derivations', amongst which a few are probably productive, some, e.g. the '-ly' forms which are 'adverbs' 'derived' from 'adjectives' are almost certainly purely allomorphic, but the vast majority is not productive. In respect of the latter, Munla's basic mistake is that he has dealt with them at all, i.e. that he has considered them as morphological complexes. Having done this, it is only logical that he should have arrived at allomorphs, but the fact that he has also established a number of different signs, or at least the way he has arrived at those, again exhibits an error, i.e. that of admitting assumptions that are a priori irrefutable as hypotheses. As I have indicated, this points in his case towards a conceptualist posture. Munla establishes for a particular group of derivatives (I shall here, for the sake of argument, ignore the 'productivity' issue) six different monemes, which he calls: 'participant', 'manipulator', 'author', 'specialist', 'occupational', and 'habitual'. That these are the six monemes is launched as a hypothesis. Such a hypothesis is however, a priori an irrefutable one. If one has, for instance, '-ist' of which one is satisfied that it has 'sign' status, one has to assume that there is only one sign involved. This is potentially refuted by aid of the definitions for ' h o m o n y m y ' and connected

204

Linguistic description: General principles, and phonology

ones, as I have said. But this is not what Munla has done. Furthermore, one has to assume that '-ist' does not have the same denotation as, say, '-er', though this may, of course, be refuted. If it is refuted, one has still to assume that two signs are involved, i.e. that this is a case of 'synonymy', which may be refuted by showing that it is a matter of contextual variance, which would lead to the conclusion that the two are allomorphs. This too, Munla has not done. Instead, he has implicitly or explicitly assumed 'homonymy' or 'allomorphy', as the case may be, in advance, and he has offered such hypotheses for refutation, overlooking the fact that the theory does not provide a means for refuting such hypotheses. As I have argued, if we hypothesize, say, twenty homonyms 'tree', this is irrefutable, unless, of course, we find in a proper fashion (i.e. by starting at one, then two, then three, etc.) that there are, say, twenty-one. Munla now establishes within the range of those six monemes a vast number of allomorphs. I shall simplify this, as only the principle is relevant here. In fact Munla does it in a much more sophisticated way than I can represent it here. I shall tabulate it by putting the different monemes horizontally, and — as I said: in a generalizing and simplificatory fashion — the different forms vertically. In the resulting slot I shall give one example. It may be the only one encountered, or it may be just one of a number of possible examples. The one with an asterisk is my own addition. According to this list, which is Munla's simplified, the 'participant' moneme has six allomorphs, three of which are homomorphs of allomorphs of the 'manipulator' moneme, three of the 'author' moneme, three of the 'specialist' moneme, two of the 'occupational' moneme, and one of the 'habitual' moneme. The 'manipulator' moneme has six allomorphs, two being homomorphs of allomorphs of the 'author' moneme, one of the 'specialist' moneme, two of the 'occupational' moneme, and one of the 'habitual' moneme. Also, three are homomorphs of allomorphs of the 'participant' moneme, but this was indicated already. I shall not mention in the following that which has already been indicated. The 'author' moneme has five allomorphs, three being homomorphs of all (i.e. three) allomorphs of the 'specialist' moneme. Two are homomorphs of two of the three allomorphs of the

t-l

a>

-man

-ster

-cian

chairman

musician

assistant

cyclist

poet

rhymester

theoretician

novelist

fabler

author

repairman

logician

psychologist

geographer

specialist

spinster

typist

occupational

trickster

*rapist

habitual

and potential

cabby

boatman

charioteer

lutenist

violinist

fifer

worker

spalder

manipulator

participar

Further notes on 'productivity' pitfalls 205

ε1

i

1

206

Linguistic description:

Genera! principles, and

phonology

'occupational' moneme, and of both allomorphs of the 'habitual' moneme. Of the three allomorphs of the 'specialist' moneme, one is 'homomorphic' with one of the allomorphs of both the 'occupational' and the 'habitual' moneme (which latter are, of course, therefore homomorphs in respect of one another). The three allomorphs of the 'occupational' moneme have two homomorphs in common with the 'habitual' moneme. So, we have the situation that those six signs all have homomorphs in common, and that only '-enist', '-y', and 't ~ m' as in 'lutenist', 'cabby' and 'poet' escape from being homomorphous with some allomorph of some other sign. Though this is not so, it could in principle have been the case that the six signs were completely homonymous, i.e. every allomorph of any one sign could have had five homomorphs, i.e. one with each of the other signs. And this could, in principle, have extended over vast sections of the lexicon. I am not sure what should be regarded more serious here: the 'conceptualism' involved, which has made this type of reasoning not only possible but logically compelling, or the fact that Munla did not see that such absurd conclusions should have made him wonder whether the theory he employed was not a completely inappropriate or inadequate one, or whether perhaps there was something seriously wrong with the application of that theory. Conceptualism is, of course, something that is entirely incompatible with this type of theory. As to the absurdity of the conclusions, I owe, as I have said, a debt of gratitude here, as it has made me think about the problems outlined. The above five assumed 'monemes' are only one example to be found in the thesis, but it is full of other examples of a similar kind of conceptualism, misapplication of the hypothetico-deductive method, and the failure to see that only full productivity can be productivity in a structural sense, and that otherwise all we can do is to list the forms in question as unanalysable within the systemology.

Further notes on 'productivity'

and potential pitfalls

207

8. Some further remarks It is of course not the case that in actual linguistic description, one always follows step by step, and one by one, all the logical procedures involved in faithfully applying the hypotheticodeductive method. But if one can establish that, if one had applied those procedures, the result would have been different, or even if there is serious doubt as to whether the result would not have been different if the criteria had been faithfully applied, or if it is clear, as in Munla's case, that there is no precise awareness of the logic behind the method, we are fully entitled to reject the results. Let us try to trace what is logically involved in coming to the conclusion that, say, 'the man in that chair may have bought a horse' is a syntactic complex. First we have to establish by refutation of the contrary that when we observe it as a phenomenon, it represents an instance of a speech-sound, that, further, though other such instances may have the same distinctive function, there are those that have different distinctive function. If we are less interested in actual phonological form than in the signum aspect, we may skip some further procedures and by refutation of the contrary come to the conclusion that the form in question has distinctive function at the grammatical level, and that the instance in question therefore corresponds to an utterance. It must consequently be a member of a signum. If we can establish by refutation to the contrary that there is denotation, this signum has the status of a 'sign'. We must now assume that it is a simple sign. This can be refuted (but only if the assumption of productivity is not refuted) by showing that there is valid commutation. If in this way we have established that it is complex, we must assume that it is a morphological complex. This can be refuted by criteria which we have discussed earlier. Once we have established syntactic complexes, we have to assign them to particular structural models which we establish. The criteria used in establishing these models are based on considerations of consistency, adequacy, and simplicity. Though there are still a number of theoretical notions involved which enable us to refute hypotheses, there comes a time when we

208

Linguistic description: General principles, and phonology

have to arrive at structures that we call descriptive models. These structures are basically arbitary, but their choice is limited, often severely so, by the above-mentioned considerations of consistency, adequacy, and simplicity. Such models are always in terms of their immediate constituents. We shall deal with them in extenso when we discuss syntax, but here I shall give the models involved in describing our example of a syntactic complex. To the left I shall give it in terms of structure, to the right in terms of implementation of that structure. 1st level of analysis: subject

the man in that chair extended -> predicative syntagm

-> may have bought

object

a horse

2nd level of analysis: (a) ["article Ί [numeral] [adjectival] [supplements]

the

J

> nominal (singular)

Φ Φ

> man J

[in that chair]

(b) [modal] [tense] [aspect] [modality] -I

> predicative

[may] [have]

Φ Φ

> bought

(c) art/numeral η [adjectival] [supplements]- 1

a π > nominal

Φ Φ

> horse J

The notion 'simultaneity'

209

3rd level of the analysis: functional (preposition)
chair

There are all sorts of representational conventions applied here which will be explained later. Examples are the square brackets, which indicate commutability with φ, except for the article and numeral in a nominative syntagm (singular), where it means that one of the two must be present. Where I put 'supplements' there should actually be several positions indicated, such as 'locative supplement', 'relative supplement', 'possessive supplement', 'qualitative supplement'. The notation 'x/y', here 'art/numeral' (in. the context of the indefinite article), indicates an archiposition (to be explained later). The terms are arbitrary and only refer to types of position (syntactic function), not to actual items. For instance in 'article' position one can find all sorts of non-articles, e.g. 'that', 'my', 'John's', 'my father's sister's daughter's' etc. This should give some indications of what I mean, indications which should suffice for the time being.

9. The notion 'simultaneity' Morphological complexes are self-contained simultaneous bundles of monemes. Monemes as such are, just as distinctive features in phonology, entities that can only occur in relations of simultaneity with one another. A morphological complex, if it can combine with other morphological complexes to constitute other self-contained complexes, is a syntagmatic entity in grammar. Some monemes are independent in the sense that they can occur

210

Linguistic description: General principles, and phonology

on their own to constitute syntagmatic entities, i.e. somemonemes may on the next level correspond to syntagmatic entities. Some may not even be combinable with other monemes at all, e.g. 'in', 'the', etc. Others may be, but are not always, e.g. 'book', 'work', etc. Minimum syntagmatic entities in grammar are called 'pleremes'. In many cases they correspond to the traditional notion 'word', but in a significant number of cases they do not. One should consult the relevant definitions, given in Chapter VIII. The same can be said of 'monemes', which in practice, but not always, correspond to 'morphemes', a term used by other linguists. The term 'monemes' was probably first used by Henri Frei, and is current among functionalists. Syntagmatic entities are entities that are orderable, or have immediate constituents that commute with orderable entities. But also entities that are intrinsically uncombinable at the tactic level are regarded as syntagmatic entities. An example of the latter would be 'yes', were it not that it can occur in conjunctive constructions (see Chapter VI). But there are examples in languages, e.g. the so-called 'question particle': ma, in Pekingese. In grammar, syntagmatic entities may, however, just as monemes, stand in a relation of simultaneity with other such entities as well. This is the case when there is not sub-ordination ( - > ) , but coordination (), or inter-ordination (< — > ) (see Chapter VI). Syntagmatic entities in phonology are called 'phonotactic entities'; syntagmatic entities in grammar are called syntactic, or plerotactic, entities. Though the notion 'simultaneity' (albeit for purely practical reasons defined in the theory) is not a notion that can be used to refute the contrary, i.e. 'simultaneity' is hypothesized, and can only be refuted by demonstrating ordering relations, it is in many cases counter-intuitive enough to some persons to warrant some, albeit rather informal, explanation. In the first place, 'ordering', does not mean physical ordering, but rather functional ordering, i.e. asymmetry of relations between entities in direct relation. Nor does simultaneity, i.e. nonordering, mean physical simultaneity. In fact, especially in morphology, which is the level on which we deal with simultaneous bundles of monemes, physical simultaneity is rare, but also in phonology one finds cases of physical non-simultaneity, i.e. sequential order, also called linearity. Examples of the latter

The notion'simultaneity'

211

are 'affricatives' in some languages, including German /pf/ as in Pferd. Because of the fact that in morphology there is usually linearity, some linguists, for instance French functionalists, do not distinguish what we call 'morphology' from 'syntax'. Their syntax, thferefore, roughly covers our 'morphology' as well as 'syntax', and perhaps even more (such as the sentential level, which, if dealt with at all, is regarded by most linguists as part of syntax). In our theory, linearity is not a matter of structure, but of realization. But, as Pilch (1964), has pointed out for phonology (but it is valid for morphology as well), simultaneity in realization must imply non-ordering, as there is no way in which ordering can be expressed, other than in a linear, or other spatial or temporal, way. The factor of 'allomorphy' (if one forgets that it is 'classes', not single members, that matter) may sometimes cloud the issue, and such cases as 'au', 'du', 'des', etc. may seem to be counter-examples. Note, though, that the 'complexity' here is not a matter of constituency, i.e. there are no direct grammatical relations between the 'parts'. The 'linearity' requirement implies, in fact, that we must in language have a level of morphology, as distinct from syntax, as there are such complexes as 'men', 'wrote', 'thought', etc. These simply could not be syntagms. It would, however, be absurd to regard them as simple entities, but, say, 'horses', 'walked', 'worked', etc. as syntagms, not intrinsically different in that respect from, say, 'the book on the table', 'John may have been drinking too much wine', etc. Besides, if one's theory is to have any semiotic significance at all, and ours is a semiotic theory, one has to consider that there are semiotic systems, e.g. the system of traffic-signs, that have a morphology, but not a syntax, whereas other systems, e.g. number-writing have a syntax, but not a morphology. I hope this highlights the intrinsic inadequacy of not distinguishing the two levels: there is an important difference between complexity within pleremes (roughly 'words') and complexity between pleremes. I believe that one of the good features of Bloomfieldianism, and some other schools, has been that they distinguish between an analysis of words (morphology), and an analysis of syntagms (syntax). On the phonematic level, few people have great difficulty with the simultaneity concept. Perhaps this is in the first place because physical simultaneity is the rule, rather than the exception, but

212

Linguistic description: General principles, and phonology

it is possibly also because within moneme-complexes there tend to be differences in intuitive semantic importance between constituents, and also differences in specialization. As to difference in intuitive semantic importance, this is in fact rather a difference in distinctive import. If a moneme belongs to an extremely large, perhaps virtually infinite, paradigm (e.g. 'noun', Verb', 'adjective', etc.), the set of monemes to which the moneme in question is directly opposed is extremely large, and hence the distinctive function of that item is great. It has a conspicuous functional load, and this has repercussions as to the intuitive semantic importance. In, say, 'worked', the moneme 'work' intuitively has a greater degree of 'meaning' than the 'past tense' moneme has, so it is difficult to conceive that the relation between 'work' and 'past tense' is the same as that between 'past tense' and 'work', i.e. that we have here a symmetrical relation nevertheless. On top of this may come the fact that some people find it difficult to ignore that there is linear order, i.e. to accept that this is structurally (though not realizationally) irrelevant. Nevertheless, and this is the only thing that counts, we cannot refute the only possible hypothesis here that it is as much 'work' and 'past', as it is 'past' and 'work'. If we put it in this way, it may even intuitively be more acceptable that this is a case of simultaneity. Some people may have difficulties with such items as 'unkindness', but also here — and this is the only thing that really matters — the simultaneity hypothesis remains unrefuted. Moreover, can one really say that intuitively it is necessarily more 'unkind' + '-ness', than it is 'un' + 'kindness'? The fact that 'un' and 'ness' (though this is not completely true, as one can have 'unness', just as one can have 'butness', 'ifness', etc.) are occurence dependent is, of course, irrelevant. Otherwise one would have to regard most 'words' consisting of more than two monemes, and phonemes consisting of more than two distinctive features, as tactic complexes. Suleiman (1984) (apparently linking it with (pseudo) criterion (c) of 'Pseudo-composites') comes along those bogus lines to the conclusion that 'unkindness' is syntactic, thereby, in fact, refuting the argument on which his conclusion is based (i.e. the immediate constituents are not the ultimate constituents). That is, if his conclusion were true, then his argument would be false, as in that case there would be 'parallel determination', i.e. un — > kind, ness

The notion 'simultaneity'

213

and the immediate constituents would be the ultimate constituents. I give this example in order to show how damaging it can be (a) to use notions in the theory for issues to which they have no relevancy, and (b) to overlook the logical consequences of what one is doing — not to mention failing to see the absurdity of it all. I have assumed here that 'un-' is productive (unlike '-ant' in Suleiman's other example: 'assistant'). Though the use of 'un-' is restricted in practice, it may, perhaps, in principle combine with any 'adjective'. As we have seen, 'wrote' must by necessity be a case of simultaneity, unless one regards it as simple, which would in the end violate considerations of adequacy, as well as simplicity, and perhaps under most theories consistency as well. More difficult is perhaps the question of specialization (in a different sense of this term from when I used it in connection with 'non-productivity') and its implications for further complexity. In syntax we have, as we shall see, the concept of nuclearity, which is very much connected with the concept of ordering relations. One of the properties of syntactic nuclei, e.g. 'man' in 'the man in the room', 'buy' in Ί buy a car', 'in' in 'in the park', is that they determine the distribution of the complex in question. The complex 'the man in the room' has the distribution of nominal complexes, Ί buy a car' that of predicative based complexes, 'in the park' that of prepositional syntagms. The constituents, 'in the room'. T , 'a car', and 'the park' have little, if anything to do with the potential distributional deployment of those sytagms. That such items as 'the', and to a certain extent, 'very' in 'very good' do is due to the fact that they imply a particular type of nuclei, but this is only the case when they are present. Their absence, e.g. 'books' rather than 'the books' or 'good' rather than 'very good' distributionally speaking makes little difference. Now if we look at morphological complexes, such monemes as ' b o o k ' in 'books', 'work' in 'worked', 'orum' in "Romanorum' (genitive plural 'of the Romans'; only the 'genitive' component is relevant here) also determine the distribution, just as nuclei in syntax do. If My' adverbs, such as 'greatly', etc. were not allomorphs (positional variants) of the corresponding adjectives, also '-ly' would be such an entity. There are, however, many things that they do not have in common with nuclei in syntax. The only thing I can see that they have in common with

214

Linguistic description: General principles, and phonology

syntactic nuclei is the specialization, which can be regarded as being of a lexical kind. And also in syntax one can regard the distributional aspect of nuclei to be lexically determined, as most distributional features seem to be, e.g. difference between 'transitive' and 'intransitive' based syntagms. In any case, such distributional facts do not refute assumptions of simultaneity. But the term 'nucleus' implies that there are 'peripheral' entities as well, and a relation between nuclear and peripheral entities cannot be a relation of simultaneity. Therefore, it would be inconsistent in morphological complexes to speak of 'nuclei'. At best they can be called 'quasi-nuclei'. They have some fringesimilarity with nuclei in syntax.

10. The phoneme as a simultaneous bundle of distinctive features As I have said already, the notion of simultaneity is most easily understood in the context of phonology. That phonemes are simultaneous bundles of distinctive features is readily accepted by most linguists who accept an analytical level below 'phoneme', i.e. a level of so-called 'distinctive features'. Now one has to be careful here, as, especially with American linguists, the term 'distinctive feature' tends to be understood either in a quasiphonetic, or in a Jakobsonian sense. Jakobson, as I explained earlier, has introduced a particular type of universalism, according to which there is a closed set of potential distinctive features from which each language, as it were, selects its own set. Though those features are basically phonetic and descriptive of the phenomena, they are said to be distinctive, not only because they are arranged in oppositional pairs, but also because they are only regarded as adopted by the language in question if the members of such a pair are in fact functionally opposed in that language (see Jakobson — Halle 1956). When functionalists speak of 'distinctive features', they are much closer to Trubetzkoy's (1929) ideas. What I and other functionalists mean is simply features which differentiate phonemes from one another. This implies that such features have

The phoneme as a simultaneous bundle of distinctive features

215

to be established in the context of phonemes. Phonemes, in their turn, have to be established in the context of phonotagms, so before I turn to the actual notion 'distinctive feature', I have to explain how one arrives at phonotagms, and what they are. The data, on which we base every aspect of our description are actually in the form of sentence-utterances. Though we frequently make shortcuts and focus upon parts, or even particular aspects of parts, of sentence-utterances, ultimately everything is logically speaking referred back to the sentence. I shall not go into that here, but at the level of the sentence the distinction between 'signum' and 'utterance' loses some (though not all) of its importance. This is because every sentence is unique, mainly because of intonational features. But, as I said, I shall not deal with this here. It will be dealt with in Chapter VI. A sentence consists of (though not as 'constituents') a base and sentential features. The base corresponds to (note that I do not say 'is') syntactic entities, not necessarily together forming a syntagm. Sentential unity is not the same as syntactic unity. It is the base we are interested in, as 'phonotagms' correspond to forms, or parts of forms, of entities of the base. All the rest we may ignore for our present purpose. If we compare sentences, and especially the bases of sentences with one another we find that differences in phonetic realization between certain 'chunks' may ipso facto lead to differences in the potential messages conveyed by sentences that have those bases. On the basis of formal criteria we find that within certain 'chunks' of phonological form, which we now have established in this crude oppositional manner, there are distributional restrictions but there are no distributional restrictions outside those 'chunks'. That is we have established phonological 'chunks' that are self-contained. It is these that we call phonotagms, or (instances o f ) distributional units. One of the alternative definitions for 'distributional unit' is 'minimum type of structure within which the distribution of cenotactic (phonotactic) entities can be described completely and exhaustively'. As we have not arrived at these phonotactic entities yet, the 'chunks' we select are, to begin with, tentative phonotagms. In description we often have to proceed in this way, fitting together, as it were, the pieces of a jig-saw puzzle. Another definition, which is equivalent, but which highlights a different aspect, is 'self-contained bundle of

216

Linguistic description: General principles, and phonology

positions in a cenological system (here: phonological system)'. Originally I have adopted the convention of calling 'distributional unit' also an instance of the more abstract model. Nowadays, however, I prefer the term 'phonotagm' for this, as it is nicely paired with 'syntagm' and fits in with the term 'phonotactics', a term which was itself inspired by terminology used in stratificational linguistics. As positions in instances of distributional units, i.e. phonotagms, are filled by phonemes, we can establish phonemes by comparing phonotagms. There again we have to use the proper hypotheticodeductive method, i.e. we assume that an observed formal difference is not functional, until this is refuted. By successive refutation in this manner we arrive at an inventory of phonemes. Once we have such an inventory we can compare phonemes with one another and detect functional differences in exactly the same manner. I must point out here yet another feature of our procedures. Commutation both in grammar and in phonology, is not only always between 'minimal pairs', but it is also always in the minimum context. That it must be between 'minimal pairs' is clear, as otherwise one cannot assess where the difference resides. That it must be in the minimum context is a necessary rider. If one enlarges the contexts, it is possible in a great number of cases to find what look like suspensions of distinctive function. For instance in English 'tall-boy' the feature 'unvoiced' at the beginning does not commute in any attested case with 'voiced'. It is clear that this is purely accidental. The same is true for the feature 'unvoiced' at the beginning of, say, 'talk'. This is within a phonotagm, unlike the case of 'tall-boy', but unless one can establish a structural rule according to which the phonological from /dork/ is impossible, i.e. if one cannot establish /dork/ as a structural impossibility one still has to regard this as accidental. Cases of this in syntax are even more abundant, as there the factor of 'semantic incompatibility because of the organization of 'the real world' plays a role. If, however, either in phonology, or in grammar, one is able to establish a structural rule, the phenomenon is usually described under the heading of distribution. The exception to this is in phonology (and theoretically speaking also possibly in grammar) when one establishes a case of neutralization. In the case of phonology we then speak of

The phoneme as a simultaneous

bundle of distinctive features

217

an archi-phoneme. I shall come back to this notion. But at any rate the minimal pairs for the establishment of phonemes are pairs of phonotagms, and the minimal pairs for the establishment of distinctive features are pairs of phonemes. Similarly, pleremes (roughly: words) are established within syntagms that do not themselves consist of syntagms, and monemes are established within 'pleremes'. Otherwise, one might, for example, come to the absurd conclusion that there is, by lack of distinctive function, no past tense mqneme in 'worked' within the syntagm 'John worked last Saturday at home'. It seems so obvious, but this is again an area within which o f t e n the wrong conclusions are drawn. Let us now compare the phonemes /b/, /p/, /v/, /f/, and /m/ in English. We also compare these with other phonemes, e.g. /d/, /t/, jdI, jQj, and /n/. Phonetically we could describe, say, /b/ as 'lenis' and 'voiced', unlike say /p/, / f / , /1/, and jQj. It is also bilabial, unlike /d/, /t/, /3/, jQj and /n/, and unlike /v/ and / f / , which are labio-dental in most cases, rather than bilabial. It is also 'unaspirated' unlike /p/ and /1/ in many, though not in all cases. Unlike /v/, / f / , /3/ and jQj it is occlusive. Unlike / m / and /n/ it is non-nasal. We may find other phonetic peculiarities, but these suffice for the moment. We have, then, say, for /b/ as possible candidates for distinctive features the set: lenis, voiced, bilabial, unaspirated, occlusive, and non-nasal. The hypothesis that any of these is a distinctive feature is a priori irrefutable, and therefore not valid. We therefore assume of none of them that it is distinctive, though we know that there are one or more distinctive features involved, as we have already established /b/ as a phoneme, distinct from other phonemes. Only when our assumption of non-distinctiveness is refuted, do we assert that the phonetic feature in question here has distinctive function and is therefore a distinctive feature. The refutational instances come from the minimal pairs test. We ask ourselves, when, say, assessing the distinctiveness of 'lenis' if there is any phoneme in the language that is not 'lenis', but that is 'voiced', 'bilabial', 'unaspirated', 'occlusive' and 'non-nasal'. We do not find such a phoneme. But if we ignore the feature 'voiced', or we do the same test for 'voiced' and ignore the feature 'lenis', we find /p/, which is 'fortis' rather than 'lenis' and 'unvoiced' rather than 'voiced', but which is for the rest exactly the same as /b/. We have then established that there is no functional distinction between 'lenis' and 'voiced'

218

Linguistic description: General principles, and phonology

but that either, or b o t h together, can be regarded as a distinctive feature. It is arbitrary whether we call the feature 'lenis' or 'voiced', or whether we give it any o t h e r name. Let us call it 'voiced'. I have in anticipation o f findings when we deal with /p/ already ignored 'aspirated' for two reasons: firstly it only occurs with 'unvoiced' and 'occlusive', and is therefore not a distinctive feature in respect o f t h e m , and secondly it is dubious anyway that it could be a candidate, as generally speaking it only occurs before vowels in stressed syllables, when it is not preceded in the same phonotagm by /s/. We now turn t o 'bilabial' and ask if there is a p h o n e m e that is not 'bilabial', but otherwise the same as jbj. There is, namely /d/, which is 'voiced, occlusive', and 'apical' rather than 'bilabial'. Again leaving out 'unaspirated', we now turn t o 'occlusive'. Is there a p h o n e m e that is e x a c t l y the same as / b / , but which is not occlusive? T h e r e is none at first sight. One is at this stage not sure if 'nasals' are distinctively 'occlusive' as well, and under those circumstances the phoneme that c o m e s nearest t o being a candidate for a minimal distinction is /v/. However, the difference is that /v/ tends to be 'labiodental', rather than 'bi-labial'. It is now found that the difference between 'labio-dental' and 'bilabial' exactly corresponds t o a difference between 'fricative' and 'occlusive' in the case o f labiality, and we now regard the two as one and the same feature and call it just 'labial'. Turning t o 'non-nasal' we find that this is an a u t o m a t i c property o f both 'occlusive' and 'fricative' and therefore not a distinctive feature. We see then that / b / is completely characterized by assigning to it the distinctive features 'voiced', 'occlusive' and 'labial'. It is 'voiced' because o f jbj ~ /p/, 'occlusive' because o f / b / ~ j\j, and 'labial' because o f jbj ~ Id/. T h e p h o n e m e /p/ is exactly the same as / b / , except that it is not voiced; /v/ is the same as jbj, e x c e p t that it is not occlusive; /d/ is the same as / b / , e x c e p t that it is not labial. Because the pairs differ in only one feature, they are called minimal pairs. As we c a n n o t establish functional ordering-relations between 'voiced', 'occlusive', and 'labial', which would in this case be an a priori impossibility anyway because o f simultaneity in realization ( c f . Pilch 1 9 6 4 ) , / b / is a simultaneous bundle o f these distinctive features. As it is a simultaneous bundle, it is irrelevant in which order we put the distinctive features in the presentation. T h e most

The phoneme as a simultaneous

bundle of distinctive features

219

usual seems to be 'labial, occlusive, voiced' in writing, or 'voiced, labial, occlusive' in speaking, with 'voiced' as a sort of 'adjective' to 'labial' and 'occlusive'. That 'voiced' traditionally is written last, or is used as an adjective, seems well-motivated, as it reflects what phoneticians call a secondary articulation. Traditionally, Prague-school scholars have tended to regard the so-called 'orders', which tend to reflect place, rather than mode of articulation as slightly more important than the so-called 'series', which tend to reflect mode of articulation. But it should be said that there is here some parallel with morphology, where we have seen that certain monemes may be more determining to the distribution of the whole. Namely the features of the 'series', i.e. 'occlusive', 'fricative' and 'nasal' tend to be more important in that respect. Occlusives, fricatives, and nasals, tend to have different distributions, whereas labial, apical, and dorsal occlusives tend to have similar distributions, and the same goes for fricatives and nasals respectively. It might therefore seem that we might as well, for convenience sake, list them in the order 'series, orders, secondary', e.g. 'occlusive, labial, voiced'. But against this is the consideration that what are traditionally called secondary articulations reflect differentiations within series, rather than within orders. It is therefore a matter of good practice, in descriptions, to take this into consideration, if not in the order we list distinctive features, then in the way we set up phoneme-tables. It is reflected in the habit of talking about 'voiced occlusives', rather than, say, 'voiced labials', let alone labial or occlusive 'voiceds'. I have already referred earlier to the importance of 'secondary' articulations in this respect. There are, however, strong structural (if not phonetic) arguments in favour of regarding also such distinctions as between 'dental', 'hissing', and 'hushing' (e.g. / 3 / ~ / z / ~ / 2 / in English, see below) as being similar to distinctions within series. The differences in realization between thpse too could, phonetically speaking, be regarded as 'secondary articulations'. In fact, also these distinctions are prone to neutralization, e.g. in final position in such widely divergent languages as Korean and Thai. See below for the consequences in respect of setting up phoneme tables, which deviates from the practice I have followed in earlier publications. Let us now return to /p/. The establishment of distinctive

220

Linguistic description: General principles, and phonology

features of phonemes belonging to an integrated system becomes progressively easier as we proceed. The distinctiveness of the feature 'labial' is established, i.e. the hypothesis of its non-distinctiveness is refuted, by the comparison /ρ/ ~ /t/, as /t/ is exactly the same as /pi, except that it is not labial. The distinctiveness of 'unvoiced' is established by the comparison /ρ/ ~ /b/, and that of 'occlusive' is established by comparing it with / f / . It is easy to establish in the same way the distinctiveness of'labial', 'fricative', and 'voiced' for /v/, and of 'labial', 'fricative' and 'unvoiced' for /f/. Turning to /m/, we have seen already that any assumed occlusivity is a necessary by-product of 'nasal', so it would be redundant to mention it when we have already mentioned 'nasal'. Possible candidates for distinctiveness apart from 'nasal' may seem to be 'labial' and Voiced'. The feature 'labial' is established by the comparison /m/ ~ /n/, /n/ being exactly the same as /m/, except that it is not labial. But with regard to the feature 'voiced', we do not find a phoneme that is 'labial' and 'nasal', but not 'voiced'. It is therefore not a distinctive feature. But the feature 'nasal' is. Each of the phonemes /b/, /p/, /v/, / f / , is labial, but not nasal. That they are either occlusive, or fricative, and voiced or unvoiced as well is irrelevant, as those features have already been ruled out as being distinctive features within /m/, and in fact, as will be seen when we progress, within any of the nasals. We can say, in fact, that the difference between /ml and /b/ is the same as that between /m/ and /p/, or /v/, or /f/. So in a sense we might distinguish between nasal and non-nasal, and within the latter between occlusive and fricative, were it not that 'nonnasal' is concomitant with 'occlusive' as well as 'fricative', and it is therefore not a distinctive feature. Mentioning it would therefore be redundant. This is why, for instance, for English we have a three-way distinction between occlusives, fricatives, and nasals.

11. Phoneme tables as practical descriptive devices An easy representation of sets of phonemes that have distinctive features in common is, as I have said, by way of phoneme tables. I shall not go into great detail here as to problems and solutions

Phoneme tables as practical descriptive devices

221

with regard t o such tables, as I have done this in extenso elsewhere (see Mulder 1968, and especially Mulder 1978, which latter is reprinted in Mulder — Hervey 1980. It should, however, be n o t e d that I have n o w some better solutions, by which several of the problems referred t o in Mulder 1978, and Mulder — Hervey 1980, no longer exist). Before I refer again t o t h e consonant table of English, I shall present one for D u t c h , which in first instance exhibits fewer complications t h a n t h e English one.

Occlusive

Fricative Nasal

voiced

unvoiced

voiced

unvoiced

Labial

b

Ρ

V

f

m

Apical

d

t

ζ

s

η

g

X

Dorsal

k

1

There are also t h e consonants /h/, /1 / and jxj, which fall outside t h e system, because for these we can only establish one distinctive feature each. F r o m this table we can easily read the distinctive feature make-up of each of the phonemes. N o t e t h a t t h e symbol jgj here represents a 'dorsal voiced fricative', not a 'dorsal voiced occlusive' as in usual English representations. We see that, as in D u t c h there is n o f u n c t i o n a l distinction between 'voiced' and 'unvoiced' in 'dorsal occlusives', neither of those can be said t o be distinctive features, in /k/. T h o u g h in m o s t , b u t n o t all, realizations 'unvoiced', this is no m o r e a distinctive feature, than say 'voiced' is in 'nasals' in English. In fact 'nasals' in English may on occasion be 'unvoiced' in realization, and they may on occasion be 'labiodental', whereas 'fricatives' m a y b e 'bi-labial'in realization. An example is the word ' n y m p h ' where e i t h e r ' m ' m a y be realized labio-dentally, or ( t h o u g h rarely so) ' p h ' may be realized bilabially. As / k / in D u t c h does n o t exhibit differentiation between 'voiced' and 'unvoiced' it represents b o t h of these features in comparison with djt and bjp, i.e. it is exactly as different f r o m

222

Linguistic description:

General principles, and

phonology

/d/ as it is from /t/, and as different from /b/ as it is from /p/. This is a similar situation to the case of archiphonemes, with the difference that in the case of archiphonemes there is differentiation in at least one other context. I have called such cases as /k/ in comparison with djt or bjp hyperphonemes. In a similar fashion I have coined the term 'hyper-features' where there is differentiation between features in some cases but not in others. For further details I refer to Mulder 1978, reprinted in Mulder Hervey 1980. Actually, the fact that paper is two-dimensional is a drawback in the representation of phoneme-tables. In the phoneme-table for Dutch there are three dimensions involved, first 'labial, apical, and dorsal', then, as the second dimension 'occlusive, fricative, and nasal', and built upon the 'occlusive/fricative' part of this (with the exception of the /k/ part) the dimension 'voiced, unvoiced'. Taking into consideration things that have been said above, especially the 'secondary articulation' issue, and the notions 'hyperphoneme' and 'hyper-feature', we can now come to a much better representation of the English consonant system. The phonemes /h/ and /1 / have, as we shall see, only one distinctive feature each, and I classify /r/ as a semi-vowel. The table becomes then as follows:

Occlusive

Fricative Nasal

voiced labial

b

unvoiced Ρ

dental hissing

d

t

hushing dorsal

g

k

voiced

unvoiced

ν

f

9

θ

ζ

s

ζ

§ χ

m

η

1

Phoneme tables as practical descriptive devices

223

We can, as we have done for /k/ in Dutch, call /d, t, n/ 'hyperphonemes' in comparison with /3, 0, z, s, z, §/, and we can call 'dental/hissing/hushing' a hyper-feature. To avoid clumsiness one may call this hyper-feature 'apical', as long as one does not forget that here 'apical' equals 'dental/hissing/hushing' (as a logical product). In a similar way we can call /x/ a 'hyperphoneme', and 'voiced/unvoiced', which in this case is just 'fricative', a 'hyper-feature' in respect of /x/. In the table I have included /x/, which is current in Scottish and Irish varieties of English, not just in dialects, but in varieties of Standard English. If we leave this out, we still have a difficulty, i.e. the terms 'occlusive' and 'nasal' do not have the same meaning as elsewhere, as here 'occlusive' and 'nasal' are not opposed to 'fricative', and hence they do not have the same distinctive function, i.e. phonological identity, as the features of the same name have in the context 'labial' or 'apical'. For solutions to problems of this kind see Mulder 1978, reprinted in Mulder — Hervey 1980. The phoneme-table for the Peking dialect of Chinese can be used to highlight some other points of procedure. It is as follows:

Unaspirated

Aspirated

Fricative

labial

b

Ρ

f

apical

d

t

hissing

ζ

c

hushing

i

c

s

dorsal

g

k

h

Nasal m

s η

Τ

Here the hyper-features (products of features elsewhere in opposition) are 'apical/hissing' for /s/, and 'apical/hissing/hushing' for /n/. The phoneme /s/ is a hyperphoneme in comparison with Id, z, t, c/, and the phoneme /n/ a hyperphoneme in comparison with /d, z, z, t, c, c, s, S/. The realization of the unaspirated consonants is, in fact, fairly fortis, and ««voiced, as in Thai generally even more/br//.s than that of the aspirated consonants. This creates

224

Linguistic description: General principles, and phonology

a difficulty for many Anglo-Saxon speakers, who tend to reproduce the unaspirated consonants (as in English) as lenis and voiced. This native bias probably deludes their ears in hearing a voiced quality where there is none. The letters one chooses for representing phonemes are arbitrary, but not unmotivated. One of the considerations is often availability on the type writer. One should not read any direct pointers to realization in the choice of the letters themselves. In this table for Pekingese Chinese 'z', 'c', '2', and 'c' indicate affricates; /z/ is roughly [ts], /c/ is roughly [ts'J, whereas /2/ and /C/ are the corresponding 'hushing' phonemes, in realization involving, roughly, retracting and flattening the tongue, though with scarcely any palatalization. The same goes for /§/, as opposed to /s/, which latter is not unlike English [s]. The phoneme /h/ in Pekingese is not a glottal aspiration, but a clear dorso-pharyngal fricative in most realizations and can therefore unproblematically be classified here. Though 'unaspirated' and 'aspirated' correspond phonetically to occlusives, the feature 'occlusive' is not a distinctive feature, and hence not represented in the table. It is already implied by 'unaspirated' and 'aspirated'. The following is a table such as is used in Jakobsonian types of phonology, but which I have adapted to our different concept of 'distinctive features'. It illustrates, perhaps better than the usual functionalist tables do, what is actually involved. In this table the top-row lists the phonemes as in the previous table.

b p f

m d t

n z c s Z C S g k h i ]

ι ι ι ι ι ι ι I ι ι I I ι + + + +

Unaspirated Aspirated Fricative Nasal

+ - -

ι Ο ι ι Ο ι ι + ι ι + + + ι Ο I I

Labial Apical Hissing Hushing Dorsal

Ο ι ι ι ι ι ι + + + + I ι Ο I I I Ο ι ι ι + + + I ι ι + I ι Ο I I I + + + + ι ι ι I ι ι I I ι ι I I I -

+ - -

+ -

- +

- -

+ -

- -

- - + - 0 0 0 - - + - - + - - + - . + - - + 0 0 0 0 0 0 +

Phoneme tables as practical descriptive devices

225

The column lists all the features involved, and in the slots we find either '+', signifying that the phoneme in question possesses that feature, '—' indicating that the latter feature is functionally opposed to the feature corresponding to this '—', and Ό ' that it is not opposed to the feature corresponding to this Ό ' . We see that the table not only indicates the features of which the phoneme in question is composed, but also the distinctive function of those features. Note how the hyperphonemes /n/ and /s/ are rendered in the table in terms of the hyper-features 'apical/hissing/hushing' and 'apical/hissing' respectively. As a further illustration I shall give the Jakobsonian type table for the consonants of English, with the inclusion of /x/. b p v f

m d t

5 0 n g k x r ) z s

i s

1 ι 1 1 1 1 1 ι 1 1 1 1 1 + + + + +

Dorsal

1 1 1 1 + + + + 1 1 1 1 1 1 1 1 1 1

Occlusive Fricative

+

1 ι 1 1 1 1 1 ι + + + + + 1 1 1 1 1

Labial Dental Hissing Hushing

1 ι + + 1 1 1 ι + 1 1 + + 1 1 1 1 1 + + 1 1 1 1 1 ι + 1 1 + + 1 1 1 1 1 -

+ -

+

+

-

+

+

-

-

-

+

+

-

+

+

-

-

-

+

-

-

+

-

+

+

+

Nasal

Voiced Unvoiced

+ - +

- 0

+ - +

- 0

+ - +

- + - + 0 - + - + 0 - +

0 + - +

+ 0 -

+ -

+

One should note in this table that in the case of /x/ both 'voiced' and 'unvoiced' have a '+' notation, and so have in the case of /d, t, η/ the features 'dental', 'hissing', and 'hushing'. Though the usual functionalist type of table more directly shows the proportionality between phonemes, and its use is in fact mainly justified by this, the Jakobsonian type table is more directly revealing as to the identity of the distinctive features involved. None of the two tables is better than the other. They simply serve different purposes.

226

Linguistic description:

General principles, and

phonology

It is seldom the case that one can bring all 'consonants' of a language together in a scheme. Not fitting into the scheme at all are in English 'h' and ' Γ . That is, we cannot establish for either 'h' or Ί ' features that are functionally opposed to other features, i.e. distinctive features. Suppose that for 'h' we were to pose aspirateness, then we still could not find any phoneme that is the same as /h/, except that it is not 'aspirated'. The only feature one could pose for /h/ is /h/-ness, and the same goes for /I I, whose feature is /l/-ness. In Southern standard British English, but not in Northern dialects and Scottish Received Pronunciation, /r/ is not a 'consonant', but a 'semi-vowel', as we shall see. But even in those British dialects where one has both /1 / and /r/ as consonants, which phonetically one could call 'liquids' in those dialects for most of their realizations, one cannot establish more than one distinctive feature for either. One could find another phoneme than /I / which has a 'liquid' type of realization and differs from /1 / only in one respect, i.e. instead of being 'lateral' it is 'retroflex'. The phoneme in question is, of course, jrj. But one cannot find a phoneme that is 'lateral' but not 'liquid', nor one that is 'retroflex', but not 'liquid'. That is to say that 'lateral/liquid' is only one distinctive feature and so is 'retroflex/liquid'. To put it in another way, one cannot find a phoneme that is 'lateral', but not /1 /, or one that is 'retroflex', but not /r/. So we see that 'lateral', 'lateral/liquid', and /l/-ness, are one and the same, and so are 'retroflex', 'retroflex/liquid', and /r/-ness in dialects where /r/ is a consonant. How many distinctive features there are bundled into one phoneme depends on the number of dimensions involved, and there is only a distinctive feature corresponding to a particular dimension if there are at least two phonemes pertaining to the dimension in question.

12. Consonants, vowels, semi-vowels, and semi-consonants A distinction between 'consonants' and 'vowels' is one of the controversies within phonetics. Though there is no difficulty in

Consonants,

vowels, semi-vowels, and semi-consonants

227

classifying m o s t of t h e p h o n e m e s with t h e o n e o r with t h e o t h e r , t h e r e are o f t e n b o r d e r l i n e cases. Also, it is d i f f i c u l t t o establish a p h o n e t i c d e f i n i t i o n t h a t is fully s a t i s f a c t o r y . F o r instance, if o n e regards as vowels p h o n e m e s t h a t in realization c o m e straight f r o m t h e glottis, w i t h o u t having any i m p o r t a n t c o n s t r i c t i o n s in t h e m o u t h , it might be d i f f i c u l t n o t t o regard [ h ] as a vowel, t h o u g h in m o s t languages it behaves m o r e like c o n s o n a n t s d o . Also fairly n a r r o w vowels, like English /i/ a n d / u / , G e r m a n and F r e n c h / y / ( o r / i i / ) , in m a n y realizations c o m e close t o w h a t o n e w o u l d r a t h e r call ' c o n s o n a n t s ' . I shall n o t a t t e m p t a n y p h o n e t i c d e f i n i t i o n , b u t p r o p o s e t o classify p h o n e m e s as 'vowels 1 w h e n t h e y always s t a n d in a n u c l e a r p o s i t i o n (see b e l o w ) in a distributional unit ( w h i c h in p r a c t i c e implies t h a t t h e y c o r r e s p o n d t o a syllabic nucleus). I p r o p o s e t o call ' c o n s o n a n t s ' t h o s e p h o n e m e s t h a t never s t a n d in n u c l e a r p o s i t i o n , i.e. t h a t a l w a y s stand in peripheral positions. This leaves us w i t h a possible c a t e g o r y of p h o n e m e s t h a t m a y stand in e i t h e r p o s i t i o n . In m a n y languages t h e s e include p h o n e m e s t h a t we a p p r o p r i a t e l y r e p r e s e n t by /i/ a n d / u / , b u t t h e r e are s o m e t i m e s o t h e r s , e.g. / y / ( o r / ü / ) in s o m e languages, a n d / r / in S o u t h e r n English, as we shall see. In s o m e languages, o n e m a y have s o m e , p h o n e t i c a l l y s p e a k i n g , m o r e ' c o n s o n a n t - l i k e ' p h o n e m e s , e.g. / m / , / η / , / l / , etc. t h a t m a y o c c u r in n u c l e a r p o s i t i o n , b u t also in peripheral positions. If t h e y b e h a v e in t h a t respect e x a c t l y like /i/ a n d / u / , etc., we have t o regard t h e m as being of t h e s a m e k i n d . It is t h e n a r b i t r a r y w h a t we call t h e m , p e r h a p s e i t h e r 'semi-vowels' o r ' s e m i - c o n s o n a n t s ' . B u t , as is usually the case in such languages, if s o m e , say /i/ and / u / , have g r e a t e r a f f i n i t y w i t h n u c l e a r i t y , t h a n o t h e r s , say /ml, / n / a n d I I I , we o u g h t t o distinguish b e t w e e n t h e m , e.g. by calling t h e o n e class 'semi-vowels' a n d t h e o t h e r ' s e m i - c o n s o n a n t s ' . S u p p o s e we have a n imaginary language in w h i c h we have all f o u r categories: c o n s o n a n t s / p / a n d / t / , vowels / a / a n d / o / , semic o n s o n a n t s /ml a n d / n / , a n d semi-vowels /i/ a n d / u / . S u p p o s e t h e r e are t h r e e positions. We can t h e n h a v e : / p a t / , / p o t / , / p m t / , / p n t / , / p i t / , / p u t / , / t a p / , / t o p / , etc., / p a / , / p o / , etc., / a t / , / o t / , etc., / a / , jo/, /m/, / n / , .HI, / u / , / i a t / , / i o t / , etc., / p a i / , / p o i / , etc, / p a m / , / p o m / , etc., / m a p / , / m o p / , etc., / m i / , / m i p / , / m i n / , / n i m / , etc. T h a t is, if t h e r e is / a / o r / o / , t h e n o n e of t h o s e is t h e nucleus. If t h e r e is I'll o r / u / , b u t n o t / a / o r / o / , t h e n I'll o r / u / is t h e nucleus. If t h e r e is / m / o r / n / , b u t n o / a / , / o / , /i/, o r / u / t h e n / m /

228

Linguistic description:

General principles, and

phonology

or /n/ is the nucleus. To have a syllable there must be either /a/, or /o/, or j\/, or /u/, or /m/, or /n/. But we see that there is a hierarchy of affinity with nuclearity or its converse. The phonemes /a/ and /o/ have more affinity with nuclearity than /i/ or /u/, and the latter have more affinity with nuclearity than Im/ or /n/. The phonemes /p/ and /t/ have no affinity wjth nuclearity at all, so we can reverse all this and say that /p/ and / t / have more affinity with peripheralness than /m/ and /n/, and so on. The phonemes /a/ and /o/ have no affinity with peripheralness at all. All this may seem trivial, but there is some parallel with syntax. There are pleremes that have in practice greater affinity with nuclearity than others, e.g. 'verbs' more so than 'nouns', 'nouns' more so than 'adjectives' (to put it all in a very rough and crude way), and in communication this is a great pragmatic aid to the understanding of the message conveyed, reducing ambiguity, and other factors which otherwise could make communication more difficult. It is also an important aid to the describer of languages, for instance when he establishes syntactic models o n t o which speech-phenomena can be mapped. In other words, just as in phonology there are categorial phonemic clues as to phonotactic structure, in grammar there are lexical clues as to syntactic structure.

13. The notions 'position', 'nuclearity', and 'peripheralness' in phonology Because what we have agreed to call 'vowels' only occur in nuclear position, but what we have agreed t o call 'consonants' only occur in peripheral positions, vowels and consonants do not commute with one another. Semi-vowels and semi-consonants do commute with vowels, and may commute with consonants. The reason why they must commute with vowels, but not necessarily with consonants is that there is only one nuclear position in a distributional unit, but generally more than one peripheral position. To give an idea of what I mean by 'positions', let us look at such forms as 'sprinkle' or 'scramble'. Apart from some additions

The notions 'position'nuclearityand

'peripheralness' in phonology

229

which are either forms of monemes, or derive from allomorphy or from historical accidents (see below) — I shall call these 'morphemes', which more or less coincides with American usage — the above forms may seem at first sight to correspond to the maximum of filled positions in phonotagms. That is, the distributional unit underlying phonotagms in English basically seems to contain seven positions. It can be seen that I use the term 'morpheme' for convenience sake, i.e. not for the form of a sign, nor for a phonological form as such, but for such phonological entities or clusters of phonological entities as are there because of grammatical or diachronic peculiarities which, of course, are accidental from the point of phonology. For instance, if one has a noun of a particular form, it is almost certain that another form exists which corresponds to a plural; if there is a verb of a particular form, one tends also to have the third person singular, as well as the past tense of that verb. In addition I shall call morphemes (remember that this is a term of pure convenience) those phonological features that are actually forms of pseudomonemes, as they do not partake in productive processes. Accordingly one can have 'sprinkles' and 'scrambles' (third person morpheme); 'sprinkled' and 'scrambled' (form of a past tense moneme); and 'besprinkled' (form of a pseudo-moneme 'be-', as well as a form of a past tense moneme), etc. As the basic distributional unit I shall regard the set of underlying positions of phonotagms that do not exhibit those accidental extensions. I shall call the positions: pre-explosive, first explosive, second explosive, nuclear, first implosive, second implosive, and third implosive respectively. The term 'explosive' in realization means: progressive opening towards the nucleus, the term 'implosive' means progressive closing away from the nucleus. These terms are inspired by Saussurean ideas about the syllable, and reflect the basic 'ballistic' type of realization for syllables, a realization which tends to go from silence, or maximum closure, via highest point of sonority, or maximum opening, to silence or maximum closure again. It is on this basis that the positions are established, but once established it is not necessarily the case that the succession of phonemes is in agreement with this, and it may be that the phonological syllable, or rather basic distributional unit, does not coincide with an actual phonetic syllable. According to what I just said, one should regard as the most natural fillers

230

Linguistic description: General principles, and phonology

for the most peripheral (i.e. most away from the nucleus) positions the 'occlusives'. After that the continuants such as fricatives, nasals and liquids, and nearest to the nucleus such phonemes as semi-vowels. Though semi-vowels, when in peripheral position, are usually found (though not always) to be nearest to the nucleus, the order of consonants may vary, in which case there is often a parasitic i.e. non-functional [a] type sound involved in the realization. Both 'sprinkle', and 'scramble' are examples, i.e. there may be in realization a parasitic [a] sound between [k] and [1], and [b] and [1] respectively. Setting up the positional scheme we can map phonotagms onto it in the following manner. I shall use for pre-explosive, first and second explosive, nuclear, and first, second and third explosive the abbreviations pre, 1, 2, η, 1, 2, 3, respectively. The capital letters represent archi-phonemes, to be explained later. The sign V indicates a 'potential' filler. When a filler stretches over more than one position, there is an archi-position, to be explained presently. The reason why I have set up the scheme like this, i.e. with seven basic positions, is that I could not think immediately of 'basic' forms that had more than seven phonemes. I could start off with seven, as by finding the forms 'sprinkle'and 'scramble', the hypothesis that there are less than seven was implicitly refuted. The hypothesis as it stands is, then, a basic distributional unit of seven positions. A research-student, however, Fawzi El-Shakfeh, has pointed out that a form such as 'fore-castle' is /fOukSl/. Other, more ordinary, forms of this kind include 'counsel', i.e. /kaunsl/, and several others, so that the hypothesis of having only three implosive positions seems to be refuted. It has, therefore, to be investigated if the semi-vowels, should not be put in a separate position, e.g. 'scramble' may have to be mapped as follows, in comparison with, say, 'fore-castle', 'counsel', and 'sample' (the latter suggested by M.A.L. Lamb).

The notions 'position', 'nuclearity', and 'peripheralness' in phonology

231

pre

1

2

η

1

2

3

scramble

s

Κ

r

a

Ν

b

1

sprinkle

s

Ρ

r

i

Ν

k

1

ram

φ

Φ

r

a

m

Φ

Φ

ramble

φ

Φ

r

a

Ν

b

1

pram

φ

Ρ

Γ

a

m

Φ

Φ

write

φ

φ

r

a

i

Φ

t

pray

φ

Ρ

r

e

i

Φ

Φ

stray

s

τ

r

e

i

Φ

Φ

strain

s

τ

r

e

i

η

Φ

straight

s

τ

r

e

i

Φ

t

u

e

1

f

θ

Γ

e

i

t

η

θ

Φ

twelfth

t

straighten (?)

s

τ

length

φ

say

s

φ

Φ

e

i

Φ

Φ

aim

φ

φ

Φ

e

i

m

Φ

rhyme

φ

φ

τ

a

i

m

Φ

brown

φ

b

r

a

u

η

Φ

small

s

m

0

r

1 «

Φ

slower (?)

s

1

ο

u

Φ

Γ

sloper (?)

s

1

ο

u

Ρ

r

boater (?)

φ

0

u

t

Γ

e

i

Φ

Φ

υ

i

Φ

t

$

i

R

u

Φ

Φ

1

b

Φ

h

hay which (Scottish) stew

e

h s

Τ

232

Linguistic description: General principles, and phonology

pre

1st

2nd f

Φ

η

1st

2nd

3rd

4th

ο

u

k

S

1

Φ

k

Φ

a

u

η

s

1

s

Κ

r

a

Φ

m

b

1

s

Φ

Φ

a

r

m

Ρ

1

All this is, of course, tentative. I only wish to give some idea as to how the phonologist may proceed. I deliberately do not immediately give any final solutions, nor is it likely that I could reach any final solutions, short of a complete investigation into and a description of English phonology. If we adopt the above, we shall have to investigate what the repercussions will be in respect of archipositions. For instance: is the 'zero' in the first position of 'fore-castle' really to be 'zero', or does /f/ stand in an archi-position representing three positions, i.e. 'pre' and first and second explosive? In order to work this out, it would, of course, not be correct to hypothesize that there is i.e. that the place could be filled, as this would be a priori irrefutable. As I have said repeatedly, the fact of not finding something can never constitute a refuting example. Such examples as 'sphere' seem to refute the hypothesis that 'pre' could not be filled in principle, i.e. the impossibility of filling it here could be due to an accidential gap. But there may, of course, be a structural rule involved. This has to be investigated. The 'context' of an archiposition may, in principle, of course, be any part of the distributional unit (this is why it is called 'distributional' unit), albeit that the 'immediate' context is likely to be more important in that respect. Examples such as 'friend' seem to refute the hypothesis that the second explosive position could not be filled. This leaves us with yet another plausible possibility, i.e. that 'pre' cannot be filled if 'e 2 ' is filled, and that 'e 2 ' cannot be filled if 'pre' is filled, i.e. there could be, in different contexts, two types of archi-position involved, namely 'pre/e 1 ' and 'e 1 /^ 2 '. Though examples are scarce, there are such cases as 'sphragistics' which make even this doubtful, i.e. by extrapolation we may assume that — though there are hardly

The notions 'position', 'nuclearityand

'peripheralness' in phonology

233

any examples — it is structurally not impossible to have all three positions filled. In the same way we have in the first scheme to correct 'length', 'slower' and 'sloper', as an archi-position for /If cannot be established by proper refutation. In the first scheme I have put a question mark after 'straighten', 'slower', 'sloper', and 'boater'. That is, at that point it had to be considered whether these do not exhibit cases of 'morphemes'. In fact, we can safely assert that they do. Remember that the term 'morpheme' has no actual grammatical connotations, but it indicates purely phonological entities, whose appearance is accidental, in the sense that they correspond either to factors pertaining to grammar, or sometimes to historical, word-forming processes, or other accidental features from the point of view of phonology. Such features seem in an accidental way to upset the 'normal' pattern. Though strictly speaking we should treat them on an equal footing with other features, e.g. set up a distributional unit incorporating anything that can occur phonologically speaking, if we did so we would have to face enormous complications in presentation. The notion 'morpheme', as I employ it, enables me to satisfy the requirement of adequacy without sacrificing too much simplicity. I therefore add to the distributional unit established above, before and after the already established positions, (now including El-Shakfeh's fourth implosive position), a position, which can be further subdivided — i.e. one can establish 'minor' distributional units for these positions — which I shall call 'morphemes'. On these I shall map certain forms, including 'straighten', 'slower', 'sloper', and 'boater'. All this is, as I repeatedly said, tentative, and only meant to give an indication as to how to go about setting up distributional units, which is itself necessary for the establishment of phonemes by commutation. This part, though connected with phonematics, is a necessary preliminary step for the study of phonotactic distribution, a thoroughly neglected field of study. It harbours, even at this stage, enormous complications. For instance, in English there are so many ancient and new loan words, which seem to fit with great difficulty even in the above type of scheme. To give just a few examples, the many forms with 'ver-', e.g. 'verbenaceous', Vermiculation', 'vertebration'. Less cumbersome are forms that have a simply analysis (unlike the previous examples, which are not so straightforward) into more than one

234

Linguistic description: General principles, and phonology

morphemes

pre

1st

2nd

η

1st

2nd

3rd

4th

morphemes

bestrewn

br

s

Τ

r

R

u

Φ

Φ

Φ

η

besprinkled

br

s

Ρ

Γ

i

Φ

Ν

k

1

d

believers

br

φ

Φ

1

R

i

Φ

rz

unspeakable

rn

s

Ρ

Φ

R

i

k

b

1

Φ

straighten

Φ

s

Τ

τ

e

i

Φ

Φ

t

η

strained

Φ

s

τ

Γ

e

i

Φ

η

Φ

d

slower

Φ

s

φ

1

0

u

Φ

Φ

Φ

r

lengthen

Φ

Ι Φ

φ

1

e

Φ

θ

η

sloper

Φ

s

φ

1

0

u

Ρ

Φ

Φ

r

boater

Φ

φ

b

Φ

0

u

t

Φ

Φ

r

houses

Φ

a

u

ζ

Φ

Φ

rz

h

ν

η

distributional unit. In this respect, for English one can assume that in a distributional unit there can only be one phoneme in nuclear position within the basic distributional unit, and the only phoneme that can occur in a nuclear position in one of the 'morpheme' attachments is /τ/. But with /r/ one has to be careful. The same realization occurs among others as (a) the [a]like under-articulation of one of the other vowels, (b) a parasitic non-functional [a] between certain consonants, and (c) as a genuine [a] like realization of a phoneme, e.g. 'below ~ blow' i.e. /brlOu/ ~ /blOu/. In English we can usually recognize the latter by the fact that in that case there is free variance in realization between [ 9 ] and [I], and anything in between may be heard, though for some speakers there is opposition in some cases, e.g. 'affect' ~ 'effect' (Example from M.A.L. Lamb, unpublished). We should not regard as a 'morpheme' in the context of the distributional unit a form or part of a form that can be interpreted

The notions 'neutralization'and'archiphoneme'

235

as an instance of a basic distributional unit. In this respect it is necessary to consider 'syllables' and 'distributional' units as not necessarily 'exclusive' entities. That is, for instance in 'blowing' one can recognize the syllables, and indeed phonotagms /blOu/ and /uii)/, i.e. /u/ belongs to both (cf. Mulder 1968). Syllables and distributional units, when physically inseparable, are like adjacent rooms which have walls in common. I have dealt with this extensively in Mulder 1968, which as far as phonology is concerned can be regarded as to a surprisingly large extent upto-date, and in Mulder and Hurren 1968, Mulder 1974, and Mulder 1978; the above can be read to supplement the treatment of phonological description in the present volume. Mulder 1968 is, however, considerably superseded, at least in parts, with respect to the introduction and the chapter on grammatical background. Of course, as far as the present volume deviates from anything I have written in the past, it has to be taken as representing my latest ideas on the matter, also in respect of phonology.

14. The notions 'neutralization' and 'archiphoneme' The suspension of distinctive function between, and hence the disappearance of the identity of, one or more distinctive features within a phoneme in a particular position is called 'neutralization'. What is left of the phoneme is called an 'archiphoneme'. A point of debate between linguists, and even among functionalists, has always been whether one really needed this old Prague-school notion in the description of languages. The opponents of the notion usually claim that the phenomena in question could as well be described under the heading of 'distribution', as the non-occurrence of certain phonemes in certain contexts has to be described under that heading anyway. When a certain phoneme does not occur in certain contexts there are, for the adherents of the notion, three possibilities: (a) it is accidental, i.e. one cannot establish a structural rule for the non-occurrence; in that case it is called an 'accidental gap'; (b) there are phonemic distributional constraints; (c) there is neutralization. One regards it as a case of neutralization if the phonemes involved are members

236

Linguistic description: General principles, and phonology

of a 'correlation', i.e. if they have at least one feature in common, and it is that feature or those features which in cases of neutralization is (or are) not affected. For instance, if in some contexts one can have 'occlusive labial voiced' as well as 'occlusive labial unvoiced', but in another context one can can have only one of the 'occlusive labial' phonemes, one has to concede that the opposition between 'voiced' and 'unvoiced' is suspended in that context. It is irrelevant in this respect whether the realization is voiced, or unvoiced, or anything in between. The fact that in some cases, in some languages, the realization of an archiphoneme may not coincide with that of one of their factors is another strong reason to maintain the notion 'archiphoneme', as description in terms of distribution in those circumstances becomes difficult, and of necessity arbitrary. The essential difference between 'distributional restriction' and 'neutralization' is, however, that in the former whole phonemes are affected, but in the latter only features of phonemes. This is quite naturally correlated with a difference in the type of phonetic factors that can be regarded as being the 'cause' of the non-occurrence of particular phonemes in particular contexts. In the case of distributional restrictions it is more often than not that — apart from 'position' within the phonotagm — primary articulation factors are involved, whereas in the case of neutralization it is usually secondary features, such as voiced ~ unvoiced, aspirated ~ non-aspirated, ingressive ~ non-ingressive, pre-glottalized ~ non-pre-glottalized, fortis ~ lenis, etc. that are involved, but also primary features may be involved. But there seems to be a distinct hierarchy: if primary features are involved, the third dimension tends to be involved as well. A typical example of distributional restriction is the fact that in English no fricative other than /s/ can precede any phoneme in any of the explosive positions. That is to say, in fact, that /s/ before the nucleus is the only phoneme that can stand in that position. As there seems to be a definite distinction of type between factors involved in cases of 'restricted distribution' as opposed to 'neutralization', there seems to be a real advantage in keeping them apart, and the establishment of archiphonemes is an adequate way of doing this. As, moreover, it is often difficult, if not impossible, to judge from the realizations which member of a correlation should be chosen if one were to reject the 'archiphoneme' solu-

The notions 'neutralization'and

'archiphoneme'

237

tion, it seems to be wise to adopt it and avoid having to make an arbitrary choice. For a functionalist, even more important than all this should be the fact that not recognizing archiphonemes would be inconsistent with one of the very definitions of 'phoneme', i.e. 'a simultaneous bundle of distinctive features'. It is clear that, for instance, the bundles (labial, fricative) and (labial, voiced, fricative) are not the same. If the latter equals /v/, the first does not, and hence they are qua phonemes not the same. They do not, however, occur in the same context, and are, therefore, not mutually opposed. It is the latter which makes archiphonemes special, and the fact that there are phonemes in which — as bundles of features — they are properly included. Intuitively speaking, one could say that the notion 'neutralization' deals with the distribution of distinctive features, not with that of phonemes. In fact, if one accepts that an archiphoneme represents its factors (i.e. the phonemes involved), the distribution of the phonemes themselves is not affected at all. Some linguists have suggested establishing archiphonemes only in those cases where morphological processes indicate a 'deeper' difference. An example is 'Hand' in German, which is realized [hant], but the plural is Hände which is realized [henda]. This is actually akin to the idea of 'morpho-phoneme', an idea which I reject, as we have seen earlier. But it is even worse than the latter, as it brings morphology into phonology proper, rather than into a border-line area, as morpho-phonologists do. If we admit morphology into phonology we will no longer be able, without begging the question, to describe the forms of signa in phonological terms, as this form is already taken into consideration in these phonological terms. Consequently we cannot deal with allomorphy in an adequate manner. The same, mutatis mutandis, is true if one uses semantic considerations in grammatical description, as Munla (1981) seems to have done. Moreover, this leads to begging the question when giving a semantic interpretation of grammatical structures. Broadly speaking, we find two kinds of archiphoneme in languages. In one, the context is its position within the phonotagm, or zero is preceding or following it. The other is subject to constraints of homorganity. Examples of the former are in Dutch, and in German dialects in which there is a distinction between

238

Linguistic description: General principles, and phonology

voiced and unvoiced, in respect of occlusives and fricatives at the end of a phonotagm. In that case the realization is generally unvoiced, but as unvoiced is not opposed to voiced, there is no identity for that feature. Another example is the lack of opposition between /s/ and /§/ in German at the beginning in 'pre-explosive' position. An exception may seem to be in cases of an immediately following vowel (e.g. [Se:rg] (or sjeira]) ~ [ze:r], i.e. 'Schere' ~ 'sehr' ('scissors' ~ 'very'). I render the former, however, by /Siera/, as we shall see, for reasons of overall greater simplicity, though a representation */&erg/ would be equally adequate. Examples of the 'homorganity' type are in English the neutralization of Voiced ~ unvoiced' after 'voiced' or 'unvoiced' phonemes, or after archiphonemes which result from such a neutralization, the neutralization of 'labial' and 'dorsal' in nasals before labials and dorsals, the non-distinction between /e/ and jo/ (i.e. the neutralization of 'spread ~ rounded') before /u/, and even the non-distinction between /r/, /i/, or /u/ (i.e. 'neutral', 'spread', 'rounded') immediately before /i/, or /u/, etc. We shall see examples of the latter two below. An archiphoneme, then, can be regarded as a phoneme in particular subsystems which represents, i.e. is equivalent to, two or more phonemes of the overall system. There are three necessary conditions for the establishment of an archiphoneme: 1. In at least one context the phonemes to which the archiphoneme is equivalent are opposed to one another. This makes the notion 'archiphoneme' different from that of the notion 'hyperphoneme' (see above). 2. The phonemes involved should have at least one distinctive feature in common. Otherwise the phenomena involved have to be described under 'restricted distribution'. 3. The archiphoneme in question should actually occur in at least one context, and be opposed in that context to other phonemes or archiphonemes, but not to any of its factors (i.e. the phonemes involved), and not exclusively to 'zero'. In the latter case criterion (2) would be automatically violated, or at least arbitrarily the description would be divided between 'neutralization' and 'restricted distribution'. If we take as an example /s/ as in English /sPriNkl/, where /s/ is exclusively op-

The notions 'neutralization'and 'archiphoneme'

239

posed to φ it would be impossible to decide non-arbitrarily what the factors are, if we regarded /s/ rather as /S/, i.e. as an archiphoneme. If we did, we could still not escape relegating some of it under 'restricted distribution', because of criterion (2), but it would be arbitrary to decide which part, i.e. there may be multiple neutralization involved, i.e. 'voiced ~ unvoiced', and/or 'occlusive ~ fricative ~ nasal', and/or 'labial ~ apical ~ dorsal', but there still remains the non-opposition to /h/ and /l/ when /s/ is followed by another phoneme in explosive position. It would, then, be impossible to decide non-arbitrarily where neutralization ends and restricted distribution begins, i.e. which are the actual phonemes involved in the presumed neutralization. An archiphoneme, in terms of distinctive features is a simultaneous bundle of distinctive features common to two or more phonemes. That is, logically speaking it is included in each of the phonemes, as it is an intersection of classes of features that make up particylar phonemes. We can represent this by the following Venn-diagram, which represents the neutralization of opposition between voiced and unvoiced, in certain contexts, between /d/ and /t/ in English:

The archiphoneme /T/ is a simultaneous bundle of the features 'apical' and 'occlusive'. If we add to this 'voiced', we have Id/, if we add 'unvoiced' we have /t/. The archiphoneme /T/ represents both /d/ and /t/. Therefore, strictly speaking, we cannot say that /b/ does not distributionally occur after /s/ in /sPriNkl/. It does, but represented by /P/, and /p/ likewise. Archiphonemes are conventionally written by the capital letter of one of their factors. This is, of course, not an absolute law. There may be cases in which it is more convenient to use another symbol. But for the sake of clarity we should stick to it in most cases. Which factor we choose is irrelevant, but it is wise to reckon with sensitivities

240

Linguistic description: General principles, and phonology

of linguistically unsophisticated people. I know of a case of a research-student who made his professor furious, because in his thesis he had represented a supposed neutralization between /r/ and /l/ in a certain language by /R/. After he had changed /R/ into /L/ (which involved quite a bit of work on the type writer), the professor in question was fully satisfied. This shows how important it may be in practice to make the right arbitrary choices.

15. The description of vowel systems It should be clear that what I am trying to do in this, and also in the next chapter, is not merely present the theoretical foundations for phonological and grammatical description, but also give practical hints as to how to go about tackling the problems. Such hints are, of course, based on experience. On the one hand, if one has dealt with many descriptive problems in many languages, one knows, almost intuitively, what to expect. This is itself, of course, connected with the very nature of 'communication by language' and it is, via the requirement of appropriateness, already to a certain extent reflected in the theory and methodology. But what I mean here is much more material features, features directly connected with the language under investigation. One can often almost predict what to find, and the further one proceeds, the more prediction becomes possible. This may sound Procrustean, and it would be so, were it not that a strict adherence to the hypothetico-deductive method as outlined above will free one from such dangers. There is something else that comes from experience - and it is part of my own experience that I try to convey here. It is our aim to come to solutions that are both adequate, and as simple as possible. And the solutions have to be consistent with one another. This is what we aim at. We do not claim to present the truth, but to describe the facts in such a manner that they become clear, and this we do in a consistent, adequate, and as simple a manner as possible. Where experience comes in is in certain rules of thumb which (we know this by experience), if they can be applied consistently, guarantee a maximum degree of simplicity

The description of vowel systems

241

within complete adequacy. For instance, we know that there are in languages certain phonemes with fairly narrow opening in realization, e.g. [i], [y], [u], [I], [a], and that some of those seem to be closely akin to slightly more consonantal realizations, e.g. [j], [UJ, [w], [v], [v], [i] (the Southern English /r/) etc., which only seem to occur in peripheral realizations. The rules of economy, i.e. of simplicity, in description imply that, in the absence of opposition between the two types of realization (which we assume, and, if unrefuted, we accept as corroborated), we regard, say, [i] and [j], etc. as realizations of the same phoneme. Now it may be that in the process of description we find reasons for not doing this, which could only be that either the metahypothesis of adequacy, or that of simplicity, or both, or even the meta-hypothesis of consistency is refuted. But to begin with, it would be extremely unwise not to start off by assuming something that is likely to lead to the simplest solution, provided, of course, that it is a valid hypothesis to make. However, many of us are victims of our own prejudices, usually due to our own linguistic habits, which is often a conditioning factor for what we actually hear our informants say, but also to our interpretation of what they are saying. For instance, if a Dutchman hears pronunciations of, say, the German word, Wasser 'water', he hears on average as the pre-vocalic element a [v]. Dutch realizations [v] are opposed to realizations [v], the former being a variant of the /u/ phoneme, the latter being the normal realization of the /v/ (voiced labial fricative) phoneme. He will therefore have no difficulty in assuming that the 'W' of 'Wasser' is phonemically /u/, and this puts him descriptively in an advantageous position. He may, of course, be ultimately knocked off this position for other reasons having to do with consistency, adequacy, and/or simplicity, but until such time as he is he should stick with it. For a speaker of French, things are slightly more difficult, as the realization of /v/ (voiced labial fricative) in French may approach a [v] type of realization, as sometimes in ville 'town'. The greatest difficulty seems to exist for the average speaker of English. As their peripheral realizations of the /u/ phoneme are generally distinctly bi-labial, they actually hear [ν], i.e. a labio-dental fricative, rather than [v], i.e. a labio-dental approximate. Not that this should matter, as there is no reason

242

Linguistic description: General principles, and phonology

why [u] and [v] could not be variants of the same phoneme (cf. Pike 1974), but as in English realizations [v] almost invariably are /v/, the fact of hearing wrongly leads on to the next step, i.e. that of interpreting wrongly, that is according to the knowledge of one's own language. The German stands, to begin with, in a neutral position, as no matter what is to be the outcome, there is either no /v/ phoneme in the language, or the /u/ phoneme does not occur in peripheral positions/*. But if he is experienced enough, he will start with the simplest of the assumptions, i.e. that it is a variant of the /u/ phoneme. This is immediately supported by the fact that in German no other fricatives except /s/ and /z/ exhibit an opposition Voiced ~ unvoiced', and in Southern dialects or pronunciations of Standard High German, such an opposition even for occlusives is disappearing, or has disappeared already completely. Also, after initial [s] or [§] there is never an opposition Voiced' ~ 'unvoiced', but */v/ would not partake in this neutralization, which makes it doubtful that it could belong to the correlation at all. Further support to follow is that /u/ and /i/ remain similar in distribution if one assumes [v] to belong with /u/, e.g. otherwise one cannot directly compare Bayern ~ Bauern, Jahr ~ war, 'Bavaria ~ farmers', year ~ was', Schere ~ Schwere, i.e. /Siere/ ~ /Suere/, 'scissors ~ heavy' etc. Note that the opposition between /s/ and /§/ is neutralized in initial position, resulting in the archiphoneme /S/. Indeed, it has been suggested by someone who refused to recognize peripheral /u/ in German, to pose /au/ as a single phoneme, the only diphthong in the language, and he suggested all sorts of other solutions — partly on further wrong assumptions and procedures —, such as regarding [j] in initial position as a variant of the archiphoneme /X/, product of the neutralization of opposition between /x/ (the 'ach-laut') and /9/ (the 'ich-laut'). As a further consequence he proposed to regard, say, Eichen, 'oak trees', as phonemically /agn/, rather than /aiXn/, thereby forgetting that the latter does not rhyme with 'Mama-chen' in the once popular song 'Mama-chen gib mir

*

Nevertheless Germans tend to hear something strange, a sort of overfricativation, in many British pronunciations of German 'w'. I have asked informants about this, and I have also heard it said spontaneously by a German Lektor in a British university.

The description of vowel systems

243

ein Pferdchen, ein Pferdchen war mein Paradies'. Regarding [j] as a variant of /X/ is itself based on the wrong assumption that no dorsal fricative occurs initially. In fact, though it is rare, one does find this. In Northern dialects (and these are the only dialects in which one might find occasional [ v ] , rather than [v] realizations) one finds, say, 'chemisch' and 'Chemie' pronounced with [5], and though 'Chiemsee' at the Chiemsee itself is pronounced with initial [ k ] , one does find for this [x] or [9], i.e. an unvoiced dorsal fricative of a kind, as far south as Frankfurt. Not to mention that it is rather implausible (i.e. it is materially inadequate to maintain) that the mere fact of being 'prevocalic' or 'postvocalic' could account for the difference in realization between [j] and [x] or [ ς ] . All this shows what difficulties one is likely to land in if one is prejudiced on the basis o f one's own language, or otherwise prejudiced, and especially if one does not start o f f with the simplest assumption. None of these complications will arise, if one starts o f f with assuming that [ v ] , or, if one really cannot hear [ ν ] , that [v] is a variant of the juj phoneme. O f course, in a language like German, in the absence o f opposition, realizations in peripheral positions may vary over a much wider range, e.g. [v], [ v ] , [ w ] , than, say, in Dutch where [v] and [v] have to be kept distinct, or, say, in English, where [v] and [w] have to be kept distinct. In Dutch it would not matter if [v] and [w] were confused, though it would sound odd, nor would it matter if in English [v] and [v] were confused. In fact, it seems that, unlike for a Dutch speaker, for an Anglophone the difference between [v] and [v] is so slight that it is likely to pass unnoticed. I am, therefore, not implying that [v] realizations in initial and non-final implosive positions do not occur in some German dialects* (though they are much scarcer than is assumed by most English ears) but that this is irrelevant. Even if the realizations were always [ v ] , this could very plausibly be a member of the /u/ phoneme in German. Another rule of thumb, based on experience, in the description of vowel-systems is to avoid establishing a so-called 'correlation of

*

E.g. after some 'longish' realizations of a phoneme in nuclear position, •such as in the personal name 'Uwe', or in 'braven', i.e. /bräun/ as opposed to both /braun/ and /brausn/. Even 'devoicing', may occur, e.g. before.the diminutive '-chen' in morpheme position.

244

Linguistic description: General principles, and phonology

length', i.e. a straight opposition between 'long' and 'short' vowels. If 'long' and 'short' is the only phonetically plausible way to characterize the differences, it is still simpler (because it saves setting up a number of otherwise unnecessary phonemes) to regard the items in question as clusters of two phonemes, i.e. to treat them in the same way as one would treat 'geminations', i.e. clusters of two identical consonants. Indeed, it would be inconsistent to treat the two differently, unless there is a very good reason, e.g. similarity in distribution to 'short' vowels, as, for instance, in Thai. If, however, one has already established the 'narrow' phonemes, e.g. in German, or English, /i/ and /u/, as 'semi-vowels', i.e. phonemes that may stand either in nuclear, or in peripheral position, one would not usually have a 'correlation of length' for those phonemes, but clusters of two, of which one stands in nuclear, the other in peripheral position. There are, of course, exceptions, as again in Thai, where they clearly both stand in nuclear position, commuting with short vowels, semi-vowels, and diphthongs. Both in English and in German there is neutralization of opposition — because of the requirement of 'homogeneity' between semi-vowels in nuclear position and immediately following semi-vowels in peripheral position. That is, both in English and in German, after /i/ one can only have an [i] -like or [j ] -like realization (I keep 'interjections' out of consideration), and in German after /y/ it is rather indistinctive whether the realization is /i-j/-like, /u-w/-like, or, indeed, /y-u/-like. In Southern English, one of the semi-vowels is (as we shall see) /r/, and the same goes for /r/ in Southern English. I shall in German symbolize the archiphoneme by /I/. If one wishes one can use the symbol /:/ instead, in order to satisfy the feelings of those who regard it as elegant to use phonemic representations which, as much as possible, reflect phonetic realizations. In English, it works out simpler to regard the neutralization as residing in the nuclear position. As before /i/ or /u/ the opposition between /r/, /i/, and /u/ is neutralized, I pose an archi-phoneme /R/, and write /dRu/ and /bRi/ for 'do' and 'be' respectively. On the contrary, in German, the neutralization should be regarded as residing in the peripheral position. The reason is that nowhere else is /y/ attested in peripheral position, and we can, therefore regard /i/ and /u/, but not /y/ as the factors of the archiphoneme /I/. There are some

The description of vowel systems

245

more advantages but this will suffice. In German we have, besides /i/, /y/, and /u/, the sequences /il/ as in /ilr3/ Ihre 'your' (as opposed to /ira/ irre 'err'), /yl/ as in /hylts/ hüte 'to guard' (as opposed to /hyts/ Hütte 'hut'), and /ul/, as in SuIxT/ sucht 'seeks' (as opposed to Sucht 'disease', which is /SuxT/). The minimal pairs in German are scarce. Having done away with a correlation of length (which saves us three phonemes, at the cost of one archi-phoneme) we now look at the other vowels of German, and find [£] ~ [e:], [oe ~ 0:], [0 ~ o:] and [a ~ a:]. All of these are functionally opposed. In all of these there is a conspicuous difference in vowel-quality, though, perhaps least so in [a ~ a:]. In the latter the most conspicuous feature is usually a difference in length, and also in the other cases there usually is the impression of a quantitative difference. Some of this may, however, be due to contextual features, such as a difference between so-called 'closed' and 'open' syllables, and neutralization is often involved. I shall not go into this here, but there are a number of archiphonemes that can be established. It may seem, therefore, that one has a choice between calling the difference one of 'length', or one of 'quality'. There are, however, two reasons why we should opt for quality. The first is perhaps merely a consideration of 'elegance'. It is after all the case that the most constant factor in the realization is one of 'quality' rather than 'length'. Even if one were to pronounce the 'long' phonemes shorter than the 'short' ones, one would still be able to recognize the proper phoneme from the realization. The other reason is of more importance. If one were to establish a correlation of length for the more open vowels, one would for reasons of consistency be more or less forced to do the same for the narrow vowels, and this would unnecessarily violate the principle of 'simplicity' on many scores. Not only would this lead to the establishment of another four, i.e. in total six, extra phonemes, but also one would still have to accept the occurrence of 'semi-vowels' in peripherial positions elsewhere. Unless, of course, one does not recognize the existence of semi-vowels in German, as in the treatment I have already criticized above. The conclusions I have come to for German lead to a very simple vowel-system. The only vowel that thus far has not found a place is /a/. For simplicity reasons, this can be placed in the only gap left in the pattern, i.e. /i/ and /u/ are semi-vowels, and

246

Linguistic description: General principles, and phonology

place /a/ in a correlation with /a/ and /ä /. I agree that there is something counter-intuitive in such a treatment, because some realizations of /a/, especially before /r/, in German do not strike one as particularly narrow. One should, however, remember that the phonetically inspired 'labels' are no more than 'labels', and should be taken as classificatory rather than descriptive. Also the labels 'lax' and 'tense' in the scheme below are not fully satisfactory, and neither are 'front' and 'back'. Accordingly, as long as this remains unrefuted, I feel entitled to put /a/ vertically in a correlation with /a/ and /ä/, and horizontally in a correlation with /i/, /y/, and /u/, as follows:

front

medium

back

neutral

narrow

i

y

u

lax

ε

oe

0

a

tense

e

0

ο

ä

There is, of course, much more to be said about the vowelsystem of German. In particular, neutralization is an important issue, and fairly complicated. Also other problems, e.g. dialect differences, are insufficiently dealt with in the above. For instance among speakers of Rheinland dialects there tends to be an opposition between [e:], [£], and [£:]. For a full description of German, much research would have to be done. The above is offered as an exemplification of sensible descriptive techniques, and at the same time as a first approximation to such a description.

16. Some further practical hints for establishing vowel-systems Owing to the fact that the syllable tends to have a 'ballistic' shape, i.e. from closed to gradually more open, to closed again, it is to

Some further practical hints for establishing vowel-systems

247

be expected that 'semi-vowels' are, when in peripheral positions, nearest to the nucleus. It follows, therefore, that a good technique in the description of vowel-systems is to set up tables in which one tries various combinations of vowels with semi-vowels. I have done this (in the case of Southern Standard English in combination with Hurren) in several publications (see Mulder — Hurren 1968, Mulder 1968, Mulder 1974). Here follow examples, some taken from earlier publications. The neutralization of opposition in Southern Standard English between /i/, /u/ and /r/ is, as I said, symbolized by /R/, whilst in Scottish English (in fact the data came from Angus and Perthshire and to a certain extent from Dundee, which is different from the rural areas) the neutralization of opposition between /i/ and /u/ is symbolized by /I/. In Scottish, /r/ is not a semi-vowel, whilst /a/ is a separate vowel-phoneme, not a semi-vowel either, but it partakes in the neutralizations leading to the archi-phoneme /I/. I have, where otherwise ambiguity could arise, put a dot underneath the phoneme in nuclear position, to indicate that position.

Γ+

i+

u+

-

+u

+i

+r

a

ram

iak

uag

man

kau

bai

par

e

rest

iel

uel

men

dOu

bei

per

ο

rod

iot

uog

kot

dOu

boi

por

Γ

rrm

iri]

urn

3r

dRu

bRi

Ρ"

i

rid

iir

uil

pil

dRu

bRi

Pir

u

ruk

piur

uul

pul

dRu

bRi

pur

t

The left hand (a, e, o, r, i, u) refers to the phoneme in nuclear position of the syllables in the corresponding line. The top (r+, i+, u+) indicate these semi-vowels in explosive (i.e. pre-nuclear) position in the corresponding column, whereas (+u, +i, +r) refer in the same way to these semi-vowels in implosive (i.e. post-nuclear) position. The phoneme sequences in the table refer to the phonol-

248

Linguistic description: General principles, and phonology

ogical forms of the following words in Southern Standard British English: ram

yak

wag

man

cow

buy

par

rest

yell

well

men

dough) Λ )

bay

pair

rod

yacht

wog

cot

dough)

boy

pore

rum

young

won

the

do)

be)

pun-

)

)

rid

year

will

pill

do)

be)

)

)

rook

pure

wool

pull

do)

be)

peer poor

In order to account for all relevant occurrences of the vowels and the phonemes /i,u/ in Scottish English in a similar way to Southern English, we have to set up the following table. The approximate phonetic realization is written underneath each phonological form in square brackets. The realizations of such phoneme-groups (not represented in the table) as /-iei-, -ill-, -uli-/ follow the same pattern as that of /-ei-, -Ii-/, e.g. /ieil/ ~ /reik/ is [jel] ~ [rek], /ulil/ ~ /rlid/ is [wi:l] ~ [ri:d], except that in the case of pre-nuclear /i/ the sequence /Iir/ is not [ij(9)r], but [i:r]. That is /flir/ is usually approximately realized [fij(3)r], but /ilir/ is realized [ji:r]. The symbol [(a)] means: extremely slight, often virtually absent [ a ] . The forms /ieil, ulil, ilir/ are the phonological forms of the words 'Yale' (actually not a word, but a proper name), 'weal', and 'year', respectively. The symbol [y] represents roughly a realization as French 'u', as in pur, or North-German 'ü', as in für, though the realization in the Scottish dialect is usually impressionistically some what 'duller', i.e. slightly more back. The symbol [u] stands for a realization between [y] and [u], i.e. approaching that of Τ in Turkish kadln 'woman', but not as 'spread'. The symbol [i] stands for a realization between [i] and (short) [e], not unlike that in Western Dutch pronunciations of 'i' in dik 'thick, fat'. Note a difference between front [a] and back [a]. Final /t/ is often realized as a slightly apicalized glottal stop. The other symbols will be familiar enough to most readers.

[pur]

pur

Μ H < δ 1 Ο. —'

fir [fir]

[hors]

l-l kH U> ω 1 a,1 Λ —

hors

+l-l

par [par]

Some further practical hints for establishing vowel-systems

*

l-l +

l-l rt ·"• 'ST •-Η π « Ό Ό 1

Ι-ι .a

ι-Η

"

£ £

Λ 9CU 1— 1

*

+

1 s ·° X> —1

l-l

+ur

l-l rt

+3

1

1

3 " o —Ι

_ "3




X

ι—H·

N l

-Ö

Ο

glottal

Ν

dorsal

voiced vase (very

> big)

I shall come back to this. I should, however, like to point out here already that the above is a conflation of two levels of analysis, and that the fact that there is an arrow between 'very' and 'big' is irrelevant to the first level. I have pointed this out already in the previous chapter, but I cannot repeat it enough. We always consider only one level at a time. Let me give a further example:

We see here that 'lost' has three arrows pointing to it and that this, therefore, is likely to be (and in fact is) the 'nucleus nucleorum', though not the nucleus at the first level of the analysis. The nucleus at the first level is the syntagm 'has lost', of which

Direct and indirect tactic and non-tactic relations

279

the nucleus is 'lost'. Indeed, at the first level, in terms of their immediate constituents, and the relations between them we have: the little girl ->

has lost.

her very nice bracelet The difference between a ' ] > ' type, and a ' — I — > ' type of presentation is that the former refers to parallel or conjunctive determination, the latter to diverse or disjunctive determination. 'Determination' is the tactic relation from 'peripheral' to 'nuclear'. The converse of this relation, i.e. from 'nuclear' to 'peripheral', is called 'government'. In the last example, 'the little girl' and 'her very nice bracelet' are said to determine 'has lost', and conversely 'has lost' is said to govern 'the little girl' and 'her very nice bracelet'. Determining entities are said to depend in respect of their tactic function on the presence of the entity governing them. This is called 'functional dependency', as opposed to 'occurrence dependency', as we shall see. In the case of 'diverse' determination, the relation between one of the peripheral entities and the nucleus, and that of at least one other of the peripheral entities and the nucleus, is different. That is, the hypothesis that the relations are not different is refuted. But it can only be refuted if the classes of items that can stand in the two, or more, peripheral positions in question are the same. In the above case, which is an example of a pattern subject >

predicative

object this is indeed the case. Any item that can stand in 'subject' position can also stand in 'object' position, and vice versa, at least in principle. There may be allomorphic variation, as between Τ and 'me', 'he' and 'him', 'she' and 'her', etc., or there may be semantic reasons for unlikely occurrences, but in principle one could have: her very nice bracelet (subject) -> the little girl (object)

has lost

280

Linguistic description II: Syntax and the sentential level

for instance in an "Alice in Wonderland" situation, where a bracelet goes for a walk with a little girl, but, being a bit careless, loses the little girl. I said there may be allomorphic variation, but if there is a difference in morphological structure matters are, of course, different. For instance in Latin, ego amo filiam Ί love the girl' is a case of parallel determination, not of diverse determination, as it can be shown that 'filiam' is morphologically complex, i.e. different at the sign-level from, say, 'filia' (which could stand in 'subject', but not in 'object' position), and therefore the position-classes are not the same. Hence, the hypothesis that we have parallel determination remains unrefuted. We have parallel determination if it remains unrefuted, and only diverse determination if the hypothesis of parallel determination is refuted by producing two identical peripheral classes, items in which determining the same nucleus. As we have in our example two items with two arrows pointing to it, we have at least two complex immediate constituents, which can be analysed at the second level. In fact, we have three, as there is also an arrow from 'has' to 'lost', and 'has' is not further connected with any of the other constituents. So we have on the second level: (1) the girl little (2) her -> bracelet very nice (3) has

> lost.

Of course, if there is only one peripheral constituent, the distinction between 'parallel' and 'diverse' determination does not come into play. But we see here, already, that we cannot simply 'deduce' the immediate constituent structure from the pattern of direct and indirect relations. That is, as far as these relations are concerned, it could well have been the case that we have here three immediate

Direct and indirect tactic and non-tactic relations

281

constituents determining 'lost', i.e. the little girl (subject) her very nice bracelet (object)

> lost

*has (auxiliary) We can, however, by valid commutation, show (i.e. refute a hypothesis to the contrary) that 'has lost' is a constituent, whereas we cannot do the same for 'the little girl lost', and 'lost her very nice bracelet'. Some may, in fact, think that the latter is a constituent, e.g. a VP in Chomsky's sense, but there are reasons why this would be incorrect, as we shall see. As there is also an item with only one arrow pointing towards it, there is yet a third level of analysis, only affecting 'very nice', i.e. very —

> nice

We have already seen that there is no straightforward one-one relation between direct and indirect relation patterns and immediate constituent structure in terms of functional dependency (i.e. the patterns we have shown). But we can turn this round. The direct and indirect relation patterns actually belong to the 'protocols', or at least they are nearer to the protocols than actual syntactic patterns are. That is, they are to a certain extent a record of observed facts. Sometimes there may be some doubt, and considerations of consistency, adequacy, and simplicity may even be applied here, but generally speaking these are the supposed facts, and they can therefore be used to test the assumed material adequacy of the descriptive patterns. The syntactic patterns we set up must be in agreement with, i.e. they must not be materially inadequate with regard to, the protocolization, including the more or less observed direct and indirect relation patterns. The latter can be mapped onto the syntactic patterns, and given the syntactic patterns we can reproduce (if you wish, call it 'generate') the

282

Linguistic description

II: Syntax and the sentential

level

relation patterns. That is (if we conflate levels), from the > girl little ->

(has

very

nice

> lost)

her > bracelet (very

> nice)

we can derive

bracelet,

but not vice versa. There is yet one more reason why it is not 'vice versa', namely, direct relations do not only obtain between peripheral entities and nucleus, but also (a) between nuclei of immediate constituents and (b) between immediate constituents that, unlike in the above examples, do not stand in a relation of sub-ordination. I shall deal with this later. Here we are only concerned with (a). Let us now consider a syntagm such as 'he fell under the bridge', and indicate the pattern of direct relations: he

fell

under

the

bridge

It may be counter-intuitive to some people that I regard 'under' as nuclear in respect of 'bridge', which is, of course, nuclear in respect of 'the'. Indeed, most linguists have referred to items such as 'bridge', in such constructions as 'under the bridge', as the head or something of that kind. To my knowledge, I have been the first to consider so-called 'prepositions' and 'conjunctions' as the most important entity, i.e. as the nucleus, of such constructions. But this idea is actually a straightfoward logical consequence of adopting Martinet's concept of a 'functional' (Martinet

Direct and indirect tactic and non-tactic

relations

283

1960, but I knew of this well before the work was published via personal conversation) and applying it to syntax, which, unlike Martinet, I separate from morphology. Martinet's distinction between 'functionals' and 'modifiers' has always struck me as one of the most important and valuable features of his theory at that time. The reason why others may regard items such as 'bridge' in the above construction as the most important is perhaps because they 'feel' that 'bridge' has more meaning, and, in fact, they are right. That is, 'bridge' belonging to a huge, if not infinite, paradigm of items that are mutually opposed has a perhaps infinitely greater distinctive load than 'under' (let alone 'the') has, and this is reflected in its meaningfulness. Nevertheless, it is 'under' that characterizes the syntagm in question, that determines the distribution of the syntagm and that governs the function of 'the bridge' in respect of the wider syntagm of which 'under the bridge' is a constituent. It is, in short, the identity element for the relations within the syntagm 'under the bridge'. It is therefore, syntactically speaking, the most important element within 'under the bridge', and it is this which I call 'nucleus'. But it has to be admitted that the arrow from 'bridge' to 'under' is put there by implication, and does not derive from simple protocolization. I shall come back to this. In the above example one sees clearly that nuclearity could never have been assigned to 'bridge', as one could not maintain that there is any meaningful direct relation between 'fell' and 'bridge'. Such a relation is indirect, i.e. via 'under'. This is another characterization of 'nucleus', i.e. the entity via which the relations within the wider syntagm pass, hence it is the 'identity-element' for such relations. Now let us look at the pattern of relations. We do see that 'fell' must be the 'nucleus nucleorum'. Because it is 'under', and not 'bridge' that is directly connected to 'fell', we see that 'under' is the nucleus at the second level, and 'bridge', being directly connected to 'under', is the nucleus at the third level. But what we cannot tell at this stage is whether 'he' and 'under the bridge' are immediate constituents within the same construction, or whether'under the bridge'determines 'he fell', or whether, indeed 'he' determines 'fell under the bridge'. Now, one of the rules in description is that we keep things as simple as possible, i.e. we hypothesize the simplest solution, unless, of course, this involves inconsistency with decisions we had

284

Linguistic description II: Syntax and the sentential level

to, or will have to, make elsewhere, or if it is inconsistent in respect of the whole of the approach. We regard such a hypothesis as corroborated unless and until it is refuted by demonstrating the inadequacy of the solution in question. In the above case, it is obviously simpler to have one construction with three immediate constitutents than to have first a construction with two immediate constituents, e.g. 'he > fell' or 'fell < under the bridge', which is then determined by 'under the bridge' or 'he' respectively. Unless we have good linguistic reasons to do otherwise we should initially hypothesize that the construction is as follows: he >

fell

[under the bridge] It is, of course, until shown otherwise (we have seen this above) a case of parallel determination. I shall call the 'under the bridge' part 'complement' and the 'fell' part the 'predicative' and we pose, therefore, the following structure: subject ->

predicative

[complement] The square brackets indicate that the item in question is a socalled 'expansion', i.e. that it commutes with 'zero'. I shall come back to this later. Please ignore the brackets for the time being. We may now notice that this construction is compatible with the meaning in a situation that someone is, or walks, by a bridge and falls underneath it, perhaps because the surrounding is slippery. That is, the construction is materially adequate in respect of such a meaning. In that case, 'under the bridge' commutes with 'under the table', 'into the river', 'against the wall', and other such likely or unlikely items. In particular in English we are lucky in this respect as there is a difference between 'in' and 'into', which helps us by commutation in deciding whether such items as 'under

Direct and indirect tactic and non-tactic relations

285

the bridge' actually do stand in this position. The structure for 'he fell into the river' could only be he >

fell

[into the river] However, there is also a homonym of, i.e. a different syntagm, 'he fell under the bridge', which is not compatible with this construction, i.e. this construction is not materially adequate in respect of it. This is in, say, the situation that someone walks under a bridge and falls. Unless there is homonymy involved in one or more of the 'words' used (the only candidate for this would be 'under'), the construction must be a different one. It is theorematic in the present theory that the denotation of a whole is the resultant of the denotation of each of its parts and the grammatical (in this case 'syntactic') relations between those parts. Therefore, if there is a difference in denotation of the whole, this must either be due to a difference of one or more of its parts (in this case 'homonymy'), or the syntactic structure must be different. Consequently, unless we can establish 'homonymy' (by refutation of the contrary, which here we cannot), it would be materially inadequate if we did not establish a different structure. Now, the above is to be used as an adequacy test, not in order to set up the structure in question. For setting up the structure in question we have to use syntactic procedures. These procedures involve commutation trials. The homonym of our first construction for 'he fell under the bridge' will have the structure: (he

> fell)
.Φ

fell)


fell [there]

[under the bridge]

< — / — > indicates a syntactic relation of coordination (see below). I should, perhaps, say at this stage that neither the linear order in the representations, nor in fact the linear order as such in realization, is relevant. Though linear order may be a pointer to syntactic structure, e.g. in the case of 'John loves Mary' versus 'Mary loves John', for syntactic structure as such, linear order is irrelevant. This is different in phonology where the relations are purely formal, and therefore involve all the formal features, including linear order. Syntactic relations are, if you wish to call it this, of a more abstract or esoteric kind. Their adequacy test is in the message, rather than in phonetic realization, as is the case with phonology. Coming back to our example, it is also useful to note that in the second interpretation of 'he fell under the bridge', 'he fell' commutes with any 'predicative based syntagm', e.g. 'he is playing cards', 'he is a fool', 'he kisses Jane', 'Peter buys a horse', 'John may have been able to play a game of bridge', etc. All this demonstrates the peripheraliness of 'under the bridge' to 'he fell', by refuting that it is a constituent at the same level as 'he' and 'fell'. How important commutation is in assessing syntactic structure can be further demonstrated by using a Japanese example, Hiroshima wa kaki wa umai 'in Hiroshima oysters are delicious' literally 'Hiroshima oysters delicious', where the post-position 'wa', which commutes with almost any post-position, marks 'topicality' (i.e. 'as for Hiroshima, as for oysters, delicious'). In that example

Direct and indirect tactic and non-tactic

relations

287

'Hiroshima wa' not only can be demonstrated not to be the 'subject', by commuting it with Hiroshima de 'in Hiroshima', but the rest commutes with any predicative based syntagm, e.g. watasi ga mizu wo nomimasu Ί drink water', watasi ga anata ga suki da Ί like you', etc. I shall come back to these examples in another context. We now come to the question as to which of the direct and indirect relations are direct or indirect tactic relations. All other relations are, of course, non-tactic. The matter is actually very simple, albeit that whether a relation is tactic or non-tactic is not immediately given in the protocols, but only appears after the immediate constituents have been established. We have seen above how we may arrive at those immediate constituents. The definition of 'tactic relations' is 'constructional relations (whether ordering or not) between syntagm^tic entities, as immediate constituents, in combinations' (def. 7° ). It is, therefore, clear that relations between nuclei of complex immediate constituents, though direct relations, are not direct tactic relations. Relations between a nucleus and a peripheral entity at the same level of analysis are direct tactic relations, and relations between peripheral entities at the same level of analysis are, of course, indirect tactic relations. Indirect relations between peripheral constituents and between those and nuclei, on mutually different levels of analysis, are without interest, and only mentioned for the sake of completeness. They may affect relations of another kind, such as contextual connotational relations, but are from the point of view of syntax irrelevant. For instance in a syntagm such as: the [big] < - - y — > [black] [on < (the > table)] a

> vase > has

> crack

i.e. 'the big black vase on the table has a crack', there are direct tactic relations between 'the' and 'vase', 'big black' and 'vase', 'on the table' and 'vase', 'big' and 'black', 'on' and 'the table', 'the' and 'table', 'a' and 'crack', 'the big black vase on the table' and 'has' and 'a crack' and 'has'. There are indirect tactic relations

288

Linguistic description II: Syntax and the sentential level

between 'the', 'big black' and 'on the table', and between 'the big black vase on the table' and 'a crack'. There are direct non-tactic relations between 'big' and 'vase', 'black' and 'vase', 'vase' and 'has', 'vase' and 'on', 'table' and 'on', and 'crack' and 'has'. There are relations between any other pair of items, but they are indirect non-tactic, and of no interest syntactically speaking. I shall come back to the type of relation between 'big' and 'black', which we encountered already before in 'there under the bridge', in the next section.

3. Functional dependency versus occurrence dependency Thus far, I have been mainly dealing with 'functional dependency', and, within functional dependency, with just one type of relation, i.e. subordination. In passing, we have come across one more type of relation (i.e. 'coordination'), and I have also used 'square brackets' for so-called 'expansions'. The latter refer, in fact, to an entirely different type of relation, which is of a distributional rather than syntactic nature. It is amazing that the difference between functional dependency and occurrence dependency has not been recognized in the past, and has, as far as I know, always been confused, even by the greatest linguists, including Hjelmslev and Martinet. If there are linguists who have not confused those quite different types, I am not aware of it. Though I have not given a separate definition for 'functional dependency', nor for 'occurrence dependency', such definitions can be theorematically derived from the definitions I have given for each of the three types of the former, and each of the three types of the latter. Within functional dependency between two immediate constituents there are, logically speaking, three possibilities, i.e. (a) The one depends on the other for its function, but not vice versa. This is written a > b, or b < a, but order is, as I have said, irrelevant. It is called subordination. The converse of this would, of course, be superordination, but as this is automatic we can ignore it.

Functional dependency versus occurrence dependency

289

(b) None of the two depends on the other for its function. This is written a b. It is, of course, the conflation of a —/—> b (i.e. a does not depend on b) and b —/—> a (or, equivalently, a b). This is called coordination. Its semantic import in English, and in most other languages as far as I know, and as far as it occurs, can more or less be described as 'and-ness', though this is not a theoretical necessity. I have come across two 'exotic' languages, where I have had the impression that the semantic import of this relation could better be described as 'or-ness'. The fact that in English, and many other languages, it is 'and-ness' has led to regrettable confusions regarding the nature of the conjunctive 'and', and its equivalent in other languages. Because 'and' as a sign in, say, 'John and Peter', happens to denote a relation between that which is denoted by John, and that which is denoted by Peter, and there happens to be a syntactic relation with roughly the same meaning, it has been confused with that relation. One has regarded 'and' as denoting a syntactic relation rather than, as it should be, a semantico-material relation, or even worse, one has regarded 'and' as standing for, or being, or implementing, that syntactic relation. Consequently, some have even regarded 'and' as not being a sign at all, but an element with a purely syntactic function, a mere 'coordination' element. Unnecessary to say that, under this theory, 'and' is to be regarded as a 'functional', an ordinary 'conjunctive', and will be treated accordingly. I shall come back to this. (c) Each of the two depends on the other for its function. This is written a < > b, conflation of a > b and b > a (or, equivalently, a < > b). This is called interordination. It is, again, at least to someone who systematically uses logical methods to support his thinking, amazing that other linguists have not recognized this situation. Almost every linguist distinguishes between subordination and coordination, but the possibility of interordination is never referred to. It is understandable that inductivists have not 'discovered' it, as it is relatively rare in languages, and where it appears it is not always very conspicuous. Indeed, I did not myself 'discover' or 'notice' it until after I realized that it was one of the three logical possibilities with regard to a dependency relation, in this case a functional dependency relation, between two immediate constituents in direct tactic relation. It will be clear that the notion, just as that of 'occurrence depen-

290

Linguistic description II: Syntax and the sentential level

dency', only refers to entities in direct relation of immediate constituency, i.e. in 'direct tactic relation'. Examples of interordination are, in English: either a sinner


or a saint

let


working (idem)

him


no pay

>

or in German: wären wir schneller gelaufen < > hätten wir nicht ein Taxi nehmen müssen (if we had walked quicker, then we would not have had to take a taxi). The so-called 'subjunctive mood' of 'wären' and 'hätten' signals that this is an interordinative construction. Also the English translation is, because of 'if . . . then' interordinative, but it is rather contrived stylistically. More normal would be not to have 'then', in which case there is subordination, but 'had we not χ < > we would have y' is again interordination. Many cases of interordination in English have synonyms or near synonyms which exhibit subordination. Even in the case of 'either Λ: orjy', we have a nearly synonymous 'x or y\ This may have created another difficulty, or perhaps even aversions, which has prevented 'interordination' from having been 'discovered'. Once, however, we recognize the possibility of 'interordination', we have to be quite mechanistic with regard to its establishment. That is, we have to treat 'either χ or y' and 'χ or y\ as well as 'if χ then y* and 'if x, y* as syntactically different, and even as belonging to quite different types of syntactic relationship. In order to avoid confusion, I should like to point out that technical terms have to be treated as unanalysable, i.e. as specialized and fossilized. When we call something 'functional dependency' we mean something very specific. One should not regard 'functional' here as an adjective, especially not in the sense of 'all semiotic features are functional', and one cannot argue that the form is a misnomer as in the case of coordination there is

Correlations between functional dependency and occurrence dependency

291

no (i.e. φ) dependency. I do recognize within the type 'functional dependency' three logical possibitites, i.e. a

> b = unilateral functional dependency (subordination)

a a


predicative

>

nomimasu

[mizu wo]

[object]

But both 'ga' and 'wo' (apart from their difficult to paraphrase denotation, marking 'subject' and 'object' respectively) commute with 'wa' (which apart from its difficult to paraphrase denotation could be said to mark the 'topic'). In fact, 'wa' commutes with any post-position governing an entity that under the circumstances could be topicalized, e.g. de 'in, by', ni 'in' e 'towards', etc. If we now commute, 'ga' with 'wa', the relations within the structure, i.e. the syntactic structure, remain the same, though, of course, there is a difference in denotation ('wa' and 'ga' have a different denotation). The same is the case if we commute 'wo' with 'wa'. [watasi wa]

[watasi ga] >

[mizu wo]

nomimasu

>

nomimasu

[mizu wa]

The only difference between those is in the 'topicalization' of 'watasi' and 'mizu' respectively, but this is a matter of semantics, not syntax. It would now be totally absurd if we claimed that when we commute in the above examples 'wo' with 'wa', and 'ga' with 'wa', respectively, we would suddenly have a completely different structure, i.e. a case of diverse determination, i.e. [watasi wa] not:

* [watasi wa] - > nomimasu but:

[mizu wa]

- > nomimasu [mizu wa]

Yet this has been suggested. I can only assume that one was misled by the fact that the English translation of all four cases remains the same, i.e. Ί drink water', the only difference in Japanese being 'no topicalization', 'topicalization of watasi', topicalization of mizu', and 'topicalization of both', and that

298

Linguistic description II: Syntax and the sentential level

it seems that in the latter case 'sequential order' is the means by which the two peripheral positions are distinguished. But 'sequential order' is a matter of realization, not of structure. In this case there are even enough contextual clues for alleviating any ambiguity that might arise to make even 'mizu wa watasi wa nomimasu', without a difference in meaning, possible. It would, probably be hardly more unconventional than say, 'mizu wo watasi ga nomimasu'. If the latter were more likely to be encountered than the former, this could be explained by the fact that the more conventional sequential order provides an extra clue in realization, in addition to the lexical clues, for understanding the meaning of the whole. In the latter example there are already two clues, i.e. the lexical one, and the use of 'ga' and 'wo'. But all of this is relatively unimportant. What is important is the absurdity in both asserting that commutation with 'wa' does not alter the structure, and asserting that it does alter it radically. There is even a greater absurdity in the fact that one can commute one of the two, but, if one then commutes the other, one has to claim that no commutation has taken place, as commutation requires identity of context (including 'structure'). A similar suggestion, i.e. of 'diverse determination' has been made for our earlier example Hiroshima wa kaki wa umai, 'in Hiroshima oysters are delicious', or, in freer but more correct translation 'Hiroshima has delicious oysters', for which we would have, via commutation, the on all levels equivalent structures 'Hiroshima de kaki ga umai', 'Hiroshima wa kaki ga umai', and 'Hiroshima de kaki wa umai'. But here the misconception is compounded by also not using commutation, by which it can be shown that the structure is in fact (representing two levels): 'Hiroshima wa > (kaki wa > umai)'. We have seen some of this earlier. Equally, or perhaps even more absurd is the alternative suggestion to regard it as a case of a 'nested subject' construction, as dealt with in section (2) of the previous chapter, i.e. subject 1st level 2nd level

* Hiroshima wa kaki wa

> predicative > kaki wa umai > umai.

All this serves to show that, when we deal with syntactic (i.e. syntagmatic) relations, we should not lose sight of the paradigmatic

The notion'semi-cluster'

299

relations. That is, we should work with 'position-classes', not with single members which those classes may have in common, and we should not disregard 'commutation' in the assessment of syntactic (or, for that matter, phonotactic) structure. An important other point that emerges from this is that one must distinguish ambiguity in realization from structural ambiguity. This was exemplified by the two structures corresponding to 'he fell under the bridge'. From ambiguity in general one has to distinguish 'indeterminacy' which is similar, but has no structural implications. It is the coincidence of realizational and structural ambiguity.

6. The notion 'semi-cluster' I have not defined, not even implicitly in a theorematic way, the notion 'semi-cluster' in the postulates, as it is a relatively unimportant — though nevertheless convenient — notion. I have referred to semi-clusters in those cases where there is a cluster of two or more phonemes in a single position or archi-position, and within that position or archi-position there is no reversibility. Though being two phonemes (it is important to remember this), they act in a certain respect (i.e. the positional one) as single phonemes. This is not the case with /sTr/, as in 'strength'. Having established the distributional unit, one regards a phoneme as standing in a particular position if it can be mapped onto a particular position. But there are cases that in a cluster of two or more phonemes (or archiphonemes) one cannot assign a position (or archi-position) to each of them. An example is /t$/ as in /tj>er/ ('chair'). There are three pre-nuclear positions involved, i.e. pre, e^, and e^. However in case of pre-nuclear (t$l, there can be no phoneme preceding it, nor following it explosively. As there are two phonemes which occupy a place equivalent to three positions, we can under those circumstances not assign any of the two phonemes to a specific position. All there is left is to regard /tS/ as a whole as standing in a position equivalent to those three positions, i.e. in an archiposition. In that case we may say that /t£/ here acts as a single phoneme. It does not extend over more than one position or archiposition, and within that position it acts in

300

Linguistic description II: Syntax and the sentential level

the absence of permutability (there is no /ST/ in explosive position) as a simultaneous bundle. It is still two phonemes from the point of view of the whole phonotagm, but it acts as a single phoneme (unlike /sTr/ which extends over more than one position, i.e. over three positions in this pre-nuclear context). To mark it as something special I have called this a semi-cluster, and written it /tS/. I do not, however, as I have said, attach great theoretical significance to it. We may now investigate how the acceptance of this notion might affect someone's opinion as to the existence of parallel determination in the English phonotagm. This has in fact been suggested, as I said earlier. All I can see is that one might consider/t/ and /S/ as here standing in a relation of coordination. Also the distinction between sub-ordination, coordination, and interordination, is a superfluous distinction for phonology. But admittedly we can regard /tSI as being one immediate constituent, and let us then assume that it is describable as /1 < —/·—> S/, so that one could say that f t / , as well as /S/ stand in a direct relation with the nucleus. But this would still not be a direct tactic relation, as neither /t/, nor /S/, but only /t$/ as a whole, are immediate constituents. I shall not go into the unlikely possibility of there being a language in which no two position classes are the same. The chances are that, in the absence of distinctive function, phonemes in different positions are mere variants, or that, again in the absence of distinctive function, the supposed phonemes are not composed of distinctive features, and as the whole supposed phonotagm would be a simultaneous bundle, there would in fact not be phonotagms, and not even phonemes, as their establishment presupposes phonotagms. In the first case, different positions would share the same classes after all, and in the second case, the semiotic system in question (i.e. the supposed 'language') would not be a proper language under the relevant definitions. We would then, perhaps, have discovered the first natural language that is not a proper language. Thus, if one wishes, for reasons of purely academic interest, to compare phonotagms with syntagms, one could represent,

The notion'plereme'

301

say, the basic phonotagm of English in the following way (see Chapter V): morphemes pre 1st Ε 2nd Ε 1st I

> Ν

2nd I 3rd I 4th I morphemes This would however, as I have said already, not serve any useful descriptive purpose. It would, moreover, graphically speaking, give a much less clear picture of the situation peculiar to phonology than an 'onion-shaped' type of representation would. But asserting that there is parallel determination in English phonotactics would be absurd. Even asserting that parallel determination could exist in any phonotagm in any language would probably turn out to be absurd, but there might be some merit in investigating such a claim. An investigation of this kind, however, should be based on a thorough and complete understanding of all the issues involved, in order to be, even academically speaking, of any real value.

7. The notion 'plereme' In phonology, the minimum syntagmatic, i.e. tactic, entity is the phoneme. In semiotic systems in general, where the term phoneme

302

Linguistic description II: Syntax and the sentential level

is not always appropriate, it is called ceneme. Comparable to the ceneme in cenological (in spoken language: phonological) systems is the plereme in plerological (i.e. grammatical) systems. The usual term for this is 'word', but this is a fairly imprecise term, as it has, at least in ordinary language, strong realizational connotations. Even most linguists who have employed the notion 'word' have had in their definitions some realizational features in mind, and this has often had unfortunate consequences for tenability and consistency. Some, e.g. Martinet, have rightly, for that reason, rejected the notion 'word' altogether, without, however, replacing it by some other notion. When people talk about 'words', they usually refer to entities that are unified in shape, i.e. not discontinuous, and one is also reluctant to call 'word' an entity such as genitive "s', realized by /z/, /s/, /S/, or /iz/, in spite of the fact that it is clearly a tactic entity, not just a moneme. As 'king's' in 'the king's horses' is the same in realization as 'kings' in 'the kings of France', they are both treated as one word, which in itself may already create a dilemma with regard to the determination by 'the' in the case of 'the king's horses'. Even greater would the problem be in such cases as 'the king of England's horses'. The latter example shows clearly that Ύ governs more than 'England' alone. Note that "s' here has to be regarded as the nucleus in 'the king of England's', i.e. that 'the king of England' is peripheral to "s\ It may appear that 'the' stands in direct relation with 'horses', but it certainly stands in direct relation with 'king'. To say that 'horses' is in direct relation via "s' with the rest, so that the relation of 'horses' with 'king' is indirect, and that with 'the' is even more indirect, i.e. first via "s' and then via 'king', is perfectly adequate in respect of the message. To say, however, that 'the' stands in direct relation with 'horses' would, however, prevent us from being able to account for the obvious direct tactic relation between 'the' and 'king'. Perhaps it could be regarded, but this would in this case be rather trivial, as a case of 'functional amalgamation' (see below). To imply that 'the' does not stand in direct tactic relation with 'king' has however to be rejected because of (albeit indirect) inadequacy. But let us come back to the point. Most adherents of the notion 'word' would probably regard German hinaufgehen 'to go up' as one word, but 'geht' and 'hinauf in Sie geht die Treppe hinauf 'she goes up the stairs' as two

The notion'plereme'

303

words. And what to do with 'garage door'. The layman would regard them probably as two words, because they are written with a space in-between, but the linguist might wrongly consider it as one. The equivalents in German and Dutch, even Hottentottensoldatententententoonstelling 'Hottentot soldiers tents exhibition', would probably count as one word. One could, of course, give a precise and adequate definition of the term 'word' for the purpose of linguistics, but this would (a) be rather different from the traditional notion, and (b) be likely to cause errors in thinking, as the familiar ordinary language notion might interfere. It is for this reason that I have coined a new term, a term which at the same time suggests the isomorphism that exists between minimum tactic (and maximum 'matic') entities in both cenology (phonology) and plerology (grammar). I have in the past still used the term 'word' in contra-distinction with the term 'grammateme', not for an entity, but for different, but complementary ways of looking at the allomorphsof pleremes. I got this idea actually from one of Martinet's (about 1958, in his lectures at the Sorbonne) examples 'la grande montagne blanche' (/la gräd mötan bias/) which he used to explain why the notion 'word' should be rejected. He claimed, rightly, that the units involved are actually /l/, /grä/, / . . . a . . . d . . . mötan . . . §/ and /blä/, i.e. that / . . . a . . . d . . . S/ was merely a matter of 'concord', caused by the 'gender' of 'montagne'. Thus, the moneme in this instance was / . . . a . . . d . . . mötan. . . S/, rather than just /mötan/. That is, it is a case of a discontinuous moneme. Apart from the imprecision involved in calling an allomorph of a moneme a moneme, this view struck me as correct. To give another example, one can regard 'he . . . s' in 'he works' as one single entity from the syntactic point of view. But in the French example, more so than in the English one, it is also clear that the concord involved is in addition due to the signs 'definite article', 'grand', and 'blanc' themselves. Therefore an equally valid view was to regard as the four allomorphs involved /'la/, /gräd/, /mötan/ and /bläS/, rather than as /!/, /gräd/, / . . . a . . . d . . . mötan . . . s/, and /blä/ as in the other view. Note that I write '/la/', etc., rather than /la/, etc., as, unlike Martinet, I distinguish between the notations for 'allomorph' ('/x/') and 'phonological form' (/x/). I have called the first view, i.e. recognizing within the class of allomorphs discontinuous entities, the 'grammateme' view, and the

304

Linguistic description II: Syntax and the sentential level

other the 'word' view. To some classes only the 'grammateme' view applies, to some only the 'word' view, and to yet others, as in the above example, both views are correct. From the structural point of view, in grammatical description, all this is irrelevant, as we are concerned with grammatical entities, not with their allomorphs. The plereme is a simultaneous bundle of one or more monemes, and it is the minimum syntactic entity. The form the plereme takes, discontinuous or not, in amalgamation with other pleremes or not, is irrelevant. The latter is a matter of allomorphy, i.e. of phonological realization, not of syntactic structure. It is therefore, also irrelevant in what way we represent the plereme in setting up structures. We have to use our common sense and good taste in that respect. Usually it is most convenient to use the standard orthography, i.e. to present them as 'words' according to the ordinary writing conventions, only to deviate from this if it interferes with clarity. As I have referred above to the difference between jxj and '/jc/.' I shall explain here the conventions I have adopted. The signum is represented by 'JC', the phonological level by /x/, and the phonetic level by [JC], as is usual in linguistics. Any two of these can be used in combinations. We have then: [*]

/[*]/ Μ

phonetic form allophone phonological form allomorph

'x'

signum.

With a bit of ingenuity one can easily make up some more representations. If one were to symbolize 'image' by 0c>, one could have j{x)j for 'allophonon' and '0t>' for 'utterance'. The notion 'allomorphon' one could symbolize by '[JC]'. But all of those are primarily theoretical notions. They may be useful in definitions, but would scarcely be used in descriptions. It is even doubtful whether one could represent a particular image or a particular utterance at all in any satisfactory way.

Syntactic description

305

8. Syntactic description In order to do syntactic description one must in the first place have an entity of which one assumes that it is a well-formed syntagm of a kind. It is possible that in the course of the attempt to describe an entity, this assumption, i.e. the hypothesis for the well-formedness of that entity, is refuted, but one does start off with at least tentative syntagms. Furthermore, by using criteria discussed in the previous chapter, one does know which are the pleremes in this syntagm. That is to say, one has a tentative syntagm in terms of pleremes. For further protocolization it is necessary to know where there are direct relations between those pleremes. The rest is as has been outlined in previous sections of the present chapter. We proceed in setting up patterns, paying special attention, not only to consistency, but also to adequacy and simplicity. One of the rules in this game, having to do with consistency, adequacy, and simplicity, is that we do not set up two different structures for what can adequately be accounted for by one. Examples for this were given in the previous chapter in connection with the so-called 'nested' subject in Japanese. Further examples will emerge in the following. But equally important it is to set up different structures for what is different. This means, as we have seen, even though there we introduced other considerations, that for the two different denotations of 'he fell under the bridge' we need two different structures. It also means, as we shall see below, that the structure assigned to 'Mary loves a fool' must be different to that of 'Mary is a fool', as it can be demonstrated, by demonstrating difference in semantic import of the syntactic relations, that these syntactic relations do not have the same identity in the two cases. Of course, when we proceed in the syntactic (or any other for that matter) description of a particular language we gradually build up a stock-pile of structures, and we can map new utterances we encounter directly onto those structures without going through the tedious procedure of starting every time from scratch. That is, syntactic analysis of a language becomes gradually more easy, the more of that language we have described. In actual fact, on the one hand all this is easier than one might think, as the number of different patterns to be found in any language is very limited.

306

Linguistic description II: Syntax and the sentential level

On the other hand, it is amazing how many surprises there often are, and this adds again to the difficulty. At any rate, the most difficult stage is setting up the most important patterns, that is, building up one's initial stock pile. In Chapter VII I shall analyse some examples from English. This will further clarify the descriptive principles and the ensuing methodology involved.

9. Concord and congruity Though allomorphy is non-functional, and therefore irrelevant from the point of view of grammatical structure as such, in realization it may be an important clue with regard to the relations between entities, and therefore with regard to the structure. There are, of course, many other clues, lexical ones (as we have seen earlier), accent and intonation patterns, and even connotative stress and connotative modulation. In fact all the clues that enable us to understand the sentence-utterance, which ultimately determines the identity of the tentative syntagm under review. Though in syntactic analysis we abstract from the notion 'sentence' as such, we have to know the precise identity of the syntagm in question, and this presupposes that we can imagine a sentenceutterance, which corresponds to a particular sentence in the base of which this syntagm can be thought of as partaking. To come back to 'allomorphy', the fact that, for instance, in English, and other European languages, the predicative often occurs in a form that 'agrees' with the so-called 'person' of the subject, can be a clue as to which entity is the subject, just in case there is more than one nominal, and therefore more than one candidate, among the immediate constituents. Another clue may, of course, be the order in which the constituents are realized, but in a language such as Latin, e.g. in the case of ego amo filiam Ί love the girl', which could as well be realized 'filiam ego amo', 'amo ego filiam', or 'amo filiam ego', the fact that 'amo' agrees with 'ego' but not with 'filiam' shows 'ego' to be the subject. However, as we shall see, this is not a case of 'concord' which is a kind of 'allomorphy', but of 'congruity'. At this level, however, these two play exactly the same role. A case of real 'concord',

Concord and congruity

307

i.e. of allomorphy of a kind, is the distinction between 'work' and 'works' in Ί work', and 'he works' respectively. A case of real concord we also had in Martinet's example 'la grande montagne blanche'. In order to be able to talk about concord, and the same is true for 'congruity', we have to adopt the 'word', rather than the 'grammateme' type of view. Though we can say that 'he works' exhibits the allomorphs '/hRi · · · S/' and '/urrk/' (the grammateme view), only if we say equally correctly, that the allomorphs are '/hRi/' and '/urrkS/' can we say that the allomorph of the predicative is in agreement or concord with the subject. The term 'concord' (another term is 'agreement', but I prefer to reserve this for both 'concord' and 'congruity', see below) refers to a situation in which allomorphic variation is due to the presence and function of another item. Now let us return to our Latin example 'ego amo filiam'. We cannot say that the particular form 'amo' is due to the presence of a first person subject, i.e. 'ego'. Even if we leave 'ego' out we may have 'amo'. In this particular case, the form 'amo' still implies that semantically speaking it is a matter of 'first person', i.e. '0' here can only commute with 'ego'. As, in addition, it is debatable whether the difference between 'ego amo filiam' and 'amo filiam' is really denotational, or merely connotational, one might perhaps think that by extending our definition of concord somewhat (i.e. by including 'commutation with first person subject') we might still regard the form 'amo' as purely allomorphic. This would, however, lead to serious complications and inconsistencies elsewhere, as the above situation is far from general. For instance in English, we have 'he has worked', 'he is working', 'my father's book', etc. In the first example the form 'worked' is in agreement with the presence of 'has' which is an allomorph of 'to have' in an 'auxiliary' position, which I call 'tense' (see below). In the second example, 'working' is in agreement with 'is', an allomorph of 'to be' in another 'auxiliary' position. In the third example, the form 'my father's', as opposed to 'my father' is dictated by the fact that it stands in the article position determining a nominal. In these cases, neither '-ed' nor '-ing', nor'Y is functional in the context in question, as they do not commute with anything, not even with 'zero'. With regard to '-ed' and '-ing', however, in another function (i.e. position), i.e. as an 'adjectival', it can be demonstrated by commutation that they are functional (i.e. distinctive) at the

308

Linguistic description II: Syntax and the sentential level

grammatical level, and that they therefore correspond to monemes. Though in practice we may regard such forms as 'worked' and 'working' in the first two examples as just contextual variants of 'to work', it would not be entirely correct to call them 'allomorphs' of 'to work', as there is a difference in terms of monemes. This is what I call 'congruity' or 'congruence'. It is for all practical purposes the same as 'concord', but it involves difference of structure, rather than of mere phonological form. The case of 'my father's' may at first sight seem more dubious, as moneme-identity for "s' may seem here more difficult to assert. Though there is a difference between, say, 'father's figure' and 'father figure', there is here no valid commutation between "s' and '0', as the structures are not comparable. But we can establish that 'my father' and 'my father's' are different denotationally, and therefore have a different identity at the sign level. The second is a synonym of 'of my father', whereas the former is not, just as there is a difference in denotation between 'amo' and other forms of amare 'to love'. But there still remains one point to consider. The e n t i t y Ύ is the nucleus in 'my father's', i.e. the entity via which 'my father' is in relation with 'book'. It stands indirect tactic relation with 'my father', and in direct non-tactic relation with 'father', as well as with 'book'. It has therefore a tactic, even nuclear, function. Such cases are not regarded as cases of 'congruity', i.e. agreement of a kind. But it does illustrate just as cases of congruity do, an extremely important point. It is imperative in assessing the identity of an item (a) to take valid commutation of the entities of the context into consideration, and (b) to establish the distinctive function of the item in question within a construction in which that item is an immediate constituent. If we do not adhere to the principle stated in (a), we might describe as being structural, conditions that are purely accidental, e.g. simple cases of non-occurrence for no structural reason. In respect of (b), if we were to widen the context sufficiently, i.e. beyond immediate constituent structure, we might well find all sorts of cases where distinctive function is suspended, and description might ultimately become so messy that it would no longer be feasible. The golden rule of thumb here is — and this ho?ds for phonology as well as grammar — always to establish the distinctive identity of an item in the smallest possible wellformed context. Distinctive features are established in phonemes,

Syntactic patterns in English

309

phonemes in phonotagms, monemes in pleremes, and pleremes in syntagms of a single structural level. Never in a wider context.

10. Syntactic patterns in English Most linguists have attached special importance to the type of minimum syntagm that contains a main verb together with a subject, and perhaps also a compulsory or not compulsory object, and perhaps some other entities. Scholars as widely apart as Martinet and Chomsky have regarded this type of syntagm as a priori sentential. Martinet has called it 'phrase ind6pendent' (independent sentence), and Chomsky has even gone so far as to claim that every sentence in first instance consists of a nounphrase (NP), and a verb-phrase (VP), which latter may, as an object, contain in its turn a noun-phrase. Many other linguists have had similar ideas. Such ideas have necessarily lead to an acceptance of 'linguistic' notions that have more to do with psychology than with linguistics. Such terms as 'psychological subject' and 'implied predicate', in cases that cannot be denied to be sentence-utterances, but in which there is no subject, or no verb, or neither, can hardly be regarded as referring to linguistic entities in the sense of those being observable in the speech phenomena themselves. This has created many dilemmas for linguists who do not adhere to conceptualism, but nevertheless with few exceptions (e.g. some Dutch linguists, such as Uhlenbeck and de Groot) they have not been able to provide a solution to this problem. Only those few linguists who regard 'sentences' as entities of an ontologically different kind than syntagms can easily escape from the above difficulties. It can then be recognized that something is a sentence if and only if it is a model onto which actual utterances can be mapped, utterances, that is, which carry a complete self-contained message. In languages, such utterances are marked by a particular intonation, and there may be other features which mark it as a sentence. It becomes then irrelevant to being a sentence whether the base of that sentence (corresponding to syntactic entities) itself corresponds to a well-formed syntagm, or not, or whether any part of the base corresponds to a

310

Linguistic description II: Syntax and the sentential level

well-formed syntagm or not. It is then ijo longer necessary for every sentence to contain explicitly or implicity (sic) a subject and a verb as minimum requirements. Sentences can be: 'John!', 'Yes.', 'Voilä, un livre.', and even, in written language, titles of books, headings in newspapers, and so on. It goes, therefore, without saying that the 'subject and verb' type of syntagm cannot be connected to the notion sentence any more than most other types of syntagm, or indeed single pleremes can. If we are to assign it a special status, it has to be based on something else. I base it myself mainly on descriptive convenience, but there are some other factors. If, for instance, in a complex syntagm there is somewhere a verb, and/or a subject (a 'subject' is far from compulsory in many languages), the syntagm in which they figure 'tends' to occupy a central place. Furthermore, most other types of syntagm, and the same goes for pleremes, provided that they can stand in syntactic relation with other syntactic entities at all, can occur as expansions in positions within, or peripheral to, that syntagm. It is, therefore, convenient to start with this type of syntagm, which has many sub-types, as we shall see, and then to proceed to the other types of syntagm. I shall call this type of syntagm in English 'predicative based syntagm', abbreviated PBS. The 'definitions' are not definitions in a theoretical sense, but rather generalizing descriptions for English.

(a) The predicative

based syntagm

(PBS)

A PBS has a predicative as its nucleus. It has one or more peripheral positions, of which one is 'subject'. It should be understood that the terms used for positions are mere arbitrary labels. They are names for those positions, and may not be appropriate for every item that commutes in such a position. I also should like to reiterate that the terms are construction-specific, e.g. a term such as 'subject', or such as 'complement', in one type of construction should not be taken as referring to the same identity as the same terms 'subject' or 'complement' in other types of construction. Except in the imperative PBS, the subject is compulsory, i.e. it is not an expansion. Again except in the imperative PBS (and in the 'subjunctive'), the form of the predicative, or of one of

Syntactic

patterns

in English

311

its auxiliaries (if any), is in concord with the 'person' of the subject. According to whether the nucleus of a PBS may or may not contain a direct and/or indirect object, and/or a complement, I divide the PBS into several sub-types. The identity of the entity in nuclear position determines the sub-type. That is, it is equally correct to speak about, say, a transitive or intransitive predicative, as it is to speak about a transitive or intransitive PBS, though the two should, of course, be distinguished. I have adopted the convention of distinguishing between the prefix 'in-' and the prefix 'non-' in the nomenclature. Whereas, for instance 'transitive' implies must have an object, 'intransitive' implies cannot have an object, and 'nontransitive' implies may have, but must not have an object, e.g. 'John eats [soup]' is nontransitive, 'John gives him a book' is indirect transitive. 'John buys [him] a book' is non-direct transitive. The compulsory nonoccurrence of the indirect object is only indicated by the omission of either 'indirect' or 'non-direct'. The same applies to the notion 'complementary', i.e. 'complementary' implies there must be a complement, and 'non-complementary' implies there may be a complement. However, as it seems that all PBS types may have certain types of complement in addition (of a qualitative or modality type, to be further investigated) I omit this in the labelling. We shall come across this in a later section, when we deal with examples, but in the patterns I ignore it for the time being. The type of complement I have in mind is easily mistaken for an "adjunct' (see below), but commutation may show that it is not. An example is 'John is seen [by Harry] [near the river]' where at least in one of the possible interpretations 'near the river' is a 'complement' rather than an 'adjunct'. But as I said, further investigation into such cases is needed. A syntagm such as 'John put it on the table', where 'on the table' is said to stand in complement position, is called 'complementary transitive'. 'John lays it out on the table' (to 'lay out' is the nucleus) is 'non-complementary transitive'. There is also a type 'non-complementary intransitive', and some more types, as we shall see. (aa) The intransitive PBS Examples: 'John sleeps', 'the man walks', 'my son's cat sleeps and snores', 'John is a fool', 'John is foolish'.

312

Linguistic description II: Syntax and the sentential level

subject John the man my son's cat John John —

> > > > > >

predicative sleeps walks sleeps and snores is a fool (see below for explanation) is foolish (see below for explanation)

(ab) The transitive PBS Example: 'the man hit his dog', subject object

— > predicative

the man his dog

-> hit

(ac) The non-transitive PBS Examples: 'Mary eats soup and potatoes', 'Mary eats'. subject [object]

-> predicative

Mary -> eats [soup and potatoes] (ad) The indirect transitive PBS Example: 'John regards Mary to be a fool', subj' indirect object

> predicative

direct object John Mary to be a fool

> regards

Syn tactic patterns in English

313

(ae) The nondirect transitive PBS Examples: 'John buys Mary a book', 'John buys a book', subj' [indirect object]

> predicative

direct object John [Mary]

> buys

a book One may object that one could have 'he buys and I sell', i.e. that the object is not compulsory. It is doubtful, however, if 'buys' and the same goes for 'sells' here is not a homonym of the other 'buys'. At any rate, this would call for a separate investigation. As it concerns semantics I shall not go into it here. Perhaps this example is in actual fact a non-complementary nondirect transitive, i.e. subj' [indirect object] -> predicative direct object [complement] as one can have 'John buys [Mary] a book [for her father]'. I shall however keep here things as simple as is adequate for the purpose. Such questions would, of course, have to be resolved in a full description of English syntax. The purpose of the present, and also the next section is only exemplification. If this amounts to a rather extensive treatment of English syntax, it is an extra, though contingent, bonus.

314

Linguistic description II: Syntax and the sentential level

(af) The complementary transitive PBS Example: Ί put it on the table', subj' > predicative

object complement I it

-> put

on the table (ag) The non-complementary transitive PBS Example: 'John lays it out [on the table] \ 'John lays it out', subj' > predicative

object [complement] John

> lays out

it [on the table]

Another example is 'John throws the bottle [into the river]'. (ah) The non-complementary intransitive PBS Example: 'John jumped [into the river]'. subj' —>

predicative

[complement] John — > jumped [into the river]

Syntactic patterns in English

315

Note that, unlike in cases where there is both a subject and an object position, this is a case of parallel determination. As 'subject' and 'object' correspond to identical position-classes, if there are both these positions in a syntagm, there must be diverse determination. This implies, of course, that Ί ~ me', 'he ~ him', she ~ her', and 'we ~ us' are regarded as allomorphs, which is both adequate and the simplest solution because of commutation with nominals or nominal syntagms (dealt with below). (ai) The complementary intransitive PBS A case of this type is listed by Suliman Hadj'-Mohamed (HadjMohamed 1980), i.e. 'he stepped into the garden'. He has, however, overlooked that this, and also the previous type, is not a case of diverse, but of parallel determination. Hadj-Mohamed has applied my methods to Kamali Arabic and compared the patterns established with English patterns. In doing this he has discovered four more types for English than I had already established at that time, plus one type with which I tend to disagree (see below). His treatment contains an extensive discussion of the various patterns. Another example is 'he is in the house', where 'is' is a verbal predicative, not a copula, as some linguists seem to think. (aj) The non-direct non-transitive PBS This example too is from Hadj-Mohamed, i.e. 'she teaches [the boys] [physics]'. (ak) The optional indirect transitive PBS This is the term Hadj-Mohamed has used for a type in which either an indirect, or a direct object must be present, but both may be present. More or less in consistency with a notation I use in the 'nominal syntagm (singular)' (see below), he represents it as follows (presentation slightly changed by me): subject indirect object direct object

he >

predicative

her the story

—>

told

316

Linguistic description II: Syntax and the sentential level

(al) The optional complementary transitive PBS Also from Hadj-Mohamed, i.e. 'she argued [the case] with him', 'she argued the case [with him]. subject object

she > predicative

complement

the case

> arguec

with him

It is, perhaps, debatable if this is a correct example, i.e. 'he argued' may have to be regarded as well-formed, unless we can establish 'homonymy'. I list the type, however, as a possibility. (am) The non-direct non-complementary transitive PBS. Also from Hadj-Mohamed, i.e. 'she left [him] a message [in his pigeonhole] subject [indirect object] -> predicative object [complement] she [him] ->left a message [in his pigeonhole] (an) The non-instrumental intransitive PBS This is, like the next type, a so-called 'passive'. An example is: 'John is seen [by Harry]'. The reason why I have not regarded 'by Harry' here as an ordinary 'complement' is, as we shall see in the next type, because the position classes are not mutually exclusive,

Syntactic patterns in English

317

and moreover, an entity such as 'by Harry' does not commute with items we have called 'complements' in other structures. A counter-example may seem to be 'he was killed by the river', which is ambiguous. Because there is homonymy involved here, either the syntactic relations must be different, or one or more of the 'signs' employed must be different, or both (this is theorematic). In my opinion it is 'both'. The structures are certainly different, i.e. in the case of 'by the river' indicating a location, 'by the river' determines the whole PBS, i.e. it is outside the PBS. But there is also homonymy in respect of 'by', as one 'by' is more or less synonymous with 'in the vicinity o f and the other with 'by means o f . subj'

John > predicative

[instrumental]

> is seen [by Harry]

(ao) The non-instrumental non-complementary intransitive PBS Example: 'The bottle is thrown [into the river] [by John]'. This is another passive, and again a case of parallel determination, as there are no two identical classes, i.e. the bottle [by John]

> is thrown

[into the river] subject [instrumental]

> predicative

[complement] Other passives can be added, corresponding to other types of 'transitive', amongst them the 'non-instrumental transitive', e.g. 'Mary is bought a book [by John]', corresponding to the 'indirect transitive'. Also interesting is 'Mary is regarded to be a fool [by John]', corresponding to the 'indirect transitive' (see (ad)). Unless we regard 'to be a fool' here as an allomorph of 'a fool' (or as a case of congruity), we would have to concede an anomaly, i.e. that in such cases the 'object' position class is not isomorphic

318

Linguistic description II: Syntax and the sentential level

with the 'subject' position class. This would have implied 'parallel determination', rather than 'diverse determination'. (ax) The copulative PBS This is not a separate type, but just another intransitive, including the non-instrumental, and the non-instrumental non-complementary transitive, but not the non-complementary and the complementary intransitive. It is only on the second level of analysis, i.e. when we analyse the predicative, that there is a difference with other intransitives, albeit that the non-instrumental and the noninstrumental non-complementary intransitive, i.e. the so-called 'passives', do not have a non-copulative counterpart. Nevertheless, though the 'passives' have a different pattern from any of the others on the first level of analysis, on the level where we speak of 'copulative' (as opposed to 'verbal') they are of the same structure, i.e. copula < complement ('complement' here, of course, not being comparable with 'complement' in other types of construction). It is only on yet another level down, where we look at the 'complement' within the copulative syntagm that they become different again, e.g. John

> (is
(is
fool))

John > (is
bought

5vhicli [for . . . wedding]

(5c) Extended quasi-verbal modal Remember that the pattern of this is similar to that of an ordinary EPS, see (4c). [may]

[modal] [tense] [aspect]

> quasi-verbal modal

Φ

— > appear to

Φ

[modality]

Φ

(5d) Copulative syntagm That it is of the passive type, i.e. involving one of the passive patterns, see (2a), is here no longer relevant. (5e) Adverbial

determination

As I have said before, the nature and syntactic deployment of so-called 'adverbs' is in need of further research. The type exemplified by 'just' is especially interesting, as it has a very wide distribution and a very wide range of potential peripheralness. It generally determines what follows. Though it may seem materially adequate to regard it as merely determining 'a flaw', this would

Further examples of syntactic analysis and additional remarks

397

interfere with the further analysis. Therefore I adopt the at least equally adequate solution of regarding it as determining the whole of the nominal syntagm. One may briefly consider it to determine 'flaw', just as 'a' and 'in the pattern' do, but this would lead to a revision of the nominal syntagm pattern. To regard it as solely determining 'a', though seemingly not materially inadequate, seems a bit farfetched. adverbial

>

nominal syntagm

just

>

a flaw in the pattern.

(6a) Nominal syntagm (singular); 'the' in 'article' position. (6b) Nominal syntagm (singular); 'my' in 'article' position. (6c) Prepositional syntagm for
Mary', which we regard as the nucleus of a conjunctive construction, the peripheral entity being the conjunctive syntagm 'and < Peter'. But the first mistake in this is thinking that there is similarity of semantic import between a relation of coordination, and the syntactic relation (of subordination) in a conjunctive construction. In fact, the similarity in meaning between 'Lucy < — / — > Mary' and, say 'Lucy < and Mary' is due to different things, a syntactic relation in the first and the use of 'and' in the second. This is quite accidental. If there had been 'or' instead, there would have been no similarity. But there is similarity if we have, say, 'Lucy Mary or Peter'. This exposes the second mistake, i.e. it refutes the assumption of adequacy of analyzing the example (representing two levels) as '*(Lucy < — / — > Mary) < (and < Peter)'. This is immediately clear if we compare this with '*(Lucy < — / — > Mary) < (or < Peter)'. The ' < — / — > ' relation, if there were one, would have a different semantic import in the two cases, which is absurd. It is clear that 'and' or 'or' function between 'Lucy' and 'Mary' as much as they function between those items and 'Peter'. That is, we have here a case of conflation, i.e. '(Lucy < ( and < Mary)) < (and < Peter)', or

Further examples of syntactic analysis and additional remarks

409

'(Lucy < ( Γ0ΓΙ < Mary)) < (or < Peter)'. It would, however, be equally adequate and simple to present those as 'Lucy < ( | and < (Mary < (and < Peter)))', etc., or, indeed, if we have more complex cases, say 'Lucy Mary Peter and Gretchen' to present this in several different ways. That is, we have here again an 'onion-shaped' situation. If we had, say, 'Lucy and Mary and Peter and Gretchen', or the same, but with 'or' this would not make any difference, except that there is no conflation. I propose therefore to write this pattern simply as Lucy < ( 1 and | < Mary) < ( [and] < Peter) < (and < Gretchen), etc., and call it a multiple conjunctive construction. That is, because the division into levels is arbitrary, I regard it as being two, rather than a layer of three or four, etc. levels. I still do not leave out the brackets completely, and regard it as one level, as this would be inconsistent with the distinction between 'conjunctive constructions' and 'conjunctive syntagms'. I should like to say one thing more. I am often asked by 'intuition-free' people (in itself not a bad state of mind), why I analyze 'John and Mary' as 'John < (and < Mary)', rather than as '(John > and) > Mary'. The latter is as simple, and seems equally adequate, though to most people counter-intuitive. Now, even if it were completely arbitrary which of the two to choose, it would still be considerate to choose the one which is least counter-intuitive, either to oneself, or to a majority of speakers. But it is good to find some reasons why intuitions are likely to be as they are, i.e. to reduce them to reasoned intuitions. In the first place, we notice that sentences tend to begin with 'and', 'or', or 'but', more than that they end with conjunctives. Though in a sentence almost anything may happen, one regards intuitively sentences as more well-formed (no matter how fictitious this is in actual fact) the more elements within their base correspond to well-formed syntagms. A sentence such as 'But I told you so.' would be regarded as perfect, but a sentence such as Ί told you so, but . . .' as imperfect, and in fact unfinished. Note also that clause-borders tend to be immediately before conjunctives, rather than immediately after. In realization there is a general tendency to have pauses coinciding with the borders of syntactic entities, and the more complex they are, the more conspicuous the pause (which is either a para-phonotactic

410

Further examples of syntactic analysis and additional remarks

or a para-syntactic feature) usually is. This points to 'John < (and < Peter)', rather than '(John > and) > Peter'. The realization of, say, 'John and Peter and Mary and Isobel and Irma' is rather likely to be 'John . . . and Peter . . . and Mary . . . and Isobel . . . and Irma' than 'John and . . . Peter and . . . Mary and . . . Isobel and . . . Irma' (let alone 'Irma and'). All this is sufficient reason to regard the pattern to be 'Isobel < (and < Irma)', rather than '(Isobel > and) > Irma'. For most people all this will go without saying, but I have been challenged on this on several occasions by people who — in a sense justly, but there is a danger in it as well — reject anything that has the smell of 'intuition'. But if at the end of the day a choice is arbitrary, it is advisable to follow one's intuitions, rather than go against them. Completely out of the question would be an analysis into three immediate constituents, as this would imply that 'and' (or 'or') were nuclear, which is inconsistent with the fact that it is part of an 'expansion', and also because this would imply diverse determination. The latter would be inadequate in respect of the message conveyed. Let us now proceed with the analysis of our example. (1)

PBS (intransitive) 'the . . . language'

> 'is . . . either'

(2a) 'Nominal syntagm', with 'the' in 'article' position, 'Sanskrit language' in nominal position. (2b) Copulative syntagm 'is'
'refined'

This is ambiguous. If we take the second reading we have a slightly different analysis at the next level, i.e. 'more' > 'exquisitely'. I shall, however, take the first reading. (6b), (6d), (6f) Comparative syntagms 'complement JC'

'than'
'refined'

(8a), (8b) Compound nominals, as (3a), but involving conflation. 'Greek

>

language

', 'Latin

>

language

It should be noted that 'either' in ( 6 0 is a 'pronoun'. It acts as a 'functional', but of a very special kind in such constructions as 'either χ or y'. the latter is an example of interordination, i.e. 'either JC < > or y\ unlike, say, 'x < or y\ which is a straightforward example of a conjunctive construction. Yet 'or >>' in both cases must be considered a conjunctive syntagm. But what is 'either *'? It only occurs in this type of construction, and seems in all respects to be equivalent to the construction with 'or'. In the absence of evidence to the contrary I consider 'either JC' to be a conjunctive syntagm as well. I have in an earlier publication, tried to solve this by posing here, bilateral conflation, i.e. '(either . . . |or| < x) < > (or . . . jeitherf < y). (See Mulder 1980, published in Mulder - Hervey 1980). This is, however, unsatisfactory, as it is difficult to see what 'either . . . or' as a conjunctive in both parts of the interordinated construction could possibly denote. There is not such a sign as 'either. . . or', unless, rather artificially, we create it in the syntactic part of our description. A better solution may seem to be to regard 'either' as a homomorph of 'either' as a pronoun, and as an allomorph of 'or'. The difference in meaning between 'either' and 'or',

Further examples of syn tactic analysis and additional remarks

413

which may appear to be perhaps one of connotation rather than denotation, could then be accounted for by the difference in construction, i.e. 'interordination' versus 'conjunctive construction', and by the fact of having twice the sign 'or' (allomorphs 'either' and 'or'), rather than once. The latter would explain that in the 'either χ < > or y' construction the meaning of 'disjunction' is connotatively stronger than in the case of 'x < or y\ Though we are not allowed to use this as an argument, it is interesting to know that in German we have in a similar fashion 'entweder JC' < > 'oder y', but in Dutch we have 'of x' < > 'of y', both translating in English as 'either χ or y'. Especially the latter, though, as I said, it cannot be used as an argument, may seem to lend some plausibility to this solution, as the structures are of a similar kind ('entweder JC < > oder y', 'of χ K. > ofy'). However, all this rests on the acceptance of homonymy between 'either' as a pronoun, and 'either' as a conjunctive. And what about 'either' in, say, 'he does not know that either'? The homonymy thesis presupposes that we have refuted the valid hypothesis, i.e. that there is sign-identity between the two (or the three), and for that we need denotational criteria. But this is exactly where our difficulty resides. What does 'either' mean? Let us investigate this, and see whether via this another solution suggests itself. We have, of course, to be careful not to base our solution on 'meaning', but we have to find an independent solution which is consistent with it. From such constructions as 'both χ and y' and their meaning as compared with 'either JC or y\ we may reach the conclusion that whatever the meaning of 'either' is, the meaning of 'both' is to 'and' as the meaning of 'either' is to 'or', i.e. 'both': 'and' :: 'either' : 'or'. A similar proportionality pertains to 'neither' and 'nor'. The third use of 'either', e.g. in 'he does not know that either' is always with 'negation', i.e. 'not . . . either', so we can equate this with 'neither'. Its 'and' or 'both' equivalent is, semantically speaking, 'not both', or one of its synonyms. Now the meaning of 'both' can be roughly paraphrased as 'the one and the other', 'the one and also the other', or even, in certain contexts, 'also', 'too' or 'as well', or, at least conceptually speaking, those meanings are implied. A contrary of 'he does not know this either' is 'he knows this too', i.e. in rough paraphrases 'he neither knows

414

Further examples of syntactic analysis and additional remarks

this nor this' versus 'he knows both this and this'. Also 'neither' and 'both' occur as pronouns ( Ί want neither', Ί want both', Ί want either'). It appears then, in comparison with 'both', which can be paraphrased as '(and) χ and not x\ 'either' can be parapphrased as '(or) χ or not x', and 'neither' as '(or) not χ or not not x. If we now have the construction, say, 'both John and Mary', we have jc, as well as 'not x' specified, i.e. they are 'John' and 'Mary' respectively. So we have 'both John and (not John, i.e.) Mary'. In a similar fashion we have 'either John or (not John, i.e.) Mary', etc. That is the contrary is 'implied' (to put it no more strongly than that) in the denotation of 'both', 'either' and 'neither', and this gives a 'disjunctive' flavour to their denotations as compared with those of 'and', 'or', and 'nor'. That is, we have established 'both ~ and', 'either ~ or', 'neither ~ nor' as different pairs of signs. They are not pairs of allomorphs, nor is there homonymy involved in the use of those forms in other contexts. The deployment of 'both', 'either', and 'neither' is different from that of 'and', 'or' and 'nor' respectively, so strictly speaking we should not regard them as conjunctives. They must in their usage in the 'interordination' pattern, however, be regarded as 'functional' of a kind. But they do not belong to any of the already established classes. Because of their connection with the conjunctives 'and', 'or', and 'nor', however, both denotationally and syntactically, I shall regard them as a sub-class of 'conjunctives', and call 'both x', 'either x', and 'neither x' conjunctive syntagms, i.e. 'both < x', etc. But an important difference is that these conjunctive syntagms do not partake in conjunctive constructions. Another respect in" which 'both', 'either', and 'neither' differ from 'and', 'or' and 'nor' is that the former belong also to the category of 'pronouns', whereas the latter do not. That is, they belong to the intersection of two distribution-classes. 'Neither' (or 'not . . . either'), because of its 'not'implication, has in addition also 'adverbial' deployment, e.g. in Ί do not know it either', 'neither does he know it'. The same is true of 'nor does he want to come'. In the latter two cases there is functional amalgamation, as 'neither' and 'nor' in this usage are conjunctives as well. This supersedes what I have said in Mulder 1980, but some of it is not all that far removed from what I had in mind at that time. The above example, i.e. 'the Sanskrit language, etc.' is interest-

Further examples of syntactic analysis and additional remarks

415

ing because of its regularity and its 'nesting' appearance. It is good practice, before starting to analyze an example, to have a good view over the general structure, and especially — just as in a piece of music which one is going to play for the first time — to pay attention t o the repetitions and variations within repetitions. F.

'although

4b

entirely4c

2a the

4a

factory

3a

4c

was

J 3 a destroyed 3a by 4d the

i 3a,almost ^explosion

i t J 3b the 4e workers •^managed to 4 f g e t 5blI ' 3 b11 I I 5b I "^working 3b again 2 b w i t h i n 5 c a 6 w e e k '

(1) EPBS, with 'although determining an EPBS. (2a) Conjunctional

explosion' in 'condition' position,

syntagm

'although'
PS

[adverbial complement] the factory [by the explosion]

-> was destroyed

[almost entirely] Adverbial types of complement can occur with any type of PBS. They are always expansions, as far as I have come across them. This means that to every label for a type of PBS we could add 'non-adverbial complementary', or something similar. It would, however, be rather superfluous to do this. Note the use again of the vertical broken line, which indicates that there are four, not

416

Further examples of syntactic analysis and additional remarks

five, immediate constituents, i.e. that one of those (i.e. was . . . destroyed) is discontinuous. (3b) PBS (transitive), with optional adverbial complement. subject object

the workers > EPS

[adverbial complement]

it [again]

> managed to get working.

It is, as I have said before, often difficult to decide whether an item is a 'complement' or an 'adjunct'. In cases as the above, where the item is realized immediately after the verb-group, the 'complement' thesis is most likely to be correct, but there is still ambiguity involved in many such cases. Much less ambiguous are such cases as 'again the workers managed to get it working', 'the workers again managed to get it working', and 'the workers managed again to get it working'. Unless in such cases there are lexical, or, in realization, prosodic clues to the contrary, one is dealing with an 'adjunct'. As I have said, 'adjuncts' in English tend to be rather free in sequential realization. If one is not sure about an item of this kind when it occurs in an 'ambiguous' place in the realization, one might try whether changing its place to a less ambiguous 'adjunct' position (in a realizational, not a syntactic sense) is possible with regard to the meaning we know the corresponding utterance to have. If we are dealing with functional syntagms (especially prepositional syntagms) as complement, we always find a strong connection between the denotation of the 'verb' and that of the functional, e.g. 'jump into', 'jump over', 'fall off', etc. This can also be used as a clue as to whether we are most likely dealing with a complement or not. And one should not forget that when faced with ambiguity, we may arbitrarily choose the solution we fancy most. Notice again the vertical broken line. (4a) Nominal syntagm (4b) Adverbial

'almost'

(singular)

determination

> 'entirely'

Further examples of syn tactic analysis and additional remarks

417

(4c) Copulative predicative (passive) (4d) Prepositional syntagm 'by'
PS

Φ

[aspect]

Φ

[modality]

Φ

- — > get working

There are several things to be said in connection with this analysis. In the first place, one has probably already realized that the 'verbs' corresponding to 'quasi-verbal modals' seem invariably to be 'transitive' verbs. As this type of modal is characterized by an allomorphic 'to' (after the verb-form), and 'to' (before a predicative, a PBS, or a semi-PBS), just as '-ing' forms, also occur in a capacity as 'substantifiers', e.g. 'to be or not to be', there may seem to be some arbitrariness in regarding them as 'modals', rather than as 'transitive' main verbs, with, in this case, 'to get working' as object. One could, of course, decide that, if commutation with an '- ing' form is possible, one should regard it as an object, e.g. 'managed getting it working', but perhaps not *'wants getting it working', though this might occur as 'sub-standard'. Note the difference between 'wants to feed', and 'wants feeding'. But in the first place there are numerous cases in which it is difficult to make a decision on those grounds, especially as 'sub-standard' is not a precise concept. In the second place, the 'modal' solution is probably appropriate to all cases, the 'object' solution only to a sub-set of those. Therefore, one should give priority to the 'modal' solution, and only adopt the other one if there are strong unequivocal reasons for this, e.g. if the two solutions would correspond to differences in meaning. A second point I should like to raise, but this applies to EPS in general, is that it is semantically irrelevant which of the Verbal' entities in the EPS is morphologically complex in that it contains a 'past tense' moneme. In English, and many other languages, a

418

Further examples of syntactic analysis and additional remarks

denotation 'past', whether realized by syntactic, by morphological, or adverbial means, calls for 'congruity'. Congruity, though it may have grammatical (semi-realizational) implications is actually a matter of 'semantic compatibility'. In some languages, e.g. Malay, (but the same is perhaps true for a majority of the world's lanuages) where there is no 'conjugation' in our sense, the meaning of Ί came already', or Ί have already come' is expressed by Saya sudah datang Ί already come'. Ί am already coming' is expressed by Saya sudah mau datang ( Ί already want come'. But in English, if one has 'already', the PS or EPS to which it is ultimately related must either involve a syntactic, or morphological past tense, or both. Which constituent in an EPS is morphologically marked for 'past' (in case of the morphological past tense) is semantically irrelevant (but there are strict rules for its grammatical realization), e.g. 'wanted to come', 'wanted to be coming', 'was coming', 'did come', 'came', etc., but not *'wanted to came', * 'wants to came', *'wanted to was coming', etc. In the case of 'aspect' (syntactic or morphological) there is no such congruity, except between syntactic and morphological aspect itself, but not vice versa. E.g. 'he constantly works', as well as 'he is constantly working'. The item 'working' in 'get working' ('aspect' in English has an 'iterative' denotation) involves morphological aspect. But one can also have 'to get to work' (no 'aspect'), 'getting to work' ('aspect' marking 'to get'), and 'getting working' ('aspect' marking both), with corresponding denotational differences. (5a) Nominal syntagm (singular) (5b) Compound predicative 'get'
'working'.

(5c) Prepositional syntagm 'within'
quasi-verbal modal

[have] Φ [not]

> dared to

420

Further examples of syntactic analysis and additional remarks

(4b) EPBS with 'in the early . . . champion', in 'time' position. (5a) PBS (intransitive). Τ

> 'used . . . instructor'

(5b) Coordination, 'in . . . thirties' < — / — > 'when . . . wrestling champion'. As it stands, in the absence of strong prosodic clues, it is ambiguous whether 'when . . . champion' is not a case of 'a-syntactic relation' (here perhaps to be called 'apposition'). Even with 'intonational' clues, it could remain ambiguous. In syntactic analysis one should give priority to the possibility of establishing syntactic relations over posing an 'a-syntactic relation'. Theoreticaly, there is a third ambiguity, i.e. 'when . . . champion' directly determining 'thirties', but to assume this possibility is in this particular context farfetched. But one does have such cases as 'the time when he arrived', where 'when he arrived' stands in 'relative supplement position' in a nominal syntagm. The item 'used to be' is a fossil. Note that even the 'past tense' is not really 'past tense' as it is merely part of the form of the plereme. There is no 'present tense' equivalent, so the apparent 'past tense' is not functional. Of course, it is a feature of its denotation, but this is grammatically irrelevant. (6a) Copulative syntagm. 'used to be' < (6b) Prepositional syntagm. 'in'
predicative

[aspect]

[have] Φ

> kicked up such a fuss

[not]

[modality]

(4a) PBS (transitive), involving conflation subject

John > EPS

object

-> had said t h a t . . . reason

(4b) Conjunctive syntagm 'and'
EPS object

-> known t h a t . . . reason

Further examples of syn tactic analysis and additional remarks

42 3

(7a) EPS, involving conflation in respect of 'had' in tense position [modal]

Φ

tense

—>

predicative

had

[aspect]

Φ

[modality]

Φ

> known

(7b) Relative syntagm 'that' < (8)

'Jane . . . reason'

PBS (non-complementary intransitive)

subject

Jane

> EPS

[complement]

> had . . . listen

[to reason]

(9a) EPS, with 'had been inclined to' in 'modal' position, determining 'listen'. (9b) Prepositional syntagm. 'to'
consideration

[that. . . conveyed] [of adequacy] Φ

(5a) Adverbial determination. Adverbial determining a conjunctior 'just'

> 'as'

(5b) Prepositional syntagm. 'in'
predicative

[adverbial] thg [with . . .conveyed]

— > has to do

[directly] The item 'has to do' is a fossil, i.e. a single plereme. (7a) Prepositional syntagm. ' o f '
nominal

[relative supplement] [other supplements] the Φ [actual or potential]

> message

[conveyed] Φ If it had been 'the conveyed message', rather than 'the message conveyed', the item 'conveyed' would have been in 'adjectival' position, as it commutes with items in that position. (9) Conjunctive construction, 'actual'< (10) Conjunctive syntagm. 'or'
EPS

-> do not deny that. . . black

object

(3a) EPS, with 'do not' in 'modality' position, if we assume that 'do not' is realized with a 'neutral' accent, and is equivalent to 'don't'. If we assume there to be an accent on 'do', the analysis involves an archiposition for 'do', and is as follows: [modal/ tense/aspect ] [modality] [do]

-> predicative

> deny

[not] In that case 'do' is as in 'we do deny'. I shall, however, assume the former, i.e. 'do not' in 'modality' position. The item 'of course' is a fossil. (3b) Relative syntagm. 'that' < (4)

'due . . . /blak/'.

EPBS, with 'due . . . blackbird' in 'condition' position determining 'some . . . black'.

Further examples of syntactic analysis and additional remarks

(5 a) Conjunctive

431

construction

'due to . . . /blak/'
EPS

complement some . . . speakers blackbird

—> may well associate

with black (6a) Conjunctional syntagm 'due to'
nominal

[some . . . most] Φ

> speakers

Φ (6d) EPS, with 'may' in 'modal', and 'well' in 'modality' position. Note that 'well' here commutes with 'not' in that position, so it has to be regarded as determining 'associate' in an equivalent way. (7a) Interordination. 'the form' < and 'proper name', e.g. 'Doctor' < this with (E3a).

> '/blak/' Not unlike 'title' > 'Johnston'. Compare

(7b) Nominal syntagm (singular), with 'the' in 'article' position. I have not kept to the exact level numbering here, as I wanted to mark the whole of 'the form /blak/' as something special, and deal with it in one analysis, i.e. '(the > form)' < > '/blak/'. (8a) Conjunctional syntagm. 'to' < 'even . . . a knowledge . . . blackbird'. The item 'to' here is an allomorph of 'due to', even though it is a free variant. But it cannot be said to be a synonym, as the form 'due to', is not in free variance with 'to' at (6a). This is one of the few cases in which an allomorph is a free variant, but it is in a negative way a contextual variant as well, as it cannot occur in all contexts. It cannot consistently be said to be a case of conflation, i.e. *' due to', as 'due' is not a plereme, not even a moneme, but mere part of the form. One could, however, call it 'quasi-conflation'. (8b) Conjunctive construction, 'some'
nominal

[qualitative supplement] a Φ [of the . . . blackbird]

—> knowledge

(10b) Conjunctive syntagm. 'or' < (11a) Prepositional syntagm. 'of
'most'.

(12a) Nominal syntagm (singular), with 'the' in 'article', and 'of blackbird' in 'possessive supplement' position. (12b) Coordination, ' p e r h a p s ' < — / — > 'even'. (13) Prepositional syntagm. Note that 'blackbird' is here not a nominal, but a 'quote'. It refers to a particular word, not to an entity denoted by that word. This concludes the analysis. Some points in the analyses of the above eleven examples will inevitably raise eyebrows, either because to some people some decisions may be counter-intuitive, and others may ask 'how do you know that this is correct, and that there could not be a better solution?'. Let us first deal with the 'intuition' issue. Intuition is largely a matter of conditioning. It is subjective, and therefore not a valid criterion, though it may be an incentive for looking for a better solution, as a suspicion of incorrectness has been created. As I said, however, intuition is largely a matter of conditioning, and none of my analyses are any longer counter-intuitive to myself and most of my students, though some may have been counterintuitive in the past. We have got used to this kind of analysis, and what one is used to is generally not counter-intuitive. As to the question 'how do you know?', the answer is that I do not know. I 'know' nothing. Presuming otherwise would bring me into conflict with the hypothetico-deductive philosophy in any of its forms. I do not claim, nor attempt, to present the truth,

434

Further examples of syntactic analysis and additional remarks

only the best solution. My sole criteria are those of 'appropriateness', which implies 'consistency', 'adequacy', and 'simplicity'. It is therefore up to the critic to find better solutions here and there (and I am sure that they sometimes may be found, as I have pointed out repeatedly myself). Such solutions should, of course, be consistent with the present theory, and score on points of 'adequacy' and/or 'simplicity'. Of course, if 'inconsistency' can be pointed out the criticism is inherently valid. But even in that case, one should first see whether it is a minor case of changing a single adjective, or some other word, without affecting the theory or description as a whole, rather than a case of wholesale rejection. I have had myself to make minor revisions in the postulates during the research connected with writing this book. But I have here again to issue a warning against 'tinkering'. Alternatively, the critic can, of course, create a theory himself, a theory which constitutes an advance over the present one, or which has a different scope, but where it overlaps with the present theory it is as consistent and adequate. Though armchair-criticism and (by necessity superficial) reviewing has an important function in the development of our discipline, it cannot — unless it explores weaknesses in the epistomological foundations — be regarded as substantial criticism of either theories or descriptions. Only the two alternatives mentioned can, and it is those that I most welcome.

Chapter VIII: Postulates for axiomatic linguistics.

Revised version The postulates for Axiomatic Functionalism, as they were published in Mulder, 1980, and circulated before that in mimeographed form from 1974 onward, are at the moment of publishing this about sixteen years old. In the course of these sixteen years they have been subjected to discussion, been the basis for teaching the essential features of the theory, and — most importantly - used in the solution of descriptive problems over a very wide range. That is, their consistency, adequacy, and simplicity have been severely tested over that period, and — though there have been no indications as to a major lack of consistency — 'adequacy' and 'simplicity' have turned out to be in need of improvement. Also a few terminological inconsistencies had to be ironed out. In the first place, some of the defined notions were hardly ever used in practice, whereas other notions, sometimes rather informally and implicitly (i.e. theorematically) defined in the so-called 'explanations', have proven to be important enough to be included in the formal (i.e. specifically numbered) definitions. Secondly, as I have now written the present work, many of the explanations, which were included because they did not appear elsewhere, can now be omitted, or considerably shortened, or changed. Furthermore, as, when producing the 1974 set, my attention was focussed on syntax, and therefore some of the definitions were in the first place geared to this, it was noted that certain definitions could be given a wider scope — and so made more powerful — by making small changes, e.g. by replacing such terms as 'chain' (a syntactic term) by 'construction' (a general term), and by other minor

4 36

Postulates for axiomatic linguistics

changes of that kind. Also the order of presentation has been changed here and there, and Axiom D, i.e. 'all semiotic systems contain sentences', reads now 'All semiotic systems contain sentences, constituted by a base and para-syntactic features'. That is, the necessity for sentences to have the latter is now also duly axiomatized. In order not to disrupt the relation with earlier writings, the changes are kept to a minimum. To add further clarity, however, I shall insert some extra comments and also provide a separate alphabetical index to the definitions, in order to facilitate the reading. Here then follows the new set of postulates: Axiom A. All features in semiotic sets are functional. Def. la.

'Functional' for 'separately relevant to the purport of the whole of which it is a part'. Def. lb. 'System' for 'self-contained set of features with a common purport'. Def. l b l . 'Self-contained' for 'representing all relative dependencies of its members (or constituents), as members (or constituents) of the set (or combination), in question'. The notions 'functional' and 'self-contained' can, of course, be applied to 'combinations (of items)' as well as to 'sets'. In the case of 'combinations', the term 'members' has to be replaced by 'constituents'. Def. lc. 'Semiotic system' for 'system of conventions for communication'. Alternative definition to Def. 5. Axiom A states the point of view of the theory, i.e. 'the functional principle'. The above definitions give an interpretation to the axiom, and provide an ontological definition of 'semiotic system' (see Chapter III, 1, 2). The reason why the term 'semiotic system' is not already used in the axiom is that otherwise the recognition of 'functionality' for features in sub-systems of semiotic systems (e.g. 'phonology', 'phonotactics', 'grammar', 'syntax', etc. in a natural language) would be precluded. Def. l c l . 'Features' for 'elements, analytical properties of elements or relations between elements or analytical properties of elements'. Def. Ic2. 'Entity' for element, or 'discrete analytical property of element'.

Revised version

437

Def. Ic3. 'Semiotic entity' for 'entity in semiotic system'. Def. Icl-lc3 provide instructions for the consistent usage of the terms 'feature', 'entity', and 'semiotic entity'. Such terms as 'element', and also 'item' remain undefined, i.e. they are to be regarded as 'primitive terms', to be used in their 'ordinary language' sense. Def. Id. 'Communication' for 'subjective (i.e. involving choice or optionality) conveyance of information'. This rules out 'labels', 'names', or 'designations', not to be confused with 'communicating' these or about these, from being 'communication' in our sense. Axiom B. Semiotic systems contain simple, and may contain complex ordered, and/or complex unordered signa and figurae. Def. 2.

'Index' for 'item or class of items with informationvalue'. Def. 2a. 'Information-value' for 'specific set of potential interpretations'. Def. 2a 1. 'Signum' for 'Sign or Symbol'. Alternative definition: 'semiotic entity with both form and information-value'. Def. 2a2. 'Sign' for 'signum with wholly fixed conventional information-value' Alternative definition: 'index possessing the property of denotation'. Def.2a2a. 'Denotation' for 'wholly fixed conventional informationvalue of index in semiotic system'. Def. 2a3. 'Symbol' for 'signum with not wholly fixed conventional information-value, i.e. to which a temporary information-value can be attached by a definition'. Def. 2a3a. 'Proper symbol' for 'symbol with partially fixed conventional information-value'. Def. 2a3b.'Nonce symbol' for 'symbol with wholly non-fixed conventional information-value, i.e. with no fixed information-value at all'. Axiom Β is the most powerful axiom of the whole theory. It harbours (after being given an interpretation by means of definitions that follow it) the theory of semiotic systems (which is one of the sub-theories), as well as almost the whole of the systemology of any semiotic system (i.e. for natural languages both phonolo-

438

Postulates for axiomatic linguistics

gy and grammar), with the exception of the paratactic sub-systems. The latter are covered by Axioms C and D (see Chapter III, 3, 4, and Chapters V and VI) The systemology is unfolded in definitions 2-16c, together with the basic methodology for both plerological (grammatical) and cenological (for spoken language: phonological) descriptions. Definitions 2-2a3b and 2b develop that part of the theory of indices (again a sub-theory, see Chapter VI, 1) which is relevant to semiotic systems. Def.2a4. 'Plerological entity' or 'Grammatical entity' for 'entity in systemology corresponding to a signum'. Def. 2a4a. 'Plerematics' or 'morphology' for 'complex unordered plerological system'. Def. 2a4b.'Plerotactics' or 'syntax' for 'complex ordered plerological system'. Def.2a4c. 'Plerological (grammatical) system' for 'system of plerological entities'. Also simply called 'plerology'. Def. 2b. 'Figura' for 'semiotic entity which has only form'. Def. 2b 1. 'Cenological entity' for 'entity in systemology, corresponding to a figura'. Def.2bla. 'Cenematics' for 'complex unordered cenological system'. Def.2blb. 'Cenotactics' for 'complex ordered cenological system'. Def. 2b 1 c.'Cenological system' for 'system of cenological entities'. Also simply called 'cenology'. Def. 2bl d.'Cenological form' for 'notion in Signum-theory, corresponding to feature potentially belonging to cenological system'. Def. 2b2. 'Systemology' for 'cenological system and plerological (grammatical) system' or for 'the sub-theory dealing with the description of the systemology (in the above sense) of semiotic systems' (the same goes for cenematics, cenotactics, etc., i.e. for the sub-systems of systemology). Def. 3a. 'Phonology, Phonematics, Phonotactics, Phonological entity, Phonological system, Phonological form, etc.' for 'Cenology, Cenematics, Cenotactics, etc. in natural language'. For other semiotic systems often convenient terms can be coined, e.g. 'graphology', etc., for 'writingconventions', etc.

Revised version

439

The terms 'plerological' and 'grammatical' are synonymous, and 'plerematics' and 'plerotactics' are synonymous with 'morphology' and 'grammar' respectively. The terms with 'cene-' and 'ceno-' apply to any semiotic system, the terms with 'phone-' and 'phono-' only to natural (spoken) languages. Def. 3b. 'Articulation' for 'cenotactics or plerotactics'. Def. 3c. 'Double articulation' for 'cenotactics and plerotactics'. Def. 3c 1. 'Language' for'semiotic system with double articulation'. Def. 3c2. 'Proper language' for 'semiotic system with a cenology containing both a cenematics and a cenotactics, and a plerology (grammar) containing both a plerematics (morphology) and a plerotactics (syntax)'. All natural languages known to date are proper languages, but not necessarily vice versa. Natural languages, in addition, incorporate para-tactic systems, but so do all other semiotic systems, though for the latter this is often trivial. This is, because the same realizations may correspond to para-cenotactic entities from the one point of view, but to para-plerotactic entities from another. Also, especially in simple systems, tactic realizations may coincide with para-tactic ones, in which 'tactic' and 'para-tactic' are mere different aspects of the same thing. One needs, however, to recognize both aspects, as they account for different realizational features. That is, the tactic notions, e.g. 'ceneme', 'plereme', etc., as models, are not set up in a way which allows them to account for such contrastive (but nevertheless 'functional' from an overall point of view) features as 'pause', 'juncture', etc. In natural language, and many 'language-connected' semiotic systems, however, all this is far from trivial. Def. 3c2a.'Interlocking' for 'in systemology the one sub-system providing the forms of the entities of the other subsystem' (a cenological system and plerological system interlock in this way), or for 'the one sub-system providing the basic entities of the other sub-system' (an inventory of simple cenological entities and a cenematics, a cenematics and a cenotactics, as well as an inventory of simple plerological entities and a plere-

440

Postulates for axiomatic linguistics

Def. 4a. Def. 4b. Def. 4b 1. Def. 4b2. Def. 5.

Def. 6a. Def. 6b.

Def. 6c.

matics, and a plerematics and a plerotactics interlock in this way). There is a kind of interlock between 'cenotactics' and 'para-cenotactics', and between 'plerotactics' (syntax) and 'para-plerotactics' (para-syntax, or the sentential level), but this is of a different kind. Cenotactics and plerotactics provide the entities that correspond to elements in the base of para-cenotactic and para-plerotactic entities respectively 'Simple system' for 'system without combinations of entities'. 'Complex system' for 'system with combinations of entities'. 'Unordered system' for 'complex system without ordering relations between entities' 'Ordered system' for 'complex system with ordering relations between entities'. 'Semiotic system' for 'communication system with a systemology, i.e. with a cenological system interlocking with a plerological system' (alternative definition to Def. lc). We can, therefore, have semiotic systems where either the cenological system, or the plerological system, or both, are simple, unordered, and/or ordered, which leads to various types. Definitions 3b-5 are specially geared to clarifiying the notion 'proper language'. All natural languages are proper languages, but there is at least a theoretical possibility that the reverse does not hold. Ordering relations' for 'asymmetrical relations between entities in combinations (constructions)'. 'Relations of simultaneity' for 'symmetrical relations between entities in combinations (constructions)'. By Axiom A, only functional criteria may be brought to bear in deciding whether a relation is symmetrical or asymmetrical. 'Construction' or 'Combination' or 'complex' for 'selfcontained complex of entities in cenological or plerological system'. The terms imply such notions as: 'complex ceneme', 'cenotagm', 'complex para-cenotactic entity', 'complex plereme', 'syntagm', or 'complex sentence' (i.e. consisting of more than one clause as its

Revised version

441

immediate constituents), and 'self-contained bundle of immediate constituents'. Def. 7a. 'Paradigmatic' for 'the oppositional aspect of semiotic entities'. Def. 7a 1. 'Paradigmatic relations' for 'relations of opposition between members of sets'. Def. 7a2. 'Commutation' for 'alternation (or: choice) between semiotic entities (or 'zero' and semiotic entities) in functional opposition as immediate constituents, in a given context'. Def. 7a3. 'Distinctive function' for 'the set of oppositions in which an entity may partake'. In symbols: a ~ (fcUcUcO, which states the distinctive function of α, in case the set of oppositions a enters is (a ~ bt a ~ c, a ~ d), and no other. In fact, a ~ ( b u c u d ) = a ~ b u a ~ c u a ~ d . Def. 7a4. 'Neutralization' for 'suspension of opposition between members of a correlation in given contexts, and governed by those contexts'. Def. 7 a4a.'Correlation' for 'set of tactic entities which have an immediate constituent in common'. Def.7a4b.'Archiceneme (in natural language: archiphoneme)' for 'cenotactic entity resulting from neutralization'. In principle there could also be 'archipleremes'. Alternative definition: 'Simultaneous bundle of distinctive features in particular contexts, common to two or more cenemes in other contexts, i.e. equalling the intersection of those cenemes'. Def. 7b. 'Syntagmatic' for 'the ordering aspect of semiotic entities'. Def. 7b 1. 'Syntagmatic relations' for 'ordering relations between semiotic entities in combinations (constructions)'. Though the term 'permutation' may seem the syntagmatic equivalent of 'commutation', it is used in a realizational, rather than structural sense, though there may be structural implications. I use it as a primitive term, i.e. I refrain from defining it. Def. 7b2. 'Syntagmatic entity' for 'tactic entity'. This implies that it is orderable, and/or has constituents that commute with orderable entities, provided it is not intrinsically

442

Postulates for axiomatic linguistics

uncombinable at the tactic level (e.g. such as 'yes' in English), except in conjunctive constructions (e.g. 'yes or no'). It will be clear, therefore, that 'distinctive features' and 'monemes' are not syntagmatic entities, but 'phonemes' and 'pleremes', and anything more complex (but below the para-tactic levels), are. Def. 7c. 'Tactic' for 'cenotactic' or 'plerotactic (syntactic)'. Def. 7c 1. 'Cenotactic entity' for 'syntagmatic entity in cenological systems'. Def. 7c2. 'Tactic relations' for 'cenotactic relations or plerotactic relations'. Alternative definition: 'constructional relations (whether ordering or not) between syntagmatic entities, as immediate constituents in combinations (constructions)'. Note that tactic relations are not necessarily syntagmatic relations, though the reverse holds. 'Tactic relations' are relations between syntagmatic entities. Def. 7d. 'Plerotactic (syntactic) entity' for 'tactic relations in plerological (grammatical) system'. Def. 7e. 'Plerotactic (syntactic) relations' for 'tactic relations in plerological (grammatical) system'. Def. 7el. 'Cenotactic relations' for 'tactic relations in cenological system'. Def. 7f. 'Constructional relations' for 'relations between immediate constituents'. Definitions 6a-7f lay the foundation for further terminological developments necessary for an effective description of semiotic systems. Def. 7f 1. 'Constituents' for 'entities (of the same kind, i.e. of the same level of abstraction) in self-contained combinations'. Def. 7fl a.'Immediate constituents' for 'constituents that are not constituents of constituents within the combination in question'. Def. 7flb. 'Ultimate constituents' for 'constituents that are minimum (i.e. simple) entities at the level in question'. Ultimate constituents may be immediate constituents. It is theorematic that in cenematics and plerematics (morphology), in contradistinction with cenotactics and plerotactics (syntax), immediate constituents are always

Revised version

Def. 7g.

Def. 7h.

Def. 8a.

Def. 8al.

Def. 8a2. Def. 8a3.

Def. 8a4. Def. 8b.

443

at the same time ultimate constituents. 'Positions' for 'divisions within a tactic construction, such that in every such division an entity, as an immediate constituent of that construction, can stand and alternate, (i.e. commute) with other entities, or with φ\ Alternative definition: 'divisions within a construction corresponding to immediate constituents as relata in tactic relations'. 'Archi-position' for 'the intersection of two or more positions'. In phonotactics, intersection can only occur between adjacent positions. In syntax, intersection can occur between any two or more peripheral positions, but it can only occur in the case of parallel determination, not in the case of diverse determination. See Chapters V and VI. 'Ceneme (in language: phoneme)' for 'self-contained bundle of one or more distinctive features as its immediate, and at the same time ultimate, constituents'. Alternative definitions: 'Minimum syntagmatic entity in cenological system', 'minimum cenotactic entity'. For natural language, 'cene-' and 'ceno-' in definitions can be replaced by 'phone-' and 'phono-'. 'Cenematic complex' for 'complex ceneme'. A complex ceneme is a cenematic complex, as opposed to a cenotactic complex. A complex cenological entity is either cenematically or cenotactically complex. Or it may, of course, be para-cenotactically complex. 'Distinctive feature' for 'minimum cenological entity'. 'Hyper-feature' for 'distinctive feature in a particular phonematic context, equivalent to two or more distinctive features in at least one other phonematic context'. 'Hyperphoneme' for 'phoneme consisting of, or containing, one or more hyper-features'. See Chapter V. 'Plereme' for 'self-contained (by definition: simultaneous) bundle of one or more monemes as its immediate, and at the same time ultimate, constituents'. Alternative definitions: 'minimum syntagmatic entity in plerological (grammatical) system', 'minimum plerotactic (syntactic) entity'. From the point of view of the set of allomorphs,

444

Postulates for axiomatic linguistics

one can distinguish within 'plereme' between 'word' and 'grammateme'. If we set up the set of allomorphs in such a way that all allomorphs are continuous, i.e. uninterrupted in realization, they are properly called 'words', otherwise it is better to refer to them as 'grammatemes'. The distinction has, however, only importance from the point of view of realization. It has no structural importance. Def. 8b 1. 'Plerematic (morphological) complex' for 'complex plereme'. A complex plereme is a plerematic (morphological) complex as opposed to a plerotactic (syntactic) complex. A complex plerological entity is either plerematically (morphologically) complex or plerotactically (syntactically) complex. Or it may, of course, be complex at the sentential level, i.e. consist of more than one clause. Def. 8b2. 'Moneme' for 'minimum plerological (grammatical) entity'. Monemes are the grammatical analogues of 'distinctive features'. Def. 9a. 'Distributional unit' or 'cenotagm (in natural language: phonotagm)' for 'self-contained bundle of positions in cenotactics', or for 'instance of a self-contained bundle of positions in cenotactics'. Alternative definition for 'distributional unit' in the former, i.e. abstract, sense: 'minimum type of structure within which the distribution of cenotactic (natural language: phonotactic) entities can be described completely and exhaustively'. This is to say that nothing outside such a structure can determine the distribution of immediate constituent entities within the structure. It is possible that we may in some cases have to describe further the distribution of types of distributional unit in respect of one another. In fact, I tend to use the term 'distributional unit' in the more abstract sense, and the term 'cenotagm' (phonotagm) for an instance of a distributional unit. Def. 9a 1. 'Distribution' for 'the set of occurrences of an entity in constructional relations with other entities as immediate constituents in the same construction'. Def. 9b. 'Syntagm (plerotagm)' for 'self-contained bundle of positions in plerological (grammatical) system' or for

Revised version

Def. 9 b l . Def. 10. Def. 10a.

Def. 10b.

Def. 11a.

Def. l i b .

445

'instance of a self-contained bundle of positions in plerological (grammatical) system'. Alternative definition for 'syntagm' in the first sense: 'underlying structure of a self-contained bundle of plerotactic (syntactic) entities', and in the second sense: 'self-contained bundle of plerotactic (syntactic) entities, as immediate constituents', or 'plerotactic complex'. 'Syntactic (plerotactic) entities', for 'pleremes or syntagms'. 'Syntagmeme' for 'paradigmeme in a particular position, and in its capacity of standing in that particular position'. 'Paradigmeme' for 'member of a self-contained set of entities in functional opposition in a given context'. Alternatively: 'member of a paradigm'. 'Paradigm' for 'set of entities in functional opposition in a given context'. Definitions 7f-10b refer to general notions in the description of semiotic systems (see Chapter V and Chapter VI, 1-7), though definitions 8a, 8al, 8a2, and 9e are more relevant to cenology (in natural language: phonology), whereas definitions 8b-8b2 are more relevant to plerology (grammar). 'Relation of sub-ordination (or: determination)' for 'direct tactic asymmetrical relation of functional dependency'. Its converse is super-ordination (or: government). If a and b are in direct tactic relation, and a is for its tactic function dependent on b, but not vice versa (in symbols: a > b), a is said to be subordinate to b, and b super-ordinate to a. Furthermore, a is said to be standing in peripheral, and b in nuclear position in the construction in question. 'Super-ordination (or: government)' and the notions 'nuclear' and 'peripheral' are hereby defined as well. Alternative definition: 'direct tactic relation of unilateral functional dependency'. 'Relation of coordination' for 'direct tactic (by implication: symmetrical, and therefore simultaneous) relation of mutual functional independency'. If a and b are in direct tactic relation, and a is for its tactic function independent of b, and vice versa, a and b

446

Postulates for axiomatic linguistics

Def. 11 c.

Def. 12a.

Def. 12b.

Def. 12c.

Def. 13a.

are said to be coordinated (in symbols: a Alternative definition: 'direct tactic relation of bilateral functional independency'. 'Relation of interordination' for 'direct tactic (by implication: symmetrical, i.e. simultaneous) relation of mutual functional dependency'. If a and b are in direct tactic relation, and a is for its tactic function dependent on b, as well as vice versa, a and b are said to be inter-ordinated (in symbols: 'a < > b)\ Alternative definition: 'direct tactic relation of bilateral functional dependency'. Note that defs. 11 b and 11 c, unlike 11a, refer to non-syntagmatic relations between syntagmatic entities. 'Relation of unilateral occurrence dependency' for 'relation such that one of two entities in direct relation which are immediate constituents in a construction, can occur in the construction in question, whilst the other is φ\ In symbols: [a]b or a[b], the square brackets indicating the occurrence dependent entity, i.e. in the above case it requires the other item, or an item in the same position, for its occurrence, but not vice versa. The dependent item is called an 'expansion'. 'Bilateral occurrence independency' for 'relation such that either of two entities in direct relation which are immediate constituents in a construction can occur in the construction in question, whilst the other is φ\ In symbols [λ] [6]. Both items are 'expansions' in respect of one another. 'Bilateral occurrence interdependency' for 'relation such that neither of two entities in direct relation which are immediate constituents in a construction can occur in the construction in question, whilst the other is φ\ In symbols ab. Neither of the items are 'expansions' in respect of one another. 'Nucleus' or 'governing entity' for 'in a relation of subordination, the identity element for the tactic functions of the other elements', (cf. Def. 11a). In symbols: b > a, [6] > a, a < b, or a < [6], in which a is the nucleus. That is, the tactic relations of the other immediate constituents depend for their

Revised version

447

tactic function on their relation with the nucleus. Def. 13b. 'Peripheral entity' or 'governed entity' or 'determinant entity' for 'in a relation of subordination, a nonnuclear immediate constituent', i.e. in symbols: b, in the above representations. See Def. 11 a and 13a. Def. 13c. 'Expansion' for 'immediate constituent that commutes with φ\ In symbols [al > b, in which a is an expansion; also [a] < — / — > [b], i.e. coordination, in which both a and b are expansions. The term 'expansion' is always used with reference to another entity with which the other entity is in direct tactic relation. Complex expansions may contain entities that are themselves not expansions, e.g. [[a] > b] > c, or even [a > b] > c. Def. 14a. 'Diverse determination' or 'disjunctive determination' for 'self-contained complex of tactic relations, such that two or more peripheral immediate constituents are subordinated to a nucleus, but demonstrably in different ways'. I.e. aR x c and frR^c, where a and b are peripheral, c is nuclear, and R x and R y represent different tactic relations. In symbols: John

a > c Example: b

Mary

i.e. 'John likes Mary'. A both sufficient and necessary condition is that at least two of the peripheral entities in question belong to the same distribution-class. Because in cenotactics all relations must involve time and space (on which al our sensory perceptions depend, and without which there could not be 'form' at all) in a functional capacity, in cenotactics there is in the above case diverse determination by logical necessity. There are other logical reasons, but this is for the present purpose sufficient. See Chapters V and VI for detailed explanations. Def. 14b. 'Parallel determination' or 'disjunctive determination' for 'self-contained complex of tactic relations, such that no two immediate constituents can be demon-

448

Postulates for axiomatic linguistics

strated to determine the nucleus in different ways'. This implies that one has to assume parallel determination until this is refuted by the successful application o f Def. 14a. In symbols: a >

c

the

Example

b

> vase

blue

i.e. 'the blue vase'. Both sufficient and necessary condition: All peripheral entities, or their commutants, belong to different distribution classes. Def. 14c. 'Underlying syntactic structure' for 'abstract representation of a syntactic complex in terms of positions, with or without indication of occurrence dependency'. Def. 15. 'Direct relation' for 'relation between constituents (not necessarily immediate constituents) that is not a relation via other constituents'. Def. 16a. 'Conflation' for 'a single realization corresponding to more than one constituent (not immediate constituents) having a similar tactic function, in a construction'. Example: 'John likes but I hate him', where 'him' functions in two different, but equivalent, positions. Representation: 'John likes him but I hate him', or, representing the syntactic structures in question: John - - > likes) < — (but him

> hate))

< — ι, him

The unshaded box around 'him' indicates that there is conflation at that point. In the older ( 1 9 8 0 ) version o f the postulates this was called 'ellipsis'. Def. 16b. 'Functional amalgamation' for 'one single entity having different tactic functions (and therefore corresponding to more than one constituent) on different levels of the analysis'. In phonology an example is a single phoneme standing in a 'final' position in one phonotagm, but at the same time in an 'initial' position in another (adjacent) phonotagm. In syntax an example is: 'the

Revised version

449

man who bought that'. Representation: 'the man fiyliQj bought that'. In an analysis of English 'who bought that' is firstly a 'relative syntagm' with the structure 'who < bought that', and on a subsequent level a so-called 'transitive predicative based syntagm', i.e.

tyy/j

F^ZI —>

predicative

object

*m

->

bought

that

The shaded box indicates that there is functional amalgamation. Def. 16c. 'Antecedence' or 'post-cedence' for 'a case in which a syntagm from the structural (but not from the realizational) point of view is only partly well-formed, but the 'missing' entity is represented elsewhere'. E.g. 'the book he give me', in which 'he gave me' is represented he me

•

> gave

The 'antecedent' entity is here 'the book'. The empty box indicates that the construction is a semi-x (where * stands for the type of construction in question, in this case an 'indirect transitive predicative based syntagm'). Note that the non-well-formedness is merely a matter o f structural analysis, not o f actual deficiency. There may be cases in which the antecedent is merely 'implied'. Definitions 11a to 16c further develop the whole of syntax. All the revelant notions are presented, and the methodology is developed in fairly great detail. Axiom C. Cenological entities may have para-cenotactic features and plerological entities may have para-syntactic features. Def. 17. 'Para-tactic features' for 'para-cenotactic or para-

450

Postulates for axiomatic linguistics

syntactic features'. In natural language these are usually, but (from a functional point of view) inappropriately, lumped together under the term 'prosody'. This is because their phonetic substance is usually 'pitch' or 'stress', or a mixture of the two. But also 'pause' or 'juncture' plays a role in this respect. Another type of para-tactic feature, frequently encountered in natural language, is difference in sequential order, i.e. permutation of the tactic entities involved. E.g. 'can he do it' versus 'he can do it'. This should not be confused with realizational permutation as a means of expressing syntactic relations, e.g. 'John hit 'Mary' versus 'Mary hit John'. The latter have to do with the actual syntactic construction, and are therefore not para-tactic. The term para-tactic implies that the features involved are not inherent in the form of the tactic constituents and their arrangement, i.e. that they are not merely realizational on the tactic level. E.g. there is nothing in the phonemic constituence of the form of the word 'blackbird', nor in the fact that there are two phonotagms that can account for the fact that it represents a unit on a higher than tactic level with an accent (in neutral realization) on the first syllable, and nor is the fact that an entity represents, say, a clause, inherent in the conglomeration (which is not even 'constituency') of tactic entities that correspond to its base. Only in the most simple cases, as, for instance, the Morse code, could one claim that the pause before and after a cenotagm is purely realizational, by arbitrarily establishing nuclear and peripheral positions, but it would be extremely complicated to account for it in this way, rather than saying that 'cenotagm' and 'sentence' in the Morse code coincide. This also preserves the generality in that one would not have to say that either the notion 'sentence' or the notion 'cenotagm' in the Morse code are purely realizational. The latter would logically follow. This does not mean that I do not agree that for some semiotic systems recognizing para-tactic levels is trivial, but in principle they can be recognized for any semiotic system. Not

Revised version

451

all semiotic systems have, however, distinctive paratactic features (see below). Def. 18. 'Para-cenotactic features' for 'features corresponding to cenological form, accompanying, but not determining the identity of cenotactic entities'. Of course, a cenotactic entity in combination with such features assumes an identity of its own on the para-cenotactic level. In cases where this is trivial, they are only different entities from different points of view, just as, for instance, a plereme is a maximum entity from the morphological, but a minimum entity from the syntactic point of view. Def. 18a. 'Contrastive para-cenotactic features' for 'para-cenotactic features with the function of groupment over and above cenotactic groupment'. Def. 18b. 'Distinctive para-cenotactic features' for 'para-cenotactic features that are in a relation of direct opposition with one or more other cenotactic features, or with φ'. A typical example is distinctive 'tone', as, for instance, in Chinese. Trivially, unless there is no one-one correspondence (in which case it would not be trivial), also the phonological forms of distinctive intonations are distinctive para-cenotactic (para-phonotactic) features, whilst the intonations themselves are para-syntactic features. One should not be misled by the terminology in thinking that 'contrastive para-tactic' features are not functional. They are, as so many other things, e.g. syntactic relations, distinctive in a systemic, not in a directly oppositional, sense. 'Distinctive para-tactic features' are, however, distinctive in the latter sense. Def. 19. 'Para-syntactic features' or 'para-plerotactic features' for 'features accompanying, but not determining the identity of, syntactic entities or conglomerations of syntactic entities'. Syntactic enties or conglomerations of syntactic entities in combination with para-syntactic features assume an identity of their own on the sentential level, i.e. they become sentential entities (i.e. 'sentences' or 'clauses', see below, Def. 21,21a). Def. 19a. 'Contrastive para-syntactic features' for 'features with the function of groupment over and above syntactic

45 2

Postula tes for axioma tic linguistics

Def. 19b.

Def. 20.

Def. 20a.

Def. 20b.

groupment'. In language the suspensive, i.e. 'comma' or 'semi-colon' intonation type belong to this, but also the 'distinctive types such as 'finality' ('full stop type'), 'finality with emphasis' ('exclamation-type'), and 'question', though having a directly distinctive function, correspond, from another point of view, to contrastive features, as they too provide groupment. 'Distinctive para-syntactic features' for 'para-syntactic features that are in a direct relation of opposition with other para-syntactic features or with φ\ It is possible that the 'finality' (full stop type) could be regarded as , but there seems little profit in doing this. 'Para-tactic entities' for 'tactic entities or conglomerations of tactic entities, together with accompanying para-tactic features, such that the whole assumes an identity on a level different from the tactic level'. Alternative definition: 'entity constituted by a base and para-tactic features'. 'Base' for 'in a para-tactic entity, the total complex of those features that correspond (on another level) to tactic entities'. 'Complex para-tactic entity' for 'self-contained entity constituted by two or more para-tactic entities, together with further accompanying para-tactic features'. Axiom Β and related definitions deal with the paracenotactic (in natural language: para-phonotactic) features and with the para-plerotactic (para-syntactic) features. To the former belong, in natural languages, such features as 'tone' (in so-called tone-languages), which is a distinctive para-phonotactic feature, and accent and juncture, which are contrastive paraphonotactic features (see Chapter III, 3, 4; see also Chapter V, 20). Para-plerotactic features, such as intonation in natural languages, are sentential features, i.e. they pertain to sentences and clauses (which are the sole two types of sentential entity), albeit that not all sentential features are para-plerotactic features. But, as the next axiom (AxiomD) states, para-plerotactic features are obligatory as far as sentences are concerned (see Chapter VI, 11). Axioms B, C, and D together

Revised version

453

cover the whole of cenology (in natural language: phonology) and plerology (grammar), i.e. the whole of the systemology (see Chapters V and VI). One has to be careful to distinguish from the paratactic (i.e. para-cenotactic and para-plerotactic) features such realizational features as I have called 'connotative stress', which is physically similar to realizations of 'accent', and 'connotative modulation', which is similar to realizations of 'intonation'. These affect 'connotation', not 'denotation', and could perhaps profitably be dealt with under the heading of 'pragmatics'. Without extending the theory by adding one or more axioms, together with definitions, this is outside our scope. It is doubtful whether the theory itself could be so extended without coming into conflict with the functional principle and its interpretations, but it would be worthwhile investigating this. For the time being, however, if we wish to indulge in matters of 'pragmatics' we should use or set up for it a separate and independent theory. Because of physical similarities and physical simultaneity, even between the actual para-tactic features themselves, it is for this area of research clearer than ever that one cannot base one's description on matters of realization (the inductive approach). One might as well measure the weight or the commercial value of a load of books, and, solely on the basis of this, try to arrive at the weight or the price of every single book in the load. It is clear that only the inverse method may lead to exact results. Axiom D. All semiotic systems contain sentences, constituted by a base and para-syntactic features. Def. 21. 'Sentence' for 'plerological entity (by definition corresponding to a signum) with such features that it cannot be a feature (constituent, or other feature) of another plerological entity? Alternative definition: 'signum such that is a self-contained vehicle for conveying messages'. Def. 21a. 'Clause' for 'immediate constituent (perhaps the only one) of a sentence'. A clause is therefore, by implication,

454

Postulates for axiomatic linguistics

also a para-syntactic entity, and as all para-syntactic entities are sentential entities, though not vice versa, a clause is a sentential entity. Def. 21b. 'Sentential features or entities' for 'such features or entities as determine particular signa to be sentential entities'. These are not necessarily para-syntactic features, though the latter are necessarily sentential features. Def. 21c. 'Sentential markers' for 'occurrence dependent sentential features or entities that are not para-syntactic features or entities, but correspond to syntactic features or entities'. That is, such features or entities determine, just as para-syntactic features do, particular entities to be sentential entities. But there is no sentential entity without para-syntactic features. It is therefore the latter, rather than sentential markers, which are not compulsory, that are the actual features that make something to be sentential. I refrain from formally defining other types of sentential entity, as more research has to be done, and the adequacy of the ideas I have at present has to be tested over a wide range of languages. Note that neither the notions 'sentential feature' and 'para-syntactic feature', nor the notions 'sentential entity' and 'para-syntactic entity' are equivalent, but the notions 'sentential level' and 'para-syntactic level' are equivalent by mutual implication. As I have said, this concludes the systemology (see Chapter VI, 11). It sets the sentential level apart from the rest of grammar, especially syntax. This separation is one of the more conspicuous features of axiomatic functionalism. It effectively removes the dilemma of well-formedness versus non-well-formedness in syntax and many another dilemma which tends to plague linguists of other persuasions. Axiom E. There may be a many-to-one relation between cenetic form and figura (allophony), and between cenological form and signum (allomorphy), and vice versa Qiomophony and homomorphy respectively). Def. 22. 'Image' for 'model for the unique form of a singular

Revised version

455

realization of a cenetic feature' (in natural language: speech-sound). Symbolized: i. Def. 22a. 'Cenetic form (in natural language: phonetic form)' for 'class of impressionistically similar images'. Symbolized :f Formal definition 'f = {i}', the braces indicating that it is a self-contained class (of i). Def. 23. 'Cenological (phonological) form' (symbolized: p) for 'a particular self-contained class of one or more particular cenetic (phonetic) forms (i.e. ( f j ) , each member f in its capacity of having a particular distinctive function d \ Formal definition: p1 = where the superscripts i stands for any integer, and can be read as 'a particular' or 'a specific', and i . . . η indicates 'each one of a particular class'. As much of the following is in the first place relevant to natural language, I shall use further the appropriate terminology. It will be easy enough to translate this into more general semiotic terminology, or to coin new terms as required. Def. 23a. 'Allophony' for 'one phonological form, the realization of which corresponding to two or more phonetic forms'. Formal definition: 'f l Rd l ~ PRd 1 ', where also j stands for an integer, signifying 'a particular', but i Φ j, and ~ can be read as 'compared with'. Def. 23a 1 'Allophone' for 'one of the terms of 'allophony' as a comparison'. If there is no allophony a term of that kind is properly called a 'phone', i.e. '^Rd 1 ' is a 'phone', but often for this also the term 'allophone' is used. Def. 24. 'Signum (symbolized S)' for 'the conjunction of a particular expression and a particular content which mutually imply one another'. Formal definitions: Έ & C' or { p*""' "Rd*} & {diRpi · · 1 1 } . As Ε implies C and vice versa, and either implies S and vice versa, any two of those are equivalent by mutual implication. It is therefore, in practice, admissible to use the formula for 'expression', i.e. [ p 1 · • "Rd 1 j , instead of the more lengthy formula, for 'signum' itself. The same holds for when we are dealing with the notions 'allomorph' etc. Def. 24a. 'Expression (symbolized E)' for 'a particular selfcontained class of one or more particular phonological forms (i.e. ( p ) ) each member in its capacity of having

45 6

Postulates for axiomatic linguistics

a particular distinctive function d \ Formal definition E1 = [p 1 • · · "Rd 1 j . As ρ has already distinctive function incorporated, i.e. p1 = {f 1 • · · "Rd 1 } , the distinctive function d in 'Expression' (or in 'Content', or in 'Signum') is by implication distinctive function at a further level, i.e. on the signum-level. Def. 24b. 'Content (symbolized C)' for 'a class of one particular distinctive function d being and in its capacity of being the particular distinctive function of each member of a particular self-contained class of phonological forms'. Alternative definition: 'the converse of expression'. Formal definition: ' {d'Rp 1 · · · n } \ Def. 24b 1 'Allomorphy' for 'one signum, the realization of which corresponding to two or more phonological forms'. Formal definition: 'p*Rdl ~ p J Rd 1 , i Φ j \ Def.24bla 'Allomorph' for 'one of the terms of 'Allomorphy' as a comparison'. If there is no allomorphy, a term of that kind is properly called a morph, i.e. ' p ^ d 1 ' is a 'morph', but often for this also the term 'allomorph' is used instead. Def. 25.

Def. 26.

Def. 27.

Def. 28.

Def. 29.

'Homophone' for 'allophone of one figura, the realization of which corresponding to that of an allophone of another figura'. Formal definition of 'homophony': ' f R d 1 ~ f*Rdj, i Φ j \ 'Homomorph' for 'allomorph of one signum, the realization of which corresponding to that of an allomorph of another signum'. Formal definition of'homomorphy': 'piRd1- p'Rdj,i Φ j\ 'Homonym' for 'total class of allomorphs of one signum, in comparison with, and the realizations of its members corresponding to, those of the total class of another signum'. Formal definition of homonymy: { p l · · · "Rd 1 } ~ { p i - - - n R d j } , ΐΦ j'. 'Synonym' for 'signum, in comparison with and having the same intrinsic information-value (denotation) as another signum'. Formal definition of synonymy: {p(i...n)iRdi JRDi ^ {p(i - · • n >JRd j } RD\ i Φ j, and where D stands for 'denotation'. 'Utterance' for 'model for the unique form of a singular

Revised version

457

realization of a signum'. Formal definition: '(iRd^Rd 1 '. Axiom Ε and the ensuing definitions deal with the general ontology or 'signum-theory', which is extensively dealt with in Chapter IV, 2. The next axiom, i.e. Hervey's Axiom F, together with definitions (see Hervey, 'Postulates for axiomatic semantics', in Mulder and Hervey 1980) provides the basis for semantics (see Chapter IV, 1). Axioms A-F together cover the whole of the theory. Axiom A states, as I have said, the all pervading point of view. Axioms B, C, and D, deal with the systemology, Axiom Ε with the signumtheory, and Axiom F with semantics. In order to complete the set, I am giving here Hervey's (in Mulder — Hervey 1980:203) axiom for semantics: Axiom F. Signa may be realized an unlimited number of times (in actual communication), each resulting utterance denoting a denotatum which may belong to a potentially infinite denotation class. The postulates are the theory in a nutshell, as it were. Though they may not contain everything in every detail, most of the other things that can be regarded as belonging to the theory are theorematic, or logically implied. One could regard the postulates as the formally codified nucleus of the theory. There may be some freedom outside those, as long as this freedom is used in consistency with the postulates, and the rules of further consistency, adequacy, and simplicity are observed. For the complete set of postulates for semantics I refer to Hervey, in Mulder and Hervey 1980.

Bibliography

Batög, Τ. 1967

The axiomatic method in phonology (London: Routledge & Kegan Paul). Boch6nsky, I.M. 1959 Die zeitgenössischen Denkmethoden (Bern: Francke Verlag). Chomsky, Noam 1957 Syntactic Structures (The Hague: Mouton). 1965 Aspects of the theory of syntax (Cambridge, Mass.: The M.I.T. Press). Chomsky, Noam — Stuart Hampshire 1968 "Noam Chomsky and Stuart Hampshire discuss the study of language", The Listener 30.5.1968:687-691. Cook, Walter Α., S.J. 1969 Introduction to tagmemic analysis (New York: Holt, Rinehart & Winston). Gardiner, Sir Alan 1951 The theory of speech and language (Oxford: Clarendon Press). Gardner, Sheena 1985 Parasyntax and the sentential level in axiomatic functionalism (Ph.D. thesis, University of St. Andrews). Graham, D.C. 1954 Songs and stories of the Ch 'uan Miao (Smithsonian miscellaneous collections, vol. 123, no. 1, Washington D.C.) Hajd-Mohammed, Suleiman 1976 The phonology of Kamali Arabic (M. Litt, thesis, University of St. Andrews). 1980 Predicative based syntagms in Kamali Arabic compared with similar patterns in English (Ph.D. thesis, University of St. Andrews). Halliday, Michael A.K. 1961 "Categories of the theory of grammar", Word XVII:241-292. Harris, Zellig S. 1951 Methods in Structural Linguistics (Chicago: University of Chicago Press).

460

Bibliography

Hervey, Sändor G.J. 1970 Functional semantics (D.Phil, thesis, University of Oxford). 1975 "Semantics in axiomatic functionalist linguistics" Proceedings of the deuxieme colloque international de linguistique fonctionelle (Clermont-Ferrand), 145-152. 1975 "Grammar and semantics in axiomatic functionalist linguistics", Lingua 36:47-67. Reprinted in Mulder - Hervey 1980:188-202. 1979 Axiomatic semantics - A theory of linguistic semantics (Edinburgh: Scottish Academic Press). 1980 "Postulates for axiomatic functionalist semantics", Mulder Hervey 1980:201-211. "Denotation and non-denotation". Mulder - Hervey 1980:212-236. 1982 Semiotic perspectives (London: George Allen & Unwin). Hervey, Sandor - Jan W.F. Mulder 1973 "Pseudo-composites and pseudo-words — sufficient and necessary criteria for morphological analysis", La linguistique 9:4770. Reprinted in Mulder - Hervey 1980:122-144. Hjelmslev, Louis 1953 Prolegomena to a theory of language (translated from the Danish by F.J. Whitfield) (Madison: University of Wisconsin Press). Hockett, Charles F. 1958 A course in modern linguistics (New York: Crowell Collier & Macmillan Inc.). Jakobson, Roman — Morris Halle 1956 Fundamentals of language (The Hague: Mouton). Kemöny, J.G. 1959 A philosopher looks at science (Princeton: D. Van Nostrund Co. Inc.). Lamb, Sydney M. 1962 Outline of stratificational grammar (Berkeley: University of California Press). Lockwood, David G. 1972 Introduction to stratificational linguistics (New York: Harcourt Brace Jovanovich). Martinet, Andri 1949 Phonology as functional phonetics (London: University of Oxford Press). 1960 Elements de linguistique generale (Paris: Armand Colin). 1961 A functional view of language (Oxford: Clarendon Press). 1965 La linguistique synchronique (Paris: Presses Universitäres de France). Mulder Jan W.F. 1965 "Some operations with sets in language", Foundations of language 1:14-29.

Bibliography

1966

461

Sets and relations in phonology - A theory of linguistic description with special reference to Pekingese (D.Phil, thesis, University of Oxford). 1968 Sets and relations in phonology - An axiomatic approach to the description of speech (Oxford: Clarendon Press). 1969 "On the art of definition, the double articulation of language, and some of the consequences", Forum for modern language studies 5:103-117. 1971 "Linguistic sign, word, and grammateme", La linguistique 7: 93-101. 1973 "From sound to denotation" Folia Linguistica VI 1 /2:64-72. 1974 "Descriptive adequacy in phonology, and the vowel-phonemes of the Scottish dialects of Angus and Perthshire compared with the Southern English system", La linguistique 10:71-91. Reprinted in Mulder - Hervey 1980:88-103. 1975 "Syntax in axiomatic functionalist linguistics", Proceedings of the second international colloquium of functional linguistics (Clermont — Ferrand), 109-115. Reprinted in Mulder — Hervey 1980:145-151. 1978 "Phoneme tables and the functional principle", La linguistique 14:3-27. Reprinted in Mulder - Hervey 1980:104-121. "Some difficult cases in syntactic description", Mulder — Hervey 1980:161-175. 1987 "Effective methodology and effective phonological description", La linguistique 23:1942. 1988 "The simplicity of languages", Papers on language and mediaeval studies presented to Alfred Schöpf (Frankfurt, Peter Lang): 159-176; Mulder, Jan. W.F. — Sändor Hervey 1971 "Index and signum", Semiotica IV/4:324 - 338. Reprinted in Mulder - Hervey 1980:176-187. 1972 Theory of the linguistic sign (The Hague: Mouton). 1975 "Language as a system of systems", La linguistique 11:3-22. Reprinted in Mulder - Hervey 1980:73-87. 1980 The strategy of linguistics (Edinburgh: Scottish Academic Press). Mulder, Jan W.F. - H.A. Hurren 1968 "The English vowel-phonemes from a functional point of view, and a statement of their distribution", La linguistique 1:43-60. Munla, Muhammad S. 1981 Aspects of a functional description of English morphology (Ph.D. thesis, University of St. Andrews). Ob ana, Yasuko 1983 Syntactic properties of "wa" and "ga" in Japanese (M.Litt. thesis, University of St. Andrews).

462

Bibliography

Pike, Kenneth L. 1956 Phonemics - a technique for reducing language to writing (Ann Arbor: University of Michigan Press). Pilch, Herbert 1964 Phonemtheorie (Basel: S. Karger). Popper, Karl R. 1965 Conjectures and re filiations (2nd ed. London: Routledge & Kegan Paul). 1965 The logic of scientific discovery (4th ed. London: Hutchinson & Co.). Saussure, Ferdinand de 1915 Cours de linguistique generale (Paris: Payot). Suleiman, Muhammad Y.T.H. 1983 Modern trends in morphology - a critical study (Ph.D. thesis, University of St. Andrews). Trubetzkoy, N.S. 1939 Grundzüge der Phonologie (Travaux du cercle linguistique de Prague 7). Uhlenbeck, E.M. 1973 Critical comments on transformational generative grammar 19621972 (The Hague: Smits).

Index to the definitions

allomorph, 24bla allomorphy, 24b 1 allophone, 23a 1 allophony, 23a antecedence, 16c archiceneme, 7a4b archiphoneme, 7a4b articulation, 3b

construction, 6c constructional relations, 7f content, 24b contrastive para-cenotactic features, 18a contrastive para-syntactic features, 19a coordination, 1 lb correlation, 7a4a

base, 20a cenematic complex, 8al cenematics, 2bla ceneme, 8a cenetic form, 22a cenological entity, 2b 1 cenological form, 23 cenological form, 2bId cenological system, 2b lc cenology, 2blc cenotactic entity, 7c 1 cenotactic relations, 7el cenotactics, 2b lb cenotagm, 9a clause, 21a combination, 6c communication, Id commutation, 7a2 complex para-tactic entity, 20b complex system, 4b complex, 6c conflation, 16a conjunctive determination, 14b constituents, 7fl

denotation, 2a2a direct relation, 15 disjunctive determination, 14a distinctive features, 8a2 distinctive function, 7a3 distinctive para-cenotactic features, 18b distinctive para-syntactic features, 19b distribution, 9a 1 distributional unit, 9a diverse determination, 14a double articulation, 3c entity, lc2 expansion, 13c expression, 24a features, lcl figura, 2b functional amalgamation, 16b functional dependency, 11a, l i b , 11c functional, la

464

Index to the definitions

governed entity, 13b governing entity, 13a grammatical, see plerological homomorph, 26 homomorphy, 26 homonymy, 27 homophone, 25 homo phony, 25 image, 22 immediate constituents, 7fla index, 2 information-value, 2a interlocking, 3c2a interordination, 11c language, 3c 1 moneme, 8b2 morph, 24bla morphological complex, 8b 1 morphology, 2a4a neutralization, 7a4 nonce symbol, 2a3b nucleus, 13a occurrence dependency, 12a, 12b, 12c ordered system, 4b2 ordering relations, 6a para-cenotactic features, 18 para-phonotactic, see para-cenotactic para-plerotactic features, 19 para-syntactic features, 19 para-tactic entities, 20 para-tactic features, 17 paradigm, 10b paradigmatic relations, 7al paradigmatic, 7a

paradigmeme, 10a parallel determination, 14b peripheral entity, 13b permutation, 7b 1 phone, 23al phone-, 3a phonematics, 3 a phoneme, 8a phonetic form, 22a phono-, 3 a phonological form, 23 phonology, 3a phonotactics, 3a phonotagm, 9 a plerematic complex, 8b 1 plerematics, 2a4a plereme, 8b plerological entity, 2a4 plerological system, 2a4c plerology, 2a4c plerotactic entities, 9b 1 plerotactic entity, 7d plerotactic relations, 7e plerotactics, 2a4b plerotagm, 9b positions, 7g post-cedence, 16c proper language, 3c2 proper symbol, 2a3a self-contained, l b l semiotic entity, 1 c3 semiotic system, lc semiotic system, 5 sentence, 21 sentential entities, 21b sentential features, 21b sentential markers, 21c sign, 2a2 signum, 24 signum, 2a 1 simple system, 4a simultaneity, 6b

Index to the definitions

sub-ordination, 11a symbol, 2a3 synonym, 28 synonymy, 28 syntactic entities, 9b 1 syntactic entity, 7d syntactic relations, 7e syntagm, 9b syntagmatic entity, 7b2 syntagmatic relations, 7b 1 syntagmatic, 7b syntagmeme, 10

465

syntax, 2a4b system, lb systemology, 2b2 tactic entity, 7b2 tactic relations, 7c2 tactic, 7c ultimate constituents, 7 f l b underlying syntactic structure, 14c unordered system, 4b 1 utterance, 29

General index

a-syntactic relation 364, 374 abstract 140,142 accent 381 adequacy 57, 5 8 , 6 2 , 6 4 , 6 9 adequacy of detail 58 adequacy of range 58 adequacy of scope 58 adjectival 335 adjunct 416 adjunct positions 359 adjuncts 334,358 adverbial determination 346 adverbial relatives 359 agreement 308, 325 algebra 102, 137, 138 allomorph 120, 158, 159, 161, 163167,304 allomorphic level 119 allomorphic variation 279 allomorphon 167, 304 allomorphs 307 allomorphy 80, 91,162,164, 306 allophone 156, 158, 163,167, 304 allophonon 166,167,304 allophony 79,91, 162,164 ambiguity 299 Amoy Hokkien 272 analogy 200 analytical models 14 Anglophone 243 antecedence 356, 357 apposition 364 appropriateness 55, 64, 65, 68, 69 Arabic 264

arbitrariness 68 archi-position 299,339, 340 archi-positions 232,257 archi-tones 272 archiphoneme 235,236-238, 239 archiphonemes 178, 179 arithmetic 102,137 article 335 articulation 115 articulations 93 aspect 325, 328 auxiliaries 325, 328 axiomatic functionalism 66 axiomatic system 65,66, 67 axioms 39, 65, 67, 69, 180 Bangkok Thai 272 base 118,215, 371-373,377 basic distributional unit 268 bilateral functional dependency 291 bilateral functional independency 291 bilateral occurrence dependency 291 bilateral occurrence independency 291 black box philosophy 31, 33 borrowings 140 Cantonese 272 cenematics 105, 112 ceneme 119 cenetic form 154

468

General index

cenological inventory 112 cenological system 99 cenology 90 cenotactics 90,112 Chinese 223, 368 Chomsky, Noam 25, 33, 309 classes of phenomena 52 classificatory devices 107, 135 classificatory labels 108 clause 368,375 clause bases 374 clauses 92, 117 cleft sentences 364 co-referentiality 365 communicative economy 110 commutation 100, 192, 216, 284, 285 comparative constructions 340,341, 343-345 comparative syntagms 330, 340342,345 comparatives 332 comparison 163 complement 416 complementary intransitive PBS 315 complementary transitive PBS 314 complementation 163 complex nominals 345 complex ordered systems 97 complex systems 93, 94, 96, 116 complex unordered systems 96 complex verbal predicative 324 compound nominals 345 compound verbs 324 conceptualist 202 conceptualism 23, 206 concord 306-308, 326 conflation 344,357 congruence 308 congruity 306-308, 325,326, 345 conjunctional syntagms 330, 331

conjunctions 332 conjunctive 289 conjunctive constructions 331,332, 340,376-379,408,413 conjunctive determination 279 conjunctive syntagms 330, 331, 340,414 conjunctives 332, 377, 378, 414 connotation 149, 164 connotative modulation 381 connotative stress 381 consistency 55, 56, 58, 62, 64, 69 consonant tables 221 consonants 227, 228 constructional relations 95 constructional semantics 382 content 158,160, 161 contextual variants 72 contrastive para-tactic features 100 conventional indices 125, 136 Cook, W.A. 31,32 coordination 289,292 copula 320, 332 copulative (E)PS 324 copulative PBS 318 copulative predicatives 319, 323, 328 copulative syntagms 330 core-linguistics 4 CVC-type of notation 261 de Groot, W. 309 deductive system 65 definitions 39,65, 67, 180 denotables 84, 138,150 denotation 74, 84, 138, 148, 149,163-165 denotational approach 73 denotational import 102 denotational semantics 80 denotationally motivated 142,143 denoting 74,138, 149 derivations 198, 203

General index

description 37, 53, 169, 181 descriptive linguistics 5, 71 descriptive statements 4 0 , 4 1 descriptive supplement 335 determinants 330 determination 279 determining relatives 359 direct non-tactic relations 288 direct relation 281 direct relations 277, 282, 283, 386 direct tactic relations 287, 302 discontinuous 120 discourse 379 discourse analysis 377 disjunctive determination 279 distinctive features 113, 117, 214, 2 1 7 , 2 1 8 , 2 2 0 , 221,239 distinctive function 84, 217 distinctive para-tactic features 100 distribution 257 distributional unit 234, 235 distributional units 215, 216, 257 diverse determination 174, 192, 2 7 9 , 2 8 0 , 2 9 2 , 295, 296 Dutch 221, 222, 237, 241,243 eclecticism 57 economy 109, 110 El-Shakfeh, Fawzi 230 emphatic 380 English 219, 222, 226, 238, 241, 243,247,250, 301,326,330 entering tone 272 EPBS 332 EPS 323 equivalence 162 events 151 existential import 39 expansions 288 expression 158,160, 161 extended predicative based syntagm 332 extended predicative syntagm 325

469

extension of classes 48, 49 facts 151 feasibility 64 fields of phenomena 51, 52 figura 166 figurae 89, 105, 154 final 380 first articulation 115 form 154, 155 formal axiomatic system 66 formal logic 103 fossilization 196, 199 free accent 270 free variance 164 free variants 72 French 241 French functionalists 211 functional 83, 84 functional amalgamation 330, 336, 337, 344, 357, 358 functional dependency 279, 288, 291 functional principle 6, 55, 71 functional splitting 345 functional syntagms 329-331 functionalists 55, 214, 237 functional 283, 320, 330 Gabjanda, James 266 gemination 271 generative semanticists 24 generic 338 genitive 331 genitive "s' 373 genitive syntagms 330, 340 German 237, 238, 241-246, 262 glottal stop 260 government 279 grammar 71, 79, 9 0 , 9 1 grammateme 119, 120, 303 grammatical forms 93 grammatical system 115 grammatically simple 114

470

General index

graphemes 120 Hadj Mohamed, Suleiman 264 Hervey, S.G.J. 72, 382 Hielmslev, Louis 39, 119 Hockett, Charles F. 27 homomorph 163 homomorphy 162, 164 homonym 163 homonymy 150, 160, 162, 164, 285 homophone 163 homophony 162,164 hyper-features 222, 223 hyperphonemes 222,223 hypotheses 181-183 hypothetico-deductive method 36, 42, 180, 207 iconic 106, 107,140, 142 idealization 256 identity element 283 image 153-156, 158, 166, 167, 304 images 151, 165 immediate constitutent 102 immediate constituents 87, 283, 287 imperative 348, 351, 365, 366 impersonal "it" 321 indepenant syntagms 366 independent syntagms 371, 372 indeterminacy 299 indices 125 indirect relations 277, 281 indirect tactic relations 287 indirect transitive PBS 312 inductive method 180 inductivist 181,182 infinite 379 information-value 148 inherently generative 26 instrumentalism 180

instrumentalist 35 intension of classes 48,49 interlock 96 interordination 289, 290, 292, 413 interpreted axiomatic system 66 intonation 80, 92, 368, 369, 374, 380,381,382 intransitive PBS 311, 318, 319 intrinsically independent syntagms 366 Jakobson, Roman 25, 28, 214 Jakobsonian-type table 224, 225 Japanese 266, 268-270, 296, 297 Jones, Daniel 250, 251 juncture 80, 92 Kemöny, J.G. 31 Korean 219 labels 135, 136,140,141, 146 Lamb, M.A.L. 230, 234 Lamb, Sidney S. 370 language 90 Latin 280 linear order 286 linguistic description 5, 51, 170 linguistic descriptions 53 linguistic theories 53 linguistic theory 51 locative supplement 335, 336 Lockwood, D.C. 33 Malay 418 Martinet, Andr^ 55, 87, 88, 89,109, 119, 282,302, 309,329,330 material adequacy 58, 59, 61 material axiomatization 66 meta-hypotheses 43, 53 methodology 67, 179, 180 Miao 271 minimal pairs 218

General index

471

modal 325, 328 modality 325, 328 modals 327 models 14, 67, 151, 180 modifiers 283 monemes 113, 117, 209, 210, 217 morph 158, 159, 163 morpheme 252 morphemes 229, 233 morpho-phoneme 11,76 morphological analysis 117 morphological complex 192 morphological complexes 209 morphologically complex 114 morphology 80, 90, 91, 105, 110, 112, 113,211 morphon 11 morphs 165 Morse code 97-99 multiple conjunctive constructions 409 musical notation 103, 105

non-instrumental non-complementary intransitive PBS 317 non-instrumental transitive PBS 317 non-morphologically combinable syntactic entity 373 non-productivity 198 non-tactic 287 non-transitive PBS 312 nonce-symbols 123-125, 140, 145, 146 not wholly fixed conventional 103 notions 67 nuclear 102, 279 nuclear position 228 nuclearity 228 nucleus 101, 278, 283 nucleus nucleorum 102, 1278,283 number-writing 100, 101, 137, 211, 292 numeral 335 numerals 98

naive realist 182 natural indices 125 natural languages 50, 90 negation 163 neo-Bloomfieldian 27 neutralization 235, 236, 238, 239, 257,272 nominal syntagms 335, 338 non-complementary intransitive PBS 314 non-complementary non-direct transitive PBS 313 non-complementary transitive PBS 314 non-direct non-complementary transitive PBS 316 non-direct non-transitive PBS 315 non-direct transitive PBS 313 non-emphatic 380 non-instrumental intransitive PBS 316

observables 181 occurrence dependency 279, 288, 291, 342 onomatopoeic 143 ontology 72, 74, 77, 78, 151, 512 operationalism 180 opposition 163 optional complementary transitive PBS 316 optional indirect transitive PBS 315 ordered complexity 97, 116 ordered systems 93, 98, 109, 116 orders 21 para-cenotactic 89 para-phonotactic 89, 114 para-phonotactic analysis 117 para-phonotactic entities 92, 117 para-phonotactic level 373 para-phonotactics 80, 91, 373 para-plerotactics 80, 91

472

General index

para-syntactic 89 para-syntactic entities 92 para-syntactic features 114, 117, 371, 374,375, 377 para-syntax 91 para-tactic features 92, 100, 118 para-tactic levels 90, 110 parallel determination 174, 279, 280,292, 295-297 parameters 181 participle forms 394 passive 316, 317 "past tense" moneme 417 pause 92 PBS 310 Peking-dialect 223 Pekingese 252, 255, 268, 272,368 Pekingese Chinese 224 performance 256 peripheral 279 peripheral entities 101 peripheral positions 228 peripheralness 228 permutation test 191, 192 phenomena 51,52 phone 158,163 phonematic analysis 117 phonematically complex 114 phonematics 80, 91,110,113 phoneme 119,237 phoneme tables 220, 257 phonemes 113, 117, 214,217 phonetic entities 166 phonetic form 153, 154, 156, 158, 304 phonetics 91 phonological description 256, 258 phonological form 157, 158, 166, 167,304 phonological forms 93 phonological system 115 phonological words 92 phonologically simple 114

phonotactic analysis 117 phonotactic entities 210 phonotactically complex 114 phonotactics 80, 90,91,110, 113 phonotagms 80,113,117, 215, 217 •Pike, Kenneth 242 Pilch, Herbert 211,218 Plerematic inventory 112 plerematics 80,90, 91, 110 plereme 119, 120,301, 302 pleremes 113, 117, 118, 210, 217, 389 plerological entities 90 plerological system 99 plerotactic entities 210 plerotactics 80, 90, 91, 110 Popper, Karl 36, 41, 181 position 101, 228 position classes 296 positions 216, 228, 229,332 positivist 181, 182 possessive supplement 335 postulates 69 pragmatic 381 pragmatics 377, 379, 382 Prague school 235 predicative 325 predicative based syntagm 310 predicatives 323 prepositional syntagms 330, 331, 360 prepositions 331 primitive statements 65 productivity 193, 194, 196-198, 200, 202 pronominal relatives 359 pronominals 340 proper language 90 proper names 108, 123, 137, 340 proper symbol 123, 124, 140 proper symbols 125 prosody 80, 89 protocolization 47

General index

protocols 45, 46, 151, 281 pseudo hypothetico deductive 42 pseudo-composites 183, 184, 196 pseudo-productivity 193, 194, 197, 202 pseudo-words 183, 184, 199 puns 150 qualitative supplement 335 quasi-aspectual modals 327, 328 quasi-copulative modals 327, 328 quasi-plereme 343 quasi-subject 321 quasi-verbal modals 327, 328 quotes 139, 140 realizational levels 9 0 , 9 3 recursivity 109, 110, 118 relation 126 relation of correspondence 73 relative functionals 355, 359, 360 relative supplement 335, 336, 356, 359 relative syntagms 330, 337, 355, 360, 362 relatives 330, 331, 355 relator 126 restricted distribution 239 Rheinland dialects 246 Scottish English 247 second articulation 115 secondary articulations 219 self-containedness 84-87 self-evident truth 70 semantic import 95, 276, 277 semantics 72, 77, 78, 80 semi-cluster 299 semi-consonants 228 semi-distributional units 266 semi-PBS 356 semi-vowels 227, 228, 230 semiotic system 90 semiotic systems 83, 93, 96, 99

473

sentence 89, 118, 215, 306, 365, 373,375 sentence bases 374 sentences 9 2 , 1 1 7 sentential analysis 117 sentential entities 114, 365 sentential features 114, 117, 215 sentential level 80, 91, 92,114, 371, 373,374 sentential markers 367-369, 371, 372, 378 sentential semantics 382 series 219 sign 123, 124, 140, 150, 161 sign-theory 72 sign-theory 124, 152 signa 89, 93, 105, 125, 136, 154, 165 signalling devices 125, 130-132 signifiant 160 signify 160 signs 125 signum 123, 124, 135, 140, 150, 155, 158, 161, 166, 167,304 signum level 165 signum-theory 72, 77, 78 signum-theory 124, 151, 152, 153 simple dependent sentential entity 372 simple independent sentential entity 372 simple morpho-syntactic entity 372 simple non-syntactic entity 372 simple systems 93, 94, 116 simplicity 62, 64, 69 simultaneity 94, 209-214 sound symbolism 142 Southern English 247 Southern German 263 speculative 35 speculativism 180 speech phenomena 52 statements of realization 116

474

General index

stratificational linguistics 27, 370 structural 381 structural level 119 structural levels 90 structuralism 71 structure 154, 155 sub-disciplines 79, 80 sub-ordination 101,282 sub-systems 96 sub-theories 80 sub-theory 79 subjunctive 348-351 subordination 288 substantification 404 substantifiers 417 supplements 334, 335 suspensive 380 syllables 229, 235, 246 symbol 123, 124, 138, 140 symbolic logic 137 symbols 125 symptomatic indices 125, 127, 131, 132 synonyms 149, 163, 165 synonymy 162, 164, 165 syntactic analysis 117 syntactic complex 191, 192 syntactic entities 191, 210 syntactic entity 372 syntactic level 92 syntactic relations 191, 276, 277 syntactic well-formedness 374 syntactically complex 114 syntagmatic entities 117, 191,208, 210 syntagms 92, 101, 113, 117, 217 syntax 80, 90, 91, 105, 110, 112, 113,211 system 133 systematic phoneme 11 systemic economy 110 systemology 71,72,77-79, 90, 91 systems 85

tablature 106 tactic 287 tactic relations 287 tactically independent syntagms 366 tags 367,371 tense 325,328 Thai 219, 244-246, 258, 260, 261, 268, 367 theorems 69 theoretical statements 40 theory 37, 39, 5 4 , 7 0 , 1 8 0 , 181 theory of indices 72, 80, 124 theory-based 56, 256 Tibeto-Burmese 272 tone 92 "tone" language 271 tones 272 traffic signs 96, 211 transformationalists 27 transformationalism 35 transitive 322 transitive PBS 312 Troubetzkoy 214 truth 70, 240 Uhlenbeck, E.M. 88-90, 96, 110, 119,309 unilateral functional dependency 291 unilateral occurrence dependency 291 universalism 214 universals 22 unlimited 379 unordered complexes 116 unordered complexity 116 unordered systems 93, 110, 116 upper level of distinctive realization 256 utterance 74, 150, 153, 155, 167, 304, 306 utterance level 138

General index

475

utterances 151, 165, 215

wholly fixed conventional 103, 148, 164

verbal (E)P(S) 324 verbal predicatives 323, 324, 328 vowel-systems 240 vowels 227, 228

"whow" words 351, 355, 361,365, 366

well-formed sentence 118 well-formedness 85, 118, 375

Yulu 266

word 119, 120,303 writing conventions 99

George M. Horn

ffl Essentials of j^j Functional Grammar

=d ΠΤΙ

I ΓΠΙ

m

m im ΙΊΊΙ

m

1988.15.5 χ 23 cm. X, 404 pages. Cloth ISBN 311011286 8 (Trends in Linguistics. Studies and Monographs 38)

[mi ΙΎΙ

A Structure-Neutral Theory of Movement, Control, ana Anaphora

|

This monograph presents a theory of universal grammar which is structure-neutral in the sense that none of the basic principles and rule schemata associated with the major grammatical processes, movement, control, and anaphora, are formulated in terms of syntactic structure. Consequently, they apply equally to configurational and nonconfigurational languages, which are quite different at the level of syntactic structure, but which display essentially the same patterns in these three areas. The central component of the proposed theory is the Functional Component. The basic assumption is that all predicates have ordered argument structures. Both the principles which constrain the patterns of movement, control, and anaphora, and the rule systems which produce these patterns are formulated in terms of ordered argument structures. Illustrative examples are drawn from a number of Indo-European and non-Indo-European languages.

m

m fm

lmJ

mouton de gruyter Berlin · New York