On Grammar (Collected Works of Mak Halliday Series)

On Grammer

M. A. K. Halliday

Continuum

On Grammar

Collected Works of M. A. K. Halliday Volume 1: On Grammar Volume 2: Linguistic Studies of Text and Discourse Volume 3: On Language and Linguistics Volume 4: The Language of Early Childhood Volume 5: The Language of Science Volume 6: Computational and Quantitative Studies Volume 7: Studies in English Language Volume 8: Studies in Chinese Language Volume 9: Language and Education Volume 10: Language and Society

Volume 1 in the Collected Works of M. A. K. Halliday

On Grammar M. A. K. Halliday Edited by Jonathan Webster

Continuum The Tower Building, 11 York Road, London, SE1 7NX 370 Lexington Avenue, New York, NY 10017-6503 First published 2002 by Continuum Reprinted 2003, 2005 쑔 M. A. K. Halliday 2002 All rights reserved. No part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording or any information storage or retrieval system, without permission in writing from the publishers. British Library Cataloguing-in-Publication Data A catalogue record for this book is available from the British Library. ISBN 0-8264-4944-1 (hardback) Typeset by SetSystems Ltd, Saffron Walden, Essex Printed and bound in Great Britain by CPI Bath

CONTENTS

Preface

vii

Acknowledgements

ix

Introduction: a personal perspective Professor M. A. K. Halliday

1

SECTION ONE: EARLY PAPERS ON BASIC CONCEPTS Editor’s Introduction

17

1 Some aspects of systematic description and comparison in grammatical analysis

21

2 Categories of the theory of grammar

37

3 Class in relation to the axes of chain and choice in language

95

4 Some notes on ‘deep’ grammar

106

5 The concept of rank: a reply

118

Appendix to Section One

127

SECTION TWO: WORD–CLAUSE–TEXT Editor’s Introduction

155

6 Lexis as a linguistic level

158

7 Language structure and language function

173

contents

8 Modes of meaning and modes of expression: types of grammatical structure and their determination by different semantic functions

196

9 Text semantics and clause grammar: how is a text like a clause?

219

10 Dimensions of discourse analysis: grammar

261

SECTION THREE: CONSTRUING AND ENACTING Editor’s Introduction

289

11 On the ineffability of grammatical categories

291

12 Spoken and written modes of meaning

323

13 How do you mean?

352

14 Grammar and daily life: concurrence and complementarity

369

15 On grammar and grammatics

384

Bibliography

419

Index

433

vi

PREFACE

For nearly half a century, Professor M. A. K. Halliday has been enriching the discipline of linguistics with his keen insight into this social semiotic phenomenon we call language. His scholarship has advanced our understanding of language as an activity which is both rational and relational, systemic and semantic, dynamic and diverse. Building on the legacy of his mentor, Professor J. R. Firth, Halliday approaches language from the vantage point of meaning and purpose, and provides a sound theoretical framework for dealing with questions about how and why we come to use language as we do for being and becoming who we are. Halliday’s work has long attracted a wide audience, which includes linguists, educators, computer scientists and policy makers. What many find appealing in the man and his scholarship is his rejection, on the one hand, of the elitism typical of certain other schools of linguistics, while on the other hand embracing the study of that which powers language and also conditions our ways of thinking and behaving. In this series, we present the collected works of Professor M. A. K. Halliday in ten volumes. Covering a wide range of topics related to language and linguistics, these are: 1 2 3 4 5 6 7

On Grammar Linguistic Studies of Text and Discourse On Language and Linguistics The Language of Early Childhood The Language of Science Computational and Quantitative Studies Studies in English Language vii

preface

8 Studies in Chinese Language 9 Language and Education 10 Language and Society Halliday approaches language from above, from below and from roundabout (see Chapter 15, Section 15), but not from a distance. His collected works, as presented in these ten volumes, reflect his characteristic balance between formulating and applying linguistic theory. The depth of his insight into language as system is highlighted in such volumes as the present one On Grammar and the third volume On Language and Linguistics. The strength of his commitment to the study of language as it is actually used is demonstrated in subsequent volumes dealing with Linguistic Studies of Text and Discourse, The Language of Early Childhood and The Language of Science. The breadth of Halliday’s interest in all things Language is glimpsed in the volumes Studies in English Language and Studies in Chinese Language. The application of his knowledge and experience as linguist and social scientist is visited in volumes Computational and Quantitative Studies, Language and Education and Language and Society. The first volume contains fifteen papers, with the addition of a new piece entitled ‘A personal perspective’, in which Professor Halliday offers his own perspective on language and linguistic theory as covered in his collected works. The papers are divided into three sections according to topic, and within each section the papers are ordered according to the date they were written (which does not always correspond to the date of publication). The first section presents early papers (1957–66) on basic concepts such as category, structure, class and rank. Interestingly, the second section highlights how over the span of two decades (mid-1960s to mid-1980s) Halliday developed systemic theory to account for linguistic phenomena extending upward through the ranks from word to clause to text. The third section includes more recent work in which Halliday discusses the issues confronting those who would study linguistics, or as Firth described it ‘language turned back on itself ’.

viii

ACKNOWLEDGEMENTS

We are grateful to the original publishers for permission to reprint the articles and chapters in this volume. Original publication details are provided below, and also at the beginning of each chapter. ‘Some aspects of systemic description and comparison in grammatical analysis’ from Studies in Linguistic Analysis, published by Blackwell Publishers, 1957, pages 54–67. Reprinted by permission of Blackwell Publishers. ‘Categories of the theory of grammar’ from Word, 17(3), 1961, published by the Linguistic Circle of New York (now the International Linguistic Association), pages 241–92. Reprinted by permission of the author and the International Linguistic Association. ‘Class in relation to the axes of chain and choice in language’ from Linguistics, vol. 2, 1963, published by Mouton (now Mouton de Gruyter), pages 5–15. Reprinted by permission of Mouton de Gruyter. ‘Some notes on ‘deep’ grammar’ from Journal of Linguistics 2(1), 1966, published by Cambridge University Press, pages 57–67. Reprinted by permission of Cambridge University Press. ‘The concept of rank: a reply’ from Journal of Linguistics 2(1), 1966, published by Cambridge University Press, pages 110–18. Reprinted by permission of Cambridge University Press. ‘Lexis as a linguistic level’ from In Memory of J. R. Firth, published by Longman, 1966, pages 148–62. Edited by C. E. Bazell, J. C. Catford, M. A. K. Halliday and R. H. Robins. ‘Language structure and language function’ from New Horizons in Linguistics, published by Penguin Ltd, 1970, pages 140–65. Edited by John Lyons. 쑕 M. A. K. Halliday, 1971, collection 쑕 John Lyons, 1970. Reprinted by permission of the Penguin Group (UK). ix

acknowledgements

‘Modes of meaning and modes of expression: types of grammatical structure and their determination by different semantic functions’ from Function and Context in Linguistic Analysis: Essays Offered to William Haas, edited by D. J. Allerton, Edward Carney and David Holdcroft, published by Cambridge University Press, 1979, pages 57–79. Reprinted by permission of Cambridge University Press. ‘Text semantics and clause grammar: some patterns of realization’ from The Seventh LACUS Forum, edited by James E. Copeland and Philip W. Davies, published by LACUS, 1981, pages 31–59. Reprinted by permission of LACUS. ‘How is a text like a clause?’ from Text Processing: Text Analysis and Generation, Text Typology and Attrition (Proceedings of Nobel Symposium 51), edited by Sture Allen, published by Almqvist and Wiksell International, 1982, pages 209–47. Reprinted by permission of Almqvist and Wiksell International. ‘Dimensions of discourse analysis: grammar’ from The Handbook of Discourse Analysis, Vol. 2: Dimensions of Discourse, published by Academic Press Inc., 1985, pages 29–56. ‘On the ineffability of grammatical categories’ from The Tenth LACUS Forum, edited by Alan Manning, Pierre Martin and Kim McCalla, published by LACUS, 1984, pages 3–18. Reprinted by permission of LACUS. ‘Spoken and written modes of meaning’ from Comprehending Oral and Written Language, published by Academic Press Inc., 1987, pages 55–82. Reprinted by permission of Academic Press, Orlando, Florida. ‘How do you mean?’ from Advances in Systemic Linguistics: Recent Theory and Practice, edited by Martin Davies and Louise Ravelli, published by Pinter, 1992, pages 20–35. ‘Grammar and daily life: concurrence and complementarity’ from Functional Approaches to Language, Culture and Cognition, edited by Teun A. van Dijk, published by John Benjamins Publishing Co., 2000, pages 221–37. Reprinted by permission of John Benjamins Publishing Co. ‘On grammar and grammatics’ from Functional Descriptions: Theory in Practice, edited by Ruqaiya Hasan, Carmel Cloran and David G. Butt, published by John Benjamins Publishing Co., 1996, pages 1–38. Reprinted by permission of John Benjamins Publishing Co.

x

INTRODUCTION: A PERSONAL PERSPECTIVE

The volumes in this series will contain a selection of my writings on language, extending over the half century beginning in 1951. A few, including one or two items written specially for the series, have not been published before, and many of those that were published appeared in rather inaccessible places. The papers are arranged according to topic, beginning with the present volume, which is oriented towards grammatical theory. But the topical arrangement will tend to be fairly loose, partly because my writing has always been inclined to drift, and partly because both the editor and I prefer it to be that way – weak boundaries have always been characteristic of my approach. I have never really thriven in a discipline-based structure of knowledge. It was a feature of my century – the late and rather unlamented twentieth, perhaps mercifully short in Eric Hobsbawm’s conception of it – that it began by erecting walls between the disciplines, and it is proving difficult to demolish these walls now that they have come to be constraining rather than enabling. They had been enabling to start with, at least for the newly founded social (and ever newer semiotic) sciences; sociologists, psychologists and linguists had to be able to lock each other out while sorting out and investigating their own chartered domains. So in the mid-century many linguists sturdily proclaimed the independence and autonomy of the discipline of linguistics, and one could sympathize with their anxiety, because language was everybody’s business and there would always be outsiders looking over their shoulders and telling them how to do their job – or, more usually, telling them they were simply wasting their time. I think I myself once made some reference to this; if so, it will turn up in Volume 3. But 1

on grammar

what I did stress was how much linguistics had in common with other scholarly pursuits and, when it came to asking questions about language, I always found myself lining up with the outsiders. It seemed fine for us, as linguists, to determine the content or domain of our own discipline – sociologists studied society, psychologists studied . . . whatever psychologists do study, linguists studied language.1 That’s what we were there for. But it did not seem fine, to me at least, for us to determine what questions should be asked about that domain. I was interested in what other people wanted to know about language, whether scholars in other fields or those with practical problems to be faced and solved – including that undervalued and under-rewarded group who have to be both scholars and practical problem-solvers, namely teachers. There was nothing surprising about this last perspective: not only had my parents both been teachers, but I myself had taught languages for thirteen years before transferring myself into a teacher of linguistics. But even before that, while still at school, I had been trying to find out about language – because I was keen on literature and wanted to understand why its language was so effective, what was special about it. There is no separating one’s personal history from the academic paths one pursues, nor any way of detaching cause from effect in explaining one’s chosen approach to a field of study. One way or another, I have always found myself asking the kinds of questions about language that arose, as it were, from outside language itself. Of course linguists always have been located, and located themselves, within some broader context; there is nothing unusual about that. But at any given “moment” in space-time, there are likely to be only a few predominant motifs by which the context of linguistic scholarship is defined. This may even be legislated from on high, as when Stalin writing in 1950 (or Chikobava, writing on his behalf) instructed Soviet linguists to get on with the job of demonstrating the linguistic unity of the Slav nations – reasonably enough, since the Soviet Union had just taken them all over. Usually it is determined by less overtly political factors: by particular social movements and demands, or notable advances in knowledge in some other field. The present era provides a noteworthy example of the latter. Since about 1985 there has been spectacular progress in the field of neuroscience; the combination of new technology – positron-emission tomography, magnetic resonance imaging and its derivatives – with new insights in evolutionary theory and its contributing disciplines has transformed the way we understand the human brain, how it has 2

a personal perspective

evolved in the species and how it develops in the individual from birth (and before) to maturity. And this new understanding has radically redefined the place of language. It is now clear that language and the brain evolve together, and that these develop together in infancy and childhood. The development of the brain is the development of the ability to mean; as in every aspect of human history, so in the ontogeny of the individual human being the material and the semiotic interpenetrate, as complementary aspects of the characterology of the species (McCrone; Edelman; Deacon; Dawkins; Jones). To say this is not to proclaim that the human species is unique in this respect or that no other species has evolved, or could evolve, a similar type of higher-order semiotic. On the contrary. The work of Duane Rumbaugh and Susan Savage-Rumbaugh has brought out the point that the bonobo chimpanzees can operate with sets of arbitrary symbols in a way that is analogous to our own system of wording (lexicogrammar). They lack an analogous vocal apparatus, but that is beside the point. It is tempting to assume that they have been following the same evolutionary path and are simply less far advanced along the way – this is the assumption that prompts questions like ‘what age have they reached, in terms of a human child?’ But this assumption is probably wrong, or at least misleading, if it is used to describe an adult chimpanzee in terms of an immature human; the adult bonobo’s brain is fully wired up in terms of the construing of experience, and enacting of social relations, that constitute bonobo culture. The question can fairly be asked about bonobos brought up from birth in a human-like semiotic environment, like Kanzi, and it is too early to say yet whether Kanzi and the other youngsters’ development of the power of meaning tracked that of human children and stopped at a certain level or whether it was proceeding along a somewhat different route. It might be argued that such new knowledge about how the brain functions, and how it evolves and develops, has no significance for the way linguists describe and explain language, especially at the ‘inner’ strata of lexicogrammar and phonology (wording and sounding). Possibly; although even here it seems to me to set certain constraints and more importantly perhaps to favour certain explanations over others. It suggests “systems thinking” rather than compositional thinking (Matthiessen), grammatical logic rather than formal logic (Sugeno 1995), fuzzy and probabilistic categories rather than clearly bounded and deterministic ones. Since the brain is more like a jungle than like a computer (Edelman 1992), it disfavours representations of grammar and phonology that are influenced, however indirectly or subconsciously, 3

on grammar

by the way that computers happen to be being designed and operated at this particular moment in technological history. But there are two aspects of the work of linguists that surely are impacted directly by any new understanding of the brain. One is that of those branches or special fields within linguistics that relate closely to neuroscience: developmental linguistics (the study of “language acquisition”), and pathological and clinical linguistics (the study of language disorders of all kinds). Both of these fields have immense “practical” applications, in education and in medicine, that contribute to human well-being. I have not worked in the latter area. But I have written a certain amount on child language development, based on intensive research of the ‘diary’ type (Bates, in Rumbaugh and Rumbaugh) and it was clear already when I started that this must have mutual implications with the study of the brain. This was in the early 1970s, before the explosion of knowledge in neuroscience; I was not able to make any sensible use of what little I had read then about brain functioning. I simply drew attention to two factors that had emerged very clearly from my own researches: first, that the language surrounding a child was rich and highly structured, very different from the formless and impoverished quality that was being asserted about spoken language at the time; secondly, that before mother tongue came child tongue (I called it “protolanguage”), which had a different structure from “adult” (post-infancy) language – so if children were born endowed with a grammar-shapen brain, why did they first construct a language of quite a different type, which had no grammar in it at all?2 Today matters have changed, and students of child language development can hardly avoid taking note of what has been found out about the development of the infant brain. The other aspect of the work of linguists that is impacted by neuroscience is a more macro one: the modelling of the system of language as a whole. The overall construction of language as system was very much part of the enterprise of twentieth-century linguists from Saussure onwards and reached a high point with Hjelmslev’s Prolegomena (1961), first written in 1942. Since then it has become backgrounded, for various reasons: the subject expanded into a colony of subdisciplines, or branches, which seemed not to need any general perspective; in the west, at least, including the USA where most of the research was being done, Chomsky’s post-Bloomfieldian model became dominant and was not open to challenge; and the general post-modern ethos was in any case hostile to comprehensive accounts (they were seen as “totalizing”), often in fact to theorizing of any kind. 4

a personal perspective

An exceptional figure in this period was Sydney Lamb, who took over Hjelmslev’s vision and continually revised and refined it in the light of his own thinking and his own research. Lamb set out quite explicitly to model language in terms of neural structure and neural processes and, having been ignored or rejected by mainstream linguists for many decades, he has now come into his own. His Pathways of the Brain is a major work of linguistic scholarship that is fully compatible with the new thinking in neuroscience. It also brings out how essential it is to model the linguistic system as a whole if linguistics is to be taken seriously among the sciences rather than being set aside as a somewhat eccentric pastime for grammarians and philosophers of mind. Since my own thinking is in many ways close to that of Sydney Lamb, and we collaborated for some time in the 1960s, this is perhaps a good point of departure for the next step in the argument. By talking about the intellectual environment in which the study of language is pursued, I have foregrounded the context of neuroscience because that is where major advances have recently been made. But linguistics has many frontiers. If we express these in disciplinary terms, they would include sociology, anthropology, legal studies, psychology, history, politics, literature, fine art and music, computer science and physics, as well as education, medicine and biology, already mentioned. Put like that, they amount to a dull and rather forbidding catalogue; so let me make the same point in more concrete and friendly terms. Once having begotten language, as a species, and as individuals, we are stuck with it; we can’t get rid of it, and we can’t do anything without it. Language and the brain are co-created; they can also be codamaged, and co-destroyed. A particular part of language – say its vocabulary – can grow up with a particular part of the brain, but it can be dislodged and migrate to somewhere else. And whenever we process language, every region of the brain is involved. The same holds good, by analogy, for language and society: language creates society, by enacting social relationships and, by the same token, language disrupts and destroys. A particular language co-evolves with a particular culture, but another language can come along and usurp its place, and the culture survives. And whenever we use our language, all aspects of the culture are invoked. Language creates and maintains the law; it also functions to challenge and to subvert it. Human history is the interplay of material and linguistic forces, enabling and constraining, colluding and conflicting by turns. Literature attempts to transcend language, but has to use language to do it (“reading between the lines” is still reading). Visual art, music and dance are independent of language – 5

on grammar

but you have to know language in order to understand them. Computers are built to a logic derived by design from grammar; they will have to think grammatically if they are going to advance any further. And while language is subject, like everything else, to the laws of physics, the laws of physics are themselves construed in language in a specially designed form known as mathematics, which evolved as the language of measurement. The brain, in other words, is only one of many phenomena that can serve as the point of vantage from which language is viewed and explained. It is one that happens to be particularly favourable just at present, because of the success in brain science. But any other perspective – literary, social, physical, logico-philosophical or whatever – is equally valid and language will look somewhat different from each of these different vantage points. Some will obviously be more relevant than others for particular research applications: an audiologist, for example, looks at language as a physical system (i.e. system-&-process), taking account of the physical properties of the sound wave; and again there is a special branch of linguistics, speech science, where knowledge about language as a physical system is one of the central concerns. The fragmentation of linguistics into a family of subdisciplines reflects and institutionalizes these various angles of approach. If we take it that, whereas “branches” of technology deal with different parts of a system, or different stages of a process, branches in science tend to deal with different aspects of one of the same system-&-process, then it is in linguistics that this tendency reaches its furthest point.3 I used to think that language, or at least the core layers of language, lexicogrammar and phonology, would have to be modelled and described differently in all these different contexts, at least for purposes of different applications or different research goals. This was the view expressed in ‘Syntax and the consumer’ (Halliday 1964). This approach was partly taken as a defence against the dominant elite, for whom linguistics was “a branch of theoretical psychology” (Chomsky) – in the words of Ross ‘I take it for granted that the goal of linguistics is [sic] to explicate the difference between the human brain and that of an animal’. I was taken to task for suggesting that there might be more than one way of modelling and describing a language (Wales). My problem was, however, that I could not concentrate my vision. Unlike Sydney Lamb, who chose his point of vantage and then stuck to it, I was constantly jumping around to see what language looked like when viewed from the other side. To the extent that I favoured any one angle, it was the social: language as the creature and creator of 6

a personal perspective

human society, as expounded by my teacher J. R. Firth and by my friend and colleague Basil Bernstein. But by nature, and also by experience, I was (and have always remained) a generalist. So while consciously I was trying to model language as a social phenomenon, in fact I was acting against my own advice and trying to look at language from every possible vantage point in turn.4 Most linguists, it seemed to me, looked at language only from the inside, claiming the right to formulate their own questions about it – which was why linguistics seldom interested practitioners in other fields. This was also, of course, a perfectly valid perspective. But it did bring with it certain problems. When I was being trained as a dialect fieldworker, by my other great teacher Wang Li (then Professor of Linguistics at Lingnan University, Canton), there were still no tape recorders. We had to transcribe responses directly into IPA script, which was excellent training for my later investigation of child language. Professor Wang was able to acquire a primitive version of the same thing – a wire recorder, but it was not much use, because the wire was always breaking and would end up as a ball of wire wool fit only for scouring a wok. That was in 1949–50. Not that linguistics had had no base in technology. There was already high quality instrumentation for acoustic analysis (spectrograph, oscillograph, mingograph), as well as various techniques for investigating the articulatory mechanism of speech. Gramophone records were widely used in language teaching: when I was taught Chinese for the armed services at the University of London in 1942–43, the Department had its own recording equipment on which students could register their own performance and compare it with the recorded model. There were archives of spoken language on disk and even on cylinder, including dialect survey material in a number of different languages. But there was no technology for capturing authentic speech, natural conversation in the interactive situations of daily life, nor for managing an extensive body of text. As a consequence, linguistics had hardly any data. In that respect it was about where physics had been at the end of the fifteenth century, before technology had evolved to enable physicists to observe and to conduct experiments. Linguists either relied on the kind of manicured discourse that is produced in writing and in prepared and selfmonitored speech, or else invented data for themselves from their own intuition of the language, and they had no way of processing large quantities even of that. For linguistics, the two most important advances in the latter half of 7

on grammar

the century were technological ones: the invention of the tape recorder and the evolution of the computer. The tape recorder made it possible to record natural speech. The computer made it possible to process large quantities of data. The two together have given us the modern computerized corpus, with natural speech as a significant component, on which we can undertake quantitative analyses on a statistically significant scale. As a bonus, the computer enables us to test our descriptive generalizations, through text generation and analysis (“parsing”), and to observe and represent sound waves in a wealth of complementary perspectives. These resources have transformed (or at least are in process of transforming) the way language looks from the inside. Patterns are being revealed that we have known must be there, because there was a gap where the approaches from the lexical and the grammatical poles of the lexicogrammar converged, but which we could not see: the nature of grammatical logic is beginning to be understood; the semogenic (meaning-creating) power of discourse is coming into view, both in monologic and in dialogic mode; quantitative mechanisms of linguistic change are beginning to appear on the agenda. From all this it should be possible in the next decade or two to crack the semiotic code, in the sense of coming fully to understand the relationship between observed instances of language behaviour and the underlying system of language – something that has eluded us up till now, so that we have even turned the two into different disciplines, calling only one of them “linguistics” and labelling the other “pragmatics”. Some people will feel threatened by this new understanding. We know this because there are those who already do. To bring to light the systems and processes of society is already threatening enough, as witness the panic reactions to Bernstein thirty years ago when he demonstrated how social class structures are transmitted, but semiotic systems and processes are even nearer the bone. As long as linguists confined their attention to dead languages, codified texts or sanitized examples like John kissed Mary and It’s cold in here, no one would feel really at risk. But when grammar extends to the study of the meaningcreating power of everyday real life talk, it starts to become dangerous. Some people feel worried that the grammarian is someone who knows what they are going to say next and even if they can be reassured that that is not what theory is about, it is scarcely less threatening (apparently) to be told what proportion of positive to negative clauses they are going to use in their speech. And for others, just to be faced with a record of real life conversation can be unnerving; they feel embarrassed 8

a personal perspective

and ill-at-ease at what seems to invade the interactants’ privacy and strip away their elaborately constructed social identities. Others, even if they do not feel threatened or embarrassed, might still want to ask why it matters. Why do we need to bring this extra dimension into our understanding of language? Isn’t it enough to play the traditional part of a grammarian or a phonologist and join in the endeavour – itself an enterprise that has notched up considerable successes – of broadening our knowledge of the history, typology and structure of the world’s many languages? After all, there is more than enough work here to occupy the community of linguists, even if it was enlarged many times over, in meeting all the theoretical and practical demands in education, multilingualism and multi-culturalism, ecolinguistics, language maintenance, translating and interpreting, forensic linguistic work and so on. Why do we need a huge computerized corpus of authentic data, which in any case will be available only for a small number of the world’s major languages? There are it seems to me two answers to such questions – or two parts to what is ultimately a single answer. One is to complete the record of a language or rather, since it can never be complete, to make it more comprehensive and more accurate. This is what Quirk had in mind when, in launching the first systematic modern corpus, the Survey of English Usage at University College London, he described it as moving towards ‘an N.E.D. of English usage’. It was taken for granted that one of the goals of lexicography was to put on record ‘all’ the words of a language; it was natural to set the same target for the lexicogrammatical patterns in which the words are used. The other part of the answer is perhaps something of a paradox – or is made to seem paradoxical by “corpus linguists” when they describe themselves as “mere data-gatherers”: to upgrade our theory – to improve our theoretical understanding of the nature and functioning of language. If it is true, as is so often proclaimed, that the balance of people’s activities is going to shift more and more from the material to the semiotic domain, leaving machines and robots to do the material business, then the demands on language and its satellite systems are going to go on increasing and hence, inevitably, the demands on theory of language. Our world consists of these two grand phenomenal domains, matter and meaning. The science of matter is physics; the science of meaning is linguistics.5 There are fads and fashions in every field of study and linguistics is no exception. In the 1960s, it was almost impossible to get published any analysis of a text. The worst insult that could be paid to a linguist 9

on grammar

was to say that he or she was “data-oriented”. Data were said to be irrelevant to the serious study of language; the actual language used by real people, especially spoken language, was dismissed as impoverished and unstructured, a mere matter of performance that could tell us nothing about the true object of description, which was linguistic structure – the rules generating the set of idealized sentences that constituted the ideal speaker’s competence or knowledge of the language. This monolithic Cartesian culture maintained its stability by constantly re-examining its own foundations, finding newer and more elegant ways of going over the same ground. The idealogy that pervaded it and the conditions it brought about have been well described by de Beaugrande. Since it excluded any reference to the social context of language, it was necessary to invent a new field called sociolinguistics and a new kind of competence called “communicative competence” to go with it (Hymes 1971). And when a change of fashion brought discourse on to the agenda, an analogous development took place. After one or two attempts to handle text within the same formalist framework had proved vain, pragmatics was brought to life as an independent disciplinary base (and channel for getting things published) and suddenly everybody was “into discourse”. A number of factors came together to ensure the success of the pragmatics enterprise, which has released an enormous amount of energy and raised to theoretical status discursive issues such as implication, relevance and politeness. Having grown up in opposition to linguistics, pragmatics has largely dispensed with grammar; what theoretical input it has had has been drawn from strands in philosophy and sociology rather than linguistics. In view of its undoubted achievements this may not seem to matter. Perhaps I am just being old-fashioned in deploring this split between two aspects of what to me is a single enterprise: that of trying to explain language. It seems to me, however, that both parts of the project are weakened when they are divorced one from the other. The problem is, that if you don’t know the system you can’t understand the text. Discourse is the form in which linguistic systems are instantiated. From this point of view, pragmatics is the instantial aspect of semantics: the semantics of the instance, in other words. To put this in the opposite perspective, the system is the meaning potential that lies behind every instance of discourse. Children construct the system, in fact, from very large numbers of discourse instances, which in the typical case are fluent, well organized and related (and hence relatable) to their instantial situational contexts. 10

a personal perspective

If linguistics had not fragmented in this way, with the system of language being represented as if it had some mode of being of its own, unrelated to any text, and texts being expounded as if they were innocent of any underlying system, it would have been much harder for academics from other fields to dismiss language as irrelevant to their own – and, by implication, everybody else’s – concern.6 One influential exponent of this position has been Bourdieu (Hasan 1999). Bourdieu is an expert in exploiting the power of language to proclaim that language has no power. This carries the comforting message that therefore you needn’t bother to analyse it. Since grammar is difficult and analysing the lexicogrammar of a text requires a great deal of time and thought, any message that draws attention away from language will always be gratefully received. The argument being offered is that if you take account of the centrality of language you are being “logocentric”: anything “centric” is condemned without being tried. In practice the problem is exactly the opposite – we might call it ‘centrifugal’: whenever in dealing with any issue people are brought face to face with language, they will choose to avoid engaging with it if they can. To come back to the issue of pragmatics: I am not implying, obviously, that discourse does not depend on factors such as inference, knowledge of the universe and the like. Again, children incorporate these into their language games. I remember one of our rhymes: Johnny wondered which was louder, Dynamite or blasting powder. He bought some powder and struck a light; He hasn’t yet tried the dynamite.

This sort of thing was often used to challenge those of inferior understanding (i.e. younger siblings, etc.).7 But I don’t think it is sensible to treat these features as if they were of a different order of reality from language. They are all phenomena of, and operations in, meaning. What we have to do is extend and enrich our semantics to the point where we can handle these things as part of the system and process of language. Such a task should now be on the agenda. There is a lot of work to be done before the grammatics reaches the point where it can account for inferential relations like these: explaining why Johnny has not yet tried the dynamite, and how the properties of silk purses and sows’ ears are analogous to those of the imagined Harry after Maggie’s makeover and Harry as he now is. But the relations to be accounted for are relations of meaning; they are rather more complex than the singular semantic relations that have always been 11

on grammar

familiar in the lexicogrammar, like hyponymy and polarity and voice, but still ultimately of the same kind. It should be possible to extend the power of semantic representation so that such ways of reasoning can be integrated into our model of language, rather than being treated as if they were separate operations in the brain. And it will be necessary to do this, I think, in order to achieve the kind of “intelligent computing” that is envisaged by Michio Sugeno in his work at the Brain Sciences Institute in Tokyo. What the following chapters do is to illustrate some of the steps in my own thinking that have led me in this general direction. The steps that have seemed to me perhaps most critical in this endeavour might be summarized as: the unity of lexicogrammar; the priority of the view ‘from above’, from meaning and function; the move into systemics (system networks), freeing the grammar from the restrictions imposed by structure; the metafunctional foundation, disentangling the strands of meaning that are woven together in the syntax; the construction of language by children, from protolanguage to mother tongue; the decoupling and recoupling of lexicogrammar and semantics – the phenomenon of grammatical metaphor; the conceptualizing of the relation between system and text (instantiation) and the probabilistic nature of linguistic systems. Some of these will be treated separately in later volumes: in particular Volume 4 on child language and Volume 5 on grammatical metaphor and the language of science. Others will appear in various contexts and under a variety of headings. I am often asked about my views on “linguistic universals”. The answer is that I follow Hjelmslev and Firth in distinguishing theoretical from descriptive categories. The theoretical categories, and their interrelations, construe an abstract model of language (and other semiotic systems); they are interlocking and mutually defining.8 The theory that is constituted in this way is continually evolving as it is brought to bear on solving problems of a research or practical nature. (No very clear line is drawn between “(theoretical) linguistics” and “applied linguistics” – except institutionally where, for example, an education authority will give teachers release time and professional credit for a degree course called “applied linguistics” but not for one called “linguistics”.) Descriptive categories are categories set up in the description of particular languages. When people ask about “universals”, they usually mean descriptive categories that are assumed to be found in all languages. The problem is that there is no mechanism for deciding how much alike descriptive categories from different languages have to be before they are said to be “the same thing”. There is a method, 12

a personal perspective

based on the (theoretical) category of system, for matching up descriptive categories across languages but they are not claimed to be universal, and no grand hypothesis stands or falls by their “universality”. The unity of human language, and its relation to the human brain, is proclaimed by the multifaceted architecture of the theory. A volume of typological studies organized around the theoretical category of metafunction will serve to illustrate this standpoint (Caffarel, Martin and Matthiessen 2002). My own interpretation of the grammar of modern English will be found in An Introduction to Functional Grammar (Halliday 1985 and later editions). Other descriptive papers on English and on Chinese will be presented in Volumes 7 and 8. A theory-based account of the ideational semantics of English is in Construing Experience through Meaning (Halliday and Matthiessen 1999). I doubt whether any of the present volumes would have appeared without the enthusiasm, energy and efficiency of “my” editor, Dr Jonathan Webster, of the City University of Hong Kong. He brought the whole project to life, convincing me that it was worthwhile and convincing the publishers that it could actually come to fruition. It has been a pleasure being driven along by his momentum. My thanks also to the publishers, especially to Janet Joyce, who despite years of my ineffectual attempts to get started never lost patience with me or faith in the enterprise, and to Robin Fawcett, who set the whole thing going and provided many rounds of valuable suggestions and advice.

Notes 1. Psychologists, in fact, study psychology – the domain is defined by the discipline, rather than the other way round. Hence the rather odd locutions like “criminal psychology”, meaning the mind-set, or psyche, of criminals, rather than psychological theories that criminals have devised. I was once put down rather scathingly by a psychologist for suggesting that their domain of study might be the human psyche. 2. See Volume 4 in this series. 3. Language is a system of meaning (a “semiotic” system); and semiotic systems are of the fourth order of complexity, being also physical and biological and social. This means that one and the same linguistic phenomenon (whether “a language” or a single utterance by one speaker) will appear in all these various guises. 4. I hope it will be clear that I am not seeking either to justify this approach or to apologize for it. These bits of personal history are brought in simply to provide a context, to explain the way the papers in these volumes

13

on grammar wander throughout the highways and byways of language. If there has been any consistent motif, it has been ‘now how would this (phenomenon and its explanation) seem to someone who is interested in language for some other reason, different from the one that prompted me to explore it?’ 5. Strictly speaking, of course, it is semiotics; but semiotics has not yet evolved into a general theory of meaning and it seems likely that, for the time being at least, the way forward is by extending linguistics into other semiotic systems. I use “meaning” rather than the term “information” (the term imported from those who work on matter) because information is only a sub-class of meaning; it is the part that can be measured, whereas, unlike matter, meaning in general is not open to measurement (though systemic linguistics offers one way in; see Volume 5). 6. We are of course accustomed to linguistics being dismissed in this offhand way: “linguists always . . . (or . . . never . . .), so you needn’t bother yourself with what they write”. This is irksome but does little real harm – linguists will go on writing anyway. What I am talking about here is the assertion that language has no relevance – for example to social and political processes, and to anyone’s intervention in them. Such assertions can do a great deal of mischief. 7. I have, alas, no tape recordings of my grandmother, who died in 1959, in her mid-nineties. She belonged to the last generation, within my own culture, who spoke unselfconsciously in proverbs. A proverb was a theory of experience, but it was a commonsense theory, not a designed theory, and so construed in commonsense grammar, as one of a class of instances rather than a higher order abstraction. A snatch of dialogue might run like this (the example is invented): Harry’s no good; he’ll never carry corn. That business of his’ll never thrive, believe me. I don’t know; he might pull through. And Maggie’s certainly trying to buck him up a bit; she’s set her mind to that. She can’t change him, however hard she tries. You can’t make a silk purse out of a sow’s ear. It’d take more than Maggie to make anything out of him. A task for the grammatics is to show the relationship between the proverbial construct and the remainder of the discourse. 8. They constitute, in Firth’s formulation, “a general linguistic theory applicable to particular linguistic descriptions, not a theory of universals for general linguistic description” (Firth 1957: 21).

14

SECTION ONE EARLY PAPERS ON BASIC CONCEPTS

This page intentionally left blank

EDITOR’S INTRODUCTION

In this first section we look at five papers written and published by Professor Halliday over a ten-year period from 1957 to 1966. The basic concepts, which form the foundation of Halliday’s systemic theory, are elaborated in these papers. These include such fundamental categories for a theory of grammar as unit, structure, class and system. He also addresses the relations of these categories to each other and to the data in terms of scales of abstraction: rank, exponence and delicacy. When asked to compare his own approach with those of other linguists who helped shape not only his own thinking but also the discipline of linguistics as a whole, Halliday notes Firth’s interest in varieties of a language, Hjelmslev’s focus on language as a whole and Jakobson’s search for universals across all languages. Already in his early writings, Halliday draws on the insights of these and others to construct a theory of grammar grounded in the linguistic analysis and description of particular languages, which acknowledges the primacy of meaning and the need for systematicity. Published in 1957, the first paper in this section, ‘Some aspects of systematic description and comparison in grammatical analysis’, discusses theoretical considerations which developed out of the body of ideas that went into his doctoral dissertation. Building on and extending the general linguistic principles established by Firth and other scholars, Halliday demonstrates the application of formal methods of linguistic description to New Chinese (Modern Pekingese). Identifying formal linguistic methods as being derived from structural linguistic theory, Halliday maintains that ‘a complete analysis at the grammatical level, in a particular description in which all forms of the language are related to systems set up within the language itself, requires establishment of grammatical 17

early papers on basic concepts

categories, ordered as terms in interrelated systems and having as exponents the substantial (phonic or graphic) segments of the text’. Clearly influenced by Firth’s teaching and his scholarship, Halliday draws on Firth’s approach to formulate a General Linguistic theory, which is concerned with how language works at the level of grammar. While some have referred to Halliday’s approach as neo-Firthian, such a characterization does neither scholar justice. Realizing that his theoretical approach, as outlined in ‘Categories of the theory of grammar’ (Chapter 2), diverged from that of his mentor, Halliday sought to have the opportunity to discuss the paper with Firth. Firth’s sudden passing, however, prevented this from happening. In this paper, Halliday sets out the following fundamental categories for the theory of grammar: unit, structure, class and system, which relate to one another and to the data along three distinct scales of abstraction, including rank, exponence and delicacy. Halliday prefaces his discussion by stating what he regards as ‘given’, among which he includes the following: a. Texts, or observed language events, are the data to be accounted for, whether spoken or codified in writing. b. Description consists in relating the text to the categories of the theory. Description is not theory; rather it is a body of method derived from and answerable to the theory. c. Linguistic events should be accounted for at a number of different levels. While the primary levels are form, substance and context, a complete framework of levels requires certain further subdivisions and additions, including substance, whether phonic or graphic, form on two related levels of grammar and lexis, and context, which is an interlevel relating form to extratextual features. d. The study of phonic substance belongs to a distinct but related body of theory, namely General Phonetics. Phonology, on the other hand, relates form and phonic substance, i.e. where linguistics and phonetics interpenetrate. e. Language has both formal meaning and contextual meaning. Formal meaning is the information of information theory; contextual meaning relates to extralinguistic features. f. We must distinguish not only between theory and description, but also between description and presentation, being the way the linguist expounds the description. Elaborating on each of the fundamental categories for the theory of grammar, Halliday describes units as pattern carriers. The scale on which units are ranged in the theory is called rank. Structures are ‘the 18

editor’s introduction

ordered repetition of like events that make up the patterns’. There are both primary and secondary structures, distinguished in terms of delicacy, or depth of detail. Whereas class involves ‘the grouping of like events by their occurrence in patterns’, system deals with ‘the occurrence of one rather than another among a number of like events’. To help the reader better understand the application of the categories of grammar, Halliday presents a framework of categories for the description of another very familiar kind of patterned activity, namely, eating a meal. Looking back on this chapter after forty years, Professor Halliday provides some background from his personal history to help readers better understand his very careful concern for assigning things to categories: Struggling with the grammar of Chinese, and then of English, in the conceptual-categorial frameworks which were then available (traditional grammar, linguist’s descriptions of languages, Jespersen and Wang Li, Firth’s system–structure theory, Pike, Fries, Hill, Hockett, etc.), I was constantly finding that the categories were unclear: you would find a label attached to some patch or other, but with no indication of what kind of category it was supposed to be and the whole battery of technical statements never added up to a coherent picture of the whole. I felt I needed to know where I was at any moment and where any descriptive statement that I made fitted in to the overall account.

In a paper appearing in Linguistics in 1963, ‘Class in relation to the axes of chain and choice in language’ (Chapter 3), Halliday discusses the relation of class to structure, the chain axis and class in relation to system, the choice axis. Class is related to two kinds of structure found in language: the place-ordered, in which a limited number of different elements occur non-recursively, and the depth-ordered or recursive structure. Rankshift, for example, refers to a type of recursive structure which cuts across the scale of rank. It is in the fourth paper in this section, ‘Some notes on “deep” grammar’, appearing in the Journal of Linguistics (1966), that Halliday explores more fully the notion of systemic description involving a selection from among the possibilities recognized by the grammar. The relationship between structural and systemic description may be understood in terms of syntagmatic and paradigmatic relations. Appearing in the same issue of the Journal of Linguistics is the last paper in this section, ‘The concept of rank: a reply’, in which Halliday replies to arguments against rank grammar put forward by P. H. Matthews. As Halliday explains, a rank grammar ‘specifies and labels a fixed number of layers in the hierarchy of constituents such that any 19

early papers on basic concepts

constituent can be assigned to one or other of the specified layers, or ranks’. On one point both Halliday and Matthews agree, namely, that rank grammar is a hypothesis about the nature of language. As Halliday argues, it is a hypothesis worth making both for its descriptive advantages and for the questions that follow from it. Attached as the appendix to this section (pp. 127–50) is a description of English, originally prepared in 1964 for a course which Professor Halliday gave at the LSA Summer Institute and which later appeared in Kress (1976).

20

Chapter One

SOME ASPECTS OF SYSTEMATIC DESCRIPTION AND COMPARISON IN GRAMMATICAL ANALYSIS (1957)

1

Descriptive and historical, particular and comparative

The description of a language employs, at the grammatical level as at all other levels, systems of related categories. Such categories as are established in the description of the grammar of a language may be referred to forms of the language itself, or to forms of another language (or other languages) or to non-formal-linguistic concepts. The last of these points of reference is clearly of a different order from the other two: there can be no universal formal-linguistic categories (there might theoretically be categories formally identified as common to all languages studied heretofore, but such identification is not yet a practical possibility), while non-formal-linguistic categories, if they are to figure in the description at all, must be implicitly regarded as universal.1 Unless it is supposed that the sole domain of linguistic science is the study of the evolution of linguistic forms, the improvement of the methods of linguistic description remains one of the tasks of the linguist. In recent decades, striking advances have been made on the basis of General Linguistic theory in the development of descriptive techniques, especially of the first type: the description of a language in terms of categories established within the language itself (Firth 1951). As in the interrelated branches of any discipline, there is a constant mutual contribution between descriptive and historical studies in linguistics; and it is not surprising that many recent developments in the former have been founded on work done in languages where it has not yet been possible to establish a series of phonological-lexical and phonological-morphological First published in Studies in Linguistic Analysis (Special Volume of the Philological Society), 1957, Oxford: Blackwell, pp. 54–67.

21

early papers on basic concepts

correspondences as a basis for genetic groupings; this especially in Britain, where we have a long tradition of the description of the languages of Asia, Africa and the Pacific. In such languages, for historical as well as for descriptive purposes, improved methods have been demanded; but the need for a general theory of description, as opposed to a universal scheme of descriptive categories, has long been apparent, if often unformulated, in the description of all languages. A distinctive contribution of the twentieth century has been the progress towards its achievement. The sort of descriptive statement which has been the fruit of these achievements is too often characterized negatively by opposition to a historical (the synchronic–diachronic dichotomy) or to a comparative statement. This is probably due in part to the very fact that the techniques have been applied to languages which have no ‘history’ (that is, no written document of the past) and even no script, and which have not been satisfactorily organized into families by the comparative historical method. It is not unnatural that what is new in descriptive techniques should have been emphasized by its being contrasted with the historical methods which, in the modern period at least, developed earlier; but, while the comparative historical, like all other scientific methods, will benefit by question and scrutiny, it is no essential part of modern descriptive linguistics that it should reject the achievements of the past, still less that it should deny linguistic history as a field of scientific study. If we consider general linguistics to be the body of theory which guides and controls the procedures of the various branches of linguistic science, then any linguistic study, historical or descriptive, particular or comparative, draws on and contributes to the principles of general linguistics. A simple scheme for ordering the branches of linguistic science controlled by general linguistics might recognize two dimensions: diagrammatically, the horizontal represents the aim of the linguist, descriptive or historical, the vertical the scope of the material, particular (one language text) or comparative (a finite number of language texts greater than one). Thus: Descriptive Particular

Comparative

—

—

—

22

Historical — — — — — — — — — — — —

systematic description and comparison

Any of these types of study may be undertaken with the use of formal linguistic methods: that is, by the methods of what is sometimes called ‘structural linguistics’. (If such a term is to be used, it should perhaps be taken to refer neither to a branch nor to a particular school of linguistics but to that body of general linguistic theory which controls the application of formal linguistic techniques.) This inclusion of historical studies in the field of application of formal linguistic methods rests on the acceptance of the possibility of arranging language texts according to a time–construct. On a completely a-historical view, there will be only one vertical axis, as any number of texts treated in a single statement could only be material for comparative study: in the type of comparatism envisaged by Allen, in which ‘time has no direction and there is no becoming’ (Allen 1953: 106), diachronic has of course no meaning. In this scheme it is envisaged that there might be a difference in the treatment of material consisting of more than one language text according to whether or not the texts are arranged on a time-scale and treated as exponents of the same language at different periods (for example, by the modification of the description in such a way as to present a continuum in which the systematic ordering of the texts corresponds to their ordering in time). A historical study is then formed out of material provided by a series of descriptive studies. The distinction between this method, in which any linguistic form is placed in its descriptive context (systematized) and the systems in which it operates are treated historically, and the type of historical study which ‘structural’ linguistics has excluded – in which is traced the evolution of particular forms without descriptive systematization – might be reflected in the addition to the diagram of a third vertical dimension (in fact a breakaway from the second), perhaps ‘evolutionary’. If one wishes to seek an opposition between ‘structural’ linguistics and comparative philology, it must surely depend not on the acceptance or non-acceptance of history but on the type of historical approach. So may the social anthropologist study either the evolution of kingship in a particular tribe or group of tribes, or the place of the institution of kingship in the structure of a given society at different periods of its history. The “structural” linguist does not handle > or <; but a clear formulation of how these concepts are defined in terms of active participants in linguistic situations would render them less inaccessible to him and so clarify the relation between the two types of historical study. A third horizontal axis in terms of the scope of the material might be the “universal”: the question is whether this is at present, to use 23

early papers on basic concepts

Firth’s expression, ‘on the agenda’, since what is formal when particular or comparative tends to become imaginative when universal. One might then summarize diagrammatically: Descriptive Particular

Comparative

Universal

—

—

Historical — — — — — — — —

—

Evolutionary

— — — —

—— —— —— ——

——

The “structural” linguist, while attempting to develop descriptive methods that are general, in the sense that they are scientific methods universally valid in linguistic description and forming part of general linguistic theory, will be unwilling to claim universality for any formally established category; since, while, for example, it may be convenient in the description of all languages so far studied to give the name ‘verb’ to one class of one unit, this is not a universal statement: the ‘verb’ is redefined in the description of each language. Even if any one category can be identified (presumably contextually) across all languages studied so far – and this is still a long way off: possibly, if a unit ‘word’ can be satisfactorily universalized, the word-class ‘personal pronoun’, as relatively easy to identify contextually, may be the first – such a category is universal only in the limited range of what is and has been, not what might be. At present any universal system of categories must rest on other than formal linguistic criteria: if such can be provided, for example, by mathematics, so much the better – the “structural” linguist will not “reject” it, but he cannot be expected to provide it within his own terms of reference. What “structural” linguistics has done has been to concentrate, partly in redress of the balance in linguistics, on the first of the methods of description outlined in the first paragraph above. It is in fact by the study of systems and structures within the framework of particular description that this body of theory, and the methods derived from it, 24

systematic description and comparison

have earned the name of “structural” linguistics. The term is perhaps most favoured, in Britain at least, by those who do not apply these methods. One sometimes comes across an expression of “disagreement with” “structural” linguistics. But one might as well disagree with nuclear physics – in an age when the atom, no less than the infinitive, can be split. The transference of grammatical categories is a dead horse no longer to be flogged; but it may be noted that it is a formal analysis (and one, moreover, made without reference to forms of other languages) which identifies a Latin word–class “verb” and sets up for it a tense–system. The transference of parts of this system to the Modern Chinese verb (Mullie 1937: 2 ff.) leads to an analysis of quite a different type, in which reference is made either to the forms of another language (for example Latin) or to universal categories (which may or may not be universalized from Latin). It is at least justifiable to ask whether an analysis of Chinese made on the same principles as those applied to Latin, with reference to the forms of Chinese alone, might not be equally valid. Likewise the belief in the impossibility of formal grammatical analysis of some languages has little currency today, though a question concerning the methods to be applied, which was a subject for discussion at the Seventh International Congress of Linguists in 1952, did begin by querying the possibility:2 the attempt, however, needs no justification.

2

Particular description in grammar: units and classes

A complete analysis at the grammatical level (Firth 1935; 1951: 121), in a particular description in which all forms of the language are related to systems set up within the language itself, requires the establishment of grammatical categories, ordered as terms in interrelated systems and having as exponents the substantial (phonic or graphic) segments of the text. Such categories are of two types, which we may call units and classes. The units are defined by interrelation in terms of extent: unlike a system of classes, whose terms are both collectively exhaustive and mutually exclusive, the single system of units forms a hierarchy in descending progression, such that each term is defined as n times the succeeding term, that is, as consisting of one or more members of the succeeding term (exponentially, every exponent of a given unit is statable either as (coextensively) a single exponent or as a sequence of exponents of the unit next in succession). The units are established in the grammar, by formal grammatical criteria, though in the delimitation of exponents of each unit within 25

early papers on basic concepts

the text other criteria, phonetic or graphic, may contribute and may be taken, where to do so is compatible with the general aims of simplicity and comprehensiveness, as the primary or even sole criteria (for example, punctuation or spacing in a written text, features of intonation of pause in a spoken text, all of which then enter into the grammatical description). It is probable that in the description of any language at least two units will be required: these would be such as could be named the sentence and the word.3 A descending order of procedure seems preferable not only for the presentation (where indeed it may be varied for a particular purpose), but also for the analysis of the grammar, where in such a hierarchic progression the classification made at the level of each unit will itself determine the classes that are to be set up for the lower units. One may begin by establishing, and delimiting the exponents of, that unit (which we may call the “sentence”) which, while within the scope of grammatical statement – not so extensive as to be incapable of systematic analysis – is yet enabled to operate as the linguistic action of participants in a situation: which is, in fact, “living language” and constitutes the unit of analysis at the contextual level. In the subsequent establishment and classification of the lower units, the statements made about each unit will be related to values set up in the structure of the higher unit.4 For each unit there will then be set up systems of classes, formally established in the grammar and exhaustive, such that statements may be made which are valid for all exponents of a given unit. These classes are set up independently of structure: that is to say, a unit (for example clause) having been established, it is then classified by reference to various sets of formal criteria (for example presence of, or ordering of, certain formally defined elements); each set of criteria permits the establishment of one system of classes (“clause–classes”). Such mutually independent systems of classes of any unit are referred to as ‘dimensions’: thus one dimension of clause–classes might be the aspect dimension, with a system, say, of two terms, perfective and imperfective. There may be any number of dimensions of classes for each unit, and the system of any dimension may admit of a neutral or ‘unmarked’ term, but each dimension will by itself form an exhaustive system of classification such that every exponent of a given unit may be placed within it. Thus it might be that all words are either red or blue or yellow and all words are either square or round or neutral in shape. Two dimensions of classes are implicit in the taxonomic hierarchy of the unit system: as characteristic of each unit except the lowest there appears the dimension compound / simple, while for each unit except 26

systematic description and comparison

the highest there may be set up the complementary system of free / bound; the free member is exponentially identified with the simple member of the unit next above. How far these dimensions of classes enter into the description will depend on the extent of their determination of other features. Each unit is characterized by certain structures. The structure is a syntagmatic framework of interrelated elements, which are paradigmatically established in the systems of classes and stated as values in the structure.5 The terms in the system of any one dimension may operate as values in the structure of the unit next above: for example, one could state the structure of a sentence in terms of the “aspect” dimension in the clause (if the elements in the sentence structure are stated in terms of the relevant values P and I, then it might be that the structure IP is found to occur but not PI). At the same time the sentence might also be stated in terms of the values F and B the exponents of which are segments of the text as systematized in a free / bound dimension of clause–classes. Similarly if a unit “word” is established there will be dimensions of word–classes the terms in which operate as values in clause structures: given a verb / noun / adverb system of word–classes, it might be that the structure ANV and NAV were admitted in the clause but NVA excluded. The formal criteria by which the classes are established and their exponents identified may be interior to a class (as when a clause is identified as “bound” through the presence within it of a member of a certain word–class) or exterior to it (as when a clause is identified as non-sentential in virtue of its being internal to another clause (in the sentence structure), or a word as a “verb” in virtue of certain categories and combinations to which the class “verb” is uniquely susceptible). The type of grammatical description outlined here may be briefly illustrated from the description of New Chinese (Modern Pekingese). It should be emphasized that what follows is a summary and necessarily incomplete: it is, however, abstracted from a descriptive statement of Modern Pekingese grammar, one among many possible such statements, which does aim to be exhaustive. Four units are recognized: sentence, clause, word, and character,6 in descending progression. All except the first admit distinction into classes of free and bound: the free clause operates as the ‘simple’ term in the system of sentence classes, from which the bound clause is excluded; and so throughout. Other dimensions of classes, of which the majority are established at the two inner levels of clause and word, include the following, each of which is exhaustive at that level:

27

early papers on basic concepts clause–classes: 1. verbal / nominal. 2. ergative / passive / active (= neutral in voice). 3. perfective / imperfective / non-perfective (= neutral in aspect). word–classes: 1. verbal: free verb / pro-verb / prepositive / auxiliary / post-positive. 2. nominal: free noun / pronoun / determinate / auxiliary / postpositive. 3. adverbial: free adverb / conditional / conjunctive / particle. Of the remaining classes, the chief is the substantive / attributive system of word–classes, restricted to free verbs and free nouns and forming the basis for the classification of free verbs as either transitive or intransitive. As an example of the formal criteria employed in the establishment of classes, reference may be made to the dimension of voice in the clause. The ergative voice is characterized by the basic structure (N)vNV, where V = free verb, N = free noun, v = bound verb, the exponent of v in this structure being the prepositive verb baˇ, rarely jia¯ng. The passive is marked by the basic structure (N)vNVa, where a = bound adverb, the exponent of v being the prepositive verb s, that of a the particle di. All clauses with structure other than these are active, or neutral in voice. As an example of a bound word–class, the postpositive verb is placed in the system of word–classes as follows: the verbal word–class is one of a three-term system verbal / nominal / adverbial, and is itself made up of the two terms free / bound, the free forming a system of the two terms free verb / pro-verb, the bound admitting the three terms prepositive / auxiliary / postpositive. The postpositive verb, being bound, operates only in a group. The group (which alternatively might be admitted as a term in the unit system) comprises the three terms verb group / noun group / complex group, characterized as operating in the clause structure with value identical with that of the corresponding free word: the exponent of V in a clause may be free verb, pro-verb, or verb group, while to the complex group corresponds the free adverb. The postpositive verb operates either in the verb group (defined as vV, Vv, or vVv, where v is auxiliary, v postpositive verb), or postpositive complex group (which always follows VN and has structure vN). The postpositive verb admits further classification into three types, two of which operate in the verb group and are distinguished from one another by the admission by one of certain combinatory possibilities excluded by the other, and the third in the postpositive complex group.

Before leaving consideration of the particular description of a language in terms of categories set up within the language itself, I might 28

systematic description and comparison

mention one further instance of the application here of formal methods of analysis. This concerns the determination of the occurrence of a member of a particular grammatical class by formal linguistic but nongrammatical features. The presence of a certain form in a given unit in the syntagm may render probable the occurrence in a subsequent unit of a member of a particular class; this is in fact a form of contextual determination, but it may be stated, partially at least, in terms of only the linguistic (verbal) action in a context of situation – the source of the determination may be found to be in what might be called ‘context of mention’. This requires the two-term system “given” / “new”, the given being that which has been mentioned in the preceding linguistic context. Modern Pekingese shows some correlation between this dimension and word order in the clause, the position of the given being regularly precedent to that of the new. This may not only determine the relative position of words where the basic clause structure is unaffected (for example the relative position of preverbal free noun or pronoun and preverbal adverb) but also permit the prediction of the occurrence of a particular class of clause. A correlation is observable between the occurrence of marked (not active) voice in a clause and the presence of a marked opposition between given and new in the preceding context. Within the categories of marked voice, it frequently appears that in an ergative clause, both nouns are given and the verb is new; while in a passive clause, the verb is given and the (directly) preverbal noun is new. The correlation is more clearly observable if context of mention is taken to include not only repetition of the term mentioned but also reference (for example, pronominal or synonymic). With such “context of reference” we are well on the way towards context of situation;7 in a spoken text, categories of given and new established in the context of situation, though further removed from the level of grammatical description as usually envisaged, may be expected to display such correlation with grammatical categories to at least as great an extent as in a written text, where contextual reference is restricted to the linguistic context (since there is no independent context of situation for any unit less than the whole text). Indeed the contextual categories of given and new may aid in the identification of grammatical categories: a certain category might be identified as the form taken by the given or that taken by the new.

29

early papers on basic concepts

3

Language under description, language of description

While perhaps modern general linguistics would recognize the establishment of categories within the language under description itself as the basis of a particular description, reference to the forms of another language, including the language of description, may be made without infringing the requirements of formal analysis. In any other than a monolingual description of a language text there arises in any case the specific problem of the relation of the forms of the language under description to the language of description. Since language is used to describe language, if in a formal descriptive grammar it is desired to exclude from consideration as far as possible all forms that do not belong to the l.u.d., the nearest approach is that outlined above, where the aim is achieved through the creation of a metalanguage whose terms, whatever the context of their previous usage, are to be taken as defined only with reference to the text under description. All identification of categories either comparative or universal is thereby excluded. It may, however, be desirable in a given instance or for a given purpose to relate the forms of the l.u.d. to particular forms of the l.o.d. The two languages may then be seen to impinge on one another at various points. At one extreme, it is possible to make a descriptive grammar of the l.o.d. using the same procedures as are applied to the l.u.d. and subjecting the two descriptions to a systematic comparison (as envisaged by Allen 1953: 88 ff.). A comparison of the systems of categories would first establish what categories were comparable; the latter would then be compared so as to permit the identification of terms within the system of each category. At the other extreme, in dealing with a limited language text it may be possible to make a complete translation of the text under description, a translation of the type that may be characterized as contextual – one such as to reproduce, as nearly as possible, the creative effect in the given situation of the original. This may be useful where the l.o.d. is not itself described and the linguist does not consider it to be within his terms of reference to make a description such as would permit a systematic comparison. It may then be found that certain categories of the l.u.d. show a regular translation equivalence to certain unsystematized but formally defined elements in the language of the translation (for example, to one term in the system of clause–classes in the l.u.d. might correspond regularly in English translation a verb form in -ing). Such a form of statement, limited though its application, may some30

systematic description and comparison

times be found preferable to reference to undefined categories of the l.o.d. or to extralinguistic universal categories. It may be possible to exemplify the former type of reference to the l.o.d. in a comparison between systems (of the two languages) which, although no complete description has been made of the l.o.d., may yet be identified as comparable on the basis of non-grammatical criteria. The category of personal pronouns is perhaps the most readily susceptible of such treatment, since, even if the category is not grammatically identified in its place in the system of word–classes in the two languages, the terms of the pronominal system may be contextually identified with reference to persons participating in a situation as speaker, addressee and other participant. In a descriptive grammar of Modern Pekingese written in English, one might compare the personal pronoun systems of the two languages, Chinese being taken as the point of reference, as follows: Reference

Chinese

English

1

woˇ

I

33

ta¯men

they

2

nıˇ

22,2(2)3(3)

nıˇmen

3

ta¯

12(2), 13(3), 12(2)3(3)

woˇmen

12(2), 12(2)3(3)

za´men

} { }

you he she we

Contextual reference is as follows: 1, speaker; 2, addressee; 22, addressees; 3, other person; 33, other persons. Since what is under consideration here is the use of comparison in particular description (and not a comparative study as such), no calculation of degree of relationship is necessary. Identification has been permitted between terms operating in systems with different numbers of terms: in any calculation of degree of relationship this point would of course be crucial. Prima facie, such identification would seem justifiable where the criteria for the identification were contextual. The identification of terms by grammatical criteria in a comparative study poses separate problems which are touched upon briefly in the next section. 31

early papers on basic concepts

As a conclusion to the foregoing it may be remarked that reference to the forms of the l.o.d. is invaluable for pedagogical purposes, where its further systematization might prove helpful, as well as in those instances where a single l.o.d. is common to linguists working in widely divergent fields.8

4

Descriptive comparison

In a descriptive comparison there is no implication of genetic relationship; but beyond the application to particular description permitted by comparison of this type (since any l.u.d. may be compared with any l.o.d.) systematic comparison itself connotes a wider purpose, seen initially in the establishment of degrees of relationship. Here the grammatical identification of terms across languages is an essential basis of comparison; and while much can be done on contextual criteria (though not as much as in the establishment of lexical systems for comparative purposes) some formal procedure must be found in order that statement of degrees of relationship may assume a general significance. Possibly, however, we should not impose a strict demarcation between identification of grammatical terms on grammatical and on non-grammatical (e.g. contextual) criteria. Allen, rightly rejecting both terminological identification and identification by translation, considers that identification by formal grammatical criteria (as opposed to the situational–contextual criteria which are available for the semantic identification of lexical items) seems not to be possible (Allen 1953: 99, 100). Some attempt at identification may be made with the Chinese dialects, where speakers of more than one dialect constantly make such identifications in practice, with or without phonological resemblances: for example, from the personal pronouns, a Cantonese speaker equates ne˛i with Pekingese nıˇ and, as readily, ke˛udei with Pekingese ta¯men. This identification in practice demonstrates the¯ contextual basis of the identification; its validity on grammatical criteria may be tested by reference to the place of the class “pronoun” in the system of word–classes. When we find that it is possible to describe both Modern Pekingese and Modern Cantonese in terms of the same units of sentence, clause, word and character; that at the word level we can set up for both a three-term system of classes, verbal, nominal and adverbial, and that one term in the system of nominal word–classes, the auxiliary noun, enters in both into the noun group with identical position and value in the structure, we can regard the class of auxiliary noun in the two dialects as comparable, and are then justified in seeking to identify 32

systematic description and comparison

certain terms in the two systems of auxiliary nouns.9 In this comparison, both grammatical and contextual criteria must be invoked: the Pekingese system has an unmarked term (ge) which is absent from the Cantonese system, which would exclude the grammatical identification of any terms whatsoever, since each marked term in the Pekingese system must be grammatically identified as excluding not only the other marked terms but also the unmarked term, to which latter the Cantonese has no parallel. On contextual criteria, however (including linguistic– contextual, i.e. collocational), the marked terms of the Pekingese system may be regarded as forming a distinct sub-system comparable with the Cantonese system, so that a term can be identified in the use of the auxiliary noun Pekingese zhı¯, Cantonese (graphically and historically identical) ze¯k, we find the following instances (with specimen English equivalents): Pekingese yı´ ge go˘u yı` zhı¯ go˘u (na` ge) go˘u

Cantonese ya¯t ze¯k ga´u ‘a dog’ (in general context or ‘given’) ya¯t ze¯k ga´u ‘one dog’ (in specific context or ‘new’) ze¯k ga´u ‘the dog’

The non-identification of the first and third instances will not prevent the contextual identification of the second if the total spread within that sub-system can be shown to be the same.10 Some such procedure, for the identification of terms in grammatical systems, once established, the field of application of descriptive comparison is seen to be very wide. A comparison of systems in two languages such as Chinese and English, which at first sight seems to be of little purpose beyond its application to the particular description of one of the two, provides material for a formal systemic typology when compared with other such comparisons – for, if with Allen ‘relationship is not of languages but of systems’, the significance of such relationship is that typology is not of languages but of systems. But without going so far from the scope of comparative studies as usually envisaged, one may seek a field for the application of systematic comparison both in the comparison of languages considered to be genetically related and in the study of languages in geographical proximity where there is no material for genetic groupings. With regard to the former, any discussion of the nature and implications of genetic relationship may be avoided by the choice of a simple instance of systematic comparison in the phonology of the dialect areas of Chinese. In the Modern Chinese dialects, in the system of nasal finals in the phonological unit the syllable may have two or three terms, or 33

early papers on basic concepts

there may be only one nasal final. If we compare these systems in certain dialects of the Mandarin, Wu and Yu¨eh dialect groups, we find that it is prosodically linked, and identical in number of terms, with the system of plosive finals or, if the latter is absent, with that of vocalic finals. Cantonese shows the final plosive system p/t/k with prosodically identical nasals m/n/ŋ; in Shanghai, where there is a single final plosive, the glottal closure, there is one nasal final, with varying point of contact; while in Pekingese, where there are no final plosives, the two terms of the final nasal system are prosodically identical with the oral finals, so that n:ŋ::i:u.11 The calculation of the degree of relationship will depend on the number of final systems set up for each dialect and the number of these in which all terms can be identified; but the comparison immediately suggests the historical interconnection between the disappearance of the final plosive system and the elimination of one term from the final nasal system in the dialects where these have occurred. With regard to the study of languages in geographical proximity in an area where no, or only partial, genetic groupings have been established, systematic comparison can determine whether the question of grammatical affinity is to be posed at all, and if so in what form. The superficially apparent affinity among the languages of certain areas has long been the basis of traditional typology. An instance of how it may be demonstrated or disproved, initially as a function of systems, might be found in the grammar of East Asian languages. It seems possible to set up in, for example, Pekingese, Cantonese, Vietnamese, Siamese and Malay systems of nominal word–classes which would permit the identification as comparable of certain terms, including the category discussed with reference to Pekingese and Cantonese above and which may be called the auxiliary noun. If the place of the auxiliary noun in the system of word–classes, and the terms in the various systems of the auxiliary noun, can be compared as between these languages on the basis of adequate criteria, we may determine whether or not a formal description of these languages would reveal for these and other systems anything that could be regarded as grammatical affinity – such as, that is, to exclude from such affinity any systems set up for other languages in other areas. Only if such affinity could be established among a significant number of systems in the grammar of these languages would it be possible to raise further questions of general grammatical affinity and – as yet only an interesting speculation – of geographical gradation. Such a systemic comparison may help to resolve the difficulty that, on the one hand, there appears an obvious but unformulated grammatical similarity among the East Asian languages as contrasted, for example, 34

systematic description and comparison

with the Indo-European languages, while on the other hand there exists the quite justifiable scepticism among linguists either as to how this is to be explained, in comparative historical terms, where there are no correspondences, or as to whether there is anything to be explained (or, in systematic terms, to be stated comparatively) at all.

Notes 1. The study of the “form of the content” by plerematics, as envisaged by Hjelmslev, would set up particular or comparative (i.e. non-universal) categories; but the criteria for such categories, while as yet inadequately defined, would also be such as to be regarded as formal-linguistic. A method of classification of words on the basis of universal categories of relation and description is to be found in the work of Brøndal (1948). 2. Seventh International Congress of Linguists, Preliminary Reports, London, 1952, p. 53: ‘Can a purely formal grammatical analysis be carried out on languages such as Chinese, in which all or nearly all the words are invariable, and if so, on what principle?’ 3. Provided the principle of particular formal description is adhered to, the choice of current terms seems preferable to the creation of new ones. There is, however, no reason why, especially in the initial stage of the process of analysis, completely non-committal terms should not be employed. The practice of some linguists of talking (at least to themselves) about e.g. red words and blue words can equally well be extended, for example, to “strings” and “bits”. 4. Compare the descending order within the levels of linguistic analysis in which meaning may be stated, as suggested by Firth (1951: 121). For an example of the employment of this method of descending analysis in grammatical statement, cf. Robins (1953) and Halliday (1956). 5. As pointed out by Robins (1953: 109), Firth has indicated how system and structure require to be kept separate in General Linguistic theory. I have attempted to follow Firth’s view of the two as distinct but related concepts. 6. The linguistic unit of which the written character is the graphic symbol corresponding to the syllable at the phonological level. The use of the term ‘character’ follows the Chinese practice of calling both the linguistic and the graphic units by the single name tzuˇ (zı`). 7. The concept of ‘context of situation’ here intended is as developed by Firth in linguistic analysis, where it ‘is best used as a suitable schematic construct to apply to language events’ and should be regarded as ‘a group of related categories at a different level from grammatical analysis but rather of the same abstract nature’ (Firth 1952: esp. 7). 8. Professor Allen has kindly drawn my attention to the paper entitled

35

early papers on basic concepts ‘Transfer grammar’ (Chavarria-Aguilar 1954), which is an interesting discussion of structural comparison in language teaching. See also the important article by Harris (1954): ‘Transfer grammar’. 9. In the same way, a comparison of case–systems as envisaged by Allen must depend on the prior identification of the noun in the two languages compared, whether or not on the criterion of its place in the system of word–classes. 10. It may be argued that the easy identification on graphical and phonological grounds of the two words ‘dog’, on the collocation with which the identification of zhı¯ / ze¯k depends, is in fact (or would be, when applied outside the dialects of a single language) a falling back on vague translation criteria. I should, however, maintain that this is where the contextual criteria enter into the picture: if the terms to be compared cannot themselves be definitely identified by precise contextual reference, then they may be formally related (for example by collocation) to other forms which can. A more difficult instance would be the terms in the aspect systems (of clause–classes) in Pekingese and Cantonese; here again, once the systems are shown to be comparable, one would proceed by seeking to identify particular terms through their collocation with contextually identifiable forms. 11. In Pekingese the prosodic identity is in terms of “y and w”: in the determination of the quality of the vowel, final i = n, u = ŋ . Cf. J. R. Firth (1948: esp. 136, reference to Firth’s ‘The Chinese monosyllable in a Hunanese dialect (Changsha)’) with B. B. Rogers: “The prosodic diacritica (of the Hunanese syllable) included . . . yotization and labiovelarization, symbolized by y and w”.

36

Chapter Two

CATEGORIES OF THE THEORY OF GRAMMAR (1961)

There have been in the main two approaches to description in modern linguistics: the “textual” and the non-textual or, for want of a better word, “exemplificatory”. More recently a third has been added, primarily in grammar but lately also in phonology, the “generative” (strictly “transformative–generative”, since generation does not presuppose transformation). Some linguists have gone so far as to suggest that transformative generation should replace other types of description1 as a linguistic method of making statements about language.2 Others, myself included, feel that all three approaches have a fundamental place in linguistics; that they do different things, and that the third is a valuable supplement to the first two.3 Description is, however, not theory. All description, whether generative or not, is related to General Linguistic theory; specifically, to that part of General Linguistic theory which accounts for how language works. The different types of description are bodies of method which derive from, and are answerable to, that theory. Each has its place in linguistics, and it is a pity to deny the value of textual description (which is appropriate, for example, in “stylistics”, the linguistic study of literature) just because certain of the methods used in description are found to be inadequate. My purpose in writing this paper is to suggest what seem to me to be the fundamental categories of that part of General Linguistic theory which is concerned with how language works at the level of grammar, with brief reference to the relations between grammar and lexis and between grammar and phonology. The theory sketched out here First published in Word, 1961, 17(3), pp. 241–92.

37

early papers on basic concepts

derives most of all from the work of J. R. Firth.4 At the same time I do not of course imply that I think Professor Firth would necessarily have found himself in accord with all the views expressed, which in some places depart from his own; nor do I underestimate the debt to my present colleagues and the many others whose work I have, obviously, drawn on.5 No excuse is needed, I think, for a discussion of General Linguistic theory. While what has made linguistics fashionable has been, as with other subjects, the discovery that it has applications, these applications rest on many years of work by people who were simply seekers after knowledge. It would not help the subject if the success of these applications led us into thinking that the theoretical problems were solved and the basic issues closed.

1

Starting-point

It will perhaps be helpful if the point of departure is first made clear. The following is a summary of what is taken as “given” for the purposes of this paper. 1.1 One part of General Linguistic theory is a theory of how language works. It is from this that the methods of Descriptive Linguistics are derived. 1.2 The relevant theory consists of a scheme of interrelated categories which are set up to account for the data, and a set of scales of abstraction which relate the categories to the data and to each other. The data to be accounted for are observed language events, observed as spoken or as codified in writing, any corpus of which, when used as material for linguistic description, is a “text”.6 1.3 Description consists in relating the text to the categories of the theory. The methods by which this is done involve a number of processes of abstraction, varying in kind and variable in degree. It is the theory that determines the relation of these processes of abstraction to each other and to the theory.7 1.4 The theory requires that linguistic events should be accounted for at a number of different levels: this is found to be necessary because of the difference in kind of the processes of abstraction involved.8 38

categories of the theory of grammar

1.5 The primary levels are form, substance and context. The substance is the material of language: phonic (audible noises) or graphic (visible marks). The form is the organization of the substance into meaningful events: meaning is a concept, and a technical term, of the theory (see below, 1.8). The context is the relation of the form to non-linguistic features of the situations in which language operates, and to linguistic features other than those of the item under attention: these being together “extratextual” features. 1.6 The complete framework of levels requires certain further subdivisions and additions, and is as follows: (a) Substance may be either phonic or graphic. (b) If substance is phonic, it is related to form by phonology. (c) If substance is graphic, it is related to form by orthography (or graphology),9 either (i) if the script is lexical, then directly, or (ii) if the script is phonological, then via phonology. (d) Form is in fact two related levels, grammar and lexis. (e) Context is in fact (like phonology) an interlevel relating form to extratextual features. 1.7 The study of phonic substance belongs to a distinct but related body of theory, that of General Phonetics. Since phonology relates form and phonic substance, it is the place where linguistics and phonetics interpenetrate. Linguistics and phonetics together make up “the linguistic sciences”.10

Figure 1 Levels of language

39

early papers on basic concepts

1.8 Language has formal meaning and contextual meaning. Formal meaning is the “information” of information theory, though (i) it can be stated without being quantified and was in fact formulated in linguistics independently of the development of information theory as a means of quantifying it,11 and (ii) formal meaning in lexis cannot be quantified until a method is found for measuring the information of non-finite (“open”) sets (see below, 2.1 and 8.2). The formal meaning of an item is its operation in the network of formal relations. 1.9 Contextual meaning, which is an extension of the popular – and traditional linguistic – notion of meaning, is quite distinct from formal meaning and has nothing whatever to do with ‘information’.12 The contextual meaning of an item is its relation to extratextual features; but this is not a direct relation of the item as such, but of the item in its place in linguistic form: contextual meaning is therefore logically dependent on formal meaning.13 1.10 It follows from 1.8 and 1.9 that, in description, formal criteria are crucial, taking precedence over contextual criteria; and that the statement of formal meaning logically precedes the statement of contextual meaning.14 1.11 Finally, it is necessary to distinguish not only between theory and description but also between description and presentation. Presentation, the way the linguist expounds the description, varies with purpose, and relative merit is judged by reference to the specific purpose intended. Description depends on the theory; theoretical validity is demanded, and relative merit is judged by reference to comprehensiveness and delicacy.15

2

Grammar

2.1 Grammar is that level of linguistic form at which operate closed systems.16,17 Since a system is by definition closed, the use of the term “closed” here is a mnemonic device; but since “system” alone will be used as the name of one of the four fundamental grammatical categories (see below, 6) it is useful to retain “closed system” when referring to the system as the crucial criterion for distinguishing grammar from lexis. A closed system is a set of terms with these characteristics: 40

categories of the theory of grammar

(a) The number of terms is finite: they can be listed as A B C D, and all other items E . . . are outside the system. (b) Each term is exclusive of all the others: a given term A cannot be identical with B or C or D. (c) If a new term is added to the system this changes the meaning of all the others.18 Any part of linguistic form which is not concerned with the operation of closed systems belongs to the level of lexis. The distinction between closed system patterns and open set patterns in language is in fact a cline; but the theory has to treat them as two distinct types of pattern requiring different categories. For this reason General Linguistic theory must here provide both a theory of grammar and a theory of lexis, and also a means of relating the two. A description depending on General Linguistic theory will need to separate the descriptions of the two levels both from each other and from the description of their interrelations. This paper is primarily concerned with the theory of grammar, though reference will be made to lexis at various points. 2.2 The fundamental categories for the theory of grammar are four: unit, structure, class and system. These are categories of the highest order of abstraction: they are established, and interrelated, in the theory. If one asks: “why these four, and not three, or five, or another four?”, the answer must be: because language is like that – because these four, and no others, are needed to account for the data: that is, to account for all grammatical patterns that emerge by generalization from the data. As the primary categories of the theory, they make possible a coherent account of what grammar is and of its place in language, and a comprehensive description of the grammars of languages, neither of which is possible without them. Each of the four is specifically related to, and logically derivable from, each of the others. There is no relation of precedence or logical priority among them. They are all mutually defining: as with theoretical categories in general, “definition” in the lexicographical sense is impossible, since no one category is defined until all the others are, in the totality of the theory.19 The order chosen here for exposition is therefore simply that which seemed the easiest: namely the order in which they are listed above. The relation of these categories to each other and to the data involve three distinct scales of abstraction, those of rank, exponence and delicacy; these are considered separately (see below, 7) but have also to be 41

early papers on basic concepts

referred to in connection with the categories. In discussing these I have used the terms “hierarchy”, “taxonomy” and ‘cline’ as general scale– types. A hierarchy is taken to mean a system of terms related along a single dimension which must be one involving some form of logical precedence (such as inclusion).20 A taxonomy is taken to mean a special type of hierarchy, one with two additional characteristics: (i) there is a constant relation of each term to the term immediately following it, and a constant reciprocal relation of each to that immediately preceding it; and (ii) degree is significant, so that the place in order of each one of the terms, statable as the distance in number of steps from either end, is a defining characteristic of that term.21 A cline resembles a hierarchy in that it involves relation along a single dimension; but instead of being made up of a number of discrete terms a cline is a continuum carrying potentially infinite gradation. 2.3 In this view of linguistics description is, as already emphasized, a body of method derived from theory, and not a set of procedures. This has one important consequence. If description is procedural, the only way of evaluating a given description is by reference to the procedures themselves: a good description is one that has carried out the right procedures in the right order, but for any more delicate evaluation external criteria have to be invoked. Moreover, every language has to be treated as if it was unknown, otherwise procedural rules will be violated; so the linguist has to throw away half his evidence and a good few of his tools. A theory on the other hand provides a means for evaluating descriptions without reference to the order in which the facts are accounted for. The linguist makes use of all he knows and there is no priority of dependence among the various parts of the description. The best description is then that which, comprehensiveness presupposed, is maximally grammatical: that is, makes maximum use of the theory to account for a maximum amount of the data. Simplicity has then to be invoked only when it is necessary to decide between fewer systems with more terms and more systems with fewer terms; and since both information theory and linguistic intuition favour the latter even this preference might be built in to the theory.22

3

Unit

3.1 Language is patterned activity. At the formal level, the patterns are patterns of meaningful organization: certain regularities are 42

categories of the theory of grammar

exhibited over certain stretches of language activity. An essential feature of the stretches over which formal patterns operate is that they are of varying extent. Abstracting out those of lexis, where the selection is from open sets, we find that the remaining, closed system, patterns are associated with stretches that not only are of differing extent but also appear as it were one inside the other, in a sort of one-dimensional Chinese box arrangement. Since language activity takes place in time, the simplest formulation of this dimension is that it is the dimension of time, or, for written language, of linear space: the two can then be generalized as “progression” and the relation between two items in progression is one of “sequence”. But there is a danger here. It is obvious that absolute measurements of linear progression belong to language substance (where one may be interested in the number of seconds, or possibly even the number of inches, occupied by an utterance). What is less obvious is that the whole dimension of progression in fact belongs to substance, and that the stretches which carry grammatical pattern – or rather the members of that abstract category that we set up to account for these stretches – have to be ranged on a dimension of which linear progression is only a manifestation in substance: a dimension we may call “order”.23 By implication, this allows that in any given instance sequence may not manifest order, or that order may have other manifestations; even if this never happens, the distinction is necessary until such time as it is shown that the theory does not need to make provision for its happening. In fact it does happen: sequence is a variable, and must be replaced in the theory by the more abstract dimension of order.24 3.2 The category set up to account for the stretches that carry grammatical patterns is the unit. The units of grammar form a hierarchy that is a taxonomy. To talk about any hierarchy, we need a conversational scale; the most appropriate here might seem that of size, going from “largest” to “smallest”; on the other hand size is difficult to represent in tables and diagrams, and may also trap one into thinking in substantial terms, and a vertical scale, from “highest” to “lowest”, has advantages here. For the moment we may use both, eventually preferring the latter. The relation among the units, then, is that, going from top (largest) to bottom (smallest), each consists of one, or of more than one, of the unit next below (next smaller). The scale on which the units are in fact ranged in the theory needs a name, and may be called rank. “Consists of”, like “unit” and “rank”, also belongs to the theory: its 43

early papers on basic concepts

realization in form varies between and within languages, and is stated of course in description.25 The possibilities are sequence, inclusion and conflation. Thus if in a given instance a unit of one rank consists of two units of rank next below, these may appear in form as one following, interrupting, or overlaying the other. Three further points about the rank relation need to be clarified. First, the theory allows for downward rankshift: the transfer of a (formal realization of a) given unit to a lower rank. Second, it does not allow for upward rankshift. Third, only whole units can enter into higher units. Taken together these three mean that a unit can include, in what it consists of, a unit of rank higher than or equal to itself but not a unit of rank more than one degree lower than itself; and not, in any case, a part of any unit.26 3.3 The number of units in the hierarchy is a feature of the description. It varies from language to language, but is fixed by the description for each language, or rather for each describendum or “e´tat de langue”. The possibility of there being only one is excluded by the theory, since a hierarchy cannot be composed of one member. It is, however, theoretically possible to conceive of a language having only two, and an artificial language could be constructed on these lines (whereas it would not be possible to construct an artificial language having only one unit). English grammar, as far as it has been studied to date, seems to require five, though further, statistical, work on grammar might yield at least one more. No special status, other than that27 presupposed by rank, is assigned by grammatical theory to any one unit. Since in any case only two, as a minimum, are required, only two would be available for special status. As it happens we can assign special status to two grammatical units by reference to other levels on a “more / less” basis. There will always be one unit which, more than any other, offers itself as an item for contextual statement because it does the language work in situations: so it might as well always have the same name: sentence. There will be another unit, always lower in rank, which more than any other (but again not exclusively) enters into another type of pattern and thus offers itself as an item for lexical statement;28 this we may as well always call the word. So, in grammatical theory, all languages have at least two units; in description, all languages have sentences and all languages have words – but the “sentenceness” of the sentence and the “wordness” of the word do not derive from the theory of grammar. Various names are available for those above, below or in between 44

categories of the theory of grammar

when they turn up. For English, for the two units between sentence and word the terms clause and phrase are generally used. It is at the rank of the phrase that there is most confusion – because there are here the greatest difficulties – in the description of English; one reason is that in English this unit carries a fundamental class division (see below, 5), so fundamental that it is useful to have two names for this unit in order to be able to talk about it: I propose to call it the group, but to make a class distinction within it between group and phrase. Below the word, English has one unit, called by the general name for the unit of lowest rank, the morpheme.29 So in the description of English the sentence30 consists of one or more complete clauses, the clause of one or more complete groups, the group of one or more complete words and the word of one or more complete morphemes. The descriptive meaning of “consists of”, and the possibilities of rankshift (including recursive rankshift), are stated as and where applicable. One distinction that is often useful is between a member of a unit that consists of only one member of the unit next below and one that consists of more than one; the former may be called simple and the latter compound, but if this is done the terms must be kept rigorously to this, and no other, use.31 3.4 The theory requires that each unit should be fully identifiable in description. This means that, if the description is textual, every item of the text is accounted for at all ranks, through the various links of the exponence chain which involve, of course, the remaining theoretical categories. If the description is exemplificatory, exactly the same is implied, except that the description proceeds from category to exponent instead of from exponent to category. It will be clear from the discussion in the next sections that there can be no question of independent identification of the exponents of the different units, since criteria of any given unit always involve reference to others, and therefore indirectly to all the others. A clause can only be identified as a clause if a sentence can be identified as a sentence and a group as a group, and so on up and down the line. For this reason description is not and can never be unidirectional: it is essential to “shunt”, and “shunting” is a descriptive method that is imposed on description by theory.

45

early papers on basic concepts

4

Structure

4.1 The unit being the category of pattern-carrier, what is the nature of the patterns it carries? In terms once again of language as activity, and therefore in linear progression, the patterns take the form of the repetition of like events. Likeness, at whatever degree of abstraction, is of course a cline, ranging from “having everything in common” to “having nothing in common”. The commonplace that no two events are ever identical, that the same thing can never happen twice, is of no relevance whatever to linguistics; as soon as description starts, however little the generalization involved, absolute identity is a necessary hypothesis, which is then built into the theory, as one endpoint of the likeness cline. Likeness, including absolute identity, is of course redefined for each level and each category. In grammar the category set up to account for likeness between events in successivity is the structure.32 If the relation between events in successivity is syntagmatic, the structure is the highest abstraction of patterns of syntagmatic relations. The scale used for talking about it, and for its graphic display, will most naturally be the orthographic scale: to those of us brought up on the roman alphabet this happens to run horizontally from left to right, which is enough reason for adopting this version of the scale. But, as in the case of the unit, it must be stressed that linear progression itself is a feature of substance. A structure is made up of elements which are graphically represented as being in linear progression; but the theoretical relation among them is one of order. Order may, but does not necessarily, have as its realization sequence, the formal relation carried by linear progression; sequence is at a lower degree of abstraction than order and is one possible formal exponent of it.33 4.2 A structure is thus an arrangement of elements ordered in places. Places are distinguished by order alone: a structure XXX consists of three places. Different elements, on the other hand, are distinguished by some relation other than that of order: a structure XYZ consists of three elements which are (and must be, to form a structure) placeordered, though they can be listed (X, Y, Z) as an inventory of elements making up the particular structure.34 A structure is always a structure of a given unit. Each unit may display a range of possible structures, and the only theoretical restriction is that each unit must carry at least one structure that consists of more than one place.35 Each place and each element in 46

categories of the theory of grammar

the structure of a given unit is defined with reference to the unit next below. Each place is the place of operation of one member of the unit next below, considered as one occurrence. Each element represents the potentiality of operation of a member of one grouping of members of the unit next below, considered as one item–grouping.36 It follows from this that the lowest unit has no structure;37 if it carried structure, there would be another unit below it. 4.3 In description, structures are stated as linear arrangements of symbols, each symbol (occurrence) standing for one place and each different symbol (item) standing for one element. Since elements of structure “exist” only at this degree of abstraction, the relation “stands for” means simply “is shorthand for”, like that of an initial: “ ‘U’ stands for ‘United’.” In a few cases traditional names exist which can usefully serve as names for elements of structure, with the initial letter as the descriptive symbol. In the statement of English clause structure, for example, four elements are needed, for which the widely accepted terms subject, predicator, complement and adjunct are appropriate.38 These yield four distinct symbols, so that S, P, C, A would be the inventory of elements of English clause structure. All clause structures can then be stated as combinations of these four in different places: SAPA, ASP, SPC, ASPCC, etc. For one type of group we have the names modifier, head, qualifier, giving an inventory M, H, Q: here, if the total range of possible structures is H, MH, HQ, MHQ, these possibilities can be stated in a single formula, where parentheses indicate “may or may not be present”, as (M) H (Q).39 In other cases, no names come ready to hand; names can be imported or coined, or arbitrary symbols chosen – colours, for example, have advantages over letters in presentation, though there are not enough of them and they have to be redefined in description for each unit. It is tempting sometimes to derive the symbols from the name given to the grouping of members of the unit next below which operates at the given element (as if one were to put V instead of P because what operates at P is the verbal group); but it is important to avoid identifying this grouping, which belongs to a different category as well as a different rank, with the element itself – therefore if this method is to be used at all it must be used all the time and a statement made to cover it.40 There are some instances where an element of structure is identified as such solely by reference to formal sequence: where the element is defined by place stated as absolute or relative position in sequence. It 47

early papers on basic concepts

is useful to indicate that here sequence is so to speak built in to structure, and this can be shown by an arrow placed over the symbols for the elements concerned. For example, in English clause structure it 씮 is a crucial criterion of the element S that it precedes P in sequence: 씮 씮 씮 structures can be stated as SPCA, SAPA, ASP, etc.41 This displays the contrast between this situation, where S is crucially defined by position relative to P, and realized sequences of elements which are not, however, defined by sequence, which may be indicated by simple linearity of the symbols.42 4.4 In the consideration of the places and elements of structure of each unit, which of course vary from language to language and from unit to unit within a language, a new scale enters, that of delicacy (see below, 7.4). This is depth of detail, and is a cline running from a fixed point at one end (least delicate, or primary) to that undefined but theoretically crucial point (probably statistically definable) where distinctions are so fine that they cease to be distinctions at all, like a river followed up from the mouth, each of whose tributaries ends in a moorland bog. Primary structures are those which distinguish the minimum number of elements necessary to account comprehensively for the operation in the structure of the given unit of members of the unit next below: necessary, that is, for the identification of every item at all ranks. (M)H(Q), and the various possible combinations of S, P, C, A, are primary structures: one cannot account for all words in group structure, or all groups in clause structure, with fewer than these elements or places. Subsequent more delicate differentiations are then stated as secondary structures. These are still structures of the same unit, not of the unit next below; they take account of finer distinctions recognizable at the same rank.43 Rank and delicacy are different scales of abstraction: primary group structures differ in rank from primary clause structures, but are at the same degree of delicacy; while primary and secondary clause structures differ in delicacy but not in rank. As the description increases in delicacy the network of grammatical relations becomes more complex. The interaction of criteria makes the relation between categories, and between category and exponent, increasingly one of “more / less” rather than “either / or”. It becomes necessary to weight criteria and to make statements in terms of probabilities. With more delicate secondary structures, different combinations of elements, and their relation to groupings of the unit next below, have to be stated as more and less probable.44 The concept of 48

categories of the theory of grammar

“most delicate grammar”, and its relation to lexis, is discussed below (see 6.3 and 7.4); but the “more / less” relation itself, far from being an unexpected complication in grammar, is in fact a basic feature of language and is treated as such by the theory. It is not simply that all grammar can be stated in probability terms, based on frequency counts in texts: this is due to the nature of a text as a sample. But the very fact that we can recognize primary and secondary structures – that there is a scale of delicacy at all – shows that the nature of language is not to operate with relations of “always this and never that”. Grammatical theory takes this into account by introducing a special scale, that of delicacy, to handle the improbability of certainty; this frees the rest of the theory from what would otherwise be the weakening effect of this feature of language. The category of structure, for example, is the more powerful because it can be used to state the patterns of a given unit comprehensively at the primary degree without the assumption that it has accounted for all the facts.

5

Class

5.1 The structure is set up to account for likeness between events of the same rank, and it does so by referring them to the rank next below. To one place in structure corresponds one occurrence of the unit next below, and at each element operates one grouping of members of the unit next below. This means that there will be certain groupings of members of each unit identified by restriction on their operation in structure. The fact that it is not true that anything can go anywhere in the structure of the unit above itself is another aspect of linguistic patterning, and the category set up to account for it is the class. The class is that grouping of members of a given unit which is defined by operation in the structure of the unit next above. It accounts for a paradigmatic relation, being a grouping of items “at risk” under certain conditions. It is related primarily to elements of structure: the first degree of classification yields classes which stand in one / one relation to elements of primary structures, and these we may call primary classes. 5.2 Class, like structure, is variable in delicacy. Clearly, in the first place, more delicate classes are the product of more delicate structures: in fact, “secondary” classes are derived from structure in two ways. Firstly the same element at different places in structure may yield distinct secondary classes. If a given unit has primary structures XY, 49

early papers on basic concepts

XYZ, YZ and XYZY, the primary classes of the unit next below are “class operating at X”, “class operating at Y” and “class operating at Z”. If, however, there is a further restriction such that in XYZY, which will now be (secondarily) rewritten XYaZYb, only a section of the members of the class at Y can operate at Ya and only another section (not necessarily mutually exclusive) at Yb, this yields as secondary classes “class operating at Ya” and “class operating at Yb”. Secondly, with increased delicacy the elements of primary structure will be differentiated into secondary elements. A primary structure generalized as X . . . nYZ, of which XXXXYZ is an instance, shows a generalized relation of X to (say) Y; but there may be internal relations within X . . . n such that XXXX is rewritten pqrs. These will yield secondary classes “class operating at p”, “class operating at q”, etc. In the second place, more delicate classes appear whenever a restriction is found which differentiates among the members of a primary class. There may be a relation of mutual determination, or “concord”, between two classes; each divides into two sections such that a member of one section of one class is always accompanied by a member of one section of the other class. Thus if the primary class at X is 1 and that at Y is 2, a structure XY must have as its exponent either 1.1+2.1 or 1.2+2.2. Secondary classes arrived at in this way in description may be referred to distinctively as sub-classes, to indicate that they are derived by differentiation from primary classes without reference to secondary structures; but it is important to state that there is no theoretical difference here. The relation between structure and class is a two-way relation, and there is no question of “discovering” one “before” the other. In any given instance there may be descriptive reasons for stating the one without the other; but all structures presuppose classes and all classes presuppose structures. 5.3 What is theoretically determined is the relation between structure and class on the one hand and unit on the other. Class, like structure, is linked to unit: a class is always a class of (members of) a given unit: and the class–structure relation is constant – a class is always defined with reference to the structure of the unit next above, and structure with reference to classes of the unit next below. A class is not a grouping of members of a given unit which are alike in their own structure. In other words, by reference to the rank scale, classes are derived “from above” (or “downwards”) and not “from below” (or “upwards”). The distinction between downward and upward movement on the 50

categories of the theory of grammar

rank scale is important in grammar, but it is a mistake to raise it to the status of a choice between different theories, which it is not. The “formal / functional” dichotomy is one of those which linguistics is better rid of;45 it is misleading to say even that classes are functionally determined, since they are set up with reference to the form of the unit next above – the whole description is both formal and functional at the same time, and “function” is merely an aspect of form. The distinction does, however, need a name, and this seems the best use for the terms “syntax” and “morphology”. Traditionally these terms have usually referred to “grammar above the word” (syntax) and “grammar below the word” (morphology); but this distinction has no theoretical status.46 It has a place in the description of certain languages, “inflexional” languages47 which tend to display one kind of grammatical relation above the word (“free” items predominating) and another below the word (“bound” items predominating). But it seems worthwhile making use of “syntax” and “morphology” in the theory, to refer to direction on the rank scale. “Syntax” is then the downward relation, “morphology” the upward one; and both go all the way. We can then say, simply, classes are syntactical and not morphological.48 (A terminological alternative is to talk of “syntactic classes” and “morphological classes”, but this has the disadvantage that it destroys the consistency in the use of the term “class”, no longer restricting it to a category of unique theoretical status. When a name is needed for morphological groupings, groupings of items on the basis of likeness in their own structure, the term “paradigm” is available.49) 5.4 In description the term “class” covers primary and secondary classes. It is often unnecessary to specify; but it is useful to state primary classes first, since these form the link between elements of structure and more delicate classes.50 For the different classes of each unit, many names are available, especially at word rank – though it is precisely the long history of terms like “verb” and “noun” which reinforces their need for rigorous redefinition in the theory. In other cases names can often be found, especially where structural restrictions permit class identification between units: if certain word–classes only (or predominantly) operate in one class of group and others only in another, one can talk of verbal group and verb (= “verb word”), nominal group and noun (word) and so on. At the same time it is safer not to allow complete terminological identity between units: “verb” alone should not serve as the name both for a class of group and for a class of word. In the use of symbols for classes, figures have the advantage of 51

early papers on basic concepts

avoiding confusion with elements of structure. This is not only theoretically desirable, because of their different status; it has descriptive value in that the theoretical one / one relation between elements and classes allows for instances where two different elements of structure, standing in different relation to each other or to a third, yield primary classes the membership of which is coextensive: these then form a single primary class derivable simultaneously from two elements of structure.51 If letters are used for elements of structure, and figures for classes, the relation between the two can be demonstrated by the use of a colour code.

6

System

6.1 Up to this point the theory has accounted for three aspects of formal patterning: the varying stretches that carry patterns, the ordered repetition of like events that makes up the patterns and the grouping of like events by their occurrence in patterns. What remains to be accounted for is the occurrence of one rather than another from among a number of like events. The category set up for this purpose is the system.52 This falls under the general definition of system given above (2.1). But this does not yet state its place in grammatical theory, its relation to the other fundamental categories. The class is a grouping of items identified by operation in structure: that is, what enters into grammatical relations of structure is not the item itself considered as a formal realization53 but the class, which is not a list of formal items but an abstraction from them. By increase in delicacy, the primary class is broken down into secondary classes of the same rank. This set of secondary classes now stands in the relation of exponent to an element of primary structure of the unit next above. This gives a system of classes. If class 1 is the primary class (say of the group) operating at X in (clause) structure, and this has secondary classes 1.1, 1.2 and 1.3, then 1.1, 1.2 and 1.3 form a system of classes operating at X. X is now shown to presuppose a choice – a choice that is implied by the nature of the class (as a grouping of items) but that is displayed first still in abstraction, by reference to the category of class itself.54 6.2 Systems of secondary classes thus allow the description to remain at a high degree of abstraction while displaying at each step, each increase in delicacy, a more finely differentiated range of choice. This 52

categories of the theory of grammar

is the value of the concept of “sub-class” (above, 5.2), since wherever a choice among a finite number of mutually exclusive possibilities is found to occur within a class one can recognize a system whose terms have the nature and degree of abstraction of the “class”: their relation to secondary elements of structure is implied but need not be stated. Thus the system provides what is, in the order in which the categories are presented here, the final requisite for the linking of the categories to the data.55 Through the system, in one of two ways (below, 6.3), the description can now account for the formal exponents, the items identified in linguistic form – and through them is linked, by description at other levels, to the ultimate exponents in substance. But there is a crucial point here. Any category can be linked directly to its exponents:56 a given formal item can be at one and the same time, and in the same sense, an exponent of a unit, a structure, an element of structure, a class and a term in a system.57 At the same time the aim of grammar is to stay in grammar: to account for as much as can possibly be accounted for grammatically, by reference to the categories of grammatical theory. This, since it implies maximum generalization and abstraction, means that one proceeds from category to exponent by the longest route that is compatible with never going over the same step twice. Here it is important to avoid theoretical confusion between the scales of exponence and rank. Since the relation of class to structure is syntactical (as defined above, 5.3), if one derives classes from structures then “remaining in grammar” means moving step by step down the rank scale until the lowest unit is reached.58 So it might appear as if going down the rank scale is the same thing as going down the exponence scale. But it is not. The sentence stands in exactly the same relation to its exponents as does the morpheme; one can move over (at right angles, so to speak) at any rank, and the categories of class, structure and system remain at the same degree of abstraction whatever unit they are associated with. The deriving of class from structure is of course merely one way of stating a theoretical relation which could equally be viewed from the other end;59 in this case “remaining in grammar” would mean going up the rank scale and it would appear (equally erroneously) that it was only at the rank of the sentence that one reached the exponents. 6.3 There comes a point, however, when one is forced out to the exponents; and this happens in one of two ways. In the first case the 53

early papers on basic concepts

description yields a system in which the formal exponents themselves operate as terms. Here we have gone all the way in grammar; the formal items are grammatically contrastive (and do not belong in the dictionary60). In the second case the description yields a class where no further breakdown by grammatical categories is possible, a class whose exponents make up an open set. Here we must leave grammar; the relations between the exponents must be accounted for as lexical relations. Neither of these endpoints of grammar is restricted to the rank of any one unit. The exit to lexis tends to be associated predominantly – but probably never uniquely – with one unit, which for this reason is called in description the “word” (above, 3.3). The system of exponents also tends to operate at the lower end of the rank scale; but this, although predictable as economy of resources, is not a theoretical restriction: the rank distribution of formal exponents in systems is a descriptive feature which, in a five-unit description, may be expected to involve at any rate the three lower units. The theoretical place of the move from grammar to lexis is therefore not a feature of rank61 but one of delicacy. It is defined theoretically as the place where increase in delicacy yields no further systems; this means that in description it is constantly shifting as delicacy increases. The grammarian’s dream is (and must be, such is the nature of grammar) of constant territorial expansion. He would like to turn the whole of linguistic form into grammar, hoping to show that lexis can be defined as “most delicate grammar”. The exit to lexis would then be closed, and all exponents ranged in systems. No description has yet been made so delicate that we can test whether there really comes a place where increased delicacy yields no further systems: relations at this degree of delicacy can only be stated statistically, and serious statistical work in grammar has hardly begun. It may well be that the nature of language is such that this “most delicate grammar” will evaporate in distinctions which are so slenderly statistical that the system has, in effect, been replaced by the open set. For the moment it seems better to treat lexical relations, where even the identification of the items concerned by grammatical means is extremely complex, as on a different level, and to require a different theory to account for them. 6.4 A brief illustration, from English grammar, of the four fundamental categories. The exponent of the element S in primary clause structure is the primary class nominal of the unit group. The primary 54

categories of the theory of grammar

structure of the nominal group is (M . . .n)H(Q . . .n). The primary classes of the unit word operating at H and M are respectively the noun and the pre-noun. The element M can be broken down into secondary structures composed of any combination of the elements D, O, E in that order, whose exponents are the secondary classes deictic, numerative and adjective of the word. Deictics include a number of systems whose terms are the formal exponents themselves, with further secondary classes separating, for example, “all / both / such / half ” (class 1, at Da) from “a / the / some, etc.” (class 2, at Db) and various sub-classes involving concord. Adjectives likewise include a number of secondary classes, but in most of these the exponents form open sets, and for further treatment of these grammar hands over to lexis.62

7

Rank, exponence and delicacy

7.1 In relating the categories to each other and to their exponents, the theory needs to operate with three scales of abstraction, the scales of rank, exponence and delicacy. These need to be kept apart from each other, and also from two other things: regressive structures,63 which are handled in the description by rankshift, and the range of levels by which grammar is distinguished from other aspects of language in the first place. The separation of levels has been taken as a starting-point, and it is not the purpose of this paper to explore this part of the theory. My own view is that, whatever is decided for presentation, which will vary with purpose, both in theory and in description it is essential to separate the levels first and then relate them. The theoretical reason is that different kinds of abstraction are involved, and therefore different categories. In description, the attempt to account for the data at all levels at once results in a failure to account for them fully at any level. If one rejects the separation of levels and wishes, for example, to combine grammatical and phonological criteria to yield a single set of units, the description becomes intolerably complex. The attempt to delimit grammatical units on phonological criteria in English has recourse to such things as a system of juncture in which any of the exponents in substance can be an exponent of any of the terms in the system. The exponents of the phonological unit which carries contrasts of intonation in English, the tone group, are in fact not coextensive with the exponents of any one or even any two of the grammatical units: if one describes English grammar with the tone group as a unit, when it can be exponentially coextensive with every one of what a 55

early papers on basic concepts

“monolevel” grammar would regard as five different units, one fails to show what is in fact the relation between the grammar and the phonology of the language.64 A meaningful account of how grammar and phonology are related in English must be based on a prior separate statement of the two.65 As said above (1.4), it is misleading to think of the levels as forming a hierarchy. They represent different aspects of the “patternedness” of linguistic activity. In this respect phonology is an “interlevel”, since there is an added degree of dependence here: the patterns of substance, and those of form, are each fully definable in their own terms, by internal reference, so to speak; whereas the patterns of phonology, being the organization of substance in form, although requiring to be stated separately because of the distinct nature of the abstractions involved, and the fact that (not surprisingly) the patterns are carried by different stretches of exponents, are only significant with reference to the two levels of form and substance. Context is an interlevel in a different sense, since it relates language to something that is not language; it is an interlevel because it is not with the non-language activity itself that linguistics is concerned but with the relation of this to language form. Though there is no precedence or priority, there is of course order among the levels, as determined by their specific interrelations; and in the study of language as a whole form is pivotal, since it is through grammar and lexis that language activity is – and is shown to be – meaningful. 7.2 The scale of rank has been discussed (above, 3) with reference to the unit, the basic category which operates on this scale. The syntactic (‘downward’) determination of classes is a feature of the theory, so that with respect to the category of class the rank scale appears as one of logical precedence running from highest to lowest unit. But this precedence applies to class criteria only, so that even in the theory it is not a one-way relation: the theory itself embodies “shunting” (moving up and down the rank scale) as crucial to the interrelation of the categories. In description, all statements presuppose shunting; the description of the sentence cannot be complete until the description of the morpheme is complete, and vice versa. In presentation, of course, procedure varies according to purpose and scope: downward presentation seems easier to make clear, but this may well be overridden – for example if grammar is an adjunct to lexical description, as sometimes in the statistical study of lexis. Rank is distinct both from exponence and from delicacy. A shift in 56

categories of the theory of grammar

one never by itself entails a shift in either of the others. The reason why rank is often confused with other scales is that there are cases where a shift in rank does accompany a shift in something else; but this is always by virtue of the logical relations among the categories involved. The fact that by moving from structure to class, which is (or can be) a move on the exponence scale, one also moves one step down the rank scale, is due to the specific relation between the categories of class and structure, and not to any inherent interdetermination between exponence and rank. The descriptive relevance of keeping the scales distinct is that it is important to be able to display what happens if one shifts on one scale, keeping the other two constant.66 7.3 Exponence is the scale which relates the categories of the theory, which are categories of the highest degree of abstraction, to the data.67 Since categories stand in different relations to the data, it might seem necessary to recognize four different scales of exponence, one leading from each category. In fact, however, exponence can be regarded as a single scale. In the first place, each category can be linked directly by exponence to the formal item: it is in fact a requirement of the theory that any descriptive category should be able to be so linked. This may be stated by way of exemplification, as when we say “the old man is (an example of ) an exponent of S in clause structure”. This is, however, not a description of the element S, since by relating it to its exponent at a stage when it was not necessary to do so we should have lost generality (cf. above, 6.3). So instead of throwing up the grammatical sponge and moving out to lexis while this is still avoidable, the description takes successive steps down the exponence scale, changing rank where necessary, until (at the degree of delicacy chosen) it is brought unavoidably face to face with the formal item. In the second place, therefore, the step by step move from any one category to the data can proceed via any or all of the other categories. This is then a move down the exponence scale, and at each step, given that delicacy is constant, one category is replaced by another (either with or without change of rank, according to which category is replaced by which). While therefore the categories are distinct, they are interrelated in such a way that the relation of exponence has the status of a single scale. The ultimate exponent in form is the formal item. This has then to be related, in turn, to the substance. But this relationship, though it may also be called “exponence”, entails a new scale, in which the 57

early papers on basic concepts

nature of the abstraction is different and the formal item is now at the other end – it is itself the abstraction, and is not in any way delimited or categorized by grammar.68 When grammar reaches the formal item, either it has said all there is formally to be said about it or it hands over to lexis. The formal item is the boundary of grammar on the exponence scale. It is not of course the boundary on the rank scale: whenever the formal item is anything other than a single morpheme (whether in closed system like “seeing (that)”, or in open set like “pickup”) the grammar can be taken further down in rank, since it can state the structures in terms of elements whose exponents are words and morphemes. But seeing (that) enters into a system at group rank, while pickup emerges from the grammar as a word, though being a lexical item it would not necessarily be an exponent of any whole grammatical unit. Once it has been taken over by lexis, the grammatical categories, and the grammatical exponence scale, no longer impinge on it (see below, 8.2). 7.4 Delicacy is the scale of differentiation, or depth in detail. It is a cline, whose limit at one end is the primary degree in the categories of structure and class. In the theory, the other limit is the point beyond which no further grammatical relations obtain: where there are no criteria for further secondary structures, or systems of secondary classes or formal items. In description, delicacy is a variable: one may choose to describe a language without going beyond the primary degree, still being comprehensive in rank and exponence and making use of all the categories of the theory. Each subsequent increase in delicacy delays the move to the exponents (cf. above, 7.3) and thus increases the grammaticalness of the description. The limit of delicacy is set by the means at one’s disposal. In well-described languages, such as English, any extension in delicacy beyond what is already known requires either or both of largescale textual studies with frequency counts and complex secondary classification based on multiple criteria, criteria which often cut across each other and may have to be variably weighted. And, as suggested above (6.3), a point will perhaps be reached where probabilities are so even as to cease to be significant69 and classes so delicately differentiated that the description will have to decide on crucial criteria and ignore the others,70 thus setting its own limits. Delicacy is distinct from rank and the limit of delicacy applies at the rank of all units, for example differentiation of clause structures and of classes of the group. At one stage, therefore, it becomes a limit on the 58

categories of the theory of grammar

grammatical differentiation of items which then remain to be lexically differentiated: it sets an endpoint to grammar where lexis takes over. Here the scales of delicacy and exponence meet. The endpoint set to grammar on the exponence scale is where abstraction ceases: one has to move from abstract category to exponential item. That set on the delicacy scale is where differentiation ceases: the set of exponents of each class, and of each element of structure, permits no further, more delicate groupings. If the formal items are still not ranged in systems, the implication in either case is that further relations among them are lexical. Whether or not, as discussed above (6.3), grammatical delicacy can reach a point where there is a one / one category–exponent relation (where each element of structure, and each class, has only one formal item as exponent), when all formal relations, including those among what are now treated as lexical items, can be accounted for by the grammatical categories and stated grammatically – in other words, whether or not, ultimately, all linguistic form is grammar, we do not know. At present, lexical items must be treated separately, and lexical relations established in their own right. These lexical relations do not depend on grammatical categories71 (so they are not yet “most delicate grammar”) and they have their own dimensions of abstraction (so not yet “most exponential grammar”). So there must be a theory of lexis, to account for that part of linguistic form which grammar cannot handle.

8

Lexis

8.1 This section is intended merely to bring lexis into relation with grammar, not to discuss the theory of lexis as such. As has been pointed out (above, 3.3. and 6.3), there is no one / one correspondence in exponence between the item which enters into lexical relations and any one of the grammatical units. It is for this reason that the term lexical item is used in preference to word, “word” being reserved as the name for a grammatical unit, that unit whose exponents, more than those of any other unit, are lexical items. Not only may the lexical item be coextensive with more than one different grammatical unit; it may not be coextensive with any grammatical unit at all, and may indeed cut right across the rank hierarchy. Moreover, since the abstraction involved is quite different, what is for lexis “the same” lexical item (that is, different occurrences of the same formal item) may be a number of different grammatical items, so it is 59

early papers on basic concepts

not true that one lexical item always has the same relation to the rank hierarchy. So that, in English, (i) a lexical item may be a morpheme, word or group (at least); (ii) a lexical item may be assigned to no rank, being for example more than a word but less than a whole group, or even both more and less than a word – part of one word plus the whole of another, sometimes discontinuously; and (iii) one and the same lexical item may in different occurrences cover any range of the possibilities under (i) and (ii).72 This does not mean that lexical items cannot be identified in grammar; it means that they are not identified by rank. They are identified, as has been suggested (above, 6.3), by their being unaccounted for in systems. But it is an additional, descriptive reason (additional, that is, to the theoretical one that lexical items lend themselves to different relations of abstraction) for keeping grammar and lexis apart. When the two have been described separately, the next stage is to relate them; and it is here that the complex relation between lexical item and grammatical unit must be accounted for. This is exactly parallel to what was said above (7.1) about grammar and phonology; and, of course, it applies equally to phonology and lexis, where, after separate description, is displayed the relationship between the lexical item and the categories of phonology. 8.2 The task of lexis can be summed up, by illustration, as that it has to account for the likelihood of wingless green insects and for the, by contrast, unlikelihood of colourless green ideas.73 As in grammar, we shall expect language to work by contrasting “more likely” with “less likely” rather than “possible” with “impossible”; but, as has often been pointed out,74 this particular type of likelihood is not accounted for by grammar, at least not by grammar of the delicacy it has yet attained. It is, however, too often assumed that what cannot be stated grammatically cannot be stated formally: that what is not grammar is semantics, and here, some would add, linguistics gives up.75 But the view that the only formal linguistics is grammar might be described as a colourless green idea that sleeps furiously between the sheets of linguistic theory, preventing the bed from being made. What are needed are theoretical categories for the formal description of lexis. It seems that two fundamental categories are needed, which we may call collocation and set.76 The first basic distinction between these and the categories of grammar is that in lexis there are no scales of rank and exponence. There is no hierarchy of units; therefore no rank scale. There is only one degree of abstraction – a set is a set of formal items 60

categories of the theory of grammar

and a collocation is a collocation of formal items; therefore no exponence scale (exponence there is, of course, but it is a simple polarity). Only the scale of delicacy remains; sets and collocations can be more and less delicate. There is an analogy with the categories of grammar, an analogy due to the nature of language as activity. Collocation, like structure, accounts for a syntagmatic relation; set, like class and system, for a paradigmatic one. There the resemblance ends. Collocation is the syntagmatic association of lexical items, quantifiable, textually, as the probability that there will occur, at n removes (a distance of n lexical items) from an item x, the items a, b, c . . . Any given item thus enters into a range of collocation, the items with which it is collocated being ranged from more to less probable; and delicacy is increased by the raising of the value of n and by the taking account of the collocation of an item not only with one other but with two, three or more other items. Items can then be grouped together by range of collocation, according to their overlap of, so to speak, collocational spread. The paradigmatic grouping which is thereby arrived at is the set. The set does not form a closed system, but is an open grouping varying in delicacy from “having some (arbitrary minimum) collocation in common” to subsets progressively differentiated as the degree of collocational likeness set as defining criterion increases.77 In lexis, as in grammar, it is essential to distinguish between formal and contextual meaning.78 Once the formal description has identified the categories and the items, these can and must be treated contextually. The formal item of lexis, the lexical item, is unrestricted grammatically; grammatical categories do not apply to it,79 and the abstraction of the item itself from a number of occurrences (including, for example, the answer to the question whether one is to recognize one lexical item or more than one) depends on the formal, lexical relations into which it enters. The nature of these relations is such that formal statements in lexis require textual studies involving largescale frequency counts: not of course of the frequency of single items, but of items in collocation. Since these are no longer difficult to undertake, it should not be long before we find out much more about how language works at this level.

61

early papers on basic concepts

9

An analogy

9.1 Eating, like talking, is patterned activity, and the daily menu may be made to yield an analogy with linguistic form. Being an analogy, it is limited in relevance; its purpose is to throw light on, and suggest problems of, the categories of grammar by relating these to an activity which is familiar and for much of which a terminology is ready to hand. The presentation of a framework of categories for the description of eating might proceed as follows: Units: Daily menu Meal Course Helping Mouthful Unit: Daily menu Elements of primary structure Primary structures Exponents of these elements (primary classes of unit ‘meal’) Secondary structures Exponents of secondary elements (systems of secondary classes of unit ‘meal’) System of sub-classes of unit ‘meal’

E, M, L, S (“early”, “main”, “light”, “snack”) EML EMLS (conflated as EML(S)) E: 1 (breakfast) M: 2 (dinner) L: 3 (no names available; see secondary classes) S: 4 ELaSaM ELaM EMLbSb EMSaLc La: 3.1 (lunch) Lb: 3.2 (high tea) Lc: 3.3 (supper) Sa: 4.1 (afternoon tea) Sb: 4.2 (nightcap) E: 1.1 (English breakfast) 1.2 (continental breakfast)

Passing to the rank of the “meal”, we will follow through the class “dinner”: 62

categories of the theory of grammar

Unit: Meal, Class: dinner Elements of primary structure Primary structures

Exponents of these elements (primary classes of unit “course”) Secondary structures Exponents of secondary elements (systems of secondary classes of unit “course”)

Systems of sub-classes of unit “course”

씮

F, S, M, W, Z (“first”, “second”, “main”, “sweet”, “savoury”) MW MWZ MZW FMW FMWZ FMZW FSMW FSMWZ FSMZW (conflated as (F(S)MW(Z)) F: 1 (antipasta) S: 2 (fish) M: 3 (entre´e) W: 4 (dessert) Z: 5 (cheese*) (various, involving secondary elements Fa–d, Ma,b, Wa–c) Fa: 1.1 (soup) Fb: 1.2 (hors d’oeuvres) Fc: 1.3 (fruit) Fd: 1.4 (fruit juice) Ma: 3.1 (meat dish) Mb: 3.2 (poultry dish) Wa: 4.1 (fruit*) Wb: 4.2 (pudding) Wc: 4.3 (ice cream*) Fa: 1.11 (clear soup*) 1.12 (thick soup*) S: 2.01 (grilled fish*) 2.02 (fried fish*) 2.03 (poached fish*) Wb: 4.21 (steamed pudding*) 4.22 (milk pudding*)

Exponential systems operating in meal structure

63

Fc: grapefruit / melon Fd: grapefruit juice / pineapple juice / tomato juice Ma: beef / mutton / pork Mb: chicken / turkey / duck / goose

early papers on basic concepts

At the rank of the “course”, the primary class “entre´e” has secondary classes “meat dish” and “poultry dish”. Each of these two secondary classes carries a grammatical system whose terms are formal items. But this system accounts only for simple structures of the class “entre´e”, those made up of only one member of the unit “helping”. The class “entre´e” also displays compound structures, whose additional elements have as exponents the (various secondary classes of the) classes “cereal” and “vegetable”. We will glance briefly at these: Unit: Course, Class: entre´e Elements of primary structure Primary structures Exponents of these elements (primary classes of unit “helping”) Secondary structures Exponents of secondary elements (systems of secondary classes of unit ‘helping’)

J, T, A (“joint”, “staple”, “adjunct”) J JT JA JTA (conflated as J((T) (A))) J: 1 (flesh) T: 2 (cereal) A: 3 (vegetable) (various, involving – among others – secondary elements Ja,b, Ta,b, Aa,b) Ja: 1.1 (meat) Jb: 1.2 (poultry) Ta: 2.1 (potato) Tb: 2.2 (rice) Aa: 3.1 (green vegetable*) Ab: 3.2 (root vegetable*)

And so on, until everything is accounted for either in grammatical systems or in classes made up of lexical items (marked *). The presentation has proceeded down the rank scale, but shunting is presupposed throughout; there is mutual determination among all units, down to the gastronomic morpheme, the “mouthful”. Like the morpheme, the mouthful is the smallest unit, and all eating activity can be broken down into mouthfuls. Like the morpheme, it is neither more nor less fundamental than any other unit. But how little it reveals, if description proceeds in one direction from it, of the complexity of the whole activity! Our predecessors thought of language as an organism, and drew 64

categories of the theory of grammar

their analogies from evolution. We reject this as misleading; but no less misleading is its familiar substitute, according to which language is an edifice and the morphemes are the bricks. Perhaps if language had been thought of as activity we should never have heard of “morphemics”. It is unfair on one who was among the greatest figures in linguistic science to call the “bricks and edifice” view of language “Bloomfieldian” – and I do not know that Bloomfield ever used this analogy.80 But if I may use “Bloomfieldian” as a shorthand name for a view of language, and a body of descriptive method, though of course with many individual varieties, which has had wide currency for the last quarter of a century and owes a very great deal to Bloomfield’s work, I would like to consider what seem to me to be certain questionable features of “Bloomfieldian” linguistics. What is being called in question is really the theory (and perhaps the analogy) of how language works.

10

The seven sins

10.1 From the point of view of the present theory, there are seven features of what is here labelled “Bloomfieldian” method which would perhaps justify critical comment.81 These are: (1) (2) (3) (4) (5) (6) (7)

confusion of “level” with “rank” confusion of “rank” with “exponence” conflation of different level levels overemphasis on lowest unit reluctance to shunt reluctance to shade distribution of redundancy

The confusion of level with rank takes a specific form: the relation between different units at one level (morpheme . . . sentence) is conflated with the relation between two different levels, grammar and phonology. Ranged on a single scale (“from phoneme to utterance”) are (i) the move between phonology and grammar (which, moreover, always goes from the “interlevel” to the level; cf. below, 10.3) and (ii) the move between units (which, moreover, always goes from the lowest to the highest; cf. below, 10.5) in (a) phonology and (b) grammar. In description, this leads to unwanted complexity and thus weakens the power of the grammar: one must be free to recognize grammatical units whose exponents in substance both overlap with and completely fail to coincide with the units carrying phonological contrasts.82 In the theory, there is a confusion of abstractions: the abstrac65

early papers on basic concepts

tion involved in relating one unit to another at the same level is quite different from that involved in relating one level to another. The sense in which a sentence “consists of ” morphemes is stated in description with reference to its definition in the theory, and is totally different from the sense in which a morpheme “consists of ” phonemes – indeed it is doubtful if there is any meaningful sense in which a morpheme consists of phonemes. (The contrastive use of “morph” and “morpheme” is designed to build in an extra stage to account for the two kinds of abstraction. But it does not get over the first (descriptive) difficulty; nor in fact does it solve the theoretical one, since morph and phoneme differ in the extent and kind of formal determination underlying their phonological abstraction.83) 10.2 The relation among the units also tends to be confused with the relation between a category and its exponent(s). It is assumed that in moving up the rank scale, from morpheme to sentence, one is also moving up the exponence scale. It is true (as said above, 6.2 and 7.3) that in comprehensive description, in order to display the full grammaticalness of language, one takes the final step on the exponence scale at the lowest rank possible (though this, as already shown, is by no means always the morpheme),84 and this is probably the reason for the confusion of rank with exponence. But the scales of rank and exponence are again different dimensions of abstraction, and one can link any unit directly to its formal exponent (and through this to its exponent in substance): the relation of an exponent of the unit “sentence” to the category of sentence is exactly the same as that of an exponent of the unit “morpheme” to the category of morpheme. 10.3 The conflation of levels referred to is the conflation of grammar and phonology, which follows, though is distinct, from the confusion of level with rank (above, 10.1). The theoretical basis of this criticism is complex but crucial. Any distinction in substance may (i) be a free variant, in which case it is formally meaningless, or it may (ii) carry a distinction in form, with meaning either at the grammatical or at the lexical level.85 All formal distinctions presuppose some distinction in substance,86 and once a distinction in substance is shown to carry a formal distinction it must be accounted for in phonology.87 But no relation whatever is presupposed between the categories required to state the distinction in form (grammar or lexis) and the categories required to state phonologically the distinction in substance which 66

categories of the theory of grammar

carries it.88 For example: the units may not be coextensive, and a variable relation of phonological unit to grammatical unit may be the very thing that carries a formal system (cf. above, 10.1 n. 82); or a single system in grammar may be carried by different phonological distinctions, say two of its terms by tone and a third by addition of a segment; or a phonological system, such as tone associated with a given unit (recognized as phonological because it carries some formal distinction), may carry different formal distinctions, part grammatical and part lexical, or terms in different grammatical systems.89 The categories required by the grammar, and the criteria for these, should come from within grammar. They are set up to provide a description that is comprehensive, consistent and maximally powerful. In the definition of, for example, the unit “clause”, the requirement is that it should yield classes and structures which make possible the description of the sentence, the group and so on in terms of their structures and classes: hence the mutual definition of all units and of all grammatical categories (and, procedurally, until a description is “comprehensive” (primary delicacy at all ranks) all parts of it remain subject to revision). A grammatical category is not required to be identifiable by reference to a particular feature of substance stated phonologically: it merely carries the potentiality of being stated in phonological terms through a long chain of exponence.90 (The starting point here of course is the theory of grammar, so that what is being considered, and objected to, is the identification of grammatical categories on phonological criteria. For linguistic theory as a whole, the question must also be formulated the other way round: do we derive phonological categories from formal ones? For example, in the last case mentioned in the first paragraph of this section, would we state one phonological system of tone or more than one? The nature of phonology as an “interlevel” suggests that its categories should be derived from those of form; and this is usually done in prosodic phonology, as developed by Firth and others, though it is not an inherent requirement of prosodic method.91 Inter-level dependence in this direction is theoretically justifiable, since the role of phonology is to account for the formally meaningful organization of phonic substance. But it has descriptive dangers: first, that a system carried only by variable relation between grammatical and phonological unit may be missed, and second, that unless a comprehensive formal description has first been made, formal features may be distorted into a phonological mould – the phonologist may take the “word” class “verb” as a phonological unit, but he (or someone) must have described 67

early papers on basic concepts

the grammar first, or his “word” and “verb” may not turn out to be the grammarian’s “word” and “verb”.) To be precise, then, what is being criticized here has again both a theoretical and a descriptive aspect. In the theory, it is the use of phonologically stated features as crucial criteria for grammatical categories, as when supra-segmental phonemes are used as criteria for the category of “phrase”.92 A phrase is a phrase because it operates in the structure of the unit above it and has its own structures in terms of the unit below it. It has then to be related to the phonological categories which are arrived at by a different process of abstraction on the basis of the minimal requirement that some formal distinction is always involved. If at any point this yields a one / one relation of categories, so much the better: if the phrase turns out to be exponentially coextensive with the tone group, the latter can be used as a recognition signal for the former. But it remains a signal, not a criterion. In description, the trouble arises when phonological features are used as grammatical criteria even when they clash. If, for example, the segment which carries tone contrasts, or is bounded by juncture features, will not work as a grammatical unit, then tone and juncture are no use even as recognition signals; so to “define”, say, the clause by reference to tone or juncture one has to set up a phonological system in which any feature in substance can be an allophone of any term in the system. There will be clash; and if it is recognized in the first place that there is nothing at all surprising when, say, units carrying formal patterns do not coincide with those carrying patterns of the organization of substance, then the search for one / one phonological identification signals of grammatical categories, such as a phonological statement of clause boundary, can be abandoned as being without profit.93 10.4 In both grammar and phonology the smallest units, morpheme and phoneme, are often assigned a special status distinct from that of any of the other units. Since the description usually proceeds unidirectionally upwards (consistently in phonology, and in grammar up to about group rank; cf. below, 10.5) these units are treated first: they are then rated (uniquely) fundamental, and in phonology “suprasegmental” features are given phoneme status. The morpheme becomes a grammatical brick (here the analogy does yield an interesting metaphor!) which is used to build the larger units; and grammar – or at least that part of it that can be handled in this way – becomes “morphemics”, as phonology becomes “phonemics”. 68

categories of the theory of grammar

The result is sometimes referred to as excessive segmentation, though this is not really the essence. The objection is not that structural analysis is carried through to the morpheme; it is clearly necessary that it should be, though not to the extent of requiring one morpheme with distinct segmental exponent for each system in which the item in question is a term.94 The objection is that the smallest unit, being mistaken as fundamental by contrast to the others, is made to do much too much work: to carry features and contrasts which properly belong to larger units. The morpheme is no more and no less “fundamental”, being no more and no less an abstraction, than the sentence. It has enough patterns of its own to carry without having foisted on it those that the language has distributed elsewhere.95 10.5 With the smallest unit as fundamental, the description starts off in an upward direction. It proceeds, unidirectionally, from the morpheme, through the word to somewhere around the group.96 Not surprisingly, since there is no shunting, it proves extremely difficult to take it further along the same route. Shunting, or moving up and down the rank scale, is a part of descriptive method imposed by the theory to show the relation among the different units: to permit a unified description with links, through all categories, all the way from morpheme to sentence. In the absence of shunting, the description has to jump to the top end of the rank scale and proceed downwards from the sentence by “immediate constituent” analysis:97 still unidirectional, though with the direction reversed. The hope is that, by digging the tunnel from both ends, the two will meet in the middle. With the aid of a good homogeneous mathematics they might; but two totally different bodies of method are involved, morphemics and IC analysis, which are difficult to integrate. Moreover, the middle ranks of the grammar are often the most complex, presumably since they face both ways; so that a grammar which starts unidirectionally from the two ends will find it difficult to avoid leaving the middle ragged. Presentation can of course proceed in any direction that is desired; but it needs to be based on a description that permits – indeed presupposes – constant shunting. 10.6 It has been suggested at various places above that the theory cannot validly regard constrasts and relations that are clearcut, and statable in “yes / no” terms, as the norm in language. Even at the primary degree of delicacy the description will encounter features where “shading” is necessary: where a feature is better stated in terms 69

early papers on basic concepts

of “more likely / less likely’ and a statement is more powerful when it accounts for only 90 per cent of occurrences; and this becomes more true with every subsequent gain in delicacy.98 If this is built in to the theory as a basic property of language, there is no need to try to turn all constrasts and relations into “yes / no” ones, an attempt which may slant the choice of criteria. Criteria are sometimes chosen for this purpose which make it necessary to account for a large number of occurrences by changing them into something else, because they do not display the contrast chosen as crucial. (This device is more readily tolerated by those who give relatively little weight to textual as opposed to exemplificatory description.) For example, the exponent of the element “subject” in English clause structure is sometimes identified as being that nominal group which is in person / number concord with the verbal group. But since a large number of English clauses do not display this concord contrastively99 (quite apart from those which violate it, as does the clause I have just written), the textual application of this criterion has to rely on the substitution of all other verbal groups by one in is, was, has or simple present form and, if the two nominal groups are alike in number, on the dissimilation of one of them. If instead one says that the exponent of “subject” is that nominal group which precedes the verbal group with no other nominal group in between, this can be stated as a high probability relation in spoken English and is applicable to a textual study (in which it can be quantified) without substitution. The next step is then to make more delicate statements to account for the instances so far not accounted for, some of which turn out to be grammaticizable and others, for the moment, not. And it ceases to be puzzling why, if a foreigner (or native) says my friends is inviting the neighbour, every English speaker knows that it is the friends who are doing the inviting and not the neighbour. 10.7 The problem of “redundancy” is complex, and needs treating separately and at length; but the term has become a commonplace in description and a brief reference may be made here. The name is assigned to a number of varied phenomena, none of which is related in any clear way to the quantifiable redundancy of information theory.100 These include relations of formal categories to exponents in form, of formal and phonological features to exponents in substance, and of formal features to context. Moreover, in some instances the so-called “redundancy” is simply put in by the method of description and has no relevance to the language at all. 70

categories of the theory of grammar

“Redundancy” is assigned to what is displayed as multiple exponence: either in form, where more than one formal item is said to be the exponent of one grammatical category, or in substance, where a distinction is carried by what is said to be more than one phonetic feature. But neither of these is at all clearcut. Formal redundancy “occurs” where there is concord, but no criteria are available for identifying the two prerequisites of concord: that there is “more than one” exponent as opposed to “one”, and that these are exponents of “the same” category.101 Discontinuous morphemes, for example, may sometimes be clearly recognizable, though at others it is impossible to say “how many” exponents are present;102 but the question is irrelevant, since where the description does recognize concord this concord is itself the exponent of a distinct category of relation that is different from the category of which the form is exponent, and that has its own formal meaning.103 Redundancy in substance appears when formal or phonological distinctions are related to contrastive features.104 Here precisely the same problem arises, since it is not possible to give rigorous criteria for deciding what is “one” phonetic feature and what is “more than one”.105 Each time a new parameter or a further degree of differentiation is introduced into the phonetic statement, all its precursors are thereby made “redundant”. In extreme cases this “redundancy” becomes completely artificial, since it is simply inserted by the description. This happens when a contrast (or system) is assigned to a unit lower than that to which it is appropriate, and may result therefore from overemphasis on the lowest unit. This tends to happen more in phonology, when the phoneme is made to carry contrasts appropriate to a higher unit; one of the merits of prosodic phonology is that it avoids this error.106 But it is not unknown in grammar, where it may also arise from the use of morphological instead of syntactic criteria for classes.107 In such cases “redundancy” can only refer to the loss of power in the grammar brought about by such a description: it already follows from the theory that the “appropriate” unit for the assignment of any feature is the highest unit that can carry it without requiring the statement to be made twice. The best description therefore can be thought of as that which minimalizes artificial “redundancy”. But at the same time those instances where what is called “redundancy” is an artificial product of the description are not essentially different from, but are merely extreme cases of, the “multiple exponence” in form and substance to which the same name is applied.108 What is of doubtful validity here is the implication that there are 71

early papers on basic concepts

formal contrasts carried by “one” exponent and others carried by “more than one”, with a meaningful distinction between the two. Even if “multiple exponence” in form can be validly identified, it is itself formally meaningful; and it is arbitrary to postulate “one feature” as the norm of exponence in substance. The use of the term “redundancy” is unfortunate for two reasons. On the one hand it implies that some features, by contrast to others,109 can be recognized as carried by something more than what “would be enough” – to the extent even of suggesting that one may judge which of a number of exponents is “the” non-redundant one.110 On the other hand, even if it is possible to devise some theoretically valid criteria for “redundancy” of this kind, its relation to the redundancy of information theory is extremely complex and it would be better not to call it by the same name. The redundancy of information theory is of considerable interest to linguistics in the study of the information carried by grammatical systems; but this, as far as I know, has not yet been seriously attempted. The quantification of systems, rather than the appraisal of features as contrastive or idle, which rests on a very partial interpretation of the redundancy of information theory, seems the more useful role for the concept of redundancy in linguistics. Most, if not all, of the points made in this section can be brought together under Chomsky’s observation that “a linguistic theory should not be identified with a manual of useful procedures, nor should it be expected to provide mechanical procedures for the discovery of grammar”.111 The point is a familiar one to British linguists, who have for some time stressed the theoretical, as opposed to procedural, character of their own approach.112 But is it true that “it is unreasonable to demand of linguistic theory that it provide anything more than a practical evaluation procedure for grammars”?113 This it must do. But it can be asked to do more: to provide a framework of logically interrelated categories (so that it can be evaluated as a theory, and compared with other theories) from which can be derived methods of description, whether textual, exemplificatory or transformative– generative, which show us something of how language works.

Notes 1. It is in no way to deny the fundamental importance of Chomsky’s work (1957) and elsewhere, if we suggest that the readiness of linguists who had previously worked in the “Bloomfieldian” tradition to abandon these methods in favour of Chomsky’s is in part due to their lack of

72

categories of the theory of grammar

2.

3.

4.

5.

6.

7.

8.

theoretical foundation. The point of view adopted here is that transformation-generation is a type of description which, like other types, depends on but does not replace a theory. Even Chomsky (1961) seems to imply that a textual study cannot be theoretical. But a grammar of one short text may be based on theory; and any theory-based grammar, transformational or not, can be stated in generative terms. Those linguists who have followed up the work of Firth have always tended to give more weight to textual description than have those following Bloomfield, since for the former meaning and the statement of meaning have always been integrated in the theory. Cf. Firth (1957a): “The object of linguistic analysis as here understood is to make statements of meaning so that we may see how we use language to live” (p. 23; cf. also p. 11). Professor Firth died on 4 December 1960. I had just completed this paper and was planning to show it to him on the following day. Although he had not seen it and was in no way directly responsible for any of the opinions formulated here, the influence of his teaching and of his great scholarship will, I hope, be clearly apparent. See especially Firth (1955; 1957a; 1957b, chapters 9, 10, 14–16; 1957c). Of major importance to me have been discussions, both on linguistic theory as a whole and on the specific subjects mentioned, with J. C. Catford, J. O. Ellis, A. McIntosh (lexis and “delicacy” – the latter concept is of his origination), J. M. Sinclair (English grammar) and J. P. Thorne (logical structure of linguistic theory, and the work of Chomsky). As used by Firth (1957b: 225). Here “text” refers to the event under description, whether it appears as corpus (textual description), example (exemplificatory) or terminal string (transformative–generative). The set of these abstractions, constituting the body of descriptive method, might be regarded as a “calculus”, since its function is to relate the theory to the data. It is important to distinguish between calculus (description) and theory; also between description and the set of generalizations and hypotheses by which the theory was arrived at in the first place. The latter precede the theory and are not susceptible of “rigorization”; though we may distinguish the logical stages of observation– generalization–hypothetization–theory, keeping Hjelmslev’s (1953: 8) distinction between “hypothesis” and “theory”; cf. Allen (1953: 53). Here we are concerned with the stages, once the theory is formulated, of theory–description–text. Since the theory is a theory of how language works, it does not matter whether the levels are considered levels of language or levels of linguistics (theory or description): it comes to the same thing. Cf. Firth (1957b): “We must expect therefore that linguistic science will also find it necessary to postulate the maintenance of linguistic patterns and

73

early papers on basic concepts

9. 10. 11.

12.

13.

14.

systems . . . within which there is order, structure and function. Such systems are maintained in activity, and in activity they are to be studied. It is on these grounds that linguistics must be systemic” (p. 143, cf. also pp. 187, 192). Cf. McIntosh (1956). Professor McIntosh followed this up subsequently in a study of the underlying theoretical problems. This figure is a schematic representation of §§1.5–1.7. Cf. Firth (1957b): “A nominative in a four-case system would in this sense necessarily have a different ‘meaning’ from a nominative in a twocase or in a fourteen-case system, for example” (p. 227). The article from which this is quoted, “General linguistics and descriptive grammar”, was published in 1951; but Firth’s view of the “dispersal of meaning”, that (i) form is meaningful and (ii) formal meaning is distinct from contextual meaning, antedates this by some time; it is in fact already clear, though without the precise formulation of formal meaning, in “The technique of semantics” (1935), also reprinted in Firth (1957b). Some of what has been written on information theory and language is vitiated by the confusion between these two levels of meaning; cf. my reviews of Whatmough (Halliday 1958) and Herdan (Halliday 1959a). It is doubtful whether, even if contextual meaning can ever be quantified, it has anything to do with “information”; the latter is a function of the operation of (a term in) a system, and a linguistic item can never be a term in a contextual system even if such a thing can be rigorously described. Cf. below, 10.7. The reason why “context” is preferred to “semantics” as the name of this interlevel is that ‘semantics’ is too closely tied to one particular method of statement, the conceptual method; cf. Firth (1957a: 9–10, 20). The latter, by attempting to link language form to unobservables, becomes circular, since concepts are only observable as (exponents of) the forms they are set up to “explain”. The linguistic statement of context attempts to relate language form to (abstractions from) other (i.e. extratextual) observables. Cf. Firth (1957a): “References to non-verbal constituents of situations are admissible in corroboration of formal linguistic characteristics stated as criteria for setting up . . . word–classes” (p. 15). The approach to context from the other end, that is from non-language, has been developed in an important monograph by William E. Bull (1960), as what he (perhaps unfortunately, in view of the formal use of “system”) calls “systemic linguistics”. The difficulty of this method lies in deciding on what Bull calls “those features of objective reality which are pertinent to the problem” (p. 3), since this can only be known by reference to linguistic forms: cf. e.g. “it may be assumed that normal people automatically divide, on the preverbal level, all events into three categories: those anterior to PP (point present) . . . , those simultaneous with PP

74

categories of the theory of grammar

15.

16. 17.

18.

. . . , and those posterior to PP” (p. 17); “The languages of the dominant world cultures use vector formulas, and the discussion which follows is therefore concerned only with the structure of a hypothetical tense system based on the vector principle” (p. 20); “The system, of course, would break down if a plus form were to be used to describe a minus event or if a form indicating anticipation were used for recollection” (p. 24). This does not invalidate the approach; it does suggest that it will have to be part of a study of context which starts from form as well as from “objective reality”, as phonology works both from form and from substance; context, like phonology, is in a real sense an interlevel. Theoretical validity implies making maximum use of the theory (see below, 2.3 and 6.2). It is not necessary to add a separate criterion of “simplicity”, since this is no use unless defined; and it would then turn out to be a property of a maximally grammatical description, since complication equals a weakening of the power of the theory and hence less grammaticalness. It should perhaps also be mentioned here that the distinction between methods of description and discovery procedures is here taken for granted throughout (cf. below, 2.3). We are not concerned with how the linguist “finds out” how an event is to be described. This is no more capable of scientific exposition than are the steps by which the theory was arrived at in the first place – in fact less, since the latter can at least be formulated, while the former can only be summed up in the words of the song: “I did what I could”. Cf. Firth (1957a: 22), and above, 1.8 n. 11; Garvin (1957); see below, 6.3 n. 60; Robins (1959). “Grammar” is also the name for the study of grammar; as with “level” (above, 1.4 n. 8), it is unnecessary to distinguish between “the grammar” of a language and “grammar” in theory and description – though a distinction is often made between “lexis” and “lexicology”, the latter being the study of lexis. Again, not a set of discovery procedures, but a set of properties of what the linguist accounts for grammatically. The grammar of a language can only be “defined” as that part of the language that is accounted for by grammatical description. The reference is, of course, in formal meaning: it is form that is under discussion. It may always happen that the addition of a new term changes the contextual meaning of at least one of the others, since terms that are formally mutually exclusive are likely to carry contextual distinctions; but this is not a property of a system. The “addition” of a new term is not of course considered as a process (though historical change is one type of instance of it): it may be displayed in any comparison of two related systems. For example, two possible systems of first and second person pronouns used by different speakers of Italian (quoted in oblique disjunct form; I = “interior to social group”, E = “exterior . . .”).

75

early papers on basic concepts 1 1+ 2I 2I+ 2E 2E+

19.

20.

21.

22.

me noi te voi lei loro

me noi te

}

voi

(The distinctions made in written Italian are ignored, since they would not affect the point.) The difference in format meaning is a function of the different number of terms: in system one me excludes five others, in system two only three. In contextual meaning only terms of the second group are affected. Cf. Firth (1957a): “Moreover, these and other technical words are given their “meaning” by the restricted language of the theory, and by application of the theory in quoted works” (p. 2). This is true of descriptive categories too: “noun” can no more be defined in a glossary than “structure”. I should therefore agree with Palmer (1958) that linguistic levels do not form a hierarchy. His view is “that there are levels, but only in the widest sense, and that these are related in specific, but different, ways. The set of relationships cannot be regarded as a hierarchy, except in the loosest sense of the word”. Palmer, however, appears to reintroduce procedural hierarchy when he says, “The procedure is not from phonetics via phonology to grammar, but from grammar via phonology to phonetics, though with the reminder that the phonetic statement is the basis, i.e. the ultimate justification for the analysis” (p. 240). I would rather say that there is order among the levels, determined by their interrelations, but (a) no hierarchy, in the defined sense of the word, and (b) no procedural direction. Unfortunately Palmer excludes this use of “order”. “There is a statable order of levels . . . and, therefore, a hierarchy” (pp. 231–2, in reference to Hockett). Immediate Constituent analysis, for example, yields a hierarchy that is not a taxonomy: it does not fulfil criterion (ii). (It may not always fulfil (i): cf. Hockett (1957): “There must be also at least a few utterances in which the hierarchical structure is ambiguous, since otherwise the hierarchical structure would in every case be determined by form, and order, and hence not a “primitive”” (p. 391).) The theory thus leads to “polysystemic”-ness in description – both syntagmatically and paradigmatically. Syntagmatic polysystemic statement follows from the linking of classes and systems to places in structure (see below, 4–6), so that the question “how can we prove that the b of beak and the b of cab are occurrences of one and the same phoneme?” (Ebeling: 1960: 17) is regarded as an unreal one; cf. Henderson (1951: 132); Carnochan (1952: 78); Robins (1953: 96); Firth (1955: 93; 1957b:

76

categories of the theory of grammar

23.

24.

25. 26.

27. 28.

121) and Palmer (1958: 122–4). Paradigmatically, the “simplicity” referred to here follows from the requirement of making maximum use of the category of “system” by polysystemic or “multidimensional” statement in grammar; cf. Halliday (1956: 192). Manifestation (in substance) and realization (in form) are introduced here to represent different degrees along the scale of exponence (see below, 7.3). In this paper I have used exponent as indicating relative position on the exponence scale (a formal item as exponent of a formal category, and a feature of substance as exponent of a formal category or item); this departs from the practice of those who restrict the term “exponent” to absolute exponents in substance. As used here, formal item is a technical term for the endpoint of the exponence relation (“most exponential” point) in form: the lexical item cat, the word cat as member of the word–class of noun, the morpheme -ing (as class member operating at the place of an element) in word structure, etc.; it is thus already an abstraction from substance and will be stated orthographically or phonologically. In this formulation, exponence is the only relation by which formal category, formal item and feature of substance are linked on a single scale: hence the need for a single term to indicate relative position on the scale. Two defined positions on this scale can then be distinguished as “realization” and “manifestation”. Cf. Firth (1957a): “In these structures, one recognizes the place and order of the categories. This, however, is very different from the successivity of bits and pieces in a unidirectional time sequence” (p. 5). Cf. above, 3.1 n. 23. The two latter restrictions represent an important addition to the power of the unit as a theoretical category. The first toleration is required to account for “regressive” structures: cf. Yngve (1960: 19). As Chomsky (1957) has said, “the assumption that languages are infinite is made in order to simplify the description of these languages . . . If a grammar does not have recursive devices it will be prohibitively complex” (pp. 23, 24). Yngve makes the important distinction between “progressive” and “regressive” structures, accounting for them separately in his model. Whether or not he is right in postulating a depth limit (of about 7) for “regressive” structures, while allowing “progressive” structures to be infinitely expanded (p. 21), they do represent very different types of “infiniteness”, and are separately accounted for in the present theory, the former with, the latter without, rankshift. This determines the nature, but does not restrict the use, of the perfectly valid arbitrary limit on delicacy which the grammar can set in each case without loss of comprehensiveness. Such as the status of “being the smallest”. Cf. below, 8. The item for lexical statement is not to be identified on the grammatical rank scale; nor is it a unit at all in the sense in which

77

early papers on basic concepts the term is here used in grammar, since this use presupposes a rank scale (as well as the other terms structure, class and system in a system of related categories), which is absent from lexis. It is probably better to restrict the term “unit” to grammar and phonology: cf. Bazell (1953: 11) – though Bazell does not here consider lexical form. 29. So, for the description of English: units sentence clause group (phrase) word 앗 morpheme 앖 rank

(

30. Statistical work on grammar may yield a further unit, above the sentence: it will then be possible to set up sentence classes, and account for sequences of them, by reference to this higher unit. Similarly in phonology we need a unit in English above the tone group to account for sequences of different tones. The grammatical and phonological “paragraph” (and perhaps ‘paraphone’?) is probably within reach of a team of linguist, statistician, programmer and computer; cf. Firth (1957a): “Attention must first be paid to the longer elements of text – such as the paragraph . . .” (p. 18); Harris (1952); for Hill (1958: 406), and others, this is “stylistics”, but in the present theory it would come within exactly the same general framework of categories. 31. The “simple / compound” opposition is thus one of structure. It may, of course, happen that a given realization yields simple membership all the way up and down the rank scale. Yes may be (i.e. may be an exponent of) one sentence which is one clause which is one group which is one word which is one morpheme. 32. Cf. Robins (1953: 109); Firth (1957a, esp. 17, 30; and 1955, esp. 89, 91); Halliday (1959: 49). 33. Cf. Firth (1957a): “Elements of structure . . . share a mutual expectancy in an order which is not merely a sequence” (p. 17). Since sequence is a variable, and may or may not be an exponent of structure, we find difference in sequence without difference in structure (cf. below, 4.3 n. 42), or difference in structure without difference in sequence. I am indebted to J. M. Sinclair for a recent conversational example of the latter: orthographically, The man came(,) from the Gas Board. Phonologically (relevant units: tone group, bounded by //, and foot, by / – these are unit boundaries and have nothing to do with juncture): what was said was (tonic syllable underlined): // 1 the / man_ / came // 1 from the / Gas / Board // Grammatically, one clause, structure SP; exponent of P came, of S the

78

categories of the theory of grammar man . . . from the Gas Board, being a nominal group, structure MH +Q. What might have been said was // 1 the / man / came from the // Gas / Board //

34.

35.

36. 37.

Grammatically, one clause, structure SPA; exponents, S the man, P came, A from the Gas Board. The two are different in grammatical structure, and this difference has its exponent in phonic substance which can be stated phonologically. (That the phonological patterns, and the distinction between them, abstracted from the substance along one dimension correspond regularly (though not one / one) with the grammatical patterns, and the distinction between them, abstracted along another dimension from the same substance can be shown by the construction of other partially like clauses.) But though the difference in structure has its manifestation in substance (there can of course be ambiguity in substance, as in Hockett’s old men and women (1957: 390n.), in form the difference is not realized in sequence. In sequence, from the Gas Board occupies the same place in both instances; in order, S and A stand in different relations to P, and from the Gas Board is exponent of (part of) S in the one case and of (the whole of) A in the other. Sequence is presumably always manifested in phonic substance as linear progression; the distinction is then one of exponence, “sequence” being the name for that formal relation between formal items of which linear progression is the manifestation in phonic substance. It is useful to make a distinction in the use of symbols between an inventory of elements of a structure and a structure, by the use of commas in the former. Thus, X, Y, Z is an inventory of elements, XYZ a structure composed of these elements. Since a unit that carried only one-place structures would be unnecessary: if, for example, all words consist of one morpheme (i.e. the unit “word” has no structure containing more than one place), “word” and “morpheme” would be one and the same unit. For the name and nature of this grouping, see below, 5. Since the morpheme (i) is a grammatical unit and (ii) carries no grammatical structure, it has no structure. Cf. Palmer (1958: 229–30) (quoting Hockett 1955: 15): “ ‘Morphemes are not composed of phonemes at all. Morphemes are indivisible units. A given morpheme is represented by a certain more or less compact arrangement of phonologic material . . . If we call any such representation a morph, then it becomes correct to say that a morph has a phonological structure – that it consists of an arrangement of phonemes.’ [Hockett] recognizes that the units established at each level differ in kind, and not merely in size, from those established at other levels.” The “morph” does indeed accommodate the theoretical point (but cf. below, 10. 1 n. 83), that the units differ in kind; but in accepting Hockett’s view Palmer has not

79

early papers on basic concepts

38.

39.

40.

41.

42.

noted that, since they differ in kind, “size” cannot be abstracted as common to the two dimensions of abstraction for them to differ in. That is, a grammatical unit can only be exponentially coextensive (or not) with a phonological one: when it is, this is a descriptive accident for which the linguist can be thankful (cf. reference to Allen (1956) in 7.1 n. 64 below), but the grammar cannot be made to define the units for phonological statement (cf. the example in 4.1 n. 33 above, where two exponents of the same grammatical unit “clause” may be (systemically contrasted by being) coextensive either with one tone group or with two). And, even though we may use the categories of “unit” and “structure” both in grammar and in phonology, these are not shown to be comparable unless the two theories have the same system of primitive terms with the same interrelations. As used by Hill (1958: 256). The “definitions” of these terms (i.e. the categories themselves) are of course different, since the theory differs from Hill’s. Cf. below, 10.6. This formulaic presentation is useful as a generalized statement of an inventory of possible structures: a list H, MH, HQ, HMQ can be generalized as (M)H(Q). This particular instance is an oversimplification, since there may be more than one exponent of M and Q: the formula would then read (M . . .n)H(Q . . .n), where . . .n allows infinite progression (not regression). The real point is to avoid taking two distinct theoretical steps at once. As said below (5), the relation of class to structure is such that a class of a given unit stands in one / one relation to an element of structure of the unit next above: thus, the exponent of the element P in the structure of the unit “clause” is the class “verbal” of the unit “group”. We could – provided we did so consistently – replace the symbol P here by V, thus conflating two statements. But not only are there descriptive reasons for not doing so (cf. below, 5.4); it is theoretically invalid, since two sets of relations are involved (element of clause structure to unit “clause”, class of group to unit “group”), and if the two steps are taken at once the crucial relation of structure to class on the rank scale is obscured. If, instead, an inventory of elements is stated first, 씮 the arrow can be added (where it really belongs) in the inventory: S, P, C, A. It is then no longer required in the statement of structures, since it is presupposed. Cf. above, 4.1. In a Latin clause of structure SOP (O = object), sequence plays no part in the definition of the elements: so no arrow. But rearrangements of the elements, to give SPO, OSP, etc., can be usefully employed to state the more delicate distinctions씮beween puer puellam amat, puellam puer amat, etc. In English, where SP sequence is crucial to the definition of S (though various arrangements of C and A are possible), more delicate grammatical distinctions, such as those carried by intonation, must be shown secondarily.

80

categories of the theory of grammar 43. For example the following two exponents of the (class) nominal (of the unit) group: all the ten houses on the riverside and the finest old houses on the riverside have the same primary structure M . . . HQ (or MMMHQ). But a more delicate statement of M, still at group rank, shows distinct secondary structures, the first example having D2DbO, the second DbOE. 44. When Hockett writes (1955: 17) “In general, then, if we find continuous-scale contrasts in the vicinity of what we are sure is language, we exclude them from language (though not from culture)”, this applies (i) only to grammar and phonology, not to lexis or context (cf. Bazell 1953: 11), and (ii) only to one type of contrast, that between terms in systems. It is, indeed, a defining characteristic of a system that it cannot be a cline. But units and classes are not crucially discrete: in exponence, units display syntagmatic non-discreteness (syncretism); classes, paradigmatic non-discreteness (statable in various ways, such as multidimensional classification, assignment of an item to different classes with variable probability, etc.). 45. I would class it with other dichotomies that Firth rejects: cf. Firth (1957b), “My own approach to meaning in linguistics has always been independent of such dualisms as mind and body, language and thought, word and idea, signifiant et signifie´, expression and content. These dichotomies are a quite unnecessary nuisance, and in my opinion should be dropped” (p. 227). Cf. Firth (1957c: 2, 3 – though here, I must admit, Firth also rejects “form and substance”, which I find crucial (as levels) to an understanding of how language works. 46. Cf. Firth (1957a): “It follows that the distinction between morphology and syntax is perhaps no longer useful or convenient in descriptive linguistics” (p. 14). 47. i.e. languages in which inflexional systems are a regular feature of word structure. Free and bound are generalized class categories, linked to the generalized structure categories of simple and compound (above, 2.3): “free” is “able to stand as exponent of one-element structure of the unit next above”, “bound” is “unable to stand, etc”. A member of a “free” class can thus be exponent of a “simple” structure, while a member of a bound class can operate only in “compound” structures. 48. Other terms are of course available, like Haas’ (1954: 68ff.) “synthetic classification” and “analytic classification”. The terminological objection to the use of “class” in both (as in “form class” and “function class”) is, however, theoretically founded: if we say, with Haas (p. 68), that “we distinguish two ways of classifying” linguistic units, we imply two things: (i) a choice of (ii) procedural direction. But this is not a procedural matter, and there is no choice. All forms are to be accounted for, and this means stating both their class (linking them to the unit next above) and their own structure (linking them to the unit next below). Whether

81

early papers on basic concepts the “syntactic” groupings, of items operating alike in the structure of the units next above, and the “morphological” groupings, of items alike in their own structure, coincide or not is a descriptive variable; other things being equal, a form of description will be chosen in which they do, since the more they coincide the more grammatical the statement. But they must be terminologically permitted not to coincide without prejudice to their both being stated. Cf. Robins (1959): “When there is a conflict of classification by morphological paradigm and syntactic function, the latter is given preference in assigning words to word– classes.” (p. 109) – I would add “groups to group classes, etc.” 49. For example in the structure of the English verbal group, the words work, play operate at the same element: they are members of the same word–class. The words works, working, worked do not operate at the same element: they are not members of the same word–class. They have themselves, however, the same (primary) structure: they are members of the same paradigm. Likewise in the structure of that class of the word containing the words worked, played, the morphemes work, play operate at the same element: they are members of the same morpheme class. 50. More delicate classes derived from secondary structures are referable both as exponents to secondary structures and as subdivisions (same degree of exponence, but more delicate) to primary classes. Diagrammatically: Scale of Delicacy Least delicate

Most delicate

䉳

Highest abstraction

䉱

Scale of Exponence 䉲

䉴

Primary structure

䉴

Secondary structure

䉲

䉲

Primary class

Secondary 䉴 class

Data 51. Or nearly coextensive: the criteria for the setting up of one primary class or two are descriptive. For example, in English clause structure S and C are different elements standing in different relation to P. There is a high degree of overlap between their exponents: one primary class (class “nominal” of unit “group”) can be set up as exponent of both S and C. The lack of exact coextensiveness will be stated by secondary elements and classes, to account for (for example) the occurrence of the old hall, the old town, the old town hall, this hall / town / town hall is old, this is a hall / town / town hall, and the non-occurrence of this old is a hall, this is an old or this hall is town.

82

categories of the theory of grammar 52. Cf. Firth (1957b): “Various systems are to be found in speech activity and when stated must adequately account for such activity. Science should not impose systems on languages, it should look for systems in speech activity, and, having found them, state the facts in a suitable language” (p. 144). Cf. also references given in 4.1 n. 32, above. 53. Again, abstraction on the exponence scale. The formal item the old man is exponent of (is a member of) a class (“nominal”, of the unit “group”). The class “nominal group” is exponent of (operates at the place of) an element of structure (S or C, of the unit “clause”). The formal item itself, of course, has its own (and ultimate) exponents in phonic or graphic substance. 54. Diagrammatically (axes as in 5.4 n. 50, above): Primary structure XYZY

䉳

SYSTEM

1 (at X) 2 (at Y) 3 (at Z)

Secondary structures X Ya Z Yc and X Yb Z Yc

2.1 (at Ya) 2.2 (at Yb) 2.3 (at Yc)

Secondary classes Primary classes 55. As already stressed (above, 2.2), the order of presentation here is for convenience of exposition; the relations among the theoretical categories do not involve logical precedence. 56. Since the categories are set up precisely to account for the data that are stated as their exponents, this is not surprising. Cf. Firth (1957a): “A theory derives its usefulness and validity from the aggregate of experience to which it must continually refer in renewal of connexion” (p. 1). The relation of category to exponent can be generalized as one of abstraction; one endpoint of this relation may be any one of four categories, but there is no scale of abstraction among the categories – their relation to each other is such that the move from any category to its exponent may be made either directly or via any or all of the other categories. As said below (6.2), the route may involve rankshift; but this does not mean that rank is to be equated with exponence or that there is any distinction between different units as regards the kind or degree of their relation to their exponents. (That is to say, even if one chooses to move from “clause” to exponent via “group”, this does not mean that the group is in any sense “nearer” to the data than the clause; indeed, the move from

83

early papers on basic concepts clause to exponent via group presupposes the possibility of moving from group to exponent via clause.) 57. So, for example, the formal item were driven may be exponent of: (i) the unit “group”, (ii) the element P in structure, (iii) the class “verbal” and (iv) the term “passive” in a system of secondary classes. All these statements are interdependent: the link of exponent to each theoretical category depends on its link to all the others and on their own interrelations in the theory. Thus the unit “group” is linked to the structure of the “clause”; the class “verbal” is a class of the unit “group” and is linked to the elements of structure of the clause; the system “voice” has as terms classes of the verbal group; these classes have their own structures, etc. 58. Thus: Units: sentence clause ⬙ group etc.

structures . . . (system of ) classes . . . structures (system of) classes . . . etc.

}

exponents

59. Without in any way affecting the syntactic nature of the “class”. 60. Cf. Garvin (1957): “Morphemes of limited membership class should be listed in the grammar and morphemes which belong to classes of unlimited membership should be exemplified in the grammar and listed in the dictionary” (p. 55). 61. Except in the sense that the description will always try a move down the rank scale as a possible way of extending its power (“remaining in grammar”). But wherever the lexical item is greater than a morpheme, its further analysis by grammar into morphemes will leave its lexical relations unaccounted for. For example, in the train left ten minutes late, but made it up, made up is a discontinuous verbal group analysed as two words, one (made) of two morphemes, the other simple; but it enters into an open set qua lexical item make up, which itself is here assigned to no grammatical unit. 62. See below, 8. 63. Regressive structures can of course be regarded as forming a scale; but their description does not require the introduction of a separate scale into the theory. Cf. above, 3.2 n. 26. 64. Cf. Allen (1956), from which the following is taken: “It frequently occurs that an appropriate ‘bit’ of the corresponding phonological statement (or of the orthography) is used as a label for the grammatical unit in question . . . The price of using such labels is constant vigilance . . . Where the phonological analysis permits of alternatives, that alternative is to be chosen which is most congruent with the grammatical

84

categories of the theory of grammar

65.

66.

67.

68.

analysis; . . . important correspondences may be observed between phonology and grammar, in so far as different phonological systems . . . may be required for the exponents of different grammatical categories – but the relation between them . . . is an indirect one via the phonic data” (pp. 143–5). In spite of the different formulation, and the difference between Allen’s diagrammatic representation and those used here (above, 1.7 n. 10, and below, 10.3 n. 88), I do not think that there is any conflict between Allen’s view and the view put forward here. In which (i) one set of units is set up for grammar, and their structures, classes and systems described, and (ii) one set of units is set up for phonology and these also appropriately described. Other things being equal, a form of phonological statement will be preferred which simplifies the statement of the relations between phonology and grammar. It is here that one asks such questions as: “Is it possible to generalize any phonological features as recognition signals of a given grammatical category?” and “Is it possible to specify under what conditions a phonological unit, such as the tone group, is exponentially coextensive with one or other grammatical unit, such as the clause or group?” One may want to compare primary and secondary structures of the same (class of the same) unit: shift in delicacy only. One may want to compare classes of one unit with classes of the unit next below: shift in rank only. Or one may want to state and exemplify the classes of a given unit: shift in exponence only. Cf. Firth (1957a): “The term exponent has been introduced to refer to the phonetic and phonological ‘shape’ of words or parts of words which are generalized in the categories . . . The consideration of graphic exponents is a companion study to phonological and phonetic analysis . . . The phonetic description of exponents which may be cumulative or discontinuous or both, should provide a direct justification of the analysis. It may happen that the exponents of some phonological categories may serve also for syntactical categories. But the exponents of many grammatical categories may require ad hoc or direct phonetic description . . . The exponents of the phonological elements of structure and of the . . . terms of systems are to be abstracted from the phonic material . . . The exponents of elements of structure and of terms in systems are always consistent and cannot be mutually contradictory” (pp. 15–16). I would regard the concept of exponents as one of Firth’s major contributions to linguistic theory. References will be found throughout the writings of Allen and of Carnochan, Euge´nie Henderson, Mitchell, Palmer, Robins, and other linguists of the School of Oriental and African Studies, London, many of which are referred to elsewhere in this paper. Strictly speaking the relation of the formal item to its exponent in substance entails a two-fold relation of abstraction, one of whose

85

early papers on basic concepts

69.

70.

71.

72.

dimensions is exponence (and is therefore a prolongation of the scale which relates category to formal item). The other dimension is the abstraction, by likeness, of a “general” event (class of events, though not in the technical sense in which “class” is used here) from a large number of “particular” events, the individual events of speech activity. For theoretical purposes the exponence scale can be regarded as comprehending this dimension of abstraction, which takes place then in that part of the scale which relates formal item to exponent in substance. Cf. Firth (1957a: 2; 1957b: 144, 187). For example, the statistical study of sequences of clause–classes, which is necessary both to the statement of sentence structures and to the description of a unit above the sentence, would reveal the range, and cline, of the determination of probabilities by the occurrence of a member of each class. (Cf. my review of Whatmough (Halliday 1958).) For example, a preliminary study of about 1000 items of the put up type, the purpose of which is to reveal the systems (dimensions) relevant to the identification and classification of so-called “phrasal verbs” in English, shows that fifteen different formal criteria yield fifteen different sets of classes. That is, have not yet been shown to be dependent on grammatical categories, and must therefore be postulated to be independent until shown to be otherwise: on the general theoretical principle that heterogeneity is to be assumed until disproved by correlation. Recent work by McIntosh (1961) suggests that lexical relations may, in some cases, be better described by reference to grammatical restrictions of variable extent; if so, this will affect both the theory of lexis and the relations between the levels of lexis and grammar. Cf. below, 8.2. Two familiar examples may be cited, both exponents of the grammatical unit “clause”: (i) John ran up a big hill and (ii) John ran up a big bill: (i) SPA verbal group ran

Primary structure: Exponent of P: class: of unit: formal item:

(ii) SPC verbal Group ran up

Down the rank scale, verbal groups ran and ran up: Primary structure: Exponent of F: class: of unit: formal item:

(i) F verb finite word ran

(ii) Fpo verb finite Word ran

In (i), the lexical item is ran, exponent of both the unit “group” and the unit “word”. If after further analysis ran is a compound word with two elements of structure whose exponents are morphemes, then the lexical

86

categories of the theory of grammar item is run which is an exponent of the unit “morpheme”. In (ii) the lexical item is ran up which is exponent of the unit ‘group’ but above the rank of the word. (Orthography does not of course provide criteria for grammatical units; like phonology, it must be stated separately and then related to grammar, though if a one / one correspondence between orthographic and grammatical categories is found to work at any point, so much the better. The space in fact will not do in every case as exponent of the boundary of the grammatical unit “word” in English. Here, however, to treat (things like) run up as a single word, the only purpose of which would be to maximalize the in any case very partial correspondence between “word” and lexical item, considerably complicates the description of the verbal group.) If the analysis is taken to the morpheme, the lexical item is run up, which contains one morpheme (but not the other) from the structure of the word ran plus another word from the structure of the group – it is both more and less than a grammatical word. Even if this type of morphemic analysis is rejected, the grammatically discontinuous lexical item will appear in John was running up a big bill, where the analysis of running as a two morpheme word structure presumably would be accepted; and even where all words have simple structure, as in John may run up a big bill, the lexical item, though not discontinuous, is an exponent of no grammatical unit, since it is more than a word but less than a group. 73. Cf. Chomsky (1957: 15 ff.). 74. For example by Chomsky (1957). Chomsky, however, does not countenance the formal study of lexis. In the present view, the concept of “grammatical” and “ungrammatical” is paralleled by that of “lexical” and “unlexical”, this being the basic reason why “the notion “grammatical in English” cannot be identified in any way with the notion “high order of statistical approximation to English”” (p. 16): colourless green ideas sleep furiously is “unlexical”. Lexical meaning in the present theory is thus not the same as is meant by Chomsky on p. 108; it is one part (grammatical meaning being the other) of the formal meaning of language, and “formal meaning” is one part (the other being “contextual meaning”) of the total meaning of language. 75. This view (that linguistics excludes the study of (non-formal) meaning) is probably no longer widely held. It is not within the scope of the present paper to discuss contextual meaning; but briefly, since context relates form to extratextual features, and is (like phonology) an interlevel, the categories of context, like those of phonology, are not determined by (still less, of course, do they determine) the categories of form; but contextual statement is required to account for all (instances of the) reflexion in form of extratextual features. For the statement of context, as distinct from either “content” or “concept”, see especially Firth (1957a: 5–13; 1957b: 227; 1957c: 2, 3; 1957d).

87

early papers on basic concepts 76. Cf. Firth (1957a: 11–13, 26–7, 31); Mitchell (1958: 108 ff.); Halliday (1959: 156–75). 77. Analogous to the morphological grouping (the “paradigm”) of grammar is the lexical “ordered series of words”: cf. Firth (1957b: 228) – though Firth’s “ordered series of words” includes what I should consider a “lexical set”, namely his “lexical groups by association”, these being (by analogy) “syntactical”. 78. The traditional vehicle of lexical statement, the dictionary, states formal meaning by citation and contextual meaning by definition. The theoretical status of lexicographical definition (Firth 1957a: 11 “shifted terms”) needs to be carefully examined. 79. Cf. above, 7.4 n. 71. The point, however, is: what is to be regarded as “one” lexical item? Dictionaries, in general, mix grammatical and lexical criteria: in the Shorter Oxford Dictionary, for example, cut v. (defined as “to penetrate so as to sever the continuity of with an edged instrument”) and cut sb. (“the act of cutting”) are shown as separate items having the same relation to each other as bear v. (“to support and remove; to carry”) and bear sb. (“a heavily-built, thick-furred plantigrade quadruped . . .”). But formally (quite apart from the fact that one contextual statement will cover both cut v. and cut sb. but not both bear v. and bear sb.) the pairs are quite distinct: there is a high degree of overlap in the range of collocation of cut v. and cut sb., which is not the case with bear v. and bear sb. Collocation provides a formal criterion for the identification of the lexical item. 80. An analogy which Bloomfield did use was that of the signal: “Accordingly, the signals (linguistic forms, with morphemes as the smallest signals) consist of different combinations of the signalling-units (phonemes), and each such combination is arbitrarily assigned to some feature of the practical world (“sememe”)” (1933: 162). This runs the risk of suggesting the analogy of a code – or even that language “is” a code. If language is a code, where is the pre-coded message? Cf. my review of Herdan’s Language as Choice and Chance (1959). 81. Detailed references are not given in this section. It is recognized, as already remarked, that what is here called “Bloomfieldian” method is an abstraction from a large body of descriptive work by different linguists, within which there is considerable variety and disagreement even on basic issues. Roughly it covers the work based on what Hockett called the “item and arrangement” model. It is not of course suggested that all the points made in this section are applicable to all such studies, nor all of them to any one study, within this type of linguistics. Since I have been concerned to apply the present theory in the description of English, many of the points made here were in fact first formulated with reference to Hill (1958), which incorporates what is probably the best comprehensive account of English grammar yet published and is an example of the

88

categories of the theory of grammar method here referred to. Hill’s book has recently been the subject of a review article by Haas (1960); my present aim differs from Haas’ in that I want to consider certain features of an approach in descriptive linguistics, exemplified by Hill’s work but also by many other studies, in the light of the theory here outlined. 82. Especially since a language may make systemic use of this variable (cf. 4.2 n. 37, above, where it is suggested that it is desirable to recognize in English a grammatical system the exponent of which is precisely the contrast between coextensiveness of the grammatical unit with one and with (a sequence of) two exponents of a phonological unit). 83. Cf. above, 4.2 n. 37. Hill’s proportion, however, is morph : morpheme : allomorph :: sound : phoneme : allophone. “Every morpheme must contain one phoneme and may contain several” (p. 89) and, for English, “the occurrence of any juncture always marks the boundary of an entity larger than a phoneme. The entity thus bounded may be word, phrase or sentence, but must always be at least a morph” (pp. 93–4). The phoneme thus enters into the statement both of the unit “morpheme” and of its structure. 84. That is, the final step of the formal statement, the move to the formal item. Strictly, since formal statement includes the placing of all forms at all units, it would be more accurate to say “though it is by no means always at morpheme rank that systems of formal items are to be found”. For example in English the items when, because, if, in case, provided that, etc., though they can of course be analysed into words and morphemes, operate as items at the rank of the group and, as such, are members of a particular class of the group. 85. So if we find in three languages items in substance statable phonetically as [pata] and [pate], these may yield: L1 L2 L3

Substance [a] / [c] ’

Form - (free variant) singular / plural (grammar) ‘cat’ / ‘dog’ (lexis)

’

(Phonologically restricted variants may be of either type: the language might have [pata] and [pete], never [pate] or [peta], without prejudice to whether the contrast between [pata] and [pete] is formally meaningful (L2,3) or not (L1).) 86. It does not matter, of course, what type of distinction is made in substance: the pair above could equally be replaced by for example [pa`t] and [pa´t], or [’pata] and [pa’ta], etc. 87. Again it does not matter where: the following grammatical contrasts in

89

early papers on basic concepts English are carried by substantial features requiring very different phonological statements: Form I work : I worked they’ve worked: they’d worked

Phonology (Unit) phoneme: ” :

he was working: was he working syllable: he was / working : he / was foot: working

//1 he was / working // : // he was / working //

4 tone group:

(Contrast) addition of segment /t/ replacement of /v/ by /d/ change in sequence shift of unit boundary (so re-distribution of strong and weak syllables – change of tonicity) replacement of one term in intonation system by another

The phonological statement cannot necessarily be expected to cover the systems concerned: it would be absurd, for example, to state the affirmative: interrogative system in phonological terms – though, as said below, it must carry the potentiality of being so stated, since it presupposes a distinction in substance. 88. Phonology relates form to substance by providing for a separate statement of abstraction from substance for those features that are formally meaningful. The relevant section of the diagram in 1.7 n. 10 above could be more delicately represented as: Substance

Form phonology 앖

with the dotted line representing the (logically) final stage in which two separate statements of abstraction are related. Cf. Robins (1959: 103): “(grammatical distinctions are) not deducible from the phonetic shapes, as such, of the words concerned nor from phonological rules based on these shapes and the phonetic categories involved in them . . . While both phonological and grammatical categories are abstraction from the phonic material of utterance, their relation to the phonic material is entirely different”. 89. For example the phonological system of intonation in English, operating at the rank of the phonological unit “tone group”, carries a number of different grammatical systems operating at different units in the grammar. 90. Thus, the clause in English is defined by its operation in sentence

90

categories of the theory of grammar

91.

92. 93.

94.

95.

structure, and by its own classes and their structures. The difference between two instances, such as an affirmative clause he saw them and an interrogative clause did he see them?, is of course ultimately statable in terms of substance; but grammar is not grammar if it tries to define the class system in this way – or even to state the difference phonologically at all. No linguist would ever try to state the grammar of clause–classes by reference to phonology; yet the attempt to define the unit “clause” by reference to phonological features such as juncture is no less objectionable – and leads, not surprisingly, to a phonology in which any substantial feature is a possible exponent of any term in the phonological system! A phonological description will, in this view, be prosodic if (i) it incorporates a rank scale, with a hierarchy of units to which contrasts are assigned, and (ii) it is polysystemic, so that, for example, the /t/ in 10.3 n. 87, above is not “the phoneme /t/” of “the English language” (no such entity will be postulated) but a phoneme identified by reference to its place in the structure of the unit concerned; this would still be true whether the phonological units concerned are (partly) taken over from grammar or are set up independently in the phonology. Firth stresses the very different nature of the “phoneme” in a description of this type, and prefers to use the distinct term “phonematic unit” (cf. 1957b, Chapter 9, passim). Cf. Haas’ reference to “a structure of a number of pyramids, all inverted” (1960: 267). I personally feel that English requires a totally different set of phonological units not derived from the grammatical units. Intonation in English needs a carrier unit “tone group” to display the (phonological) system of intonation; this system, and the terms in it, can then be related to the grammar. The attempt to describe intonation in a framework of “the intonation of the clause”, “of the group”, etc. is complicated and may lead to a misunderstanding of the operation of the intonation system. But the attempt is not theoretically sinful, as would be the attempt to describe the “grammar of the tone group”. As in Hill’s description of the English personal pronouns (1958: pp. 145–8). This “playing games”, or “party linguistics”, is again linked to the confusion of levels. Cf. Haas (1960: 273). I should thus agree with Robins (1959): “The morpheme must be recognized as the minimal element of grammatical structure; but this does not imply that it is the most suitable element to bear the assignment of all the grammatical functions fulfilled by the word into whose composition it enters” (pp. 127–8). I would not follow Robins, however, when he says that “In many ways . . . the word is a unique entity in grammar, and not just a stage in the progression ‘from morpheme to utterance’ ” (p. 137). Robins rejects morpheme-based grammar but

91

early papers on basic concepts

96. 97.

98.

99. 100.

101. 102.

103.

suggests replacing it by word-based grammar; what is here suggested is a “multi-unit” grammar in which no unit is “more unique” than any other. On the implications of upward description cf. Haas (1960: 263–9). In which both the first unit and the first step must be primitive terms of the theory; cf. Quirk (1959): “We are left with the impression, if only because nothing is said to the contrary, that ‘the first split . . . into immediate constituents’ (409) is achieved intuitively”. Similarly the binarity is, as Bazell points out (1953: 5), a feature of IC theory. For Chomsky likewise the sentence is a primitive term (1957: 30, 46), though this is not necessary to transformative–generative description as such (e.g. if it incorporated a rank scale, with rules for rank transformation). Cf. the section “Surface and deep grammar” (and the concept of “valence”) in Hockett (1958: 246–9). The scale of delicacy is introduced to account for what Hockett calls “deep grammar” (the “grammatical” nature of which he rightly stresses). It is worth insisting, however, that delicacy is a cline; and that a secondary statement, while accounting for the 10 per cent of instances not accounted for in the primary statement, at the same time yields a further set of categories and relations, and these are likely in turn to account for only 80 per cent of instances (these being now more delicately differentiated); these now demand a further step in delicacy – and so on. Of these four, only one does: the dog is chasing the cat, the dog is chasing the cats, the dog chased the cat, the dog chased the cats. Hockett’s statement of the link between the two (1958: 87), “In everyday parlance, this word means saying more than is strictly necessary . . . In modern information theory, the term has much the same meaning, but freed from the connotation of undesirability, and theoretically capable of precise quantification”, may I think be taken as underlying the uses of the term referred to in this section. In my view this formulation reduces a potentially powerful concept to a status where it is neither rigorous nor useful in linguistics. Harris recognizes this in his use of the “broken morpheme” (1951: 165–7). For example in the contrast between l’oeil and les yeux, or between have gone and were going. If, in Old English, a nominal group consisting of a noun alone may carry four case / number distinctions, one with adjective and noun six and one with deictic, adjective, and noun seven, how can any two case / number forms be considered exponents of “the same” category when they occur in different structures? Harris states this distributionally (1951: 205). Hill (1958: 477) rejects the redundancy of concord in Latin, on the grounds that it “sorts out the members of the sentence element or construction for us”, but accepts it

92

categories of the theory of grammar

104.

105. 106.

107.

108.

in Bantu “where there are repeated suffixes of agreement but in which the members of the same sentence element are continuous”. Quite apart from the arbitrary choice of assignment of redundancy, this is simply a shift of criteria: Bantu concord is still the exponent of a relation (since not all contiguous items are members of the same sentence element). But even if concord and contiguity were completely mutually determined, the problem would still not arise, since there would then be no valid reason for not treating the whole complex as a single exponent. Cf. Hill’s statement of the English affricates (op. cit.: 44): “In the system we have adopted, therefore, affrication has not been mentioned, since / c/ is distinguished from /t/ by its position, and the affrication is redundant”. Ebeling (1960: 30) rightly rejects redundancy in substance: “A choice of one of the equivalent features as relevant and the other as redundant is in such cases arbitrary and, therefore, senseless”. They might perhaps be acoustic, so that all but one of the formants which distinguish [a] from [i] would be redundant? By assigning contrasts where they belong. Cf. for example Carnochan (1952: esp. 94); and all works by linguists of the School of Oriental and African Studies, London, some of which are referred to throughout this paper. Cf. also Robins (1957). If polarity in English, which belongs to the group, is assigned to the word, or morpheme, “redundancy” arises: one can ask unreal questions such as “is the negative in didn’t go, in contrast to positive ‘went’, carried by the did or the n’t or the go”? If the category of number is assigned to the unit “word” in any language that has a nominal group which can select only one number at a time, there will be artificial “redundancy” whether there is concord, negative concord or no concord at all, the “redundancy” of complementary distribution. Again, as in phonology: Cantonese, for example, has pairs of syllables in which in final position short vowel plus long nasal consonant contrast with long vowel plus short nasal consonant. If these are phonemicized as for example /a:n/, / an:/, /a/ contrasts with /a:/ and /n/ with /n:/; if as /aan/, /ann/, /a/ contrasts with /n/ in penultimate position, but /an/ and /aann/ are absent: “redundancy” in either case. If the contrast is referred to an element in the structure of a higher unit, it can be stated as a single contrast of relative duration. Another use of this same “redundancy” which has not been mentioned here is contextual redundancy. This is used, for example, by Bull (1960: 16): “Unless a language is needlessly redundant, there is little or no likelihood that any tense system uses the point tensor formulas”; p. 27: “English is extremely redundant. It almost always defines the axes while Mandarin is extremely parsimonious. It defines the axes only to avoid confusion”. In other words, the form is said to have reflected more of the context than it need have done. This has the merit of having nothing

93

early papers on basic concepts

109.

110.

111. 112. 113.

whatever to do with the redundancy of information theory. What it has to do with is not yet clear; but it does pose interesting problems for contextual description and for comparison and typology. In fact, all description of language is the description of this “redundancy”. A language without it would presumably have to have only one sound, variable in duration, and only one unit with either no structure or no class. Language activity is a progression of events in environments; and as soon as we have stated the event (as one among a defined number of possible events, this number being always less than the total number of possible events in that language – class) and the environment (this being defined as not the same as all other environments – structure) there is “redundancy”. An extreme instance is found in Hill (1958: 26n.), where we are told that every audible exponent of /+/ is redundant, the one contrastive exponent being inaudible. Cf. Haas (1957: 37). Chomsky (1957: 55n.). Cf. Firth (1955: 93, 99; 1957a: 1). Chomsky (1957: 52).

94

Chapter Three

CLASS IN RELATION TO THE AXES OF CHAIN AND CHOICE IN LANGUAGE (1963)

This paper is concerned with the nature of the class as a theoretical term in Descriptive Linguistics: that is, with the meaning of “class” in statements which are made in order to explain how a given language works. Such statements are familiar in various forms, such as “there are eight word–classes in this language: noun, verb, adverb . . .” and “clauses can be classified into independent and dependent”. I have assumed, for the purpose of the main points made in the paper, that this category of “class” is to be defined syntactically. By this I mean that the concept is introduced into the description of a language in order to bring together those sets of items that have the same potentiality of occurrence; in other words, sets of items which are alike in the way they pattern in the structure of items of higher rank. Thus, to take a typical instance from grammar, we may have morpheme classes defined by word structure, each such class being one set of morphemes having a given value in the structure of words: as, for example, the morphemes of inflexion in Latin nouns. Likewise we might have word–classes defined by group structure, or clause–classes by sentence structure. This use of the term “class”, to name a category defined in some way by its relationship to a higher structure, is by no means universal in linguistics; but it would probably be granted that some such category is necessary to linguistic description whatever name we choose to adopt for it. Syntactic classification (sometimes referred to as “functional classification”, in what is perhaps a rather misleading opposition of

First published in Linguistics, 1963, 2, pp. 5–15.

95

early papers on basic concepts

“form” and “function”) is a central feature of linguistic method, and one which it seems appropriate to discuss in the present context. The alternative to this use of the term “class” is to consider morphological classification. Here “class” would be the name given to a set of items which are alike in their own structure: that is, in the way that they themselves are made up of items of lower rank. A word–class would then be a set of words having a certain similarity in their own formation out of morphemes. In this usage there are no morpheme classes, since “morpheme” is the name given to the smallest unit in grammar, which by definition has no structure: its relation to items abstracted at other levels, such as phonemes, is not one of structure, but involves the interrelation of different dimensions of abstraction. It is important to notice that this is in the first instance a terminological alternative, not necessarily implying a different theory. It is not the case that the linguist has to choose between two different kinds of classification, the syntactic and the morphological; he has in fact to recognize both kinds of likeness. Moreover, the sets of items identified on these two criteria often coincide: we may recognize a syntactic class “noun”, for example, defined as “that class of word which operates as head of a nominal group”, and find that the items grouped together on this criterion will be the same set as would be grouped together on a morphological criterion such as “that type of word which is made up of a stem morpheme followed by a morpheme of case and a morpheme of number”.1 Indeed other things being equal, it is usually accepted as desirable that the two should coincide: when the linguist is faced, as he often is, with a choice between two descriptions, both theoretically valid and both accounting for the facts, one in which the two assignments coincide and one in which they do not, he will normally, and “intuitively”, choose the former. For example, groups in English such as this morning operate in clause structure both as Adjunct, as in “I came this morning”, and as Subject (or Complement), as in “this morning promises to be fine” (or “I’ve set this morning aside for it”). The syntactic class defined by operation as Adjunct is the adverbial group; that defined by operation as Subject or Complement is the nominal group. Syntactically, therefore, this morning could be assigned to either or both of these classes. Morphologically, however, it clearly resembles other nominal groups (the morning, this man, etc.) rather than other adverbial groups (quickly, on the floor, etc.), and this can be allowed to determine its primary syntactic assignment. There are, however, clear instances where syntactically defined sets do not coincide with morphologically defined sets; and it would 96

class in relation to chain and choice

probably be generally agreed that, whatever the status accorded to the latter, the former cannot be ignored. Syntactic likeness must be accounted for. Moreover, even where the two sets do coincide, the criteria on which they have been established, and therefore their theoretical status, is different; and it is desirable that they should not be called by the same name. It seems to me appropriate that the term “class” should be reserved for the syntactic set (the morphological set may then be referred to as a “type”), and I propose to adopt this usage here. It is also true, in my opinion, that the class thus defined, the syntactic set, is crucial to the whole of linguistic theory, since it is required to give meaning to the basic concepts of “structure” and “system”; whereas the type, or morphological set, is more a descriptive convenience whose theoretical implications are largely internal to itself. In the remaining sections of this paper, therefore, I should like to discuss two aspects of the syntactically defined set, which I shall refer to henceforward simply as the “class”. I shall be concerned with its theoretical status in linguistic description, and shall confine myself to the level of grammar. The two aspects are (1) the relation of class to structure (the “chain” axis) and to system (the “choice” axis),2 and (2) the relation of class to the two kinds of structure found in language, the place-ordered and the depth-ordered. It is the class that enters into relations of structure and of system in language. A structure is an ordered arrangement of elements in chain relation, such as the English clause structure “predicator + complement” (for example fetch the ink). While (in this instance) the ultimate exponent of the element “predicator” is fetch3 and that of the element “complement” is the ink, the direct exponents of these elements are respectively the class “verbal group” and the class “nominal group”. Similarly: a system is a limited (“closed”) set of terms in choice relation, such as the English system of “number” (for example boy / boys). While (in this instance) the ultimate exponents of the terms in the system are boy and boys, the direct exponents of these terms are the class “singular nominal group” and the class “plural nominal group”. It is useful to be able to distinguish classes derived in these two ways: they can be referred to respectively as “chain classes” (those relating to structure) and “choice classes” (those relating to system). Difficulty is sometimes caused here by the need to recognize onemember classes. These do not really constitute a problem, being no different, qua classes, from the others; but the tendency to refer to them sometimes as category (by class name) and sometimes as item (by naming the unique member) may obscure the fact that the exponence 97

early papers on basic concepts

relation (that is, the relation of item to category) is unaffected by the fact that class membership is limited to one item. For example, in English the definite article the forms a one-member class; if we are describing the particular choice, in English grammar, that is exemplified in the man and a man it does not matter whether we state the terms in the system as the / a or as “definite article” / “indefinite article” (or other more appropriate class name): it is the class that enters into the relation of choice. It may be noted in passing that the one-member class has a particular significance in linguistic theory: if grammar is taken to be that part of linguistic form in which choices are “closed”, by contrast with lexis (vocabulary) in which they are “open”, then any item which can be shown to be the unique member of its class is fully and unambiguously identified in the grammar.4 Thus the can be shown to be grammatically distinct from all other items in the English language; whereas the grammar has no means of distinguishing, say, haddock from halibut, and this distinction must be accounted for in some form of lexical statement. It is a commonplace of linguistics that on the chain axis, that involving relations of structure, the value of sequence is variable. That is to say, the sequence in which items occur may or may not be a crucial property of the structure in question. It is important to realize, however, that this “may or may not be” is something of an oversimplification. To take an example, in the English clause John saw Mary yesterday the sequence is clearly crucial in one respect: John is Subject and Mary Complement; whereas in the clause Mary saw John yesterday Mary is Subject and John Complement. The Adjunct yesterday, however, remains Adjunct even if put at the beginning: yesterday John saw Mary. Now if there was no difference in meaning between John saw Mary yesterday and yesterday John saw Mary, we should be justified in saying that this particular feature of the sequence had no significance: it made no difference whatever to the structure. This, however, is manifestly not true: there is a difference in meaning, and although it does not seem so important as that between John saw Mary and Mary saw John, it certainly cannot be ignored. This problem can be handled through the concept of delicacy. The difference between “John saw Mary yesterday” and “yesterday John saw Mary” is still a difference of structure; but it is a more “delicate” distinction than that between “John saw Mary” and “Mary saw John”. It is perhaps doubtful whether there are any instances in language where a difference in sequence makes no difference whatever to the meaning, and therefore does not need to be recognized as expounding 98

class in relation to chain and choice

a distinct structure, though we should allow for such cases in the theory. But when we say that sequence “may or may not be” significant for structure, what we mean is that it may be significant at varying degrees of delicacy, down to a point where a distinction becomes so delicate that we do not know what to say about it; in such cases we may have to be prepared to treat the particular feature of sequence as being non-significant. This has important implications for the category of “class”. For each grammatical unit (i.e. sentence, clause and so on) in each language we can recognize “primary” (least delicate) elements of structure; for the clause in English, for example, the elements Subject, Predicator, Complement and Adjunct. From these we derive our primary classes: these are the sets of items of lower rank that enter into the primary structures with the value of the elements concerned. Thus the class corresponding to predicator is “verbal group”, that to both subject and complement is “nominal group”, and that corresponding to adjunct is “adverbial group”. Where the sets of items operating as two or more elements of structure show more than a certain degree of overlap, as in the case of subject and complement – most items that can be subject can also be complement, and vice versa – these are conflated into a single primary class: thus the “nominal group” is the primary class expounding both Subject and Complement in English clause structure. Primary classes are always chain classes. That is to say, the first step (on the scale of delicacy) is to state the classes derived from the primary elements of structure. When we take the class further in delicacy, however, and recognize “secondary classes”, some of these more delicate classes are chain classes and others are choice classes. It is clear that primary classes cannot be choice classes, since we cannot account for a choice until we have established that place in structure where the choice is made: for example, the choice classes “singular nominal group” and “plural nominal group” are meaningful only in relation to the primary (chain) class “nominal group” which defines the context of the choice. For examples of secondary classes of both types we may take an element of structure of the English nominal group, the element “Determiner”. The class of word defined as operating with this value is the class sometimes known as “deictic”. [Note: Halliday later reversed these terms, using Deictic as the function (element of structure) and determiner as the class. (Ed.)] This includes some forty items that are fully grammatical (i.e. reducible to one-member classes by successive steps in delicacy: the this that which whose my its a any some either each 99

early papers on basic concepts

other same certain, etc.), together with other items forming an open set (i.e. that cannot be so reduced: John’s, etc., including compound ones as in the railway company’s property). This class of deictic may be variously subdivided along both axes. On the one hand, there are certain sets whose members can occur in combination, as in all my other friends; there are in fact three such secondary groupings, the members occurring respectively in first, second and third place in a maximum sequence. This gives three secondary chain classes which may be called “predeictic” (for example all), “deictic” (for example my) and “postdeictic” (for example other). Within each of these three classes, choices are made. There are many ways of describing these, according to what are taken to be the principal dimensions. The deictic, for example, may be “specific” / “nonspecific” (my / every); “selective” / “non-selective” (my / the); and, as a further subdivision of the class formed by the intersection of “specific” and “selective”, it may be “possessive” / “demonstrative” (my / this). These and various other systems eventually yield, by their subdivisions and intersections, one-member classes: thus my can be uniquely classified as “deictic: specific, selective, possessive: personal: first person”. Secondary classes regularly cut across each other. The systems of “specification” and “selection”, for example, form a matrix as follows: Specific

Selective

this / these that / those

which what

my your our their his / her its John’s (etc.)

whose

the Non-selective

Non-specific both all every each no neither

a some any either another

It is not uncommon to find a large number of such intersecting classes, which may be very difficult to sort out; the above is only one of many possible ways of approaching the classification of the English 100

class in relation to chain and choice

deictics. But the patterns they display are typical in their complexity: a given class breaks down by simple subdivision into a system of more delicate classes, but the same original class will also subdivide in a number of different ways, so that many dimensions of classification intersect with one another. Any given item, to be fully identified, may require to be simultaneously classified on all such dimensions. In this way it can be assigned to a “microclass”, this representing its value in respect of all the properties which have been found relevant to the way it patterns in the language. There will be, of course, a very large number of such microclasses: for example, in a computational study of English “phrasal verbs” (items like take up, put on) which is being carried out at the moment, 557 such items were found to yield 125 microclasses. Up to this point I have been concerned only with “place-ordered” structures. These are sometimes thought of as being the normal type of linguistic structure. By a place-ordered structure I mean one composed of a limited number of different elements occurring nonrecursively. Such a structure may be fully class-defining, in the sense that to each element corresponds a distinct class of lower rank: for example the clause structure “Subject + Predicator”, with classes respectively “nominal group” and “verbal group”, as in “my friends have arrived”; or it may be only partially class-defining, where two or more elements are expounded by the same class but differentiated in sequence. In this type of structure, there is no constant relation between successive (or otherwise paired) elements: for example, in the structure ‘Subject + Predicator + Complement’ (for example John saw Mary, my friends have invited me) it is not true that Subject is to Predicator as Predicator is to Complement. This is not the only type of structure found in language, and there seems no particular reason for assuming it to be the norm, especially in its pure form. Language also exhibits a different kind of structure, the “recursive” or “depth-ordered” structure. Here, as the name implies, an element of structure, or a combination of elements, is repeated “in depth”, a series of such elements (or combinations) thus forming a progression. It is doubtful whether one should set a theoretical limit to the degree of depth in recursion; rather there appears to be some logarithmic scale of diminishing frequency, so that the number of observations one would expect to have to make before recording a depth of, say, ten would be extremely high. Spoken English seems to tolerate greater depth in recursion, or at least to tolerate it more readily, than written English; and this may be true of language generally. The 101

early papers on basic concepts

following are some examples of recursive structures in English, with depth indicated by the letters of the Greek alphabet. 1) Sentence structure: he might have come if you had told him when you rang him up a b g while he was packing before he went away d e

2) Clause structure (P = Predicator; C = Complement): she made them Pa Ca

stop him asking people to help him Pb Cb Pg Cg Pd Cd

do Pe

his homework Ce

3) Verbal group structure: (he)’ll have been going to be doing (it every day for a month soon)

This cannot be shown lineally, since the elements are not discrete; it is analysable as “present in future in past in future” and can be built up as follows: he does he will do present future he will be doing he will have done present in future past in future he was going to be doing he will have been going to do present in future in past future in past in future he will have been going to be doing present in future in past in future

4) Nominal group structure: flue d

pipe g

support b

strap a

It is a characteristic of recursive structures that they cannot be used to differentiate classes. Apart from the first term in the series, which may be distinguished by class (as in example 1, above, where the alphaelement is expounded by the class “independent clause”), each element of structure represented by a term in the recursive series has as exponent one and the same class. In example 1, the class “dependent clause” operates at beta, gamma, delta and epsilon: each item could occur at every place. (The items themselves are of course not necessarily mutually substitutable in every instance, since there may be more 102

class in relation to chain and choice

delicate class restrictions such as are shown by the different forms of asking and to help in example 2; these do not affect the principle.) Recursive structures are of two types: those involving “rankshift” and those not. The examples given above do not. Examples involving rankshift are: 5) Nominal group structure and adverbial group structure, both rankshifted (q = “qualifier” in nominal group; c = “prepositional complement” in adverbial group; [ ] = boundary of rankshifted group): the peartree

in [qa

the garden [ca

the bridge [cg

over [qd

in front of [qb

the house [cb

near [qg

the river [cd] 8

6) Nominal group structure and clause structure, clause only rankshifted ([[ ]] = boundary of rankshifted clause): this is the farmer

sowing his corn [[qa1

that crew in the morn [[qb . . . that lay in the house [[ql

that kept the cock [[qa2

that waked the priest all shaven and shorn [[qg that Jack built5 [[qm ]] 10

“Rankshift” is in fact merely a name for that type of recursive structure which cuts across the scale of rank. That is to say: in non-rankshifted structures, whether recursive or not, classes of each rank enter into a structure of the rank immediately above: in English, morpheme classes in word structure, word–classes in group structure, group classes in clause structure and clause–classes in sentence structure.6 In rankshift, this relation is broken and the classes enter into a structure of their own rank or even of lower rank than themselves, as in examples 5 and 6: the first shows classes of the group in group structure and the second shows a class of the clause, in this case the relative clause, also in group structure. Sometimes, as in the case of the relative clause, there may be a distinct class which occurs only with rankshifted status; but this is not necessarily so. In example 5, the adverbial group in the garden is rankshifted to the status of Qualifier in the nominal group the peartree 103

early papers on basic concepts

in the garden; but this class can also operate as Adjunct in the clause, being thus not rankshifted, as in he sat in the garden. Where a class may occur either rankshifted or not, often ambiguity may arise, as in take that chair in the garden: is the chair already in the garden (= ‘take that chair which is in the garden’; rankshift) or not (= ‘take that chair and put it in the garden’; no rankshift)? Some of the most complex problems in the description of a language arise where the same structure combines place-ordered with recursive elements. Perhaps the most striking example of this in English is the nominal group, which is a troublesome mixture of the two types; the earlier elements are largely place-ordered, recursive elements being increasingly tolerated as one approaches the Head (the head being, e.g., houses in the two old stone houses by the river) and continuing, by rankshift, thereafter. The element immediately following the Determiner, which may be called Ordinative and defines the classes of cardinal and ordinal numeral, together with superlatives, is already marginally recursive (e.g. the first three second best hotels); with later modifying elements preceding the Head this potentiality is greatly increased, giving items such as example 4 above and familiar also in the language of headlines: holiday coach death crash inquiry verdict. Moreover, the linear succession of the items does not act as a constant in showing the depth relation: compare 5-millimetre perspex boxes (gamma beta alpha) with 6-inch perspex boxes (beta beta alpha). The description of the word–classes entering into the structure of the English nominal group is extremely complicated if one treats it as a simple placeordered structure, with classes defined for each possible position, as the various attempts to do so have already shown. On the other hand it is unsatisfactory to treat the whole thing as recursive in structure and to recognize no secondary classes beyond the primary word–class “noun”. The facts of the language here lie in between the extremes of these two types of structure, and the best description seems to be one which takes this into account. These seem to me to be some of the problems that arise in intralinguistic classification at the level of grammar. Such problems are probably most acute at this level, but similar ones arise also at other levels, notably phonology, which has other additional classification problems of its own. Lexical classification is rather a different matter, and there are reasons for preferring not to use the term “class” in talking about lexis; but this subject would require a separate paper. The field of classification as a whole is one where linguists can learn much from disciplines faced with similar or related problems. At the same 104

class in relation to chain and choice

time perhaps the experience gained within linguistics may not be wholly irrelevant to other sciences.

Notes 1. I am assuming here the more “abstract” view of grammatical categories such as morpheme; cf. Palmer (1964b: 232–7). 2. For interdisciplinary purposes I have used here the terms “chain axis” and “choice axis” in place of their less self-explanatory technical equivalents “syntagmatic axis” and “paradigmatic axis”. 3. Strictly speaking the ultimate exponent is a token (“occurrence”) of the type (“formal item”) fetch. If, however, we confine ourselves to the level of grammar we can regard the formal item as the ultimate exponent. 4. Another way of drawing the same distinction between grammar and lexis is to say that grammar is “deterministic” by contrast with lexis which is “probabilistic”; in the sense that in grammar one can distinguish what is possible from what is impossible (before assigning probabilities, if one wishes, to what is possible), whereas in lexis one can only distinguish between what is more and what is less probable. 5. In fact the verse does contain rankshifted nominal groups, as “prepositional complement” in adverbial groups (for example that lay in the house), but these have been ignored in the illustration for the sake of simplicity. 6. This restriction to “the rank immediately above” implies a particular model of grammar more specifically than do most of the other points made here. A more generally valid formulation would be “enter into a structure of higher rank”.

105

Chapter Four

SOME NOTES ON ‘DEEP’ GRAMMAR (1966)

In the representation of syntagmatic relations in language, we may distinguish between a linear sequence of classes, such as “adjective followed by noun”, and a non-linear configuration of functions, such as “modifier-head relation” or simply “modification”.1 Both of these have been referred to as “structure”, although this term has also been extended to cover paradigmatic as well as syntagmatic relations. For Hjelmslev, for whom “structure” was not a technical term (see e.g. 1961: 74), “the structural approach to language . . . [is] conceived as a purely relational approach to the language pattern” (1948: quoted in Firth 1951: 220); among others who have emphasized the relational aspect of such studies are Firth2 (1951: 227–8; 1957: 17ff.; cf. Robins 1953; Palmer 1964a), Tesnie`re (cf. Robins 1961: 81ff.) and Pike (cf. Longacre 1964: 16). Chomsky’s (1964: 32) distinction, using Hockett’s terms, between “surface structure” and “deep structure”, “structure” here going beyond syntagmatic relations, is extremely valuable and widely accepted: the surface structure of a sentence is defined as “a proper bracketing of the linear, temporally given sequence of elements, with the paired brackets labelled by category names”, while the deep structure, which is “in general not identical with its surface structure”, is “a much more abstract representation of grammatical relations and syntactic organization”. A representation involving the concepts of class and sequence may thus be said to be a representation of surface structure. Here the ordering, if each pair of brackets is said to enclose an “ordered set” of First presented at a meeting of the Linguistics Association, Newcastle, March 1965. First published in Journal of Linguistics, 1966, 2(1), pp. 57–67.

106

some notes on ‘deep’ grammar

classes, is interpreted in the usual sense of the word, as linear successivity, or sequence. Such an interpretation does not preclude discontinuity or fusion of constituents, nor is it affected by the depth of the bracketing imposed: both the more copious bracketing of IC-type representations and the much sparser bracketing of, for example, a tagmemic analysis can adequately specify the relation of sequence in a surface structure. The labelling attached to the entities specified as entering into this relation of sequence may then be “class”-type labelling and interpreted as such, the class “adjective” being the set whose members are good, bad, . . .; although functional labels have also been introduced: for example Nida (1960) states generalized syntagmatic relations, such as hypotaxis, within the framework of an IC analysis. If the representation of syntagmatic relations is merely in terms of this type of surface structure, sequence is then the only determinable relation. A considerable amount of bracketing may be introduced in order to give as much information as is possible, within this limitation, about the syntactic relations involved. Class labels do not by themselves reduce the bracketing required, since classes do not fully specify syntactic function. Such labels may be conventionally interpreted as functional, but if so their correct interpretation depends on their association with a designated pair of brackets: for example “adjective” is to be interpreted as “modifier” when attached to a particular node in the tree. This adds considerably to the syntactic information; but if the tree itself represents sequence at the surface its application is limited (cf. Palmer 1964a). It has always been recognized that the concepts of class and sequence alone are inadequate for the representation of syntagmatic relations in language. Indeed the development of modern structuralism may be seen as having taken place in the context of a tradition in which it was the more surface elements that had remained least explicit. Relational terms like “subject” and “predicate” have always co-existed with class names such as “noun” and “verb”; and the definition of the classes has rested at least in part on syntactic criteria, so that the designation of an item by its class name indicates something of its potentiality of syntactic function. Classes were not thought of as specifying actual syntactic function within a given sentence, since the theory also recognized the deeper syntactic relations into which the classes entered; the attempt to combine morphological with syntactic criteria in the definition of the classes (since morphological “types” have to be accounted for somewhere), while it may lead to difficulties, is entirely explicable within such a framework. 107

early papers on basic concepts

While the underlying syntagmatic relations have been recognized as non-linear, or at least as not manifested in the linear sequence of the linguistic items, their representation, as Palmer (1964a: 125) points out, has usually involved some form of linear notation. Since there is also a level of abstraction at which the relevant syntagmatic relation is one of sequence, it may be important to recognize that two different kinds of representation are involved. In this sense class and sequence are inherently surface concepts, specifying the items of the language and their arrangement; this is no less true of syntactically defined than of morphologically defined classes, the former being merely sets of items identified as relevant to the deeper representation. For terminological simplicity we might perhaps here follow one tradition in referring to an arrangement of classes in sequence as a syntagm, reserving the term structure for a configuration of functions. If then function-type labels such as “modifier” are introduced, whether as such or as conventional interpretations of class-type labels, they will not be located in the syntagm, since their defining environment is not stated in terms of (its) sequence. This holds true even if in a given language (say) a modifierhead structure is always realized as a syntagm of adjective followed by noun; a structure is not defined by its realizations. The ordering that is ascribed to structure may be thought of in dependency terms, or in constituency terms as an underlying sequence which does not (necessarily) correspond to syntagmic sequence, or as mere co-occurrence or absence of ordering.3 In all cases it is of a different nature from syntagmic sequence, in that the components are functions, not sets of items. If (with Lamb) we use 쐌 to represent configuration, this being interpreted as “unordered with respect to syntagmic sequence, whether or not any other form of ordering is considered to be present”, then a structural representation may take the form m:h, or interchangeably h:m (modifier–head); this contrasts with a syntagmic representation of the form adj^n (adjective followed by noun). Representation such as m^h and adj:n would then appear as mixed types, where deep (structural) labels are attached to surface (syntagmic) relations, or vice versa. These might be given conventional interpretations, perhaps for example m^h as “modifier–head structure with realization by sequence alone” (i.e. where modifier and head are realized by the same class), adj:n as “modifier–head structure with realization by class alone” (i.e. where the classes may occur in either sequence); but these would be merely a shorthand for combining two types of representation. While many other formulations are possible, the recognition in some 108

some notes on ‘deep’ grammar

form or other of two distinct types of representation, linked by some form of “realization” relation,4 is relevant to the understanding of syntagmatic patterns, and the distinction can be made and discussed solely in terms of relations on the syntagmatic axis. Clearly, however, it is relevant also to relations on the paradigmatic axis. It may be helpful to relate this point to the distinctions made by Hjelmslev and by Firth. In Hjelmslev’s terms (1961: 38–9), linguistic function embraces both relation and correlation; relation is syntagmatic, within the semiotic process, or the text, while correlation is paradigmatic, within the semiotic system, or the language. While his view of the relation between the two axes was somewhat different from that of Hjelmslev, Firth likewise makes a terminological distinction, referring (1957: 17) to syntagmatic relations as relations of structure and to paradigmatic relations as relations of system. Provided there is at least some syntactic element in the definition of the classes concerned, even a syntagmic, class-sequence representation already gives some information about the paradigmatic relations into which an item enters, but to a very limited extent. This limitation is again inherent in the nature of the class–sequence concept: paradigmatic “relatedness” depends on a functionally defined environment. Two entities can only be said to contrast if they have a functional environment in common,5 and this environment is generally specified in terms of syntagmatic function; it presupposes therefore a representation of structure – i.e., of “deep” syntagmatic relations. The structural representation thus specifies the environment both for sets of paradigmatic relations and for further networks of syntagmatic relations, those within the lower-order constituents: for example the function of “subject”, itself specified syntagmatically in clause structure, defines an environment both for the syntagmatic relation of modifier–head and for the paradigmatic relation of singular / plural. For paradigmatic relations in the highest unit there is no functionally defined environment in this sense, so that if the sentence is said to be either declarative or interrogative this either / or relation rests on other grounds: the sentence as a primitive term, or some postulated higher unit not yet structurally described; the appeal to contextually-defined sentence functions such as “statement” and “question” is not one of these, this being rather a way of saying that declarative and interrogative have no environment in common. The paradigmatic contrasts associated with a given, defined environment may be thought of as being accounted for either in a single representation of “deep” grammar, in which are incorporated both 109

early papers on basic concepts

syntagmatic and paradigmatic function, or in a separate form of statement, distinct from, but related via the specification of the environment to, the statement of syntagmatic relations. Firth’s concept of the system embodies the second approach. The system may be glossed informally as a “deep paradigm”, a paradigm dependent on functional environment; in a sense, and mutatis mutandis, the relation of system to paradigm is analogous to that of structure to syntagm as these terms were used above. One could think of “paradigmeme” as a possible tagmemic term. In Hjelmslev the “system”, likewise a paradigmatic concept, is defined as a “correlational hierarchy”, the underlying notion being that of commutation (Hjelmslev 1961: 73–4). A system is thus a representation of relations on the paradigmatic axis, a set of features contrastive in a given environment. Function in the system is defined by the total configuration: for example “past” by reference to “present” and “future” in a three-term tense system, as structural function is defined by reference to the total structural configuration for example “modifier” by reference to “head”. If paradigmatic relations are represented separately in this way, this implies that the full grammatical description of a linguistic item should contain both a structural and a systemic component. It may be useful therefore to consider the notion of a systemic description as one form of representation of a linguistic item, the assumption being that it complements but does not replace its structural description. The systemic description would be a representation of the item in terms of a set of features, each feature being in contrast with a stated set of one or more other features: being, in Firth’s terms, a “term in a system”. This is exactly the sort of characterization that has been familiar for a long time in the form of “this clause is interrogative, finite, present tense, . . .”, given that we are told somewhere in the grammar not merely what other tenses, moods, etc. are found in the language but also which of them could have occurred in this particular clause all other features being kept constant. There is, however, one modification of a traditional “systemic description” of this kind which may need to be considered. This concerns the ordering of the features listed. In the traditional version they are unordered; but if the grammar specifies not only relevant systems but also their interrelations with one another, in particular their hierarchization on what I have called elsewhere (Halliday 1961: 272; 1964: 18) the “scale of delicacy”, then partial ordering is introduced. Any pair of systems, such that a feature in one may co-occur with a feature in the other in a systemic description, may be hierarchical or 110

some notes on ‘deep’ grammar

simultaneous; if two systems are hierarchically ordered, features assigned to these systems are ordered likewise. So for example the system whose terms are declarative / interrogative would be hierarchically ordered with respect to the system indicative / imperative, in that selection of either of the features declarative and interrogative implies selection of indicative. If this is taken together with another system, unpredicated theme / predicated theme, likewise dependent on indicative, then the item John has seen the play may be represented in respect of these features as: (indicative : (declarative / unpredicated theme)) where : indicates hierarchy and / simultaneity. Then it’s John who has seen the play contrasts with it in respect of one feature: (indicative : (declarative / predicated theme)) and is it John who has seen the play in respect of two features: (indicative : (interrogative / predicated theme)) The systemic description would represent a selection from among the possibilities recognized by the grammar. As far as these examples are concerned, the grammar would show that in a given environment selection is made between indicative and imperative; and if indicative is selected, there is also simultaneous selection between declarative and interrogative and between unpredicated theme and predicated theme, the two latter selections being independent of one another. Any item thus contrasts with others in respect of such features and combinations of features as the ordering of the system permits. For any set of systems associated with a given environment it is possible to construct a system network in which each system, other than those simultaneous at the point of origin, is hierarchically ordered with respect to at least one other system. The point of origin is specified syntagmatically, so that all features are associated with a syntagmatic environment; at the same time the system network provides a paradigmatic environment for each one of the features, specifying both its contrastive status and its possibilities of combination. It is not the aim here to present in detail the properties of a systemic description, but rather to discuss it in general terms. Systemic description may be thought of as complementary to structural description, the one concerned with paradigmatic and the other with syntagmatic relations. On the other hand it might be useful to consider some possible consequences of regarding systemic description as the under111

early papers on basic concepts

lying form of representation, if it turned out that the structural description could be shown to be derivable from it. In that case structure would be fully predictable, and the form of a structural representation could be considered in the light of this. It goes without saying that the concept of an explicit grammar implied by this formulation derives primarily from the work of Chomsky, and that steps taken in this direction on the basis of any grammatical notions are made possible by his fundamental contribution. My own more specific debt here is to Lamb, whose formalization of stratification theory is based on general notions closely akin to those which I had adopted (cf. Hockett 1965: 198). The present paper, however, attempts no more than an informal discussion of the question of a grammatical description in terms of features, here based on the notion of a feature as one of a set of contrastive “terms in system”. Presenting the systemic description of a linguistic item as the underlying grammatical representation of that item would seem to imply that its paradigmatic relation to other items of the language was in some way its more fundamental property, from which its internal (syntagmatic) structure is considered to be derived. This would seem to be Hjelmslev’s view, in his discussion of system and process (1961: 39–40). But the priority which is implied is not one between paradigmatic and syntagmatic relations as such, but rather between the external relations, both paradigmatic and syntagmatic, into which an item enters (the point of origin of a system network being defined in syntagmatic function) and its internal relations of structure. If one talks of simplicity, this means the simplicity of the whole description; underlying grammar is semantically significant grammar, whether the semantics is regarded, with Lamb, as input or, with Chomsky, as interpretation. What is being considered therefore is that that part of the grammar which is as it were “closest to” the semantics may be represented in terms of systemic features. This would provide a paradigmatic environment for the “relatedness” of linguistic items, a contrast being seen as operating in the environment of other contrasts. Structure would then appear as the realization of complexes of systemic features, involving in places both neutralization and diversification as defined in Lamb’s terms (Lamb 1964a: 64). If the structural representation is not required to account for paradigmatic relations, the question of how “deep” it needs to be is determinable by reference to other considerations: it should give an adequate account of syntagmatic relations, and permit the explicit realization of the systemic description in terms, ultimately, of a sequence of classes. 112

some notes on ‘deep’ grammar

This may be illustrated from the example it’s John who has seen the play. Leaving aside variation that is immaterial to the discussion, there would seem to be three possible representations of its structure: (1) it Subject, is Predicator, John who has seen the play Complement (2) it . . . who has seen the play Subject, is Predicator, John Complement (3) it’s John who Subject, has seen Predicator, the play Complement (1) would presumably be an attempt merely to state the simplest sequence of classes in the syntagm, although it could be shown to be unsatisfactory even on class-distributional grounds. (2) is distributionally acceptable and would account adequately for the syntagmatic relations; but it fails to account for the paradigmatic relations in that it does not show the “relatedness” of this clause to John has seen the play, etc. If the structural description is required to show the paradigmatic as well as the syntagmatic relations of the grammar we need some representation such as (3) in which John is the Subject. This leads to complexity in the realization, since a nominalization of the form it’s John who seems to add no new insight elsewhere in the grammar. A more serious difficulty arises in relation to the element “Subject” in English, which is a complex element within which it is possible to distinguish three components, or features; each of these may contrast independently of the other two, although there is a general, and generalizable, tendency to co-variation among them. The three contrasts can be seen independently in (i) John has seen the play, with tonic on play, versus, respectively, (ii) the play has been seen by John (Subject as actor versus Subject as goal); (iii) the play John has seen (= “the play, John has seen, but . . .”, Subject as theme versus Subject nonthematic); (iv) John has seen the play (with John tonic; Subject as “given” versus Subject as “new”). Each of these three is related paradigmatically to the original item, and each of them contrasts with it in respect of one feature only. By a further contrast, that of “unpredicated theme” versus “predicated theme”, (iv) is related to (v) it’s John who has seen the play with John tonic.6 Thus, despite the difference in constituent structure, (v) differs from (iv) in respect of only one feature. Such patterns, where different complexes of (paradigmatic) features may be combined in what is syntagmatically one and the same element of structure, here the Subject, involve some complexity for a structural description; if they were handled in systemic terms, the structure need represent only their realization in syntagmatic relations. We could then adopt a form of structural representation such as (2) above. 113

early papers on basic concepts

The examples cited might be regarded as irrelevant on the grounds that they do not involve cognitive distinctions and therefore belong to the realm of stylistic variation. But this is to assume that it is the task of a grammatical theory to differentiate between these different types of distinction. Such problems seem to me to fall more properly within the domain of a semantic theory, where the selection of a particular variable, such as paraphrase, as a basis for the classification of distinctions is not arbitrary as it seems to be in the grammar. This is not to deny that the speakers of a language recognize some distinctions as “more important” than others, and that this may depend at least in part on a concept of paraphrase. The hierarchization of systems in delicacy, in a system network, does seem to reflect some notion of the relative importance of the systems involved; this is an instance of the convergence of semantic and distributional criteria referred to by Lyons in his important discussion of semantics and grammar (1963: Chapter 2). Even if a clear answer can always be given to the question “is a a paraphrase of b or not?”, or to other questions where this is irrelevant (for example in the distinction between John has seen the play with John tonic and with play tonic, which answer different questions), the place of a given distinction in the grammar would, as I see it, depend on its environment in terms of other distinctions, this presupposing also its syntagmatic environment, rather than on a classification of its semantic function. To return to the discussion, another relevant factor here would be the desire to incorporate into the grammar phonological realizations of grammatical features, particularly (in English) those of intonation and rhythm. Such features may be assigned a place in a syntagmatic representation, either as superfixes in a syntagm or as elements in a structure; but the assignment of, say, a pitch contour as a constituent to a specified place in a structural representation, while it may be necessitated by the realization requirements, seems in other respects rather arbitrary. Intonation, in English, provides instances of both neutralization and diversification; one and the same feature may be realized in some environments by a structural pattern and in others by intonation, and a given intonation pattern may realize different features in different environments. In other words, intonation is not predictable from its structural environment. It can, however, be shown to be predictable in the grammar if it is regarded as a form of the realization of systemic features, at the same degree of abstraction (same stage of representation) as the structural elements but without constituent status. Intonation, however, is merely a special, if clear, case of a more 114

some notes on ‘deep’ grammar

general point: namely that if a representation in terms other than of constituent structure is adopted for the statement of paradigmatic relations, and is then made to determine the constituent structure, then provided the structural description adequately handles the syntagmatic relations there is no need for everything to be accounted for at a constituent stage of representation. This is most obviously relevant to phonological features of the prosodic type, but could be extended also to items identified as being markers of, rather than elements in, syntactic relations. The crucial factor in the designation of any feature as present in the grammar would thus be its assignment to a place in the systemic network. A putative feature which could not be shown to contrast independently with one or more others at some point would not be a distinct feature; each feature that is recognized is thus a term in a system, which system is located in hierarchical and simultaneous relation to other systems. The location is “polysystemic”: the recognition of a system, and the assignment of a feature to it, depends on the potentiality of contrast in the stated environment. For example, there might seem to be a proportionality in English such that he can go is to can he go as he is wondering if he can go is to he is wondering can he go; but the related features are different in the two cases: that is, the two syntagmatic environments determine different sets of paradigmatic relations. The ordering of the systems in delicacy would thus be important in the identification of the systemic features. It would be necessary also to specify the syntagmatic environment, in order to define the point of origin of a system network. This can be done in terms of the notion of rank, where the initial identification and labelling of certain stages in a constituent hierarchy in such general terms would provide a starting point for the delimitation of different more specific environments. The designation of rank, in other words, is a possible first step in the specification of what Haas (1966: 125) calls “functional relations”, relevant here in that it makes possible the assignment of a system to a place determined solely by constituent status (for example all clauses) and allows further specification of the environment to be in terms of features: a feature x may be associated with constituents having the features y and z rather than with constituents having a given syntagmatic function. The possibility of contrast between active and passive in the clause depends on other features of the clause, not on its function in the sentence. In stratificational terms, rank defines an inner series of strata, or substrata, within the outer grammatical stratum, with each rank character115

early papers on basic concepts

ized by a different network of systems. While many, though by no means all, features would be present at more than one rank, with consequent preselection at certain points, an important distinction is to be made here between preselection, where the choice of a feature at one rank determines the choice of a feature at a lower rank but the two operate in different paradigmatic environments, and the realization of a feature at a lower rank than that in which it has its environment. The latter includes such familiar instances as the realization in the structure of the word of a choice, such as that of number, associated with the group. There seems no reason to assume a necessary relationship between the rank at which a feature has its environment and that in whose constituent structure it is realized. The relevance of the concept of rank in this connection would thus be that it is as it were neutral between system and structure. While clearly a constituency notion, reflecting here the speaker’s awareness of the hierarchical organization of linguistic items, it imposes a minimum of bracketing and in this way facilitates the interrelating of paradigmatic and syntagmatic modes of representation. The discussion of “systemic description” here has been in terms of a rank-type constituent structure, since this would be one way of defining a point of origin for a system network: each system, like each structure, would be assigned to a given rank as its most generalized functional environment. It is not implied that a description in terms of features would necessitate a rank-type constituent structure, but rather that the status of constituents in the grammar would need to be brought into the discussion. Palmer (1964: 130) wrote: “Perhaps we need a pre-grammatical statement in which order is utterly divorced from sequence”. In this paper I am following up Palmer’s conclusion by asking whether such a statement could be thought of as a representation of the “deep” grammar. If deep grammar is equated with deep structure, in the sense of being thought of as relations of the constituency type, it may be difficult to avoid connotations of sequence and to solve some of the problems Palmer raised. If the underlying “order” is thought of as systemic, the more abstract representation of grammatical relations carries no implication of sequence. Sequence can be stated by reference to these, a “linguistic element whose exponent is sequence” having a status no different from that of others. Such a description is in a sense of the WP type, with word replaced by unit, or constituent, and paradigm by system. It is not suggested that paradigmatic relations are somehow “more important” than syntagmatic ones; but merely that a description in terms of features, if it can be made explicit, may help in 116

some notes on ‘deep’ grammar

bringing the “unidimensional time sequence” of language into relation with its deeper patterns of organization.

Notes 1. This paper was first presented at a meeting of the Linguistics Association, Newcastle, March 1965. I am grateful to R. M. W. Dixon and R. D. Huddleston for their subsequent valuable comments and suggestions. 2. Hass (1966: 131) points out that Firth’s position here has been misinterpreted; this may be partly due to his use of the term “element of structure” as a functional term. 3. Firth, perhaps somewhat confusingly, reserved the term “order” precisely for this non-linear relation among the components (“elements”) of a structure, contrasting it with “sequence”. 4. I use Lamb’s term “realization” instead of the earlier “exponence”. Lamb’s term is more widely known; it also corresponds closely to my own use, whereas as Palmer (1964b) pointed out my use of “exponence” differed materially from that of Firth. 5. This may be interpreted either as “if there is at least one set of conditions under which both could occur” or as “if both could occur under the given set of conditions”. It is the latter interpretation which I take to be the basis of (one aspect of) Firth’s ‘polysystemic’ approach. Firth himself was inconsistent in referring to a ‘system’ of word–classes noun, verb, etc. 6. Since the subject normally has the feature “given”, that of “new” being realized in other elements, the realization of the feature “new” in the subject is often accompanied by its predication as in it’s John who has seen the play. This explains why it’s John who has seen the play usually, though not obligatorily, has John and not play tonic, while the opposite is true of John has seen the play.

117

Chapter Five

THE CONCEPT OF RANK: A REPLY (1966)

I am most grateful to Peter Matthews (who I am pleased to note seems to deplore, as I do, “missionary fervour” in linguistics1) for allowing me the opportunity of replying in the same issue to his criticisms of the concept of “rank” and to the various other more or less related criticisms which he has incorporated into his discussion (1966). Having embarked on a detailed reply, I soon discovered that it would take up far too much space: for example, Matthews’ strictures on my use of terminology demanded, besides a lengthy documentation of my own care in this respect and of the reasons for particular choices, a demonstration, with citations, that my own practice in regard to terminology is in no way different from that of other linguists. Yet Matthews is well aware that I could cite numerous examples of the redefinition of traditional terms, especially polysemic ones like “word”, many of them much more at variance with earlier usage than mine – and presumably therefore much more likely to “mislead” the “public”. Such tu quoque arguments seem to me trivial; and yet how else can one answer a charge such as ‘If Halliday does not mean by “word”, in particular, what the ordinary linguist on the Clapham omnibus means by “word”, why has he never said so?’ other than by pointing out, what is surely obvious, that linguists do use terms in different ways and that apparently I have said so, since that is Matthews’ starting point? Such strictures seem designed merely to defy comment. A discussion of rank does, however, demand detailed consideration. The main issue, that of total accountability, has been a subject of discussion over many years; if Matthews is here covering old ground, First published in the Journal of Linguistics, 1966, 2(1), pp. 110–18.

118

the concept of rank: a reply

that is my fault rather than his. I plead guilty unreservedly to the charge of inadequate documentation of my own work, although the relevant bibliography of others’ writings is rather longer than Matthews seems to suggest. Some of the points raised in my paper, ‘Some notes on “deep” grammar’ (see above, Chapter 4), are I think relevant to Matthews’ comments; more important in this connection is Huddleston (1965), written without knowledge of Matthews’ article. Here I shall try to take up with reasonable brevity some of the issues as Matthews himself sees them. Two assertions seem crucial to Matthews’ argument. (1) A rank-free constituency grammar is to be preferred to a rank constituency grammar. (2) All formulations of rank grammars are either theoretically insignificant or empirically unsound: unsound if they differ materially from rank-free grammars and insignificant if they do not. (1) follows from (2), except that, if they do not differ materially, this by itself gives us no reason for preferring either. Let us examine each of these in turn. By a rank grammar I mean one which specifies and labels a fixed number of layers in the hierarchy of constituents, such that any constituent, and any constitute, can be assigned to one or other of the specified layers, or ranks. The European linguistic tradition, by its use of terms such as “sentence” and “clause”, has always implied the possibility of such a grammar, although, as is well known, the absence of criteria regulating the necessary modifications of the simple constituency relation led to various difficulties. I return below to the concept of rank as suggested in such formulations as “clause used as a word” and the like. A rank grammar is, as Matthews observes, a hypothesis about the nature of language. This leads us to ask, first, whether it can be falsified, and second, whether it is worth making in the first place. Like many other hypotheses, both in linguistics and in other disciplines, its empirical falsification, given that it cannot be logically disproved, is unlikely to take the form of the discovery of a clear counter-example – in this case, of a language which it is impossible to describe in rank constituent terms. It would be likely to take the form rather of demonstrating that the hypothesis prompts no interesting new questions and leads to unnecessarily complex or otherwise unsatisfactory accounts of the facts. Neither this limitation on the conditions of its falsification nor the fact that disagreements may arise (as in all subjects) over whether a given hypothesis has in fact been shown to be inadequate of themselves deprive it of interest. It seems to me that this is a hypothesis 119

early papers on basic concepts

which is capable of being falsified in these terms – that there are, in other words, conditions under which I would regard it as having been shown to be inadequate; and that such conditions do not obtain. We must then ask whether it is worth making in the first place, and basically there are two grounds for thinking that it is: its descriptive advantages, and the questions that follow from it. Among these, I suggest, are the following. It defines a point of origin for structures and systems, so that the assignment of any item to a given rank, as also the assignment of the structures and systems themselves, becomes an important step in generalization. To show that a system operates at a given rank is the first step in stating its relationship to other systems; likewise to assign an item to a given rank is the first step in stating the systemic and structural relations into which it may enter and those which it may embody within itself. On the structure axis, rank is a form of generalization about bracketing, and makes it easier to avoid the imposition of unnecessary structure. It also serves frequently to distinguish between similar structures, for example between defining and non-defining relative clauses in English. It may contribute towards a significant measure of depth (Huddleston 1965). It provides a point of reference for the description at other levels, such as phonology. These and other considerations suggest to me that the rank hypothesis, if valid, leads to a gain in descriptive power. Among the further questions that would follow from it are these. If some such form of hierarchical organization is universal, is the number of units also a universal, or is it a typological variable (in either case it is of interest)? Are certain paradigmatic or syntagmatic relations universally associated with specific ranks? Is there any statistical association among the relative frequencies of items of different rank? Is there any type of aphasia characterized by progressive, rank-by-rank loss of grammatical structures? Is there any reason why different languages have institutionalized different grammatical units in their orthographies or the same unit in different ways? Since Matthews considers the rank hypothesis to be of no significance unless accompanied by a requirement of total accountability (chain-exhaustive assignment to constituents) at each rank, we should perhaps ask what the alternative hypothesis is. A rank-free grammar is also a hypothesis about the nature or language, one which would seem to hold that the number of pairs of brackets in a given constituent hierarchy would enable us to make no predictions about the syntagmatic and paradigmatic relations to be encountered in association with 120

the concept of rank: a reply

a given pair of brackets. In other words, in a labelled constituent structure there will be no necessary association between the labels and the distance, in number of nodes, from the top of the tree, although it would still be possible to predict that certain syntagmatic and paradigmatic features would be associated with higher nodes than others. Furthermore, assuming that the bracketing has any meaning at all, it would seem to be implied that there is an unlimited number of stages where new paradigmatic and / or syntagmatic relations can be introduced; they must be new, since for the mere repetition of existing relations grammars of both kinds require some form of recursive mechanism. The rank-free constituency hypothesis is thus harder to falsify; to me it seems less satisfactory, but all I wish to suggest here is that some form of fixed constituency hypothesis about language does seem worth making. There is of course nothing new in this idea. Let us now take up the second point that rank grammars are either insignificant or unsound. Matthews writes: “If . . . we abandon this requirement (sc. of total accountability at each rank) by permitting both downward and upward rankshift, the concept of rank is at once stripped of its theoretical significance”. Downward rankshift does not affect total accountability, and is in any case merely a formulation in rankgrammatical terms of something that is explicitly provided for in all labelled grammars. It is therefore upward rankshift; which we must consider. It may be worth noting here the contrast between the tagmemic concept of hierarchy, which does include “level skipping” (upward rankshift) and is presumably for Matthews therefore devoid of significance, although he makes no mention of it, and the alternative possibility which I put forward and which is said to be unsound. Matthews makes it clear that he is not objecting to the use of traditional rank-type terminology: ‘. . . still less will we insinuate that there are “no such things” (to put it crudely) as words, phrases and clauses. On the contrary, . . . these constructs are useful’ (p. 102). Matthews’ reference to “the sense which practising linguists have adopted (sc. for these terms) in the past” suggests that there is something common to all the senses in which they have previously been used, from which I have suddenly departed; and that this is something other than the purely negative characteristic of not being defined in terms of rank, which would hardly seem sufficient to make them useful. What this is is not defined. The fact that the grammatical tradition has a set of terms sentence, clause, phrase and word is itself suggestive. What is more relevant is that it uses formulations like that mentioned above, “a clause used as a 121

early papers on basic concepts

word”, which suggest a clear understanding that constituents of a certain type have as it were an “unmarked” place of operation in the language relative to constituents of other types. In other words it may happen that, while the “normal” rank-value of the item operating as a constituent at a given place in structure is rank x, we also find instances where an item of rank y occurs instead. This intuition, though admittedly inadequate as formulated, seems to me basically sound; the questions are how it can be made explicit, and in particular, at this point in the discussion, what it implies about “upward rankshift”. The formulation “x used as y” implies that the terms appearing in it are useful precisely because they are not merely labels for morphological types but have certain characteristic functions associated with them. In cases of (downward) rankshift, an item normally having the function of (entering into the paradigmatic and syntagmatic relations associated with) rank x characteristically “loses” these functions on taking over those of rank y: a clause operating in group structure cannot enter into direct syntagmatic relations with clauses outside the structure of that group. There are good reasons, in other words, for saying that the relevant functional environment for who came to dinner, in the man who came to dinner, is that of group structure; hence the traditional label ‘clause used as a word’. I do not see any reason, however, for saying that therefore it ‘is’ a word; I did not know that this had been suggested, although it would not be so far from traditional usage as Matthews seems to imply.2 Since, however, Matthews is not questioning the notion of (downward) rankshift as such, but merely objecting to a transfer of rank labels which has not in fact been proposed, we can ignore this as common to all grammars and concentrate on the question of total accountability. In cases of so-called “upward rankshift” the situation is less clearcut. There seem to be no strong reasons for denying that never, in the sentence Never?, simultaneously contracts relations at more than one rank. The two cases are not parallel. While the consistency of the formulation “a clause used as a word” reflects a clear division between the external and the internal relations of the item, when a word is “used as” a clause, for example run in Run, if you can!, it is also sometimes said to ‘be’ a clause; and this uncertainty, between ‘used as’ a clause and “is” a clause, arises because the item has in fact some of the functional (external) characteristics of both ranks. This explains the tendency in constituency grammars to introduce singulary branching; and total accountability in a rank grammar is one way of determining the amount of singulary branching that is required. 122

the concept of rank: a reply

Of course total accountability of constituents at all ranks is open to argument, like many other matters in linguistics; one would expect it to be taken for granted that such things were extensively discussed, although Matthews’ crusading zeal on the part of those he considers less able to look after themselves seems to have led him to think it has never been questioned. As a hypothesis it has certain things to recommend it. Matthews’ formulation “extra turns round the mulberry bush” suggests that he finds it counter-intuitive; if so, this shows that we have different intuitions, since to me the notion that an item can be a constituent at more than one defined layer is, while not new, highly illuminating. Indeed it seems to me to be one of the requirements of a constituency grammar that it should allow some singulary branching of constituents. But once singulary branching is allowed, the grammar is no longer rank-free; the question is then not ‘is there a concept of rank?” but “how is this concept defined, such that there are at least some conditions under which singulary branching occurs (i.e. upward rankshift does not take place)?”. The fact that it matches my intuitions about constituent structure to say that in John ran, John and ran are constituents both of clause structure and of group structure is neither more nor less relevant than Matthews’ account of his intuitions in the course of his own discussion. What is more relevant, however, is that there may be something to be said about such items at more than one rank. If for example yes cannot “be” a clause, then the range of intonation patterns from which it selects in the sentence Yes!, which is that characteristic of clauses, and, moreover, of clauses of a particular class, already specified, to which it can be shown to belong on other grounds, has to be stated over again for the word; and then, since it is not in fact a system of the word, further restricted to words in that particular environment. It is much simpler to let rank scale define the environment. The reason for recognizing singulary branching is basically the same as that for recognizing a scale of rank, which in principle is what singulary branching already implies: it facilitates generalization about syntagmatic and paradigmatic relations. If we are not allowed to say that John in John ran is a constituent on two layers, whether or not these are labelled “word” and “group”, generalizations about both layers become more complicated. (This is not to say that, in the examples cited by Matthews, and must be a clause, which I did not know to have been suggested or implied.) Multiple rank assignment, in fact, is merely a formulation in rank-grammatical terms of the notion that a constitute may have only one constituent; if systems and structures are stated for 123

early papers on basic concepts

each rank, constituents being assigned on this basis to the rank appropriate to them, the multiple assignment of constituents is not only simpler than the restatement of relations but also avoids making the relatively surface notion of constituency as the basis of grammatical organization. While therefore in rankshift proper the only relevant environments are defined by the terminals of the embedding relation, so that in (the man) who came to dinner we specify the internal relations by the label “clause” and the external ones by reference to (realization of an element of) group structure, in the case of proposed “upward rankshift” the item may be entering simultaneously into more than one set of relations and it is this that is brought out by its assignment of an item to more than one rank. This seems no more absurd than the assignment of an item to more one class, rather than adjusting classes so that each item figures only once or denying the relevance of classes altogether. There is an analogy in the orthographic hierarchy. In the orthographic sentence I. it does not seem strange to say that this is a sentence consisting of one orthographic word and that this word consists of one letter. (Note that I have to specify “orthographic word” here, because of the different ways in which the term “word” has traditionally been used.) If we say that the sentence consists of one letter, complications arise: we have to restate the structure of the orthographic sentence in terms of letters as well as in terms of orthographic words, thus requiring among other things a second and much more complex statement of the distribution of punctuation marks; the form I will not appear at all in the set of orthographic words, but only in that of letters, of which we must then define a subset consisting of those that can operate in the structure of the sentence; and so on. How far Matthews is objecting to all singulary branching I do not know; he may want to say that a word can consist of one morpheme but a group cannot consist of one word or a clause of one group, for example. We have seen that this question is quite different from that of (downward) rankshift; that is, that the question whether after is assigned to the rank of group as well as to that of word is quite different from the question whether of the girl I saw last night is assigned to the rank of word (which for me it is not). The important point is not whether the grammar will make it possible for after to be assigned to the rank of group, since some degree of singulary branching is usually considered desirable, but whether it will require this assignment; and if it is not required, then under what conditions it will and will not take place. To require it, by demanding total accountability at all ranks, has 124

the concept of rank: a reply

the merit of underlying simplicity and, for some people at least, intuitive adequacy, but at the cost of some surface complexity; this is something to be avoided if possible, and Huddleston (1965) has introduced an important modification into the definition of rank which goes a long way in this direction. Given an adequate representation of the underlying grammar, there is no need to insist that every element should be assigned constituent status at all; it is quite usual not to recognize intonation features as constituents, and the same considerations could apply, as Matthews points out, provided limitations were stated, to markers such as and and or. I do not know how to specify in a general formation the conditions under which accountability in constituent terms would not be required. But this problem is no more difficult for a rank grammar, which has at least the concept of total accountability to refer to as a point of departure, than for a rank-free grammar, which has not. Matthews’ proposal (p. 104), that “Any generalizations we want to make can be made, once and for all, by specifying a single element of structure at which all the relevant classes or systems may operate: it does not matter if some of these classes are classes of phrases, words or morphemes rather than clauses”, is not a solution, since in many cases it does matter; otherwise no constituency grammar would have introduced singulary branching. Matthews has shown no convincing reasons for abandoning the generalizations which the notion of rank facilitates; to the extent that his objections are purely terminological they could of course be accommodated if one knew what they were. Even if all singulary branching was excluded, while this would be failing to exploit the concept of rank it would not render it insignificant. It would remain a form of labelling of greater generality than structure labelling; and it would embody the hypothesis that there are a finite number of points of origin of distinct systemic choices in language and a finite number of layers of constituent structure at which distinct types of syntagmatic relation are contracted. There remains the question of the king of England’s hat, where Matthews has invented a problem on my behalf by insisting that either the king of England or ’s must be a word. In fact no such requirement is implied. I am not sure whether Matthews wishes to suggest, as at one point he appears to do, that there is no problem here except the one he ascribes to me; to many linguists it has seemed that there is. It is beyond the scope of a discussion of this nature to attempt to do it justice, and I do not want to argue here in favour of any one approach; but a brief comment on Matthews’ own observations seems to be called for. 125

early papers on basic concepts

It is perfectly true that the fact that something is “in some sense” a word does not make it an immediate constituent of anything. But it is surprising to be told that ‘there is no argument for treating the king of England’s . . . as (a word) . . . which does not simply take as a premise the requirement that “each unit should be fully identifiable in description” ’. The suggestion that the king of England’s should be treated as a word, which is not a new one, has not so far as I know usually rested on such grounds; in any case we may reasonably ask for more than the mere reiteration that it “is” a phrase. Many questions need to be raised: what, for example, are the implications for constituent structure of the many cases where ’s and of are not parallel, for example that hat is the king of England’s but not this hat is of the king of England? How does the grammar take account of the fact that the placement of the ’s is paralleled by the placement of the tonic (in the king of England) which likewise depends on location in the syntagm and not on the constituent structure? and so on. If Matthews is suggesting that the problem is created, or even that it is complicated, by the introduction of rank into the grammar, it seems to me that he has given no evidence of this. Here also the issue seems to be partly one of labelling. This does not make it any the less appropriate for discussion; but whatever the usefulness, or otherwise, of the concept of rank in this and other contexts, it is surely something that can be discussed in its own linguistic terms.

Notes 1. In this connection I find the concept of ‘a “neo-Firthian” ’, especially one “committed” to certain “statements”, rather extraordinary. Must we all be labelled in this way? There is an important principle at stake here: that a scholar is responsible for what he says and writes, not for what others say and write. If I express agreement with something another linguist has put forward this neither makes that linguist responsible for my views nor commits me to acceptance of the whole of his. I may be wrong, but I feel that there are undesirable limitations on this principle inherent in Matthews’ first two paragraphs. 2. I am sorry if my own formulation was unclear here. But in that case the whole of this part of Matthews’ argument rests on one purely terminological point – since even if I was appearing to suggest that such an item should be labelled a word, no further use was being made of such a suggestion.

126

APPENDIX TO SECTION ONE

TABLE 1 Tonality: Distribution of utterance into tone groups (location of tone group boundaries) Tonicity: Distribution of tone group into tonic and pretonic (location of tonic foot) Tone (primary; pitch movement on tonic)

Tone (secondary)

This paper was written between May and August 1964 and formed the substance of a course on the description of English at the University of Indiana. First published in System and Function in Language, 1976, edited by G. R. Kress, London: Oxford University Press, pp. 101–35.

127

early papers on basic concepts

Symbols // tone group boundary (always also foot boundary) / foot boundary

— tonic syllable silent ictus . . . . pause

Tone, primary and secondary, is shown by Arabic figures, alone or with diacritics, placed immediately after the tone group boundary marker.

128

appendix

TABLE 2 Tonicity: location of ‘information focus’ tonic = final lexical item (neutral) tonic = pre-final item or final grammatical item (contrastive) Tone (assuming tonality neutral): Place of clause in sentence structure: final main 1, non-final coordinate 3, non-final subordinate 4 Declarative clauses reservation: 1 unreserved, 4 reserved involvement: 1 neutral, 3 uninvolved, 5 involved agreement: 1 neutral, 3 confirmatory, 2 contradictory information: 1 one information point, 13 two information points ‘key’: 1 neutral, 1 + strong, 1 – mild Interrogative clauses, WH-type ‘key’: 1 neutral, 2 (with final tonic) mild relation to previous utterance: 1 unrelated, 2 (with WH-tonic) echo Interrogative clauses, yes/no type ‘key’: 2 neutral, 1 strong involvement: 2 neutral, 3 uninvolved, 5 involved place in alternative question: 2 first alternative, 1 second alternative specification of point of query: 2 unspecified, 2 specified Imperative clauses ‘key’ (positive): 1 neutral, 3 moderate, 13 mild ‘key’ (negative): 3 neutral, 1 strong, 13 mild force: 1 neutral, 4 compromising, 5 insistent function: 1 etc. command, 2 question ‘Moodless’ clauses (also as declarative) function: 1 answer etc., 2 question, 3 warning, 5 exclamation In the section under ‘tone’, the headings (e.g. ‘reservation’) indicate the nature of the choice, the entries under each heading representing the terms in the choice with their appropriate tone. Thus ‘1 unreserved, 4 reserved’ means ‘in this choice tone 4 indicates reservation, by contrast with one 1 which indicates no reservation’. Secondary tones are indicated where relevant.

129

1

Systems of tone

appendix

2 2.1

Systems of the clause Dependent clause

131

early papers on basic concepts

2.2

Declarative clause

132

appendix

2.3

Interrogative 1

133

early papers on basic concepts

2.4

Interrogative 2

134

appendix

2.5

Imperative

135

early papers on basic concepts

2.6

Transitivity

extensive: (effective/operative: (goal-intransitive)) John threw; Mary washed (sc. the clothes) extensive: (effective/operative: (goal-transitive: non-benefactive)) John threw the ball; Mary washed the clothes extensive: (effective/operative: (goal-transitive: benefactive)) John gave the dog a bone; Mary washed the boys their clothes extensive: (effective/middle) Mary washed (sc. herself) extensive: (effective/receptive: (agent-oriented: non-benefactive)) the ball was thrown; the clothes were washed extensive: (effective/receptive: (agent-oriented: benefactive: (goal-receptive))) the bone was given the dog extensive: (effective/receptive: (agent-oriented: benefactive: (beneficiary-receptive))) the dog was given the bone extensive: (effective/receptive: (process-oriented)) the books sold; the clothes washed extensive: (effective/operative) the sergeant marched the prisoners extensive: (descriptive/middle: (+ range)) Peter jumped the wall extensive: (descriptive/middle: (ⳮ range)) Peter jumped; the prisoners marched extensive: (descriptive/receptive) the prisoners were marched intensive: non-benefactive Mary seemed happy; Mary made a good wife intensive: benefactive Mary made John a good wife

136

appendix

2.7

Theme

Examples of tone-expounded systems in declarative clause (from spontaneous conversation) //1 this of course de/pends on the / country where they / live // //- 1 . and / this is a / bit / hard // //- 1 . it’s / rather / interesting // // . . . 1 . it’s grade / one / two / three to / nine // // . . . 1 . of / vitamins and / sugars and / salts and . . . // 1 + . in fact the / smaller ones / eat the / bigger ones // // 1 + . well / yes they’re e/normously / long // // 1 – . I’m / not / sure that it’s / worth it // // 1 . and there was a / photograph of a / rabbit // I quivering his / ears a // 1 – lovely / white / rabbit and //1 – he was a / little man / sitting behind the / rabbit // //4 . oh / mine / aren’t parasites // //4 no worse than / anyone / else // //3 six / foot //3 I don’t know // //3 . they / just find a / comfortable / place in your / gut and they //3 stick their / hooks in and //1 stay there // /5 he was a / very / famous / man //

137

early papers on basic concepts //5 . it’s / very / interesting // ///- 3 that’s / right // //- 3 . yes it’s / not the / first thing // //2 yes they / do // //- 2 I don’t / know // // 13 . that’s the / trouble with / growing bac/teria in / culture // //13 . they / change peri/odically // //53 . they / do in / some uni/versities // //53 . and it / helps / them // //4 . in the / case of the / British ex/am I //1 don’t know / whether it / does // //4 . they / may have //I + pushed the / standard / up // //3 . fascinatingly e/nough I mean the //1 facts seem to be / fairly / true // //4 oh the ma/terial was //5 excellent // //2 . you’re not / serious //2 are you // //4 . no there / was a / Russian in the / first one //2 wasn’t there // //1 . and in fact / most of the / zoo department were / there //2 weren’t they // //5 . I’m / sure they do //2 don’t they // //1 Cambridge / always / was //1 wasn’t it // / 5 oh / no that’s / very em/barrassing //1 isn’t it // //1 used to be the / habit in / China //2 did it //

2.8

First order WH clause systems

Key to examples

Examples //1 who said / that // //2 who said / that // //1 who was it / said that // //2 who was it / said that // //2 it was / who said / that // //1 who did you / see // //2 who did you / see //

1

2

3

4

5

6

7

8

9 10

0 0 0 0 0 0 0

0 0 0 0 0 0 0

1 1 1 1 1 1 1

-

0 1 0 1 1 0 1

0 0 0 1

1 1 1 1 1 0 0

0 0

0 0 1 1 1 0 0

138

0 0 1 -

appendix //2 who was it you / saw // //2 where did he / go // //2 where was it he / went // //1 who said / that // //2 who said / that // //1 who was it / said / that // //2 who was it / said / that // //1 who did you / see // //1 who was it you / saw // //2 where did he / go // //1 where / was it he / went // //1 who wants / what // //2 who’s / going to sit / where // //1 which shall I / put / where // //1 John said / what // //2 John said / what // //1 put it / where // //2 John put / what there //

2.9

0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1

0 0 0 0 0 0 0 0 0 0 0 1 1 1 0 0 0 0

1 1 1 0 0 0 0 0 0 0 0 (1) (1) (1) (1) (1) (1) (1)

0 1 0 1 0 0 1 0 0 1 0 -

1 1 1 0 1 0 1

0 0 1 0 1 1 0 0 1

0 0 0 0 1 1 1 1 0 0 0 1 1 0 (0) (0) (0) (0)

0 1 1 1 0 1 1 0 0 1 0

1 0 1 0 0 1 1 0 1 0 1 0 0 0 0 0 0 0

0 0 0 0 0 0 -

Second order WH clause systems

Key to examples

Examples //2 . did / who do it // //1 . did / who do it // //2 did he do / what // //1 did he go / where // //1 who said / what // //1 who was it / said / what // //1 what did / who say // //2 where did / who go // //2 who went / where // //2 where was it / who went //

139

1

2

3

4

5

6

0 0 0 0 1 1 1 1 1 1

1 1 0 0 -

0 1 0 1 -

0 0 0 0 0 1 0 0 0 1

0 0 0 1 1 1

0 1 1 0 0

early papers on basic concepts

3

Imperative systems: jussive

Key to examples

Examples //1 ask / John // //1 you ask / John // //1 you ask / John // //1 let’s ask / John // //1 let’s ask / John // //1 do ask / John // //13 do ask / John // //1 you / do ask / John // //1 do let’s ask / John // //13 do let’s ask / John // //3 don’t ask / John // //13 don’t ask / John // //4 don’t / you ask / John // //1 don’t / let’s ask / John // // 1 let’s / not ask / John // //13 don’t let’s ask / John // //13 let’s not ask / John // //1 you / ask / John // //5 don’t let’s / ask John // //53 . let’s / ask / John // //1 you ask //4 John

1

2

3

4

5

6

7

8

9

0 0 1 0 1 0 1 1 0 1 0 1 1 1 1 1 1 1 1 0 -

0 0 0 0 0 0 0 0 0 0 0 1 1 -

0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 0 1 0 0

0 0 0 0 0 1 1 1 1 1 0 0 0

0 0 0 1 1 0 0 0 1 1 0 0 0 1 1 1 1 0 1 1 0

0 1 1 0 0 1 0 0 1 1 1

0 0 1 1 0 0 1 -

0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 1 -

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1

140

appendix

4

Theme systems: examples

Key to examples

Examples 1 //1 John / saw the / play / yesterday // 0 0 //13 John / saw the / play / yesterday // 0 //4 John //1 saw the / play / yesterday // //1 John / saw the / play / yesterday // 0 0 //1 John / saw the / play / yesterday // 0 //13 John / saw it / yesterday the / play // //13 . he / saw the / play / yesterday / John // 0 //4 . it / wasn’t / John that / saw the / play / yesterday // 0 //13 . it was / John that / saw the / play / 0 yesterday // //4 . it / wasn’t the / play John / saw / 1 yesterday // //13 . it was the / play John / saw / yesterday // 1 0 //4 John he //1 saw the / play / yesterday // 1 //1 yesterday / John / saw the / play // 1 //4 yesterday //1 John / saw the / play // 1 //1 . the / play / John saw / yesterday // //4 . the / play //1 John saw / yesterday // 1 //4 . it / wasn’t / yesterday John / saw the / play // 1

2 -

3 0 0 0 0 0 0 0

4 0 0 0 0 0 0 0

5 -

6 0 0 0 0 0 0 1

7 0 0 0 0 0 1 0

8 0 0 0 1 1 0 0

9 00 01 10 00 00 01 01

- 0 1 1 0 0 1 00 - 0 1 0 0 0 1 01 1 1 0 0 1 1

0 0 1 0 0 0 0

1 1 0 0 0 0 0

1 0 -

0 0 0 0 0 0 0

0 0 0 0 0 0 0

1 1 0 0 0 0 0

00 01 10 00 10 00 10

0 0 1 1 0 0 1 00

Examples of theme and information systems (from spontaneous conversation) in the first month one was too ill to move adjudicator I thought they were called the sound that went floating out on the air I didn’t know I had it in me aged legal gentlemen all like pipes the metal container somehow it turns your coffee rather sour

141

early papers on basic concepts Britain it’s all roads it’s inhaling that’s harmful it’s the side that has possession is at an advantage it was that part I didn’t enjoy it’s rather good coffee this it was quite fascinating to see her it does interest me how memory works imagining some suffering is worse than experiencing it oneself //this of course de/pends on the / country where they / live // // . I / thought / cats always / ate them // // how / long do these / changes / take // // . that’s why it’s so / awful to / have to get / rid of it // // . it / looked rather / odd having those / needles // // . no / I saw the / first one // // . but / in A/merica they //layer / things // // all the / dialect forms are //marked / wrong //

Paradigm examples corresponding to those on preceding page yesterday I saw John John I saw yesterday John I saw him yesterday I saw John yesterday John yesterday he saw me John he saw me yesterday it was John that saw me yesterday it was John saw me yesterday it was John I saw yesterday he saw me yesterday John it was strange to see John it was strange how I saw John seeing John was strange //. I / saw / John / yesterday // // I saw / John / yesterday // //. I / saw / John // //. it was / strange to / see / John // //. it was / strange / seeing / John // // I saw / John / yesterday // // yesterday I // saw / John // // I saw / John // yesterday //

142

appendix

5

Systems of the verbal group 1

6

Systems of the verbal group 2

143

early papers on basic concepts

7

Verbal group

7.1 te

Verbal group 1 td

tg

tb

past in

present in

future in

past in

144

future in

ta past present future past present future past present future past present future past

1 2 3 4 5 6 7 8 9 10 11 12 13

appendix

present in

7.2

present in

past in

present in

future in

future in

past in

past in

future in

past in

present in

past in

future in

present in

future in

past in

past in

future in

past in

Verbal group 2

Tense (major system) 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

took/did take takes/does take will take had taken has taken will have taken was taking is taking will be taking was taking is going to take will be going to have taken was going to have taken is going to have taken will be going to have taken

145

present future past present future past present future past present future past present future past present future past present future past present future

14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36

early papers on basic concepts 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36

7.3

had been taking has been taking will have been taking was going to be taking is going to be taking will be going to be taking had been going to take has been going to take will have been going to take had been going to have taken has been going to have taken will have been going to have taken was going to have been taking is going to have been taking will be going to have been taking had been going to be taking has been going to be taking will have been going to be taking had been going to have been taking has been going to have been taking will have been going to have been taking

Verbal group 3

Tense (minor system: modal / non-finite) I II III IV V VI VII VIII IX X XI XII

to take, taking; can take (= 2) to have taken, having taken; can have taken (= 1, 4, 5) to be taking, being taking;* can be taking (8 =) to be, being; can be + going / about to take (= 3, 11, 12) to be, being; can be + going / about to have taken (= 6, 14, 15) to have, having; can have + been taking (= 7, 16, 17) to be, being; can be + going / about to be taking (= 9, 20, 21) to have, having; can have + been going / about to take (=10, 12, 23) to have, having; can have + been going / about to have taken (=13, 25, 26) to be, being; can be + going / about to have been taking (= 18, 29, 30) to have, having; can have + been going / about to be taking (= 19, 31, 32) to have, having; can have + been going / about to have been taking (= 28, 34, 35)

Note: 24, 27, 33, 36 have no equivalents in this system. * The form being taking is now (predictably!) attested in regular use.

146

appendix

7.4

Verbal group 4

Tense (minor system: sequent) 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36

had taken took would like (1) (1) would have taken had been taking was taking would be taking had been going to take was going to take would be going to take had been going to have taken was going to have taken would be going to have taken (7) (7) would have been taking had been going to be taking was going to be taking would be going to be taking (10) (10) would have been going to take (13) (13) would have been going to have taken had been going to have been taking was going to have been taking would be going to have been taking (19) (19) would have been going to be taking (28) (28) would have been going to have taking

147

early papers on basic concepts

8 8.1

Nominal group 1 Principal systems

148

appendix

8.2

Determiners

149

early papers on basic concepts

8.3

Determiners

150

appendix

8.4

Quantifiers

151

This page intentionally left blank

SECTION TWO WORD–CLAUSE–TEXT

This page intentionally left blank

EDITOR’S INTRODUCTION

In the second section, which includes works spanning two decades from the mid-1960s to the mid-1980s, we observe how Halliday’s developing theoretical framework is applied in the analysis and description of patterns at various linguistic levels ranging from lexical item to clause to text. The clause as lexico grammatical construct is different in kind from both lexis and text. While each is defined differently – lexical item defined by reference to collocation, clause by reference to structure, and text by reference to context of situation – they are nevertheless analogous in nature and systemic in orientation. ‘Lexis as a linguistic level’ (Chapter 6), published in 1966, first appeared in a volume dedicated to Halliday’s mentor, Professor J. R. Firth. Noting the importance Firth gave to lexical studies in descriptive linguistics, Halliday explores some of the issues involved in handling lexical patterns in language, which he regards as being different in kind, not just in degree of delicacy, from grammatical patterns. What Halliday is proposing is a lexical model “with distinct, though analogous, categories and forms of statement”. In lexical analysis, the occurrence of an item appears to have more to do with collocational restrictions than a “known and stated set of terms in choice relation”. This being the case, Halliday concludes, “even such a thing as a table of most frequent collocates of specific items, with information about their probabilities, unconditioned and lexically and grammatically conditioned, would be of considerable value for those applications of linguistics in which the interest lies not only in what the native speaker knows about his language but also in what he does with it”. Appearing in John Lyons’ New Horizons in Linguistics in 1970, Halliday’s paper on ‘Language structure and language function’ 155

word–clause–text

(Chapter 7) introduces the three grammatically relevant ‘language functions’: ideational, interpersonal and textual. Similar to the Prague linguists, Halliday’s study of grammar represents a synthesis of both structural and functional approaches. Grammar is described by Halliday as a system of available options from which the speaker or writer selects “not in vacuo, but in the context of speech situations”. An act of speech involves “a simultaneous selection from among a large number of interrelated options”. These options, otherwise referred to as the ‘meaning potential’ of language, “combine into a very few relatively independent ‘networks’; and these networks of options correspond to certain basic functions of language”. Thus the functions of language are reflected in the structure of the clause. Halliday proceeds to show how each of the functions is reflected in the structure of the English clause, beginning with the realization of ideational meaning in terms of transitivity structure, involving the linguistic expression of process, participant and circumstance. Halliday also looks at how interpersonal meaning is captured in the mood structure of the clause, and how the textual function is expressed in both thematic and information structures. In ‘Modes of meaning and modes of expression: types of grammatical structure, and their determination by different semantic functions’ (Chapter 8), published in 1979 as part of the Festschrift for William Haas, Halliday borrows Pike’s insight into language as particle, wave and field to distinguish between experiential structures which are constituency-based (particle-like), interpersonal structures which are prosodic (field-like) and textual structures which are periodic (wavelike). Halliday also recognizes, as a distinct component, the logical mode, in which “reality is represented in more abstract terms, in the form of abstract relations which are independent of and make no reference to things”. Chapter 9 is based on what originally appeared as two separate papers ‘How is a text like a clause?’ and ‘Text semantics and clause grammar: some patterns of realization’. Both were written in 1980. Making it clear that he considers a text to be a semantic rather than a formal lexicogrammatical entity, Halliday argues that ‘the elements of structure of the text are more abstract; they are functional entities relating to the context of situation of the text, to its generic properties in terms of field, tenor and mode’. But having noted that texts and clauses have two distinct natures, texts being semantic and clauses being lexicogrammatical, Halliday proceeds to point out how they are alike, metaphorically speaking. While on the one hand, clauses are the 156

editor’s introducton

constituents, or building blocks, of the text, they also have ‘evolved by analogy with the text as model, and can thus represent the meanings of a text in a rich variety of different ways’. In ‘How is a text like a clause?’, Halliday elaborates on how the textual properties of structure, coherence, function, development and character have their analogous counterparts in the organization of the clause. The final paper in this section, ‘Dimensions of discourse analysis: grammar’ (Chapter 10), published in 1985, illustrates the application of systemic-functional grammar to the analysis of a sample of spoken language, i.e. a discussion between an adult and three nine-year-old schoolgirls. The analysis is presented in ten steps, ranging from transcription of intonation and rhythm, through lexicogrammatical analysis, to description of context of situation in terms of field, tenor and mode: 1. transcription and analysis of intonation and rhythm 2. analysis into clauses and clause complexes, showing interdependencies and logical semantic relations 3. analysis of clauses and clause complexes, for thematic (Theme– Rheme) structure 4. comparison of clauses and information units, and analysis of the latter for information (Given–New) structure 5. analysis of finite clauses for mood, showing Subject and Finite 6. analysis of all clauses for transitivity, showing process type and participant and circumstantial functions 7. analysis of groups and phrases (verbal group, nominal group, adverbial group, prepositional phrase) 8. analysis of grammatical and lexical cohesion 9. identification, rewording and re-analysis of grammatical metaphors 10. description of context of situation and correlation with features of the text The goal of the analysis is to show how the text being studied derives from the linguistic system and how it comes to mean what it does.

157

Chapter Six

LEXIS AS A LINGUISTIC LEVEL (1966)

At a time when few linguists, other than lexicographers themselves, devoted much attention to the study of lexis, and outlines of linguistics often contained little reference to dictionaries or to other methods in lexicology, J. R. Firth repeatedly stressed the importance of lexical studies in descriptive linguistics.1 He did not accept the equation of “lexical” with “semantic”,2 and he showed that it was both possible and useful to make formal statements about lexical items and their relations. For this purpose Firth regarded the statement of collocation as the most fruitful approach, and he sometimes referred, within the framework of his general views on the levels of linguistic analysis, to the “collocational level”.3 The aim of this paper is to consider briefly the nature of lexical patterns in language, and to suggest that it may be helpful to devise methods appropriate to the description of these patterns in the light of a lexical theory that will be complementary to, but not part of, grammatical theory. In other words, the suggestion is that lexis may be usefully thought of (a) as within linguistic form, and thus standing in the same relation to (lexical) semantics as does grammar to (grammatical) semantics, and (b) as not within grammar, lexical patterns thus being treated as different in kind, and not merely in delicacy, from grammatical patterns. This view is perhaps implicit in Firth’s recognition of a “collocational level”.4 One of the major preoccupations of grammatical theory in presentday linguistics is the extension of grammatical description to a degree First published in In Memory of J. R. Firth, 1966, edited by C. E. Bazell, J. C. Catford, M. A. K. Halliday and R. H. Robins, Longman, pp. 148–62.

158

lexis as a linguistic level

of delicacy greater than has hitherto been attained, and it is rightly claimed as a virtue of contemporary models that they permit more delicate statements to be made without excessive increase in complexity. A grammar is expected to explain, for example, the likeness and unlikeness between this brush won’t polish and this floor won’t polish, the three-way ambiguity of John made Mary a good friend, and the nonacceptability of beautiful hair was had by Mary beside the acceptability of the last word was had by Mary. Such explanations require the recognition of distinctions which, as is well known, begin to cut across each other at a relatively early stage in delicacy, and the model has to accommodate cross-classification of this kind. The form of statement adopted (and the terminology) will of course depend on the model; but it is generally agreed that all such patterns need to be accounted for.5 As part of the process of accounting for these distinctions the grammar attempts, both progressively and simultaneously, to reduce the very large classes of formal items, at the rank at which they can be most usefully abstracted (for the most part generally as words, but this is merely a definition truth from which we learn what “word” means), into very small sub-classes. No grammar has, it is believed, achieved the degree of delicacy required for the reduction of all such items to one-member classes, although provided the model can effectively handle cross-classification it is by no means absurd to set this as the eventual aim: that is, a unique description for each item by its assignment to a “microclass”, which represents its value as the product of the intersection of a large number of classificatory dimensions. If we take into account the amount of information which, although it is still far from having been provided for any language, contemporary grammatical models can reasonably claim to aim at providing, there would seem to be two possible evaluations of it. One is that, when the most delicate distinctions and restrictions in grammar have been explained, all formal linguistic patterns will have been accounted for; what is left can only be accounted for in semantic terms. The second is that there will still remain patterns which can be accounted for in formal linguistic terms but whose nature is such that they are best regarded as non-grammatical, in that they cut across the type of relation that is characteristic of grammatical patterning. The particular model of grammar that is selected may suggest, but does not fully determine, which of these two views is adopted operationally. For example, a model which distinguishes sharply between the grammar of a language and the use of the grammar, regarding corpus-based statistical statements as proper only to the latter, 159

word–clause–text

and therefore as outside the range of validity of a descriptive statement, is less easily compatible with the second view than is a model which does not make this distinction and which allows statistical statements a place in linguistic description; nevertheless it is not wholly incompatible with it.6 Lexical statements, or “rules”, need not be statistical, or even corpus-based, provided that their range of validity is defined in some other way, as by the introduction of a category of “lexicalness” to parallel that of grammaticalness. One may validly ask whether there are general grounds, independent of any given model, for supplementing the grammar by formal statements of lexical relations, at least (given that the aim of linguistics is to account for as much of language as possible) until these are shown to be unnecessary. It may be a long time before it can be decided whether they are necessary or not, in the sense of finding out whether all that is explained lexically could also have been incorporated in the grammar; there still remains the question whether or not it could have been explained more simply in the grammar. The question is not whether formal lexical statements can be made; they are already made in dictionaries, although at a low level of generality, in the form of citations. The question of interest to linguists is how the patterns represented by such citations are to be stated with a sufficient degree of abstraction, and whether this can best be achieved within or outside the framework of the grammar. Let us consider an example. The sentence he put forward a strong argument for it is acceptable in English; strong is a member of that set of items which can be juxtaposed with argument, a set which also includes powerful. Strong does not always stand in this same relation to powerful: he drives a strong car is, at least relatively, unacceptable, as is this tea’s too powerful. To put it another way, a strong car and powerful tea will either be rejected as ungrammatical (or unlexical) or shown to be in some sort of marked contrast with a powerful car and strong tea; in either case the paradigmatic relation of strong to powerful is not a constant but depends on the syntagmatic relation into which each enters, here with argument, car or tea. Grammatically, unless these are regarded as different structures, which seems unlikely, they will be accounted for in a way which, whatever the particular form of statement the model employs, will amount to saying that, first, strong and powerful are members of a class that enters into a certain structural relation with a class of which argument is a member; second, powerful (but not strong) is a member of a class entering into this relation with a class of which car is a member; 160

lexis as a linguistic level

and third, strong (but not powerful) is a member of a class entering into this relation with a class of which tea is a member. It would be hoped that such classes would reappear elsewhere in the grammar defined on other criteria. Argument, car and tea will, for example, already have been distinguished on other grounds on the lines of “abstract”, “concrete inanimate” and “mass”; but these groupings are not applicable here, since we can have a strong table and powerful whisky, while a strong device is at least questionable. The same patterns do reappear: he argued strongly, I don’t deny the strength of his argument, his argument was strengthened by other factors. Strongly and strength are paralleled by powerfully and power, strengthened by made more powerful. The same restrictions have to be stated, to account for the power (but not the strength) of his car and the strength (but not the power) of her tea. But these involve different structures; elsewhere in the grammar strong, strongly, strength and strengthened have been recognized as different items and assigned to different classes, so that the strong of his argument has been excluded on equal terms with the strong of his car. Strong and powerful, on the other hand, have been assigned to the same class, so that we should expect to find a powerful car paralleled by a strong car. The classes set up to account for the patterns under discussion either will cut across the primary dimension of grammatical classification or will need to be restated for each primary class. But the added complexity involved in either of these solutions does not seem to be matched by a gain in descriptive power, since for the patterns in question the differences of (primary) class and of structure are irrelevant. Strong, strongly, strength and strengthened can all be regarded for this present purpose as the same item; and a strong argument, he argued strongly, the strength of his argument and his argument was strengthened all as instances of one and the same syntagmatic relation. What is abstracted is an item strong, having the scatter strong, strongly, strength, strengthened, which collocates with items argue (argument) and tea; and an item power (powerful, powerfully) which collocates with argue and car. It can be predicted that, if a high-powered car is acceptable, this will be matched by a high-powered argument but not by high-powered tea. It might also be predicted, though with less assurance, that a weak argument and weak tea are acceptable, but that a weak car is not. As far as the collocational relation of strong and argue is concerned, it is not merely the particular grammatical relation into which these items enter that is irrelevant; it may also be irrelevant whether they enter into any grammatical relation with each other or not. They may be in 161

word–clause–text

different sentences, for example: I wasn’t altogether convinced by his argument. He had some strong points but they could all be met. Clearly there are limits of relevance to be set to a collocational span of this kind; but the question here is whether such limits can usefully be defined grammatically, and it is not easy to see how they can. The items strong and power will enter into the same set as defined by their occurrence in collocation with argue; but they will also enter into different sets as defined by other collocations. There is of course no procedural priority as between the identification of the items and the identification of the paradigmatic and syntagmatic relations into which they enter: item, set and collocation are mutually defining. But they are definable without reference to grammatical restrictions; or, if that is begging the question, without reference to restrictions stated elsewhere in the grammar. This is not to say that there is no interrelation between structural and collocational patterns, as indeed there certainly is; but if, as is suggested, their interdependence can be regarded as mutual rather than as one-way, it will be more clearly displayed by a form of statement which first shows grammatical and lexical restrictions separately and then brings them together.7 If therefore one speaks of a lexical level, there is no question of asserting the “independence” of such a level, whatever this might mean; what is implied is the internal consistency of the statements and their referability to a stated model. Possible methods of lexical analysis, and the form likely to be taken by statements at this level, are the subject of another paper by J. McH. Sinclair.8 Here I wish to consider merely some of the properties of this type of pattern in language, and some of the problems of accounting for it. Clearly lexical patterns are referable in the first place to the two basic axes, the syntagmatic and the paradigmatic. One way of handling grammatical relations on these two axes is by reference to the theoretical categories of structure and system, with the class definable as that which enters into the relations so defined. In lexis these concepts need to be modified, and distinct categories are needed for which therefore different terms are desirable. First, in place of the highly abstract relation of structure, in which the value of an element depends on complex factors in no sense reducible to simple sequence, lexis seems to require the recognition merely of linear co-occurrence together with some measure of significant proximity, either a scale or at least a cut-off point. It is this syntagmatic relation which is referred to as “collocation”. The implication that degree of proximity is here the only variable does not of 162

lexis as a linguistic level

course imply how this is to be measured; moreover, it clearly relates only to statements internal to the lexical level: in lexicogrammatical statements collocational restrictions intersect with structural ones. Similarly in place of the “system” which, with its known and stated set of terms in choice relation, lends itself to a deterministic model, lexis requires the open-ended “set”, assignment to which is best regarded as probabilistic. Thus while a model which is only deterministic can explain so much of the grammar of a language that its added power makes it entirely appropriate for certain of the purposes of a descriptive grammar, it is doubtful whether such a model would give any real insight into lexis. Collocational and lexical set are mutually defining as are structure and system: the set is the grouping of members with like privilege of occurrence in collocation. Second, in grammar a “bridge” category is required between element of structure and term in system on the one hand and formal item on the other; this is the class. (This specific formulation refers to the “scale-and-category” version of a system-structure model; but it is probably true that all models make use of a category analogous to what I am here calling the “class”.) In lexis no such intermediate category is required: the item is directly referable to the categories of collocation and set. This is simply another way of saying that in lexis we are concerned with a very simple set of relations into which enter a large number of items, which must therefore be differentiated qua items, whereas in grammar we are concerned with very complex and variable relations in which the primary differentiation is among the relations themselves: it is only secondarily that we differentiate among the items, and we begin by “abstracting out” this difference. In other words there is a definable sense in which “more abstraction” is involved in grammar than is possible in lexis. Third, the lexical item is not necessarily coextensive on either axis with the item, or rather with any of the items, identified and accounted for in the grammar. For example, on the paradigmatic axis, in she made up her face one can identify a lexical item make up1 whose scatter and collocational range are also illustrated in your complexion needs a different makeup. This contrasts with the lexical item make up2 in she made up her team and your committee needs a different makeup. That the distinction is necessary is shown by the ambiguity of she made up the cast, she was responsible for the makeup of the cast. Grammatically, the primary distinction is that between made up and makeup; this distinction of course involves a great many factors, but it also relates to many other items which are distinguishable, by class membership, in the same way. If the 163

word–clause–text

grammar is at the same time to handle the distinction between make up1 and make up2 it must recognize a new and independent dimension of class membership on the basis of relations to which the previous dimension is irrelevant. Any one example can of course be handled by ad hoc grammatical devices: here for instance the potentialities of make up2 in intransitive structures are more restricted. But such clearly grammatical distinctions, even when present, are so restricted in their range of validity that the generalizing power of a grammatical model is of little value as compared with the cost, in increased complexity, of the cross-classifications involved. It may be worth citing a further example of a similar kind. We can distinguish grammatically, but not lexically, between they want the pilot to take off (= “so that they can take off” and = “they desire him to do so”): these are not necessarily distinguished by intonation, although the unmarked tone selections are different. On the other hand it is easier to distinguish lexically than grammatically between he took two days off (= “he did not work”) and (= “he reduced the time available”). In the takeoff of the president (= “his becoming airborne”) and (= “the imitation of the president”) the distinction can usefully be made both in grammar and in lexis. On the syntagmatic axis, it may be useful to recognize a lexical item which has no defined status in the grammar and is not identified as morpheme, word or group. For example, in he let me in the other day for a lot of extra work, one could handle let in for as a single discontinuous item in the grammar; but this complexity is avoided if one is prepared to recognize a lexical item let in for without demanding that it should carry any grammatical status. Similarly the ambiguity in he came out with a beautiful model may be explained, instead of by giving two different grammatical descriptions, by identifying two distinct lexical items, come and come out with (and of course two different lexical items model). It is not suggested of course that such non-coextensiveness between the items of grammar and those of lexis is the norm, but merely that for certain purposes it is useful to have a descriptive model of language that allows for it. At the same time the above considerations suggest that the lexical component requires not, as it were, a second ‘runthrough’ of the model designed for the grammar but rather a specifically lexical model with distinct, though analogous, categories and forms of statement. Nor is it suggested that the set of patterns recognized as language form is neatly divided into two types, the grammatical and the lexical. A model for the description of language form may recognize only one 164

lexis as a linguistic level

kind of pattern and attempt to subsume all formal relations within it: some grammatical models, as has been noted, envisage that it is the grammar’s task to distinguish strong from powerful as well as to distinguish a from the and “past” from “present”; while a lexicographical model in which a and the, as well as strong and powerful, are entered in the dictionary and described by means of citations could be regarded as in a similar way attempting to subsume grammar under lexis. Even where the model recognizes two distinct kinds of pattern, these still represent different properties of the total phenomenon of language, not properties of different parts of the phenomenon; all formal items enter into patterns of both kinds. They are grammatical items when described grammatically, as entering (via classes) into closed systems and ordered structures, and lexical items when described lexically, as entering into open sets and linear collocations. So in a strong cup of tea the grammar recognizes (leaving aside its higher rank status, for example as a single formal item expounding the unit “group”) five items of rank “word” assignable to classes, which in turn expound elements in structures and terms in systems; and the lexis recognizes potentially five lexical items assignable to sets. But, to take a further step, the formal items themselves vary in respect of which of the two kinds of pattern, the grammatical or the lexical, is more significant for the explanation of restrictions on their occurrence qua items. The items a and of are structurally restricted, and are uniquely specified by the grammar in a very few steps in delicacy; collocationally on the other hand they are largely unrestricted. For the item strong, however, the grammar can specify uniquely a class (subclass of the “adjective”) of which it is a member, but not the item itself within this class; it has no structural restrictions to distinguish it from other members of the class (and if the members of its ‘scatter’ strong, strength, etc., turn out to operate collocationally as a single item then this conflated item is not even specifiable qua class member); collocationally, however, it is restricted, and it is this which allows its specification as a unique item. There might then appear to be a scale on which items could be ranged from “most grammatical” to “most lexical”, the position of an item on the scale correlating with its overall frequency ranking. But these are three distinct variables, and there is no reason to assume a correlation of “most grammatical” with either “least lexical” or “most frequent”. The “most grammatical” item is one which is optimally specifiable grammatically: this can be thought of as “reducible to a one-member class by the minimum number of steps in delicacy”. Such an item may or may not be “least lexical” in the sense 165

word–clause–text

that there is no collocational environment in which its probability of occurrence deviates significantly from its unconditioned probability. In a lexical analysis it is the lexical restriction which is under focus: the extent to which an item is specified by its collocational environment. This therefore takes into account the frequency of the item in a stated environment relative to its total frequency of occurrence. While a and of are unlikely to occur in any collocationally generalizable environment with a probability significantly different from their overall unconditioned probabilities, there will be environments such that strong occurs with a probability greater than chance. This can be regarded, in turn, as the ability of strong to “predict” its own environment. As extreme cases, fro and spick may never occur except in environments including respectively to and span (the fact that to and fro accounts for only a tiny proportion of the occurrences of to, while spick and span may account for all occurrences of this item span, is immaterial to the specification of fro and spick); here it is likely that, for this very reason, to and fro and spick and span are to be regarded as single lexical items. It is the similarity of their collocational restriction which enables us to consider grouping lexical items into lexical sets. The criterion for the definition of the lexical set is thus the syntactic (downward) criterion of potentiality of occurrence. Just as the grammatical system (of classes, including one-item classes) is defined by reference to structure, so the lexical set (of items) can be defined by reference to collocation. Since all items can be described lexically, the relation of collocation could be regarded as being, like that of structure, chain-exhausting, and a lexical analysis programme might well begin by treating it in this way; but this is not a necessary condition of collocation, and if closed-system items turn out, as may be predicted, to be collocationally neutral these items could at some stage be eliminated by a “deletion-list” provided either by cross-reference to the grammar or, better, as a result of the lexical analysis itself. Once such “fully grammatical” items are deleted, collocation is no longer a chain-exhausting relation. Moreover, while grammatical structures are hierarchically ordered, so that one can recognize a scale of “rank” each of whose members is a chain-exhausting unit (text items being then fully accounted for in sentence structure and again in clause structure and so on), it does not seem useful to postulate such an ordered hierarchy for lexis. Lexical items may indeed enter into a sort of rank relation: it is likely, for example, that on collocational criteria we would want to regard stone, grindstone and nose to the grindstone each as a separate lexical item, and though triads of this kind may be rare it looks as though we need the 166

lexis as a linguistic level

categories of simple and compound, and perhaps also phrasal, lexical items, in addition to collocational span, as units for a lexical description. Since the only “structural” relation in lexis is one of simple cooccurrence, these represent a single serial relation: the item stone enters (say) into the collocation grindstone, which then does not itself collocate exactly like the sum of its parts but enters as an item into (say) the collocation nose to the grindstone, which likewise does not collocate like the sum of its parts but enters as an item into (say) the collocation he’s too lazy to keep his nose to the grindstone. The first stage of such compounding yields a morphological (upward) grouping of items, the lexical series which, like its analogue in grammar, may or may not coincide with the syntactic grouping recognized as a set: oaktree ashtree planetree beechtree presumably do operate in the same set, while inkstand bandstand hallstand grandstand almost certainly do not. The series is formed of compound items having one constituent item in common; this item, here tree and stand, is the “morphologically unmarked” member of the series and, likewise, if the series forms a set it may or may not be the “syntactically unmarked” member of the set. Equivalence or non-equivalence between series and set is an interesting feature of lexical typology: one would predict that in Chinese, for example, practically all such series do form sets (with an unmarked member), whereas in Malay and English they very often do not. The lexical item itself is of course the “type” in a type–token (item– occurrence) relation, and this relation is again best regarded as specific to lexis. The type–token relation can be made dependent on class membership: just as in grammar two occurrences assigned to different primary classes, such as ride (verb) and ride (noun), can be regarded as different (grammatical) items, so in lexis two occurrences assigned to different primary sets can be regarded as different lexical items. This can then be used to define homonymity: if the two occurrences of model in the example above are shown to differ according to criteria which would assign them to different sets then they represent two homonymous items. It is not to be assumed, of course, that grammatically distinguished items such as ride (verb) and ride (noun) may not also operate as distinct lexical items, as indeed they may; merely that if they turn out to belong to the same set they will on that criterion be said to constitute a single lexical item, as also will strong, strength, strongly and strengthen, and perhaps also (if they can be suitably delimited) noncognate “scatters” such as town and urban. This would provide a basis for deciding how many lexical items are represented by “expressions” such as form, stand and term. 167

word–clause–text

If we say that the criterion for the assignment of items to sets is collocational, this means to say that items showing a certain degree of likeness in their collocational patterning are assigned to the same set. This “likeness” may be thought of in the following terms. If we consider n occurrences of a given (potential) item, calling this item the node, and examine its collocates up to m places on either side, giving a span of 2m, the 2mn occurrences of collocates will show a certain frequency distribution. For example, if for 2000 occurrences of sun we list the three preceding and three following lexical items, the 12,000 occurrences of its collocates might show a distribution beginning with bright, hot, shine, light, lie, come out and ending with a large number of items each occurring only once. The same number of occurrences of moon might show bright, full, new, light, night, shine as the most frequent collocates. On the basis of their high probability of occurrence (relative to their overall frequency) in collocation with the single item sun, the items bright, hot, shine, light, lie, come out constitute a weak provisional set; this resembles the weak provisional class recognizable in the grammar on the basis of a single “item-bound” substitution frame – although in lexis it is relatively less weak because of the lower ceiling of generality: lexis is more item-bound than grammar. If we intersect these with the high frequency collocates of moon we get a set, whose members include bright, shine and light, with slightly greater generality. That is to say, bright, shine and light are being grouped together because they display a similar potentiality of occurrence, this being now defined as potentiality of occurrence in the environment of sun and in that of moon. The process can be repeated with each item in turn taken as the node; that is, as the environment for the occurrence of other items. The set will finally be delimited, on the basis of an appropriate measure of likeness, in such a way that its members are those items showing likeness in their total patterning in respect of all those environments in which they occur with significant frequency. This is of course very much oversimplified; it is an outline of a suggested approach, not of a method of analysis. As Sinclair has shown, however, methods of analysis can be developed along these lines. Many other factors are involved, such as the length of the span, the significance of distance from the node and of relative position in sequence, the possibility of multiple nodes and the like. One point should be mentioned here: this is the importance of undertaking lexicogrammatical as well as lexical analysis. It is not known how far collocational patterns are dependent on the structural relations into which the items 168

lexis as a linguistic level

enter. For example, if a cosy discussion is unlikely, by comparison with a cosy chat and a friendly discussion, is it the simple co-occurrence of the two items that is unlikely, or their occurrence in this particular structure? All that has been said above has implied an approach in which grammatical relations are not taken into account, and reasons have been given for the suggestion that certain aspects of linguistic patterning will only emerge from a study of this kind. But it is essential also to examine collocational patterns in their grammatical environments, and to compare the descriptions given by the two methods, lexical and lexicogrammatical. This then avoids prejudging the answer to the question whether or not, and if so to what extent, the notion of “lexicalness”, as distinct from “lexicogrammaticalness”, is a meaningful one.9 An investigation on the lines suggested requires the study of very large samples of text. The occurrence of an item in a collocational environment can only be discussed in terms of probability; and, although cut-off points will need to be determined for the purpose of presenting the results, the interest lies in the degree of “lexicalness” of different collocations (of items and of sets), all of which are clearly regarded as “lexical”. Moreover the native speaker’s knowledge of his language will not take the form of his accepting or rejecting a given collocation: he will react to something as more acceptable or less acceptable on a scale of acceptability. Likely collocations could be elicited by an inquiry in which the subject was asked to list the twenty lexical items which he would most expect to find in collocation with a given node;10 but the number of such studies that would be required to cover even the most frequent lexical items in the language is very large indeed. Textually, some twenty million running words, or 1500–2000 hours of conversation, would perhaps provide enough occurrences to yield interesting results. The difficulty is that, since lexical patterns are of low generality, they appear only as properties of very large samples; and small-scale studies, though useful for testing methods, give little indication of the nature of the final results. It is hard to see, however, how the results could fail to be of interest and significance for linguistic studies. Their contribution to our knowledge of language in general, and of one language in particular, may perhaps be discussed in relation to the use of the term “semantics”. If lexis is equated with semantics, the implication is that lexical patterns can only be described either externally (that is, as relations between language and non-language, whether approached denotatively or contextually) or lexicogrammatically (that is, in dependence on grammati169

word–clause–text

cal patterns). This restriction leaves two gaps in our understanding of language: the internal relations of lexis, and the external relations of grammar – that is, lexis (lexical form) and grammatical semantics. But linguistics is concerned with relations of both types, both internal (formal, within language) and external (contextual or “semantic”, between language and non-language); and all linguistic items and categories, whether operating in closed contrasts, like the and a, or “past” and “present”, or in open ones, like strong and powerful, enter into both. Moreover, as Firth stressed, both these types of relation are “meaningful”: it is part of the meaning of “past” that it contrasts with “present”, and it is part of the meaning of strong that it collocates with tea. The fact that the labels for grammatical categories are chosen on semantic grounds should not be taken to imply that they represent an adequate substitute for grammatical semantics; but equally the existence of traditional methods in lexical semantics does not mean that lexical items display no internal, formal patterns of their own. A thesaurus of English based on formal criteria, giving collocationally defined lexical sets with citations to indicate the defining environments, would be a valuable complement to Roget’s brilliant work of intuitive semantic classification in which lexical items are arranged “according to the ideas which they express”.11 But even such a thing as a table of the most frequent collocates of specific items, with information about their probabilities, unconditioned and lexically and grammatically conditioned, would be of considerable value for those applications of linguistics in which the interest lies not only in what the native speaker knows about his language but also in what he does with it. These include studies of register and of literary style, of children’s language, the language of aphasics and many others. In literary studies in particular such concepts as the ability of a lexical item to “predict” its own environment, and the cohesive power of lexical relations, are of great potential interest.12 Lexical information is also relevant to foreign language teaching; many errors are best explained collocationally, and items can be first introduced in their habitual environments.13 A further possible field of application is information retrieval: one research group in this field is at present undertaking a collocational analysis of the language of scientific abstracts. 14 Only a detailed study of the facts, such as that now being undertaken by Sinclair,15 can show in what ways and to what extent the introduction of formal criteria into the study of lexis, as implied by the recognition of a “lexical level”, are of value to any particular applications of linguistics. But there seem to be adequate reasons for expecting 170

lexis as a linguistic level

the results to be interesting; and if they are, this is yet another indication of the great insight into the nature of language that is so characteristic of J. R. Firth’s contribution to linguistic studies.

Notes 1. See especially Firth (1935) (reprinted in Firth 1957b). 2. Firth (1957b: 195–6): ‘It must be pointed out that meaning by collocation is not at all the same thing as contextual meaning, which is the functional relation of the sentence to the processes of a context of situation in the context of culture . . . Meaning by collocation is an abstraction at the syntagmatic level and is not directly concerned with the conceptual or idea approach to the meaning of words.’ Compare also Firth (1935) (in “lexical items”): “This (sc. the lexical) function should not be misnamed semantic”. 3. See Firth (1957a: 12). In the present paper “lexical level” has been used in preference to “collocational level” in order to suggest greater generality and parallelism with the grammatical level. 4. It is also stated explicitly by Firth (1957a: 12): “Collocations of a given word are statements of the habitual or customary places of that word in collocational order but not in any other contextual order and emphatically not in any grammatical order”. Note that here “order” refers to the “mutual expectancy” of syntagmatically related categories, such as elements of structure in grammar or phonology, and not to linear sequence: cf. ibid., pp. 5, 17 and Halliday (1961: 254–5 (see above, Chapter 2)). 5. That is, that distinctions are made which involve the recognition of more finely differentiated syntagmatic relations in the grammar, and that these in turn define further sub-classes on various dimensions within previously defined classes. 6. The place of collocational restrictions in a transformational grammar is considered by Matthews (1961). 7. For a discussion of the relation between grammatical and lexical patterns see McIntosh (1961). 8. Sinclair (1966). 9. The implication is, in effect, that “wellformedness” is best regarded as “lexicogrammaticalness”, and that a departure from wellformedness may be ungrammatical, unlexical or unlexicogrammatical. That the last two are distinct is suggested by such examples as sandy hair, sandy gold and sandy desk: sandy desk is unlexical, in that this collocation is unlikely to occur in any grammatical environment, whereas sandy gold is merely unlexicogrammatical: there is nothing improbable about golden sand. An analogous distinction is observable in cliche´s: in shabby treatment the mutual expectancy is purely lexical, and is paralleled in they treated him

171

word–clause–text

10.

11. 12. 13.

14. 15.

shabbily, a shabby way to treat him and so on, whereas the collocation faint praise is restricted to this structure, in the sense that it will not occur with similar probability under other grammatical conditions. Compare the methods used to assess the disponibilite´ of lexical items in the development of “Franc¸ais fondamental” (Gougenheim, Michea, Rivenc and Sauvageot 1956). Roget (1960: 8). Cf. Firth (1951). The following text examples (drawn from written work by learners of English) may be cited in this connection: festive animals, circumspect beasts, attired with culture, funny art, barren meadows, merry admiration, the situation of my stockings was a nightmare, lying astray, fashionable airliner, modern cosy flights, economical experience, delightfully stressed, serious stupid people, shining values, a wobbly burden, light possibility, luxurious man, whose skin was bleeding, driving a bicycle, old and disturbed bits of brick wall, a comprehensive traffic jam, her throat became sad, my head is puzzled, people touched with assurance, thoughts are under a strain, a sheer new super car. This research is being undertaken by Dr. A. R. Meetham and Dr. P. K. T. Vaswani at the National Physical Laboratory, Teddington, Middlesex. Sinclair, Jones and Daley (1970).

172

Chapter Seven

LANGUAGE STRUCTURE AND LANGUAGE FUNCTION (1970)

1

The functions of language

Why is language as it is? The nature of language is closely related to the demands that we make on it, the functions it has to serve. In the most concrete terms, these functions are specific to a culture: the use of language to organize fishing expeditions in the Trobriand Islands, described half a century ago by Malinowski, has no parallel in our own society. But underlying such specific instances of language use are more general functions which are common to all cultures. We do not all go on fishing expeditions; however, we all use language as a means of organizing other people, and directing their behaviour. A purely extrinsic account of linguistic functions, one which is not based on an analysis of linguistic structure, will not answer the question; we cannot explain language by simply listing its uses, and such a list could in any case be prolonged indefinitely. Malinowski’s ethnographic account of the functions of language, based on the distinction between “pragmatic” and “magical”, or Bu¨hler’s well-known tripartite division into the “representational”, “expressive” and “conative” functions, show that it is possible to generalize; but these generalizations are directed towards sociological or psychological inquiries, and are not intended primarily to throw light on the nature of linguistic structure. At the same time, an account of linguistic structure that pays no attention to the demands that we make of language is lacking in

First published in New Horizons in Linguistics, 1970, edited by John Lyons. Harmondsworth: Penguin, pp. 140–65.

173

word–clause–text

perspicacity, since it offers no principles for explaining why the structure of language is organized one way rather than in another. Here, therefore, we shall consider language in terms of its use. Structural preoccupations have been dominant in linguistics for some time; but the usefulness of a synthesis of structural and functional approaches has long been apparent from the work of the Prague linguists (Vachek 1966) who developed Buhler’s ideas, especially in the study of grammar. The particular form taken by the grammatical system of language is closely related to the social and personal needs that language is required to serve. But in order to bring this out it is necessary to look at both the system of language and its functions at the same time; otherwise we will lack any theoretical basis for generalizations about how language is used. It is perhaps most helpful to begin with the notion of an act of speech, regarding this as a simultaneous meaning potential of language. In speaking, we choose: whether to make a statement or ask a question, whether to generalize or particularize, whether to repeat or add something new, whether or not to include our own judgement, and so on. It would be better, in fact, to say that we “opt”, since we are concerned not with deliberate acts of choice but with symbolic behaviour, in which the options may express our meanings only very indirectly: in the same sense we may be said to “opt” between a long vowel and a short one, or between a straight arm and a bent one (where the meaning is likewise mediated through the symbolic significance of the distinction between a handshake and a salute). The system of available options is the “grammar” of the language, and the speaker, or writer, selects within this system: not in vacuo, but in the context of speech situations. Speech acts thus involve the creative and repetitive exercise of options in social and personal situations and settings (Ellis 1966; Pike 1967; Firth 1968). It is fairly obvious that language is used to serve a variety of different needs, but until we examine its grammar there is no clear reason for classifying its uses in any particular way. However, when we examine the meaning potential of language itself, we find that the vast numbers of options embodied in it combine into a very few relatively independent “networks”; and these networks of options correspond to certain basic functions of language. This enables us to give an account of the different functions of language that is relevant to the general understanding of linguistic structure rather than to any particular psychological or sociological investigation. 1. Language serves for the expression of “content”: that is, of the 174

language structure and language function

speaker’s experience of the real world, including the inner world of his own consciousness. We may call this the ideational function, though it may be understood as easily in behavioural as in conceptual terms (Firth 1968: 91). In serving this function, language also gives structure to experience, and helps to determine our way of looking at things, so that it requires some intellectual effort to see them in any other way than that which our language suggests to us. 2. Language serves to establish and maintain social relations: for the expression of social roles, which include the communication roles created by language itself – for example the roles of questioner or respondent, which we take on by asking or answering a question; and also for getting things done, by means of the interaction between one person and another. Through this function, which we may refer to as interpersonal, social groups are delimited, and the individual is identified and reinforced, since by enabling him to interact with others language also serves in the expression and development of his own personality. These two basic functions, to each of which corresponds one broad division in the grammar of a natural language, are also reflected in Bernstein’s studies of educational failure (e.g. Bernstein 1970). Bernstein’s work suggests that in order to succeed in the educational system a child must know how to use language as a means of learning, and how to use it in personal interaction; these can be seen as specific requirements on his control of the ideational and interpersonal functions of language. 3. Finally, language has to provide for making links with itself and with features of the situation in which it is used. We may call this the textual function, since this is what enables the speaker or writer to construct “texts”, or connected passages of discourse that is situationally relevant; and enables the listener or reader to distinguish a text from a random set of sentences. One aspect of the textual function is the establishment of cohesive relations from one sentence to another in a discourse (Hasan 1968). All these functions are reflected in the structure of the clause. Here we attempt to show, by reference to English, what a clause is: how it serves for the realization of a number of very general meanings, or semantic options, relating to the interpersonal, ideational and textual functions of language; and how these are expressed through various configurations of structural “roles” – functional elements such as “process” and “actor” that derive from these basic functions. For a more detailed exemplification we shall consider an aspect of ideational 175

word–clause–text

meaning, the system of transitivity; the remaining areas, which have the same formal properties, will be referred to only briefly. Any one clause is built up of a combination of structures deriving from these three functions (for the sake of brevity we shall leave out the logical component in linguistic structure, which is somewhat different in its realizations).

2

Language and experience

Since normally every speech act serves each of the basic functions of language, the speaker is selecting among all the types of options simultaneously. Hence the various sets of structural “roles” are mapped onto one another, so that the actual structure-forming element in language is a complex of roles, like a chord in a fugue: for example Sir Christopher Wren, in the clause Sir Christopher Wren built this gazebo, is at once Actor and Subject and theme (see, 13 below). Each of these three represents a value in some configuration – some melodic line, so to speak – such as “Process plus Actor plus Goal”. And all such configurations are meaningful, since what we have called the basic functions of language, looked at from another point of view, are simply different kinds of meaning. So for example there is a difference in meaning between (1i) and (1ii): (1i) She would marry Horatio. She loved him. (1ii) She would marry Horatio. It was Horatio she loved. The difference concerns the organization of the second clause as a piece of information, and it derives from the textual function. There is also a difference between (1i) and (1iii): (1iii) She would marry Horatio. She did not love him. But we cannot say that this difference is “greater” or “more meaningful” than that between (1i) and (1ii); it is merely of a different kind. The speaker does not first decide to express some content and then go on to decide what sort of a message to build out of it – whether to turn it into a statement or a question, whether to make it like (1i) or (1ii) and so on. If he did, the planning of each sentence would be a totally discrete operation and it would be impossible ever to answer a question that had actually been asked. Speech acts involve planning that is continuous and simultaneous in respect of all the functions of language. Linguistics is not as a rule concerned with the description of particular speech events on individual occasions (although it is possible 176

language structure and language function

to write a theoretical grammar of just one instance if the need arises; it usually does not). It is concerned rather with the description of speech acts, or texts, since only through the study of language in use are all the functions of language, and therefore all components of meaning, brought into focus. Here we shall not need to draw a distinction between an idealized knowledge of a language and its actualized use: between “the code” and “the use of the code”, or between “competence” and “performance”. Such a dichotomy runs the risk of being either unnecessary or misleading: unnecessary if it is just another name for the distinction between what we have been able to describe in the grammar and what we have not, and misleading in any other interpretation. The study of language in relation to the situations in which it is used – to situation types, i.e. the study of language as ‘text’ – is a theoretical pursuit, no less interesting and central to linguistics than psycholinguistic investigations relating the structure of language to the structure of the human brain. We shall consider each of the functions in turn as it is reflected in the structure of the English clause, beginning with what we have called the “ideational”. To the adult – though not, be it noted, to the child – the predominant demand that we make on our language (predominant, at least, in our thinking about language; perhaps that is all) is that it allows us to communicate about something. We use language to represent our experience of the processes, persons, objects, abstractions, qualities, states and relations of the world around us and inside us. Since this is not the only demand we make on language it is useful to refer to it specifically; hence ideational function, ideational meaning, etc. (Other terms that have been used in a similar sense are “representational”, “cognitive”, “semantic”, “factual-notional” and “experiential”.) Let us consider the expression of processes: of actions, events, states and relations, and the persons, objects and abstractions that are associated with them. For this purpose we will focus our attention on one unit of linguistic structure, namely the clause. In any language, a vast number of different processes can be distinguished; but these are reducible to a small number of process types, and the grammar of every language comprises sets of options representing broad categories of this kind. The most familiar, and simplest, model is that which groups all processes into the two categories of “transitive” and “intransitive”. Associated with each type of process are a small number of functions, or “roles”, each representing the parts that the various persons, objects or other classes of phenomena may play in the process concerned. For example, in: 177

word–clause–text

(2) Sir Christopher Wren built this gazebo we have a “transitive” clause containing three roles: an “Actor”, a “Process” and a “Goal”. (The specification of this clause, assuming just these categories, would involve (i) selection of the option ‘transitive’, from the system transitive / intransitive; which would then determine (ii) the presence of the functions “Process”, “Actor” and “Goal”; these being realized (iii) by built, Sir Christopher Wren and this gazebo respectively.)

3

Transitivity functions: process and participant roles

The roles which appear in the expression of processes are of different kinds. First there is the process itself, usually represented by a verb, for example built in (2). Then there are the participant functions, the specific roles that are taken on by persons and objects, for example Wren and gazebo; and finally there are what we may call the circumstantial functions, the associated conditions and constraints such as those of time, place and manner (Fillmore 1968, where the two together are referred to as “cases”; Halliday 1967–68). It has been customary to recognize three participant functions in English, namely “actor”, “goal” (or “patient”), and “beneficiary”. Various subdivisions and modifications have been proposed, such as the distinction between goal and ‘object of result’ (Lyons 1968: 439; cf. “factitive” in Fillmore 1968: 25) as in: (3i) the Borough Council restored this gazebo (3ii) Sir Christopher Wren built this gazebo where this gazebo is goal in (3i) but object of result in (3ii); in (3ii) the gazebo comes into existence only as a result of the process of building. Similarly, the beneficiary may be the recipient of an object, as Oliver in (4i), or the recipient of a service, as Frederick in (4ii): (4i) I’ve given Oliver a tie (4ii) I’ve made Frederick a jacket These subclassifications are not made arbitrarily; they account for systematic distinctions in the grammar, e.g. the related prepositional form is to Oliver but for Frederick in (4), and in (3) restore but not build can be substituted by do to (what they did to this gazebo was restore it). But there may be many, often contradictory, criteria to choose from (see below, Section 4); moreover, the more categories one sets up, the 178

language structure and language function

more indeterminate instances will arise – for example, is I’ve brought Percival a pullover like (3i) or (3ii)? The same function may often be expressed in more than one way, e.g. Oliver, to Oliver above. Similarly, General Leathwall is Actor throughout (5i, ii, iii): (5i) General Leathwall won the battle (5ii) the battle was won by General Leathwall (5iii) General Leathwall’s winning (of) the battle . . . This is what makes it necessary to distinguish “logical” from “grammatical” categories (Sweet 1891: 10ff., 89ff.). In Sweet’s terms, General Leathwall is the logical subject in (5i–iii), though it is the grammatical subject only in (5i). Conversely in the book sells well, the book is grammatical subject but “logical direct object”. The concepts of actor, goal and beneficiary are represented in Sweet’s account as “logical subject”, “logical direct object” and “logical indirect object” respectively. The linguistic expression of processes, and of the participants (and, by extension, the circumstances) associated with them, is known by the general term transitivity. Transitivity comes under what we have called the ‘ideational’ function of language. Actor, Goal and Beneficiary are structural functions, or roles, in transitivity; and just as the same transitivity function may be realized in more than one way, as in (5), so also the same constructional form may express different transitivity functions. Thus by the fire is Actor in (6i), Place in (6ii): (6i) it was singed by the fire (6ii) it was stored by the fire This also illustrates the conflict of criteria. In (6i), by the fire might be considered instrument rather than actor, on the grounds that it is inanimate. Fillmore (1968) distinguishes actor and instrument as, respectively, the “typically animate perceived instigator of the action” (his “agentive”; cf. below, Section 8) and the “inanimate force or object causally involved in the action”; the latter may also be grammatical subject, and if not may also be expressed by with as in (7): (7i) the key opened the door / John opened the door with the key (7ii) the door was opened with the key But with is not normally used where the action is unintentional (the window was broken with the ball is odd), nor can it be substituted in (6i). 179

word–clause–text

We need here a further distinction between instrument and (natural) force, the latter not being subject to any external intent. We might therefore list, as participant roles: (a) (b) (c) (d)

Actor (“logical subject”): prepositionally by Goal (“logical direct object”) Beneficiary (“logical indirect object”): prepositionally to / for Instrument: prepositionally with / by with the possibility for further distinctions such as (b) Goal: goal, resultant [ex. (3)] (c) Beneficiary: beneficiary recipient [ex. (4)] (d) Instrument: instrument, force [ex. (6i), (7)] where “force” may simply be equivalent to (inanimate) actor.

4

Other transitivity functions: circumstantial roles

The three main types of transitivity role – process, participant, circumstance – correspond, by and large, to the three major word (or word group) classes found in most languages: verb, noun, adverb. In English, typically, processes are expressed by verbal groups, participants by nominal groups and circumstances by adverbial groups – the last often in the form of prepositional phrases. There are also incongruent forms of expression, with functions of one type expressed by classes primarily associated with another type, as in (8): (8) dinner of roast beef was followed by a swim Here the processes of eating and swimming are expressed by nouns; the temporal relation between them by the verb follow; and of the two participants, one is omitted and the other (roast beef ) is made to qualify dinner (contrast in the evening they ate roast beef and then swam). The circumstantial functions seem less central to the process than do the participant functions; this is related to their inability to take on the role of subject. But this peripheral status is not a feature of all circumstantial elements, which can be subdivided into an “inner” and “outer” type. Within the function “place”, in: (9i) he was throwing stones at the bridge (9ii) he was throwing stones on the bridge at the bridge (the “inner” type) seems more central to the process than on the bridge: we can say what was he throwing stones at? and not (in this sense) what was he doing at the bridge? (On the other hand, we can say 180

language structure and language function

what was he doing on the bridge? and not what was he throwing stones on?) However, the sense of “inner” and “outer” is contributed to by various factors not all of which coincide. For example, in (10) the place element is obligatory in (i) but optional in (ii): (10i) he put all his jewels in the wash (10ii) he lost all his jewels in the wash In (11), there is a difference of clause type; (i) is a relational clause (see below, Section 7) whereas (ii) is an action clause (Fillmore, from whom (11) is taken, gives this as an instance of dependency between functions: the place element is “outer” if an actor is present and “inner” otherwise): (11i) John keeps his car in the garage (11ii) John washes his car in the garage

5

Inherent functions

The distinction between obligatory and optional roles helps us to relate transitivity functions to a system of clause types. As, however, this involves recognizing that an “obligatory” element may in fact be absent, we shall use the term “inherent” rather than “obligatory”. An inherent function is one that is always associated with a given clause type even if it is not necessarily expressed in the structure of all clauses of that type. (We are not here talking about ellipsis, which is a matter of textual structure.) Consider a pair of clauses such as (12): (12i) Roderick pelted the crocodile with stones (12ii) the crocodile got pelted The verb pelt, as it happens, is always associated with three participant roles: a pelter, a pelted and something to pelt with; and this holds for (ii) as well as for (i) (cf. Svartvik (1966), on “agentless agentives”). Similarly there are inherently benefactive clauses without a beneficiary, such as we’re giving a silver coffee-pot. So: (12iii) Roderick pelted the crocodile is “(inherently) instrumental”, and although no instrument is mentioned the receiver interprets the process as having an instrumental role associated with it. The same verb may occur in clauses of more than one type. But 181

word–clause–text

within one type there may be different sets, and different alignments, of participants; this is the function of the system of voice – of the choice between active and passive, though the actual patterns are more elaborate than this. The options in the voice system (simplifying somewhat) are (a) middle / non-middle (see next paragraph); if nonmiddle, then (b) “active” / “passive” (not exactly equivalent to active and passive in the verb; see Halliday, 1967–68: § I 39ff., where they are referred to as “operative” and “receptive”); if “active”, then (c) plus / minus goal; if ‘passive’, then (d) plus / minus actor. The reason for choosing one rather than another of these options lies in the textual function of language (see Sections 11 and 12 below); but which options are available to choose from depends on transitivity. Voice is concerned with the roles of Actor and Goal (but see below, Section 8), both as inherent and as actualized roles. A “middle” clause is one which has only one inherent participant, which for the moment we will continue to refer to as the “Actor”; examples are Hector sneezed, the cat washed. A “non-middle” clause is one which has two, an Actor and a Goal, but one or the other may not be actualized: if “active”, there may be no Goal, for example Mary is washing (“the clothes”), and if ‘passive’, no clear Actor, e.g. the clothes have been washed (“by Mary”). All actions are classified into those involving one participant role and those involving two; there are then different ways of presenting the situation in those cases where there are two. The point was made earlier that the notion of “participant” derives from the more fundamental concept of syntactic function, or “role”. The basic elements of transitivity structure are the various roles associated with processes; and two or more such roles may be combined in one participant, as in a reflexive clause such as John is washing (“himself ”) where John is both Actor and Goal at the same time. The elements that operate as Actor, Goal, etc. also play a part, simultaneously, in other structures of the clause, expressing aspects of the interpersonal and textual functions of language. The principle of combining a number of roles in a single complex element of structure is fundamental to the total organization of language, since it is this that makes it possible for the various functions of language to be integrated in one expression. We return to this below, Section 9.

6

Transitivity clause types: action clause

All the clauses so far considered have been concerned with actions or events, and have involved an “actor” as inherent role. Let us refer to 182

language structure and language function

this type as action clauses. Action clauses all have corresponding equative forms as in example (32) below, having do or happen in them, such as: (13i) what Lionel did was (to) jump off the roof (13ii) what happened to Lionel was that he fell off the roof The following table shows the full range of possibilities of voices in action clauses, together with the roles associated with each of them: voice (clause) middle n o active n active m passive i d passive d l passive e

{

roles

voice (verb)

example

Actor

active the gazebo has collapsed

Actor, Goal

active the Council are selling the gazebo

Actor, (Goal) active the Council won’t sell Goal

active the gazebo won’t sell

Goal, Actor

passive the gazebo has been sold by the Council

Goal, (Actor) passive the gazebo has been sold

The role in parentheses are inherent but not expressed. Not all clauses are of the “action” type. English appears to recognize three main types of process: action, mental process and relation. Mental process clauses, and clauses of relation, are associated with what are at first sight rather different sets of participant roles.

7 Transitivity clause types: mental process clauses, relation clauses In mental process clauses, such as: (14) I liked your hairstyle we cannot really talk of an Actor and a Goal; it is not possible to say, for example, what I did was like your hairstyle, or what I did to your hairstyle was like it. The inherent roles are those of a human, or at any rate animate, being whose consciousness is impinged upon, and some phenomenon which impinges upon it. Let us refer to these as the “Processer” and the “Phenomenon”. The voice potentialities are now somewhat different; among the non-middle (two participant) clauses there are two types: those having the Phenomenon as Subject in active 183

word–clause–text

voice (15i), and those having the Processer (15ii). In the first types, the passive form is much more frequent than the passive in action clauses; in the second type it is much less so: (15i) the gift pleased her / she was pleased by (with) the gift (15ii) she liked the gift / the gift was liked by her This is because the passive is a means of bringing the element governed by by into prominence as the focus of information (see below, Section 12); in (15ii) the by element, i.e. her, is the Processer, and in English this tends to be the “given” element in the situation (she must have been referred to already in the text), and thus does not appropriately carry such prominence. Mental process clauses express (a) perception, e.g. see, look; (b) reaction, e.g. like, please; (c) cognition, e.g. believe, convince; (d) verbalization, e.g. say, speak. They are distinct in that the Phenomenon – that which is perceived, reacted to, etc. – is not limited, as are the participants in action clauses, to the class of “thing”, namely persons, objects, abstractions and the rest of the phenomena on the plane of experience. What is perceived or felt or thought of may be a simple phenomenon of this kind, but it may also be what we might call a metaphenomenon: a fact or a report – a phenomenon that has already as it were been filtered through the medium of language. Here words as well as things may participate in the process. For example, in (16) all the “processed” entities are simple phenomena, or “things”: (16i) (16ii) (16iii) (16iv)

I noticed Helen over there [person] I noticed a discrepancy [abstraction] I noticed a quarrel (going on) / them quarrelling [event] I noticed what (the thing that) she was wearing [object]

In (17) and (18), however, they are metaphenomena; facts in (17), reports in (18): (17i) (17ii) (18i) (18ii)

I noticed what (the fact of what) she was wearing it worries me that you look so tired I notice the bank rate’s going up again he says the bank rate’s going up again

We could insert the fact (that) in (17) and the report (that) in (18i); not, however, in (18ii), which is a clause of verbalization, since such clauses accept only reports, and “reported speech” is the meaning of clauses of 184

language structure and language function

this type. The difference between fact and report is that a “fact” is a representation at the semantic level, where the truth lies in the meaning – (she regretted) that he had gone away; whereas a “report” is a representation at the lexicogrammatical, or syntactic, level, where the truth lies in the wording – (she said) that he had gone away. In relational clauses, the “process” is simply a form of relation between two roles. One type is the attributive, such as: (19i) Marguerite is a poet (19ii) Marguerite looks desperate where the relation is one of class membership: “Marguerite belongs to the class of poets,” “. . . the class of people who look desperate”. This is a relation between entities of the same order of abstraction but differing in generality. The other type, exemplified by (20): (20i) Templecombe is the treasurer (20ii) the treasurer is Templecombe has two functions, resembling the two terms of an equation, where the one serves to identify the other, as in x = 2. Here the two entities are alike in generality but differ in abstraction: the identifying element may be of a higher order of abstraction, as in (21i), where the treasurer expresses Templecombe’s function, or of a lower order, as in (21ii) where the fat one expresses Templecombe’s form, how he is to be recognized: (21i) (which is Templecombe?) Templecombe is the treasurer (21ii) (which is Templecombe?) Templecombe is the fat one (21i) could be interpreted in the sense of (21ii) if the committee were in view on the platform; there is in fact partial ambiguity between these two sub-types. These two major types of relational clause, the attributive and the equative, differ in various respects. The attributive are non-reversible (e.g. we can say that man is a poet but not a poet is that man), have the role Attribute which may be an adjective and is usually indefinite, express class inclusion, are usually questioned by what? or how? and are expressed by the verbs be, get, turn, keep, remain, seem, sound, look, etc. The equative are reversible (i.e. have a “voice” system), have the role Identifier which must be as noun and is usually definite, express class identity, are usually questioned by who? or which? and are expressed by the verbs be, equal, represent, resemble, stand for, etc. 185

word–clause–text

It is interesting to note that, in relational clauses, quite unlike clauses of action or mental process, the verb is regularly unstressed. This is a symptom of its much weaker function in the clause. Contrast the pronunciation of equals in (22i) and (22ii): (22i) England Equals (Australia’s Total of) 512 [action] (22ii) 29 = 512 [relation]

8

The ergative

As far as the ideational component of grammar is concerned, the English clause shows the three principal types – action, mental process and relation – and associates with each a set of different inherent roles, or structural functions. The system of clause types is a general framework for the representation of processes in the grammar; possibly all languages distinguish three such categories. We need to ask, at this point, whether the structural functions can be generalized across clause types; whether, for example, an Actor in an action clause can be shown to be equivalent to a Processer (one who does the thinking, etc.) in a mental process clause. This may be approached through a reconsideration of the functions in action clauses, a reconsideration which such clauses demand anyway. If we look at examples like (23i and ii): (23i) the sergeant led the recruits (23ii) the sergeant marched the recruits they appear to be clearly distinct, (i) being transitive, with Actor and Goal, (ii) causative, with Initiator and Actor. However, there is a problem with (23iii): (23iii) the sergeant trained the recruits Is it like (23i) or like (23ii)? Actually it is like both; (23i) and (23ii) are not really different as far as transitivity is concerned. In English no very clear distinction is made between doing something to someone and making someone do something, so that (23iii) can be interpreted in either way without any sense of ambiguity. This is why so many verbs are labelled “vb trans. & intrans.” in the dictionary. The concepts of actor and goal are not well suited to describing this situation, since with these we are forced to describe (23i) and (23ii) differently. The distinction between them is by no means entirely 186

language structure and language function

unreal, since there are verbs like lead which are normally transitive (two inherent participants) and others like march which are normally intransitive (one inherent participant). But with a large number, especially of the more frequently used verbs, either form seems equally normal: there is nothing to choose, as regards the more typical use of the verb bounce, between he bounced the ball and the ball bounced. In addition there are a number of verbs which, while themselves clearly transitive or clearly intransitive, group into pairs differing only in transitivity, so that Mary put out the fire is to the fire went out as Polly lit the fire is to the fire lit. It has been pointed out by various linguists (Halliday 1967–68: § 3; Anderson 1968; Fillmore 1968) that action clauses in English seem to be organized on an ergative rather than on a transitive (or “nominative”) basis. This means that, with any action clause, there is associated one inherent role which is that of the participant affected by the process in question. Fillmore describes this as the “semantically most neutral” function, and labels it the “objective”; I used the term “affected”, which I will retain here. In (23) the recruits has the role of “affected” in every case, even through it is Goal (if an Actor–Goal analysis is used) in (23i) and Actor in (23ii); in general, the affected is the Goal in a transitive and the Actor in an intransitive clause. We have now turned what was the borderline case, such as (23iii), into the most central clause type. This is the type in which both middle (one-participant) and non-middle (two-participant) forms are equally normal; it may be considered the “favourite” clause type of Modern English. The transitive and intransitive types – those with non-middle as norm and with middle as norm respectively – are the marginal ones, and they seem to be becoming more marginal as time goes on. Hence all the examples in (24i) have the same structure, with a Process and an Affected. Those in (24ii) also have a Causer (Fillmore’s “agentive”): (24i) they’re being led they’re being trained / they’re training they’re being marched / they’re marching (24ii) he’s leading them he’s training them he’s marching them These two ways of representing processes, the transitive and the ergative, are very widely distributed; possibly all languages display one or the other, or (perhaps always) both, in different mixtures. In English, 187

word–clause–text

the two occur side by side. The transitive system asks “does the action extend beyond the active participant or not?”, the ergative, “is the action caused by the affected participant or not?” The ergative component is more prominent now than it was in Middle English, and this appears in various ways, for example, the change from impersonal to personal forms in mental process clauses (formerly methinks, it likes me). In the modern form I like, I cannot be explained as an Actor (among other things we cannot say what he does to jam is like it); but it can be shown on various grounds to have the function Affected. As this suggests, the ergative pattern, whereby a Process is accompanied by an obligatory Affected participant and an optional Causer, is more readily generalizable than that of Actor and Goal. It extends beyond action clauses to those of mental process, and perhaps even to clauses of relation as well. We want to say that Paul has the same function in both (25i) and (25ii): (25i) Paul fears ghosts (25ii) ghosts scare Paul – not that they are identical in meaning, but that the transitivity roles are the same. This is not possible in Actor–Goal terms. But in an ergative system there is considerable evidence for regarding Paul as the ‘Affected’ participant in both cases. The ergative, therefore, represents the more general model of the transitivity patterns of Modern English – that is, of the options available to the speaker of English for talking about processes of all kinds.

9

Other dimensions of clause structure

So far the discussion has been confined to the expression of ideational meanings. We have not yet considered the structure of language in its other functions, the “interpersonal” and the “textual”. Both these functions are manifested in the structure of the clause. Certain problems that have arisen in the history of the investigation of subject and predicate provide an insight here. A sentence such as (26i) presents no problem in this respect: my mother is clearly subject and the rest predicate. But in (26ii) there seem to be three candidates for the status of subject, these beads, my mother and I: (26i) my mother gave me these beads (26ii) these beads I was given by my mother 188

language structure and language function

The solution was to recognize different kinds of subject. For Sweet, my mother was “logical subject”, I was “grammatical subject”; these beads came to be known as ‘psychological subject’. In (26i), all these coincide. The notion of subject conflates three distinct roles which, although they are typically combined into one element, are nevertheless independent of one another. We may think of this as governed by a “good reason” principle: many linguistic systems are based on this principle, whereby one option (the “unmarked” option) will always be selected unless there is good reason for selecting otherwise (cf. Jakobson 1963: 268ff.). These three “kinds of subject” relate to the functions of language as described above. The logical subject is the actor; this is a transitivity role, deriving from the ideational function. The other two have different sources, though they are no less meaningful. The grammatical subject derives from the interpersonal component in language function: specifically, it has to do with the roles taken on by the performer and receiver in a communication situation. The psychological subject belongs to the textual component; it is concerned with the organization of the clause as a message, within a larger piece of discourse. The next two sections will examine these in turn.

10

Mood

As we have said, one function of language is to provide for interaction between people, by allowing the expression of statuses, social and individual attitudes, assessments, judgements and the like; and this includes participation in linguistic interaction. Language itself defines the roles which people may take in situations in which they are communicating with one another; and every language incorporates options whereby the speaker can vary his own communication role, making assertions, asking questions, giving orders, expressing doubts and so on. The basic “speech functions” of statement, question, response, command and exclamation fall within this category (though they do not exhaust it), and these are expressed grammatically by the system of mood (cf. Sweet 1891: 105), in which the principal options are declarative, interrogative (yes / no and wh- types) and imperative. The difference between he can and can he? is a difference in the communication role adopted by the speaker in his interaction with a listener. The notion “grammatical subject” by itself is strange, since it implies a structural function whose only purpose is to define a structural 189

word–clause–text

function. Actually, just as the “logical subject” is a function defined by transitivity, so the “grammatical subject” is a function defined by mood. If we consider an example such as (27): (27) Tigers can climb trees. – Can tigers climb trees? – They can climb trees, can’t they? – No they can’t. we find that one part, tigers can, has the function of expressing mood throughout; it also typically carries the positive / negative option. It consists of the finite element of the verb, plus one nominal (noun or noun group) which is the “grammatical subject”. The function of the “grammatical subject” is thus a meaningful function in the clause, since it defines the communication role adopted by the speaker. It is present in clauses of all moods, but its significance can perhaps be seen most clearly in the imperative, where the meaning is “request you to . . .”; here the speaker is requiring some action on the part of the person addressed, but it is the latter who has the power to make this meaning “come true” or otherwise, since he can either obey or disobey. In the usual form of the imperative, this modal entity, or “modal subject” as we may call it, is the listener; and the only option is plus or minus the speaker himself, as in let’s go home as opposed to (you) go home. Hence, in a passive imperative such as be guided by your elders, although the actor is your elders, the modal subject is “you”; it is the listener who accedes, potentially, to the request, fulfilling the modal function defined by the speaker’s role.

11

Theme

The basic unit of language in use is not a word or a sentence but a “text”; and the “textual” component in language is the set of options by means of which a speaker or writer is enabled to create texts – to use language in a way that is relevant to the context. The clause, in this function, is organized as a message; so in addition to its structure in transitivity and in mood, it also has structure as a message, what is known as a “thematic” structure. (It was linguists of the Prague school who first studied this aspect of language, cf. Mathesius 1928; Firbas 1959; 1964; Svoboda 1968 and references therein.) The English clause consists of a Theme and a Rheme. The Theme is another component in the complex notion of subject, namely the “psychological subject”; it is as it were the peg on which the message is hung, the theme being the body of the message. The Theme of a clause is the element which, in English, is put in first position; in (28 190

language structure and language function

i–v) the Theme is the item outside the brackets, what is inside being the Rheme: (28i) (28ii) (28iii) (28iv) (28v)

I (don’t know) yesterday (we discussed the financial arrangements) his spirit (they could not kill) suddenly (the rope gave way) people who live in glasshouses (shouldn’t throw stones)

As we have seen, Theme, Actor and modal Subject are identical unless there is good reason for them not to be (cf. (26) above). Where they are not, the tendency in Modern English is to associate Theme and modal Subject; and this is the main reason for using the passive. The passive has precisely the function of dissociating the Actor from this complex, so that it can either be put in focal position at the end or, more frequently, omitted, as in (29): (29i) this gazebo was built by Sir Christopher Wren (29ii) this gazebo is being restored The typical theme of declarative clause is thus the modal Subject (or “grammatical subject” – this gazebo in both cases); in interrogatives, however, the picture is different. If we ask a question, it is usually because we want to know the answer, so that the typical Theme of an interrogative is a request for information. Hence we put first, in an interrogative clause, the element that contains this request for information: the polarity-carrying element in a yes / no question, and the questioning element in a “wh-” question, as in (30): (30i) didn’t (Sir Christopher Wren build this gazebo?) (30ii) how many gazebos (did Sir Christopher Wren build?) In English there is a definite awareness of the meaning expressed by putting something in first position in the clause. The Theme is the point of departure for the message; a paradigm form of it is the headword in a definition, for example a gazebo in (31): (31) a gazebo is a pavilion or summerhouse on an eminence, open for the view In addition to the selection of a particular element as the theme, the speaker has other options in thematic structure open to him (Halliday 1967–68: § 2); for example, any clause can be split into two parts by the use of nominalization, as in 191

word–clause–text

(32) the one who built this gazebo was Sir Christopher Wren where the Theme is the whole of whichever part comes first – here the one who built this gazebo.

12

Information structure

Thematic structure is closely linked to another aspect of the textual organization of language, which we may call information structure. This refers to the organization of a text in terms of the functions Given and New. These are often conflated with Theme and Rheme under the single heading “topic and comment”; the latter, however, is (like the traditional notion of “subject”) a complex notion, and the association of Theme with Given, Rheme with New, is subject to the usual “good reason” principle already referred to – there is freedom of choice, but the Theme will be associated with the Given and the Rheme with the New unless there is good reason for choosing some other alignment. In English, information structure is expressed by intonation. Connected speech takes the form of an unbroken succession of distinctive pitch contours, or tone groups; each tone group represents what the speaker decides to make into one unit of information. This is not necessarily the same length as a clause, though it often is so. The information unit consists of an obligatory New element – there must be something new, otherwise there would be no information – and an optional Given element; the main stress (tonic nucleus) marks the end of the New element, and anything that is Given precedes it, unless with good reason – which means, here, unless it is a response to a specific question, either asked or implied. The function Given means ‘treated by the speaker as non-recoverable information’: information that the listener is not being expected to derive for himself from the text or the situation. (33) illustrates the interaction of information structure with thematic structure (information unit (“i.u.”) boundaries are marked by //; main stress is indicated by bold type; 4 = falling-rising tone; 1 = falling tone): (33) //4 this gazebo //1 can’t have been built by Wren// (clause: Theme . . . Rheme. . . . . . . . . . . . . . . . . . . .) (i.u.(1): New. . . . . . .; (2): New. Given. . . . . . . . . . . . . . . .) meaning ‘I am talking (Theme), specifically, (New) about this gazebo: the fact is (Rheme) that your suggestion (Given) that Wren built it is 192

language structure and language function

actually (New) quite impossible’. No such suggestion need actually have been made for this clause to occur; one of the features of the Given–New structure is its use for various rhetorical purposes, such as bullying the listener. Given and New thus differ from Theme and Rheme, though both are textual functions, in that Given means ‘here is a point of contact with what you know’ (and thus is not tied to elements in clause structure), whereas Theme means ‘here is the heading to what I am saying’. The functions of given and new link up in turn with the functions in transitivity. It was noted earlier (see 3 above) that a number of participant roles may be expressed in either of two ways, either directly or through the mediation of a preposition, for example the Beneficiary in: (34i) I’ve offered Oliver a tie (34ii) I’ve offered the tie to Oliver The members of such a pair have the same ideational meaning but differ in information. Typically, the prepositional form of the Beneficiary is associated with the function New, the other form with the function Given; and if we assume here the expected intonation pattern, then in (34ii) Oliver is New and the tie is Given, the implied question being ‘who did you offer the tie to?’, while in (34i) a tie is New and Oliver is Given, answering ‘what have you offered to Oliver?’ (note that one of the meanings of definiteness – not the only one – is Given, hence the likelihood of the tie in (34ii)). A general principle underlies the existence of these two informationally distinct forms, one with a preposition and one without, for expressing participant roles. The textual function of language requires that, for effective communication, new information should be made grammatically explicit. New lexical content has to be backed up, as it were, by adequate quanta of grammar; specifically, it has to be made clear what is the ideational function of any new material in the discourse, and here it is the preposition that indicates the role of the unfamiliar element. The use of a preposition to specify function in the clause in just those cases where the element in question is typically New (compare the use of by with the Actor in a passive construction) illustrates how the “texture” of discourse is achieved through the interplay of varied grammatical resources expressing different facets of the total meaning.

193

word–clause–text

13

Conclusion

The subject, in its traditional sense, is thus a complex of four distinct functions, three in the structure of the clause (cf. Lyons 1968: 343–4): 1. Actor (“logical subject”): ideational 2. modal Subject (“grammatical subject”): interpersonal 3. Theme (“psychological subject1”): textual together with a fourth function which is in the structure of the ‘information unit’ 4. Given (‘psychological subject2’): textual These coincide unless there is “good reason” for them not to do so; thus in (35i) the Borough Council is Actor, modal Subject and Theme, whereas in (35ii) the Borough Council is Actor, this gazebo is modal Subject and next year is Theme: (35i) the Borough Council will restore this gazebo next year (35ii) next year this gazebo will be restored by the Borough Council No mention has been made of subject and predicate as a logical relation. We might introduce “predication” as another dimension of clause structure, with the Borough Council in (35i) being also “subject in predication” and the rest predicate; but the subject in this sense would be identical with the modal subject. The subject-predicate structure is entirely derivable from mood, and has no independent significance (cf. Fillmore 1968: 17; and Fillmore’s reference, ibid., to Tesnie¯re 1959: 103–5). As a form of generalization, it may be useful in that it expresses the fact that Actor, modal Subject and Theme are regularly associated; but it obscures the equally important fact that they are distinct and independent structure roles. The multiple function of language is reflected in linguistic structure; this is the basis for the recognition of the ideational (including logical), interpersonal and textual functions as suggested here. It is not necessary to argue that one function is more abstract, or “deeper”, than another; all are semantically relevant. The investigation of these functions enables us to relate the internal patterns of language – its underlying options, and their realization in structure – to the demands that are made on language in the actual situations in which it is used. As performers and receivers, we simultaneously both communicate through language and interact through language; and, as a necessary condition for both of these, we create and recognize discourse (the 194

language structure and language function

textual function is thus instrumental to the other two). A speech act is essentially a complex behaviour pattern which in most instances combines the ideational and interpersonal functions, in varying degrees of prominence. These very general notions in turn encompass a broad range of more specific patterns relating to the creative and the repetitive aspects of language in use.

195

Chapter Eight

MODES OF MEANING AND MODES OF EXPRESSION: TYPES OF GRAMMATICAL STRUCTURE AND THEIR DETERMINATION BY DIFFERENT SEMANTIC FUNCTIONS (1979)

1

Preamble

Let us say that a code is a system of signs having a content and an expression: for example, a traffic control code (Figure 1).

Figure 1

The relation between the content and the expression is one of realization. Then, a semiotic, or semiotic system, is a code having two or more realizational cycles in it, so that the expression of content1 (call it expression1) is itself a content (content2) that in turn has its own expression (expression2). Hence there will be at least three levels, or strata, in such a system: level one: two: three:

content1 (realized as) (realized as)

expression1 = content2 expression2 . . .

A semiotic, in other words, is a stratified, or stratal, system, in which the output of one coding process becomes the input to another. First published in Function and Context in Linguistic Analysis: A Festschrift for William Haas, 1979, edited by D. J. Allerton, Edward Carney and David Holdcroft. Cambridge: Cambridge University Press, pp. 57–79.

196

modes of meaning and modes of expression

In this sense language is a semiotic. It consists (at least) of three strata with, therefore, two realizational cycles; these are set out in Figure 2, (i) in everyday and (ii) in technical terminology. The formulation “at least” is intended to allow for the addition of further strata above the semantic system, since the semantic system itself can be regarded as the realization of some higher level semiotic. In principle this may be associated with any of a number of different orders of meaning, cognitive, social, aesthetic and other things besides. At any particular time, attention is likely to be focused on one or other of these higher orders. It follows from this that as far as the elements of a semiotic system are concerned, we may in principle consider the organization of any one part of the system from three different aspects: (1) at its own level – its relation to other elements identified at the same level as itself (2) from above – its relation to elements at the next (or some) higher level (3) from below – its relation to elements at the next (or some) lower level

Figure 2

Now it is typical of semiotic systems that the different strata are not isomorphic; there is no relation of biuniqueness (one–one correspondence) between one level and the next. This is bound to be the case in a system such as language, where the coding not only converts elements of one kind into elements of another kind – meanings into wordings into sounds – but also reduces both the size and the inventory of the basic components. By any usual definition of linguistic units, units of speech sound are both smaller than and fewer than units of form; and units of form are both smaller than and fewer than units of meaning. 197

word–clause–text

Hence when one organizational system is represented in terms of another there will be mismatches of various kinds: what Sydney Lamb called “interlocking diversifications” in the realizational process. This means that we cannot simply operate with a schema of definitions and say, for example, that the elements of this system are defined “from above”; it is not possible ever to derive the structure fully from statements of this kind. The description of any part of the system involves an interpretation of all three sets of relations into which it enters, “upward”, “downward” and “across”. This will be true no matter whether we are concerned with the most detailed specifics of the system, or with the broadest generalizations. Whatever is said in interpretation of one level has implications not only for that level but also for what is above and what is below. This provides the context for the present discussion.

2

Functional modes of meaning

Let us focus first on the semantic system, and introduce a broad generalization along the following lines. The semantic system of a natural language is organized into a small number of distinct components, different kinds of meaning potential that relate to the most general functions that language has evolved to serve. Here are the headings we shall use: IDEATIONAL EXPERIENTIAL

INTERPERSONAL

TEXTUAL

LOGICAL

The first of these is language as representation: the semantic system as expression of experience, including both experience of what is round about us in the outside world and experience of the world of consciousness that is inside us. This we are calling the ideational component. There are two subcategories: an experiential, where we represent experience “directly” in terms of happenings (actions, events, states, relations), entities that participate in these happenings (persons, animate and inanimate objects, institutions, abstractions) and circumstantial features (extent, location, time and space, cause, manner and so on); and a logical, where we represent experience “indirectly” in terms of certain fundamental logical relations in natural language – ‘and’, ‘namely’, ‘says’, ‘is subcategorized as’, etc. – which are not those of formal logic but rather are the ones from which the operations of 198

modes of meaning and modes of expression

formal logic are ultimately derived. These two, the logical and the experiential, together make up the ideational component in the semantic system: that of meaning in the reflective mode. The second main component, the interpersonal, is language as interaction: it is meaning in the active mode. Here the semantic system expresses the speaker’s intrusion in the speech event: his attitudes, evaluations and judgements; his expectations and demands; and the nature of the exchange as he is setting it up – the role that he is taking on himself in the communication process, and the role, or rather the role choice, that he is assigning to the hearer. This component is therefore both speaker- and hearer-oriented; it is interpersonal – what Hymes called “socio-expressive” – and represents the speaker’s own intrusion into the speech situation. All discourse involves an ongoing simultaneous selection of meanings from both these components, which are mapped into a single output in the realization process. But there is also a third component, which we are calling the textual, whereby the meanings of the other two kinds take on relevance to some real context. Here the semantic system enables the speaker to structure meaning as text, organizing each element as a piece of information and relating it significantly to what has gone before. If the ideational component is language as reflection (the speaker as observer of reality), and the interpersonal component is language as action (the speaker as intruder in reality), the textual component is language as relevance (the speaker as relating to the portion of reality that constitutes the speech situation, the context within which meanings are being exchanged). The textual component provides what in modern jargon we might refer to as the ecology of the text. For example, from the Walrus and the Carpenter, in Alice Through the Looking-Glass, when the Carpenter says to the Walrus Cut us another slice! the ideational meaning is the representation of a material process, cutting, in which three entities participate: the one who cuts, the thing that is cut and the one that the thing is cut for; also the place of cut in the taxonomy of actions and of slice in the taxonomy of things. The interpersonal meaning is a demand for goods-and-services, “I want you to do something for me”, embodied in the selection of the imperative mood, direct, explicit and without any special modulation. The textual meaning is the internal organization of this as a message with the focus on what is demanded, together with its relation to the preceding text 199

word–clause–text

through presuppositions – a slice of something, of which I have already had at least one.

3

Above and below the semantic system

Since this functional interpretation is a generalization about language it can be examined from all three angles of approach: from above, from below and from its own level. Presumably, if it is valid, it has some implications for all three. The main concern of this paper is with its implications for what is below; the hypothesis will be that (i) each of these semantic components typically generates a different kind of structural mechanism as its output, or realization; and that (ii) these different types of structure are non-arbitrarily related to the kinds of meaning they express. However, let us focus briefly on the other two levels at which this generalization is significant: the semantic level itself and the level above. (1) If we represent the semantic system as a meaning potential through the use of system networks, which are networks of options each with its condition of entry, these functional components appear as relatively independent sets of semantic options. Within each component, the networks show a high degree of internal constraint: that is, of interdependence among the various options involved. The selections made by the speaker at one point tend to determine, and be determined by, the selections he makes at another. For example, within the interpersonal component, there is a high degree of interdependence of this kind between the systems of mood and modality, both in terms of what can be selected and in terms of the meaning of whatever is selected. To cite one particular instance of what is a complex and quite general phenomenon, the meaning of the modality “probable” is different in interrogative mood from what it is in declarative, and it cannot combine at all with imperative. But between one component and another, there is very little constraint of this kind: little restriction on the options available, and little effect on their interpretation. For example, the choice of modality (in the interpersonal component) is quite independent of the choice of transitivity (in the ideational component): the speaker can always contribute his own judgement of probability no matter what the nature of the process he is talking about or what participants are associated with it. Hence the categories of ideational, interpersonal and textual appear 200

modes of meaning and modes of expression

clearly in the semantic system itself, as system networks each having a high degree of internal dependence but a very low degree of external dependence. Choices made within one component have a great deal of effect on other choices within the same component but hardly any effect on choices in the other components. (2) If we now look above the semantic system, to the social contexts in which meanings are exchanged, it seems to be the case that these functional components have considerable significance for the way in which the social context acts as the determinant text. Every act of meaning has a context of situation, an environment within which it is performed and interpreted. For communication to take place at all, it is necessary for those who are interacting to be able to make intelligent and informed guesses about what kinds of meanings are likely to be exchanged. They do this on the basis of their interpretation of the significance – the semiotic structure – of the situation. Let us postulate that the relevant features of a situation in which language has some place are the field of social process, the tenor of social relationships and the mode of discourse itself: that is, (1) what is going on, (ii) who are involved, and (iii) what part the text is playing – whether written or spoken, in what rhetorical mode and so on. We shall then find a systematic relationship between these components of the situation and the functional components of the semantic system. It appears that, by and large, the field – the nature of the social activity – determines the ideational meanings; the tenor – the social statuses and roles of the participants in the situation – determines the interpersonal meanings; while the mode – the part assigned to the linguistic interaction in the total situation – determines the textual meanings. In the example in Section 2, the activity going on is that of having a meal; the Walrus and the Carpenter are dining off the oysters, accompanied with slices of bread and butter, and this is what the Carpenter takes as the ideational content of his utterance. In this context the two of them are collaborators, since both have shared in the preparation of the meal; this is reflected in the interpersonal meaning as an instruction from one to the other. The text is language-in-action, directed towards furthering the activity in question; hence selection of the textual meanings makes exophoric (situational) reference to the processes and the objects involved, as well as internal reference to an earlier occurrence a loaf of bread through the collocational potential of slice. Hence the categories of ideational, interpersonal and textual appear 201

word–clause–text

to have implications for what is ‘above’, in that they represent different components of the meaning potential which are activated by different components of the social context (cf. Halliday 1977). This appears to be the basis on which interactants make predictions about the meanings that will typically be exchanged in any particular situation with which they find themselves confronted.

4

The experiential mode

For the rest of this paper we shall be concerned with what is “below” the semantic system: with the question of what kinds of structural mechanism are typically involved in the realization of these various components of meaning. The suggestion will be that here too the same categories are relevant, since they tend to be expressed through fundamentally different types of structural organization (cf. Mathesius 1964: 24–5 of Czech original).2 Let us consider the experiential function first. Here we are concerned with the semantic (linguistic) encoding of experience; particularly our experience of the processes of the external world and of the internal world of our own consciousnesses. We tend to encode such experience in terms of configurations of elements each of which has a special and distinct significance with respect to the whole. Typically, we recognize a process itself, and various more or less specialized participants and circumstantial elements. For example, suppose there is a flock of birds flying overhead. We represent this in language as something like There are birds flying: that is, a process of “flying” and, separated out from this, an entity that is doing the flying, namely “birds”. This is, certainly, one valid way of encoding it; but it is not the only one – we might have said, instead, It’s winging. If we did say this in English we would be treating the phenomenon as a single unanalysed process, not as a process plus a participant; this is, after all, what we do with It’s raining (although not, for some reason, with The wind’s blowing). No doubt it is useful to be able to talk about birds doing other things than flying, and about flying being done by other things than birds. Some languages feel the same about rain: in Chinese one says, liberally translated, “There’s rain falling”; and in one south Chinese dialect, a variety of Cantonese, there are usually two participants in the pluvial process, which is encoded as “The sky is dropping water”. So there is no reason for assuming that each particular process will always be encoded as just this or that particular configuration of 202

modes of meaning and modes of expression

elements. But we can formulate the general principle that this is how processes are represented in languages. This means that a structure which represents experiential meanings will tend to have this form: it will be a configuration, or constellation, of discrete elements, each of which makes its own distinctive contribution to the whole. We usually represent this kind of structure linguistically as a functionally labelled constituent structure as shown, for example, in Figure 3.

Figure 3

There is no particular reason at this stage why the representation should have to be linear; that is no doubt the form of the final output, after the other structures have been mapped on to it, but experiential structures are not in fact sequential and we could just as well represent this as in Figure 4. If ordering is to be introduced into the representation, there is the possibility of using

Figure 4

a dependency construct having the Process at the centre, as in Figure 5. But a more appropriate ordering would have a nucleus consisting of Process plus Goal, with the other elements clustering around it, as in Figure 6. (We will leave aside here the question of the appropriateness of the functional labels themselves.)

Figure 5

Figure 6

203

word–clause–text

Each one of these elements may have a substructure of the same general type, with different labels as, for example, in Figure 7. This again can be reinterpreted as Figure 8. The layers

Figure 7

Figure 8

introduced into the structure in this way are the source of the linguist’s classic mode of representation of constituent structure in the form of trees, or labelled bracketings. But the appropriate bracketing for functionally labelled structures is minimal, not maximal (“many ICs” rather than “few ICs”, in the terms of Hudson 1967). Maximal bracketing imposes too much structure for a functional grammar: for example, in four young oysters it is reasonable to recognize young oysters as a constituent provided the labelled elements are classes, since it is a nominal group, but in no way does this correspond to any meaningful functional constituent. The nonlinear representation implies more of a molecular model of structure, with a taxonomy similar to cell: molecule: atom: subatomic particle. Experiential meanings are typically realized as elemental structures of this type. The basic structural mechanism is that of constituency, with larger units constituted out of layered clusters or bracketed strings of smaller units, each part having its own specific function with respect to the whole. We could call this “segmental”, except that it is better perhaps to reserve the term “segment” for an element in the final output – the syntagm – that serves as input to the next realizational cycle. So let us say that experiential meanings are realized through some kind of constituent structure. This expresses the particular way in which we order our experience of reality when we want to turn it into meaning. The bounded entities that enter into constituent structures with specific functions like Process, Actor, Goal, Extent or Manner offer a presentation of reality in terms of “things” – doings by, and happenings to, persons and objects, in the environment of other persons and objects, with yet other persons and objects, and also times and places and so on, as attendant circumstances; and including various “metathings” (facts and reports), which are complex things that have already been encoded in language and so acquired a status which enables them to participate in certain types of 204

modes of meaning and modes of expression

process as objects in their own right. Such elements naturally form constituent-like structures which allow us to isolate them and continue to refer to them as discrete entities.

5

The interpersonal mode

Now let us consider interpersonal meanings. These are expressed through very different structural devices. Think of an example such as the following, uttered when somebody is sick: I wonder if perhaps it might be measles, might it d’you think? The experiential meaning is: Attribute “measles” plus Time “now”. With this the speaker has combined an interpersonal meaning of “I consider it possible”, together with an invitation to the hearer to confirm the assessment. This interpersonal meaning, however, is strung throughout the clause as a continuous motif or colouring. It appears as I wonder, perhaps, might, might and d’you think; each of these expresses the same modality, and each one could occur by itself. When they all occur, the effect is cumulative; with each one the speaker reaffirms his own angle on the proposition. The intonation contour is another mode of the realization of interpersonal meanings. It expresses the “key”, the particular tone of assertion, query, hesitation, doubt, reservation, forcefulness, wonderment, or whatever it is, with which the speaker tags the proposition. This too is continuous; and in this case there is no possibility of associating it with any segments – it is simply a melodic line mapped on to the clause is a whole, running through from beginning to end. We shall refer to this mode of realization as “prosodic”, since the meaning is distributed like a prosody throughout a continuous stretch of discourse (cf. Mitchell 1958). It is characteristic of interpersonal meanings that they are expressed in this prosodic fashion. Mood and modality, tone and key, intensity and other attitudinal meanings are typically realized through this kind of structural pattern. Swearwords and obscenities, also, may occur at any or all points in the clause; it does not matter what segments they are attached to – many writers have noted that such elements readily occur even in the middle of a word. The speaker who says Christ they beat the hell out of those bastards is in fact using a very regular and well-established resource for the expression of meanings of this kind. 205

word–clause–text

It is not difficult to see the rationale behind this mode of realization. The interpersonal component of meaning is the speaker’s ongoing intrusion into the speech situation. It is his perspective on the exchange, his assigning and acting out of roles. Interpersonal meanings cannot easily be expressed as configurations of discrete elements. They may be attached, as connotations, to particular lexical items, like bastards above meaning “people” plus “I’m worked up”; but connotations do not enter into constituent-like structural relations. The essence of the meaning potential of this part of the semantic system is that most of the options are associated with the act of meaning as a whole. Even when the meaning is realized in a single word or a phrase, this can be interpolated at more or less any point in the clause; and even when two or more such elements are present at the same time, they still do not go together to form constructions. It is much more difficult to know how to represent prosodic structures in a description of language. Usually they are either ignored or forced to fit into the constituency mould. It may be more effective to treat them like prosodies in phonology, that is as contrasting features having no place in the constituent structure (which is, after all, an experiential structure) but which are specified separately and then mapped on to the constituent structure as a distinct step in the realizational process.

6

The textual mode

The “textual”, or text-forming, resources in the semantic system generate structures of still another kind. Consider the example: Why did you let the big one get away? This clause has, in effect, two points of prominence. It is a WHquestion, which means a demand for a particular piece of information; and this demand is enunciated at the beginning, through the word why. The word why proclaims the theme of the discourse; the speaker begins by announcing ‘What I’m on about is this: I want to know something’. This is what we call thematic prominence, and in English it is associated with first position in the clause; in fact it is realized by first position, since putting something first is what gives it the status of a Theme. But there is also another point of prominence here, at the end. To be aware of it we have to consider the clause in its spoken mode, since this takes the form of tonic prominence: that is, the location of the tonic accent, which is the dynamic centre of the pitch contour, the 206

modes of meaning and modes of expression

place where the greatest pitch movement takes place. (This may be a falling movement or a rising movement or some kind of complex movement, depending on which kind of melody it is. It corresponds to what is sometimes called “primary stress”, although it is not, in fact, a stress feature.) Suppose we represent the intonation unit (the tone group) as bounded by double slash, and intermediate rhythmic units (the feet) by single slash, with tonic prominence as bold type; the likely form of the utterance would be: // why did you / let the / big one / get a/way // The meaning of tonic prominence is the focus of information; it signals the climax of what is new in the message. This kind of focal prominence can be assigned at any point in the clause; it is not realized by final position, in the way that thematic prominence is realized by initial position. But it is typically located at the end, and any other focus is “marked” and so explicitly contrastive. In the typical form of the message, in other words, the speaker puts what is new at the end. So there is a peak of prominence at the beginning, which is the Theme; and another peak of prominence, usually at the end, which is the focus of information or, simply, the New. The two are different in meaning. The Theme is speaker-oriented; it is the speaker’s signal of concern, what it is that he is on about – he may even make this explicit, by starting ‘as far as . . . is concerned’. The New is heareroriented (though still, of course, selected by the speaker); it is the speaker’s presentation of information as in part already recoverable to the hearer (the Given) and in part not recoverable (the New). These two types of prominence are independent of each other. But both contribute to the “texture”, to fashioning the fabric of the text. What these text-forming systems do is to organize discourse into a succession of message units, quanta of information such that each has its own internal texture, provided by the two systems of prominence just mentioned. The message unit corresponds, typically (i.e. in the unmarked case), to a clause. Hence it is possible in such instances to represent both thematic and focal prominence as constituent-like structures of the clause, by recognizing the functional significance of the non-prominent part. So Theme contrasts with Rheme, and New contrasts with Given, as in Figure 9. In fact, the information unit is not always coextensive with the clause; to return to the Walrus and the Carpenter, in The moon was shining sulkily (following the earlier The sun was shining on the sea, shining with all his might) 207

word–clause–text

Figure 9

we should presumably have the structure shown in Figure 10. But because a clause is typically one information (focal) unit, and the focus in the information unit is typically at the end, final position in the clause carries a potential for prominence which highlights it in the same way that initial position is highlighted by prominence of a thematic kind.

Figure 10

The structures that realize options in the textual component are what we may call “culminative” structures. They are not configurations or clusters of elements such as we find in the ideational component; nor are they prosodic chains of the interpersonal kind. What the textual component does is to express the particular semantic status of elements in the discourse by assigning them to the boundaries; this gives special significance to “coming first” and “coming last”, and so marks off units of the message as extending from one peak of prominence to the next. The effect of this is to give a periodicity to the discourse. The clause, in its status as a message, begins with prominence of one kind, thematic prominence, and ends with prominence of another kind, prominence due to information focus. The latter is expressed through the assignment of the tonic accent to a particular place in the tone group; so the prominence is also in part phonological – and can be heard. The periodicity is further reinforced by the use of conjunctives to link one sentence with another; these contribute to the texture by relating a clause cohesively to what went before it, and they also occur at the boundaries – usually the beginning, but sometimes, especially in casual speech, at the end. 208

modes of meaning and modes of expression

Perhaps the clearest instance of the periodicity of texture is to be found in poetic forms. The metric regularity of the structures that have evolved in poetry – lines with a fixed number of feet, and stanzas or genres with a fixed number of lines – expresses in symbolic fashion the regularity of the structure of discourse as an exchange of messages. This is not to suggest, of course, that the structural unit of any particular genre, such as the iambic pentameter line, functions directly as the realization of any unit in the structure of the text. On the contrary, there is usually a tension set up between the two types or modes of structure, with the periodicity of the message, deriving from the theme and information systems (Theme–Rheme and Given–New), cutting across that of the metric form. But the impact of this tension on the reader, and especially on the listener, is one of the clearest indications of the reality of the two kinds of periodic movement.

7

Particle, field and wave

Figure 11 summarizes the three types of structure we have recognized so far. It is important to stress that when we associate each of these structural types with one of the functional semantic components, we are talking of a tendency not a rule. Experiential structures tend to be more elemental in character, interpersonal structures tend to be prosodic and textual structures tend to be culminative or periodic. Furthermore this is a statement about the description of English. The functional categories themselves are universals; but the structural tendencies, though clearly non-arbitrary – we can see why it is that each should take this form – may differ very considerably from one language to another.

Figure 11

Given structures of these very general kinds, it is clear that each can be reduced to some form of constituency; but not all with the same success. Experiential structures are quite constituent-like; whereas interpersonal ones are not, and the attempt to represent them in constituency terms involves idealizing them to an extent that is tantamount to 209

word–clause–text

a form of reductionism. An example of this is the attempt to reduce an intonation contour to a sequence of pitch phonemes (which are then attached to specific places in a string). Figures 12 and 13 represent an English clause first in non-constituency terms and then in constituency terms. The clause is: On Sunday perhaps we’ll take the children to the circus, shall we? If we consider the major traditions in linguistic thought, we find, not at all surprisingly, perhaps, that those in the psycho-philosophical tradition, who are firmly committed to language as an ideational system, have usually worked with constituency models of structure:

Notes: (1) experiential : clause as representation (of process); (2) interpersonal : clause as interaction; (3) textual : clause as message Figure 12

210

modes of meaning and modes of expression

Figure 13

American structuralist and transformationalist theories, for example. By contrast, linguists in the socio-anthropological tradition, like Firth, who are interested in speech functions and stress the interpersonal aspect of language, have tended to develop prosodic models. Those in the literary tradition, concerned primarily with texture and text structure, have developed models of a periodic kind: the structure of the paragraph (topic sentences, etc.), generic structures of various kinds and of course the whole theory of metrics. It is interesting to recall here Pike’s (1959) important insight into language as particle, wave and field. Although Pike did not conceive of these in quite the same way, it seems very clear that this is what we have here: constituent (experiential) structures are particulate prosodic (interpersonal) structures are field-like periodic (textual) structures are wave-like

8

The logical mode

There is one functional component that we have omitted to take into consideration so far, and that is the logical. This is perhaps the most difficult to interpret. As far as its origin at a higher level is concerned, the logical is a subcategory of the ideational, since it is language in the representation of reality. But there are two distinct modes of representing reality in semantic terms. In the experiential mode, reality is represented more concretely, in the form of constructs whose elements make some 211

word–clause–text

reference to things. The linguistic structures actually stand as metaphors for the relations between things; and the elements that enter into and are defined by these relations are identified as Process, Actor, Goal, Extent, Manner and the like. These in turn are interpreted as “roles” “occupied by” various classes of phenomena, and these classes of phenomena themselves have names, names like moon and shine and sulky. In the logical mode, reality is represented in more abstract terms, in the form of abstract relations which are independent of and make no reference to things. No doubt these relations, which taken all together constitute what we might call the logic of natural languages, have evolved by a process of generalization out of relations between things; and some of them, for instance “and”, are not hard to interpret in concrete terms (one can lay a set of objects side by side). But unlike experiential structures, logical structures present themselves in the semantic system as independent of any particular class or classes of phenomena. They are not the source of rules about what goes where. Again we have to deal with a distinction whose boundaries are fuzzy; there are the usual doubtful cases. More interesting, however, is the question whether languages differ as to what relations they are going to treat as logical. It seems to me that they do, although this argument will depend on our being willing to accept evidence “from below” – that is, to argue that, because we can identify a particular type of structure as characteristic of the expression of logical meanings, wherever we find that type of structure we shall assume it derives from the logical component. Since we are claiming these structural manifestations are only tendencies, such an argument is only tenable on the grounds that the type of structure that is generated by the logical component is in fact significantly different from all the other three. The principle is easy to state: logical structures are recursive. But we immediately encounter a difficulty here, a difficulty that is associated with the use of the term “embedding” to cover two different types of structure-forming process. In one type – which I have referred to as rankshift – the output of one network (by the application of realization statements) produces an element of structure which is a point of entry into the same (or some higher rank) network. A typical instance of rankshift is nominalization, where a function in the structure of a clause may be filled, not by a nominal group (the congruent form) but by something that itself has the structure of a clause, for example to come and spoil the fun and That you have wronged me in: 212

modes of meaning and modes of expression

It’s very rude of him to come and spoil the fun That you have wronged me doth appear in this Another example of rankshift is a restrictive relative clause, for example the lights went out in: The day the lights went out These are not true recursive structures. The recursion-like effect that is produced is an incidental outcome of the selection, at a particular place in structure, of an item from the same rank or from a higher rank in the constituent hierarchy. Clearly, this effect may appear more than once, as in a “house that Jack built” routine; but it is strictly a nonevent – there is no function involved that we could identify as a recursive function. True recursion arises when there is a recursive option in the network, of the form shown in Figure 14, where A:x,y,z . . . n is any system and B is the option ‘stop or go round again’. This I have called “linear recursion”; it generates lineally recursive structures of the form a1 + a2 + a3 + . . . (not necessarily sequential):

Figure 14

These are of two kinds: paratactic and hypotactic. The paratactic involve relations like “and” and “equals”, which are logically transitive (A ‘&’ B, and B ‘&’ C, implies A ‘&’ C; A ‘=’ B, and B ‘=’ C, implies A ‘=’ C). The hypotactic are logically non-transitive; these include relations such as “if” and “says”, where a “if” ß and ß “if” y does not imply a “if” y, nor does a “says” ß and ß “says” y imply a “says” y. In paratactic structures, because they are transitive, recursive order is expressed by structural (linear) sequence: there is no other way in which it can be expressed – no way, that is, in which a sequence A and B and C could realize an order “A & C & B”. The only departure from this strict sequence is nesting, where a sequence A and B and C and D represents an order, say, “A & (B & C) & D”. These are typically signalled in the phonological structure, e.g.: 213

word–clause–text

// 13 soup // 3ˆ/ ham and / eggs // 3^/ apple/pie // 1^ and / tea or / coffee // i.e. A ‘and’ (B ‘and’ C) ‘and’ D ‘and’ (E ‘and’ F) In hypotactic structures there is no such restriction; recursive order is not always signalled by the sequence, which may reverse it ( ß before a) or modify it in other ways ( ß inside a, for example). In English the principal instances of ‘recursive structures’ (i.e. structures deriving from recursive options) are as set out in Table 1. We should distinguish structures of this kind from those which happen to be realized as lists, like counting. Counting 1 2 3 4 5 6 . . . is not a recursive operation, linguistically; it is simply the enumeration of a list, like Sunday Monday Tuesday Wednesday Thursday Friday Saturday, but one which happens to be endless because there is a (nonlinguistic) recursive mechanism for generating the items in the list. In Table 1

214

modes of meaning and modes of expression

the same way, the days of the week do not form a closed system, in the linguistic sense; they form a lexical set whose members happen to be limited and fixed. In both instances, the infiniteness of the set of natural numbers and the finiteness of the set of days of the week are properties not of the language but of the social system. Logical structures are different in kind from all the other three. In the terms of systemic theory, where the other types of structure – particulate (elemental), prosodic and periodic – generate simplexes (clauses, groups, words, information units), logical structures generate complexes (clause complexes, group complexes, etc.) (Huddleston 1965). The apparent exception is the sentence, which is generated by logical structures; but this is merely a terminological exception – the sentence is, in fact, simply a name for the clause complex. While the point of origin of a non-recursive structure is a particular rank – each one is a structure “of” the clause, or of the group, etc. – recursive structures are in principle rank-free: coordination, apposition, subcategorization are possible at all ranks. The more restricted ones, like tense and report, are also the ones that are nearer the borderline; they are only just logical structures. Tense is particularly interesting because it has only come into this category within the last two to three centuries, and English appears to be unique in treating tense in this way.

9

Postscript

As a postscript, it should be noted that what is conceptually one and the same kind of relation may be coded in the semantic system in more than one way, i.e. may be realized through more than one of the functional components. The “and” relation, for example, may be coded in a logical system, expressed as coordination; or in a textual system, expressed as conjunction. The same is true of “yet”, “so” and “then”; these are much more weakly represented in the logical mode, but on the other hand they can also be coded experientially. Table 2 shows different codings of the temporal “then” relation. Although the distinct types of structure discussed in this paper appear at all ranks throughout the grammar, in English at least it is at the rank of the clause that they are most clearly in evidence. As an experiential construct, the clause is the locus of transitivity: it is the representation of the processes, participants and circumstances that constitute our experience of the real world. As an interpersonal construct it is the locus of mood and modality: the speaker’s adoption and assignment of speech roles and his judgement of the validity of the 215

word–clause–text Table 2 Functional mode

Example

logical : paratactic logical : hypotactic Textual Experiential

He sang, then people applauded, then . . . After he had sung, people applauded (First) he sang. Afterwards, people applauded Applause followed his song. / His song was followed by applause

proposition. As a textual construct it is the locus of theme and, typically, of information structure: the message as expression of the speaker’s concern and his presentation of what is “news”. The clause, therefore, is a multiply structured concept; it is clause as representation, clause as interaction and clause as message. And each of these provides its characteristic contribution to the total, characteristic in terms of the kinds of structure we have been talking about. The clause is orchestrated as melody (the experiential component, constellations of different notes), as harmony (the interpersonal component, an ongoing modal progression) and as rhythm (the textual component, the beat which organizes the sound into a coherent whole). This seems to support an interpretation of grammar which is structurally neutral, not based on the concept of constituent structure as the norm to which all grammatical patterns are expected to conform. In a systemic interpretation, language is treated as a resource, a potential; the various kinds of structure are the different means of expression of this potential. “Structure” is then no longer the basic organizing concept; instead, structural representations are derived from a more abstract conceptual framework that is paradigmatic rather than syntagmatic, according to which the semantic system of a language is comprised of sets of options in meaning, each of which makes some contribution to the expression in its final shape.

10

Summary

If we study the semantic system of a language we find that it consists of three major functional components: an ideational, an interpersonal and a textual; with the ideational further subdivided into an experiential and a logical. This pattern appears clearly at the semantic level itself: within each component, the networks of systemic options are closely interconnected, whereas between one component and another there are 216

modes of meaning and modes of expression

relatively few connections. In other words, choices made in one component affect other choices within the same component but hardly at all affect the choices in other components. It also appears clearly at a higher level, in the relation between language and situation. Broadly speaking, ideational meanings reflect the field of social action, interpersonal meanings reflect the tenor of social relationships and textual meanings reflect the mode of operation of the language within the situation. But it is at the lower level, that is in their grammatical realization, that these functional components are made manifest in the linguistic structure. In English, experiential options tend to generate constituentlike structures, actually constellations of elements such as can be fairly easily represented in constituency terms. Interpersonal options generate prosodic structures, extending over long stretches (for example intonation contours), which are much less constituent-like. Textual options generate culminative structures, elements occurring at the boundaries of significant units, and give a kind of periodicity to the text, which is part of what we recognize as “texture”. Logical options generate recursive structures, paratactic and hypotactic, which differ from all the other three in that they generate complexes – clause complex, group complex, word complex – and not simple units. Systemic theory takes the system, not the structure, as the basis of the description of a language, and so is able to show how these types of structure function as alternative modes of the realization of systemic options. They are then mapped on to each other to form the syntagm, which is the “output” of the lexicogrammatical system.

Notes 1. Written language may be (i) an alternative coding of meanings (ideograms); (ii) an alternative coding of wordings (characters, as in Chinese); or (iii) an alternative coding of sounds (syllabaries and alphabets). 2. Mathesius’s observation relates to ‘functional styles’ (orientation towards different functions in the use of language), not to functionally derived components of the system; but it is pertinent nevertheless: ‘The influence of functional styles on the lexical and semantic aspects of speech was stressed especially by Gro¨ber, . . . [who] distinguishes the subjective expression . . . and the objective expression . . . The subjective expression differs from the objective both quantitatively (inasmuch as it expresses by a pause, by tone or by gesture what the latter expresses by words; and further, as it repeats what could be expressed only once) and qualitatively (by

217

word–clause–text choosing other words than factual names of the things referred to), and, finally, locally (by placing sentence elements into positions not pertaining to them in objective speech). Both ways of expression are often combined in actual speech.’ (My italics throughout.)

218

Chapter Nine

TEXT SEMANTICS AND CLAUSE GRAMMAR: HOW IS A TEXT LIKE A CLAUSE? (1981)

1

Patterns of wording in the clause

Thanks to the work of our predecessors, and especially perhaps to that of outstanding figures of mid-century linguistics such as Sapir and Whorf and Bloomfield and Firth and Hjelmslev, linguists of subsequent decades have been able to extend our concerns upwards and outwards, from the syllable, through the clause, to the text. While broadening our vision in this way we have had to ensure that we do not lose sight of the syllable when we attend to the clause, nor of the clause when we attend to the text. Being a linguist means keeping all these things in focus at once: we are trained to do this both as observers, when we listen simultaneously to the meanings, the wordings and the sounds of speech, and as theorists, when we construct representations of language as simultaneously semantics, lexicogrammar and phonology. But it is not always easy to maintain this multiple focus, because each shift of attention involves a shift in two directions, a knight’s move that is a move both upwards and outwards. This same two-dimensional relationship was described by Hockett many years ago in a paper called ‘Linguistic elements and their relations’ (Hockett

Two works are combined in this chapter: ‘Text semantics and clause grammar: some patterns of realization’, first published in The Seventh LACUS Forum 1980, 1981, edited by James E. Copeland and Philip W. Davies. Linguistic Association of Canada and the United States, pp. 31–59. ‘How is a text like a clause?’, first published in Text Processing: Text Analysis and Generation, Text Typology and Attrition (Proceedings of Nobel Symposium 51), 1982, edited by Sture Allen. Stockholm: Almqvist and Wiksell, pp. 209–47.

219

word–clause–text

1961), in which he discussed the different statuses of morpheme and phoneme and contrasted the two possible pathways between them. A clause, that is to say, is not only bigger than a syllable; it is an entity of a different kind, at another level of abstraction. And it is this second relationship, that of realization, or coding, that is the critical one; the size distinction is typically associated with it, but not obligatorily – there can be a clause encoded as a single syllable. It took a surprisingly long time to clarify this two-dimensional relationship, and to accept that the relation of syllable to clause (or of phoneme to morpheme) was not simply one of part to whole, although it should have been fairly obvious seeing that the two are separated by the Saussurean line of arbitrariness. It is much harder to establish that a similar shift along two dimensions separates the clause from the text. A text is likewise – typically but not necessarily – bigger than a clause. But it is also, and more importantly, of a different level of abstraction. A text is a semantic entity rather than a formal, lexicogrammatical one; and this distinction is less easy to draw, because between the semantics and the grammar there is no such line of arbitrariness. (I shall return to this point below.) There is a problem in discussing text, if only because a text can be such a large object: every example takes up a great deal of space. One solution to this is to write rules for generating text but never actually to generate any. Another is that once suggested by Peter Wexler when he proposed to introduce a talk with the words “This paper is about the language of this paper”, so making the same entity serve both as text and as metatext. I have usually approached the problem in a different way, by using text that is so familiar that the audience can supply the missing parts for themselves, like Mother Goose or Alice in Wonderland. We need to face up to it somehow or other; there is something discouraging about a publication where the author is insisting on the importance of context but cites nothing longer than a decontextualized clause. When text is discussed in this way, with reference made only to isolated clauses, it is perhaps being assumed that the relation of clause to text is simply one of constituency. If a text is the same kind of thing as a clause only bigger, we can reasonably use clauses as instances while making observations about text. And perhaps this approach to text in turn reflects a presemantic view of language, in which it is assumed that the linguistic system is no more than grammar and phonology; so a text must be a grammatical unit, something that consists of clauses in the same way that a clause consists of words. Since there is no line of arbitrariness between semantics and gram220

text semantics and clause grammar

mar, this view is plausible. It is natural to think of text in the sense of the wordings that realize it. But it does cause some problems. The relations between the parts of a text are not such that we can set up structures whose exponents will be clause-like entities. The elements of structure of the text are more abstract; they are functional entities relating to the context of situation of the text, to its generic properties in terms of field, tenor and mode. It is not easy to explain the nature of a text if we treat a text as if it was a macrosentence, just as it was not easy to explain the nature of a sentence when a sentence was treated as if it was a macrophoneme. I am saying this at some length (despite the fact that I have said it often enough before) because I am now going on to say the opposite, or at least what will at first sight appear to be the opposite. Having insisted that a text is not like a clause, I now intend to suggest that it is. It is not that I have changed my mind on the issue. The point is rather that, once we have established that texts and clauses are of different natures, the one being lexicogrammatical (a construct of wording) the other semantic (a construct of meaning), we can then go on to note that there are several important and interesting respects in which the two are alike. But the likeness is of an analogic kind; it is a metaphorical likeness, not the kind of likeness there is between, say, a clause and a word. Starting from Hockett’s diagram, where one axis stands for constituency and the other for realization, we can link text to clause along the diagonal; the relationship between them is there but it is an oblique one, and this determines the kind of likeness we can expect to find. 1.1

How is a text like a clause?

Text is the process of meaning; and a text is the product of that process. A text is therefore a semantic entity; it is given to us in clauses, but it is not made of clauses, in the sense of being a whole of which the clauses are simply parts. So when we speak of the problem of relating clause to text as one of getting “from micro to macro”, this is only one aspect of the relationship. It is true that texts are, on the whole, larger than clauses; what is more significant, however, is that they are one level of abstraction beyond the clause. The relationship is not so much one of size as one of overt to covert; the text is realized in clauses. In “scale-and-category” terminology (Halliday 1961), the relationship of clause to text is one of exponence as well as one of rank. This has consequences for the ways in which the properties of a text are made manifest. For example, the notion that a text has “structure” 221

word–clause–text

would imply, if a text was a lexicogrammatical entity, that the elements of structure would be “filled” by classes of the clause (perhaps with some intermediate units) in the same way that elements of structure of the clause are filled by classes of the group. But it is difficult to specify text structures in a way which represents the text simply as a higher-rank grammatical constituent; the configurations are different in kind, and the relationship to the wording is both indirect and complex. Functional elements of text structure are not translatable into strings of clauses. A text is therefore not “like” a clause in the way that a clause is like a word or a syllable like a phoneme. But by the same token, just because clause and text differ on two dimensions, both rank (size level) and exponence (stratal level), there can exist between them a relation of another kind: an analogic or metaphorical similarity. A clause stands as a kind of metaphor for a text. In this paper I shall refer to some well-known properties of a text, and then, drawing on some recent text-linguistic studies in a systemic-functional framework, try to show that these are paralleled in significant ways by properties of a clause that are in some sense (not always the same sense) analogous. The textual properties to be considered are the following: 1. 2. 3. 4. 5. 1.1.1

A text has structure. A text has coherence. A text has function. A text has development. A text has character. A text has structure

For at least some registers, perhaps all, it is possible to state the structure of a text as a configuration of functions (Hasan 1979). A generalized structural representation is likely to include some elements that are obligatory and others that are optional; and the sequence in which the elements occur is likely to be partly determined and partly variable. Most of the actual formulations of text structure that have been put forward seem to relate to one broad genre, that of narrative. The original source of inspiration for these was Propp’s theory of the folk tale. The structure of traditional oral narrative has been investigated in detail within tagmemic and stratificational frameworks, on foundations provided by Longacre and Gleason. A well-known representation of another kind of narrative is Labov and Waletsky’s structural formula for narratives of personal experience: 222

text semantics and clause grammar

Abstract ^ Orientation ^ Complication • Evaluation • Resolution ^ Coda Outside narrative registers, Mitchell (1957) set up structures for the “language of buying and selling” in Cyrenaican Arabic, recognizing three subvarieties having some common and some variant features: Market and shop transactions: Salutation ^ Enquiry as to object of sale ^ Investigation of object of sale ^ Bidding ^ Conclusion Auctions: Opening ^ Investigation of object of sale ^ Bidding ^ Conclusion Mitchell refers to these as “stages” and comments that “stage is an abstract category and the numbering of stages does not necessarily imply sequence in time”. Hasan considers that structure is a property of texts in all registers. For any register, specified at any appropriate degree of delicacy, it should be possible to state a generalized structure by reference to which any actual text can be interpreted. Her suggested formula for a particular class of transactions, retail sale in a personal service food store, is as follows: ((Greeting •) Sale initiation ^) ((Sale inquiry •) (Sale request ^ Sale compliance) ^) Sale ^ Purchase ^ Purchase closure (^ Finis) Martin (1980), who uses “functional tenor”, or rhetorical purpose, as the superordinate concept for characterizing registers, gives the following structural formula for the register of “persuasion”: Set ground ^ State problem ^ Offer solution ^ Evaluate solution (^ Personalize solution) 1.1.2

A text has coherence

One of the most frequent observations made about texts that are felt to be defective in some way is that they do not “hang together”: they “lack coherence”. A text has coherence; it forms a unity, a whole that is more than the sum of its parts. Coherence is a complex property to which many factors contribute. One way to approach it is through the category of cohesion, as defined by Halliday and Hasan (1976). Cohesion is a necessary but not a sufficient condition of coherence. The different types of cohesive relation are the fundamental resources out of which coherence is built. But the mere presence of cohesive ties is not by itself a guarantee of a 223

word–clause–text

coherent texture. These resources have to be organized and deployed in patterned ways. (a) Ruqaiya Hasan identifies a feature which she calls “cohesive harmony” (Hasan 1984). This is based on the recognition of cohesive chains of a lexico-referential kind. Such chains have been called “participant chains”, since their most obvious manifestation is in the form of sequences such as a little boy . . . John . . . he . . . he which are coreferential to a narrative participant; but Hasan points out that they are not confined to participants, nor are they necessarily coreferential. They may be “identity chains” or “similarity chains”; and the element that is chained may be of any kind – participant, human or otherwise, including institutions and abstract entities; attribute or circumstantial element; event, action or relation; fact or report; or any recoverable portion of text. What makes a text coherent is not merely the presence of such chains but their interaction one with another. In comparing texts which were judged coherent, by herself and others, with those which were not, Hasan found that in the former it was always the case that a significant majority of the tokens in each chain were functionally related with tokens in some other chain; while in the latter the ones that were related in this way were only a minority. Specifically, they were related in some experiential function – transitivity, or an extension thereof – either to each other, or identically to some third function. For example, a pair of tokens might be related to each other as Actor to Process; or by their both having the function Actor relative to some other element as Process. In other words, in order to achieve coherence there had to be not merely parallel currents of meaning running through the text, but currents of meaning intermingling in a general flow, some disappearing, new ones forming, but coming together over any stretch of text in a steady confluence of semantic force. The following illustration (Appendix 1a, pp. 247–50) shows the difference in cohesive harmony between two stories told by children (Hasan 1980). Hasan points out that these two texts differ hardly at all in the number and distribution of cohesive ties, or in the proportion of their lexical and referential tokens that appear in chains. Where they do differ is in the proportion of such tokens that occur in interaction with others from other chains; in other words, in the extent of cohesive harmony displayed. This has proved to be a significant element in discrimination between passages perceived as coherent and those where coherence is felt to be lacking. (b) Another type of cohesive relationship identified by Halliday and 224

text semantics and clause grammar

Hasan is that of conjunction, the linking of successive elements of a text by the semantic relations of ‘and, or, nor, viz, yet, so, then’: additive, including alternative and appositive; adversative; causative, and termporal. These are described and illustrated, like the other cohesive systems, with reference to cohesion between adjacent sentences. But we can identify three ways in which conjunctive relations create coherence in the more extended sense. 1. James R. Martin has shown (1992) how conjunctive relations create texture in dialogue by linking sentences that are not adjacent, spanning whatever material may intervene. His interpretation of the system of conjunction, in which he modifies the version given in Halliday and Hasan, eliminating the category of “adversative” and grouping “as against” with additive and “contrary to expectation” with consequential (causal), is expressed in the network in Appendix 2, pp. 249–51. The category of “implicit”, also not in Halliday and Hasan, accounts for those instances where the semantic relationship is present but there is no conjunction or textual (discourse) adjunct making it explicit. Martin’s analysis of a short passage of dialogue includes instances of conjunctive relations bridging a number of intermediate turns (Appendix 2). 2. Conjunctive relations may be set up between passages of any extent. Not only a turn in a dialogue, but an episode, argument, scene or any other functional element may be “picked up” conjunctively in a succeeding portion of the text. In this way the presence of conjunctive relations creates coherence over extended passages of discourse. An example of this would be a section beginning Because of this, where this refers to the whole of some preceding argument. 3. Halliday has suggested (1975) that the “textual” properties of a text – the cohesive patterns and those of ‘functional sentence perspective’ – tend to be determined by the “mode”, the function ascribed to the text in the given context of situation, the purpose it is intended to achieve. Thus the mode would determine the balance among the different types of cohesive resource – reference, ellipsis, conjunction, lexical cohesion; and within conjunction, the relative weight accorded to internal and external conjunctive relations and to the various semantic alternatives within each. In this way the kind of conjunctive relations found in the text will be characteristic of the register (as defined on the dimension of ‘mode’) to which the text belongs. An illustration of this principle is provided by Mary Ann Eiler (1979) in her study of expository writing by ninth graders in an American 225

word–clause–text

high school. She has shown how the conjunctive relation of specific instance to general principle, coded in Martin’s network as: (internal / comparative) : similarity : nonexhaustive is the major conjunctive factor giving coherence to these texts. For example, in one part of a text there occurs the sentence Odysseus’ friends and anyone else who heard his story respected him. Elsewhere in the same text we find: Heroic men are very much respected and idolized. (Either member of such a pair may come first.) There is lexical cohesion between individual items – the repetition of respect. There is cohesion between Odysseus and heroic (men), with ‘Odysseus’ being a hyponym of ‘hero’. But between the two sentences as wholes there is a conjunctive relation – itself an extension of hyponymy – such that the second one stands as a general principle of which the first one offers a specific instance; and this type of conjunction is a distinguishing feature of the sort of expository discourse she is investigating. 1.1.3

A text has function

A text unfolds in a context of situation, and has some identifiable rhetorical function with reference to that context. This is the domain of functional theories of language, insofar as these are concerned with the process (‘functions of the utterance’) as distinct from the system (‘functional components of the grammar’). The assumption of those theories that are functional in the former sense, which we may call “process-functional theories”, is that a text can be interpreted as having one or other of a small set of “rhetorical” functions – exclusively, or predominantly, or in some recognizable combination. Malinowski, starting from an ethnographic standpoint, identified the functions “pragmatic (active / narrative) / magical”. Buhler, from a psychological perspective, recognized “expressive / conative / representational”, with orientation respectively towards speaker, addressee and the rest of reality; to these Jakobson later added three more, having orientation towards the channel (“phatic” – inappropriately), the message (“poetic”) and the code (“metalinguistic”). Britton, as an educational theorist, realigned Buhler’s categories so as to group conative with representational (both being “transactional”), and added the poetic as a fourth. In the work of Desmond Morris we 226

text semantics and clause grammar

can even find an ethological categorization: “mood talking / grooming talking / information talking / exploratory talking”. All these apparently very divergent interpretations have in common one basic distinction: that between language as reflection and language as action – between discourse that is oriented towards the function of the representation of experience (Malinowski’s narrative, Buhler’s and Britton’s representational, Morris’ information talking) and discourse that is oriented towards the function of interpersonal behavior (Malinowski’s active, Buhler’s and Britton’s expressive and conative, Morris’ mood talking and grooming talking). There is also a partial recognition of a third orientation, towards an imaginative function (magical, poetic, exploratory). In work on register the rhetorical function is treated as one component in the context of situation of the text. Halliday, McIntosh and Strevens (1964) proposed a tripartite framework for interpreting the register: (i) the nature of the social process in which the text is embedded – ‘what is going on’ (field); (ii) the interpersonal relationships among the participants – ‘who are taking part’ (tenor); (iii) the role assigned to the text, including both medium and rhetorical function – ‘what part the language is playing’ (mode). Gregory (1967) separated the rhetorical function from the medium and associated it more closely with the participant relations, referring to the latter as “personal tenor” and to rhetorical function as “functional tenor”. Ure and Ellis (1979) take this one step further, recognizing four distinct categories of field, mode, formality (personal tenor) and role (functional tenor). Martin (1980) proposes to return to the rhetorical perspective and treat functional tenor, which he defines as “the purpose of the text”, as superordinate to field, mode and personal tenor. His argument is that “it is the functional tenor of a text that is responsible for its structural formula” – in other words that function (in this sense) determines structure. 1.1.4

A text has development

A text is a dynamic process; it has a semantic ‘flow’, a development of ideational and interpersonal meanings. This flow or development is carried forward by the interaction of speaker and listener; obviously so in the case of dialogue, but so also in monologic modes where the active participation of a listener still contributes to the construction of meaning. Even in written language the semogenic process is essentially of the same nature; researchers in writing theory now strongly insist on the part played by the imagined audience in the process of written composition. 227

word–clause–text

Peter Fries refers (1981) to the “method of development” of a text. Below is one of his examples, with the relevant sections of the commentary (Appendix 3, pp. 254–5). In the paper from which this is taken, Fries is interpreting the development of a text in terms that relate it to the concepts of theme and rheme. His argument proceeds in four steps: (1) the pattern of theme–rheme organization in the clause is a function of the register; (2) the pattern in the choice of theme is a function of the method of development of the text; (3) “theme–rheme” is clearly distinct from, but also clearly related to, “given–new”; (4) the theme–rheme organization of the clause “fits into a larger pattern governing the information flow in sequences of sentences in English discourse in general”. Fries regards the theme as a “ground”: “In English discourse at least there seems to be a strong tendency to set up certain information as a ground first, and then to introduce later information using that ground as a basis for evaluation and comparison”. (It is reasonable to understand “information” rather broadly here; presumably the ground may be any configuration of ideational and interpersonal meanings.) So there is a movement from the ground, to something that is defined by it as “not ground”. There is also, we may add, a complementary movement in a text, which is a movement towards rather than away from: a movement to what we may call the “point” (generalizing from Fries’ “main point”), from something that is defined by it as “not point”. Rhetorical theory has always stressed the beginning and the end: topic sentence, introductory paragraph &c. on the one hand, and culmination, climax, summation &c. on the other. But it is important to stress that this is not a single movement. A text is a kind of diminuendo – crescendo, beginning and ending with prominence; but the prominence is of two different kinds. Rather than thinking this time of a flow, a unidirectional current with a set of rapids at each end, we should perhaps change the metaphor to that of a gift, or rather an exchange, in which there is a shift of focus from donor to recipient in the course of the exchange, or rather from giving to receiving. The process begins as giving and ends as receiving; but “giving – not giving” is not the same movement as “not receiving – receiving”. Moreover although the process must start with the giving – until then there is no exchange – it need not necessarily end with the receiving, which may occur quite early and be followed up in various ways. In the development of a text, phasing out the “ground” goes along with phasing in the “point”. This pattern is one that can be repeated over as many levels in the hierarchy of constituents as the text has in it, from the entire text down to the individual clause. 228

text semantics and clause grammar

1.1.5

A text has character

A text is an instance of a particular “register”; it has the generic features characteristic of that register, associated with a particular alignment of the features of the context of situation – the “contextual configuration”, in Hasan’s terms. A text is also an individual entity, having a specific character of its own distinct from that of other texts within the same register. Some texts are highly valued as individual texts, and one of the interests of text studies is the stylistic one, the attempt to understand the unique qualities of a highly valued text, and what it is that makes it highly valued. (A) The generic character of a text is in principle predictable from its context of situation. Taking the categories of field, tenor and mode as a predictive framework, Halliday proposed that the ideational meanings of a text tend to be determined by the field, the interpersonal meanings by the tenor, and the textual meanings by the mode, suggesting that this was how listeners and readers make predictions about what is coming next – predictions that they must make if meanings are to be successfully exchanged. For illustrations of this see Halliday (1975, 1977), Halliday and Hasan (1980). An example of a register variable is provided by Jean Ure’s study of lexical density (1971). Ure shows that the lexical density of a text is a function of its level of formality, the amount of self-monitoring done by the speaker or writer; writing has a higher density than speech, with what she calls “language-in-action” having the lowest density of all. Lexical density can be defined as the number of lexical items per unit grammar (per clause, as the most natural measure), though Ure measures it as a percentage of running words; in her sample of 68 texts, comprising about 21,000 words each of speech and writing, the values range from 57 per cent (formal written) to 24 per cent (casual spoken), and all texts with a density of 35 per cent and below are dialogue. Charles Taylor (1979) has used both these measures in his study of the language of high school textbooks in New South Wales. Robin Melrose (1979) has suggested another variable that defines the generic character of a text, one that relates to the field instead of the mode. He finds that each text will tend to be characterized by a particular “message type”. Melrose distinguishes factual, phenomenal and relational message, with various subcategories; deriving these from the material, mental and relational processes of the transitivity system (Halliday 1967–68; 1975). An instance of a text with relational : attributive messages is given in Appendix 4, p. 255–9. 229

word–clause–text

According to Melrose, certain other features are associated with these message types: different patterns of theme, conjunction and lexical cohesion, and also different kinds of “message superordinate”, the summative expression described by Winter (1977); the table in Appendix 4 shows Melrose’s hypothesis about these. (B) The specific character of a text is what distinguishes it from other texts of the same genre, those features which are not predictable from the contextual configuration. This has sometimes been characterized as deviation, the text creating its own rules; but “breaking the rules” is a minor and relatively insignificant form of uniqueness. What a text does is to create its own norms, its own unique selection from the resources of the system by which it is generated. Many texts in daily life are not unique at all; the same things have been said countless times before. Such texts are often of particular interest to an ethnographer (and, one might add, to a linguist). Other texts are presumed to be unique; this class includes all those texts we think of as literature. But any text can be described and interpreted as an event that is sui generis. If the qualities that we perceive as specific to a text reside not merely in the particular combination of features selected but also in a special highlighting of some aspect of these, we usually try to relate this highlighting to our interpretation of the underlying theme, seeking the kind of semiotic convergence that would explain the particular impact that the text has on us. A text is a polyphonic composition of ideational, interpersonal and textual “voices”. The ideational voice provides the content: the things, facts and reports; processes, participants and circumstances; the logical relations of different kinds. The interpersonal voice provides the interaction: mood, modality, person, polarity, attitude, comment, key. The textual voice provides the organization: thematic and informational prominence; grammatical and lexical cohesion among the parts. The “character” of the text is its pattern of selections in these various voices, and the way they are combined into a single whole. The accompanying extract from J. B. Priestley’s An Inspector Calls (Appendix 5, pp. 257–9) is a piece of dramatic dialogue which is distinctively characterized by the foregrounding of modality. The first part is dominated by modalized assertions, which move from probability to obligation, the second part by assertions about obligations; the final speech shifts into narrative, returning at the end to the assertive mode but this time without modalities. Herein lies the movement of the play, which is concerned with social responsibility (obligation) acted out through chance (objective probability); the interplay of these 230

text semantics and clause grammar

modalities determines a strong narrative line leading up to a nononsense conclusion. In this passage the linguistic system functions as a symbol within the process; this is its characteristic contribution to our interpretation of the meaning of the play. What strikes us about these properties of a text is that all of them are also, in some sense, properties of a clause. The notion of text structure is clearly modelled on that of clause structure. A clause is a configuration of functions; so is a text. As said at the beginning, this is not to argue that a text is a larger whole of which a clause is a part. A clause is a lexicogrammatical object, a structure of wording; whereas a text is a semantic object, a structure of meaning. The resemblance is like that of clause to syllable. A syllable is a phonological object, and therefore not part of a clause; but it has structure in the same sense. With one difference, however: between clause and syllable runs the line of arbitrariness in language. In the realization of wording in sound, natural symbols are the exception. But there is no such line of arbitrariness separating the clause from the text. The realization of meaning in wording is largely “natural”, non-arbitrary. This leads us to speculate whether the text may display the same kind of multiple structuring that is found in the clause, ideational, interpersonal and textual. The representations of text structure proposed by Hasan and others suggest that we might want to interpret a text as having, potentially at least, an ideational structure relating to its field and an interpersonal structure relating to its tenor, rather than (or as well as) a single structure deriving from the mode (functional tenor) as proposed by Martin. Benjamin N. Colby and Lore M. Colby (1980) analyse traditional and other oral narratives in terms of “eidons”, which are ideational elements of text structure set up to allow for the interpretation of the text as an ethnographic document, as a window on the culture. The theory and method are set out in Colby’s study (1973) of Eskimo folk tales. The notion of the eidon Colby ascribes to Gregory Bateson’s interpretation (1936) of Iatmul culture and the Naven ceremonies. Bateson talks of the “eidos” and the “ethos” of a culture, and of “eidological” and “ethological relationships” – the realization of eidos and ethos in cognitive and affective aspects of cultural behavior. Colby’s interpretation of text structure is “eidological”, corresponding to the ideational component in the structure of the clause; he suggests the possibility of a parallel “ethological” interpretation corresponding to the interpersonal component in the structure of the clause. 231

word–clause–text

We should not press the analogy too far. But if it seemed useful to set up simultaneous structures in a text along these lines we might ask whether there is the same kind of structural variation as we find in the clause, with the eidological structure being “particulate” (represented by definable segments of the text) and the ethological being “fieldlike” (represented by overlapping prosodies in the text). (The “wavelike” periodic movement corresponding to the textual dimension of clause structure has already been referred to under 1.1.4 above; cf. further below.) (Cf. Halliday 1979.) To say that a text resembles a clause in having coherence is not to say very much, since the coherence in a clause is created by its structure, whereas coherence in a text largely depends on cohesion. Cohesion is the resource that takes over, as it were, when grammatical structure no longer holds (i.e. above the clause complex). We could point out that cohesion also obtains within clauses; we find reference, substitution, ellipsis, conjunction and lexical cohesion all operating between elements in the same clause, for example: M’s evening speech caused more fuss than his morning one had. C R L S E But this is a superficial similarity. A more significant analogy can be found with the notions of cohesive harmony and conjunctive relations discussed in 1.1.2 above. Ruqaiya Hasan’s work showed how lexico-referential motifs enter into a text not as isolated motifs but as interlocking chains having some kind of regular functional relationship with each other. But these functional relationships are relationships within the clause; and this reflects the fact that the elements in these chains themselves cannot occur as isolated entities. Names have no place in language except in function with other names; and the functions are defined within the structure of the clause. The conjunctive relations discussed by J. R. Martin (1992) are also derived from relations with the clause. Consider a series such as the following: She didn’t know the rules. She died. She didn’t know the rules. So she died. She died, because she didn’t know the rules. She died because of not knowing the rules. That she died was because she didn’t know the rules. 232

text semantics and clause grammar

That she died was caused by her not knowing the rules. Her ignorance of the rules caused her death. The same conjunctive relation, the “external causal”, can be coded in very many ways. It can appear as a relationship within the clause, realized lexicogrammatically; but it can also serve to link segments of the text, at any distance and of any extent. The kind of coherence that is achieved by the presence in the text of semantic relationships of the conjunctive kind is essentially a clause-type coherence, one that is based on relations defined systematically within the transitivity system of the clause. The notion that a text has function is again closely related to an analogous feature of the clause, also one that is coded in the lexicogrammatical system: that a clause has a speech function, realized by the mood system. The speech function of the clause – in simplest terms, as statement, question, command, offer, or a minor speech function – is represented by the grammatical categories of declarative, interrogative and so on; this is the rhetorical function of the clause, and the whole range of rhetorical functions that we assign to text are simply the “mood” of the text. (Cf. Martin (1980) for speech-functional analysis of dialogue.) The development of the text again has its analogy in the clause. This has already been made clear from the example cited in 1.1.4, since Peter Fries used the theme–rheme structure of the clause as the source from which to derive the method of development of the text. We can generalize this still further by bringing in the notion of information structure, the given–new movement within the clause. In its “textual” aspect, a clause has a wave-like periodic structure created by the tension between theme–rheme (where theme is the prominent element) and given–new (where new is the prominent element); the result is a pattern of diminuendo–crescendo, with a peak of prominence at each end. There is a balance of development (i) away from the theme, and (ii) towards the new. But these are separate movements. They are in phase in the unmarked, “default” case, where the theme is selected from what is given, and the news is put into the rheme. But they can also be out of phase, and this gives an alternative pattern of texture to the clause. Putting the two out of phase means locating the new (the focus of information) somewhere other than at the end of the rheme; this as it were changes the wave shape but does not disturb the essential periodicity. This pattern is the “method of development” of the clause. It is 233

word–clause–text

closely analogous to what takes place in a text; not only over the whole text but also in structurally defined intermediate units within the text. The classic movement of a paragraph, beginning with a topic sentence (from theme to elaboration) and culmination – having a high point, unmarkedly but not obligatorily final – in a climax (from prelude to main point), is one of the clearest manifestations of the analogy between clause and text. It is in the clause that this movement is displayed in the most systematic and clearly motivated form. Finally a clause can be said to have “character” in both the generic and the specific sense. If a text is typified by virtue of its being organized around the expression of processes of a particular type, the clause is the unit in which these processes are realized and categorized. The clause is the locus of the transitivity system: the system of material, mental and relational processes, together with their numerous subcategories. Thus analogous to the major types or genres of text are the major types or classes of the clause, each being characterized by the selection of a dominant process type. But each clause is also a unique combination, or potentially unique combination, of features deriving from the different semantic functions, ideational, interpersonal and textual. Moreover any one or other of these may be foregrounded: the clause may display an orientation towards any one, or any combination, of the various systems and their subsystems. The extract from J. B. Priestley’s An Inspector Calls (Appendix 5, pp. 257–9) gave an instance of a clause that is oriented towards a certain type of modality: interpersonal meanings are highlighted, with the speaker’s skeptical doubting as the predominant rhetorical function. The passage cited is a unique interplay between the exploration of probabilities and the assertion of obligations, and so is the entire text. No one clause can recapitulate the whole; but all contribute, and some achieve a remarkable likeness – a likeness that is possible because the systems of the clause not only embody all the semantic components from which the text is built but do so in a way that allows an infinitely varied, almost text-like balance among them. Thus the properties that we recognize in a text are also, in a transformed way, properties that we recognize in a clause. A clause is a kind of metaphor for a text – and a text for a clause. That this is possible is due to two things: one, that a text is not only (typically) larger than but also more abstract than a clause; two, that on the other hand there is no line of arbitrariness between clause and text as there is between clause and syllable. Hence it is not only in a formal sense that a text is like a clause. It is no accident that it is possible to illustrate so 234

text semantics and clause grammar

many of the relations in a text by reference to relations in a clause. The illustrations given here already contain within themselves a demonstration of this conclusion. In showing that a text has structure, coherence, function, development and character, we cannot help at the same time showing that a clause has all these things too, though in an interesting variety of different ways. Presumably this is how clauses evolved – as the most efficient means of encoding text. SUMMARY TEXT configuration of contextually motivated semantic functions

CLAUSE configuration of semantically grammatical functions

Coherence:

by cohesion (i) cohesive harmony: chains interrelate by function in (semantic) transitivity (ii) conjunctive relation: between messages or larger parts of text

by structure (i) names (things) interrelate by function in (grammatical) transitivity (ii) conjunctive relation: between parts of clause, as major or minor process

Function:

has “functional tenor” (rhetorical function as text)

has “speech function” (rhetorical function as speech act)

Structure:

Development: has “information flow”: ground - - - - > - - - - > point

{

Character:

1.2

}

generic: selects “favourite” process type as message type specific: foregrounds one or more systems

{

has “information structure”: theme - - - - > - - - - > focus / new

}

generic: selects process type specific: foregrounds selections from one or more systems

A functional interpretation

We shall be able to explore the relationship between clauses and text more thoroughly by starting from a functional interpretation of the clause; so it may be helpful to comment first on functional theories of 235

word–clause–text

language. Functional theories of language came originally from outside linguistics; the consequence was that they were only theories of the text – they had nothing to say about the system. According to such a theory, any piece of text can be assigned a particular function, in the sense that it is oriented, exclusively or at least predominantly, towards some communicative purpose. The unit that is described in this way may be a very small piece of text, realized as one clause (the functions are then “functions of the utterance”); or it may be a larger piece constituting a recognizable semiotic event. The best-known functional schemata are two dating from around 1930, the ethnographic one of Malinowski (1935) and the psychological one of Karl Bu¨hler (1934). Bu¨hler’s scheme is interesting because although extralinguistic in intent it is one that is explicitly derived from language – that is, from the linguistic system – in the first place: his tripartite framework of expressive, conative and representational functions denotes text that is oriented, respectively, towards speaker, addressee, and the rest of the universe – in other words the first person, second person and third person categories of the Indo-European verb. This is similar to the way in which various logical relations originally derived from natural language have been transformed into non-linguistic relations and then turned back on to language as explanations of linguistic forms. The interest of such functional schemata for the linguist is that the functions arrived at are not in fact simply functions of the text. If they were, they would be of limited concern; but they are more than this – they are functions that are built in to language as the fundamental organizing principle of the linguistic system. We shall not be surprised at this, if we take a Hjelmslevian view of language as system and process: if we accept that language and text are one and the same thing, and that the system evolved as a means of serving human intentions through the creating of text. It is only if we set up artificial dichotomies like langue and parole, or competence and performance, that we are surprised when a system displays properties relating it to its use. Now, despite the divergencies that separate Bu¨hler’s and Malinowski’s functional theories, from each other and from various subsequent schemata, divergencies that are a natural consequence of the different purposes for which they were devised (ethnographic, psychological, ethological, educational, etc.), there is one feature that strikes us as common to all of them. They all share in the fundamental opposition of action and reflection, the distinction between language as a means of doing and language as a means of thinking. The former is Bu¨hler’s first and second person function, Malinowski’s active function; the latter is Bu¨hler’s 236

text semantics and clause grammar

third person function, Malinowski’s narrative function. And the opposition is incorporated into the semantics of natural languages, in the form of what I have referred to as the “interpersonal” and “ideational” components. (The distinction between first and second person language, however, is not a systematic one; the two are simply different angles on the same interpersonality.) For all human beings, in all social groups, the environment in which they live has these two validations: it is something to be acted on, turned into food or shelter or other needs; and it is something to be thought about, researched and understood. Language has evolved to serve both these elementary functions. The reflective mode is coded directly as the ideational element in the semantic system. But since language is symbolic, one who speaks does not act on reality directly but only through the intermediary of a listener. Hence the active mode, when translated into a network of semantic systems, comes to be coded as interpersonal. While these two functions are given to language from the outside, as it were, by its role in human situations, in order to fulfil such roles a language has to have a third semantic component, whereby it is enabled to latch on to those situations in a systematic way. There must be a relevance function, a system of meaning potential which allows a text to cohere with its environment, both the non-linguistic environment and that part of the environment which consists of what has been said before. So there is a third component in the semantics of natural language which only an immanent linguistics will discover, since it has no transcendent motivation; this is the contextualizing function – or the “textual” function, as I have called it, because it is what makes text text, what enables language to be operational in culturally meaningful environments. Now a clause is a complex realization of all these three semantic functions. It has an ideational component, based on transitivity, the processes, participants and circumstantial elements that make up the semantics of the real world, and including the onomastic systems that classify these into nameables of various kinds. It has an interpersonal component, consisting of mood, modality, person, key and all the various attitudinal motifs that come to be organized as meaningful alternatives. And it has a textual component, the “functional sentence perspective” (thematic and news-giving systems) and the cohesive resources of reference, ellipsis and conjunction. Each of these components makes its contribution to the total make-up of the clause. What we identify as a clause is the joint product of functional-semantic processes of these three kinds. 237

word–clause–text

But what is the nature of the contribution, in each case? We are accustomed to thinking of this in structural terms: that each semantic component generates its own particular tree, a configuration of parts each having a distinct function with respect to the organic whole. The ideational component generates “actor–action–goal”-type structures: configurations of Process, Medium, Agent, Beneficiary, Range, Extent, Location, Manner, Cause and so on. The interpersonal component generates so-called “modal-propositional” structures: configurations of Subject, Finite, Modality, Predicator, Complement and Adjunct. And the textual component generates thematic and informational structures, configurations of Theme and Rheme, and Given and New; as well as cohesive elements of a non-configurational kind. We can represent all these in structural terms, using the linguist’s traditional notion of structure: the simplest of all possible forms of organization, that of parts into wholes. Because this notion of constituent structure is so simple, it is natural that a linguist should want to do as much as possible with it. And it can be made to do quite a lot. But there comes a point where it ceases to be appropriate; where moulding the facts so that they fit the notion of constituency will distort them rather than just simplifying them. With a multifunctional interpretation of the clause we reach this point. As outlined in Chapter 8, the contributions that are made by the three functional-semantic components to the form of the clause are of three rather different kinds. As far as the ideational systems are concerned, these do tend to generate part–whole structures; they are realized by organic configurations which themselves, and whose constituents, are reasonably clearly bounded, such that it can be specified where one clause element leaves off and the next one starts. But this is not nearly so true of interpersonal systems. Interpersonal systems tend to generate prosodic patterns that run all the way through the clause: not only intonation contours, though these provide a clear instance, but also reiterations of various kinds like those that are typical of modality in English, e.g. surely . . . can’t . . . possibly . . . can . . . d’you think in: Surely they can’t possibly be serious about it can they d’you think ? Textual systems generate patterns that differ from both of these, culminative patterns formed by peaks of prominence; and since these peaks typically appear at the beginning or the end of the clause, where there is a sequence of clauses they result in a kind of periodicity, a movement from a clause-initial peak via an off-peak medial state to a clause-final peak which is then sustained to form the initial peak of the 238

text semantics and clause grammar

succeeding clause. Thus a clause is at one and the same time particle, field and wave, as Pike suggested more than twenty years ago (Pike 1959), although the details of this interpretation are not quite the same as those worked out by Pike. Now, the significance of this step in our interpretation lies not only in establishing that these three distinct patterns of realization go to make up the English clause, but also in the fact that they appear to be non-arbitrary; this is clearly important when we come to ask whether such tendencies are found in every language. The grammar of languages is a natural grammar; as I expressed it earlier, there is no line of arbitrariness between semantics and grammar as there is between grammar and phonology. If the clause is at once particle, field and wave this is because the meanings it has to express have different semiotic contours, to which these three realizational forms correspond in a natural, non-arbitrary way. The particular nature of ideational structures reflects the relative discreteness of the phenomena of our experience. Consider cows eat grass: we know where the cow begins and ends, what eating is and is not, which part of reality consists of grass and which part consists of other things. Many of our perceptions are schematized into entities that are bounded in this way, and the constituent-like form of the wording reflects this fact: the word cow has an outline because the object cow has an outline. Of course not all experience is like this; indeed I have always tended to emphasize the unboundedness of many phenomena, the indeterminacy and the flux; and I share Whorf’s view that language itself, once it has been constituted in this way, strongly influences the forms that our perceptions take. Nevertheless there is a basic fit between the discreteness of words and the discreteness of things; otherwise we should not be able to talk about the things at all, or explain contrastively those instances where the fit of words to things is less than perfect. By contrast, the interpersonal kind of meaning is a motif that runs throughout the clause; and this is represented by lexicogrammatical or phonological motifs that are likewise strung unboundedly throughout. The speaker’s attitudes and assessments; his judgements of validity and probability; his choice of speech function, the mode of exchange in dialogue – such things are not discrete elements that belong at some particular juncture, but semantic features that inform continuous stretches of discourse. It is natural that they should be realized not segmentally but prosodically, by structures (if that term is still appropriate) that are not particulate but field-like. The linguist’s tree is an inappropriate construct for representing structures of this kind. 239

word–clause–text

Thirdly, the undulatory movement by which textual meanings are encoded in the English clause may again be in some sense a natural form for their representation. All the patterns I have been discussing vary from language to language, as is very obvious; those of English merely provide one specific instance of something that seems to be a general tendency in the expression of meanings of each kind. The English clause, as a message, is a movement from prominence to prominence, a diminuendo that is then picked up and becomes a crescendo; but the prominence is of two different kinds. It is a movement away from a Theme, to something that we can characterize as non-Theme; that is the diminuendo aspect. It is a movement towards a New, from something that we can characterize as non-New; that is the crescendo aspect. But Theme is not the same as non-New, nor is New the same as non-Theme; there are two movements here, not one. Their relationship is less automatic than the above formulation implies, and they can be combined in other ways besides; what is described here is just the unmarked, typical form. The essential point is that the two types of prominence differ; and that they differ as speaker to listener. The Theme is speaker-oriented prominence: it is “what I am on about” (grammarians used to call it the psychological subject). The New is hearer-oriented prominence: it is “what I present as news to you”. The English clause is textured by this shift in its orientation, from speaker-prominence to listener-prominence. Each clause is in this sense a kind of gift, one move in an exchange, symbolized by the change of perspective from me to you. So when Alice says: it turned into a pig in answer to the Cheshire Cat’s question What became of the baby?, she begins with the Theme it (‘I’m going to tell you about something’) and ends with the New a pig (‘here’s what is news to you’). In this case, Alice has obligingly chosen as her Theme the thing that the Cat had asked her about, namely the baby, realized by the anaphoric reference item it; and she has kept the news, its change of state, till last. Alice is being helpful, keeping the wave pattern of the dialogue in phase. But she need not do this; the Theme is the speaker’s choice, and in any case there is not always a ready-made candidate for thematic status. Compare the following instance: “How am I to get in ?” Alice repeated, aloud. “I shall sit here,” the Footman remarked, “till to-morrow – ” 240

text semantics and clause grammar

The Footman, just like the rest of us, favours himself as an unmarked Theme. So the patterns of wording in the clause, which is the basic unit of lexicogrammatical organization, display a variation that derives from the different kinds of meaning they express; and the structural shape is in each case a natural product of the semantic functions. A functional grammar is an interpretation of the primary semiotic purposes that language has evolved to serve, and of the different ways in which meanings relating to these different purposes tend to be encoded (and the patterns just described are only tendencies). When we go on to observe the developmental processes whereby young children construct their language, we gain a further insight into the steps by which grammar may have evolved on the way towards its present form.

2

From clause to text

Since the functions that we have called ideational, interpersonal and textual are components of the semantic system, and since a text is a semantic unit, it follows that these components will be present in the text just as they are in the lexicogrammatical entities, the wordings, by which the text is realized. In this sense, then, a clause is bound to be like a text: it originates in the same meaning potential. But to say this is to say no more than that both derive from the linguistic system – a point that is perhaps worth making, since text is still sometimes treated as if it had no source of its existence in language, but is nevertheless not saying a very great deal. The problem to be solved is how features from these semantic components are represented, on the one hand in clauses and on the other hand in texts, and with what kind of systematic relationship between the two, such that the clause can function as the principal medium through which meanings of such different kinds, and differing domains, are coded into an expressible form. In this latter half of the paper I will suggest two different facets of the clause-to-text analogy, which correspond to the two axes of the relationship of clause to text that I referred to earlier: their relationship in size, and their relationship in abstraction. To go from text to clause involves a move along the axes both of composition (constituency) and of realization. I shall consider the size dimension first. (A) Do we find, extending over a whole text, patterns that are like those we find in the clause? Let us take each of the three functional components in turn. 1. Ideational. Like a clause, a text has an ideational structure, with 241

word–clause–text

something of the same particulate kind of organization: it is possible to recognize functional constituents of a text, always allowing (as in the clause) for some variation in sequence and a certain amount of overlapping. These structural elements have been identified most clearly, perhaps, in narrative; and the researches of Pike, Longacre and their co-workers on the one hand, and of Gleason and his colleagues on the other, have provided a rich body of empirical findings about the structure of narrative in languages and cultures from every continent. Other genres have been less thoroughly studied. Sinclair and Coulthard (1975), in their study of the structure of classroom discourse, set up a rank scale, a hierarchy of constituents each with its own configuration of functional elements. Ruqaiya Hasan (1979) considers that this structural organization is a general feature of texts of all genres; in her studies of transactional discourse she recognizes optional and obligatory elements, variations in sequence, recursive options and the like, all of which make the text structure look rather similar to the ideational structure that is characteristic of the clause. We can sum this up by saying that, in at least some genres, and perhaps in all, a text is a configuration of functional elements, collectively representing some complex construct of experience and typically realized as discrete, bounded constituents in a partially determinate sequence. Within the ideational component there is a category of conjunctive relations of the types of “and, or, nor, viz, yet, so then”, which can be coded in a great variety of different ways. They appear in many forms within the clause and even within clause constituents; most typically, perhaps, they link clauses in a hypotactic or paratactic clause complex. But they also function as semantic links over longer passages of discourse. Martin (1992) has interpreted these relations in a generalized system network and suggested how they may be accounted for as an aspect of the ideational structure of a text. 2. Interpersonal. A clause has an interpersonal pattern of organization, including a modal structure (mood, modality and key) which expresses its character as a speech event. In the same way a text has a unified character as a rhetorical event. In a recent study, Melrose (1979) makes the suggestion that a text or portion of a text derives its character from the type of process, in the transitivity system, that is predominant in it: material including action, event, behaviour; mental, including perception, reaction, cognition; verbal; or relational, including attributive, equative, existential. This “type of process” is of course an ideational category; but a clustering of processes of the same kind expresses the rhetorical design of the text rather in the same way that a 242

text semantics and clause grammar

particular complex of ideational features such as “I am certain” or “I want to know” functions as the interpersonal motif of a clause. The bridging concept in this case is the field of discourse, which is the aspect of the context of situation of a text by which the transitivity selections in it tend to be mainly determined (Halliday 1977); the “field” is defined as the nature of the social action in which the text is playing some part, and this naturally limits the range of possible parts that are open to it to play (Appendix 4, pp. 255–7). 3. Textual. That discourse displays some kind of a wave form, with peaks of prominence at both ends, has been a commonplace of rhetorical theory ever since it was first hypothesized that a text has a beginning, a middle and an end. (This is perhaps one of the few examples of a verifiable hypothesis in linguistics, though characteristically in order to be verified it has first to be trivialized.) The concept of the paragraph is based on the notion of culminatives, with terms such as initiating, introductory or topic sentence referring to movement downwards from a beginning, and terms such as culminating, summative or focal sentence referring to movement upwards to an end. The diminuendo–crescendo pattern we find in the clause is thus also present in the paragraph, and probably in other text units as well: a text can justifiably be thought of as a construct of waves within waves. And this nesting of wave-like structures one inside another is characteristic also of lexicogrammatical organization: among the constituents of the clause in English, endocentric word groups (verbal groups and nominal groups) display this same kind of movement from speaker prominence to listener prominence. So when a linguist says to his editor I have been going to finish my three brilliant articles for you ever since the beginning of the year, the verbal group have been going to finish goes from the speakerprominent deictic have, locating the process in speech time, to the listener-prominent lexical item finish, saying what the process actually is; and the nominal group my three brilliant articles likewise goes from a deictic my, locating the object in speaker space, to a lexical item articles, again giving the main piece of information. This is essentially the same complex movement as that from Theme to non-Theme and from nonNew to New in the clause. So both text and clause can be seen to participate in a multilayered pattern of organization in which the movement has this same underlying periodicity repeated over structures of differing extent. So much for what we might call the metonymic aspect of the relation of clause to text. Now we turn to the metaphorical: where the feature that we have identified in the clause is not being repeated on a 243

word–clause–text

larger canvas, as in the instances just considered, but rather is standing as the realization of something else that is a feature of the text. However, not only is it realizing a text feature, but also, given the naturalness of fit that we were able to establish between the grammar and the semantics, it has a similarity, in some transformed way, to the feature which it serves to realize. (B)1. Ideational. Cohesion, as defined by Halliday and Hasan (1976), is the semantic resource through which textual coherence is realized. A text displays cohesion; and this cohesion is achieved by means of a variety of features of the clause, which serve to relate one clause to others that constitute its context. However, while cohesion is a necessary condition of textual coherence, it is not by itself sufficient to guarantee it; and in her subsequent studies Ruqaiya Hasan (1984) has been comparing pairs of texts, of similar nature and origin, where one is judged coherent and the other not, in order to establish what are the differences between them. She has one set of texts which are stories told by children; she has also examined texts from schizophrenic patients, including a pair of texts from one particular patient, one when undergoing treatment and the other when the same patient was judged as having been cured. In each case all the texts display typical chains of identity or similarity, ongoing representations of some participant or some other element of the semantic structure – a process, perhaps, or an attribute, or a complex concept of some kind. Now, in the texts judged to be coherent, these lexicoreferential chains were systematically interrelated: a majority of the occurrences in any one chain were related to occurrences in some other chain. They were systematically related, that is to say, in the ideational structure of the clause; for example as Agent to Process, or Attribute to Carrier, or by their both having the same role with reference to some other element, such as both being carriers of the same attribute. In the texts that were judged to be non-coherent, on the other hand, although the proportion of lexicoreferential occurrences entering into cohesive chains was no less than in the coherent texts, only a minority of these occurrences were cross-related in this way; in general, the recurring elements ran alongside each other through the text but without intermingling to any extent. The coherence of the text appears to be the product of this “interchaining”. If a text is coherent there is a movement of related particles through a succession of clauses, so that not only do the individual particles persist from one clause to another, but the structural configurations, though not remaining static, also preserve a recognizable continuity. Just as individual elements form a clause not as isolated 244

text semantics and clause grammar

entities but as roles in a structural configuration, so chains of elements form a text not as isolated chains but as role-chains in an ongoing configurational movement (Appendix 1, pp. 247–50). 2. Interpersonal. How does one recognize the unique rhetorical flavour of a text? Partly at least from the overall pattern of interpersonal features of the individual clauses. A text has its own character as an intersubjective event, and this tenor of discourse is manifested primarily through the cumulative force of the options taken up in the interpersonal systems of meaning. In Priestley’s play An Inspector Calls, the underlying theme, or rather one of the underlying themes, is that of social responsibility: we are all members of one body. This confers obligations on all of us, each one towards others. In the course of the play these obligations are acted out – or rather the consequences of their not being fulfilled are acted out – through the step-by-step uncovering of a chain of irresponsibility, compounded by sheer chance and observed through a confusion of prejudices and doubts. Now, the three conceptual fields of probability, opinion and obligation together comprise the semantic raw material of the complex system of modality in the grammar of English. It is not surprising therefore that the underlying semiotic of the play is worked out metaphorically, at a critical point in the action, through the highlighting of modal selections within the clause, backed up by lexical choices from the same semantic fields. The clauses in this key passage, each with its own small momentum, combine to produce a powerful semantic movement, a motif first of chance and then of duty, both hedged around by opinion, and culminating, after a narrative monologue serving as commentary, in a burst of direct assertion in which the modalities are finally swept away. As audience we respond to this movement even though the events which call it forth are in themselves trivial, no more than an argument over the identification of a photograph. Here the interpersonal features of the clause stand as a metaphor for the social semiotic of the text, as an exploration of the complex symbolic structures binding men to their fellow men (Appendix 5, pp. 257–60). 3. Textual. The last two examples suggest that we can ‘read off’ significant aspects of the semiotic quality of a text from looking at the transitivity and modal features that predominate in the individual clauses. When we come to consider the rhetorical organization of a text, this too can be discovered from a reading of the clause patterns, in this case those having to do with functional sentence perspective: what are the elements that function predominantly as theme, and what are the elements that function predominantly as news. In his study of 245

word–clause–text

the thematic organization of discourse, Fries (1981) has shown how these patterns realize the development of the paragraph. Examining a tightly constructed paragraph by Lytton Strachey, Fries found three lexicosemantic chains, one having to do with the opposition of wisdom and chance, one with the English constitution, and one with political apparatus in general; of the three, the former was overwhelmingly associated with initial position in the clause, the second with final position, while the third showed no particular pattern of distribution. Fries points out that this reflects the rhetorical interpretation of the paragraph as having the “wisdom versus chance” motif as its method of development and the English constitution as its main point. Thus the mode of discourse is manifested in the same cumulative manner by the ongoing selections, in each clause, from the thematic and informational systems, those comprising the “textual” element in the meaning potential of the clause (Appendix 3, pp. 254–5). I am not suggesting, of course, that listeners and readers process text in a conscious manner, parsing each clause as they go along. On the contrary, speaking and understanding are, as Boas and Sapir always insisted, among the most unconscious of all the processes of human culture. The conscious task is that which falls to the linguist, when he tries to find out how text is organized. Listeners and readers make predictions – they have a good idea of what to expect; if they did not make these predictions, with a greater than chance probability of being right, they would not be able to understand each other. It is the organization of a text, and in particular the relation of a text, as a semantic unit, to a clause as the primary lexicogrammatical unit through which it is realized, that makes such prediction possible. The linguistic analysis of text is a necessary step in the interpretation of how meanings are exchanged. A clause, while it realizes directly only a very small unit of text (sometimes referred to as a “message unit”), stands also as the realization of a text as a whole, or some structurally significant portion of it, in the indirect, metaphorical sense that these examples suggest. The former is its automatic function, as determined by the system of the language. The latter is what Mukarˇovsky´ (1977) recognized as “deautomatization”: still, of course, part of the potential of the linguistic system but deployed in a metagrammatical way, conveying meaning by the act of systemic choice instead of (in fact always as well as) by the act of realization. A clause is a text in microcosm, a “universe of discourse” of its own in which the semiotic properties of a text reappear on a miniature scale. This is what enables the clause to function as it 246

text semantics and clause grammar

does. What are clauses for? – to make it possible to create text. A clause does this effectively because it has itself evolved by analogy with the text as a model, and can thus represent the meanings of a text in a rich variety of different ways.

Appendices Appendix 1 from Hasan (1980). Text A 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13.

once upon a time there was a little girl and she went out for a walk and she saw a lovely little teddybear and so she took it home and when she got home she washed it and when she took it to bed with her she cuddled it and she fell straight to sleep and when she got up and ( ) combed it with a little wirebrush the teddybear opened his eyes and ( ) started to speak to her and she had the teddybear for many many weeks and years and so when the teddybear got dirty she used to wash it and every time she brushed it it used to say some new words from a different country and that’s how she used to know how to speak English Scottish and all the rest

Text B 1. 2. 3. 4. 5. 6. 7. 8. 9. 10.

the sailor goes on the ship and he’s coming home with a dog and the dog wants the boy and the girl and ----they don’t know the bear’s in the chair and the bear’s coming to go to sleep in it and ----they find the bear in the chair they wake him up ----and (-----) chuck him out the room and (-----) take it to the zoo the sailor takes his hat off 247

word–clause–text

11. and the dog’s chased the bear out the room 12. and the boy will sit down in their ----- chair what the bear was sleeping in Underlined items are those which enter into lexico-referential chains. Broken underlining indicates that one item incorporates more than one token, for example ----they referring to the girl, the boy, the sailor and the dog. Empty underlining within parenthesis ( ), (-----) indicates a token or tokens presupposed by ellipsis. The number of lexico-referential tokens in the two texts is not very different: 66 in Text A, 56 in Text B. But whereas 43 of those in Text A (65 per cent) occur in chain interaction, the comparable figure for Text B is only 20 (36 per cent). Text A thus displays considerably greater cohesive harmony. When subjects were asked to judge the coherence of the two texts, Text A was consistently rated “more coherent” than Text B.

248

Chain Interaction – Text A home home

왗 왘

girl girl girl girl girl girl girl girl girl girl girl girl

왗 왘

went got

왗 왘

took had

왗 왘

왗 왘

왗

왘

teddybear teddybear

took-to-bed fell-to-sleep got-up washed combed washed brushed lovely dirty

왗 왘

왗 왘

teddybear teddybear teddybear teddybear teddybear teddybear speak teddybear teddybear

왗

왘

say speak

왗

왘

words English Scottish All-the-rest

Chain Interaction – Text B chair chair

bear bear 왗

go come

왗

왘

go-to-sleep sleep

왗

왘

bear bear

chuck chase

왗

왘

bear bear

왘

chair chair

왗

왘

room room

sailor sailor

Each rectangle corresponds to one chain; subdivisions in the chain are indicated by boxing. Each box contains those items which are in a constant functional relation (shown by a double-headed arrow) to items in some box in a different chain; for example in Text A, between girl (4) and the box containing washed . . . brushed there is an actor– action relation; between the latter and teddybear (4), a relation of action–goal.

text semantics and clause grammar

Appendix 2 from James R. Martin (forthcoming) 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18.

B: Lips are a must. They’re in fashion. So . . . what are you using in your skin care? A: Oh I just – I don’t know. Something my mom gets: Ponds or something. B: Yes. A: I don’t know. B: Well really uh that’s not good enough really. You want something that’s going to treat the skin. You need to cleanse your skin well uh to use a good toner A: Hmm. B: and moisturiser is a must and of course then you can go into the make-up. But if you do all these things your skin will start to improve. A: Yeah. B: You’re finding a few little spots under your skin, aren’t you?

Talking Shop: scene 21 Halliday and Poole (1978).

251

word–clause–text

Figure 1 Conjunctive relations in Talking Shop (scene 21)

252

text semantics and clause grammar

Figure 2

253

word–clause–text

Appendix 3 from Peter Fries (1981). 1. A. The English Constitution – that indescribable entity – is a living thing, growing with the growth of men, and assuming ever-varying forms in accordance with the subtle and complex laws of human character. 2. B. It is the child of wisdom and chance. 3. C. The wise men of 1688 moulded it into the shape we know. 4. C. but the chance that George I could not speak English gave it one of its essential peculiarities – the system of a cabinet independent of the Crown and subordinate to the Prime Minister. 5. C The wisdom of Lord Grey saved it from petrification and set it upon the path of democracy. 6. C. Then chance intervened once more. 7. D. A female sovereign happened to marry an able and pertinacious man, 8. D. and it seemed likely that an element which had been quiescent within it for years – the element of irresponsible administrative power – was about to become its predominant characteristic and change completely the direction of its growth. 9. C. But what chance gave chance took away. 10. D. The Consort perished in his prime, 11. D. and the English Constitution, dropping the dead limb with hardly a tremor, continued its mysteric life as if he had never been. Queen Victoria, Lytton Strachey (p. 192) [The above] is a well constructed paragraph which contains within it three lexical systems; the first concerns living, growing, changing, the second system concerns wisdom versus chance and the third system concerns concepts having to do with government. From reading the paragraph it is clear that the main point of the paragraph is that the English constitution is living, growing and changing, that the paragraph is developed via the opposition between wisdom and chance and that the lexical system having to do with government plays no particular role within the structure of the paragraph. On examining the paragraph one finds that the terms having to do with living, growing and 254

text semantics and clause grammar

changing typically occur within the rhemes of the component sentences of the paragraph. The terms having to do with wisdom and chance, with certain exceptions which can be explained, occur within the themes of the component sentences. The terms having to do with the form of government occur more or less equally within the themes and rhemes of the component sentences of the paragraph. Thus the consistent placement of the terms of a lexical system inside or outside the themes of the component sentences of the paragraph affect the perceived role of that lexical system within the paragraph as a whole. [Hence] a) the lexical material placed initially within each sentence of a paragraph (i.e. the themes of each sentence of a paragraph) indicates the point of departure of the message expressed by that sentence, and b) the information contained within the themes of all the sentences of a paragraph creates the method of development of that paragraph. Thus if the themes of most of the sentences of a paragraph refer to one semantic field (say location, parts of some object, wisdom vs chance, etc.) then that semantic field will be perceived as the method of development of the paragraph. If no common semantic element runs through the themes of the sentences of a paragraph, then no simple method of development will be perceived.

Appendix 4 From Robin Melrose (1979). The remaining eleven sections deal with every aspect of life, regulating it at every stage and aspect, ordering everything, forcing everything into a symmetrical pattern: the cities are uniform, married life is strictly controlled, education is minutely prescribed. Philosophy is confined within rigid limits, the fine arts somewhat less so (. . .) This planned paradise is enforced by drastic penal laws. Machinery of government is paternalistic and pyramidal. It is based on division into families, tribes, cities and provinces, and, in the case of the different crafts and professions, on units of ten. To each unit of work is assigned its “master” (. . .) Each paterfamilias over fifty is a senator, each family in turn provides a tribal chief, each town in turn a city chief. Subordinate senates of cities are controlled by the Supreme Senate. At the head of the state is the General. Totalitarianism, Leonard Schapiro (pp. 87–8)

255

word–clause–text

Theme: The remaining eleven sections ; the cities ; married life ; education ; Philosophy ; This planned paradise ; Machinery of government ; It ; To each unit of work ; Each paterfamilias over fifty ; each family ; Subordinate senates of cities ; At the head of state Lexical Cohesion: Group A : regulating ; ordering ; forcing ; uniform ; controlled ; prescribed ; confined ; planned ; enforced ; penal laws ; paternalistic and pyramidal This is a particularly clear example of an attributive message. It begins with a summation, “every aspect of life”, with the general noun “aspect” acting as Head of a nominal group. It is this summation which determines the Theme of the clauses that follow: thematic prominence is assigned precisely to aspects of life, so that there is a relationship of superordinate to hyponym between summation and Theme in the message, reinforced by Theme in the last five clauses, which is in a relationship of hyponym to “machinery of government”, itself an aspect of life. There is no explicit conjunction of interest : the chief conjunctive relationship is an implicit one, of the internal additive type. More worthy of attention is the lexical string of Group A. Just as Theme was determined by the summation, so the lexical string of Group A is determined by the non-finite clauses dependent on the clause of which the summation is an element. Together with the three verbs in these non-finite clauses, nine lexical items of the message proper constitute a string of synonyms, near-synonyms, and collocates – and of these nine, six function as Complement, and so form a kind of pattern in the Rheme. Thus it may be seen that in this attributive message the summation and the clause complex of which it is an element are both closely related to the message proper that follows: the summation is hyperonymously linked to the Theme, and the clause complex (or, more precisely, the non-finite clause) is synonymously connected – with one exception – to the Rheme. Or, to put it another way, it is most often the case that “aspects of life” are encoded in the Subject, while “regulation” is realised in the Complement.

256

text semantics and clause grammar

Characteristics of message types Table adapted from Melrose’s Table 4, p. 50. Factual

Phenomenal

Relational

Type of Process

material (doing, happening)

mental (seeing, feeling, thinking); verbal (saying)

relational (being – attribute, identity)

Characteristic Theme

main participant

cognizant / sayer or phenomenon / discourse

synonym or hyponym of summative element

Typical Conjunction

external temporal

external or internal additive or temporal

internal additive or adversative

Summative Element

general noun, of which message is meronym

general noun, of which message is hyponym

general noun + expansion, of which message is meronym

Appendix 5 from M. A. K. Halliday (1982). Mrs. Birling: I think we’ve just about come to an end of this wretched business. Gerald: I don’t think so. Excuse me. (He goes out. They watch him go in silence. We hear the front door slam.) Sheila: (to Inspector) You know, you never showed him that photograph of her. Inspector: No. It wasn’t necessary. And I thought it better not to. Mrs. Birling: You have a photograph of this girl? Inspector: Yes. I think you’d better look at it. Mrs. Birling: I don’t see any particular reason why I should – Inspector: Probably not. But you’d better look at it. Mrs. Birling: Very well. (He produces the photograph and she looks hard at it.) Inspector: (taking back the photograph) You recognize her? 257

word–clause–text

Mrs. Birling: No. Why should I? Inspector: Of course she might have changed lately, but I can’t believe she could have changed so much. Mrs. Birling: I don’t understand you, Inspector. Inspector: You mean you don’t choose to do, Mrs. Birling. Mrs. Birling: (angrily) I meant what I said. Inspector: You’re not telling me the truth. Mrs. Birling: I beg your pardon! Birling: (angrily, to Inspector) Look here, I’m not going to have this, Inspector. You’ll apologize at once. Inspector: Apologize for what – doing my duty? Birling: No, for being so offensive about it. I’m a public man – Inspector: (massively) Public men, Mr. Birling, have responsibilities as well as privileges. Birling: Possibly. But you weren’t asked to come here to talk to me about my responsibilities. Sheila: Let’s hope not. Though I’m beginning to wonder. Mrs. Birling: Does that mean anything, Sheila? Sheila: It means that we’ve no excuse now for putting on airs and that if we’ve any sense we won’t try. Father threw this girl out because she asked for decent wages. I went and pushed her further out, right into the street, just because I was angry and she was pretty. Gerald set her up as his mistress and then dropped her when it suited him. And now you are pretending you don’t recognize her from that photograph. I admit I don’t know why you should, but I know jolly well you did in fact recognize her, from the way you looked. And if you’re not telling the truth, why should the Inspector apologize? And can’t you see, both of you, you’re making it worse? (She turns away. We hear the front door slam again.) An Inspector Calls, J. B. Priestley (Act 2) In the text, obligation is tied to judgements of probability: there are opinions relating to duties, and, as a minor motif, duties relating to opinions. The two themes are closely interwoven. We have already seen that this is a projection into the text of a relation that exists between them in the system. The scales of “possible–certain” and “allowed–required” both typically combine with a common semantic 258

text semantics and clause grammar

feature, that of “subjective”, in the sense of representing the speaker’s judgment; and this is symbolized by the use of modal verbs as one form of the realization of both. The significance of the lexicogrammatical selections in the text can only be fully revealed by a consideration of their value in the semantic system. Textually, the passage under discussion centres around the scrutiny and recognition of a photograph. The words and structures which, in their automatic function as the “output” of semantic choices, carry forward the movement of the text, also become de-automatized and so take on a life of their own as engenderers of meaning. Example of modalized clause complex Inspector: Of course she might have changed lately, but I can’t believe she could have changed so much. Clause 1

polarity positive modality low / (indicative : probability) / (subjective : congruent)

Clause 2

polarity negative : transferred modality high / (a) can (imperative : inclination) / (subjective : congruent) (b) I . . . not believe (indicative / probability) / (subjective : explicit) (c) could (indicative / probability) / (subjective : congruent)

259

indicative (MODALITY) ‘it is’ probability temporality ‘it either is ‘it both is or isn’t’ and isn’t’

modal

categorical

positive high median low

imperative (MODULATION) ‘do!’ obligation inclination ‘you do!’ ‘me do!’

is

do

certain

must be

always

required

must do

desperate

probable

will be

usually

supposed

will do

keen

possible

can be

sometimes

allowed

can do

able

categorical 䉲 negative high: must ought to need has/had to median: will would shall should low: can could may might

isn’t

don’t subjective, congruent: must &c. (modal auxiliary verbs) subjective, explicit: I think &c. (mental process clauses) objective: certain(ly) &c. (modal adverbs, etc.)

Chapter Ten

DIMENSIONS OF DISCOURSE ANALYSIS: GRAMMAR (1985)

This chapter presents a brief sketch of a lexicogrammatical text interpreter for Modern English, in terms of systemic-functional grammar. The grammar is in general neutral between spoken and written English, but the text used for illustration is taken from spoken language; it is a discussion among an adult and three nine-year-old schoolgirls. Here is the text in standard orthography and punctuation (Hasan 1965: 65): Adult: Do you – when you have a small baby in the house, do you call it ‘it’, or do you call it ‘she’ or ‘he’? Elsie: Well if it’s just – if you don’t know what it is I think you ought to call it ‘it’, because you don’t know whether you’re calling it a boy or a girl, and if it gets on and if you start calling it ‘she’ then you find out that it’s a boy you can’t stop yourself cause you’ve got so used to calling it ‘she’. Lacey: Em – Mrs. Siddons says that if – if some neighbour has a new baby next door and you don’t know whether it’s a he or a she, if you refer to it as ‘it’ well then the neighbour will be very offended. Tilly: Well if it’s in your family I think you should call it either ‘he’ or ‘she’ or else the poor thing when it grows up won’t know what it is. Adult: Well what did Mrs. Siddons suggest you should do if . . . your neighbour has a baby and you don’t know whether it’s a boy or a girl? First published in The Handbook of Discourse Analysis, Vol. 2: Dimensions of Discourse, 1985. London: Academic Press, pp. 29–56.

261

word–clause–text

Tilly: Elsie: Lacey: Elsie:

She didn’t. I don’t suppose she knew. Call it ‘the’. Hello, The! Oh, I know. Call it ‘baby’.

Systemic grammar is an analysis–synthesis grammar based on the paradigmatic notion of choice. It is built on the work of Saussure, Malinowski and Firth, Hjelmslev, the Prague school, and the American anthropological linguists Boas, Sapir, and Whorf; the main inspiration being J. R. Firth. It is a tristratal construct of semantics (meaning), lexicogrammar (wording), and phonology (sound). The organizing concept at each stratum is the paradigmatic system: A system is a set of options with an entry condition, such that exactly one option must be chosen if the entry condition is satisfied. Options are realized as syntagmatic constructs or structures; a structure is a configuration of functional elements – functions or function bundles. The functions are motivated (nonarbitrary) with respect to the options they realize; the grammar as a whole is motivated with respect to the semantics. The only line of (relative) arbitrariness is that between content and expression (between the lexicogrammar and the phonology). A text in systemic-functional grammar is an instantiation of the system (in the Hjelmslevian sense of “the linguistic system”). (Note that “instantiation” is not the same thing as “realization”; the two concepts seem to be confused in Saussure.) Text may be studied as process or as product; in either case, interpreting a text means showing how it derives from the system and therefore why it means what it does. It is not possible here to present the networks of systems from which the text is derived, since that would involve representing large portions of the grammar. Instead we employ structural notations, with brief discussion of some of the options from which the structural functions are derived. The analysis is given in 10 steps, with a short commentary on each. The 10 steps are as follows: 1. transcription and analysis of intonation and rhythm 2. analysis into clauses and clause complexes, showing interdependencies and logical-semantic relations 3. analysis of clauses, and clause complexes, for thematic (Theme– Rheme) structure 4. comparison of clauses and information units, and analysis of the latter for information (Given–New) structure 5. analysis of finite clauses for mood, showing Subject and Finite 262

dimensions of discourse analysis

6. analysis of all clauses for transitivity, showing process type and participant and circumstantial functions 7. analysis of groups and phrases (verbal group, nominal group, adverbial group, prepositional phrase) 8. analysis of grammatical and lexical cohesion 9. identification, rewording and reanalysis of grammatical metaphors 10. description of context of situation, and correlation with features of the text

1

Transcription and analysis of intonation and rhythm

The text is transcribed orthographically with notation for intonation and rhythm:

Figure 1

263

word–clause–text

Figure 2 Systems and Notation for Intonation and Rhythm

2

Clauses and clause complexes

The text is analysed into clause complexes, showing the interdependencies and logical-semantic relations among their constituent (ranking, nonembedded) clauses. Clause complex 1 1.1 ⳯b 1.2 1 1.3 +2

⳯b ^ a ( 1 ^ +2 ) ||| when you have a small baby in the house || || do you call it || || or do you call it she or he ||

Figure 3a

264

dimensions of discourse analysis a ( ⳯b ( a ^ ‘b ) ^ a) ^ ⳯b ( 1 (a ^ ‘b ) ^ +2 ( ⳯b ( 1 ^ +2 ( 1 ^ ⳯ 2 (a ^ ‘b ) ) ) ^ a ( a ^ ⳯b ) ) ) ||| well if you don’t know || 2.1 a ⳯b a || what it is || 2.2 a b ‘b || I think you ought to call it it || 2.3 a a || because you don’t know || 2.4 ⳯b 1 a || whether you’re calling it a boy or a girl || 2.5 b 1 ‘b || and if it gets on || 2.6 b +2 ⳯b 1 || and if you start calling it she || 2.7 b 2 b +2 1 2.8 b 2 b 2 ⳯2 a || then you find out || 2.9 b 2 b 2 2 ‘b || that it’s a boy || || you can’t stop yourself || 2.10 b 2 a a || ’cause you’ve got so used to calling it she ||| 2.11 b 2 a ⳯b Clause complex 3 a ^ ‘‘b ( ⳯b ( 1 ^ + ( a ^ ‘b ) ) ^ a ( ⳯b ^ a ) ) ||| Mrs. Siddons says || 3.1 a || that if some neighbour has a new baby next door || 3.2 ‘‘b ⳯b 1 || || and you don’t know || 3.3 b b +2 a || whether it’s a he or a she || 3.4 b b 2 1b || if you refer to it as it || 3.5 b a ⳯b || well then the neighbour will be very offended ||| 3.6 b a a Clause complex 4 1 ( ⳯b ^ a ) ^ ⳯2 ( a ( a 具具 ⳯b ) 典典 ^ 1b ) ||| well if it’s in your family || 4.1 1 ⳯b || I think you should call it either he or she || 4.2 1 a || or else the poor thing 具具 典典 won’t know || 4.3 ⳯ 2 a a 具具 when it grows up 典典 4.4 2 ⳯b || what it is ||| 4.5 2 a ‘b Clause complex 5 ‘‘b a 具具 ) a ( 典典 ^ ⳯b (1 ^ +2 ( a ^ 1b ) ) ) ||| well what 具具 典典 you should do || 5.1 ‘‘b a 具具 did Mrs. Siddons suggest 典典 5.2 a || if your neighbour has a baby || 5.3 b ⳯b 1 || and you don’t know || 5.4 b b +2 a || whether it’s a boy or a girl ||| 5.5 b b 2 ‘b Clause complex 6 6.1 ||| she didn’t ||| Clause complex 7 7.1 ||| I don’t suppose she knew ||| Clause complex 8 8.1 ||| call it the ||| Clause complex 9 9.1 ||| hello the ||| Clause complex 10 10.1 ||| oh I know ||| Clause complex 11 11.1 ||| call it baby ||| Clause complex 2

Figure 3b Clauses and Clause Complexes

265

word–clause–text

The major portion of the text consists of five turns, each made up grammatically of one clause complex. These contain, respectively, 3, 11, 6, 5, and 5 clauses. They show a preference for hypotactic (17) over paratactic (7) interdependencies; the predominant logical-semantic relation is that of enhancement, typically hypothetical (11 instances) – the discussion centers around what to do if and when a certain situation arises. Of the 15 other instances, 5 are extension, “and / or”; the remainder (8) are projection, of which 2 are saying (the teacher as Sayer) and 6 are knowing, mainly negative and with a generalized Senser “(if) you don’t know” – this being an aspect of the problem under discussion. (The two instances of I think [new addition] are metaphorical modalities, not projections, as can be shown by adding a tag: the tagged form of I think you should call it he or she is shouldn’t you? not don’t I?) All these clause complexes are not only complex but also impeccably well formed, as is typical of casual spontaneous speech (including that of children!). So much for the “reasoning” component of the discussion. The remainder consists of “suggesting,” partly humorous and partly serious, and here the turns are short, one or two clauses each. The clause complexes are even shorter, since each consists of just one clause. There is no parataxis or hypotaxis. (6.1 and 7.1, She didn’t. I don’t suppose she knew [new addition] could be considered to form a paratactic elaboration, given the tone concord; but the latter, though necessary, is probably not a sufficient condition.) As far as the organization of the information is concerned, a comparison of the two transcriptions shows that in the majority of instances one clause is one information unit, this being the unmarked (default) situation in English. This holds throughout, with the following systematic exceptions: (1) seven out of the eight projections have both projecting and projected clause on one tone group, for example, // if you / don’t know / what it / is // – this is the locally unmarked form; and (2) one clause consists of two information units, one for the Theme and one for the Rheme: // if some / neighbour has a // new / baby next / door // – this is the predominant pattern when two information units are mapped on to one clause. For the analysis of theme and information structure see the next two sections.

266

dimensions of discourse analysis

3

Theme

Each clause, and each clause complex characterized by rising dependency (b ^ a), is analysed for thematic structure: Clause Theme Clause

Textual

Interpersonal

Figure 4 Clause Theme

267

Topical

Markedness of topical theme

word–clause–text

The Theme is the (speaker’s) point of departure for the clause. It is realized in English by position in the sequence: thematic elements are put first. Hence the thematic structure of the clause is Theme ^ Rheme. Each of the three metafunctional components of the content plane – ideational, interpersonal, textual – may contribute thematic material. The textual Theme is some combination of continuative (for example oh, well), conjunctive (for example then, if) or relative (for example that, which). The interpersonal Theme is modality (for example perhaps), interrogative mood marker (WH-element or Finite verbal auxiliary), or Vocative. The topical Theme is any element functioning in the transitivity structure of the clause. The typical sequence is textual ^ interpersonal ^ topical, and the Theme in any case ends with the topical element: in other words, the Theme of a clause extends up to the first element that has a function in transitivity. The unmarked Theme for any clause is determined by the choice of mood: Subject in declarative, WH- or Finite element in interrogative, [zero] in imperative and minor clauses. Semantically, the unmarked Theme is the natural starting point for the particular speech function: in a question, “this is what I want to know” – the information-seeking (WH-) or polarity-carrying (Finite) element; in a statement, “this is the entity on which the argument rests” (Subject). The ongoing choice of clause Themes reveals the method of development of the text. In the example, every clause has an unmarked topical Theme. At first, the impersonal you predominates, followed later by specific third person participants: the teacher (Mrs. Siddons), the neighbour, and the baby. Many are preceded by textual Themes, continuative and / or conjunctive. Thus the text develops as a discussion of a general topic with particular personalities brought in to carry it forward, the whole being linked together both dialogically and logically. At the higher rank of the clause complex, on the other hand, the logical structure of the argument becomes the dominant motif: here there are a number of marked Themes, in the form of hypotactic (dependent) clauses introduced by if. The picture is that of a shared discourse being developed as a logical generalization with hypothetical cases, without much concern for attitudinal rhetoric (the only interpersonal Themes are questions and uncertainties), and with some concretization towards the end.

268

dimensions of discourse analysis

4

Information structure

Each information unit is analysed for information structure (numbers refer to clause complexes, letters to information units):

Figure 5

269

word–clause–text

In the “New” column, the focal element is outside the brackets. Square brackets enclose other New material; within this, the items shown within curved brackets are those previously mentioned. The New is what the speaker presents as being for the listener to attend to: “this is what’s news”. It may be previously unknown, or contrary to expectation, or picked out for special prominence. It is realized by means of the tonic accent. Phonologically, spoken discourse in English consists of a linear succession of tone groups, each characterized by one intonation contour or tone. The tone group, in turn, consists of a tonic segment that carries the characteristic tone contour: 1, falling; 2, rising; 3, level (phonetically realized as low rising); 4, falling–rising; 5, rising–falling; 13, falling plus level; 53, rising–falling plus level. The tonic segment begins with the tonic accent, which embodies the distinctive pitch movement. Optionally, the tonic segment may be preceded by a pretonic segment that also forms part of the same tone contour. Both tonic and pretonic segments display a range of more delicate options within each tone: wide fall, narrow fall, low pretonic, high pretonic, and so on. Grammatically, the tone group realizes a unit of information, which is one piece of news, so to speak. It consists of a New component optionally accompanied by a component that is Given. Typically, the New comes at the end; but unlike thematic structure, information structure is not realized by the order in which things are arranged, but by tonic prominence – the New is the element containing the tonic accent. The particular word on which the tonic accent falls is said to carry the information focus. Anything after the focal element is thereby marked as Given, while anything preceding it may be Given or may be New (there are rather subtle intonational and rhythmic variations serving as signals). An information unit is not necessarily coextensive with a clause, but that is its unmarked status: each ranking clause (i.e. independent or dependent, but not embedded) is typically one information unit. The principal systematic variants are (1) two clauses / one information unit: failing dependencies, that is, a ^ b sequences; (2) one clause / two information units: thematic focus, that is, // Theme // Rheme // information patterns. Analysis of the New elements will reveal the “main point” of the text. In the example it is to do with babies, what sex they are, and how they are to be referred to in cases of doubt. Subsidiary points of 270

dimensions of discourse analysis

attention are the baby’s growing up, the children’s understanding and obligations, and the adult world’s possible displeasure. The Theme in a clause is what is prominent for the speaker; it is “what I am on about”. The New in an information unit is what is made prominent (by the speaker) to the listener; it is “what you are being invited to attend to”. When clause and information unit are mapped on to each other, the result is a wavelike movement from speaker to listener, the diminuendo of the speaker’s part being as it were picked up by the crescendo of the listener’s part. The effect of this movement is cumulative over the text as a whole. The present text is typical in the way that the sequence of Themes represents the “method of development” of the dialogue, while the sequence of News represents the “main point” of the discussion, with each speaker contributing her part to the construction of the overall pattern – all unconsciously, of course. This interplay of two distinct waves of prominence is possible because Theme–Rheme and Given–New are not (as often conceived) one single structure, but two distinct structures that interpenetrate. As a result, they can vary independently, allowing for all possible combinations of the two kinds of texture. In unit 2h, for example, Elsie might have chosen a different distribution by combining thematic and information prominence (mapping New onto Theme): ’cause you’ve got //.1. so / used to / calling it / she // This would have strongly highlighted used to, as a marked focus, and marked calling it she as Given; the effect of the latter would have been to bring out the repetitive facet of calling it she, thus reinforcing its cohesion with 2e, but by the same token to deprive it of its status as a main point for attention. The interaction of the thematic and informational systems is the clause grammar’s contribution to the creation of texture in discourse.

5

Mood

Each finite clause is analysed for mood; its Subject and Finite element are shown, together with any modality:

271

word–clause–text

Figure 6

The system of mood expresses the speech function of the clause. Typical patterns of realization are as shown in Figure 7 (where
means ‘is typically realized as’). 272

dimensions of discourse analysis













Figure 7

In addition there are the minor speech functions of exclamation, greeting, and call, realized by minor clauses, and by exclamatory declaratives. Offers, commands, and statements may be tagged, for example, I’ll help you, shall I? The tag makes explicit the speaker’s request for the listener to perform his complementary role: accept offer, carry out command, acknowledge and confirm statement. There are no tagged clauses in the text under consideration. The Subject expresses the participant in respect of which the particular speech function is validated: performer in the case of goods-&-services, bearer of the argument (i.e. the one on whom the validity is made to rest) in the case of information. In a declarative, the Subject is typically also the Theme (hence “unmarked Theme”; see the discussion of Theme above); but whereas the Theme is a discourse (textual) function, displaying patterns over the text as a whole, the Subject is an interpersonal function having significance just for the particular exchange. Here it is frequently the impersonal you, showing that these statements are to be arguable as statements that are valid in general; in other cases, it is the baby, Mrs. Siddons, or the hypothetical neighbour. The Finite element expresses the deicticity of the process, by reference to either (1) speaker-now (primary tense: past, present, or future) or (2) speaker judgement (modality: probability, usuality, obligation, inclination, or ability; high, median, or low value). Almost all the finiteness in this text is combined with present tense; the children’s text proceeds as a series of declaratives, some independent and some dependent, the mood-bearing constituent of which consists of generalized Subject plus Finite present (and there is very little secondary tense). 273

word–clause–text

This is typical of logical argument; and it is interspersed with interrogatives as the adult prompts the children to explore further. There is very little modality in the text. In the Finite element, apart from one “ability” form can’t, in you can’t stop yourself, there are just three expressions of obligation (ought and two instances of should). Other than in combination with finiteness, there are again only three modalities, in this case expression of probability: (I suppose and two instances of I think). As it happens, in two paired instances the two kinds of modality are associated: where the speaker expresses a judgement of obligation, she qualifies it with a judgement of probability, “it may be that it should be so”. Thus when the children are making rules, they are also being tentative about them.

6

Transitivity and process types

Each clause is analysed for transitivity, showing process type, Process, Medium, other participant functions, and circumstantial elements:

274

dimensions of discourse analysis

Figure 8

Transitivity is the representation, in the clause, of the experiential component of meaning: specifically the processes, the participants in them, and the attendant circumstances. 275

word–clause–text

This text is a discussion of a problem: what to do if a certain situation arises. The majority of the clauses in it relate to the situation, the problem, its solution, and the process of problem-solving: (i) (ii) (iii) (iv) (v)

possessive / attributive: intensive / attributive: intensive / identifying: cognitive: verbal:

have + baby (⳯ 3) be + male / female (⳯ 5) call+ baby+ he / she (⳯ 11) know, find out (⳯ 8) say, suggest (⳯ 3)

The remaining six clauses include three characterizing the baby (two material: grow, get on; one circumstantial: be in + family), one characterizing the neighbour (affective: be offended), and two others, one minor (greeting: hello) and one a WH- process (do what). There are three major types of process in English: Type I, doing (material and behavioural); Type II, sensing / saying (mental and verbal); Type III, being (relational and existential). They are distinguished in the grammar in various ways; the principal distinctions are as follows:

Figure 9

276

dimensions of discourse analysis

In the example text, the clauses in (i–iii) above are all type III, relational; the issue is one of being, partly attribution (having a baby, which in this text means possessing it, not bearing it; being male or female) and partly identification (being the name of). Of these, (i) and (ii) are middle – there is a Medium (the neighbour, the baby) but no Agent; (iii) however is effective – there is an Agent in the identification process, always represented as you, but moving from personal “you” in clause complex 1 to impersonal “you” thereafter. The clauses in (iv) and (v) are type II, mental-verbal; (iv) are cognitive, with the Medium (Senser) being you, I, Mrs. Siddons or the baby when it grows up; (v) are verbal, the Medium (Sayer) being Mrs. Siddons. All interactants, real and hypothetical, are involved in thinking about the problem – including the baby, at some future date, if a solution is not reached now; and the teacher has put the problem into words. A summary of process types and the relevant participant functions follows:

Figure 10

277

word–clause–text

7

Other group features

Groups and phrases are analysed with respect to features that are relevant to the inquiry:

Figure 11

The English verbal group carries a recursive three-term (past / present / future) tense system of the type “present in past in . . .” where any tense selection may become the point of reference for another one, subject to certain restraints that limit the total number of 278

dimensions of discourse analysis

possible combinations. In the “full” (finite) system of tenses the number is 36; from this is derived, by a neutralizing of certain tenses in the “past” series (he said she had arrived corresponds to she arrived, she has arrived, and she had arrived), the 24-tense “sequent” system; and from this in turn, by a parallel neutralization in the “future” series (to be about to depart corresponds to will depart, is going to depart, and will be going to depart), is derived the “non-finite” system which has just 12. This last is also the system that applies if the Finite verbal element is a modality (e.g., should, must), since that eliminates the primary tense choice. Most of the verbal groups in this text are simple present tense; not only because of the general nature of many of the propositions, but because most of the processes are other than “doing” ones, and therefore have simple present, not present in present, as their unmarked choice. Furthermore, even the material ones are dependent (if it gets on, when it grows up), which again requires simple present. Nominal Group. The only nominal groups with structure other than simply Head / Thing are the following:

Figure 12

Just as the verbal group further specifies the process, in respect of tense, polarity, and so on, the nominal group further specifies the entity represented by the Head noun. There are similarities between the two types of word group; but the nominal group has much more lexical material, since entities have a more developed taxonomic organization. Hence the nominal group has a functional structure Deictic– Numerative–Epithet–Classifier–Thing–Qualifier, with left–right ordering 279

word–clause–text

from the most instantial, situation-bound to the most permanent characteristic, modified by a purely syntactic principle which puts anything that is embedded at the end. There is little elaboration of nominal groups in the text. Most of them are simply personal pronouns, functioning cohesively; those that have noun as Head contain just enough specification to establish the general point being made, for example, a small baby, your neighbour, the house. The only Qualifier is the nonfinite clause to calling it she following the Attribute used; and this is a metaphor for a modality “have so usually been calling it she”.

Figure 13

8

Grammatical and lexical cohesion

The text is analysed for grammatical and lexical cohesion:

280

dimensions of discourse analysis

Figure 14

The headings given in the key are the basic lexicogrammatical resources for creating texture between clause complexes. In fact they function also within the clause complex – they are simply indifferent to grammatical boundaries; but they have greater force when linking one clause complex with another because of the absence of structural links. In this example, therefore, only inter-complex instances have been noted. What all types of cohesion have in common is that every instance presumes some other element in the text for its interpretation; and 281

word–clause–text

hence a tie is set up between it and what it presumes. In reference, what is presumed is some semantic representation: of a participant, for example, as when it refers back to “a small baby”, but also of a semantic construct of any extent. In substitution and ellipsis, on the other hand, what is presumed is a lexicogrammatical representation, some piece of wording that has to be retrieved, as when she didn’t requires the restitution of suggest (anything); this is a different kind of textual retrieval and rarely extends beyond one clause complex. Conjunction refers to the nonstructural representation of logical-semantic relations that may also be expressed structurally; for example, on the other hand, semantically related to paratactic but and hypotactic although. Lexical cohesion is created by the repetition of a lexical item (e.g., call . . . call); the use of a synonym (e.g., call . . . refer to); the use of a high-frequency collocate (e.g., house . . . family); or the use of a hyponym or superordinate – an item within the same lexical set but differing in generality (e.g., baby . . . [poor] thing, baby . . . boy, girl). The sample text is characterized by dense lexical repetition and personal reference; the discourse unfolds around a small number of entities that are constantly being referred to. When a new instance is brought in, the link is achieved by collocation: in the house . . . next door. There is very little conjunction, because the logical-semantic relations are realized by the hypotactic organization of the clause complex: The reasoning is systematic and explicit. There is also little ellipsis, which comes in only when the reasoning gives way to a more dialogic pattern with shorter exchanges.

9

Grammatical metaphor

Grammatical metaphors are identified, reworded, and reanalysed. Most examples of adult English contain some instances of grammatical metaphor: clauses in which one type of process is represented in the grammar of another; for example, the fifth day saw them at the summit “on the fifth day they arrived at the summit”, or guarantee limited to refund of purchase price of goods “we guarantee only to refund the price for which the goods were purchased”. Children’s speech is largely free of grammatical metaphors of this kind; this is in fact the main distinction between child and adult language. There are no examples of it in the present text. There are however certain grammatical metaphors that have been built into the language, so that the metaphorical version has become the norm; for example, she gave a nod “she nodded”, he has a long nose 282

dimensions of discourse analysis

“his nose is long”. One very common type of these is the use of a mental process (cognitive) clause to express a modality, such as I think, I don’t believe. It was pointed out above that the tagged form of I think it’s broken is I think it’s broken, isn’t it?, not I think it’s broken, don’t I?, showing that this is a metaphor for it’s probably broken. We can use one of these as an example: || I don’t suppose mental: cognitive negative

|| she came ||| a ˆ ´ b material positive

reworded as: ||| she probably didn’t come material modalized:probability/median negative

|||

(single clause)

In cases like this it saves time if the analysis moves directly to the nonmetaphorical version, since the rewording is quite automatic. In other instances, however, the principle is as follows: 1. Analyse the clause as it stands. 2. Reword it, in nonmetaphorical form. 3. Analyse the reworded version. Both analyses figure in the interpretation. Sometimes it takes several steps in rewording to reach a nonmetaphorical version, and there may be more than one possible route; all are potentially relevant to an understanding of the text.

10

Context of situation

The context of situation of the text is described in terms of field, tenor, and mode. The “field” is what is going on: the nature of the socialsemiotic activity. The “tenor” is who are taking part: the statuses and mutual roles of the interactants. The “mode” is what part the language is playing: the rhetorical and communicative channels. This description is then used to interpret the lexicogrammatical features of the text. Field.

A general, imaginary problem of verbal behaviour: how to refer to a baby whose sex is unknown, without offending against the parents, the baby (later in life), or the language. 283

word–clause–text

Tenor. Adult and three children: adult (neither teacher nor parent) interviewer, informal; children self-conscious but relaxed. Speech roles: adult questioning, children suggesting. Mode. Informal spontaneous speech. Dialogue: question-andanswer exchanges. Exploratory; hypothetical and logical in orientation, moving towards (partly humorous) resolution. These features determine the choice of register: that is, the kinds of meanings that are likely to be exchanged. Like the rest of the linguistic system, the patterns are probabilistic: given these features of the context of situation, we can make semantic (and therefore lexicogrammatical) predictions with a significant probability of being right – that, after all, is precisely what the interactants themselves are doing all the time. What makes this possible is what makes it possible for a child to

Figure 15

284

dimensions of discourse analysis

learn the language in the first place: the systematic relationship between these categories of the situation and the metafunctions of the content system. By and large, characteristics of the field predict experiential meanings, those of the tenor predict interpersonal meanings, and those of the mode predict textual meanings. In analysing a text, we identify those features of the lexicogrammar which in a text-generation program might reasonably be expected to be called upon if the situation was represented in this way.

11

Conclusion

Three final points should be made about an outline of this type. One is that it is just an outline: Obviously, the analysis under every heading could be developed much further in delicacy, and other headings could be added. The guiding principle is to select and develop whatever is needed for the particular purpose in hand. There are many different purposes for analysing a text, and the scope and direction of the analysis will vary accordingly. Often we may want to scrutinize only one or two features, but to follow them through to a considerable depth. Secondly, a text analysis is a work of interpretation. There are relatively few absolute and clearcut categories in language; there are many tendencies, continuities, and overlaps. Many actual instances can be analysed in two or more different ways, none of which can be ruled out as impossible; some may be less sensible than others, and so can be discarded, but we may still be left with valid alternatives. Especially in a literary text it is to be expected that we will find clauses with multiple grammars; but this is not a distortion of the system – it is a richer use of its natural resources. All analyses may need to figure in the interpretation. Thirdly, the lexicogrammatical analysis is only a part of the task. It is an essential part; all text is made of language, and the central processing unit of the linguistic system is the lexicogrammar. But just as the grammatical system does not itself create text – text is a semantic creation, with the grammar functioning largely (though not entirely) as the automatic realization of the semantic choices – so the analysis of the grammar does not constitute the interpretation of a text. (There has been some misunderstanding on this point, for example in the use of cohesion as a method of text analysis. Cohesion is an essential property of texts, but it is the way the cohesive resources are deployed that makes the difference between text and non-text, and between one text and another.) 285

word–clause–text

Editor’s note: Examples of text analysis based on the grammatical principles outlined in this chapter are presented in Volume 2, Linguistic Studies of Text and Discourse. See Hasan (1985/1989) and Martin (1992) for theoretical discussion and illustration of the place of grammar in the analysis of discourse.

286

SECTION THREE CONSTRUING AND ENACTING

This page intentionally left blank

EDITOR’S INTRODUCTION

Linguistics like other sciences requires a metalanguage for representing its object of study, which in the case of linguistics is language itself. However, as Halliday points out in ‘On the ineffability of grammatical categories’ (Chapter 11), unlike other sciences, linguistics is “language turned back on itself”, to use Firth’s expression. The problem lies, Halliday argues, in the nature of language as object. Because language is an evolved system, not a designed system, it rests on principles that are ineffable. Its very richness, “its power of distilling the entire collective experience of the culture into a single manageable, and learnable, code . . . puts its categories beyond the reach of our conscious attempts at exegesis”. This richness is most apparent in unconscious, spontaneous, un-self-monitored language, or as Halliday notes, “our ability to use language depends critically on our not being conscious of doing so”. In ‘Spoken and written modes of meaning’ (Chapter 12), originally published as a chapter in Comprehending Oral and Written Language (1987), Halliday elaborates further on the differences between unconscious and spontaneous spoken discourse, and its more conscious and self-monitored counterpart, written language. Arguing against assumptions that written language is syntactically more complex or more richly endowed than spoken, Halliday maintains that each is highly organized and complex in its own way: “Written language tends to be lexically dense, but grammatically simple; spoken language tends to be grammatically intricate, but lexically sparse”. Halliday describes written language as “crystalline” and spoken language as “choreographic”. Comparing the two, Halliday notes how speaking and writing impart their own character onto the world they depict. Written language objectifies. “A written text is an object, so what is represented in writing tends to be given the form of an object. But when one talks, 289

construing and enacting

one is doing; so when one talks about something, one tends to say that it happened or was done.” By means of grammatical metaphor, written language symbolically distances “the act of meaning and its counterpart in the real world”. Each plays complementary roles when it comes to using language to acquire knowledge and reflect on one’s experience. Halliday maintains that this complementarity must figure into any attempt at developing a linguistic theory of learning. In ‘How do you mean?’ (Chapter 13), appearing in Advances in Systemic Linguistics: Recent Theory and Practice (1992), meaning is taken as a mode of action occurring at the intersection of the conscious and material modes of experience. In particular, Halliday examines the evolution of protolanguage into language, or how the two-dimensional semiotic that defines the mammalian experience evolves into ‘a semiotic of a new kind: a stratified, tri-stratal system in which meaning is “twicecooked”, thus incorporating a stratum of “pure” content form. The two dimensions of protolanguage, a minimal semiotic system, include “the ‘inner’ dimension of reflective / active, ‘I think’ as against ‘I want’, and the ‘outer’ dimension of intersubjective / objective, ‘you and me’ as against ‘he, she, it’ ”. The third dimension results from the introduction of grammar, or as Halliday describes it: ‘a purely symbolic mode of being between these two interfaces’. The processes of instantiation and realization make possible this dynamic open system we call language. In the two final works in this section, ‘Grammar and daily life’, which first appeared in Functional Approaches to Language, Culture and Cognition (2000), a Festschrift for Sydney Lamb, and ‘Grammar and grammatics’, published as Volume 121 of the series Current Issues in Linguistic Theory (1996), Halliday describes how grammar enables us, unconsciously, to construe our reality, and interpret our experience, while grammatics makes it possible for us to reflect consciously on how this theory of our human experience works. Halliday introduced the term “grammatics”, in contradistinction to grammar, to distinguish between a particular stratum of a natural language and the study of this stratum. “Thinking grammatically”, or “using grammatics to think about what grammar thinks about the world”, may help us better understand this ‘grammatical energy’ or ‘grammatical logic’ that powers language and also conditions our attitudes to each other and to the world around us. “To be a linguist”, Halliday writes, “is inevitably to be concerned with the human condition”, and those who ‘think grammatically’ will be better prepared not only to address issues of social injustice and inequality, but also to contribute to the development of new applications of linguistics such as intelligent computing. 290

Chapter Eleven

ON THE INEFFABILITY OF GRAMMATICAL CATEGORIES (1984)

1

The problem of the ineffable

We live in an age of growth, in which every day more and more things come into our lives; and things, and all their parts, need names. So more and more words come in with them – new words, or new ways of exploiting, embellishing and combining the old ones; and in this way the balance is maintained. There is no sign that our onomastic resources are drying up; indeed we are likely to run out of natural resources long before we run out of names for the things we make out of them. But there are signs that the things are becoming more resistant to being named. There is no natural way of referring to the various small plastic objects that lie around the house, or the toys we give our children, or the furniture we now have to assemble for ourselves. They no longer fit our taxonomies. We live in modules, sit on units and entertain ourselves with systems. Behind these nameless objects is a technology and a science that produce them; and there, less visible to us, is another realm of things that have to be named. Many of them are abstract things, the categories and concepts of a theory; and some of these also prove recalcitrant to ordinary onomastic processes – they only come to be ‘named’ by some mathematical formula, like a function of the co-ordinates of x and y, the integer over psi1 and psi2, and so on. But somehow they have to be enmeshed in language; otherwise they are not brought under control. First published in The Tenth LACUS Forum, 1984, edited by Alan Manning, Pierre Martin and Kim McCalla, pp. 3–18. Amsterdam: John Benjamins.

291

construing and enacting

In this respect linguistic categories are no different from those of other theories; they too need names. It is true that in one instance a theory evolved without them: Chinese phonology, in its first few centuries, had neither terms nor definitions – and still managed to give an account of the syllable that was explicit enough to enable later scholars to reconstruct the system. But western linguists were always more namebound; they either created a terminology, or borrowed it from some other field. The earliest known Arabic grammarian, AlKhalil ibn Ahmad, took various terms over from architecture; his great pupil Sibawayhi, who had been trained as a lawyer and so placed a high value on names and definitions, replaced them with a more systematic terminology derived from legal models (see Carter 1981). In ancient Greek linguistics, by contrast, the technical terms evolved out of everyday language. The process was a gradual one, extending over three or four centuries; and in the course of this time the original terms had moved some distance from their non-technical meanings, evolving as the theory evolved. Thus, in everyday language, onoma meant ‘a name’, rhe¯ma meant ‘a saying’; logos meant ‘speech, discourse’ and grammatike¯ meant ‘writing’. As grammatike¯ evolved into ‘grammar’, onoma came to mean ‘a noun’; rhe¯ma became first ‘rheme’, in the Prague sense, and then ‘a verb’; while logos came to mean ‘a sentence’. Here it was the folk linguistic terms for forms of discourse which became the source of technical nomenclature in grammar. Once having reified these abstract categories by naming them, the Greek grammarians went on to ask what the names meant. What ‘is’ a noun? they wondered. At first, this was a question of: how do I recognize a noun when I see it? how do I know whether something is a noun or not? But before long the questions came to be asked in the other direction: what ‘is’ a noun, in the sense of what function does it serve? At this second stage, instead of treating ‘noun’ as the Value and then supplying a Token for it, the definition now treats ‘noun’ as the Token and seeks to supply a Value for it (for the terms Token and Value see Halliday (1985: Chapter 5); also this volume, Chapter 7, p. 173). Instead of ‘a noun is a word that inflects for number and case’, we have ‘a noun is the name of a person, creature or thing’. This is a decoding definition, one which embodies a notion of ‘what the category means’. To define a linguistic term by encoding is relatively simple: one hops along the realizational chain of grammatical categories until reaching some form of output. Defining a noun in this way would involve, altogether, three steps: (1) a move in rank – a noun inflects 292

on the ineffability of grammatical categories

for case; (2) a move in delicacy – case is nominative, genitive, dative or accusative; (3) a move in exponence – the accusative case ends in – n. But how does one define by decoding? how do we say what a grammatical category means? To maintain some kind of parallel, saying what a category means would imply relating it to something that is observable on the content plane – to some aspect of experience, in the context of the culture. But this is a very different, and a very difficult, task. Let us return for a moment to the encoding type of definition. Between a grammatical category and its Token – that by which it is realized – runs the familiar line of arbitrariness in language; so the metalanguage in which the category is ultimately represented is of a different order from the category itself. It consists of letters, or phones, or some abstraction from these: phonological features, in a typical case. There may be some magic gateway of biuniqueness between these two worlds, as happens in a language where every morpheme is mapped into a syllable; but this, while it is a bonus to the linguist, definitely does not lead him to say that a morpheme and a syllable are ‘the same thing’. Hence in an encoding definition the category and its interpretation are clearly distinct; there is no danger of a statement such as ‘a noun is a word ending in –us’ being tautological. But between a grammatical category and its Value – that which it realizes – there is no such line of arbitrariness. Grammars are ‘natural’, in the sense that wordings are related iconically to meanings; and this, in fact, is how the name of the category was arrived at in the first place. Hence in a decoding definition there is no mechanism for insulating the category from its interpretation. A noun is called a ‘name’ because it means a ‘name’. So to define a noun by saying that it is the name of something (and the gloss ‘of a person, creature or thing’ adds nothing; it simply lists everything nameable) is, at first sight, merely tautologous. This apparent tautology is one that is discussed by Michael Reddy in his paper ‘The conduit metaphor’ (1979), in which he develops Whorf’s theory of the metaphor of the container in languages of the Standard Average European mould. According to Whorf, western languages characteristically employ an extended spatial metaphor whereby mental processes and relations become ‘objectified’, as illustrated in his famous passage (1956: 146): I grasp the thread of someone’s argument, but if its level is over my head my attention may wander and lose touch with the drift of it. 293

construing and enacting

This is accompanied by an antinomy of count / mass, which produces expressions such as a pane of glass, conceptualized as ‘a pane with glass in it’ (like a cup of tea), and extending to time construed as a container as in an hour of bliss. Reddy takes this argument one step further and shows how the metaphor determines the way we talk about communication, with ideas, meanings and emotions being packaged inside words and sentences and piped along a conduit. He lists about 150 such expressions, like get your thoughts across, his feelings came through, the lines are empty of meaning. All this, according to Reddy, is innocent enough. But it may become pathological, when the conduit metaphor pervades the whole terminology of linguistics, because the effect of this metaphor is that all words for kinds and quantities of discourse, like poem, message, text, are split into two meanings: (i) the content, and (ii) the container. In our terms, there is a stratal polysemy: these words refer both to a piece of meaning (semantic stratum) and to a piece of wording (lexicogrammatical stratum). Reddy’s view is that as long as one remains within the conduit metaphor the effect is still benign, because the one sense is then metonymic to the other: that is, the meaning is ‘contained in’ the wording. So text2 ‘wording’ ‘contains’ text1 ‘meaning’, and no great harm is done. It is if one tries to escape from the metaphor that this kind of polysemy becomes pathological. It will have been noticed that this stratal polysemy is precisely the process by which the original Greek terms had come to be extended so as to serve in the grammar: starting as names of semantic (or presemantic) categories, they were then transferred to become names for the lexicogrammatical categories by which the former were (typically) expressed. So logos2 ‘sentence’ is that by which logos1 ‘discourse’ is realized. Likewise with onoma and rhe¯ma: onoma2 ‘noun’ is what realizes onoma1 ‘name’ – while with rhe¯ma there is an additional step: rhe¯ma1 ‘saying’ is realized by rhe¯ma2 ‘Rheme’, which is in turn realized by rhe¯ma3 ‘verb’. It is not hard to see why this happens. It soon becomes obvious, once one begins to be aware of language as an object and starts to investigate its central processing unit, that the categories of this coding system are not arbitrary, but relate systematically to the meanings. It is natural, therefore, to name them by reference to their semantic function. With terms for classes and units, like those above, the polysemy is confined within language. Both ‘noun’ and ‘name’ are linguistic entities; the difference is simply that one is grammatical, the other 294

on the ineffability of grammatical categories

semantic. The same is true of the pair ‘sentence’ and ‘discourse’. But with the majority of grammatical terms – those for functions, systems and terms in systems (features) – this does not hold. These too have to be imported into grammar from outside; but in this case they come not from folk semantics but from outside language altogether. When categories of this type come to be named, the terms that are introduced for the purpose interpret the categories by reference to some aspect of extralinguistic experience. A typical example would be a complex of categories such as ‘(system) number: (features) singular / plural’. Consider a label such as ‘plural’. This derives from the ideational meaning of the category: it expresses a relation between that category and the speaker’s experience of the world. The term ‘plural’ is the name of this relationship. But the same term is also used as the name of the grammatical category which realizes this relationship; a noun will be said to be ‘plural (in number)’. And this can cause problems. Typically in the history of western linguistics the reasoning has proceeded as follows. In the morphology we are presented with a certain category, let us say a two-term system, of the form ‘a : x/y’; thus, ‘all a are either x or y’. By inspecting, or perhaps introspecting, typical contexts in which these forms are used we recognize a redundancy, such that x redounds with one of a set of countable things and y redounds with two or more. The abstract labels ‘a : x/y’ are then replaced by the substantive labels ‘number : singular / plural’. This is of course an idealized model of the process; such abstract labels have never been used, as far as I know, at least until modern times. But it helps to bring the issue into focus. What happens next? The categories that have been labelled in this way then come to be reified and the question is asked what they mean. The answer given is: ‘singular means one of a thing, plural means more than one’. In fact, these are definitions of the words singular and plural; but they are made to serve as definitions of the metawords, the terms of the metalanguage. Next, the terms are evaluated for their predictive power: will they correctly predict text from situation, or situation from text? Given a plural form, will it refer to more than one of a thing? Given more than one of a thing, will it be referred to by a plural form? The answer this time is: yes, with a certain degree of probability – high enough for many purposes, but inadequate for some, and disturbing for those who like their categories pure. This then gives rise to a theory of ‘core meanings’: a term of the metalanguage is said to represent the ‘core meaning’ of the category. With this defence, in 295

construing and enacting

spite of fruit and furniture, scissors and oats (or their counterparts in the language in question), ‘singular’ and ‘plural’ can continue to be used. This problem – that of interpreting a symptom and then labelling its interpretation – is common to all sciences. It arises in any realm of discourse involving explanation and abstraction. Somehow, a metalanguage has to be created, and created out of natural language, in order to assign a Value to a Token. I go to my doctor with a swelling somewhere on my body. He looks at it, and pronounces ‘You’ve got an oedema.’ What is oedema? – Greek for ‘swelling’. But what he is diagnosing is a more abstract swelling – it is that which is manifested by the swelling on my body. An oedema is the Value of which a swelling is the Token. The use of a different term in this way allows for stratal diversification: not all swellings realize oedemata, and not all oedemata are realized as swellings. The relationship is a probabilistic one – and hence invites further, more delicate investigation. Where do such terms come from? Ideally, they come from a parallel but distinct semiotic. It should be a natural language, in order to maintain the non-arbitrariness of the relation between the symptom and the underlying condition – given swelling and oedema we would predict that, in default of any special circumstances, they will refer to the same thing. But it should be a different language, or at least a different sub-language, in order to allow for instances where they do not. And it should be a higher status code, in order to symbolize a higher order of abstraction (and also the social value of abstract knowledge). The ideal source of a metalanguage is thus the ‘high’ variant in a diglossia. A word of a natural language that is at one remove from primary reality, such as ancient Greek, or classical Chinese, or Sanskrit, is appropriate for symbolizing a phenomenon that is at one remove from primary observation. But when it comes to metalinguistic matters, linguistics presents a special case. It is not just another science. It is ‘language turned back on itself’, to use Firth’s (very British) expression; or, in Weinreich’s (very American) formulation, ‘language as its own metalanguage’. As a consequence, where other sciences need two terms, we need three: one for the phenomenon, and two for the metaphenomenon, one grammatical and the other semantic. To return to the example of number: we need to be able to say that the grammatical category of ‘plural’ typically expresses (realizes) the semantic category of, say, ‘manifold’, which typically expresses (redounds with) more than one of a thing. But notice what has happened. The grammatical category of ‘plural’ 296

on the ineffability of grammatical categories

was set up in the first place to account for a morphological phenomenon: suppose this had been in English, then the –s / –z / –iz of cats, dogs and horses. At this point, therefore, we ought to have come round in a circle: –s / –z / –iz means –s / –z / –iz. But instead we have tried to escape from the circle by finding a gloss for –s / –z / –iz – that is, an exact synonym for it, in natural language wording; and that is an extremely difficult thing to do. We might try glossing it as more than one, or several, or many; but the trouble is we don’t actually say I like more than one cat, or I like many cat – we say I like cats. The meaning of the –s on cats is impossible to gloss in natural language, except by means of itself. The category is, quite simply, ineffable.

2

Difficulties with the subject

Why should this be so? One hypothesis might be that natural languages are not good things for glossing with; in that connection, Reddy remarked, ‘As a metalanguage, English, at least, is its own worst enemy.’ We can certainly point to some deplorable habits that English has, both in its vocabulary and in its grammar. For example, we frequently use the same lexical item to stand both for the study of a particular phenomenon and for the phenomenon itself, as when we talk of someone’s psychological make-up instead of their psychic make-up. It can be disastrous for students of linguistics (not to mention the general public) that grammar is both the name of a stratum in language and the name for the study of that stratum; and likewise with phonology and semantics. Not even the conduit metaphor excuses a ragged polysemy such as these. Even worse are some of English’s grammatical pathologies. For our metalinguistic vocabulary, we usually draw on some parallel semiotic as already illustrated, bringing in new words so as to be freed from the accumulated associations of the old ones. (The freedom is often shortlived, since the new term may soon be borrowed into the daily language, like the psychological above.) But for the grammar of our metalanguages we are usually content to stick with the everyday forms of English; and this can lead to serious misconstructions – such as the following, perpetrated by myself, when I wrote some time ago: the Theme of an English clause is the element that is put in first position. Now I meant this as Value ^ Token, with is meaning ‘is represented by’. But all such clauses in English, if they have the verb be, are 297

construing and enacting

ambiguous; and this one was frequently misread as Token ^ Value, with is taken to mean ‘represents’. In other words, a clause that was intended to say how the Theme in English is to be recognized was taken as a statement of how it is to be defined – one of the most fundamental confusions in linguistics. It would all have been avoided if the verb be had had a passive; I should, therefore, have created the appropriate metagrammar and written: The Theme of an English clause is been by the element that is put in first position. So there are problems in using natural language as a metalanguage, for whatever purpose: its logical and ideational systems were not designed for the task. Some combinations of features may be realized in ways that are ambiguous, others may carry a baggage of unwanted corollaries, and so on (this does sometimes lead to the creation of minor neologisms in the grammar, like the prepositional phrases that appear in the language of mathematics (I mean mathematical English) such as the inequations over O, symmetry about a certain point for various angles of rotation). And using natural language as a metalanguage for natural language itself is likely to inflate the problems still further, since whatever shortcomings it has are compounded by the factor of selfreference – the metalanguage being a form of the same semiotic system that it is also being used to describe. The problem of self-reference is a familiar one; nevertheless it is not the central issue. The real problem lies in the nature of language as object, and particularly the nature of lexicogrammar. It is not so much that language is not good for glossing with. The problem is rather that language is not good for being glossed. Let us take, as an example, the category of Subject. This has always been one of the most obscure and controversial categories in western grammatical theory. Here is Jespersen on the subject (1909–43, Volume 3: 206–7): The subject cannot be defined by means of such words as active or agent; this is excluded by the meaning of a great many verbs, e.g. suffer (he suffered torture), collapse, as well as by passive constructions . . . How are we to distinguish between the subject and the object (or the objects)? The subject is the primary which is most intimately connected with the verb (predicate) in the form which it actually has in the sentence with which we are concerned; thus Tom is the subject in (1) ‘Tom beats John’, but not in (2) ‘John is beaten by Tom’, though both sentences indicate the same action on the part of Tom; in the latter

298

on the ineffability of grammatical categories sentence John is the subject, because he is the person most intimately connected with the verb beat in the actual form employed: is beaten. We can thus find out the subject by asking Who (or What) followed by the verb in the form used in the sentence: (1) Who beats (John)? Tom / (2) Who is beaten (by Tom)? John. There are also outward signs which sometimes, but not always, assist us in recognizing the subject, viz . . .

As if this was not confusion enough, the category of subject is subsequently used in the interpretation of further categories (1909–43, Volume 7: 122–3): The generally indefinite character of sentences with there is . . . shown by the ‘subject’, which in the majority of cases is indefinite . . . Not infrequently, when the subject is seemingly definite, the underlying notion is really indefinite as shown by the indefinite article after of: it was not long before there shone in at the door the ruddy glimmer of a lantern.

– showing incidentally that Jespersen also failed to understand the meaning of the, which is another ineffable category in the grammar of English. To confine ourselves to Subject, however, here is a brief extract from a discussion by a grammarian concerned with a nonwestern language (Chao 1968: 69): The grammatical meaning of Subject and Predicate in a Chinese sentence is topic and comment rather than actor and action.

and this is accompanied by a footnote saying: Note that we are using the terms ‘topic’ and ‘comment’ as semantic terms and not as grammatical terms as used by many writers in discussing Chinese grammar.

It is only a very short step from here to the assertion that the Subject ‘has no meaning’. The implication is: whatever it is that is functioning as Subject in any instance has meaning as actor, or has meaning as topic; but as Subject it has none – the category of Subject has no meaning in itself. In this view, Subject is a grammatical function whose only function is to be a grammatical function. Such a view is enshrined in the terminology, in the term ‘grammatical Subject’ (used for example in Sweet 1891); this is in contradistinction to ‘logical Subject’ (i.e. Actor) and ‘psychological Subject’ (i.e. Theme). Compare the later Prague school interpretation, with ‘syntactic structure’ (Subject–Predicate) contrasted with ‘semantic structure’ 299

construing and enacting

(Actor–Action) and with ‘functional sentence perspective’ (Theme– Rheme) (Danesˇ 1966). The Subject is conceived of as being ‘purely’ grammatical – that is, as arbitrary, not realizing any semantic features. I have always rejected this view. In my opinion the category of Subject is no less ‘meaningful’ (semantically motivated) than other functional categories in the grammar. Nor in the last analysis is it any more obscure than other categories. It is just as impossible to arrive at an adequate gloss for functions such as Actor, Agent, Goal (‘logical direct object’), Range (‘logical cognate object’), Topic, Theme, New; or for grammatical features such as definite, passive, irrealis, equative, personal, human, modal – in fact more or less everything in the grammarian’s pharmacopoeia. It would be easy to pour scorn on the whole enterprise of trying to gloss such categories at all, in a metalanguage drawn from natural language. But there are sound and respectable reasons for wanting to do so. The original impetus for semantic glosses on the grammar comes from the desire to explain the observed formal patterns: why is this particular noun in the nominative case, or in the genitive? Why is this element put first, or after that one? Why is this verb in the passive? To answer questions like these one has to postulate more functional, semantically oriented generalizations; these are then used to predict further instances that have not yet been observed. Then subsequently the technique is extended to the interpretation of ‘unknown’ languages, either for descriptive or for pedagogical purposes. We can give an illustration again from Chao (1968: 448): A necessary condition for the use of jiann is that the first verb be for an event which happens to the ‘actor’ without his volition. Thus, there is no ⬚mhojiann ‘feel (by hand) for, so as to feel, – feel’, since the act of feeling with one’s hand is considered more active than the reception of the ‘distant senses’.

A much more extended example would be Whorf’s famous discussion of tenses (or, in his later term, assertions) in Hopi: factual or presentpast (later reportive), future (later expective), generalized or usitative (later nomic) – which initially takes up half a page, but whose semantic and ideological interpretation is the subject of an entire article (‘An American Indian model of the universe’). Such discussions typically involve implicit, or sometimes explicit, contrast with the language in which they are written; in this case, English. More recently, there has evolved a third context for semantic glosses: research in text generation, in the framework of artificial intelligence. 300

on the ineffability of grammatical categories

One project in which they play an essential part is the Penman project at the Information Sciences Institute (Mann and Matthiessen 1983; Mann 1983). In Penman the motive power is provided by the grammar, Nigel, which is a systemic grammar consisting of a network of some hundreds of options. At each choice point a Chooser is activated; the Chooser consults the environment (the Knowledge Base) for instructions on which way to go. The Chooser’s questions are referred to an Inquiry Operator; and they take forms such as the following: Is this concept inherently multiple, i.e. a set or collection of things, or is it unitary? Is the process one which conceptually has some sort of entity which causes the process to occur? Does this represent a concept which the speaker expects the hearer to find novel, not previously mentioned or evoked, and thus does not expect the hearer to identify uniquely by reason of the attention which it currently holds, its inherent uniqueness in culture, or its association with an identifiable entity? All such glosses are attempts to get at the grammar beneath the skin; and they may be supported by a variety of different beliefs. First, it may be assumed that all grammatical generalizations have some significance at a higher stratum; or alternatively, that some are simply housekeeping devices and have no semantic function. Secondly, those grammatical categories that are regarded as semantically significant may be thought of as universal, or as particular to the given language, or as particular to a given register, a functional variety of a language. (These would represent fairly well the respective views of Jakobson, Hjelmslev and Firth.) Thirdly, it may be held that every such category has one meaning that is common to all its manifestations, and the problem is to find the right semantic generalization to cover all cases; or alternatively, that some categories at least are polysemous, so that their meaning varies in ways that are not predictable from the context (cf. Ikegami 1980: 59). Fourthly, there is a range of beliefs about the place of grammar, and the need to postulate some higher level semiotic system (‘semantics’, ‘semology’, ‘the conceptual level’, etc.) to which grammatical categories can be related in a systematic and in some sense ‘natural’ way. Positions taken on these issues may complicate the task of semantic interpretation: for example, if categories are assumed to be universal, and yet are established at an insufficiently abstract level. But whatever beliefs are held about them, grammatical categories 301

construing and enacting

will remain ineffable. Some of the more recalcitrant ones are categories that Whorf originally called ‘covert’: ‘having no mark other than distinctive reactances with overtly marked categories’. But by no means all are of this kind – there is nothing covert about definiteness in English, for example. The phenomenon we are concerned with has more to do with Whorf’s follow-up notion of a ‘cryptotype’. Whorf remarks of these that ‘they easily escape notice and may be hard to define, and yet may have profound influence on linguistic behavior’ (1956: 92). Among cryptotypes in English Whorf cites gender, transitivity (of the verb), inherence (of the adjective), and various more delicate categories, such as that of verbs that may be phrasalized with up. There is, of course, a connection between the two senses in which Whorf is using ‘crypto-’: a category may be hard to define precisely because it is hidden from view. But hidden from whose view? It is not because they are hidden from the linguist that grammatical categories are hard to define; once the linguist has found them, the fact that they had escaped his notice ceases to matter. The significance of this concept of a cryptotype is that it is something that escapes the notice of the speakers of the language. Franz Boas long ago drew attention to the unconscious nature of language, contrasting it in this respect with the other meaning systems of a culture; and although his observations have often been quoted, it seems to me that their significance is seldom fully taken into account. There is a fundamental relationship between the unconsciousness of language and the nature of its semantic categories. I have often pointed out, in the many years since I began the study of informal speech, that it is only in the most spontaneous, un-self-monitored kinds of discourse that a speaker stretches his semantic resources to the utmost (cf. Halliday 1966). This does not happen in formal speech; and it certainly does not happen in writing. It is in unconscious spoken language that we typically find the truly complex sentences, with their labyrinths of hypotaxis and all their projections and expansions, from which, while we blunder through such sequences often losing ourselves completely when we are engaged in the planned self-monitoring discourse of an academic lecture, we emerge in good order and with every node unravelled provided we are completely unaware of what we are saying and attending only to whatever it is we are involved in at the time. (That sentence is best taken orally, at high speed.) Our ability to use language depends critically on our not being conscious of doing so – which is the truth that every language learner has to discover, and the contradiction from which every language teacher has to escape. 302

on the ineffability of grammatical categories

Why is there this apparent contradiction, such that only when we cease to attend to the process of meaning can we ever master the ability to mean? The immediate cause is no doubt the dynamic nature of the medium – as with walking, or riding a bicycle: once you think about it you fall over, and you only succeed when you no longer have to try. But above and beyond this is a more abstract phenomenon, a specific property of unconscious spoken language which distinguishes it from all conscious discourse, spoken and written. While the complexity of conscious language is dense and crystalline, formed by a closely-packed construction of words and word clusters, the complexity of unconscious language is fluid and choreographic. Conscious language achieves its creative force mainly by lexical means; and lexical items are semantically close to experience. Unconscious language depends much more for its creative force on grammar – and grammatical categories are far removed from experience. To quote Whorf again, grammatical categories ‘represent experience . . . but experience seen in terms of a definite linguistic scheme’ (1956: 92). The meaning of a typical grammatical category thus has no counterpart in our conscious representation of things. There can be no exact paraphrase of Subject or Actor or Theme – because there is no language-independent clustering of phenomena in our experience to which they correspond. If there was, we should not need the linguistic category to create one. If language was a purely passive partner, ‘expressing’ a ‘reality’ that was already there, its categories would be eminently glossable. But it is not. Language is an active participant in the semogenic process. Language creates reality – and therefore its categories of content cannot be defined, since we could define them only by relating them to some pre-existing model of experience, and there is no model of experience until the linguistic categories are there to model it. The only meaning of Subject is the meaning that has evolved along with the category itself.

3

How children become grammatical

Meaning is formed in action; people create meaning, by exchanging symbols in shared contexts of situation. The symbols evolve along with the meanings; there is just one process taking place here, not two, though we have to interpret it as if it was two. We cannot observe this process as it took place in the history of the community, since that would be coextensive with the evolution of the human species. But we can observe it happening in the history of a human 303

construing and enacting

child. Semogenesis begins well before the mother tongue, as the infant creates his own protolanguage or ‘child tongue’: he constructs this symbolic system, in interaction with those who share in his meanings, for the twofold purpose of doing and thinking that characterizes all such systems; and in the same process he also constructs the objects of his action and his reflection. But the child tongue has its limitations, for both these purposes; so he moves into the adult mode, and takes over the mother tongue with its ready-made grammatical categories. The symbols of the mother tongue, which have been around him from the start, now become his reality, at once a part of, and a key to, the complex phenomena of his experience. Language and culture are construed as one. Does a child, then, know what a Subject is? We cannot ask him; nor can we set up a test situation to find out – if only because children, given an unnatural task, will respond with unnatural behavior. (It is not intended to suggest that there is any contrast here with adults. The problem lies again in the unconscious nature of linguistic processes, which adults cannot reproduce experimentally either.) Nevertheless it is clear, surely, that a child does know what a Subject is, because he uses one a hundred times a day. We only have to listen to a five-yearold in ordinary, real life situations, and we will hear the categories of the grammar that we find most difficult to explain, deployed in their appropriate semantic roles. What we observe, of course, even with a tape recorder on permanent duty, is only a limited set of instances. We have to infer the system that lies behind them; for language (if I may be allowed to invert Chomsky’s famous dictum) is an infinite system that generates only a finite body of text. But what we can observe is already very convincing. If I assert that a five-year-old knows what a Subject is, it is because I have listened to children for many years, and heard them talking in clauses which have Subjects. In my own detailed record of one particular child, there are about 2,500 of them; but since child language studies became fashionable there has been an abundance of such material available, if one does not feel one can rely just on one’s listening. Now, any one of these clauses could have had the appropriate Subject by chance. Moreover, since no linguistic category is chosen in isolation – in choosing the Subject one is always making other choices besides (and this will apply whatever category is used as illustration) – in any one instance we could always claim that the appropriateness of the Subject was a consequence of some other choice. But if countless 304

on the ineffability of grammatical categories

children produce countless clauses each with an appropriate Subject, the probability that this represents a systematic choice on the part of the speaker amounts to a virtual certainty, because there is no other single choice with which the Subject is always associated. Can we also find a critical instance in which the category of Subject is highlighted? It can never be insulated from other meaningful choices, for reasons already given. But we can set up a pair of agnate clauses which differ primarily in respect of which element is selected as the Subject. Consider the following pair: Why was that given to mummy now? Why was mummy given that now? These have the same transitivity structure: Goal / Medium that, Recipient mummy, Process give, Cause why, Time now. They have the same thematic structure, with why as Theme; and the same information structure, with focus of New on now. The only significant difference lies in the choice of the Subject: in the first example that, in the second example mummy. We can display the Subject prominently by turning the clause into the declarative and adding a tag or response: That was given to mummy now, wasn’t it? – No it wasn’t. Mummy was given that now, wasn’t she? – No she wasn’t. The Subject in English can always be recognized in this way: it is that element that turns up (in appropriate pronominal form) in the repeat. This not only enables us to identify the Subject; it also makes it clear what the Subject means, and why the speaker chooses that particular entity to figure as Subject of the clause. That is the entity that he wants to appear in tag or response – or rather, the entity that he wants to carry the meaning that is realized by its potential for appearing in tag or response, whether or not any tag or response is actualized. (Compare the reason for putting something in clause-final position: so that it carries the meaning that is realized by its potential for bearing the unmarked focus of information, whether or not that is where the focus actually falls.) How does this relate to the speech of our five-year-old? A particular child is unlikely to produce any particular wording; but he could produce instances of either of these structures if the occasion arose. He would not, of course, produce them one after the other, since we do not talk in paradigms; whichever occurred would be in its appropriate syntagmatic environment. On the other hand, I doubt whether he 305

construing and enacting

could produce either of them under test conditions, or discriminate between them if presented to him; but then neither could an adult speaker. The fact that we cannot bring to consciousness the difference in meaning between two related forms which we nevertheless keep systematically apart tells us nothing about the semantics of the language. It is not really surprising that the child controls the semantics and grammar of the Subject in his language. Assuming that he has been actively listening for some 31⁄2 out of his five years, and assuming that he has been exposed to a tally of about 200 Subject-featuring clauses a day (and these are very modest assumptions), then he has heard anything up to a quarter of a million Subjects in the course of his life. All of these, moreover, have been functional in some context of situation. If we take note of Jay Lemke’s observation (1985) that ‘meaning is created at the intersection of the material and the discursive’, and put it together with Whorf’s ‘sense of the cumulative value of innumerable small momenta’, we will not be surprised that a grammar can be learnt in this way. I have been insisting for many years that a child’s semiotic experience is extraordinarily rich – not at all the farrago of featureless fragments that we are taught spoken language consists of! Much of it comprises repetitions, like Come and have your lunch; but repetition is itself a semogenic factor, since it allows the child to model the language as a probabilistic system. Every instance, whether repetitive or unique, is a configuration of meanings of different kinds, available to the child both for storage as coded text and as evidence for construing the system that lies behind. There is no mystique in a child’s ability to construct a language on the evidence of what he hears. But that which makes the category of Subject learnable is also that which ensures that it will be ineffable. How can we generalize, in a single definition, or even in an article or a book, the whole of the shared experience of Subjecthood of the adult speech community – or even that of one novitiate member of it? If a language had been a designed system, matters would have been different. Designed systems are designed so as to be effable; in fact, effability is a necessary condition of design. You cannot design unless the principles can be made explicit. But a language is an evolved system; and evolved systems rest on principles that are ineffable – because they do not correspond to any consciously accessible categorization of our experience. Only the relatively trivial meanings of a natural language are likely to be reducible to (meta-)words. Fundamental semantic concepts, like those underlying Subject, or Theme, 306

on the ineffability of grammatical categories

Actor, New, definite, present, finite, mass, habitual, locative, are, in an entirely positive way, ineffable. Can we then do nothing to make such categories explicit? If we try to describe any semiotic system ‘from below’, first reifying the forms through which it is expressed and then asking what these forms mean, we will simply get a relabelling of each formal category. The description will be a gloss on its name (since the name was an attempt to capture its meaning in the first place), together perhaps with a gloss on its relationship to other categories that were themselves formally established. To pursue our earlier example from grammar, a noun will be defined as ‘that which names a person, living being or inanimate object (gloss on the name “noun”); which can be a participant in an action or event (gloss on the expression “subject or object of a verb”), and may be single or multiple (gloss on “singular or plural in number”)’. This can always be done – whether or not it is useful – with categories like that of ‘noun’ which are classes: they are the ‘output’ categories of the grammar, lists of items that can figure at particular places in the syntagm. A category such as ‘noun’ makes no direct contact with semantics, other than in this restricted sense of semantics as a commentary on the meaning of the forms. But the ‘input’ categories of the grammar – the systems, such as ‘mood’; their features, such as ‘indicative’; and the functions, such as ‘Subject’ – cannot be so readily glossed in this way: they relate directly to the semantic system that is ‘above’ the grammar, that which interprets the ideologies of the culture (Lemke’s ‘activity structures’ and ‘thematic systems’), and codes them in a wordable form. To understand these categories, it is no use asking what they mean. The question is not ‘what is the meaning of this or that function or feature in the grammar?’; but rather ‘what is encoded in this language, or in this register (functional variety) of the language?’ This reverses the perspective derived from the history of linguistics, in which a language is a system of forms, with meanings attached to make sense of them. Instead, a language is treated as a system of meanings, with forms attached to express them. Not grammatical paradigms with their interpretation, but semantic paradigms with their realizations. So if we are interested in the grammatical function of Subject, rather than asking ‘what does this category mean?’, we need to ask ‘what are the choices in meaning in whose realization the Subject plays some part?’ We look for a semantic paradigm which is realized, inter alia, by systematic variation involving the Subject: in this case, that of speech function, in the interpersonal area of meaning. This recalls Firth’s 307

construing and enacting

notion of ‘meaning as function in a context’; the relevant context is that of the higher stratum – in other words, the context for understanding the Subject is not the clause, which is its grammatical environment, but the text, which is its semantic environment. For this very reason, it is difficult to give a brief illustration. But here is a piece of dialogue, that has been doctored so as to keep it short, which displays something of the meaning of the Subject. The speaker is telling a story of a sporting experience, and he says: I caught the first ball, I was beaten by the second; the third I stopped – and by the fourth I was knocked out. Let us identify the different grammatical variables, and establish clause by clause what has changed and what has remained constant. 1 2 3 4

Subject

Actor

Theme

I I I I

I ball I ball

I I ball ball

Subject = Actor = Theme Subject = Theme; ⫽ Actor Subject = Actor; ⫽ Theme Actor = Theme; ⫽ Subject

The speaker keeps the listener’s attention by varying the Theme and the Actor: 1 2 3 4

Theme

Actor

about me (‘I . . .’) about me (‘I . . .’) about the ball (‘the 3rd’) about the ball (‘by the 4th’)

what I did (‘did you?’) what happened to me (‘were you?’) what I did (‘did you?’) what happened to me (‘were you?’)

Clause 1 has the speaker in all three functions of prominence: interpersonal (I as Subject), ideational (I as Actor) and textual (I as Theme). Clause 2 is marked by ideational modesty (this is what happened to me, not what I did), and clause 3 is marked by textual modesty (now I’ll tell you about the ball, not about myself). In clause 4, the speaker gains further merit by ending on a doubly modest note, in which he is neither the Actor nor the Theme. But in regard to the speech function, the picture is quite different. The speaker retains himself in the role of Subject throughout. There is no sign here of interpersonal modesty; the assertion is made to rest on I every time, and the listener’s response, correspondingly, must always have a you in it – Did you?, Were you? In other words, every step in 308

on the ineffability of grammatical categories

the narrative has to be validated by reference to ‘me’; and that is the meaning of Subject. ‘Subject’ is not the same as ‘Actor’; nor is it the same as ‘Theme’. But it is far from being devoid of meaning. It is quite possible to have a clause in which all three of these functions are dissociated from one another: that teapot Theme

my aunt Subject

was given by the duke Actor

This helps to give a sense of the different meaning of each. This particular one is made up; but the structure of which it is an instance is entirely familiar. (Except perhaps to philosophers of language, who tend to be disturbed by departures from the ideal of John hit Paul. Once when I was giving a seminar, a participant refused to accept that such a clause was possible in English, with a non-Subject nominal in fronted position. So I asked him a question – I forget what the question was, but any question would have served; and when he answered, I said ‘Yes – that answer I’ve been given by other people too.’ Needless to say, he raised no eyebrow; he did not notice that I had used a clause with the structure he had just rejected as impossible. When I pointed this out to him, he seemed to think I had cheated – perhaps by bringing real language into the discussion.) Obviously, there would be no sense in dissociating the Subject from both Actor and Theme if it did not embody a meaning of its own, distinct from either of the other two.

4

Talking about the ineffable

What I have been trying to show with this illustration is that while, with a category like Subject, it is impossible to answer the question ‘what does it mean?’, this does not signify that it has no meaning. The problem of ineffability is common to all grammatical categories; there are various reasons why some may seem less problematic than others, but it is an illusion to think that any can be exhaustively defined. And this, as I remarked above, is not because of the shortcomings of natural language for serving as a metalanguage, real though such shortcomings are. Rather the converse: it is the very richness of natural language, its power of distilling the entire collective experience of the culture into a single manageable, and learnable, code that puts its categories beyond the reach of our conscious attempts at exegesis. This leads us back to the question of the Grundbedeutung. The categories we have been considering have been categories of the grammar: grammatical systems and structures, and their component 309

construing and enacting

features and functions; and since grammar is the central processing unit, where meanings of different kinds are brought together as wordings, we expect its categories to be valid for the language as a whole. All uses of English involve ‘Mood : Subject + Finite’, or ‘tense : past / present / future’; and these are assumed to be in some sense ‘the same thing’ in all contexts, since otherwise we would not be looking for definitions of them. No such constraint figures in our conception of semantics. The grammar is the grammar: it has internal organization, of a metafunctional kind; it has some special purpose sub-grammars; and it has considerable indeterminacy – but there is such a thing as ‘the grammar of English’. We do not operate with a separate grammar for each register. No doubt we can also conceive of such a thing as ‘the semantics of English’; but we also feel that (at least at the present state of our knowledge) it is not counter-productive to envisage a more restricted domain for semantic generalizations, and to operate with semantic sub-systems each relevant to a specific universe of discourse. In principle, the domain of a semantic description may be anything from ‘the whole language’ down to a single text. At one end of the scale, I have found it useful to set up a semantic system relating to just one dialogue of 35 words long; this was a child–adult dialogue, and the purpose was to explore what meaning potential the child must have in order to be able to construe such a discourse (see Appendix, p. 313). Geoffrey Turner’s (1973) semantic networks define a rather broader range of texts, such as mother–child control patterns in specific experimental situation types. More general again is Ruqaiya Hasan’s (1983) ‘message function’ network, which describes spontaneous interaction between children and parents, for the purpose of investigating the development of children’s learning patterns. At the other end of the scale, J. R. Martin’s (1983a, 1992) conjunction networks are like grammatical networks in that they are set up for the language as a whole. When we describe semantic systems, we are saying what it is that ‘preselects’ the grammatical categories: what choices in meaning call on what features in the grammar for their realization. It is by this process that the grammatical categories are defined; when this is done, there is no need to gloss them further. Once the semantic system is made explicit, it can only be misleading to attach separate semantic descriptions as glosses to the categories of the grammar. At the same time, if the semantic system is set up only for a restricted domain, some particular register variety, then the meanings of any 310

on the ineffability of grammatical categories

grammatical categories that figure distinctively in that variety will appear thereby less ineffable. For example, we have no general definition of ‘future’ as a category of English grammar; its effability measure is decidedly low. But when this category figures in the register of weather reporting and forecasting, the semantics of that variety makes only limited demands on it, for realizing the meanings that are engendered by that particular context. The category of ‘future in the register of weather forecasting’ is much less resistant to being glossed than the general category of ‘future in English’. This interpretation of semantic systems is a kind of functional semantics, and it derives from the twentieth century functional semantic traditions of Boas, Sapir and Whorf, of Malinowski and Firth, and of Mathesius and the Prague school. These were three groups of scholars with very different orientations, but their work was complementary in significant ways. While each had a well-rounded view of language, they emphasized, respectively, the ideational, the interpersonal and the textual aspects of meaning. For Malinowski, language was a means of action; and since symbols cannot act on things, this meant a means of interaction – acting on other people. Language need not (and often did not) match reality; but since it derived its meaning potential from use, it typically worked. For Whorf, on the other hand, language was a means of thought. It provided a model of reality; but when the two did not match, since experience was interpreted within the limitations of this model it could be disastrous in action – witness the exploding petrol drums. Mathesius showed how language varied to suit the context. Each sentence of the text was organized by the speaker so as to convey the message he wanted at that juncture, and the total effect was what we recognize as discourse. Their work provides the foundation for a systematic functional semantics which enables us to bridge the gap between the context of culture and the language, and between the context of situation and the text. This is how we can become aware of the meaning of grammatical categories. As a final step, let us summarize some of the alternative principles that can be adduced for talking about the ineffable: A. Metonymic: the use of some semiotic system as a descriptive metalanguage (the ‘parasemiotic’ principle) 1. Parallel semiotics within the same language (a) everyday language as folk linguistics (possible where there is a shift of metafunction but probably not otherwise) 311

construing and enacting

(b) open-ended definitions from theories about language (not theories of language), e.g. rhetoric (c) self-contained, technical systems of definition within linguistics itself, e.g. Hjelmslev’s 2. Translation and commentary in another language, such as Whorf on Hopi: ‘A first approximation to the meaning of the . . . projective is “does with a forward movement” ’ (1956: 103) – a (presumably non-existent) example would be a grammar of English in (non-westernized) Hopi 3. Deaf sign as metalanguage: the visual modality of sign gives it a different semiotic potential (the semantic field of a sign is in general greater than that of a lexical item; the sign is susceptible of a greater range of modification; there is greater potential for iconicity) 4. Non-linguistic semiotic systems (a) representational: some meanings, at least, can be depicted, e.g. Kluckhohn and Leighton on Navaho (1962: 284): two forms of a verb, one meaning ‘doing repeatedly to the same Goal’, the other ‘doing repeatedly to different Goals’, ‘glossed’ by two picture series each of three events, one person kicking another – the one being kicked either falls lower and lower (same person kicked repeatedly) or stays in the same posture (different person each time) (b) Non-representational: e.g. music as metalanguage, e.g. Chandola (1970: 145) suggests that many aspects of ragas can be compared with language, and gives an example of a musical pun – but music might lend itself more to glossing dynamic aspects of the grammar such as information structure B. Metaphoric: the use of theoretical models 1. Interpretative metaphors for particular features of language; e.g. Whorf’s interpretation of assertion in Hopi as ‘two grand cosmic forms’ (1956: 59n.); transitivity and ergativity as complementary modes of representing experience 2. A general theory of register and genre, from which semantic categories can be derived in a principled way (Martin, in press) 3. General theories of meaning in language, e.g. Lemke’s (1985) concept of ‘making meaning’ in terms of activity structures and thematic systems, and of ‘metaredundancy’ 312

on the ineffability of grammatical categories

4. Perhaps as a combination of the metaphoric and the metonymic we should cite Eco’s recent novel The Name of the Rose (1980), where instead of using language to talk about literature he turns the tables and uses literature to talk about language Whether or not language has the property that is sometimes claimed for it, of being able to interpret all other semiotic systems (and I see no reason to assume that this is so), there are certainly limitations on the ability of language to interpret itself. We may have to move outside language, to some parallel or higher order semiotic which, since it is not itself language, can be represented in language and then refracted to become a metalanguage for representing language. All such interpretation is ultimately circular; but in linguistics, we have tended to operate within circles that are pathologically small. Until we can create a greater distance between the semiotic object and the metasemiotic, grammatical categories are bound to remain ineffable.

Appendix Nigel at 5;4 (from Halliday (1984)) Nigel: Shall I tell you why the North Star stays still? Father: Yes, do. Nigel: Because that’s where the magnet is, and it gets attracted by the earth; but the other stars don’t so they move around. //2 shall I / tell you / why the / North / Star / stays / still// //1 yes // .1 do// //4 . because / that’s //1 where the / magnet / is and it //1 gets at/ tracted //1 by the / earth //4 . but the / other / stars //4 don’t so / /4 they //1 move a/round //

1.

Ideational

1.1

Experiential

Lexicogrammar: clause – transitivity [ Nigel ] shall I

tell

you

Sayer Process: Receiver Medium Verbal Beneficiary

why

the North Star stays

Cause Carrier Medium Actor

313

still

Process: Range relational intensive Process: material

construing and enacting that

’s

where

Identified Process: Token relational circumstantial Medium

it

the magnet is

Identified Value

they

move

around

Actor

Process: Place material

Medium Attribute Carrier Range Medium

Process: relational circumstantial

gets attracted by the earth the other stars don’t [get attracted by the earth]

Goal Process: Medium material

Actor Agent

Goal Medium

Process: material

Actor Agent

Process type

Process

Medium

2

do

stay still

North Star Cause: Why?

4

be at relational: identifying circumstantial: locative

is (represents)

North Star Location: where magnet is

material: middle

冀

4 be at relational: attributive [[ ]] circumstantial: locative

冁 is (has attribute) magnet

Other elements

Location: at North Star

5

do to material: effective

attract

North Star Agent: earth

6

do to material: effective

(not) attract

other stars Agent: earth

7

do

move

other stars

1.2

Logical

material: middle

Lexicogrammar: clause complex – interdependency, logical-semantic relation

‘why?’ . . . because and but so 314

on the ineffability of grammatical categories

Network: experiential

clause

{

PROCESS TYPE VOICE

[Nigel] 1–2 [Father] 3 [Nigel] 4–7

왘

왘

material relational verbal

31 32 33

effective middle

41 42

33.42 31.42 33.42 32by.41 [[3ay.42]]

{

31.41

왘

attributive identifying

a b

왘

intensive circ.: place possessive

x y z

31.41

31.42

Network: logical

clause complex

{

INTERDEPENDENCY (‘TAXIS’)

왘

paratactic hypotactic

expansion LOGICAL-SEMANTIC RELATION

123... abg...

왘

alaborating extending enhancing

= + ⳯

왘

locution idea

“ ‘

왘

projection

Semantics: experiential and logical: informal gloss 2 North Star not move ) 4 ( 5 6 7

cause? magnet be-at cause: add (effect): earth attract earth not attract contrast: move effect:

North Star North Star other stars other stars

‘a does not do x, because a has property p, but not-a have a not property p, so not-a do x’ Nigel at 5;4 (from Halliday (1984)) Nigel: Shall I tell you why the North Star stays still? Father: Yes, do. Nigel: Because that’s where the magnet is, and it gets attracted by the earth. But the other stars don’t so they move around. 315

construing and enacting

2 2.1

Ideational semantics Experiential reactants (‘magnet’) North Star objects

왘

왘

stars 왘

bodies

other stars earth

PHENOMENA

locating (‘be at’) processes

왘

move motion Moving (‘do’)

왘

왘

not-move (‘stay still’) force

alternatively:

moving (‘do’)

2.2

{

positive 왘

negative uncaused (motion) 왘

caused (force)

Logical contrastive (‘but’) additive

LOGICAL SEMANTIC RELATIONS

왘

positive (‘and’) 왘

effect (‘so’) causal

왘

unknown (‘why’) cause

왘

known (‘because’) 316

on the ineffability of grammatical categories

Interpersonal semantics COMMODITY ROLE

(proposal) goods-&-services

(proposition) information

give

offer

statement

demand

command

question

give

{

ROLE IN EXCHANGE

왘

demand tentative (‘shall I? / will you?’) VEIN COMMODITY EXCHANGED

왘

왘

goods-&-service (proposal)

neutral (‘let . . . / let me!’) definitive (I’ll . . . / you . . .’)

information (proposition)

2

Interpersonal

Lexicogrammar: clause – mood shall

I

Finite

Subject

Mood

tell you Residue

do Finite mood why

the North

WH/ Adjunct Resi-

Subject Mood

Star

stays

still

Finite due

that

’s 317

construing and enacting

it the other stars they

gets attracted don’t move around

Subject Mood

Finite Residue

Network: interpersonal declarative indicative

01 yes/no 02

왘

interrog

왘

major

clause

WH- 03

{

Mood

왘

1st person imperative

왘

1st/2nd persons 05 2nd person

MOOD PROJECTION

ELLIPSIS

04

direct

11

indirect

12

full

21

06

왘

왘

elliptical 22

[Nigel] 1 [Father] 3 [Nigel] 2

02.11.21 06.11.22 03.12.21

yes / no interrogative ‘offer, tentative’ 2nd person imperative ‘command, neutral’ WH- interrog, indirect ‘question, projected’

4–7 01.11.21/22 declarative, full / elliptical

318

‘statement’

on the ineffability of grammatical categories

Semantics: informal gloss

3

Textual

3.1

Structural (THEME and INFORMATION)

Lexcogrammar: clause – theme Shall

I

Intertopical personal Theme

tell you

why

the North Star stays still

Rheme

Intertopical personal Theme

Rheme

because

that

’s where the magnet is

and

it

gets attracted by the earth

but

the other stars

don’t

so

they

move around

structural Theme

topical

Rheme

1

discourse theme [first clause]: offer (interpersonal)

2

clause Themes: (interpersonal) question (topical) the North Star

4–7 clause Themes: (structural) because, and, but, so

(topical) [speaker]

(topical) the North Star, the other stars

319

construing and enacting

Lexicogrammar: information unit – information Shall I tell you why the North Star stays still

because That’s Where The magnet is

Focus Given Focus 왗

New

and it gets attacted

by the earth

Focus

Focus

Given 왗

New

왗

New

but the other stars Focus

New

New Given

Focus 왗

don’t

Focus

New

New

Focus New

Network: theme systems unmarked focus

unmarked TOPICAL THEME

TONICITY

왘

marked

320

왘

so they

Focus

move around

왗

New

marked focus

on the ineffability of grammatical categories

INTERPERSONAL THEME

TEXTUAL THEME

3.2

{ {

왘

mood finite — WH-

왘

modality —

TONALITY

unmarked (clause = info.unit) marked thematic (other) fresh

STATUS 왘

vocative —

왘

continuative —

왘

structural —

왘

conjunctive —

왘

contrastive

Cohesive

stays still

S A

1

왗 왘

the North Star |R that (=there) 왗 2 왘 the magnet |C |R 3 1 gets attracted it 왗 왘 왗 왘 the earth E

E

S H the other stars |R 1 move around 왗 왘 they

SA: SH: C: E: R:

왘

1

왗 왘

Key: synonymy – antonyms synonymy – cohyponyms collocation ellipsis reference

don’t Ø

3

왗 왘

Ø

1: Process + Medium 2: Process + Location 3: Process + Agent columns: lexical chain boxes: referential chain

321

construing and enacting

4. 4.1

Field Context of culture (the system)

Copernican – Galileo – Newtonian universe: heavenly bodies move: movement is the natural state of (such) things: non-movement is exceptional. Natural phenomena are subject to general ‘laws’: exceptions need to be explained: explanation is in terms of cause-&-effect. Construing this universe (Nigel at 5 years perhaps [escentric: geostatic]).

4.2

Context of situation (the instance)

Child constructing frequent of cosmology, rehearsing information derived from teacher (probably supplemented by charts / pictures; possibly with referent in direct experience of night sky): observational stars move, North Star doesn’t move; problem: why is North Star exceptional?; explanation: held by magnet.

5. 5.1

Tenor Context of culture (the system)

Family: parent–child as hierarchical relationship (age / generation): Parent as authority (‘+knowledge’ and ‘+power’). Class: middle class, intellectual: Role relationship personal rather than positional, hence (i) child can impart knowledge (which may be corrected); (ii) child announces intent (but seeks permission, which may be refused). - i.e. both forms of authority negotiable.

5.2

Context of situation (the instance)

Child (5 years) and parent interacting: child (i) makes explicit and (ii) seeks approval for interaction to impart knowledge: (a) displaying knowledge (boasting) (b) seeking confirmation. Parent approves: child proceeds to do so (would have done so anyway). 322

Chapter Twelve

SPOKEN AND WRITTEN MODES OF MEANING (1987)

1

Spoken language and education

It seems to me that one of the most productive areas of discussion between linguists and educators in the past quarter century has been that of speech and the spoken language. Twenty-five years ago, when I launched the “Linguistics and English Teaching” project in London, which produced Breakthrough to Literacy and Language in Use, it was still rare to find references to the place of spoken language in school, or to the need for children to be articulate as well as literate. Dell Hymes had not yet introduced “communicative competence”; the words oracy and orality had not yet entered the field (Andrew Wilkinson’s Some Aspects of Oracy appeared in 1967); David Abercrombie (1963) had only just published his ‘Conversation and spoken prose.’ Language, in school, as in the community at large, meant written language. The word language itself was hardly used in educational contexts. In the primary school, there was reading and writing; in the secondary school there was English, which meant literature and composition. Not that a classroom was a temple of silence; but the kind of spoken language that had a place, once a pupil had got beyond the infant school, was prepared speech: reading aloud, drama, debating – language that was written in order to be spoken, or at least was closely monitored in the course of its production. Spoken language in its natural form, spontaneous and unselfconscious, was not taken seriously as a medium of learning. First published in Comprehending Oral and Written Language, 1987. Orlando, FL: Academic Press, pp. 55–82.

323

construing and enacting

Among linguists, by contrast, the spoken language had pride of place. One learnt in the first year of a linguistics course that speech was logically and historically prior to writing. The somewhat aggressive tone with which linguists often proclaimed this commitment did not endear them to educators, who sensed that it undermined their authority as guardians of literacy and felt threatened by a scale of values they did not understand, according to which English spelling was out of harmony with the facts of the English language – whereas for them it was the pronunciation that was out of step, being a distorted reflection of the reality that lay in writing. The linguists’ professional commitment to the primacy of speech did not, however, arise from or carry with it an awareness of the properties of spoken discourse. It arose from the two sources of diachronic phonology (the study of sound change) and articulatory phonetics (the study of speech production), which came together in twentieth century phonological theory. This was an interpretation of the system of speech sounds and of the phonological properties of the stream of speech; it did not involve any attempt to study the grammar and semantics of spoken as distinct from written language. As early as 1911, in his discussion of functional variation in language, Mathesius (1964) was referring to “how the styles of speech are manifested in the pronunciation of language, in the stock of words, and in syntax” (p. 23), and to “the influence of functional styles on the lexical and semantic aspects of speech” (p. 24); and it is clear that “speech” for him (parole) did encompass both spoken and written varieties. But it was not until the 1950s, with the appearance of tape recorders, that natural speech could become the object of systematic study. The notion of “spoken text” is still not easily accepted, as can be seen from the confusion that prevails when spontaneous speech is reduced to writing in order to be analysed. Spoken language came to figure in educational discussions in the context of language in the classroom: the language used by teachers to structure, direct and monitor their students’ progress through the lesson. But the emphasis was on verbal strategies rather than on the text as a document; the investigators of the fifties and early sixties were not concerned with the particular place of spoken language in the learning process. It was assumed, of course, that students learnt by listening; but the expository aspects of the teacher’s language were given little attention, while the notion that a student might be using his own talk as a means of learning was nowhere part of the picture. Probably it would have been felt that the principal means of learning through the spoken language was by asking questions; but studies of the early seventies (for 324

spoken and written modes of meaning

example the Toronto research reported in Five to Nine) revealed that students seldom do ask questions – not, that is, while they are occupying their student role (i.e. in class). It is the teachers that ask the questions; and when they do so, both question and answer may be somewhat removed from the patterns of natural dialogue.

2

Complexity of natural speech

Already half a century earlier Franz Boas (1911) had stressed the unconscious character of language, unique (as he saw it) among the phenomena of human culture. Boas’ observation was to be understood in its contemporary context as a characterization of the language system (langue); not that, writing in 1911, he could have read Saussure’s Course in General Linguistics, any more than Mathesius could have done; but the unconscious was in the air, so to speak, and playing a critical role in the conception of systems as regularities underlying human behaviour. But Boas may also have had in mind the unconsciousness of the behaviour itself: the act of speaking (acte de parole) as an unconscious act. The lack of conscious awareness of the underlying system, and the difficulty that people have in bringing it to consciousness, are things which language shares with other semiotic systems – for example, social systems like that of kinship; what is unusual about language is the extent to which even the manifestation of the system, the actual process of meaning, remains hidden from observation, by performer and receiver alike. In that respect talking is more like dancing, or even running, than it is like playing chess. Speaker and listeners are of course aware that the speaker is speaking; but they are typically not aware of what he is saying, and if asked to recall it, not only the listeners but also the speaker will ordinarily offer a paraphrase, something that is true to the meaning but not by any means true to the wording. To focus attention on the wording of language is something that has to be learnt – for example if you are studying linguistics; it can be a difficult and somewhat threatening task. About 30 years ago, as a result of being asked to teach English intonation to foreign students, I began observing natural spontaneous discourse in English; and from the start I was struck by a curious fact. Not only were people unconscious of what they themselves were saying; they would often deny, not just that they had said something I had observed them to say, but also that they ever could say it. For example, I noticed the utterance it’ll’ve been going to’ve been being tested every day for the past fortnight soon, where the verbal group will have been 325

construing and enacting

going to have been being tested makes five serial tense choices, present in past in future in past in future, and is also passive. This passed quite unnoticed by both the speaker and the person it was addressed to; yet at the time it was being seriously questioned whether a simple verb form like has been being tested, which one can hear about once a week, could ever occur in English. Five-term tense forms are, predictably, very rare – one can in fact make a reasonable guess as to how rare, on the basis of observed frequencies of two- and three-term tense forms together with the constraints of the tense system; but they are provided for within the resources of the spoken language. Another instance I observed was they said they’d been going to’ve been paying me all this time, only the funds just kept on not coming through. Other things I noted regularly included present in present participial non-finites like being cooking in I never heard you come in – it must have been with being cooking; marked thematic elements with reprise pronoun, as in that poor child I couldn’t get him out of my mind; and relatives reaching into dependent clauses, such as that’s the noise which when you say it to a horse the horse goes faster. These are all systematic features that people are unaware that they incorporate in their speech, and often deny having said even when they are pointed out; or at least reject as unsystematic – after “I didn’t say it”, the next line of defence is “well it was a mistake”. But of course it was not a mistake; it was a regular product of the system of spoken English. But perhaps the most unexpected feature of those early observations was the complexity of some of the sentence structures. Here are two examples from recordings made at the time: (i) It’s very interesting, because it fairly soon is established when you’re meeting with somebody what kind of conversation you’re having: for example, you may know and tune in pretty quickly to the fact that you’re there as the support, perhaps, in the listening capacity – that you’re there, in fact, to help the other person sort their ideas; and therefore your remarks, in that particular type of conversation, are aimed at drawing out the other person, or in some way assisting them, by reflecting them, to draw their ideas out, and you may tune in to this, or you may be given this role and refuse it, refuse to accept it, which may again alter the nature of your conversation. (ii) The other man who kicks is the full-back, who usually receives the ball way behind the rest of his team, either near his line or when somebody’s done what the stand-off in the first example was doing, kicked over the defenders; the full-back should be able then to pick it up, and his job is 326

spoken and written modes of meaning

usually to kick for touch – nearly always for touch because he’s miles behind the rest of his side, and before he can do anything else with the ball he’s got to run up into them, before he can pass it, because he can’t pass the ball forward, and if he kicks it forward to another of his side the other man’s automatically off-side. And you get a penalty for that, do you, the other side? Depending on whether it’s kicking or passing forward. Passing forward – no, it’s a scrum. If you kick it forward and somebody else picks it up that will be a penalty. And if not, if the other side picks – If the other side picks it up that’s all right; but the trouble is this is in fact tactics again, because you don’t want to put the ball into the hands of the other side if you can avoid it because it’s the side that has possession, as in most games of course, is at an advantage. Examples such as these were noteworthy in two respects. One was that they embodied patterns of parataxis (combining with equal status) and hypotaxis (combining with unequal status) between clauses which could run to considerable length and depth. The other was that they were remarkably well formed: although the speaker seemed to be running through a maze, he did not get lost, but emerged at the end with all brackets closed and all structural promises fulfilled. And this drew attention to a third property which I found interesting: that while the listeners had absorbed these passages quite unconsciously and without effort, they were difficult to follow in writing.

3

Lexical density

These two examples have been around for a long time; so let me turn to some recent specimens taken from recordings made by Guenter Plum to whom I am indebted for drawing them to my attention. In these spontaneous narratives Plum regularly finds sequences such as the following: 1A I had to wait, I had to wait till it was born and till it got to about eight or ten weeks of age, then I bought my first dachshund, a black-and-tan bitch puppy, as they told me I should have bought a bitch puppy to start off with, because if she wasn’t a hundred percent good I could choose a top champion dog to mate her to, and then produce something that was good, which would be in my own kennel prefix. 327

construing and enacting

This displays the same kind of mobility that the earlier observations had suggested was typically associated with natural, unselfconscious speech – which is what it was. I asked myself how I would have expressed this in writing, and came up with two rewordings; the first (1B) was fairly informal, as I might have told it in a letter to a friend: 1B I had to wait till it was born and had got to about eight or ten weeks of age; that was when I bought my first dachshund, a black-and-tan bitch puppy. By all accounts I should have bought a bitch puppy at the start, because if she wasn’t a hundred percent good I could mate her with a top champion dog and produce a good offspring – which would carry my own kennel prefix. My second rewording (1C) was a more formal written variant: 1C Some eight or ten weeks after the birth saw my first acquisition of a dachshund, a black-and-tan bitch puppy. It seems that a bitch puppy would have been the appropriate initial purchase, because of the possibility of mating an imperfect specimen with a top champion dog, the improved offspring then carrying my own kennel prefix. The aim was to produce a set of related passages of text differing along one dimension, which could be recognized as going from “most likely to be spoken” to “most likely to be written”. How such variation actually correlates with difference in the medium is of course problematic; the relationship is a complicated one, both because written / spoken is not a simple dichotomy – there are many mixed and intermediate types – and because the whole space taken up by such variation is by now highly coded: in any given instance the wording used is as much the product of stylistic conventions in the language as of choices made by individual speakers and writers. Here I am simply moving along a continuum which anyone familiar with English usage can readily interpret in terms of “spoken” and “written” poles. The kind of difference that we find among these three variants is one that is often referred to as a difference of ‘texture’, and this familiar rhetorical metaphor is a very appropriate one: it is as if they were the product of a different weave, with fibres of a different yarn. But when we look behind these traditional metaphors, at the forms of language they are describing, we find that much of the difference can be accounted for as the effect of two related lexicosyntactic variables. The written version has a much higher lexical density; at the same time, it has a much simpler sentential structure. Let us examine these concepts in turn. 328

spoken and written modes of meaning Table 1 Lexical Density of Texts 1A, 1B, and 1C

1A 1B 1C

(1) Lexical items

(2) Running words

(1:2)

(3) Clauses

(1:3)

23 26 25

83 68 55

1:3.6 1:2.6 1:2.2

13 8 4

1.8:1 3.3:1 6.3:1

The lexical density is the proportion of lexical items (content words) to the total discourse. It can be measured in various ways: the ratio of lexical items either to total running words or to some higher grammatical unit, most obviously the clause; with or without weighting for relative frequency (in the language) of the lexical items themselves. Here we will ignore the relative frequency of the lexical items and refer simply to the total number in each case, providing two measures (Table 1): the number of lexical items (1) as a proportion of the number of running words, and (2) as a proportion of the number of clauses. Only non-embedded clauses have been counted (if embedded clauses are also counted, then each lexical item occurring in them is counted twice, since it figures in both the embedded and the matrix clause – i.e., both in the part, and in the whole of which it is a part). The figures are given to the nearest decimal. As Jean Ure showed (1971), the lexical density of a text is a function of its place on a register scale which she characterized as running from most active to most reflective: the nearer to the “language-in-action” end of the scale, the lower the lexical density. Since written language is characteristically reflective rather than active, in a written text the lexical density tends to be higher; and it increases as the text becomes further away from spontaneous speech. Jean Ure measured lexical density as a proportion of running words; but as is suggested by the figures given above, if it is calculated with reference to the number of clauses the discrepancy stands out more sharply. Thus in the example given above, while the number of lexical items remained fairly constant and the number of running words fell off slightly, the number of clauses fell steeply: from 13, to 8, to 4. In other words, the lexical density increases not because the number of lexical items goes up but because the number of non-lexical items – 329

construing and enacting

grammatical words – goes down; and the number of clauses goes down even more. Let us attempt a similar rewording the other way round, this time beginning with a passage of formal written English taken from Scientific American: 2A Private civil actions at law have a special significance in that they provide an outlet for efforts by independent citizens. Such actions offer a means whereby the multiple initiatives of the private citizens, individually or in groups, can be brought to bear on technology assessment, the internalization of costs and environmental protection. They constitute a channel through which the diverse interests, outlooks and moods of the general public can be given expression. The current popular concern over the environment has stimulated private civil actions of two main types. 2B is my attempt at a somewhat less “written” version; while 2C is in another step nearer to speech: 2B Private civil actions at law are especially significant because they can be brought by independent citizens, so enabling them to find an outlet for their efforts. By bringing these actions, either as individuals or in groups, private citizens can regularly take the initiative in assessing technology, internalizing costs and protecting the environment. Through the use of these actions as a channel, the general public are able to express all their various interests, their outlooks, and their moods. Because people are currently concerned about the environment, they have been bringing numerous private civil actions, which have been mainly of two types. 2C One thing is especially significant, and that is that people should be able to bring private civil actions at law, because by doing this independent citizens can become involved. By bringing these actions, whether they are acting as individuals or in groups, private citizens can keep on taking the initiative; they can help to assess technology, they can help to internalize costs, and they can help to protect the environment. The general public, who want all kinds of different things, and who think and feel in all kinds of different ways, can express all these wants and thoughts and feelings by bringing civil actions at law. At present, people are concerned about the environment; so they have been bringing quite a few private civil actions, which have been mainly of two kinds. 330

spoken and written modes of meaning Table 2 Lexical Density of Texts 2A, 2B, and 2C

2A 2B 2C

(1) Lexical items

(2) Running words

(1:2)

(3) Clauses

(1:3)

48 48 51

87 101 132

1:1.8 1:2.1 1:2.6

5 12 17

9.6:1 4.0:1 3.0:1

Table 2 shows the relative lexical density of the three variants of Text 2. Again, the number of lexical items has remained fairly constant; the variation in lexical density results from the increase in the total number of words – which means, therefore, in the number of grammatical words. This, in turn, is related to the increase in the number of clauses – where, however, the discrepancy is again much more striking.

4

Grammatical intricacy

We have characterized the difference in general terms by saying that written language has a higher lexical density than spoken language; this expresses it as a positive feature of written discourse and suggests that writing is more complex, since presumably lexical density is a form of complexity. Could we then turn the formulation around, and express the difference as a positive characteristic of spoken language? To say that spoken discourse has more words in it, or even more clauses, does not seem to convey anything very significant about it. We need to look at how the words and clauses are organized. Let us consider a shorter example of a pair of texts related in the same way, one “more written” (Text 3A), the other “more spoken” (Text 3B). I have constructed these so that they resemble the originals of Texts 1 and 2; but they are based on a natural example occurring in two texts in which a person had described the same experience twice over, once in speech and once in writing. More “written”: 3A Every previous visit had left me with a sense of the risk to others in further attempts at action on my part. More “spoken”: 3B Whenever I’d visited there before I’d end up feeling that other people might get hurt if I tried to do anything more. 331

construing and enacting

The first version (3A) is one sentence, consisting of one clause: a “simple sentence” in traditional grammar. The second version (3B) consists of four clauses (assuming that ended up feeling and tried to do are each single predicators); but these too have to be transcribed as one sentence, since they are related by hypotaxis – only one has independent status. These four clauses form what is called in systemic grammar a clause complex (for analysis and notation see Table 3): Table 3

Figure 1

The structural representation of this clause complex is given in Figure 1. The lower lexical density of Text 3B again appears clearly as a function of the number of clauses. But the significant factor is not that this text consists of four clauses where Text 3A consists of only one. It is that Text 3B consists of a clause complex consisting of four clauses. The clauses are not strung together as one simple sentence after another; they are syntactically related. Looked at from the point of view of the sentence structure, it is the spoken text that appears more complex than the written one. The spoken text has a lower degree of lexical density, but a higher degree of grammatical intricacy. 332

spoken and written modes of meaning

Figure 2

Let us return to Text 1, in its original spoken form (Text 1A). This consisted of 13 clauses. However, these 13 clauses were not strung out end to end; they were constructed into a small number of clause complexes of mixed paratactic and hypotactic construction: arguably just one clause complex throughout. Here is its interpretation as one clause complex:

Figure 3

Sequences of this kind extend to a considerable length and depth in parataxis and hypotaxis. A typical pattern is one in which both these kinds of “taxis”, or interdependency, occur, with frequent alternation both between the two and also among their various subcategories, as in the example here. The relationships between successive pairs of clauses in Text 1A are set out in Table 4. 333

construing and enacting Table 4

Other examples from the same source but from different speakers show similar patterns; there are, obviously, individual differences (including perhaps in the preference for one or other type of interdependency), but the same free-flowing intricacy is noticeable all the time, as in Texts 4–6: 4 Roy was always interested in dogs and unfortunately he’d never had the opportunity to have a dog of his own, just because of circumstances – where he lived and what not, and so I bought him a Shepherd pup, which was supposedly, you know, pure-bred Shepherd, but unfortunately people sold it because it didn’t have papers with it, so it was a ‘pup’. 5 Now how I got a German Shepherd was that I worked with a veterinary surgeon, as I’ve told you before, and there used to be a lady that brought her Shepherds along to the clinic and I used to admire them greatly, and she said, ‘Well,’ she said, ‘if you get married I’ll give you one as a wedding present,’ so immediately I bustled around looking for someone to marry so I could get a Shepherd given to me for a wedding present, you see, so that’s how that worked out well, not quite! However I got my Shepherd and he was my first dog, mainly because when I was a youngster I always wanted a dog but I lived with grandparents who wouldn’t have dogs or cats and I was a very frustrated animal lover at that stage of the game, so as soon as I got out on my own I sort of went completely berserk! 6 So we rang up the breeder, and she sort of tried to describe the dog to us, which was very hard to do over the phone, so we went over to have a look to see what they were like, and we bought Sheba, because at that stage Bob was away a lot on semitrailers with the army and it used to get quite 334

spoken and written modes of meaning

bad with the exercises – you’d have prowlers and perverts through the married quarters, so if we, you know, got a dog, which we could do because it didn’t matter what sort of dog anyone had, it’d bark and they wouldn’t bother us.

5

Types of complexity

Two distinct points need to be made here, and both of them run counter to received attitudes towards spoken language. One is that speech is not, in any general sense, ‘simpler’ than writing; if anything, it is more complex. There are, of course, many different kinds of complexity, and we have already noted one measure – lexical density – whereby speech will appear as the simpler of the two. But the patterns we have been illustrating, which are the patterns of the organization of the clause complex, referred to above as grammatical intricacy, would seem to be at least as central to any conception of complexity; and in this respect, speech appears as the more complex. The “syntactic complexity expected in writing”, with which Deborah Tannen (1982) introduces her discussion of oral and literate strategies, does not turn out to be a characteristic of written discourse. Of course, there are many other variables. Some writers achieve considerable intricacy in the structure of the clause complex; it can be learnt and consciously developed as a style. Some forms of spoken discourse, on the other hand, militate against it: rapid-fire dialogue presents no scope for lengthy interdependencies – complex semantic patterns can be construed between interactants, but usually without being realized in syntactic terms. And the categories of “written” and “spoken” are themselves highly indeterminate – they may refer to the medium in which a text was originally produced, or the medium for which it was intended, or in which it is performed in a particular instance; or not to the medium at all, but to other properties of a text which are seen as characteristic of the medium. So it is important to indicate specifically which variable of discourse is being referred to, when one variety is being said to display some distinctive characteristic. My point here is to question the assumption that written language is syntactically more complex than spoken, and to suggest that, as far as one particular kind of syntactic complexity is concerned – the intricacy (I do not want to call it “structure” because that assumes a particular interpretation) of the sentence or “clause complex” – this is more a characteristic of the most unconscious spontaneous uses of language. The more natural, un-self-monitored the discourse, the more intricate 335

construing and enacting

the grammatical patterns that can be woven. Usually, this kind of discourse will be spoken, because writing is in essence a more conscious process than speaking. But there are self-conscious modes of speech, whose output resembles what we think of as written language, and there are relatively spontaneous kinds of writing; spoken and written discourse are the outward forms that are typically associated with the critical variable, which is that of consciousness. We can use the terms spoken and written language, to refer to the idealized types defined by that variable. Spoken and written language, then, tend to display different kinds of complexity; each of them is more complex in its own way. Written language tends to be lexically dense, but grammatically simple; spoken language tends to be grammatically intricate, but lexically sparse. But these buts should really be ands, because the paired properties are complementary, not counterexpectative. It is hard to find a form of expression which will show them to be such; I have usually had recourse to metaphors of structure versus movement, saying for example that the complexity of written language is crystalline, whereas the complexity of spoken language is choreographic. The complexity of spoken language is in its flow, the dynamic mobility whereby each figure provides a context for the next one, not only defining its point of departure but also setting the conventions by reference to which it is to be interpreted. With the sentence of written language, there is solidarity among its parts such that each equally prehends and is prehended by all the others. It is a structure, and is not essentially violated by being represented synoptically, as a structural unit. With the clause complex, of spoken language, there is no such solidarity, no mutual prehension among all its parts. Its mode of being is as process, not as product. But since the study of grammar grew out of writing – it is when language comes to be written down that it becomes an object of study, not before – our grammars are grammars of the written language. We have not yet learnt to write choreographic grammars; so we look at spoken language through the lens of a grammar designed for writing. Spoken discourse thus appears as a distorted variant of written discourse, and not unnaturally it is found wanting. For example, Chafe (1982) identifies a number of regular differences between speech and writing: writing is marked by more nominalization, more genitive subjects and objects, more participles, more attributive adjectives, more conjoined, serial and sequenced phrases, more complement clauses, and more relative clauses; all of which he sum336

spoken and written modes of meaning

marizes by saying, ‘Written language tends to have an ‘integrated’ quality which contrasts with the fragmented quality of spoken language’ (p. 38). The general picture is that of written language as richly endowed, while speech is a poor man’s assemblage of shreds and patches. But Chafe has described both speech and writing using a grammar of writing; so it is inevitable that writing comes out with positive checks all round. Not that he has no pluses on the spoken side: speech is said to have more first person references, more speaker mental processes, more I means and you knows, more emphatic particles, more vagueness like sort of, and more direct quotes – all the outward signs of language as interpersonal action. Chafe summarizes them as features of “involvement” as opposed to “detachment”; but they are items of low generality, and negative rather than positive in their social value. This leads me to the second point that, as I remarked above, runs counter to our received attitudes towards speech. It is not only that speech allows for such a considerable degree of intricacy; when speakers exploit this potential, they seem very rarely to flounder or get lost in it. In the great majority of instances, expectations are met, dependencies resolved, and there are no loose ends. The intricacy of the spoken language is matched by the orderliness of spoken discourse.

6

The myth of structureless speech

Why then are we led to believe that spoken discourse is a disorganized array of featureless fragments? Here it is not just the lack of an interpretative grammar for spoken language, but the convention of observing spoken discourse that we need to take into account. Speech, we are told, is marked by hesitations, false starts, anacolutha, slips and trips of the tongue, and a formidable paraphernalia of socalled performance errors; these are regularly, more or less ritually, cited as its main distinguishing feature. There is no disputing the fact that these things occur, although they are much less prevalent than we are asked to believe. They are characteristic of the rather self-conscious, closely self-monitored speech that goes, for example, with academic seminars, where I suspect much of the observation and recording has taken place. If you are consciously planning your speech as it goes along and listening to check the outcome, then you naturally tend to lose your way: to hesitate, back up, cross out, and stumble over the words. But these things are not a particular feature of natural spontaneous discourse, which tends to be fluent, highly organized and 337

construing and enacting

grammatically well formed. If you are interacting spontaneously and without self-consciousness, then the clause complexes tend to flow smoothly without you falling down or changing direction in the middle, and neither speaker nor listener is at all aware of what is happening. I recorded this kind of casual discourse many years ago when studying the language spoken to and in the presence of a small child, and was struck by its fluency, well-formedness, and richness of grammatical pattern. Interestingly, the same feature is apparent at the phonological level: spontaneous discourse is typically more regular in its patterns of rhythm. However, while the myth of the scrappiness of speech may have arisen at the start from the kind of discourse that was first recorded, it has been perpetuated in a different way – by the conventions with which it is presented and discussed. Consider, for example, Beattie (1983: 33): Spontaneous speech is unlike written text. It contains many mistakes, sentences are usually brief and indeed the whole fabric of verbal expression is riddled with hesitations and silences. To take a very simple example: in a seminar which I recorded, an articulate (and well-known) linguist was attempting to say the following: No, I’m coming back to the judgements question. Indeterminacy appears to be rife. I don’t think it is, if one sorts out which are counterexamples to judgement. But what he actually said was: No I’m saying I’m coming back to the judgements question (267) you know there appear to (200) ah indeterminacy (1467) appears to be rife. I don’t think it is (200) if one (267) if one sorts out which are counterexamples (267) to judgement, I mean observing. Here, the brief silences (unfilled pauses) have been measured in milliseconds and marked (these are numbers in brackets) and all other types of hesitation – false starts, repetitions, filled pauses and parenthetic remarks put in italics. It is these hesitations (both filled and unfilled) which dominate spontaneous speech and give it its distinctive structure and feeling.

In other words: when you speak, you cannot destroy your earlier drafts. If we were to represent written language in a way that is comparable to such representations of spoken language, we should be including in the text every preliminary scrap of manuscript or typescript, with all the crossings out, misspellings, redraftings and periods of 338

spoken and written modes of meaning

silent thought; this would then tell us what the writer actually wrote. Figure 4 is a specimen.

Figure 4 Written discourse

Now, there are undoubtedly research purposes for which it is important to show the planning, trial and error, and revision work that has gone into the production of a piece of discourse: it can have both educational and clinical applications. This is as true of writing as it is of speech: written material of this kind has been used in neuropsychiatry 339

construing and enacting

for most of a century. But for many purposes the discarded first attempts are merely trivial; they clutter up the text, making it hard to read, and impart to it a spurious air of quaintness. What is much more serious, however, is that transcribing spoken discourse in this way gives a false account of what it is really like. It may seem a harmless piece of self-gratification for a few academics to present spoken language as a pathological phenomenon; one might argue that they deceive nobody but themselves. But unfortunately this is not the way. Just when we are seeing real collaboration between linguists and educators, and the conception of “language in education” is at last gaining ground as a field of training and research, it seems we are determined to put the clock back to a time when spoken language was not to be taken seriously and could have no place in the theory and practice of education. Let us recapitulate the argument. Speech and writing as forms of discourse are typically associated with the two modal points on the continuum from most spontaneous to most self-monitored language: spontaneous discourse is usually spoken, self-monitored discourse is usually written. We can therefore conveniently label these two modal points “spoken” and “written” language. Spoken and written language do not differ in their systematicity: each is equally highly organized, regular, and productive of coherent discourse. (This is clearly implied once we recognize them both as “language”.) Discourse in either medium can be characterized by hesitation, revision, change of direction, and other similar features; these tend to arise when attention is being paid to the process of text production. Since highly monitored discourse is typically written, these features are actually more characteristic of writing than of speech; but because most written text becomes public only in its final, edited form, the hesitations and discards are lost and the reader is shielded from seeing the process at work. Where they are likely to remain in is precisely where they occur least, in the more spontaneous kinds of writing such as personal letters. (Not all discourse features that are regarded as pathological, or assigned negative value, are of this self-monitoring kind. One form of discourse that has received a lot of critical attention is casual conversation, where the well-recognized characteristics are those of turntaking, such as interruptions and overlaps. But the strictly linguistic “deviations” of casual conversation are mainly systematic features that would not seem deviant if we had a grammar that took into account the specifically “spoken” resources of the linguistic system.) Spoken and written language do differ, however, in their preferred 340

spoken and written modes of meaning

patterns of lexicogrammatical organization. Neither is more organized than the other, but they are organized in different ways. We have already identified the principal variable. Spoken language tends to accommodate more clauses in the syntagm (to favour greater “grammatical intricacy”), with fewer lexical items in the clause. Written language tends to accommodate more lexical items in the clause (to favour greater “lexical density”), with fewer clauses in the syntagm. (This does not imply, of course, that the average number of clauses per clause complex will be greater in spoken language, because there may also be a tendency towards very short ones, especially in dialogue. It would be better to say that the greater the intricacy of a clause complex the more likely it is to be a product of spontaneous speech.) We must now return to this distinction in order to look through and beyond it.

7

A closer look at the difference

Let us illustrate with another passage of written discourse (Text 7): Thus the sympathetic induction of people into a proper and deep understanding of what Christianity is about should not be bracketed simply with the evangelizing aim to which I referred earlier. It is not absolutely incompatible with that aim, however, for the following reason. What counts as indoctrination and the like depends upon a number of criteria, to do with the degree to which a teacher fails to mention alternative beliefs, the tone of voice used, the lack of sympathy for the criticisms levelled at Christianity or Humanism and so on. A dogmatic teacher or lecturer differs from an open one. The nondogmatic teacher may be tepid; the open one may be fervent. Fervour and indifference are not functions of closedness and openness. (Smart 1968: 98) This has the high lexical density that is typical of written language: 52 lexical items, 8 clauses, density 6.5 (ignoring embedded clauses; if embedded clauses are counted, then 66 lexical items, 19 clauses, density 4.7). Let us make this explicit by setting it out clause by clause: clause complex boundary clause boundary embedded clause lexical items shown in boldface

||| || [[ ]]

||| Thus the sympathetic induction of people into a proper and deep understanding of [[what Christianity is about ]] should not be bracketed 341

construing and enacting

simply with the evangelizing aim [[to which I referred earlier]]. ||| It is not absolutely incompatible with that aim, however, for the following reason.||| [[What counts as indoctrination and the like]] depends upon a number of criteria,|| to do with the degree [[to which a teacher fails to mention alternative beliefs]] , the tone of voice [[used]], the lack of sympathy for the criticism [[levelled at Christianity or Humanism]] and so on. |||A dogmatic teacher or lecturer differs from an open one.||| The non-dogmatic teacher may be tepid;|| the open one may be fervent.||| Fervour and indifference are not functions of closedness and openness.||| To see how this lexical density is achieved, we can look at the first clause. After the cohesive thus, it begins with a nominal group the sympathetic induction of people into a proper and deep understanding of what Christianity is about. The Head is induction; the Postmodifier consists of a series of alternating embedded prepositional phrases and nominal groups, mainly one inside the other, and ending with an embedded clause: | the sympathetic induction [ of [ people ] ]] [ into [ a proper and deep understanding [ of [[ [ what ] < Christianity | is > about ]] ] ] ] | group or phrase boundary | embedded group or phrase [ ] enclosed elements <> (the prepositional phrase what . . . about is discontinuous, the items Christianity and is being enclosed within it) This nominal group contains a large amount of lexical information; and if we take this passage as a whole we find that out of the 52 lexical items the only ones that do not occur in nominal groups are bracketed, simply, depends, do, and differ. It is a characteristic of written discourse that most of the lexical information is encoded in nominal form: that is, in nominal groups, with their structure potential of Head (typically a noun or adjective), Premodifier (typically adjectives and nouns), and Postmodifier (typically embedded phrases and clauses, which then have further nominal groups inside them). Not every instance of a nominal group has a complex structure, of course; the remaining ones in this passage range from: | the lack [ of [sympathy [ for [ the criticisms [[ levelled | at [ Christianity or Humanism ] ]] ] ] ] ] | 342

spoken and written modes of meaning

which like the first one involves considerable embedding, to simple nominal groups such as tepid, the open one, fervour and indifference. But it is the potential for extended structures of this kind which enables the nominal group to take over the main burden of the lexical content of the discourse. So while spoken English is marked by intricacy in the clause complex, written English is marked by complexity in the nominal group. Since the lexical items have to go somewhere, lexical density is accompanied by its own characteristic resources within the grammar. The key factor is the structure of the nominal group; and within that, the critical resource is that of embedding, because of its open-endedness – the recursive function which generates sequences like: implicit [ in [ the argument [ about [ the necessity [ of [ the parahistorical approach [ to [ religious studies ] ] ] ] ] ] ] ] (Smart 1968: 98) If now we construct a more ‘spoken’ variant of one of the long nominal groups taken from Text 7, we might arrive at something like the following: ||| people can be sympathetically persuaded a || so that they understand properly and deeply xb a || what Christianity is about ||| b ⬘ b where the structure is xb a b ⬘ b. In place of the embedding, which is a nominalizing device, we have hypotaxis, which is a form of interdependency between clauses; and this points up the difference between the two variants. This difference is obscured, on the other hand, if the grammar fails to distinguish between embedding and hypotaxis. Traditional grammar lumped them together, under the heading of subordination, and treated them both as embedding (noun clause, adjectival clause, adverbial clause). In other words, being a grammar of written language, it recognized only the category that was characteristic of written language. This ambiguity is in fact still present in the concept of embedding, which is why I have often employed the term rankshift to refer just to embedding in the strict sense, and so distinguish it from the interdependency relation of hypotaxis, where one element is dependent on another but is not a constituent of it. Hypotaxis is more like parataxis than it is like embedding; and both are characteristic of spoken rather than written language. So in order to do justice to the particular mode of organization of both spoken and written discourse, the grammar 343

construing and enacting

needs to distinguish between the constituency relation of embedding, or rankshift, where one element is a structural part of another, and the dependency relation of ‘taxis’, where one element is bound or linked to another but is not a part of it. Either of these relations can be reduced to a form of the other one, but only at the cost of distorting the nature of discourse. The distinction between embedding and hypotaxis – between, for example, the conviction [[that he failed]] / [of failure] and was convinced || that he had failed; or between the effect [of such a decision] would be [[ that no further launchings could take place]] and if they decide that way || no further . . . – is an important one; but it is really an instance, and a symptom, of a more general and fundamental divergence. As always, when we talk about these phenomena, and when we illustrate them, they will appear as dichotomies: either this way or that. As always, however, at least in the present context (but also in most issues that have to do with language), they must be seen as tendencies – more or less continuous variation along a line, but with most actual instances (most texts, in this case) tending towards one pole or the other. The divergent tendency that is manifested in the distinction of hypotaxis and embedding is one that can be expressed in terms of the familiar opposition of process and product. Written language represents phenomena as if they were products. Spoken language represents phenomena as if they were processes (see the discussion in Martin 1984b). In other words: speaking and writing – each one makes the world look like itself. A written text is an object; so what is represented in writing tends to be given the form of an object. But when one talks, one is doing; so when one talks about something, one tends to say that it happened or was done. So, in Text 3 above, the written variant tells the story in nouns: visit, sense, risk, attempt, action; whereas the spoken version tells it in verbs: visited, ended up feeling, might get hurt, tried to do. This is to look at it from the point of view of the writer or speaker. For reader or listener, there is a corresponding difference in the way the discourse is received. To the reader, the text is presented synoptically: it exists, spread out on the page. So the reader is predisposed to take a synoptic view of what it means; behind it is a tableau – like the pictures from which writing originally evolved. But when one is listening, the text reaches one dynamically: it happens, by travelling through the air. So the listener is predisposed to take a dynamic view of what it means; behind it is a film, not a picture.

344

spoken and written modes of meaning

8

Grammatical metaphor

Where then in the linguistic system do spoken and written discourse diverge? A language, if it is not written down, consists of three interrelated subsystems: a semantic system (meanings), coded into a lexicogrammatical system (wordings), recoded into a phonological system (sounds). A language that has a writing system has an alternative form of expression: visual symbols as well as sounds. In such a language, a written text could, in principle, be a spoken text that has been written down (a transcription); here the written version is a transcoding of something that has already been coded in sound. Most writing is not like this. Secondly, a written text could be an alternative expression of a given wording: in this case meanings are coded as words and structures (“wordings”), which are then expressed either in sound or in writing. If this was the norm, there would be no systematic difference between spoken and written texts; the medium would not be a significant register variable. But there are such differences; so, to some extent at least, spoken and written discourse must represent alternative wordings. In this third case, meanings are coded either as “speakable wordings” or as “writeable wordings”, the former appropriate to the dynamic nature of the text process, the latter appropriate to the synoptic nature of the text product. This is the sort of interpretation we have been offering. But is it the whole story? There is still a fourth possibility – that speech and writing can diverge already at the semantic level, so that spoken and written discourse embody different meanings. Is there any sign that this can happen? It would of course be only a very partial effect; no one has suggested that the two derive from different semantic systems (or even two different lexicogrammatical systems, for that matter). But we should consider the possibility that there is some flowback into the meaning. Consider the last sentence of Text 2, in its original written form (2A): The current popular concern over the environment has stimulated private civil actions of two main types. We “translated” it into something more speech-like as: At present, people are concerned about the environment; so they have been bringing quite a few private civil actions, which have been mainly of two kinds. 345

construing and enacting

But this could be wrong; it may have meant: At present, people are concerned about the environment; so there have been mainly two kinds of action being brought by private citizens. There is no way of deciding: by reference to the spoken version, the written version is simply ambiguous. Compare the following, also from a written text: A further complication was the 650-ton creeper cranes poised above the end of each 825-foot arm. Does this mean: Above the end of each 825-foot arm there were poised 650-ton creeper cranes, and they made the work more complicated. or does it mean: . . . and this made the work more complicated. (i.e., not the cranes, but the fact that they were poised where they were)? Another example is: Slavish imitation of models is nowhere implied. This could be reworded either as it is nowhere implied that models have been slavishly imitated, or as . . . that models should be slavishly imitated. Examples of this kind could be added to indefinitely; they arise because nominal constructions fail to make explicit many of the semantic relations that are made explicit in clause structure. Written discourse conceals many local ambiguities of this kind, which are revealed when one attempts a more “spoken” paraphrase. But the final sentence of Text 2 illustrates another significant feature of written language, which can be seen in the wording popular concern over the environment has stimulated private civil actions. We reworded this as people are concerned about the environment, so they have been bringing private civil actions. The original is one clause with the verb stimulate representing the Process; in other words, the thesis is encoded as a single happening, and what happened was that A brought about B. But A and B are themselves nominalized processes. The meaning of stimulate here is as in pruning stimulates growth. The spoken version represents the thesis as two distinct processes, linked by a relation of cause; cf. if the tree is pruned, it will grow. Here one kind of process has been dressed up by the grammar to look like a process of a different kind – or, in this instance, two 346

spoken and written modes of meaning

processes, one mental and one material, have been dressed up as one which is neither. This coding of a semantic relation between two processes as if it was the single process is very common in writing; the sentence immediately preceding Text 2A contained another example of the same thing, here with the verb leads to: A successful tort action leads to a judgment of damages or an injunction against the defendant company. But this is just one type of a more general phenomenon, something that I call grammatical metaphor (Halliday 1985, Chapter 10). Written language tends to display a high degree of grammatical metaphor, and this is perhaps its single most distinctive characteristic. Here are three further examples of grammatical metaphor taken from various written sources, together with suggested rewordings which are less metaphorical: Issue of the specially-coded credit cards will be subject to normal credit checking procedures. “Credit cards have been specially coded and will be issued only when credit has been checked in the normal way.” Strong Christmas sales were vital to the health of the retail industry, particularly in the present depressed climate. “Unless many goods were sold at Christmas the retail industry would not be healthy, particularly when the economy is depressed as it is now.” He also credits his former big size with much of his career success. “He also believes that he was successful in his career mainly because he used to be big.” In all these examples nominalization plays a significant part, as it does in many types of grammatical metaphor; so it is perhaps worth stressing that nominalization is well motivated in English. It is not simply a ritual feature that has evolved to make written language more ambiguous or obscure; like the passive, which is another feature whose functions are widely misunderstood, nominalization is an important resource for organizing information. Take the example youth protest mounted, which is not a headline but a complete sentence from a feature article. We might reword this as more and more young people protested, or young people protested more and more; but the only way to get the combination of youth and protest as the Theme of the clause is by means 347

construing and enacting

of a nominalization (not necessarily such a laconic one; it might have been the protests of the young people, but this is still a nominalizing of the process). So while there is a price to be paid, in that the information being conveyed may become mildly (and sometimes severely) ambiguous, there is also a payoff: more choice of status in the discourse. In terms of systemic theory, there is a loss of ideational information, but a gain in textual information. This of course favours the specialist: you need to know the register. If you do not know the register you may misinterpret the thesis, so the fact that it is highly coded as a message is not very helpful to you; but if you do know it you will select the right interpretation automatically, and the additional “functional sentence perspective” is all tax-free profit. Some nominalizations of course cannot be denominalized, like private civil actions at law or an injunction against the defendant company. These are abstractions that can enter into the structure of a clause – civil actions can be brought, an injunction can be issued – but cannot themselves be coded as finite verbs. Much of our environment today consists of such abstract entities and institutions; their representation in nominal form is no longer metaphorical – if it ever was – and they have become part of our ideology, our way of knowing about the world we live in. Patterns of this kind invade the spoken language and then act as infiltrators, providing cover for other metaphorical nominalizations – which are still functional in speech, but considerably less so, because spoken language has other resources for structuring the message, such as intonation and rhythm. Grammatical metaphor is not confined to written language: quite apart from its tendency to be borrowed from speech into writing, there are specific instances of it which seem clearly to have originated in speech – most notably the pattern of lexically empty verb with the process expressed as “cognate object” (Range) as in make a mistake ‘err’, have a bath ‘bathe’, give a smile ‘smile’. But in its principal manifestations it is typically a feature of writing. Writing – that is, using the written medium – puts distance between the act of meaning and its counterpart in the real world; so writing – that is, the written language – achieves this distance symbolically by the use of grammatical metaphor. It is often said that written discourse is not dependent on its environment; but it would be more accurate to say that it creates an environment for itself (see Nystrand 1987), and this is where it depends on its metaphorical quality. If I say technology has improved, this is presented as a message; it is part of what I am telling you. If I say improvements in technology, I present it as something I expect you to take for granted. 348

spoken and written modes of meaning

By objectifying it, treating it as if it was a thing, I have backgrounded it; the message is contained in what follows (e.g., . . . are speeding up the writing of business programmes). Grammatical metaphor performs for the written language a function that is the opposite of foregrounding; it backgrounds, using discourse to create the context for itself. This is why in the world of writing it often happens that all the ideational content is objectified, as background, and the only traces of process are the relations that are set up between these taken-for-granted objects. I recall a sentence from the O.S.T.I. Programme in the Linguistic Properties of Scientific English (Huddleston, Hudson, Winter and Henrici 1968) which used to typify for us the structures found in scientific writing: The conversion of hydrogen to helium in the interiors of stars is the source of energy for their immense output of light and heat.

9

Ways of knowing and learning

In calling the written mode metaphorical we are of course making an assumption; in fact each mode is metaphorical from the standpoint of the other, and the fact that the spoken is developmentally prior – the individual listens and speaks before he reads and writes – while it means that the language of “process” is learnt first, does not guarantee that it is in any sense “closer to reality”. It might be a hangover from an earlier stage of evolution, like the protolanguage that precedes the mother tongue. But personally I do not think so. I am inclined to think the written language of the future will go back (or rather forward) to being more processlike; not only because the traditional objectlike nature of written discourse is itself changing – our reading matter is typed into a memory and fed to us in a continuous flow as the lines follow each other up the screen – but also because our understanding of the physical world has been moving in that direction, ever since Einstein substituted space-time for space and time. As Bertrand Russell expounded it (1925: 54), We are concerned with events, rather than with bodies. In the old theory, it was possible to consider a number of bodies all at the same instant, and since the time was the same for all of them it could be ignored. But now we cannot do that if we are to obtain an objective account of physical occurrences. We must mention the date at which a body is to be considered, and thus we arrive at an ‘event’, that is to say, something which happens at a given time. 349

construing and enacting

Meanwhile, grammatical analysis shows spoken and written English to be systematically distinct: distinct, that is, in respect of a number of related tendencies, all of which combine to form a single package. But it turns out to be a semantic package: the different features that combine to distinguish spoken and written discourse can be shown to be related and encompassed within a single generalization, only when we express this generalization in semantic terms – or at least in terms of a functional, meaning-oriented interpretation of grammar. Speech and writing will appear, then, as different ways of meaning: speech as spun out, flowing, choreographic, oriented towards events (doing, happening, sensing, saying, being), processlike, intricate, with meanings related serially; writing as dense, structured, crystalline, oriented towards things (entities, objectified processes), productlike, tight, with meanings related as components. In their discussion of the comprehension and memory of discourse, Hildyard and Olson (1982: 20) suggested that meaning is preserved in different ways by speakers and listeners: Readers and listeners may tend to extract different kinds of information from oral and written statements. Listeners may tend to recall more of the gist of the story and readers may recall more of the surface structure or verbatim features of the story.

In other words, the listener processes text largely at the level of meaning, the reader more, or at least as much, at the level of wording. But this is specifically a function of the medium in which the text is received, rather than of the linguistic features of the code that lies behind it. The notion of different ways of meaning implies, rather, that there are different ways of knowing, and of learning. Spoken and written language serve as complementary resources for acquiring and organizing knowledge; hence they have different places in the educational process. Teachers often know, by a combination of intuition and experience, that some things are more effectively learnt through talk and others through writing. Official policy usually equates educational knowledge with the written mode and commonsense knowledge with the spoken; but teachers’ actual practice goes deeper – educational knowledge demands both, the two often relating to different aspects of the same phenomenon. For example: definitions, and structural relations, are probably best presented in writing; demonstrations of how things work may be more easily followed through speech. The two favourite strategies for describing the layout of an apartment, reported in the well-known study by Linde and Labov (1975), would seem to 350

spoken and written modes of meaning

exemplify spoken and written modes of symbolic exploration. We may assume that speech and writing play different and complementary parts in the construction of ideologies (Hasan 1986), since each offers a different way of knowing and of reflecting on experience. Considerations of this kind are an essential element in any linguistic theory of learning. The development of such a theory is perhaps the most urgent task of educational linguistics; and certain components of it can already be recognized: (1) the child’s construction of language, from presymbolic communication through protolanguage to the mother tongue; (2) the processing of new meanings into the system; (3) the interaction between learning elements that are ready coded and learning the principles of coding; (4) the relation between system and process in language; (5) the unconscious nature of linguistic categories; (6) the social construction of reality through conversation; (7) linguistic strategies used in learning; (8) the development of functional variation, or registers; (9) the relation between everyday language and technical language; and (10) the development of generalization, abstraction, and metaphor. The absence of any general theory of learning based on language has been a significant gap in educational thinking and practice. This provides an important context for our current concern, since the complementarity of spoken and written language will certainly be a central issue in any learning theory which has language as its primary focus.

351

Chapter Thirteen

HOW DO YOU MEAN? (1992)

I realize that the title might well prompt someone to ask, ‘How do you mean, “How do you mean?”?’ I could have written, ‘How are meanings made?’ – although I prefer the more personalized version. The question is meant theoretically; but, like so many theoretical questions, it becomes relevant in practice the moment we want to intervene in the processes we are trying to understand. And some processes of meaning are involved in more or less everything we do. I shall need to talk about two fundamental relationships, those of realization and instantiation; so let me begin by distinguishing these two. Instantiation I take to be the move between the system and the instance; it is an intrastratal relationship – that is, it does not involve a move between strata. The wording fine words butter no parsnips is an instance, or an instantiation, of a clause. Realization, on the other hand, is prototypically an interstratal relationship; meanings are realized as wordings, wordings realized as sound (or soundings). We often use the term to refer to any move which constitutes a link in the realizational chain, even one that does not by itself cross a stratal boundary (for example, features realized as structures); but the phenomenon of realization only exists as a property of a stratified system. To anticipate the discussion a little, I shall assume that realization may be formalized as metaredundancy, as this is defined by Jay Lemke (1985). Instantiation I shall define by making reference to the observer; it is variation in the observer’s time depth. Firth’s concept of exponence is the product of these two relations: his “exponent” is both instantiation and realization.1 First published in Advances in Systemic Linguistics: Recent Theory and Practice, 1992, edited by Martin Davies and Louise Ravelli. London: Pinter, pp. 20–35.

352

how do you mean?

form of consciousness action

reflection

1st/2nd person

regulatory

interactional

3rd person

instrumental

personal

domain of experience

Figure 1 The protolanguage ‘microfunctions’

I shall take it that meaning is not a uniquely human activity; rather, it is part of the experience of at least some other species, obviously including the so-called “higher” mammals. In humans, meaning develops, in the individual, before the stage of language proper; it begins with what I have called “protolanguage”. So where does this mammalian experience come from? It probably evolved out of the contradiction between the two primary modes of experience, the material and the conscious. Material processes are experienced as ‘out there’; conscious processes are experienced as ‘in here’. We can see in observing the growth of an individual child how he or she construes this contradiction in the form of meaning. The child constructs a sign, whereby the one mode of experience is projected on to the other. In my own observations this took the form of what I coded as “v.h.p.s.” (very high-pitched squeak), Nigel’s first sign that he produced at five months old; I glossed it as “what’s that? – that’s interesting”. In other words, Nigel was beginning to construe conceptual order out of perceptual chaos: ‘I am curious (conscious) about what’s going on (material)’. This impact of the material and the conscious is being transformed into meaning by a process of projection, in which the conscious is the projecting and the material the projected.2 But there are two possible modes of such projection – two forms that the consciousness may take: one, that of reflection, ‘I think’, and the other that of action, ‘I want’ – one the way things are, and the other the way they ought to be. There are also, as it happens, two domains of the material experience: one, that of ‘you and me’, and the other that of ‘the rest (it, them)’. So, once the process begins (at around eight months, with Nigel), what is construed into meaning is not a single sign but a two-dimensional semiotic space constituting a sign system (Figure 1). We can justifiably refer to such a sign system as a 353

construing and enacting

“language”; but since it lacks the essential properties of an adult language, I preferred to label it more specifically as a protolanguage. The four quadrants of the space I referred to as “microfunctions”.3 In these terms, then, the microfunctional meanings of the protolanguage evolve through the projecting of the material on to the conscious, in a single twodimensional construction. And this becomes possible because the conscious mode of experience is the social mode. We have often pointed out that it takes two to mean; but we still tend to refer to consciousness as if it was an individual phenomenon, with the social as an add-on feature. I would prefer the Vygotskyan perspective, whereby consciousness is itself a social mode of being. In the act of meaning, then, the two modes of experience, through the projection of the one by the other, become fused and transformed into something that is new and different from either. We can think of this as creating a “plane of content” in the Hjelmslevian sense. If we look at this process dynamically, it is meaning-creating, or semogenic. If we look at it synoptically, as a relation construed by this process, it is semantic; and it appears as an interface (our original notion of semantics as “interlevel” was relevant here),4 one ‘face’ being the phenomena of experience. We often refer to these phenomena collectively as “the material”, as if the only form of experience was what is ‘out there’. But this is misleading. Our experience is at once both material and conscious; and it is the contradiction between the material and the conscious that gives these phenomena their semogenic potential. The other ‘face’ is the meaning – the signified, if you prefer the terminology of the sign. Many years ago I did my best to gloss the child’s protolinguistic meanings using “adult” language as metalanguage, and found myself forced into using glosses like ‘nice to seeyou, and  let’s look at this picture together’ for Nigel’s protolinguistic [ [ dɔ` [ dɔ` [ dɔ`]. This was a way of identifying these signs; I then interpreted them in terms of the microfunctional categories just referred to. But those categories themselves were not interpreted further. I think they can now be explained at this somewhat deeper level, as the intersection of the two modes of projection with the two domains of experience. But in order for meaning to be created there has also to be a second interface, a transformation back into the material, or (again, rather) into the phenomenal – this time in its manifestation in the meaning subject’s own body: as physiological processes of articulation or gesture. This is the phonetic / kinetic interface; the “expression plane”, in Hjelmslev’s terms. Since there can be no meaning without expression (meaning is intersubjective activity, not subjective), the act is “doubly articulated”, 354

how do you mean?

in Martinet’s terminology: it is the transduction of the phenomenal back into the phenomenal via these two interfaces of content and expression. (Transduction not transformation, because as Lamb (1964) pointed out many years ago in transformation the original is lost, ceases to exist. And again I am suggesting that we should conceive of it as phenomenal rather than material, since both the ‘outer’ faces, that of the content substance on the one hand and that of the expression substance on the other, embody both the material and the conscious modes of being.) What is construed in this way, by this total semogenic process, is an elastic space defined by the two dimensions given above: the ‘inner’ dimension of reflective / active, ‘I think’ as against ‘I want’, and the ‘outer’ dimension of intersubjective / objective, ‘you and me’ as against ‘he, she, it’. (Again, there is a naming problem here; we could say that the ‘out there’ dimension is that of person / object, provided we remember that “object” includes those persons ‘treated as’ object, i.e. third persons. Instantially, this means any person other than whoever is the interlocutor at the time; systemically it means any person not forming part of the subject’s (the child’s) meaning group.) This two-dimensional ‘elastic space’ defines what I have called the mammalian experience. Obviously I am begging lots of questions by calling it mammalian; but I am using this as a way of saying that it is a potential we hold in common with other creatures, which I think is rather important. It is a rich semogenic potential; but it is also constrained in certain critical respects. In our own specifically human history, in both phylogenetic and ontogenetic time, it comes to be deconstructed – or rather deconstrued – and reconstrued as something else, this time in the form of a potential for meaning that is effectively infinite, or at least unbounded (to use an analogue rather than a digital mode of expression). This reconstrual is the explosion into grammar. If we keep to the ‘interface’ conception, it is the evolution of an interface between the interfaces. If we put it in terms of even more concrete metaphors, what happens is that an entirely non-material (again, better: non-phenomenal) system is slotted in between the two material / non-material (phenomenal / non-phenomenal) systems that are already in place. By means of this critical step, protolanguage evolved into language. This step of reconstrual could not be taken with an inventory of single signs, but only with a sign system – a semiotic that is already (two-)dimensional. It operates not on the terms but on the oppositions, the paradigms that we have been able to identify as reflection / action and person / object (or intersubjective / objective). By ‘grammatical355

construing and enacting

izing’ the process of meaning – reconstruing it so that the symbolic organization is freed from direct dependence on the phenomenal, and can develop a structure of its own – the collective human consciousness created a semiotic space which is truly elastic, in that it can expand into any number of dimensions. (We will model this more explicitly in a moment.) The immediate effect is to re-form the reflection / action opposition into a simultaneity, such that all acts of meaning embody both – i.e. both reflection and action – not just as components, but as sets of options, each constituting a distinct dimension of choice. In other words they now evolve into the metafunctional categories of ideational and interpersonal. The ideational is the dimension that is primarily reflective (the construction of experience), the interpersonal that which is primarily active (the enactment of social processes5). (But note that each engenders the other mode as a secondary motif: the way we construe experience (by verbal reflection) disposes us to act in certain ways, e.g. as teachers structuring the role relationships in the learning process, while the way we construct our social relations (by verbal action) enables us to represent – to verbalize – what the resulting social order is like.) What has made this possible is what I called just now the ‘explosion into grammar’ – an explosion that bursts apart the two facets of the protolinguistic sign. The result is a semiotic of a new kind: a stratified, tristratal system in which meaning is ‘twice cooked’, thus incorporating a stratum of ‘pure’ content form. It is natural to represent this, as I have usually done myself, as ‘meaning realized by wording, which is in turn realized by sound’. But it is also rather seriously misleading. If we follow Lemke’s lead, interpreting language as a dynamic open system, we can arrive at a theoretically more accurate and more powerful account. Here the key concept is Lemke’s principle of metaredundancy.6 Consider a minimal semiotic system, such as a protolanguage – a system that is made up of simple signs. This is based on the principle of redundancy. When we say that contents p, q, r are “realized” respectively by expressions a, b, c, what this means is that there is a redundancy relation between them: given meaning p, we can predict sound or gesture a, and given sound or gesture a we can predict meaning p. This relationship is symmetrical; “redounds with” is equivalent both to “realizes” and to “is realized by”. Let us now expand this into a non-minimal semiotic, one that is trirather than bi-stratal. The expressions a, b, c now realize wordings l, m, n while the wordings l, m, n realize meanings p, q, r. In terms of redundancy, however, these are not two separate dyadic relationships. 356

how do you mean?

Rather, there is a metaredundancy such that p, q, r redounds not with l, m, n but with the redundancy of l, m, n with a, b, c; thus: l, m, n
a, b, c

p, q, r
(l, m, n
a, b, c)

Why has it to be like this? Because there is not, in fact, a chain of dyadic relationships running through the system. (If there was, we would not need the extra stratum.) It is not the case, in other words, that p
l and l
a, p, q, r is realized by l, m, n; but the system at l, m, n is sorted out again for realization by a, b, c, so that what p, q, r is actually realized by is the realization of l, m, n by a, b, c. This is the fundamental distinction between redundancy and causality. If realization was a causal relation, then it would chain: l is caused by a and p is caused by l – it would make no sense to say “p is caused by the causing of l by a”. But realization is not a causal relation; it is a redundancy relation, so that p redounds with the redundancy of l with a. To put it in more familiar terms, it is not that (i) meaning is realized by wording and wording is realized by sound, but that (ii) meaning is realized by the realization of wording in sound. We can of course reverse the direction, and say that sounding realizes the realization of meaning in wording: p, q, r
l, m, n

(p, q, r
l, m, n)
a, b, c

For the purpose of phonological theory this is in fact the appropriate perspective. But for the purposes of construing the ‘higher’ levels, with language as connotative semiotic realizing other semiotic systems of the culture, we need the first perspective. Thus when we extend ‘upwards’ to the context of situation, we can say that the context of situation s, t, u redounds with the redundancy of the discourse semantics p, q, r with the redundancy of the lexicogrammar l, m, n with the phonology a, b, c. Thus: s, t, u
(p, q, r
(l, m, n
a, b, c)) (cf. Figure 2). Once the original protolinguistic redundancy has been transformed into metaredundancy in this way, the relation becomes an iterative one and so opens up the possibilities for construing, not only the context of situation, but also higher levels such as Hasan’s symbolic articulation and theme in verbal art, or Martin’s strata of genre and ideology. The metaredundancy notion thus formalizes the stratal principle in semogenesis. What makes meaning indefinitely extendable is the evolutionary change from protolanguage to language – whereby instead of 357

construing and enacting

Figure 2 Metaredundancy

a simple plane with two interfaces to the material (the phenomenal), we have constructed a semiotic space, a three-dimensional (potentially n-dimensional) system in which there is a purely symbolic mode of being between these two interfaces. It is this that we call grammar, or more explicitly lexicogrammar. Without this semiotic space, situated in the transduction from one purely symbolic mode to another, and hence not constrained by the need to interface directly with the phenomenal, we could not have a metafunctional organization in the grammar, and we could not have the phenomenon of grammatical metaphor.7 The metaredundancy theory explains the ‘stratal’ organization of language, and the semiotic principle of realization. It explains them synoptically: by treating realization as a relation. Now, a system of this kind could still remain fully closed: it could be a circular, self-regulating system without any form of exchange with its environment. But a language, as Lemke pointed out, is a dynamic open system; such systems are not autostable, but metastable – they persist only through constantly changing by interpenetration with their environment. And in order to explain a system of this kind we have to complement our synoptic 358

how do you mean?

interpretation with a dynamic one. This leads us into the other critical concept, that of instantiation. Consider the notion of climate. A climate is a reasonably stable system; there are kinds of climate, such as tropical and polar, and these persist, and they differ in systematic ways. Yet we are all very concerned about changes in the climate, and the consequences of global warming. What does it mean to say the climate is changing? Climate is instantiated in the form of weather: today’s temperature, humidity, direction and speed of wind, etc., in central Scotland are instances of climatic phenomena. As such they may be more, or less, typical: today’s maximum is so many degrees higher, or lower, than average – meaning the average at this place, at this time of year and at this time of day. The average is a statement of the probabilities: there is a 70 per cent chance, let us say, that the temperature will fall within such a range. The probability is a feature of the system (the climate); but it is no more, and no less, than the pattern set up by the instances (the weather), and each instance, no matter how minutely, perturbs these probabilities and so changes the system (or else keeps it as it is, which is just the limiting case of changing it). The climate and the weather are not two different phenomena. They are the same phenomenon seen by two different observers, standing at different distances – different time depths. To the climate observer, the weather looks like random unpredictable ripples; to the weather observer, the climate is a vague and unreal outline. So it is also with language;8 language as system, and language as instance. They are not two different phenomena; they are the same phenomenon as seen by different observers. The system is the pattern formed by the instances; and each instance represents an exchange with the environment – an incursion into the system in which every level of language is involved. The system is permeable because each instance redounds with the context of situation, and so perturbs the system in interaction with the environment. Thus both realization and instantiation are involved in the evolution of language as a dynamic open system. Now the relation of system to instance is in fact a cline, a continuous zoom; and wherever we focus the zoom we can take a look into history. But to know what kind of history, we have to keep a record of which end we started from. To the system observer, history takes the form of evolution; the system changes by evolving, with selection (in the sense of ‘natural selection’) by the material conditions of the environment. This is seen most clearly, perhaps, in the evolution of particular sub359

construing and enacting

Figure 3 A model of semogenesis

systems, or registers, where features that are functionally well adapted are positively selected for; but it appears also in the history of the system as a whole once we look beyond the superficial clutter of random fluctuations into the grammar’s cryptotypic core. To the instance observer, on the other hand, history is individuation: each text has its own history, and its unique meaning unfolds progressively from the beginning. (Note that the probability of any instance is conditioned both systemically (a register is a resetting of the overall probabilities of the system) and instantially, by the transitional probabilities of the text as a Markoff chain.) Given any particular feature – say grammatical metaphor – we may be able to track it through both these histories, the phylogenetic – its history as it evolves in the system; and what we might call the “logogenetic” – its history as it is built up in the course of the text. There is of course a third kind of history, the ontogenetic, which is different again – the cladistic model here is one of growth. This too is a mode of semogenesis; and we could follow through with the same example, asking how grammatical metaphor comes into being in the developmental history of a child. These are in fact the three modes or dimensions of semohistory – the phylogenetic, the ontogenetic and the 360

how do you mean?

Figure 4 Postulated examples of semogenic evolution in relation to some systems of Modern English. (Note. Those on the right are labelled merely for identification, not in terms of their systemic features in the grammar.)

logogenetic; in the dynamic perspective, we can ask: how did this meaning evolve, in the system? how did it develop, in the learner? and how did it unfold, in the text? In all these histories, the meaning potential typically tends to increase. (Where it decreases, this is generally catastrophic: the language dies out, 361

construing and enacting

Figure 5 Microfunctional systems (Nigel’s protolanguage)

or is creolized; the individual dies, or becomes aphasic; the text comes to an end, or is interrupted.) Now, the mechanism of this increase of meaning potential may be modelled in the most general terms as in Figure 3. Nesbitt and Plum (1988; see also Halliday 1991) showed how to do this in a corpus-based study of ‘direct speech and indirect thought’ (the intersection of speech projection and thought projection with the interdependency system of parataxis and hypotaxis). This and other postulated examples of evolutionary semogenesis are set out in Figure 4. These involve relations between strata (semantics realized in grammar); and they suggest how metaredundancy becomes dynamic – through shifting probabilities, as the values change instance by instance. In other words the permeability of the system depends on the metaredundancy relation: this is the only way it can be nudged along. Thus where a closed system is self-regulating (autostable) and circular, an open system is other-regulated (metastable) and helical. And it is through the combination of these two relations or processes, instantiation on the one hand and realization on the other, that the system exchanges with its environment, creating order in the course of this exchange and so increasing its potential for meaning. Thus the possibility of meaning – of acting semiotically – arises at the intersection of the material (or phenomenal) with the conscious, as the members of a species learn to construct themselves (“society”) in action and to construe their experience in reflection. These two dimen362

how do you mean?

Figure 6 Nigel’s first stratifications (age one year two months)

sions – action / reflection, and you + me / other – define a semiotic field. At first this is a plane, a rubber sheet so to speak, elastic but twodimensional (this is the protolanguage phase); having just two surfaces, interfaces between the conscious and the two facets of the material (“content purport” and “expression purport”), such that meaning consists in making the transduction between them. The simple signs of the protolanguage shape themselves into a sign system as they cluster in the four quadrants of this semiotic plane; there is thus already a protosystem network, which we could set up in an idealized form as in Figure 5. (I use the variables that emerged from my own Nigel data; this should be compared with studies by Clare Painter (1984) and by Jane Oldenburg (1987), where other systemic variables may appear more prominent.9) We must leave open the question of what variables are the ones in respect of which the protolinguistic system is typically construed, but I think it will be a fairly small set. Nigel’s seemed to be (1) in instrumental: polarity, (2) in regulatory: intensity, (3) in interactional: mode of being, or process type, relational / behavioural; (4) in personal: mode of consciousness, cognitive / affective. This two-dimensional plane is then deconstrued and evolves into an n– dimensional space, as the activity of meaning becomes dialogically dynamic and metafunctionally complex: that is, it becomes possible to mean more than one thing at once, and to construe meanings into text. I have written elsewhere about how Nigel took the first step in this transformation, using the semogenic strategy already described (combining two functionally distinct variables); it is summarized in Figure 6. 363

construing and enacting

Here for the first time Nigel is selecting in two systems of meaning at once, and by this token the initial move into grammar has been made. Through the second year of life this new stratified system will gradually replace the protolinguistic one, and all meanings (except for a few protolanguage remnants that persist into adult life like hi! and ah! and yum! and ouch!) will come to be stratally and metafunctionally complex. So in Nigel’s first exemplar, just cited, we have (i) proto-metafunctions (proto-ideational – different persons; proto-interpersonal – seeking / finding), and (ii) proto-strata, with the meaning ‘first’ construed as wording (the ideational as contrasting names; the interpersonal as contrasting mood) and ‘then’ (re)construed as sounding (names as articulation, mood as intonation). At this second interface the child can now combine the segmental and the prosodic choices, in this way both realizing and also iconically symbolizing the two different modes of meaning that are combined at the first interface. The resources for making meaning are now in place. It is probably not a coincidence that, as the ideational grammar evolved, so in the system of transitivity the field of processes was construed into different process types along precisely the lines that (if my understanding is right) went into the making of meaning in the first place. If meaning arises out of the impact of the conscious and the material, as mutually contradictory forms of experience, then it is not surprising that when experience is construed semantically, these two types of process, the material and the conscious, should come to be systematically distinguished. But there is a further twist. The semogenic process, as we saw, involves setting up a relationship between systems such that one is the realization of the other – that is, they stand to each other in a relation of Token and Value. This Token–Value relationship is set up at both interfaces, and it is also what makes it possible to prise the two apart and wedge in a grammar in between. Here then we find the third of the kinds of process construed by the grammar: the relational process, based on identifying a Token with a Value. The grammar of natural language, in its ideational metafunction, is a theory of human experience; thus it may reasonably be expected to take as its point of departure the very set of contrasts from which its own potential is ultimately derived. Let me return once again, finally, to the suggestion that meaning is a mode of action engendered at the intersection of the material (or phenomenal) and the conscious, as complementary modes of experience. Now, the effect of this impact is to construe order. By the act of meaning, consciousness imposes order on the phenomena of experience. 364

how do you mean?

When protolanguage evolves into language, with the stratal dimension of realization, meaning becomes self-reflexive: and in two senses. On the one hand, it imposes order on itself: the textual metafunction, as Christian Matthiessen (1992) has shown, construes a reality that is made of meaning. On the other hand, we can talk about the way we mean, and examine the nature of the order our way of meaning has imposed. As well as a grammar, a theory of experience, we have a grammatics – a grammar of grammars, a theory of theories of experience, or a metatheory in one sense of this term. At this very general level, we can then examine our own notions of order. Experience now appears as an interplay of order and disorder: analogy and anomaly, in the terms of the ancient Greek debate (begun, like so many other issues in the history of ideas, with arguments about language – in this case arising from regular and irregular morphological patterns) – or order and chaos, in current terminology. Is chaos a feature of the phenomena themselves, or merely a product of the deficiency in our understanding? Are the two merely a function of the observer, so that patterns repeat if we wait for them long enough, probabilities become certainties when we know all that needs to be known? Or to put this in more specifically linguistic terms, will all the various contradictions in the grammar resolve themselves into some higher level of order? – I mean things like transitive and ergative as complementary theories of process, or tense and aspect as complementary theories of time; as well as all the other indeterminacies which arise in our polyelastic semiotic space? I am not of course setting out to answer these questions; I am merely pointing out that the meaning potential we have evolved for ourselves construes the possibility of asking them. But I will allow myself one further thought in the closing paragraphs of the paper. It is a human failing that we usually try to impose order much too soon. There are many examples of this in recent linguistics.10 The attempt fails; and we then resort to ‘theories of chaos’, trying to make sense of things while remaining instance observers – looking for ‘une the´orie de la parole’, so to speak.11 Such constructs are ultimately selfcontradictory; but they serve as a way of reformulating the questions and allow us to move back a bit, to shift our stance. A good example of the overimposition of order through language is provided by a designed, or semi-designed, system like the language of science. Having construed a reality that is technological (in the true sense of this term: a reality constructed not out of techne but out of the logos, or discourse, of techne), scientists themselves are now finding their language – that is, their own scientific metalanguages – too rigid and determinate, and are 365

construing and enacting

seeking ways to restore the balance: a discourse with which to construe experience in terms of indeterminacy, of continuity and of flux. Now by comparison with the languages of science, the ordinary evolved language of everyday life has many of these properties. It is oriented towards events rather than objects, and is in many respects fluid and indeterminate. But it is important to be aware that even our most unconscious everyday language also imposes an order which we may need to re-examine and to deconstruct. To return to the weather: we can say it’s raining or we can say it’s snowing – but we have to decide between them. We may accommodate an intermediate form, it’s sleeting, this cuts up the continuum more finely, but still into discrete parts – ‘rain or sleet or snow’.12 (Contrast in this respect the semantics of sign, such as Auslan, which often allows a more continuous interpretation of experience – though of course it is constantly being modified under the influence of spoken language (Johnston 1990).) In other words, while the order – that is, the particular mix of order and chaos – that our grammar construes has served us well, and continues to do so, it is not necessarily the most functional for all times and all circumstances; especially at times of rapid change like the present, we may need to hold it up to the light and see how it works.13 It is easy to remain unaware of the stories our grammar is telling us. One thing I have been trying to do, in this paper, is to use the grammar to think with about itself. Not just in the usual sense, of using language as its own metalanguage; of course I am doing that, because there is nothing else I can do. I mean this more specifically in the sense of using what I have called the grammatics – the concepts that we have developed in order to interpret the grammar – as a means towards understanding the nature and evolution of language as a whole. The strategy is that of treating language as ‘other’ – as if it was a different kind of semiotic that the grammar was being used to explore.14 Thus I have found it helpful to think of meaning as the way consciousness (that is, mental processes), by a type of projection, construes a relationship (that is, a Token = Value identity, or a nested series of such identities) between two sets of material processes (those of our experience, at one end, and those of our bodily performance – gesture, articulation – at the other). I do not know how useful anyone else will find this strategy. But at least it is something I can answer with, the next time anyone says to me, “How do you mean?”

366

how do you mean?

Notes 1. For Firth’s concept of ‘exponence’ see especially his ‘Synopsis of Linguistic Theory’ in Firth (1957a). 2. I am speaking here of phylogenesis; but the process is recapitulated in the growth of the individual, where it can be observed in the form of behaviour. A child experiences certain phenomena as ‘out there’ – as lying beyond the boundary between ‘me’ and ‘non-me’: some perturbation seen or heard, like a flock of birds taking off, or a bus going past, or a coloured light flashing. At the same time, he also experiences a phenomenon that is ‘in here’: his own consciousness of being curious, or pleased, or frightened. At first these two experiences remain detached; but then (perhaps as a result of his success in grasping an object that is in his line of sight – in Trevarthen’s terms, when “pre-reaching” becomes reaching, typically at about four months) a spark flies between them by which the material is projected on to the conscious as ‘I’m curious about that’, ‘I like that’ and so on. Now, more or less from birth the child has been able to address others and to recognize that he is being addressed (Catherine Bateson’s “proto-conversation”). The projection of the material on to the conscious mode of experience maps readily on to this ability to address an other; and the result is an act of meaning – such as Nigel’s very highpitched squeak, which he first produced at five months, shortly after he had learnt to reach and grasp. 3. Other microfunctions were added as the protolanguage evolved by degrees into the mother tongue; but these were the original four. See Halliday (1975, 1978). 4. At first labelled, somewhat misleadingly, the level of “context”. See the discussion of levels in Halliday, McIntosh and Strevens (1964). See also Ellis (1966). 5. Based on giving and demanding – that is, on exchange. Initially this meant the exchange of goods-and-services; but eventually, by a remarkable dialectic in which the medium of exchange became itself the commodity exchanged, it extended to giving and demanding information. By this step, meaning evolved from being an ancillary of other activities to being a form of activity in its own right. 6. See the chapters entitled ‘Towards a model of the instructional process’, ‘The formal analysis of instruction’ and ‘Action, context and meaning’ in Lemke (1984). 7. It is impossible to have metaphor in a protolanguage at all, unless one chooses to call “metaphor” (or perhaps “proto-metaphor”) what is taking place when, for example, Nigel transfers a particular sign [gωg gωg gωg] from ‘I’m sleepy’ to ‘let’s pretend I’m going to sleep’. See Halliday (1975: Chapter 2). 8. The analogy should not, of course, be pressed too far. Specifically, while

367

construing and enacting

9. 10. 11.

12.

13. 14.

the relation of instantiation holds both for language / speech (langue / parole) and for climate / weather, that of realization does not. It could be said that climate is in fact modelled as a stratified system (in the semiotic, not the atmospheric sense!); but this would be using ‘stratified’ with a significantly different meaning. See Painter (1984), Oldenburg (1987). For an investigation of Chinesespeaking children see Qiu (1985). As pointed out by John Sinclair (1992). Note that current “chaos theory”, as in Gleick’s book Chaos, is not a theory of chaos in this sense; rather, it is establishing a new kind of principle of order. But note Tigger’s defence in Winnie-the-Pooh: “You shouldn’t bounce so much.” “I didn’t bounce; I coughed.” “You bounced.” “Well, I sort of boffed.” In a recent paper (Halliday 1990) I suggested that our present grammars are in some respects environmentally unsound. As is done by Michael O’Toole in relation to other semiotics such as art and architecture; see for example O’Toole (1994). Cf. also Theo van Leeuwen (1988).

368

Chapter Fourteen

GRAMMAR AND DAILY LIFE: CONCURRENCE AND COMPLEMENTARITY (1998)

Let me first say what I mean by “grammar” in the title of the paper. I mean the lexicogrammatical stratum of a natural language as traditionally understood, comprising its syntax and vocabulary, together with any morphology the language may display: Lamb’s “lexical system”, in his current (1992: Chapter 5) ‘three-level architecture’ – in commonsense terms, the resources of wording in which the meanings of a language are construed. And here I have in mind particularly the evolved, spontaneous grammar that construes the discourse of daily life. This is not to exclude from the picture the elaborated grammars of scientific and other metalanguages; but these can only be understood as what they are: an outgrowth, supported by design, of the original grammar that is learnt at mother’s knee and on father’s shoulders. Now English is not very efficient at creating technical nomenclature, since it tends to confuse the study of a phenomenon with the phenomenon itself. So while the term “grammar” is commonly used in the way in which I have defined it, to mean the wording system, the central processing unit of a natural language, it is also used indiscriminately to mean the study of that system: grammar2 meaning ‘the study of grammar1’. Since the study of language is called “linguistics”, I have been calling the study of grammar “grammatics” in order to make the distinction clearer. A grammatics is thus a theory for explaining grammar. But is not a grammar itself also a theory? Clearly it is. A grammar is

First published in Functional Approaches to Language, Culture and Cognition, 2000, edited by Teun A. van Dijk. Amsterdam: John Benjamins, pp. 221–37.

369

construing and enacting

a resource for meaning, the critical functioning semiotic by means of which we pursue our everyday life. It therefore embodies a theory of everyday life; otherwise it could not function in this way. A grammar is a theory of human experience: or rather, let us say, it includes a theory of experience, because it is also something else besides. Like any other theory, a grammar is something to think with. It is through grammar that we make sense out of our experience, both of the world we live in (what we experience as taking place “out there”) and of the world that lives in us (what we experience as taking place “in here”, inside our own consciousness), construing a “reality” such that the one can be reconciled against the other (Matthiessen 1991; Halliday and Matthiessen 1999). During the past twenty years leading neurobiologists, such as Harry Jerison and John Allman, have been investigating the way the brain evolved; and they explain its evolution as the evolution of the organism’s resource for constructing reality. Changes in the ecological environment require changes in the representation of experience (Edelman 1992; Lemke 1993). One critical step was the evolution of the cerebral cortex, which transformed the mammalian map of the external environment. The second was the evolution of language, which added a new dimension to reality, that of introspective consciousness; this latter step is associated with the development of the prefrontal zone of the cortex, allowing a major reorganization of neural circuitry (Dunbar 1992). Linguists can show that the corresponding unique feature of human language, distinguishing it from semiotic systems of other genera and species, is that it has a grammar, an abstract stratum of coding in between the meaning and the expression. Grammar is what brings about the distinctively human construction of reality; and by the same token, grammar makes it possible for us to reflect on this construction. As a teacher I have often said to my students that they should learn to ‘think grammatically’. By this I mean that they should use the unique power of the human brain to reflect on the way their experience is construed in their grammar: use grammatics to think about what grammar thinks about the world. I suggest they might do this with problems of any kind, such as relationships with family and friends, or whether to go for the job that pays more or for the one they would more enjoy. Let me give a small example of what I mean by thinking grammatically. You’re feeling a bit down. What’s the matter, someone asks. ‘I have a headache.’ So how does the grammar construe your unfortunate condition? Of course, you construed it, using your grammatical potential; but you did so quite unconsciously, 370

grammar and daily life

in the way that it has been done countless other times by countless other people, so it is reasonable to talk about the condition being construed ‘by the grammar’. In I have a headache the grammar construes a kind of thing, called an ache; it then uses a part of the body to classify this thing, setting up a taxonomy of aches including stomachache, backache and various others. (Not all the parts of the body are allowed to ache, however; you cannot have a footache or a thighache.) The grammar then sets up a configuration of possession between the ache and some conscious being, in this case the speaker I. The speaker becomes the owner of one specimen of that complex class of things. It is not a prototypical form of possession; the possessor does not want the thing possessed but cannot get rid of it – cannot give it away, or put it back where it came from. Why then does the grammar not favour my head aches; or my head’s aching? – in which the aching is a process, a state of being, rather than a thing, and the entity involved in that state of being is my head rather than me. The grammar has no trouble in constructing the clause my head aches; yet it is not the most usual way in which the experience is worded. Why is I have a headache preferred instead? In English, as in many other languages (though not all), there is a particular meaning associated with being the first element in the clause. What is put first is being instated by the speaker as the theme of the coming message; it is the setting for the information that follows (Fries 1995). This pattern of the clause, a structure of “Theme + Rheme”, was apparently identified by the earliest rhetorical grammarians of ancient Greece, the sophists, who seem to have recognized in the thematic organization of the clause a potent resource for constructing legal and political discourse. In modern times it was first investigated in detail by Mathesius, the founder of the Prague school; it is a particularly prominent feature of English, appearing not only in the clause but also as a “fractal” pattern in both smaller and larger structures – inside word groups, both nominal and verbal, on the one hand and extending over a nexus of clauses on the other. The following example, taken from natural conversation, shows thematic predication of a whole clause complex (from Svartvik and Quirk 1980: 304): . . . in my last year at college I said to myself: “You want to do applied chemistry, right? What industries are now just being born which will blossom in the next quarter of a century, which is going to be my working lifetime?” And I said “Plastics, sure as the nose on your face. I’m going to get into this.” . . . 371

construing and enacting

I’m dazzled, you know . . . It’s being able to see your working life will span a period in which so-&-so is the topmost industry which I find so dazzling. Now if I say my head aches, the first element in that clause is my head: I have constructed a message in which my head is enunciated as Theme. My head is instated as what I want to elaborate on. But it isn’t; I’m the one that’s suffering, so the Theme of the clause should more appropriately be ‘me’. How does the grammar accommodate this alternative? Most naturally, by making ‘me’ the Subject, since there is a strong association of these two functions in English. The ‘ache’ becomes a thing separated from myself, something that I possess, with my head identified as its location: I have an ache in my head. Better still, if my head is used as a classifier, the ache and its location become a single complex thing; and this now occupies the culminative position in the clause: I have a headache. The flow of information here is very different from that of my head’s aching. If this was just a feature of the grammar of localized aches and pains, it might remain a curiosity, a special effect rather than a principle. But this pattern has evolved in English as the prototypical form for construing bodily qualities and states; rather than her hair is long, his throat is sore, we tend to say she has long hair, he has a sore throat, putting the person rather than the body part into the thematic role.1 And in certain other languages where initial position is thematic we also regularly find the person, rather than the body part, lodged at the beginning of the clause. The overall patterns are of course different: in particular, there may be no strong bond between Theme and Subject, and this makes it clear that the relevant function is that of Theme. We can give examples from Chinese, Russian and French. In Chinese it is possible to say woˇdi to´u te`ng ‘my head aches’, where as in the English woˇdi to´u ‘my head’ is a single element in the clause and so functions as the Theme. The preferred form, however, is woˇ to´u te`ng ‘me the head aches’, where the ‘head’ is detached from the personal pronoun; woˇ ‘me’ and to´u ‘head’ are now independent elements in the clause and only the first one, woˇ, is thematic. Again, this is the typical pattern for all such expressions in Chinese: ta¯ to´ufaˇ cha´ng ‘her the hair (is) long’, ta¯ ho´ulo´ng to`ng ‘him the throat (is) sore’ and so on. In Russian, likewise, one can say moja golova bolit ‘my head aches’; but this also is not the preferred form. Russian however displays a different pattern: u menja golova bolit ‘at me the head aches’, where again it is the ‘me’ that has thematic status. In French instead of ma teˆte me fait mal ‘my head is 372

grammar and daily life

hurting me’ one can use possession as in English: j’ai mal a` la teˆte ‘I have an ache at the head’. French also has a further device, of detaching the Theme altogether from the structure of the clause, and announcing it as a key signature at the beginning: moi j’ai mal a` la teˆte ‘me I’ve got an ache at the head’. Neither Chinese woˇ nor Russian u menja nor French moi is Subject; what they have in common is the status of Theme. At this point we might think once more of the sufferer and say to him or her: pity you’ve got a headache. But try de-construing this, in the grammar, and then re-construing it – rewording it – as my head aches; or better still my head’s aching, which makes it an external rather than an internal phenomenon. This is rather less self-centered: it is no longer a fact about me, and my inner self, but an external fact about my head. This won’t make the headache go; but it does put it in its place. It has now become a problem of my head, which is just one part of my physical make-up. One might offer this as a form of logotherapy, a kind of grammatical acupuncture. But here I just want it to serve as an instance of “thinking grammatically”. Thus the grammar enables us, unconsciously, to interpret experience; and the metagrammar, or grammatics, enables us to reflect consciously on how it does so. The grammatics, of course, is part of a more general theory of meaning: of language as a semiotic system, and of other semiotic systems brought into relation with language. Without such a general theory, the excursion into other languages is no more than a piece of tourism; it assumes significance only when we can show how this small corner of experience is construed in relation to the meaning potential of each language as a whole. But this requires much more than a purely local explanation. Taking a fragment of the grammar of daily life, and exploring it crosslinguistically in this way, still leaves it as an isolated fragment, detached from its environment in the overall system of the language. Yet this is the critical environment to take into account. The grammar construes a unitary semantic space, elastic and many-dimensioned; and whatever aspect of the grammar we are considering (such as the selection of person as Theme, in the examples above), there will usually be various other grammatical features, many of them not obviously related in any formal sense, which are associated topologically within this semantic space (cf. Martin and Matthiessen 1992). Such features may cluster into a recognizable syndrome, needing to be interpreted not piecemeal but as a whole: this is the principle of “frames of consistency” as formulated by Whorf. Illustrations of this phenomenon may be found in Hasan’s 373

construing and enacting

(1984b) ‘Ways of saying, ways of meaning’, where she shows how the grammar of Urdu construes experience as collectively shared; and in Martin’s (1988) account of “grammatical conspiracies” in Tagalog. If we are comparing the different “realities” of one language with another, it is the syndrome rather than the single feature that is likely to be significant. Side by side with such frames of consistency, however, there are also frames of inconsistency: regions where the grammar construes a pattern out of tensions and contradictions – where the different “voices” of experience conflict. To put this another way, the grammar’s theory of experience embodies complementarity as well as concurrence. Metaphorically the grammar is representing the fact that human experience is too complex, and has too many parameters, to be construed from any one angle alone. It is the combination of these two perspectives – concurrence and complementarity – that is the salient characteristic of the grammar of daily life. Let me first try to illustrate the complementarity, and then use this as a point of departure for exploring concurrence, looking at a more general syndrome of features within which the earlier, more particular example might be located. Many grammars (perhaps all) make a rather clear distinction between the two fundamental modes of human experience referred to above: between what we experience as taking place in the world outside of ourselves and what we experience as processes of our own consciousness – seeing and hearing, liking, disliking, fearing, hoping, thinking, knowing, understanding and the like. In English, the conscious or mental processes differ from the other, material kind in various respects: (1) they have a less exact present time; (2) they presume a conscious being taking part; (3) they do not fall within the scope of ‘doing’, and (4) they can project – that is, they can construe any meaning as taking place in someone’s consciousness (as “direct or indirect thought”). In addition, these inner processes display another feature not found with the grammar of processes of the external, material type: they are bi-directional. Processes of consciousness can be construed with the conscious participant, the Senser, either as object (active Complement), as in it frightens me, or as active Subject, as in I fear it; likewise it pleases/convinces/strikes me, I like/believe/notice it, and so on. These are two different and in fact contradictory constructions of the same class of phenomena. Inner experience is complex and difficult to interpret; the grammar offers two complementary models, one with the Senser in the more active role (by analogy with material processes), one with the Senser appearing 374

grammar and daily life

to be acted upon. Each of these brings out different agnate forms; the grammar of daily life, in English, accommodates both. In late Middle to early Modern English the common verbs of consciousness such as like and think changed their allegiance from the one pattern to the other: from ‘it likes/thinks (to) me’ to I like it, I think so. This happened at about the same time as the emergence of the pattern discussed earlier: I have a headache, etc. For very general processes of consciousness the grammar came to favour the type of construction in which the Senser, the participant credited with consciousness, was the Theme. What was explained above as a preference for a person rather than a part of the body as the starting point for bodily states and conditions is part of a broader picture whereby the grammar of all inner processes and physiological states tended to orient the message towards the human, or humanlike, participant – perhaps with ‘I’, the individual self, as the prototypical member of this class. This in turn leads us to another feature. At the same period of history another shift took place affecting processes of the external kind, those experienced as happening ‘out there’. In earlier English the grammatical Subject in such processes had been overwhelmingly the active participant, whether human or not (in fact the distinction between human and non-human, or conscious and non-conscious, plays no part in the construction of these processes of the external world). Thus in an arrow pierced his eye the arrow was the natural Subject, and remained Subject even if the narrative required the thing acted on to function as Theme. To use a constructed example, the pattern was that of: The king fell to the ground; his eye an arrow had pierced. with the position between different pattern:

Actor remaining as Subject even when displaced from initial in the clause. Subsequently, as already noted, this bond Subject and Actor was deconstructed and replaced by a bond, that of Subject with Theme; this gave the modern

The king fell to the ground; his eye had been pierced by an arrow. This change led to an increase in the frequency of passive verbs, which was followed by a change in the tense system as passive tenses caught up with the active ones; and various other changes took place besides. What this new alignment of grammatical forces amounted to was that relatively less prominence was being given to the structure of the experience – which partner is the doer and which the done-to, so to 375

construing and enacting

speak; and relatively greater prominence to the structure of the message – which part is the theme, and which part is the new information to be attended to. Without trying to go into all the components of the picture, let me refer briefly to three related developments. First, the grammar developed a battery of resources such that any representation of a process can be construed in all possible patterns of information flow; given ‘an arrow pierced his eye’ we have not only his eye was pierced by an arrow but also what pierced his eye was an arrow, what the arrow pierced was his eye, what the arrow did to his eye was pierce it, and so on. These evolved as different ways of dividing the clause into a thematic portion and the rest. But the construction of the message is more fluid and more complex than that simple formulation suggests. The flow of information is made up of two distinct currents: a linear movement from Theme to Rheme, and an oscillation between Given and New which is not encoded in the sequence but in which the “New” – the part presented by the speaker as ‘to be attended to’ – tends to build up at the end. And just as various features in the grammar conspire to construe the Theme, so various others come together in construing the resources for the New; and this leads in to the second of the three developments being mentioned here. Secondly, then, another feature of Modern English grammar is the motif of the “phrasal verb”; we can say he invented the whole story, but we prefer he made the whole story up; similarly you left the important part out (instead of you omitted . . .), they’ve taken the furniture away (instead of they’ve removed . . .), and so on. This is the grammar’s way of making the happening the main item of news. The news tends to come at the end of the clause; but the happening is typically a verb, and if there are two parties to it – an Actor and a Goal, say – it is hard to get the verb at the end: we cannot say he the whole story invented (we can say the whole story he invented; but that changes the thematic balance by marking the Theme). What the “phrasal verb” construction does is to split the verb into two parts, so that the second part of it can come at the end: he made the whole story up is the grammar’s suppletion for *he vented the whole story in. Thirdly, there is an analogous pattern whereby one of the other elements in the clause – one which could but would not necessarily come at the end – is marked out for news value by having a preposition added to it. If you want to tell me that you supported your brother financially you could say I gave my brother a lot of money; but if the observation is made to explain why you now need to borrow from 376

grammar and daily life

me, you say I gave a lot of money to my brother. The preposition to makes explicit your brother’s role as a participant in the process, and is added just in those positions which are prominent in the information flow (likewise if the brother appears as a marked Theme: to my brother I gave a lot of money). It is precisely this same principle which adds by to the Actor when the clause is passive: his eye had been pierced by an arrow/ by an arrow his eye had been pierced. All the features I have sketched in here are features of the grammar of daily life: some more global, some more local, but all of them characteristic of unconscious, spontaneous, everyday linguistic encounters. These, and others that could be added, form a syndrome, a concurrence of related developments, that has helped to shape the meaning potential of Modern English, giving the language its characteristic flavour – that “certain cut”, in Sapir’s terms, which makes each language unique. What all these have in common is that they tend towards giving greater prominence to the organization of discourse as a flow of information, making more explicit how each element is to be construed as part of a message. As a corollary to this, less prominence is given to the experiential patterning, much of which is in fact left implicit once the concern with the message begins to take over. Most of these effects are fairly recent in history; they reflect the changing social conditions of the language over the past five hundred years. Or rather: they do not reflect them – they help to bring them about. These features in the grammar construe the kind of discourse that can be addressed to a stranger, who does not necessarily share the same expectations and norms of interaction. They can be written down in a book that is going to be printed in thousands of copies and read by people who have never met the author and do not even know who he is. In other words, they are features of a standard language: a form of discourse in which the flow of information will typically be rendered explicit rather than being taken for granted. (Interestingly, many of these changes appear not to have taken place in the surviving British rural dialects.) Effects like these are not the result of sudden catastrophic changes. They are trends and tendencies in a long process of evolution; and at any given time they are quantitative – changes in the relative frequency with which this or that pattern is selected from within the system. The grammatics is thus a theory of probabilities, in which possible/impossible is only a special case of more and less probable – and a rather uninteresting case, because meaning is a product of choice and when something becomes impossible there is no more choice. So, for 377

construing and enacting

example, I have a headache is an instance of what is now the more probable of two agnate constructions; but in using that form the speaker is still choosing – choosing, among other things, to map I rather than my head on to the Theme. If my head aches had become obsolete, we could still have used the grammatics to explain why the structure is as it is; but the grammar would have taken over, and the significance of using I have . . . in any particular instance would have been lost. Hence the semantic features being construed in this way would gradually disappear – just as the semantic feature construed by selecting you instead of thou in Elizabethan English disappeared after thou had ceased to be an available alternative, although we can still use this history to explain why you became the sole second person form. What is it that gives language its elasticity, the facility for constantly adapting, reshaping and extending its semantic potential? The answer lies, as Lamb recognized from the start (cf. Lamb 1964), in its stratal pattern: a language is an orchestration of interrelated levels of semiosis. Lamb no longer favours the term “stratificational grammar” (1988: 4), but the stratal principle has always been critical to his thinking. What is relevant here is that Lamb always “insisted that there has to be a level of meanings that is separate from the lexico-grammatical level” (1988: 6). This embodies the evolutionary perspective that I remarked on above: the evolution of lexicogrammar was the major innovation that transformed protolanguages into languages of the adult human kind. Lamb now talks of the higher stratum as the “conceptual system” (1992: 98), and prefers to interpret this from outside language itself. As he remarks, the question “whether or not [the conceptual system] should be considered part of language is . . . relatively uninteresting”: it is absurd to draw boundaries around phenomena under study and then use these boundaries to justify one’s intellectual stance. Such metalinguistic boundaries are like the boundaries drawn by language itself, which as he says (1992: 121) “both help us and hinder us in our efforts to understand the world”. It is these arbitrary features of segmentation and of categorization, imposing syntagmatic and paradigmatic boundaries on our construction of experience, that lead to many of what Lamb calls the “thinking disorders” which arise both in everyday life and in scholarly life (both in language and in metalanguage). Such “disorders” arise at the interface between these two strata: “the semantically generated infelicities of thinking arise because of differences between concepts and the lexemes which express them” (1992: 162). I myself take the alternative approach, of treating Lamb’s “conceptual system” as part of language. This is because I do not think the 378

grammar and daily life

lexicogrammar is arbitrary in its construction of meaning. The grammar has to impose discontinuity on the flux of experience; but the human condition – our total relationship to our environment – is complex and many-faceted, so there will be indefinitely many ways of doing this, and hence differences between one language and another, and within one language at different stages in its history: some random, some resonating with variation and change in human culture. But even within one experiential domain, at any one moment in time, the grammar has to contend with conflicting and often contradictory demands; so this same interface accommodates complementarities – in a sense analogous to that in which Niels Bohr used the term to extend Heisenberg’s uncertainty principle in quantum mechanics. The grammar is unable to reduce some aspect of experience to a single construction and so introduces two distinct perspectives, two construals which are mutually contradictory and yet depend on each other to provide a theory of daily life. An example would be tense and aspect as complementary theories of time. These contradict each other: either time is a linear flow out of past through present into future, or else it isn’t. Yet many languages, perhaps all, insist that it both is and is not: in very different mixtures and proportions, but each amounting to a plausible theory for coping with the everyday world. Some of these complementarities display the further property that one of the two perspectives is construed configurationally, the other iteratively (as multivariate and univariate structures), thus foregrounding respectively the synoptic and the dynamic points of view. For example, the way the grammar constructs taxonomies of things involves both locating them in configurations of properties and modifying them by means of iterative bracketing. The construction of time in English also exemplifies this point: the system of aspect is activated once at a time, while the system of tense allows for successive reentries: present, past in present, future in past in present and so on. The essence of semiotic complementarity is that it is both objective and subjective: some domain of experience is being construed both as two phenomena and as two points of view on the one phenomenon. (The complementarity of lexis and grammar in the lexicogrammatical stratum is a metacomplementarity within the system itself.) One very pervasive complementarity is that in the grammar of agency, where the problem to be solved is: how are the processes in the external world brought about? One theory, as construed in the grammar of daily life, is that of “Actor, +/- Goal” – a “doer”, plus, optionally, something else that is “done to”. Thus: 379

construing and enacting

Don’t disturb Mum: she’s sewing. What is she sewing? She’s sewing her old jacket. Mum is the doer and the jacket is the done-to. This is a configurational model; there is no re-entry to the choice of agency. Then there’s a snap, and the grammar takes up the story again: Bother! the thread snapped. What snapped it? The machine snapped it. Yes? Who made the machine snap it? I did, of course. What made you make the machine snap it? My own impatience, I suppose. and so on. This second theory says that there is a Medium, an entity through which the process is actualized (here the thread), plus, optionally, something else as Causer that brings it about. This is an iterative model; here the agency relation is construed in such a way that it can recur. Thus there are two ways of looking at a process: one according to which participant a acts, and the action may (or may not) extend to another participant x (a is the constant, x the variable); the other according to which participant x “eventuates” (that is, permits the process to eventuate), and the event may (or may not) be brought about by another participant a (x is the constant, a the variable). The first of these (let us call it type A) is the transitive theory of processes, the second (type X) is the ergative; and probably all languages embody some tension between the two. Transitive and ergative are two points of view on the same phenomenon, that of the nature of material processes and the relationship of the participants to the process and to each other; but they are also two distinct phenomena – some processes pattern ergatively and others transitively (cf. Halliday 1967–68; Davidse 1992). This constitutes another strand in the pattern of changes that have been taking place in English: type X has tended increasingly to prevail over type A. Let us follow this up in a related corner of the grammar. When I last worked in the United States I was living in Orange County; I frequently travelled on the local bus services, and there was a notice on the buses which read: 380

grammar and daily life

Federal law prohibits operation of this bus when any passenger is forward of the standee line. If you are standing, on the bus, you are a standee. Why not a stander? You are a passenger, not a passengee (and if you cannot get on the bus you may be a bystander); but once you are a standing passenger you become a standee, and you have your standee line, and must keep behind it. What kind of participant is construed in the grammar as an ee? There are familiar ones like nominee, trainee, appointee, and more recent instances of this type like superannuee and oustee, all of which are modelled on the pattern of employee ‘person employed’. This forms one term in the transitive opposition employer/employee; the latter form was derived from the French passive participle and matched up with the English active termination -er, giving ‘the one who is acting’/’the one who is acted upon’; cf. trainer/trainee. Here the -ee is functioning as participant x in type A. Then there are some instances where rather indirect relationships are involved: biographee ‘person whose biography is being compiled’, amputee ‘person who has had a limb amputated’ (note that it does not refer to the limb; the -ee’s are all human), transplantee (I have a letter beginning “I am a heart transplantee”), ticketee (in airline parlance); and various banking terms like advisee, favouree, assignee and so on. These are modelled on words like referee ‘person to whom a dispute is referred for decision’, refugee ‘person to whom a place of refuge is offered’. Then, with escapee ‘person who escapes’ as an early model, we now have conferee and attendee ‘person attending a conference or lecture’, retiree ‘person having retired’, and returnee ‘person trying to get back to original country’. All these are like standee. When we examine them, we find that they pattern ergatively: the -ee corresponds to the function of the Medium in the process, to participant x in a process construed as type X. There is no implication that these are functioning as the Goal: a standee is not someone who has been or is being stood. If these were following type A we would have stander, returner, retirer, attender and so on. The pattern is given in Figure 1: (type A)

Actor Goal Agent Medium -er (process) Ø (type X) Ø -ee (process) er (process) -ee -er (process) -ee a x a x

Figure 1 Pattern for transitive and ergative interpretations of -ee

381

construing and enacting

In itself, each instance is trivial. It does not matter whether we write stander or standee: the message will get across. But that, in another perspective, is just the point. The word standee is an instance of a very general pattern, through which our experience is ongoingly construed and reinforced; and as such it has a dual significance. On the one hand, as an instance it perturbs, however minutely, the overall probabilities of the system. System and instance are not two separate phenomena; they are the same phenomenon seen by different observers, observing from different time depths; and, especially where the grammar is unstable (as in the present-day English transitivity system), the cumulative effect of such instances is very noticeable. On the other hand, the word standee represents one perspective within a complementarity; to understand it we have to adopt (unconsciously, as always) a particular stance towards the phenomena we experience as taking place outside ourselves. In this perspective, where standing is grouped with being trained (standee, trainee) rather than with training (trainer), agency is interpreted as ‘causing’ rather than ‘doing to’: the variable is not ‘does the action carried out by a extend to another entity x?’ but rather ‘is the process involving x caused by another entity a?’ And this is quite a different way of looking at the processes of daily life. A language is not only a mode of reflection; it is also a mode of action. Besides its ideational function, as a theory for construing our experience, it also has an interpersonal function, as a praxis for enacting our social and personal relationships. These two metafunctions are inseparably interlocked in the system of every language: the grammar does not allow us to perform in one mode without at the same time performing in the other.2 In other words, while we are constructing reality we are also acting on it through our semiotic interactions with other human beings. And this brings me back to the point from which I began, in defining grammar as the spontaneous, natural grammar with which we lead our everyday lives. It is important not to set up a disjunction here. The most abstract theory of modern physics is also a “grammar” of experience – as Lemke (1990) has shown, a scientific theory is constituted of systems of related meanings: hence as well as being something to think with, it is by the same token also something to act with. We recognize this as a feature of scientific theories: they are not ideologically neutral, and this critically affects the domains of scientific praxis. The grammar of daily life is not neutral either. I have tried to suggest elsewhere (Halliday 1990) some of the features of our everyday grammar that seem to me to condition our attitudes, to each other, to other species, and to the natural environment – certain aspects 382

grammar and daily life

of the grammar are ecologically quite unfriendly. By the same token, however, those who “think grammatically” are enabled thereby to act grammatically, whether in developing forms of praxis for educational and other professional tasks, or in combating sexism, racism and other prevailing inequalities. To be a linguist is inevitably to be concerned with the human condition; it takes a linguist of the stature of Sydney Lamb to explain how so much of what constitutes the human condition is construed, transmitted, maintained – and potentially transformed – by means of language.

Notes 1. Notice on the other hand that in the interrogative this pressure is much less strong: we readily say does your head ache? is your throat sore? as well as have you got a headache/a sore throat? This is because in the interrogative the grammar preempts the thematic slot to signal that the clause is, in fact, a question, by putting at the beginning the part of the verb that selects for ‘yes or no’, the Finite operator, does/is: does your head ache? signals ‘my message is concerned with whether it does or not’. As a result there is relatively little thematic weight left over; the difference in information flow between is your throat sore? and have you got a sore throat? is very much less noticeable than that between the agnate declarative pair my throat’s sore and I’ve got a sore throat, where the full thematic weight is felt on either my throat or I. 2. Thus the grammar signals metaphorically that meaning is a social process. We might put this together with the recent neurobiological finding by Robin Dunbar (1992), that species living in large social groups have proportionally larger cortices. “Dunbar’s explanation is that large group sizes demand greater social cohesion and hence more advanced skills for communicating . . .”.

383

Chapter Fifteen

ON GRAMMAR AND GRAMMATICS (1996)

1

The problem

Most of us are familiar with the feeling that there must be something odd about linguistics. We recognize this as a problem in the interpersonal sphere because as linguists, probably more than other professionals, we are always being required to explain and justify our existence. This suggests, however, that others see it as a problem in the ideational sphere. The problem seems to arise from something like the following. All systematic knowledge takes the form of ‘language about’ some phenomenon; but whereas the natural sciences are language about nature, and the social sciences are language about society, linguistics is language about language – “language turned back on itself ”, in Firth’s often quoted formulation. So, leaving aside the moral indignation some people seem to feel, as if linguistics was a form of intellectual incest, there is a real problem involved in drawing the boundary: where does language end and linguistics begin? How does one keep apart the object language from the metalanguage – the phenomenon itself from the theoretical study of that phenomenon? The discursive evidence rather suggests that we don’t, at least not very consistently. For example, the adjective linguistic means both ‘of language’, as in linguistic variation, and ‘of linguistics’ as in linguistic association (we never know, in fact, whether to call our professional bodies linguistic associations or linguistics associations). But a situation First published in Functional Descriptions: Theory in Practice, 1996, edited by Ruqaiya Hasan, Carmel Cloran and David G. Butt. Amsterdam: John Benjamins, pp. 1–38.

384

on grammar and grammatics

analogous to this occurs in many disciplines: objects in nature have physical properties, physicists have physical laboratories; there are astronomical societies and astronomical forces (not to mention astronomical proportions). It is easy to see where this kind of slippage takes place: astronomers observe stars, and an expression such as astronomical observations could equally well be glossed as ‘observations of stars’, or as ‘observations made during the course of doing astronomy’. Likewise linguistic theory is ‘theory of language’, but it is just as plausibly ‘theory in the field of linguistics’. To a certain extent this is a pathological peculiarity of the English language, because in English the ambiguity appears even in the nouns: whereas sociology is the study of society, psychology – originally the study of the psyche – has since slipped across to mean not only the study but also that which is studied, and we talk about criminal psychology (which means the psyche characteristic of criminals, though it “ought to mean” theories of the psyche developed by scholarly criminals). So now psychology is the study of psychology; and an expression such as Australian psychology is unambiguously ambiguous. Such confusion is not normally found for example in Chinese, where typically a clear distinction is made between a phenomenon and its scientific study; thus shehui : shehuixue :: xinli : xinlixue (society : sociology :: psyche : psychology) and so on. But one can see other evidence for the special difficulties associated with linguistics. For example, it is a feature of linguistics departments that, in their actual practice, what they teach is often not so much the study of language as the study of linguistics. (And one of the few fields where the terminological distinction is not consistently maintained in Chinese is that of grammar, where yufa often does duty also for yufaxue.) There do seem to be special category problems arising where language is turned back on itself.

2

Grammar and grammatics

In fact the ambiguity that I myself first became aware of, as a teacher of linguistics (and before that, as a teacher of languages), was that embodied in the term grammar. Here the slippage is in the opposite direction to that of psychology: grammar, the name of the phenomenon (as in the grammar of English), slides over to become the name of the study of the phenomenon (as in a grammar of English). This was already confusion enough; it was made worse by the popular use of the term to mean rules of linguistic etiquette (for example bad grammar). As a way of getting round part of the problem I started using the term 385

construing and enacting

grammatics – I think the first published occasion was in a discussion of ineffability (see above Chapter 12). This was based on the simple proportion grammatics : grammar :: linguistics : language. I assumed it was unproblematic: the study of language is called linguistics; grammar is part of language; so, within that general domain, the study of grammar may be called grammatics. But this proportion is not quite as simple as it seems. The relationship of linguistics to language is unproblematic as long as we leave language undefined; and we can do this – as linguists, we can take language for granted, as sociologists take society for granted, treating it as a primitive term. Grammar, on the other hand, needs defining. Although the word is used in a non-technical sense, as in the bad grammar example, one cannot take this usage over to define a domain of systematic study: in so far as it has any objective correlate at all, this would refer to an inventory of certain marginal features of a language defined by the fact that they carry a certain sort of social value for its speakers. We can study ethnographically the patterns of this evaluation, and their place in the social process; but that is a distinct phenomenal domain. Grammatics, in fact, has no domain until it defines one for itself (or until one is defined for it within general linguistics – exactly at what point the term grammatics takes over from linguistics is immaterial). And it is this that makes the boundary hard to draw. Since both the grammar and the grammatics are made of language, then if, in addition, each has to be used to define the other, it is not surprising if they get confused. Now you may say, as indeed I said to myself when first trying to think this through: it doesn’t matter. It does no harm if we just talk about grammar without any clear distinction between the thing and the study of the thing. They are in any case much alike: if you turn language back on itself, it is bound to mimic itself in certain respects. But this comforting dismissal of the problem was belied by my own experience. If I had become aware of the polysemy in the word grammar it was because it got in the way of clear thinking – my own, and that of the students I was trying to teach. (It does not help, incidentally, to take refuge in the term syntax, where precisely the same polysemy occurs.) There was confusion in certain concepts, such as “universals of grammar” and “rule of grammar”, and in the status and scope of grammatical categories of various kinds. But also, I suspect, a problem that has been so vexing in recent years – that of relating the system to the text (so often discourse is analysed as if there were no general principles of meaning behind it) – is ultimately part of the same overall unclarity. 386

on grammar and grammatics

3

Defining grammar

In the simplest definition grammar is part of language. If we pick up a book purporting to describe a language, or to help us to learn it, we expect to find some portion or portions of the book – but not the whole of the book – devoted to grammar. In my own work, I have operated with the concept of “lexicogrammar” (that is, grammar and vocabulary as a single unity), while usually referring to it simply as grammar for short; this is a stratal concept, with grammar as one among an ordered series comprising (at least) semantics / lexicogrammar / phonology. But whatever part–whole model is adopted, language remains the more inclusive term. But there is a further step, by which grammar is not just one among various parts of language; it is a privileged part. The exact nature of this privilege will be interpreted differently by different linguists, and some might deny it altogether; but most would probably accept it in one form or another. I would be inclined to characterize grammar in the first instance as the part of language where the work is done. Language is powered by grammatical energy, so to speak. Let me approach the definition of grammar, however, from a somewhat different angle. I shall assume here, as a general theoretical foundation, the account of language given by Lemke (1993). Lemke characterizes human communities as eco-social systems which persist in time through ongoing exchange with their environment; and the same holds true of each of their many sub-systems. The social practices by which such systems are constituted are at once both material and semiotic, with a constant dynamic interplay between the two. Note that by semiotic I mean ‘having to do with meaning’, not ‘having to do with signs’; thus, practices of doing and practices of meaning. The important feature of the material–semiotic interplay is that, as Lemke points out, the two sets of practices are strongly coupled: there is a high degree of redundancy between them. We may recall here Firth’s concept of ‘mutual expectancy’ between text and situation. Underlying the semiotic practices are semiotic systems of various kinds. In fact, we usually use the term “system” to cover both system and process: both the potential and the instances that occur; thus a semiotic system is a meaning potential together with its instantiation in acts of meaning. Now, one special kind of semiotic system is one that has a grammar in it: such a system “means” in two phases, having a distinct phase of wording serving as the base for the construction of meaning. In other words, its “content plane” contains a grammar as 387

construing and enacting

well as a semantics. We could characterize this special kind of semiotic system as a grammatico-semantic system. It is the presence of a grammar that gives such a system its unique potential for creating (as distinct from merely reflecting) meaning.

4

The emergence of grammar through time

We could locate grammatico-semantic systems within the framework of an evolutionary typology of systems, as in Figure 1. In this frame, semiotic systems appear as systems of a fourth order of complexity, in that they are at once physical and biological and social and semiotic. Within semiotic systems, those with a grammar in them are more complex than those without. physical

S

+ life = biological

S 1

+ value = social

+ meaning = semiotics1 [primary]

S 2

S 3

+ grammar = semiotic2 [higher order, i.e. grammatico-semantic]

S 4.1

4.2

Figure 1 Evolutionary typology of systems

Semiotic systems first evolve in the form of what Edelman (1992) calls “primary consciousness”. They evolve as inventories of signs, a sign being a content/expression pair. Systems of this kind, which may be called primary semiotics, are found among numerous species: all higher animals, including our household pets; and such a system is also developed by human infants in the first year of their lives – I referred to this as the “protolanguage” (Halliday 1975). Primary semiotic systems have no grammar. The more complex type of semiotic system is that which evolves in the form of Edelman’s “higher order consciousness”. This higher order semiotic is what we call language. It has a grammar; and it appears to be unique to mature (i.e. post-infancy) human beings. In other words, it evolved as the “sapiens” in homo sapiens. (I say this without prejudice; I would be happy – indeed very excited – to learn 388

on grammar and grammatics

that higher-order, stratified semiotics had evolved also with other species, such as cetaceans, or higher primates. But I am not aware of any convincing argument or demonstration that they have.1) Certain features of the human protolanguage, our primary semiotic, persist into adult life; for example expressions of pain, anger, astonishment or fear (rephonologized as “interjections”, like ouch!, oy!, wow! . . .). On the other hand, human adults also develop numerous non-linguistic semiotic systems: forms of ritual, art forms, and the like; these have no grammar of their own, but they are parasitic on natural language – their meaning potential derives from the fact that those who use them already have a grammar. (See O’Toole (1994) for a rich interpretation of visual semiotics in grammatico-semantic terms.) Thus all human semiotic activity, from early childhood onwards, is as it were filtered through our grammar-based higher order consciousness. What then is a grammar, if we look at it historically in this way, as evolving (in the species) and developing (in the individual)? A grammar is an entirely abstract semiotic construct that emerges between the content and the expression levels of the original, sign-based primary semiotic system. By “entirely abstract” I mean one that does not interface directly with either of the phenomenal realms that comprise the material environment of language. The expression system (prototypically, the phonology) interfaces with the human body; the (semantic component of the) content interfaces with the entire realm of human experience; whereas the grammar evolves as an interface between these two interfaces – shoving them apart, so to speak, in such a way that there arises an indefinite amount of “play” between the two.

5

Grammar in semiotic function

The grammar is thus the latest part of human language to have evolved; and it is likewise the last part to develop in the growth of the individual child. It emerges through deconstructing the original sign and reconstructing with the content plane split into two distinct strata, semantics and lexicogrammar. Such a system (a higher-order semiotic organized around a grammar) is therefore said to be “stratified” (Lamb 1964; 1992; Martin 1992; 1993). A stratified semiotic has the unique property of being able to create meaning. A primary semiotic, such as an infant’s protolanguage, “means” by a process of reflection: its meanings are given, like ‘here I am!’, ‘I’m in pain’, ‘let’s be together!’, ‘that’s nice’; and hence they cannot modify each other or change in the course of unfolding. By 389

construing and enacting

contrast, a stratified semiotic can constitute: it does not simply reflect, or correspond to pre-existing states of affairs. The stratal pattern of organization, with an entirely substance-free stratum of grammar at its core, makes it possible to construct complex open-ended networks of semantic potential in which meanings are defined relative to one another and hence can modify each other and also can change in interaction with changes in the ongoing (semiotic and material) environment. The grammar does not, of course, evolve in isolation; meanings are brought into being in contexts of function. The functional contexts of language fall into two major types, and the constitutive function that the grammar performs differs as between the two types. On the one hand, language “constitutes” human experience; and in this context, the grammar’s function is to construe: the grammar transforms experience into meaning, imposing order in the form of categories and their interrelations. On the other hand, language “constitutes” social processes and the social order; and here the grammar’s function is to enact: the grammar brings about the processes, and the order, through meaning. And, as we know, the grammar achieves this “metafunctional” synthesis, of semiotic transformation with semiotic enactment (of knowledge with action, if you like), by “constituting” in yet a third sense – creating a parallel universe of its own, a phenomenal realm that is itself made out of meaning. This enables the semiotic process to unfold, through time, in cahoots with material processes, each providing the environment for the other. To put this in other terms, the grammar enables the flow of information to coincide with, and interact with, the flow of events (Matthiessen 1992; 1995). This metafunctional interdependence is central to the evolution of language, and to its persistence through constant interaction with its environment. In the experiential (or, to give it its more inclusive name, the “ideational”) metafunction, the grammar takes over the material conditions of human existence and transforms them into meanings. We tend to become aware of the grammatical energy involved in this process only when we have to write a scientific paper; hence, this semiotic transformation may appear to be just a feature of knowledge that is systematic. But all knowledge is like this: to “know” something is to have transformed it into meaning, and what we call “understanding” is the process of that transformation. But experience is understood in the course of, and by means of, being acted out interpersonally – and, in the same way, interpersonal relations are enacted in the course of, and by means of, being construed ideationally. The grammar flows 390

on grammar and grammatics

these two modes of meaning together into a single current, such that everything we say (or write, or listen to, or read) “means” in both these functions at once. Thus every instance of semiotic practice – every act of meaning – involves both talking about the world and acting on those who are in it. Either of these sets of phenomena may of course be purely imaginary; that in itself is a demonstration of the constitutive power of a grammar.

6

Grammar as theory

So far I have been talking about various properties of grammar. But in “talking about” grammar, I have been “doing” grammatics – it is my discourse that has been construing grammar in this way. Naturally, I have also been ‘doing’ grammar: the properties have been being construed in lexicogrammatical terms. In other words I have been using grammar to construct a theory about itself. Every scientific theory – in fact every theory of any kind, whether ‘scientific’ or otherwise – is constructed in similar fashion, by means of the resources of grammar. A theory is a semiotic construct (see Lemke (1990) for a powerful presentation of this point). That we are able to use a grammar as a resource for constructing theories is because a grammar is itself a theory. As I suggested in the previous section, the grammar functions simultaneously as a mode of knowing and a mode of doing; the former mode – the construction of knowledge – is the transformation of experience into meaning. A grammar is a theory of human experience. Construing experience is a highly theoretical process, involving setting up categories and relating each category to the rest. As Ellis (1993) points out, there are no natural classes: the categories of experience have to be created by the grammar itself. Or, we might say, there are indefinitely many natural classes: indefinitely many ways in which the phenomena of our experience may be perceived as being alike. In whichever of these terms we conceive the matter, the grammar has to sort things out, assigning functional value selectively to the various possible dimensions of perceptual order. The grammar’s model of experience is constantly being challenged and reinforced in daily life; thus it tends to change when there are major changes in the conditions of human existence – not as a consequence, but as a necessary and integral element, of these changes. The difference between a grammar, as a “commonsense” theory of experience, and a scientific theory (such as grammatics) is that grammars 391

construing and enacting

evolve, whereas scientific theories are at least partially designed. (The nearest to an independent, fully designed semiotic system is mathematics. Mathematics is grounded in the grammar of natural language; but it has taken off to the point where its operations can probably no longer be construed in natural language wordings.) But it is still the grammar of natural language that is deployed in the designing of scientific theories (cf. Halliday and Martin 1993). In the next few sections I shall discuss some of the properties of grammars that enable them to function as they do: to theorize about human experience and to enact human relationships. In addition to their metafunctional organization, already alluded to as enabling the integration of knowledge and action, I shall mention (a) their size and ability to expand, (b) their multiplicity of perspective, and (c) their indeterminacy. In talking about these features, of course, I shall still be “doing” grammatics. Then, in the final sections, I shall turn to talking about grammatics.

7

How big is a grammar?

The semogenic operations performed by a grammar are, obviously, extremely complex. Neuroscientists explain the evolution of the mammalian brain, including that of homo sapiens, in terms of its modelling the increasingly complex relationships between the organism and its environment. This explanation foregrounds the construal of experience (the ideational metafunction); so we need to make explicit also its bringing about the increasingly complex interactions between one organism and another (the interpersonal metafunction). To this must be added the further complexity, in a grammar-based higher-order semiotic, of creating a parallel reality in the form of a continuous flow of meaning (the textual metafunction). It could be argued that, since language has to encompass all other phenomena, language itself must be the most complex phenomenon of all. While we may not want to go as far as this, there is still the problem of how language achieves the complexity that it has. Let us pose the simple question: how big is a language? (It seems strange how seldom this question is actually asked.) A simple (though not trivial) answer might be: a language is as big as it needs to be. There is no sign, as far as I know, that languages are failing to meet the immense demands made on them by the explosion of knowledge that has taken place this century. In major languages of technology and science, such as English, Russian or Chinese, there must be well over a million words in use, if 392

on grammar and grammatics

we put together the full range of specialized dictionaries – and the dictionaries can never be absolutely exhaustive. Of course, no one person uses more than a small fraction of these. But counting words in any case tells us very little; what we are concerned with is the total meaning potential, which is construed in the lexicogrammar as a whole. And here again we have to say that there seems no indication that languages are collapsing under the weight. From this point of view, then, it seems as if all we can say is that a language is indefinitely large; however many meanings it construes, it can always be made to add more. Is it possible to quantify in some way its overall meaning potential? At this point we have to bring in a specific model from the grammatics, in which a grammar is represented paradigmatically as a network of given alternatives (a “system network”). Given any system network it should in principle be possible to count the number of alternatives shown to be available. In practice, it is quite difficult to calculate the number of different selection expressions that are generated by a network of any considerable complexity. If we pretend for the moment that all systems are binary, then given a network containing n systems, the number of selection expressions it generates will be greater than n (n+1 if the systems are maximally dependent) and not greater than 2n (the figure that is obtained if all systems are independent). But that does not help very much. Given a network of, say, 40 systems, which is not a very large network, all it tells us is that the size of the grammar it generates lies somewhere between 41 and 240 (which is somewhere around 1012). We do not know how to predict whereabouts it will fall in between these two figures. So let me take an actual example of a network from the grammar of English. Figure 2 shows a system network for the English verbal group (based on the description given in Halliday 1994, but with tense treated non-recursively in order to simplify). This network contains 28 systems, and generates just over seventy thousand selection expressions – 70,992 to be exact. That is a little way over 216. (Not all the systems in it are binary.) This network is relatively unconstrained: it shows no conjunct entry conditions, and it shows an unusually high degree of independence among constituent systems – probably more than there should be, although in this respect the English verbal group is somewhat untypical of (English and other) grammars as a whole. On the other hand, it is not outstandingly delicate: it does not distinguish between can and may, for example, or could and might, or between [they] aren’t and [they]’re not; or among the 393

construing and enacting

Figure 2 The English verbal group: a simplified system network

various possible locations of contrast in a verbal group selecting more than one secondary tense. (And, it should be pointed out, the options shown are all simply the variant forms of one single verb.) So when I prepared a network of the English clause as the first grammar for William Mann’s “Penman” text generation project in 1980, which had 394

on grammar and grammatics

81 systems in it, Mann was probably not far wrong when he estimated off the cuff that it would generate somewhere between 108 and 109 clause types. Of course there are lots of mistakes in these complex networks, and the only way to test them is by programming them and setting them to generate at random. It is not difficult to generate the paradigm of selection expressions from a reasonably small network (already in 1966 Henrici developed a program for this purpose; cf. Halliday and Martin 1981), where you can inspect the output and see where it has gone wrong. But even if the program could list half a billion expressions it would take a little while to check them over. As far as their overall capacity is concerned, however, they are probably not orders of magnitude out. It has been objected that the human brain could not possibly process a grammar that size, or run through all the alternative options whenever its owner said or listened to a clause. I am not sure this is so impossible. But in any case it is irrelevant. For one thing, this is a purely abstract model; for another thing, the number of choice points encountered in generating or parsing a clause is actually rather small – in the network of the verbal group it took only 28 systems to produce some 70,000 selection expressions, and in any one pass the maximum number of systems encountered would be even less – probably under half the total, in a representative network. In other words, in selecting one out of half a billion clause types the speaker/listener would be traversing at the most about forty choice points. So although the system network is not a model of neural processes, there is nothing impossible about a grammar of this complexity – that is, where the complexity is such that it can be modelled in this way, as the product of the intersection of a not very large number of choices each of which by itself is extremely simple.

8

How does your grammar grow?

Grammars do not remain static. They tend to grow; not at an even rate, but with acceleration at certain “moments” in the history of a culture. On the one hand, they grow by moving into new domains. This happens particularly when there is an expansion in the culture’s knowledge and control: in our present era, new domains are opened up by developments in technology and science. We are likely to become aware of this when we meet with a crop of unfamiliar words, 395

construing and enacting

like those associated with the recent move into nanotechnology (engineering the very small); but the expansion may take place anywhere in the lexicogrammar, as new wording, in any form. The grammar is not simply tagging along behind; technological developments, like other historical processes, are simultaneously both material and semiotic – the two modes are interdependent. Early on in his researches into science and technology in China, Needham noted how in the medieval period, when there was no adequate institutional mechanism for keeping new meanings alive, the same material advances were sometimes made two or three times over, without anyone realizing that the same technology had been developed before (Needham 1958). On the other hand, grammars grow by increasing the delicacy in their construction of existing domains. (This has been referred to by various metaphors: refining the grid or mesh, sharpening the focus, increasing the granularity and so on. I shall retain the term “delicacy”, first suggested by Angus McIntosh in 1959.) This is a complex notion; it is not equivalent to subcategorizing, which is simply the limiting case – although also the one that is likely to be the most easily recognized. The grammar does construct strict taxonomies: fruit is a kind of food, a berry is a kind of fruit, a raspberry is a kind of berry, a wild raspberry is a kind of raspberry; these are typically hyponymic and can always be extended further, with new words or new compositions of words in a grammatical structure, like the nominal group in English and many other languages. But greater delicacy is often achieved by intersecting semantic features in new combinations; and this is less open to casual inspection, except in isolated instances which happen to be in some way striking (like certain “politically correct” expressions in presentday English). The massive semantic innovations brought about by computing, word processing, networking, multimedia, the information superhighway and the like, although in part construing these activities as new technological domains, more typically constitute them as new conjunctions of existing meanings, as a glance at any one of thousands of current periodicals will reveal. On a somewhat less dramatic scale, we are all aware of the much more elaborate variations in the discourse of environmental pollution and destruction than were available a generation ago. Even a seemingly transparent piece of wording such as smoke-free construes a new confluence of meanings; indeed the whole semogenic potential of -free as a derivational morpheme has recently been transformed. (Similar expansions have happened with -wise and -hood.) There is a special case of this second heading – perhaps even a third 396

on grammar and grammatics

type of grammar growth – in the form of semantic junction brought about by grammatical metaphor. Here what happens is a kind of reconstrual of some aspect of experience at a more abstract level, brought about by the metaphoric potential inherent in the nature of grammar. A new meaning is synthesized out of two existing ones, (a) a lexicalized meaning and (b) the category meaning of a particular grammatical class. So, for example, when [weapons] that kill more people was first reworded as [weapons] of greater lethality, a new meaning arose at the intersection of ‘kill’ with ‘thingness’ (the prototypical meaning of a noun). Much technical, commercial, bureaucratic and technocratic discourse is locked in to this kind of metaphoric mode. We can observe all these processes of grammar growth when we interact with children who are growing up (Painter 1992; Derewianka 1995). This is a good context in which to get a sense of the openendedness of a grammar. In the last resort – and in some sense that is still unclear – there must be a limit to how big a grammar can grow: that is, to the semiotic potential of the individual “meaner”; after all, the capacity of the human brain, though undoubtedly large, is also undoubtedly finite. But there is no sign, as far as I know, that the limit is yet being approached.

9

Grammar as multiple perspectives

In a stratified semiotic system, where grammar is decoupled from semantics, the two strata may differ in the arrangement of their internal space. Things which are shown to be topologically distant at one stratum may appear in the same systemic neighbourhood at the other. (See Martin and Matthiessen 1992, where the distinction is interpreted as between topological (semantics) and typological (lexicogrammar).) It is this degree of freedom – the different alignment of semogenic resources between the semantics and the grammar – that enables language to extend indefinitely its meaning-making potential (a striking example of this is grammatical metaphor, mentioned at the end of the previous section). It is also this characteristic which explains how syndromes of grammatical features scattered throughout different regions of the grammar may cluster semantically to form what Whorf called “frames of consistency”; cf. Hasan’s “ways of meaning” (1984b), Martin’s “grammatical conspiracies” (1988). This amount of “play” is obviously to be encountered across the (typically arbitrary) boundary between content and expression: we do not expect things which mean the same to sound the same – although 397

construing and enacting

there is considerable seepage, which Firth labelled “phonaesthesia” (Firth 1957). But between the semantics and the grammar, this new frontier (typically non-arbitrary) within the content plane, we expect to find more isomorphism: things which mean alike might reasonably be worded alike. As a general rule, they are: grammatical proportionalities typically construe semantic ones. But not always. On the one hand, there are regions of considerable drift in both directions; an obvious one in English is the semantic domain of probability and subjective assessment, which is construed in many different regions of the grammar – each of which may in turn construe other semantic features, such as obligation or mental process. On the other hand, there are the “syndromes” mentioned above – high-level semantic motifs which are located all around the terrain of the lexicogrammar, such as the complex edifice of meanings that goes to make up a “standard language”. People make much use of these realignments in reasoning and inferencing with language. This stratified vision of things enables the grammar to compromise among competing models of reality. As pointed out above in Section 6, a grammar sorts out and selects among the many proportionalities that could arise in the construal of experience. It does this by making adjustments among the different strata. Things may appear alike from any of three different angles: (i) “from above” – similarity of function in context; (ii) “from below” – similarity of formal make-up; and (iii) “from the same level” – fit with the other categories that are being construed in the overall organization of the system. The grammar looks at objects and events from all three angles of orientation. It takes account of their function: phenomena which have like value for human existence and survival will tend to be categorized as alike. It takes account of their form: phenomena which resemble each other to human perceptions will tend to be categorized as alike. And it takes account of how things relate to one another: phenomena are not categorized in isolation but in sets, syndromes and domains. In other words, the grammar adopts what we may call a “trinocular” perspective. It often happens that the various criteria conflict: things (whether material or semiotic) that are alike in form are often not alike in function; and the way they relate to each other may not reflect either kind of likeness. Other things being equal, the grammar tends to give some precedence to functional considerations: consider any crowded lexical domain, such as that of maps, plans, charts, figures, diagrams, tables and graphs in English; or grammatical systems that are highly critical for survival, like that of polarity in any language. But the construal of 398

on grammar and grammatics

categories must make sense as a whole. And this means that it needs to be founded on compromise. The grammar of every natural language is a massive exercise in compromise, accommodating multiple perspectives that are different and often contradictory. Such compromise demands a considerable degree of indeterminacy in the system.

10

Indeterminacy in grammar

It seems obvious that grammars are indeterminate (or “fuzzy”, to borrow the term from its origins in Zadeh’s “fuzzy logic”), if only because of the effort that goes into tidying them up. Formal logic and even mathematics can be seen as the result of tidying up the indeterminacies of natural language grammars. The typology of indeterminacy is itself somewhat indeterminate. For the present discussion I will identify three types: (a) clines, (b) blends, and (c) complementarities, with (d) probability as a fourth, though rather different case. Clines are distinctions in meaning which take the form of continuous variables instead of discrete terms. The prototype examples in grammar are those distinctions which are construed prosodically, typically by intonation (tone contour): for example, in English, “force”, from strong to mild, realized as a continuum from wide to narrow pitch movement – if the tone is falling, then from wide fall (high to low) to narrow fall (midlow to low). But one can include in this category those distinctions where, although the realizations are discrete (i.e. different wordings are involved), the categories themselves are shaded, like a colour spectrum: for example, colours themselves; types of motorized vehicles (car, bus, van, lorry, truck, limousine . . . etc.); types of process (as illustrated on the cover of the revised edition of my Introduction to Functional Grammar 1994). In this sense, since in the grammar’s categorization of experience fuzziness is the norm, almost any scalar set will form a cline: cf. humps, mounds, hillocks, hills and mountains; or must, ought, should, will, would, can, could, may, might. Blends are forms of wording which ought to be ambiguous but are not. Ambiguity in the strict sense, as in lexical or structural puns, is not a form of indeterminacy as considered here, because it does not involve indeterminacy of categorization. Blends also construe two (or more) different meanings; but the meanings are fused – it is not a matter of selecting one or the other. A favourite area for blends, apparently in many languages, is modality; in English, oblique modal finites like 399

construing and enacting

should provide typical examples, for example the brake should be on, meaning both ‘ought to be’ and ‘probably is’. There is then the further indeterminacy between an ambiguity and a blend, because a wording which is clearly ambiguous in one context may be blended when it occurs in another. A metaphor is the limiting case of a blend. Complementarities are found in those regions of (typically experiential) semantic space where some domain of experience is construed in two mutually contradictory ways. An obvious example in English is in the grammar of mental processes, where there is a regular complementarity between the “like” type (I like it; cf. notice, enjoy, believe, fear, admire, forget, resent . . . ) and the “please” type (it pleases me; cf. strike, delight, convince, frighten, impress, escape, annoy . . .). The feature of complementarities is that two conflicting proportionalities are set up, the implication being that this is a complex domain of experience which can be construed in different ways: here, in a process of consciousness the conscious being is on the one hand ‘doing’, with some phenomenon defining the scope of the deed, and on the other hand ‘being done to’ with the phenomenon functioning as the doer. All languages (presumably) embody complementarities; but not always in the same regions of semantic space (note for example the striking complementarity of tense and aspect in Russian). One favourite domain is causation and agency, often manifested in the complementarity of transitive and ergative construals. Strictly speaking probability is not a “fuzzy” concept; but probability in grammar adds indeterminacy to the definition of a category. Consider the network of the English verbal group in Figure 2 above. As an exercise in grammatics this network is incomplete, in that there are distinctions made by the grammar that the network fails to show: in that sense, as already suggested, no network ever can be complete. But it is incomplete also in another sense: it does not show probabilities. If you are generating from that network, you are as likely to come up with won’t be taken as with took; whereas in real life positive is significantly more likely than negative, active than passive, and past than future. Similarly a typical dictionary does not tell you that go is more likely than walk and walk is more likely than stroll, though you might guess it from the relative length of the entries. A grammar is an inherently probabilistic system, in which an important part of the meaning of any feature is its probability relative to other features with which it is mutually defining. Furthermore the critical factor in register variation is probabilistic: the extent to which local probabilities depart from the global patterns of the language as a whole; for example a 400

on grammar and grammatics

register of weather forecasting (and no doubt other kinds of forecasting as well), where future becomes more probable than past; or one in which negative and passive suddenly come to the fore, like that of bureaucratic regulations (Halliday 1991). Probabilities are significant both in ideational and in interpersonal meanings, as well as in the textual component; they provide a fundamental resource for the constitutive potential of the grammar.

11

Some matching features

In the last few sections I have picked out certain features of natural language grammars which a theory of grammar – a “grammatics” – is designed to account for. The purpose of doing this was to provide a context for asking the questions: how does the grammatics face up to this kind of requirement? Given that every theory is, in some sense, a lexicogrammatical metaphor for what it is theorizing, is there anything different about a theory where what it is theorizing is also a lexicogrammar? There is (as far as I can see) no way of formally testing a grammar in its role as a theory of human experience: there are no extrinsic criteria for measuring its excellence of fit. We can of course seek to evaluate the grammar by asking how well it works; and whatever language we choose it clearly does – grammars have made it possible for humanity to survive and prosper. They have transmitted the wisdom of accumulated experience from one generation to the next, and enabled us to interact in highly complex ways with our environment. (At the same time, it seems to me, grammars can have quite pernicious side-effects, now that we have suddenly crossed the barrier from being dominated by that environment to being in control of it, and therefore also responsible for it; cf. Halliday 1993). I suspect that the same holds true for the grammatics as a theory of grammar: we can evaluate such a theory, by seeing how far it helps in solving problems where language is centrally involved (problems in education, in health, in information management and so on); but we cannot test it for being right or wrong. (This point was made by Hjelmslev many years ago, as the general distinction between a theory and a hypothesis.) By the same token a grammatics can also have its negative effects, if it becomes reductionist or pathologically one-sided. The special quality of a theory of grammar, I think, is the nature of the metaphoric relationship that it sets up with its object of enquiry. If we consider just those features of language brought into the discussion 401

construing and enacting

above – the size (and growth) of the grammar, its trinocular perspective, and its fuzz – how does the grammatics handle these various parameters? To put this in very general terms: how do we construe the grammatics so as to be able to manage the complexity of language? It seems to me that there are certain matching properties. The grammatics copes with the immense size of the grammar, and its propensity for growing bigger, by orienting itself along the paradigmatic axis, and by building into this orientation a variable delicacy; this ensures that the grammar will be viewed comprehensively, and that however closely we focus on any one typological or topological domain this will always be contextualized in terms of the meaning potential of the grammar as a whole. It copes with the trinocular vision of the grammar by also adopting a trinocular perspective, based on the stratal organization of the grammar itself. And it copes with the indeterminacy of the grammar by also being indeterminate, so that the categories of the theory of grammar are like the categories that the grammar itself construes. Theories in other fields, concerned with non-semiotic systems, begin by generalizing and abstracting; but they then take off, as it were, to become semiotic constructs in their own right, related only very indirectly and obliquely to observations from experience. The prototype of such a theory is a mathematical model; and one can theorize grammatics in this way, construing it as a formal system. But a grammatics does not need to be self-contained in this same manner. It is, as theory, a semiotic construct; but this does not create any disjunction between it and what it is theorizing – it remains permeable at all points on its surface. The grammatics thus retains a mimetic character: it explains the grammar by mimicking its crucial properties. One could say that it is based on grammatical logic rather than on mathematical logic. In some respects this will appear as a weakness: it will lack the rigour of a mathematical theory. But in other respects it can be a source of strength. It is likely to be more relevant to understanding other semiotic systems: not only verbal art, but also other, non-verbal art forms, as demonstrated by O’Toole’s masterly interpretation of painting, architecture and sculpture in terms of systemic grammatics, referred to already (O’Toole 1994). And the new field of “intelligent computing”, associated with the work of Sugeno, and explicitly defined by him as “computing with (natural) language”, requires a theory that celebrates indeterminacy (it is a development of fuzzy computing) and that allows full play to the interface between wording and meaning (see section 20 below). 402

on grammar and grammatics

In the next few sections I will make a few observations about these matching properties of the grammatics, as they seem to me to emerge in a systemic perspective.

12

Paradigmatic orientation and delicacy

When many years ago I first tried to describe grammar privileging the paradigmatic axis of representation (the “system” in Firth’s framework of system and structure), the immediate reasons related to the theoretical and practical tasks that faced a ‘grammatics’ at the time (the middle 1960s): computational (machine translation), educational (first and second language teaching; language across the curriculum); sociological (language and cultural transmission, in Bernstein’s theoretical framework, for example Bernstein (1971)); functional-variational (development of register theory) and textual (stylistics and analysis of spoken discourse). All these tasks had in common a strong orientation towards meaning, and demanded an approach which stretched the grammar in the direction of semantics. There were perhaps five main considerations. i: The paradigmatic representation frees the grammar from the constraints of structure; structure, obviously, is still to be accounted for (a point sometimes overlooked when people draw networks, as Fawcett (1988) has thoughtfully pointed out), but structural considerations no longer determine the construal of the lexicogrammatical space. The place of any feature in the grammar can be determined “from the same level”, as a function of its relationship to other features: its line-up in a system, and the interdependency between that system and others. ii: Secondly, and by the same token, there is no distinction made, in a paradigmatic representation, between describing some feature and relating it to other features: describing anything consists precisely in relating it to everything else. iii: Thirdly, the paradigmatic mode of description models language as a resource, not as an inventory; it defines the notion of “meaning potential” and provides an interpretation of “the system” in the other, Saussurean sense – but without setting up a duality between a langue and a parole. iv: Fourthly, it motivates and makes sense of the probabilistic modelling of grammar. Probability can only be understood as the relative probabilities of the terms in a (closed) system. 403

construing and enacting

v: Fifthly, representing grammar paradigmatically shapes it naturally into a lexicogrammar; the bricks-&-mortar model of a “lexicon” of words stuck together by grammatical cement can be abandoned as an outmoded relic of structuralist ways of thinking. This last point was adumbrated many years ago under the formulation “lexis as delicate grammar” (see above, Chapter 2); it has subsequently been worked out theoretically and illustrated in two important papers by Hasan (1985; 1987). The principle is that grammar and lexis are not two distinct orders of phenomena; there is just one stratum here, that of “(lexico)grammar”, and one among the various resources that the grammar has for making meaning (i.e. for “realizing” its systemic features) is by lexicalizing – choosing words. In general, the choice of words represents a delicate phase in the grammar, in the sense that it is only after attaining quite some degree of delicacy that we reach systems where the options are realized by the choice of the lexical item. The lexicogrammar is thus construed by the grammatics as a cline, from “most grammatical” to “most lexical”; but it is also a complementarity, because we can also view lexis and grammar as different perspectives on the whole. The reason people write “grammars” on the one hand and ‘dictionaries’ on the other is that options at the most general (least delicate) end of the cline are best illuminated by one set of techniques while options at the most delicate (least general) end are best illuminated by a different set of techniques. One can employ either set of techniques all the way across; but in each case there will be diminishing returns (increasing expenditure of energy, with decreasing gains). To say that, as the description moves towards the lexical end, one eventually reaches systems where the options are realized by the choice of a lexical item, does not mean, on the other hand, that these are systems where there is a direct correspondence of feature to item, such that feature 1 is realized by lexical item a, feature 2 by lexical item b and so on. What it means is that one reaches systems where the features are components of lexical items. (Thus, they are like the features of a standard componential analysis, except that they form part of the overall system network and no distinction is made between features that are “lexical” and those that are “grammatical”.) Any given lexical item then appears as the conjunct realization of a set of systemic features; and “the same” lexical item may appear many times over, in different locations, much as happens in a thesaurus (where however the organization is taxonomic rather than componential). 404

on grammar and grammatics

13

A note on delicacy

Inherent in the paradigmatic orientation is the concept of variable delicacy, in which again the grammatics mimics the grammar: delicacy in the construal of grammar (by the grammatics) is analogous to delicacy in the construal of experiential phenomena (by the grammar). Since for the most part the “lexicalized” mode of realization is associated with fairly delicate categories in the grammar, we can talk of “lexis as delicate grammar” (this refers to lexical items in the sense of “content words”; grammatical items, or “function words”, like the, of, it, not, as, turn up in the realization of very general systemic features). But this is not the same thing as saying that when one reaches the stage of lexical realization one has arrived at the endpoint in delicacy. What is the endpoint, on the delicacy scale? How far can the grammatics go in refining the categories of the grammar? In one sense there can be no endpoint, because every instance is categorially different from every other instance, since it has a unique instantial context of situation. We tend to become aware of this when an instance is codified in the work of a major writer and hence becomes immortalized as a “quotation”. It seems trivial; but it may not be trivial in the context of intelligent computing, where the program might need to recognize that, say, turn left!, as instruction to the car, has a different meaning – and therefore a different description – at every instance of its use. This is the sense in which a grammar can be said to be an “infinite” (i.e. indefinitely large) system. But if we are literate, then in our commonsense engagements with language, in daily life, we behave as if there is an endpoint in delicacy: namely, that which is defined by the orthography. We assume, in other words, that if two instances look different (i.e. are represented as different forms in writing) they should be described as different types; whereas if two instances are written alike they should be described as tokens of the same type – however delicate the description, it will not tease them apart. The orthography is taken as the arbiter of paradigmatic boundaries: the way things are written determines their identity. There is sense in this: writing represents the unconscious collective wisdom of generations of speakers/listeners. And we do allow exceptions. (a) We recognize homonymy and, more significantly, polysemy, where the delicacy of categorization does not stop at the barrier created by the writing system. (b) We accept that there are systematic distinctions which orthography simply ignores: for example, in English, all those realized by intonation and rhythm. (c) And, as already noted, it 405

construing and enacting

never was assumed, except perhaps among a very few linguists, that a “function word” like of has only one location in the terrain described by the grammatics. These exceptional cases challenge the implicit generalization that the orthographic form always defines a “type” within the wording. A more explicit principle could be formulated: that, as far as the grammatics is concerned, the endpoint in delicacy is defined by what is systemic: the point where proportionalities no longer continue to hold. As long as we can predict that a : a⬘ :: b : b⬘ :: . . . , we are still dealing with types, construed as distinct categories for purposes of grammatical description. In practice, of course, we are nowhere near this endpoint in writing our systemic “grammars”. (I find it disturbing when the very sketchy description of English grammar contained in Halliday (1994) is taken as some kind of endpoint. Every paragraph in it needs to be expanded into a book, or perhaps some more appropriate form of hypertext; then we will be starting to see inside the grammar – and be able to rewrite the introductory sketch!) We are only now beginning to get access to a reasonable quantity of data. This has been the major problem for linguistics: probably no other defined sphere of intellectual activity has ever been so top-heavy, so much theory built overhead with so little data to support it. The trouble was that until there were first of all tape recorders and then computers, it was impossible to assemble the data a grammarian needs. Since grammars are very big, and very complex, an effective grammatics depends on having accessible a very large corpus of diverse texts, with a solid foundation in spontaneous spoken language; together with the sophisticated software that turns it into an effective source of information.

14

A note on the corpus

A corpus is not simply a repository of useful examples. It is a treasury of acts of meaning which can be explored and interrogated from all illuminating angles, including in quantitative terms (cf. Hasan 1992a). But the corpus does not write the grammar for us. Descriptive categories do not emerge out of the data. Description is a theoretical activity; and as already said, a theory is a designed semiotic system, designed so that we can explain the processes being observed (and, perhaps, intervene in them). A “corpus grammar” will be (a description based on) a grammatics that is so designed as to make optimum use of the corpus data available, maximizing its value as an information source 406

on grammar and grammatics

for the description. (‘Corpus-based grammar’ might be a less misleading term.) It is not a grammatics that is mysteriously theory-free (cf. Matthiessen and Nesbitt 1996). Not even the most intelligent computer can perform the alchemy of transmuting instances of a grammar into the description of a grammatical system. Corpus-based does not mean lexis-based. One may choose to take the lexicologist’s standpoint, as Sinclair does (1991), and approach the grammar from the lexical end; such a decision will of course affect the initial design and implementation of the corpus itself, but there is nothing inherent in the nature of a corpus that requires one to take that decision. A corpus is equally well suited to lexis-driven or to grammar-driven description. It is worth recalling that the first major corpus of English, the Survey of English Usage set up by Quirk at University College London, was explicitly designed as a resource for writing a grammar in the traditional sense – that is, one that would be complementary to a dictionary. The most obvious characteristic of the corpus as a data base is its authenticity: what is presented is real language rather than sentences concocted in the philosopher’s den. Typically in trawling through a corpus one comes across instances of usage one had never previously thought of. But, more significantly, any kind of principled sampling is likely to bring out proportionalities that have remained entirely beneath one’s conscious awareness. I would contend that it is precisely the most unconscious patterns in the grammar – the cryptogrammatic ones – that are the most powerful in their constitutive effect, in construing experience and in enacting the social process, and hence in the construction of our ideological makeup. Secondly, the corpus enables us to establish the probability profiles of major grammatical systems. Again, I would contend that quantitative patterns revealed in the corpus – as relative frequencies of terms in grammatical systems – are the manifestation of fundamental grammatical properties. The grammar is an inherently probabilistic system, and the quantitative patterns in the discourse that children hear around them are critical to the way they learn their mother tongues. Thirdly, the corpus makes it possible to test the realization statements, by using a general parser and, perhaps more effectively, by devising pattern-matching programs for specific grammatical systems; one can match the results against one’s own analysis of samples taken from the data. Some form of dedicated parsing or pattern matching is in any case needed for quantitative investigations, since the numbers to be counted are far above what one could hope to process manually (cf. Halliday and James 1993). Fourthly, since modern 407

construing and enacting

corpuses are organized according to register, it becomes possible to investigate register variation in grammatical terms: more particularly, in quantitative terms, with register defined as the local resetting of the global probabilities of the system.

15

Trinocular vision

The “trinocular” principle in the grammatics can be simply stated. In categorizing the grammar, the grammarian works “from above”, “from roundabout” and “from below”; and these three perspectives are defined in terms of strata. Since the stratum under attention is the lexicogrammar, “from roundabout” means ‘from the standpoint of the lexicogrammar itself ’. “From above” means ‘from the standpoint of the semantics: how the given category relates to the meaning (what it “ ‘realizes” ’)’. “From below” means ‘from the standpoint of morphology and phonology, how the given category relates to the expression (what it “is realized by”)’. What are being taken into account are the regularities (proportionalities) at each of the three strata. Since the patterns seen from these three angles tend to conflict, the resulting description of the grammar, like the grammar’s own description of experience, must be founded on compromise. This is easy to say; it is not so easy to achieve. Often one finds oneself ‘hooked’ on one oculation – obsessed, say, with giving the most elegant account of how some pattern is realized, and so according excessive priority to the view from below; then, on looking down on it from above, one finds one has committed oneself to a “system” that is semantically vacuous. If the view from below is consistently given priority, the resulting description will be a collapsed grammar, so “flat” that only an impoverished semantics can be raised up on it. On the other hand, if one is biased towards the view from above, the grammar will be so inflated that it is impossible to generate any output. And if one looks from both vertical angles but forgets the view from roundabout (surprisingly, perhaps, the commonest form of trap) the result will be a collection of isolated systems, having no internal impact upon each other. In this case the grammar is not so much inflated or collapsed; it is simply curdled. Thus the categories of the grammatics, like those of the grammar, rest on considerations of underlying function, internal organization (with mutual definition) and outward appearance and recognition. But there is more than a simple analogy embodied here. I referred above to the notion of semiotic transformation: that the grammar transforms 408

on grammar and grammatics

experience into meaning. The trinocular perspective is simply that: it is the process of transforming anything into meaning – of “semioticizing” it in terms of a higher order, stratified semiotic. Construing the phenomena of experience means “parsing” them into meanings, wordings and expressions (you only have to do this, of course, when form and function cease to match; this is why the task is inescapably one of achieving compromise). The entire stratal organization of language is simply the manifestation of this trinocular principle. Making this principle explicit in the grammatics is perhaps giving substance to the notion of ‘language turned back upon itself’.

16

Indeterminacy in grammatics

That the grammatics should accommodate indeterminacy does not need explaining: indeterminacy is an inherent and necessary feature of a grammar, and hence something to be accounted for and indeed celebrated in the grammatics, not idealized out of the picture – just as the grammar’s construal of experience recognizes indeterminacy as an inherent and necessary feature of the human condition. But construing indeterminacy is not just a matter of leaving things as they are. Construing after all is a form of complexity management; and just as, in a material practice such as looking after a wilderness, once you have perturbed the complex equilibrium of its ecosystem you have to intervene and actively manage it, so in semiotic practice, when you transform something into meaning (i.e. perturb it semiotically) you also have to manage the complexity. We can note how the grammar manages the complexity of human experience. In the first instance, it imposes artificial determinacy, in the form of discontinuities: thus, a growing plant has to be construed either as tree or as bush or as shrub (or . . .); the line of arbitrariness precludes us from creating intermediate categories like shrush. Likewise, one thing must be in or on another; you are either walking or running, and so on. At the same time, however, each of these categories construes a fuzzy set, whose boundaries are indeterminate: on and run and tree are all fuzzy sets in this sense. Furthermore, the grammar explicitly construes indeterminacy as a semantic domain, with expressions like half in and half on, in between a bush and a tree, almost running and the like. The specific types of indeterminacy discussed in Section 10 above, involving complex relationships between categories, are thus only special cases, foregrounding something which is a property of the grammar as a whole. Now consider the grammatics from this same point of view. The 409

construing and enacting

categories used for construing the grammar – things like noun and subject and aspect and hypotaxis and phrase – are also like everyday terms: they impose discontinuity. Either something is a noun or it is a verb (or . . .); we cannot decide to construe it as a nerb. But, in turn, each one of these itself denotes a fuzzy set. And, thirdly, the same resources exist, if in a somewhat fancier form, for making the indeterminacy explicit: verbal noun, pseudo-passive, underlying subject, and so on. What then about the specific construction of indeterminacy in the overall edifice constructed by such categories? Here we see rather clearly the grammatics as complexity management. On the one hand, it has specific strategies for defuzzifying – for imposing discontinuity on the relations between one category and another; for example, for digitalizing the grammar’s clines (to return to the example of “force”, cited in section 10, it can establish criteria for recognizing a small, discrete set of contrasting degrees of force). A system network is a case in point: qualitative relationships both within and between systems may be ironed out, so that (i) the system is construed simply as a or b (or . . .), without probabilities, and (ii) one system is either dependent on or independent of another, with no degrees of partial association. But, at the same time, the grammatics exploits the various types of indeterminacy as resources for managing the complexity. I have already suggested that the concept of lexicogrammmar (itself a cline from “most grammatical” to “most lexical”) embodies a complementarity in which lexis and grammar compete as theoretical models of the whole. There are many blends of different types of structure, for example the English nominal group construed both as multivariate (configurational) and as univariate (iterative) but without ambiguity between them. And the two most fundamental relationships in the grammatics, realization and instantiation, are both examples of indeterminacy. I have said that a grammar is a theory of human experience. But that does not mean, on the other hand, that it is not also part of that experience; it is. We will not be surprised, therefore, if we find that its own complexity comes to be managed in ways that are analogous to the ways in which it itself manages the complexity of the rest. In the last resort, we are only seeing how the grammar construes itself.

17

A note on realization and instantiation

I referred earlier to these two concepts as being critical when we come to construe a higher order semiotic. Realization is the name given to 410

on grammar and grammatics

the relationship between the strata; the verb realize faces “upwards”, such that the “lower” stratum realizes the “higher” one. (Realization is also extended to refer to the intrastratal relation between a systemic feature and its structural (or other) manifestation.) Instantiation is the relationship between the system and the instance; the instance is said to instantiate the system. It can be said that, in the elements of a primary semiotic (signs), the signifier “realizes” the signified; but this relationship is unproblematic: although the sign may undergo complex transformations of one kind or another, there is no intermediate structure between the two (no distinct stratum of grammar). With a higher order semiotic, where a grammar intervenes, this opens up the possibility of many different types of realization. It is not necessary to spell these out here; they are enumerated and discussed in many places (for example Berry 1977; Fawcett 1980; Martin 1984; Hasan 1987; Matthiessen 1988; Eggins 1994). But there is another opening-up effect which is relevant to the present topic: this concerns the nature and location of the stratal boundary between the grammar and the semantics. This is, of course, a construct of the grammatics; many fundamental aspects of language can be explained if one models them in stratal terms, such as metaphor (and indeed rhetorical resources in general), the epigenetic nature of children’s language development, and metafunctional unity and diversity, among others. But this does not force us to locate the boundary at any particular place. One can, in fact, map it on to the boundary between system and structure, as Fawcett does (system as semantics, structure as lexicogrammar); whereas I have found it more valuable to set up two distinct strata of paradigmatic (systemic) organization. But the point is that the boundary is indeterminate – it can be shifted; and this indeterminacy enables us to extend the stratal model outside language proper so as to model the relationship of a language to its cultural and situational environments. Instantiation is the relationship which defines what is usually thought of as a “fact” – in the sense of a physical fact, a social fact and so on. Facts are not given; they are constructed by the theorist, out of the dialectic between observation and theory. This has always been a problem area for linguistics: whereas the concept of a physical principle became clear once the experimental method had been established – a “law of nature” was a theoretical abstraction constructed mathematically by the experimenter – the concept of a linguistic principle has proved much more difficult to elucidate. 411

construing and enacting

Saussure problematized the nature of the linguistic fact; but he confused the issue of instantiation by setting up langue and parole as if they had been two distinct classes of phenomena. But they are not. There is only one set of phenomena here, not two; langue (the linguistic system) differs from parole (the linguistic instance) only in the position taken up by the observer. Langue is parole seen from a distance, and hence on the way to being theorized about. I tried to make this explicit by using the term “meaning potential” to characterize the system, and referring to the instance as all “act of meaning”; both implying the concept of a ‘meaning group’ as the social-semiotic milieu in which semiotic practices occur, and meanings are produced and understood. Instantiation is a cline, with (like lexicogrammar) a complementarity of perspective. I have often drawn an analogy with the climate and the weather: when people ask, as they do, about global warming, is this a blip in the climate, or is it a long-term weather pattern?, what they are asking is: from which standpoint should I observe it: the system end, or the instance end? We see the same problem arising if we raise the question of functional variation in the grammar: is this a cluster of similar instances (a “text type”, like a pattern of semiotic weather), or is it special alignment of the system (a “register”, like localized semiotic climate)? The observer can focus at different points along the cline; and, whatever is under focus, the observation can be from either of the two points of vantage.

18

Realization and instantiation: some specific analogies

It is safe to say that neither of these concepts has yet been thoroughly explored. Problems arise with instantiation, for example, in using the corpus as data for describing a grammar (why a special category of “corpus grammar”?); in relating features of discourse to systemic patterns in grammar (why a separate discipline of “pragmatics”?); and in construing intermediate categories (such as Bernstein’s “code”, which remains elusive (like global warming!) from whichever end it is observed – which is what makes it so powerful as an agency of cultural reproduction). (See Francis 1993 for the concept of corpus grammar; Martin 1992 for showing that there can be a system-based theory of text; Bernstein 1990 for code; Hasan 1989; 1992b for interpretation of coding orientation; and also Sadovnik 1995 for discussion of Bernstein’s ideas). As far as realization is concerned, Lemke has theorized this power412

on grammar and grammatics

fully as ‘metaredundancy’ (Lemke 1984) (and cf. Chapter 14 above); but this still leaves problems in understanding how metafunctional diversity is achieved, and especially the non-referential, interpersonal aspects of meaning; and in explaining the realization principles at work at strata outside language itself (see Thibault (1992) and Matthiessen (1993a) on issues relating to the construal of interpersonal meanings; Eggins and Martin in press, Hasan (1995), Matthiessen (1993b), on issues involving the higher strata of register and genre). I am not pursuing these issues further here. But as a final step I will shift to another angle of vision and look at realization and instantiation from inside the grammar – turning the tables by using the grammar as a way of thinking about the grammatics. One of the most complex areas in the grammar of English is that of relational processes: processes of being, in the broadest sense. I have analysed these as falling into two major types: (i) attributive, and (ii) identifying. The former are those such as Paula is a poet, this case is very heavy, where some entity is assigned to a class by virtue of some particular attribute. The latter are those such as Fred is the treasurer/ the treasurer is Fred, the shortest day is 22nd June/ 22nd June is the shortest day, where some entity is identified by being matched bi-uniquely with some particular other. (See Halliday 1967–8; 1994.) The identifying relationship, as construed in the grammar of English, involves two particular functions, mutually defining such that one is the outward form, that by which the entity is recognized, while the other is the function the entity serves. This relationship of course takes a variety of more specific guises: form / function, occupant / role, sign / meaning, and so on. I labelled these grammatical functions “Token” and “Value”. This Token / Value relationship in the grammar is exactly one of realization: the Token realizes the Value, the Value is realized by the Token. It is thus analogous to the relationship defined in the grammatics as that holding between different strata. The grammar is modelling one of the prototypical processes of experience as constructing a semiotic relationship – precisely the one that is fundamental to the evolution of the grammar itself. The attributive relationship involves a “Carrier” and an “Attribute”, where the Attribute does not identify the Carrier as unique but places it as one among a larger set. It was pointed out by Davidse (1992) that this Carrier / Attribute relationship in the grammar is actually one of instantiation: the Carrier is an instance of, or “instantiates”, the Attribute. It is thus analogous to the relationship defined in the grammatics as that holding between an instance and the (categories of the) system. 413

construing and enacting

(In that respect the original term “ascriptive”, which I had used earlier to name this type of process, might better have been retained, rather than being replaced by “attributive”.) Here too, then, the grammar is construing a significant aspect of human experience – the perception of a phenomenon as an instance of a general class – in terms of a property of language itself, where each act of meaning is an instance of the systemic meaning potential. Of course, the boot is really on the other foot: the grammatics is parasitic on the grammar, not the other way around. It is because of the existence of clause types such as those exemplified above that we are able to model the linguistic system in the way we do. The grammatics evolves (or rather one should say the grammatics “is evolved”, to suggest that it is a partially designed system) as a metaphoric transformation of the grammar itself. This is a further aspect of the special character of grammatics: while all theories are made of grammar (to the extent that they can be construed in natural language), one which is a grammar about a grammar has the distinctive metaphoric property of being a theory about itself.

19

Centricity

Since the grammatics is a theory about a “logo” system, it is “logocentric”, or rather perhaps “semocentric”: its task is to put semiotic systems in the centre of attention. In the same way, biological sciences are “bio-centric”: biased towards living things; and so on. I think it is also a valid goal to explore the relevance of grammatics to semiotic systems other than language, and even to systems of other types. The grammatics is also “totalizing”, because that is the job of a theory. Of course, it focuses on the micro as well as on the macro – the semiotic weather as well as the semiotic climate; but that again is a feature of any theoretical activity. It has always been a problem for linguists to discover what are the properties of human language as such, and what are features specific to a given language. The problem is compounded by the fact that there is more than one way of incorporating the distinction (wherever it is drawn) into one’s descriptive practice. Firth articulated the difference between two approaches: “what is being sketched here is a general linguistic theory applicable to particular linguistic descriptions, not a theory of universals for general linguistic description” (Firth 1957: 21; Firth’s emphasis). I have preferred to avoid talking about “universals” because it seems to me that this term usually refers to descriptive categories being 414

on grammar and grammatics

treated as if they were theoretical ones. As I see it, the theory models what is being treated as “universal” to human language; the description models each language sui generis, because that is the way to avoid misrepresenting it. Thus while the theory as a whole is logocentric, the description of each language is what we might call “glottocentric”: it privileges the language concerned. The description of English is anglocentric, that of Chinese sinocentric, that of French gallocentric and so on. (Note that the theory is not anglocentric; the description of English is.) This is not an easy aim to achieve, since it involves asking oneself the question: “how would I describe this language as if English (or other languages that might get used as a descriptive model) did not exist?” But it is important if we are to avoid the anglocentric descriptions that have dominated much of linguistics during the second half of the century. In practice, of course, English does exist, and it has been extensively described; so inevitably people tend to think in terms of categories set up for English – or for other relatively well-described languages. I have suggested elsewhere some considerations which seem to me relevant to descriptive practice (Halliday 1992). As far as my own personal history is concerned, I worked first of all for many years on the grammar of Chinese; I mention this here because when I started working on English people told me I was making English look like Chinese! (It seems ironic that, now that systemic theory is being widely applied to Chinese studies, the work of mine most often cited as point of reference is the descriptive grammar of English.) In my view an important corollary of the characterological approach (that is, each language being described in its own terms) is that each language is described in its own tongue. The protocol version of the grammar of English is that written in English; the protocol version of the grammar of Chinese is that written in Chinese; and so on. The principle of “each language its own metalanguage” is important, because all descriptive terminology carries with it a load of semantic baggage from its use in the daily language, or in other technical and scientific discourses; and this semantic baggage has some metalinguistic value. This applies particularly, perhaps, to the use of theoretical terms as metacategories in the description; words such as (the equivalents of) option, selection, rank, delicacy are likely to have quite significant (but variable) loadings. But the principle also helps to guard against transferring categories inappropriately. Even if descriptive terms have been translated from 415

construing and enacting

English (or Russian, or other source) in the first place, once they are translated they get relocated in the semantic terrain of the new language, and it becomes easier to avoid carrying over the connotations that went with the original. So if, say, the term subject or theme appears in a description of Chinese written in English, its status is as a translation equivalent of the definitive term in Chinese. Perhaps one should point out, in this connection, that there can be no general answer to the question how much alike two things have to be for them to be called by the same name!

20

A final note on grammatics

As I said at the beginning, when I first used the term “grammatics” I was concerned simply to escape from the ambiguity where “grammar” meant both the phenomenon itself – a particular stratum in language – and the study of that phenomenon; I was simply setting up a proportion such that grammatics is to grammar as linguistics is to language. But over the years since then I have found it useful to have “grammatics” available as a term for a specific view of grammatical theory, whereby it is not just a theory about grammar but also a way of using grammar to think with. In other words, in grammatics, we are certainly modelling natural language; but we are trying to do so in such a way as to throw light on other things besides. It is using grammar as a kind of logic. There is mathematical logic and there is grammatical logic, and both are semiotic systems; but they are complementary, and in some contexts we may need the evolved logic of grammar rather than, or as well as, the designed logic of mathematics. This reflects the fact that, as I see it, grammatics develops in the context of its application to different tasks. As Matthiessen (1991b) has pointed out, this, in general, is the way that systemic theory has moved forward. Recently, a new sphere of application has been suggested. As mentioned above in Section 10, Sugeno has introduced the concept of “intelligent (fuzzy) computing”: this is computing based on natural language (Sugeno 1995). He has also called it “computing with words”, although as I have commented elsewhere (Halliday 1995) this is really “computing with meanings”. Sugeno’s idea is that for computers to advance to the point where they really become intelligent they have to function the way human beings do – namely, through natural (human) language. This view (and it is more than a gleam in the eye: Sugeno has taken significant steps towards putting it into practice) derives ultimately from Zadeh’s “fuzzy logic”; it depends on reasoning and 416

on grammar and grammatics

inferencing with fuzzy sets and fuzzy matching processes. But to use natural language requires a grammatics: that is, a way of modelling natural language that makes sense in this particular context. Systemic theory has been used extensively in computational linguistics; and the Penman nigel grammar, and Fawcett’s communal grammar, are among the most comprehensive grammars yet to appear in computational form (Matthiessen 1991a; Matthiessen and Bateman 1992; Fawcett and Tucker 1990; Fawcett, Tucker and Lin 1993). But, more importantly perhaps, systemic grammatics is not uncomfortable with fuzziness. That is, no doubt, one of the main criticisms that has been made of it; but it is an essential property that a grammatics must have if it is to have any value for intelligent computing. This is an exciting new field of application; if it prospers, then any grammarian privileged to interact with Sugeno’s enterprise will learn a lot about human language, as we always do from applications to real-life challenging tasks.

Note 1. This is not to question the semiotic achievements of the bonobo chimpanzees (cf. Introduction, p. 3). The issue is whether their construal of human language is an equivalent stratified system with a lexicogrammar at the language is an equivalent stratified system with a lexicogrammar at the core.

417

This page intentionally left blank

BIBLIOGRAPHY

Abercrombie, D. (1963) ‘Conversation and spoken prose’, English Language Teaching 18.1. Reprinted in D. Abercrombie, Studies in Phonetics and Linguistics. London: Oxford University Press (Language and Language Learning 10). Allen, W. S. (1953) ‘Relationship in comparative linguistics’, Transactions of the Philological Society. Allen, W. S. (1956) ‘Structure and system in the Abaza verbal complex’, Transactions of the Philological Society. Anderson, J. M. (1968) ‘Ergative and nominative in English’, Journal of Linguistics 4. 1–32. Bateson, G. (1936) Naven. Stanford: Stanford University Press (2nd edn 1958). Bazell, C. E. (1953) Linguistic Form. Istanbul: Istanbul University Press. Beattie, G. (1983) Talk: An Analysis of Speech and Non-verbal Behaviour in Conversation. Milton Keynes: Open University Press. Bernstein, B. (1970) ‘A sociolinguistic approach to socialization’, in J. J. Gumperz and D. Hymes (eds), Directions in Sociolinguistics. New York: Holt, Rinehart & Winston. Bernstein, B. (1971) Theoretical Studies Towards a Sociology of Language. Vol. 1, Class, Codes and Control. London: Routledge & Kegan Paul. Bernstein, B. (1990) The Structuring of Pedagogic Discourse. Vol. 4, Class, Codes and Control. London: Routledge & Kegan Paul. Berry, M. (1957) Structures and Systems. Vol. 1, Introduction to Systemic Linguistics. London: Batsford Academic. Berry, M. (1977) Levels and Links. Vol. 2, Introduction to Systemic Linguistics. London: Batsford Academic. Berry, M. (1981a) ‘Systemic linguistics and discourse analysis: a multi-layered approach to exchange structure’, in R. M. Coulthard and M. M.

419

bibliography Montgomery (eds), Studies in Discourse Analysis. London: Routledge & Kegan Paul. Berry, M. (1981b) ‘Towards layers of exchange structure for directive exchanges’, Network 2. Bloomfield, L. (1933) Language. Chicago: University of Chicago Press. Boas, F. (1911) Introduction to the Handbook of American Indian Languages. Vol. 40.1, Bulletin of the Bureau of American Ethnology. Washington, D.C.: Smithsonian Institution. Reprinted in C. I. J. M. Stuart, Report of the Fifteenth Annual (First International) Round Table Meeting on Linguistics and Language Study (Monograph Series in Languages and Linguistics 17). Brøndal, V. (1948) Les Parties du discours, trans. by Naert. Copenhagen: Munksgaard. Bu¨hler, K. (1934) Sprachtheorie: Die Darstellungfunktion der Sprache. Jena: G. Fischer. Bull. W. E. (1960) Time, Tense and the Verb. Berkeley: University of California Press (University of California Publications in Linguistics 19). Carnochan, J. (1952) ‘Glottalization in Hausa’, Transactions of the Philological Society. Carroll, J. B. (ed.) (1956) Language, Thought and Reality: Selected Writings of Benjamin Lee Whorf. Cambridge, MA.: M.I.T. Press. Carter, M. G. (ed.) (1981) Arab Linguistics: An Introductory Classical Text, with Translation and Notes. Amsterdam: J. Benjamins (Studies in the History of Linguistics 24). Chafe, W. L. (1982) ‘Integration and involvement in speaking, writing and oral literature’, in D. Tannen (ed.), Spoken and Written Language: Exploring Orality and Literacy. Norwood, N.J.: Ablex (Advances in Discourse Processes 9). Chandola, A. (1970) ‘Some systems of musical scales and linguistic principles’, Semiotica 2. Chao, Y. R. (1968) A Grammar of Spoken Chinese. Berkeley and Los Angeles: University of California Press. Chavarria-Aguilar, O. L. (1954) ‘Transfer grammar’, in Lectures in Linguistics. Poona: Deccan College Post-graduate and Research Institute (Deccan College Handbook Series 5). Chomsky, N. (1957) Syntactic Structures. The Hague: Mouton ( Janua Linguarum 4). Chomsky, N. (1961) ‘Generative grammar’, Word 17. Chomsky, N. (1964) ‘Topics in the theory of generative grammar’, in T. A. Sebeok (ed.), Theoretical Foundations. Vol. 3, Current Trends in Linguistics. The Hague: Mouton. Colby, B. N. (1973) ‘A partial grammar of Eskimo folktales’, American Anthropologist 75.

420

bibliography Colby, B. N. and Colby, L. M. (1980) The Daykeeper: The Life and Discourse of an Ixil Diviner. Cambridge, MA.: Harvard University Press. Danesˇ, F. (1966) ‘A three-level approach to syntax’, Travaux linguistiques de Prague 1. Davidse, K. (1992a) ‘A semiotic approach to relational clauses’, Occasional Papers in Systemic Linguistics 6. Davidse, K. (1992b) ‘Transitivity/ergativity: the Janus-headed grammar of actions and events’, in M. Davies and L. Ravelli (eds), Advances in Systemic Linguistics: Recent Theory and Practice. London: Pinter. Derewianka, B. (1995) ‘Language development in the transition from childhood to adolescence: the role of grammatical metaphor.’ Unpublished PhD dissertation, Macquarie University. Doughty, P. S., Pearce, J. J. and Thornton, G. M. (1971) Language in Use. London: Edward Arnold (for the Schools Council). Dunbar, R. (1992) Summary in ‘Secret life of the brain’, New Scientist 1850 (Supplement 4). Ebeling, C. L. (1960) Linguistic Units. The Hague: Mouton ( Janua Linguarum 12). Eco, U. (1983) The Name of the Rose, trans. by W. Weaver. London: Secker & Warburg. (Italian original Il nome della rosa. Milano: Bompiani.) Edelman, G. M. (1992) Bright Air, Brilliant Fire: On the Matter of the Mind. New York: Basic Books. Eggins, S. (1994) An Introduction to Systemic Functional Linguistics. London: Pinter. Eggins, S. and Martin, J. R. (1997) ‘Genres and registers of discourse’, in T. A. van Dijk (ed.), Discourse: A Multidisciplinary Introduction. London: Sage. Eiler, M. A. (1979) ‘Meaning and choice in writing about literature: a study of cohesion in the expository texts of ninth graders.’ Unpublished PhD dissertation, Illinois Institute of Technology. Ellis, J. M. (1993) Language, Thought and Logic. Evanston Il.: Northwestern University Press. Ellis, J. O. (1966a) Towards a General Comparative Linguistics. The Hague: Mouton (Janua Linguarum). Ellis, J. O. (1966b) ‘On contextual meaning’, in C. E. Bazell et al. (eds) In Memory of J. R. Firth. London: Longman. Fawcett, R. P. (1980) Cognitive Linguistics and Social Interaction. Heidelberg and University of Exeter: Julius Groos (Exeter Linguistic Studies 3). Fawcett, R. P. (1988) ‘What makes a “good” system network good?’, in J. D. Benson and W. S. Greaves (eds), Systemic Functional Approaches to Discourse. Norwood, N.J.: Ablex. Fawcett, R. P. and Tucker, G. H. (1990) ‘Demonstration of GENESYS: a very large semantically based systemic functional grammar’, Proceedings of

421

bibliography the Thirteenth International Conference on Computational Linguistics, Helsinki, Vol. 1. Fawcett, R. P., Tucker, G. H. and Lin, Y. Q. (1993) ‘How a systemic functional grammar works: the role of realisation’, in H. Horacek and M. Zock (eds), New Concepts in Natural Language Generation. London: Pinter. Fillmore, C. J. (1968) ‘The case for case’, in E. Bach and R. T. Harms (eds), Universals in Language. New York: Holt, Rinehart & Winston. Firbas, J. (1959) ‘Thoughts on the communicative function of the verb in English, German and Czech’, Brno Studies in English 1. 39–63. Firbas, J. (1964) ‘On defining the theme in Functional Sentence Analysis’, Travaux linguistique de Prague 1. 267–80. Firth, J. R. (1951) ‘General linguistics and descriptive grammar’, Transactions of the Philological Society. Reprinted in J. R. Firth 1957b. Firth, J. R. (1948) ‘Sounds and prosidies’, in Transactions of the Philological Society. Firth, J. R. (1951) ‘Modes of meaning’, in J. R. Firth 1957b. Firth, J. R. (1952) ‘Personality and language in society’, Sociological Review 42. 2. Firth, J. R. (1955) ‘Structural linguistics’, Transactions of the Philological Society. Reprinted in F. R. Palmer (ed.), Selected Papers of J. R. Firth 1952–1959. London: Longman (Longman’s Linguistic Library). Firth, J. R. (1957a) ‘A synopsis of linguistic theory’, in J. R. Firth et al. (eds), Studies in Linguistic Analysis. Oxford: Blackwell (Special Volume of the Philological Society). Reprinted in F. R. Palmer (ed.), Selected Papers of J. R. Firth 1952–1959. London: Longman (Longman’s Linguistic Library). Firth, J. R. (1957b) Papers in Linguistics 1934–1951. London: Oxford University Press. Firth, J. R. (1957c) ‘Applications of general linguistics’, Transactions of the Philological Society. Reprinted in F. R. Palmer (ed.), Selected Papers of J. R. Firth 1952–1959. London: Longman (Longman’s Linguistic Library). Firth, J. R. (1957d) ‘Ethnographic analysis and language with reference to Malinowsi’s views’, in R. W. Firth (ed.), Man and Culture: An Evaluation of the Work of Bronislaw Malinowski. London: Routledge & Kegan Paul. Reprinted in F. R. Palmer (ed.), Selected Papers of J. R. Firth 1952–1959. London: Longmans (Longman’s Linguistic Library). Firth, J. R. (1968) Selected Papers of J. R. Firth 1952–1959 (ed. F. R. Palmer). London: Longman (Longman’s Linguistic Library). Francis, G. (1993) ‘A corpus-driven approach to grammar: principles, methods and examples’, in M. Baker, G. Francis and E. Tognini-Bonelli (eds), Text and Technology: In Honour of John Sinclair. Philadephia: J. Benjamins.

422

bibliography Fries, P. H. (1981) ‘On the status of theme in English: arguments from discourse’, Forum Linguisticum 6.1. Reprinted in J. S. Peto¨fi and E. Soo¨zer (eds), Micro and Macro Connexity of Texts. Hamburg: Buske. Fries, P. H. (1995) ‘Patterns of information in initial position in English’, in P. H. Fries and M. Gregory (eds), Discourse in Society: Systemic Functional Perspective. Norwood, N.J.: Ablex. Garvin, P. (1957) Report of the Seventh Annual Round Table Meeting on Linguistics and Language Study. Washington, D.C.: Georgetown University Press (Languages and Linguistics 9). Gougenheim, R. et al. (1956) L’e´laboration du Franc¸ais e´le´mentaire. Paris: Didier. Gregory, M. J. (1967) ‘Aspects of varieties differentiation’, Journal of Linguistics 3. 2. Haas, W. (1954) ‘On defining linguistic units’, Transactions of the Philological Society. Haas, W. (1957) ‘Zero in linguistic description’, in J. R. Firth et al. (eds), Studies in Linguistic Analysis. Oxford: Blackwell (Special Volume of the Philological Society). Haas, W. (1960) ‘Linguistic structures’, Word 16.2. Haas, W. (1966) ‘Linguistic relevance’, in C. E. Bazell et al. (eds), In Memory of J. R. Firth. London: Longman (Longman’s Linguistic Library). Halliday, M. A. K. (1956) ‘Grammatical categories in Modern Chinese’, Transactions of the Philological Society. Halliday, M. A. K. (1958) Review of Watmough: Language: A Modern Synthesis, Archivum Linguisticum 10. 2. Halliday, M. A. K. (1959a). Review of Herdan: Language as Choice and Chance, Archivum Linguisticum 11. 2. Halliday, M. A. K. (1959b) The Language of the Chinese: ‘Secret History of the Mongols’. Oxford: Blackwell (Philological Society Publications 17). Halliday, M. A. K. (1961) ‘Categories of the theory of grammar’, Word 17.3: 242–92. Halliday, M. A. K. (1964) ‘Syntax and the consumer’, in C. I. J. M. Stuart (ed.), Report of the Fifteenth Annual (First International) Round Table Meeting on Linguistics and Language Studies. Washington, D.C.: Georgetown University Press (Monograph Series on Languages and Linguistics 15). Halliday, M. A. K. (1966) ‘Patterns in words’, The Listener 75.1920. Halliday, M. A. K. (1967–68) ‘Notes on transitivity and theme in English 1–3’, Journal of Linguistics 3 and 4. Halliday, M. A. K. (1975a) ‘Language as social semiotic: towards a general sociolinguistic theory’, in A. Makkai and V. B. Makkai (eds), The First LACUS Forum. Columbia, S.C.: Hornbeam Press. Halliday, M. A. K. (1975b) Learning How to Mean: Explorations in the Development of Language. London: Edward Arnold.

423

bibliography Halliday, M. A. K. (1977) ‘Text as semantic choice in social contexts’, in A. van Dijk and J. Peto¨fi (eds), Grammars and Descriptions. Berlin: de Gruyter. Halliday, M. A. K. (1978) ‘Meaning and the construction of reality in early childhood’, in H. L. Pick and E. Saltzman (eds), Modes of Perceiving and Processing of Information. Hillsdale, N.J.: Erlbaum. Halliday, M. A. K. (1979) ‘Modes of meaning and modes of expresson: types of grammatical structure, and their determination by different semantic functions’, in D. J. Allerton, E. Carney and D. Holdcroft (eds), Function and Context in Linguistic Analysis: Essays Offered to William Haas. Cambridge: Cambridge University Press. Halliday, M. A. K. (1982) ‘The de-automatization of grammar: from Priestley’s An Inspector Calls’, in J. Anderson (ed.), Language Form and Linguistic Variation: Papers Dedicated to Angus McIntosh. Amsterdam: J. Benjamins (Current Issues in Linguistic Theory 15). Halliday, M. A. K. (1984) Listening to Nigel. Sydney: Linguistics Department, Sydney University (mimeo). Halliday, M. A. K. (1985) An Introduction to Functional Grammar, 1st edn. London: Edward Arnold (2nd edn 1994). Halliday, M. A. K. (1990) ‘New ways of meaning: a challenge to applied linguistics’, Journal of Applied Linguistics 6. Reprinted in M. Pu¨tz (ed.), Thirty Years of Linguistic Evolution. Amsterdam: J. Benjamins; and in M. A. K. Halliday, Language in a Changing World. Canberra: Applied Linguistics Association of Australia. Halliday, M. A. K. (1991a) ‘Towards probabilistic interpretations’, in E. Ventola (ed.), Functional and Systemic Linguistics: Approaches and Uses. Berlin: Mouton de Gruyter. Halliday, M. A. K. (1991b) ‘Corpus studies and probabilistic grammar’, in K. Aijmer and B. Altenberg (eds), English Corpus Linguistics: Studies in Honour of Jan Svartvik. London: Longman. Halliday, M. A. K. (1992) ‘Systemic grammar and the concept of a “science of language” ’, Waiguoyu ( Journal of Foreign Languages, Shanghai International Studies University) 2.78. Reprinted in Network 19. Halliday, M. A. K. (1993). Reprint of M. A. K. Halliday 1990. Halliday, M. A. K. (1994) Introduction to Functional Grammar, 2nd rev. edn. London: Edward Arnold. Halliday, M. A. K. (1995) ‘Computing meanings: some reflections on past experience and present prospects’. Paper presented to the second conference of the Pacific Association of Computational Linguistics (PACLING II), University of Queensland, Brisbane. Halliday, M. A. K. and Hasan, R. (1976) Cohesion in English. London: Longman. Halliday, M. A. K. and Hasan, R. (eds) (1980) Text and Context: Aspects of

424

bibliography Language in a Social-Semiotic Perspective. Tokyo: Sophia University Graduate School of Linguistics (Sophia Linguistica 6). Halliday, M. A. K. and James, Z. L. (1993) ‘A quantitative study of polarity and primary tense in the English finite clause’, in J. McH. Sinclair, M. Hoey and G. Fox (eds), Techniques of Description: Spoken and Written Discourse. A Festschrift for Malcolm Coulthard. London and New York: Routledge. Halliday, M. A. K. and Martin, J. R. (eds) (1981) Readings in Systemic Linguistics. London: Batsford Academic. Halliday, M. A. K. and Martin J. R. (1993) Writing Science: Literary and Discursive Power. London: Falmer Press. Halliday, M. A. K. and Matthiessen, C. (1999) Construing Experience through Meaning: A Language-based Approach to Cognition. London: Continuum. Halliday, M. A. K., McIntosh, A. and Strevens, P. (1964) The Linguistic Sciences and Language Teaching London: Longman (Longmans’ Linguistics Library). Halliday, M. A. K. and Poole M. (1978) Notes on Talking Shop: Demands on Language for Use with the Film Australia Production. Lindfield: Australian Film Commission. Harris, Z. S. (1951) Methods in Structural Linguistics. Chicago: University of Chicago Press. Harris, Z. S. (1952) ‘Discourse analysis’, Language 28.1,4. Harris, Z. S. (1954) ‘Transfer grammar’, International Journal of Linguistics 20. Hasan, R. (1965) The Language of Nine-year-old Children. Leeds: Nuffield Foreign Languages Teaching Materials Project, Child Language Survey (Transcript No. 2b). Hasan, R. (1968) Grammatical Cohesion in Spoken and Written English Vol. 1. London: Longman. Hasan, R. (1979) ‘On the notion of text’, in J. Peto¨fi (ed.), Text Versus Sentence: Basic Questions of Text Linguistics. Hamburg: Helmut Buske Verlag. Hasan, R. (1980) ‘The texture of a text’, in M. A. K. Halliday and R. Hasan (eds), 1980. Hasan, R. (1983) ‘Coherence and cohesive harmony’, in J. Flood (ed.), Understanding Reading Comprehension. Newark, DE.: International Reading Association. Hasan, R. (1984a) ‘The structure of the nursery tale: an essay in text typology’, in L. Coveri (ed.), Linguistica Testuale: Proceedings of the 15th Internatonal Congress of the Italian Linguistics Society. Rom: Bulzoni. Hasan, R. (1984b) ‘Ways of saying: ways of meaning’, in R. P. Fawcett, M. A. K. Halliday, S. M. Lamb and A. Makkai (eds), Language as Social Semiotic. Vol. 1, The Semiotics of Culture and Language. London and Dover: Pinter (Open Linguistics Series).

425

bibliography Hasan, R. (1984c) ‘Coherence and cohesive harmony’, in J. Flood (ed.), Understanding Reading Comprehension. Newark, DE.: International Reading Association. Hasan, R. (1985) ‘Lending and borrowing: from grammar to lexis’, in J. E. Clark (ed.), The Cultivated Australian: Festschrift for Arthur Delbridge. Hamburg: Helmut (Buske Beitra¨ge zur Phonetik und Linguistik 48). Hasan, R. (1986) ‘The ontogenesis of ideology: an interpretation of mother child talk’, in T. Threadgold et al. (eds), Semiotics, Ideology, Language. Sydney: Sydney Association for Studies in Society and Culture (Sydney Studies in Society and Culture 3). Hasan, R. (1987) ‘The grammarian’s dream: lexis as most delicate grammar’, in M. A. K. Halliday and R. P. Fawcett (eds), Theory and Description. Vol. 1, New Developments in Systemic Linguistics. London: Pinter. Hasan, R. (1989) ‘Semantic variation and sociolinguistics’, Australian Journal of Linguistics 9.2. Hasan, R. (1992a) ‘Rationality in everyday talk: from process to system’, in J. Svartvik (ed.), Directions in Corpus Linguistics: Proceedings of Nobel Symposium 82, Stockholm, 4–8 August 1991. Berlin: de Gruyter (Trends in Linguistics 65). Hasan, R. (1992b) ‘Meaning in sociolinguistic theory’, in K. Bolton and H. Kwok (eds), Sociolinguistics Today: International Perspectives. London: Routledge. Hasan, R. (1995) ‘The conception of context in text’, in P. H. Fries and M. J. Gregory (eds), Meaning and Choice in Language. Vol. 2, Discourse in Society: Systemic Functional Perspectives. Norwood: Ablex (Advances in Discourse Processes 50). Henderson, E. (1951) ‘The phonology of loanwords in some southeast Asian languages’, Transactions of the Philological Society. Hildyard, A. and Olson, D. R. (1982) ‘On the comprehension and memory of oral vs. written discourse’, in D. Tannen (ed.), Spoken and Written Language: Exploring Orality and Literacy. Norwood, N.J.: Ablex (Advances in Discourse Processes 9). Hill, A. A. (1958) Introduction to Linguistic Structures. New York: Harcourt, Brace & World. Hjelmslev, L. (1953) Prolegomena to a Theory of Language, trans. by F. J. Whitfield. Bloomington: Indiana University Press (Danish original Omkring Sprogteoriens Grundlaeggelse. Copenhagen: Munksgaard). Hjelmslev, L. (1961) Prolegomena to a Theory of Language, trans. by F. J. Whitfield. Madison, Wl.: University of Wisconsin Press (Danish original Omkring Sprogteoriens Grundlaeggelse. Copenhagen: Munksgaard). Hockett, C. F. (1955) Manual of Phonology. Baltimore: Waverly Press (Memoirs of the International Journal of Anthropological Linguistics 2, Indiana University Publications of Anthropology and Linguistics).

426

bibliography Hockett, C. F. (1957) ‘Two models of grammatical description’, in M. Joos (ed.), Readings in Linguistics. Washington: American Council of Learned Societies. Hockett, C. F. (1958) Course in Modern Linguistics. New York: Macmillan. Hockett, C. F. (1961) ‘Linguistic elements and their relations’, Language 37. Hockett, C. F. (1965) ‘Sound change’, Language 41. Huddleston, R. D. (1965) ‘Rank and depth’, Language 41. Huddleston, R. D., Hudson, R. A., Winter, E. O. and Henrici, A. (1968) Sentence and Clause in Scientific English. London: University College London, Communication Research Centre (O.S.T.I. Programme in the Linguistic Properties of Scientific English, Final Report). Hudson, R. A. (1967) ‘Constituency in a systematic description of the English clause’, Lingua 18. 225–50. Hymes, D. H. (1971) ‘Competence and performance in linguistic theory’, in R. Huxley and E. Ingram (eds), Language Acquisition: Models and Methods. London and New York: Academic Press. Ikegami, Y. (1980) ‘On some categories for describing the semolexemic structure’, in J. E. Copeland and P. W. Davis (eds), Papers in Cognitivestratificational Linguistics. Houston: Rice University (Rice University Studies, 66–2). Jakobson, R. (1963) ‘Implications of language universals for linguistics’, in J. Greenberg (ed.), Universals of Language. Cambridge, MA.: MIT Press. Jespersen, O. (1909–43) A Modern English Grammar. London: Allen & Unwin. Johnston, T. (1990) ‘A study of the grammar of Auslan (Australian sign)’. Unpublished PhD dissertation, University of Sydney. Kluckhohn, C. and Leighton, D. (1962) The Navaho (The National History Library). Garden City: Doubleday. Kress, G. R. (1976) Halliday: System and Function in Language. Selected Papers. London: Oxford University Press. Lamb, S. M. (1964a) ‘The sememic approach to structural semantics’, American Anthropologist 66. Lamb, S. M. (1964b) ‘On alternation, transformation, realization and stratification’, in C. I. J. M. Stuart (ed.), Report of the Fifteenth Annual (First International) Round Table Meeting on Linguistics and Language Studies. Washington, D.C.: Georgetown University Press (Monograph Series on Languages and Linguistics 17). Lamb, S. M. (1988) Discussion in M. A. K. Halliday, S. Lamb and J. Regan (eds), In Retrospect: Using Language and Knowing How. Claremont, CA.: Claremont Graduate School (Issues in Communication 12). Lamb, S. M. (1992) Outlines of a Cognitive Theory of Language: A Work in Progress (pre-publication draft). Lemke, J. L. (1984) Semiotics and Education. Toronto: Victoria University

427

bibliography (Toronto Semiotic Circle Monographs, Working Papers and Prepublications 2). Lemke. J. L. (1985) ‘Ideology, intertextuality and register’, in J. D. Benson and W. S. Greaves (eds), Systemic Perspectives on Discourse. Norwood, N.J.: Ablex. Lemke, J. L. (1990) Talking Science: Language, Learning and Values. Norwood, N.J.: Ablex. Lemke, J. L. (1993) ‘Discourse, dynamics, and social change’, Cultural Dynamics 6.1–2 (Language as Cultural Dynamic). Linde, C. and Labov, W. (1975) ‘Spatial networks as a site for the study of language and thought’, Language 51. Longacre, R. E. (1964) Grammar Discovery Procedures: A Field Manual. The Hague: Mouton (Janua Linguarum 33). Lyons, J. (1963) Structural Semantics: An Analysis of Part of the Vocabulary of Plato. Oxford: B. Blackwell (Philological Society Publications 20). Lyons, J. (1968) Introduction to Theoretical Linguistics. London: Cambridge University Press. Malinowski, B. (1935) The Language of Magic and Garden. Vol. 2, Coral Gardens and their Magic. New York: American Book Co. Reprinted as The Language of Magic and Gardening. Bloomington: Indiana University Press (Indiana University Studies in the History and Theory of Linguistics). Mann, W. C. (1983) ‘Inquiry semantics: a functional semantics of natural language grammar’, Proceedings of the First Annual Conference: Association for Computational Linguistics. European Chapter. Marina del Rey: Information Sciences Institute, University of Southern California (ISI/RR 83–105). Mann, W. C. and Matthiessen, C. M. I. M. (1983) Nigel: A Systemic Grammar for Text Generation. Marina del Rey: Information Sciences Institute, University of Southern California (ISI/RR 83–105). Martin, J. R. (1980) ‘How many speech acts?’, University of East Anglia Papers in Linguistics 14–15. Martin, J. R. (1983a) ‘Conjunction: the logic of English text’, in J. S. Peto¨fi and E. So¨zer (eds), Micro and Macro Connexity of Texts. Hamburg: Buske. Martin, J. R. (1983b) ‘The development of register’, in R. O. Freedle and J. Fine (eds), Developmental Issues in Discourse. Norwood, N.J.: Ablex. Martin, J. R. (1984a) Process and text: two aspects of human semiosis’, in J. D. Benson and W. S. Greaves (eds), Systemic Perspectives on Discourse: Selected Theoretical Papers from the Ninth International Systemic Workshop. Norwood, N.J.: Ablex. Martin, J. R. (1984b) ‘Functional components in a grammar: a review of deployable recognition criteria’, Nottingham Linguistic Circular 13. Martin, J. R. (1988) ‘Grammatical conspiracies in Tagalog: family, face and fate – with regard to Benjamin Lee Whorf ’, in J. D. Benson, M. J.

428

bibliography Cummings and W. S. Greaves (eds), Linguistics in a Systemic Perspective. Amsterdam: J. Benjamins (Current Issues in Linguistic Theory 39). Martin, J. R. (1992) English Text: System and Structure. Amsterdam: J. Benjamins. Martin, J. R. (1993) ‘The model’, in M. A. K. Halliday and J. R. Martin (eds), 1993. Martin, J. R. (forthcoming), ‘Lexical cohesion, field and genre: parcelling experience and discourse goals’, in J. E. Copeland (ed.), Text Semantics and Discourse Semantics: Proceedings of the Second Rice Symposium in Linguistics and Semiotics. Houston: Rice University. Martin, J. R. and Matthiessen, C. (1992) ‘Systemic typology and topology’, in F. Christie (ed.), Literacy in Social Processes. Darwin: Northern Territory University, Centre for Studies in Language Education. Mathesius, V. (1928) ‘On linguistic characteriology with illustrations from Modern English’, in Actes du premier congre`s international de linguistes a` la Haye. Reprinted in J. Vachek (ed.), A Prague School Reader in Linguistics. Bloomington: Indiana University Press. Mathesius, V. (1964) ‘On the potentiality of the phenomena of language’, trans. by J. Vachek, in J. Vachek (ed.), A Prague School Reader in Linguistics. Bloomington: Indiana University Press. (Czech original ‘O potencialnosti jevu jazykovych’, Proceedings of the Philosophy and History Section, Czech Academy of Sciences.) Matthews, P. H. (1961) ‘Transformational grammer’, Archivum Linguisticum 13.2. Matthews, P. H. (1966) ‘The concept of rank in “neo-Firthian linguistics” ’, Journal of Linguistics 2.1: 101–18. Matthiessen, C. (1988) ‘Representational issues in systemic functional grammar’, in J. D. Benson and W. S. Greaves (eds), Systemic Functional Approaches to Discourse. Norwood, N.J.: Ablex. Matthiessen, C. (1991a) ‘Language on language: the grammar of semiosis’, Social Semiotics 1.2. Matthiessen, C. (1991b) ‘Lexico (grammatical) choice in text generation’, in C. L. Paris, W. R. Swartout and W. C. Mann (eds), Natural Language Generation in Artificial Intelligence and Computational Linguistics. Dordrecht: Kluwer. Matthiessen, C. (1991c) ‘Developments in systemic functional theory: a discussion of the dimensions of change’, Network 17. Matthiessen, C. (1992) ‘Interpreting the textual metafunction’, in M. Davies and L. Ravelli (eds), Advances in Systemic Linguistics: Recent Theory and Practice. London: Pinter. Matthiessen, C. (1993a) ‘The object of study in cognitive science in relation to its construal and enactment in language’, in M. A. K. Halliday (ed.), Cultural Dynamics 6.1–2 (Language as Cultural Dynamic).

429

bibliography Matthiessen, C. (1993b) ‘Register in the round: diversity in a unified theory of register analysis’, in M. Ghadessy (ed.), Register Analysis: Theory and Practice. London: Pinter. Matthiessen, C. (1995) ‘Theme as an enabling resource in ideational “knowledge” construction’, in M. Ghadessy (ed.), Thematic Development in English Texts. London: Pinter. Matthiessen, C. and Bateman, J. (1992) Systemic Linguistics and Text Generation: Experiences from Japanese and English. London: Pinter. Matthiessen, C. and Nesbitt, C. (1996) ‘On the idea of theory-neutral descriptions’, in R. Hasan, C. Cloran and D. G. Butt (eds), Functional Descriptions: Theory into Practice. Amsterdam: J. Benjamins. McIntosh, A. (1956) ‘The analysis of written Middle English’, Transactions of the Philological Society. McIntosh, A. (1961) ‘Patterns and ranges’, Language 37.3. Melrose, R. (1979) ‘General word and particular text: a study of cohesion in academic writing’. Unpublished Phd dissertation, University of Sydney. Mitchell, T. F. (1957) ‘The language of buying and selling in Cyrenaica: a situational statement’, Hesperis 26. Reprinted in T. F. Mitchell, Principles of Firthian Linguistics. London: Longman (Longman’s Linguistics Library 19). Mitchell, T. F. (1958) ‘Syntagmatic relations in liguistic analysis’, Transactions of the Philological Society. Mukarˇovsky´, J. (1977) The Word and Verbal Art: Essays, trans. by J. Burbank and P. Steiner. New Haven, CT.: Yale University Press. Mullie, J. (1937) The Structural Principles of the Chinese Language, Vol. 2. Beijing, Peitang Lazarist Press. Needham, J. (1958) ‘The translation of old Chinese scientific and technical texts’, in A. H. Smith (ed.), Aspects of Translation. London: Secker & Warburg (Studies in Communication 2). Nesbitt, C. and Plum, G. (1988) ‘Probabilities in a systemic grammar: the clause complex in English’, in R. P. Fawcett and D. J. Young (eds). Nida, E. (1960) A Synopsis of English Syntax. Norman: University of Oklahoma, Summer Institute of Linguistics (Linguistic Series 4). Nystrand, M. (1987) The Structure of Written Communication: Studies in Reciprocity between Writers and Readers. London: Academic Press. Oldenburg, J. (1987) ‘From child tongue to mother tongue: a case study of language development in the first two and a half years.’ Unpublished PhD dissertation, University of Sydney. O’Toole, M. (1994) The Language of Displayed Art. London: Leicester University Press. Painter, C. (1984) Into the Mother Tongue: A Case Study in Early Language Development. London: Pinter (Open Linguistics Series).

430

bibliography Painter, C. (1996) ‘The development of language as a resource for thinking: a linguistic view of learning’, in R. Hasan and G. Williams (eds), Literacy in Society. London: Longman. Palmer, F. R. (1958a) ‘Linguistic hierarcy’, Lingua 7. Palmer, F. R. (1958b) ‘Comparative statement and Ethiopian semitic’, Transactions of the Philological Society. Palmer, F. R. (1964a) ‘ “Sequence” and “order” ’, in C. I. J. M. Stuart (ed.), Report of the Fifteenth Annual (First International) Round Table Meeting on Linguistics and Language Studies. Washington, D.C.: Georgetown University Press (Monograph Series on Languages and Linguistics 17). Palmer, F. R. (1964b) ‘Grammatical categories and their phonetic exponents’, in H. G. Lunt (ed.), Proceedings of the Ninth International Congress of Linguists. The Hague: Mouton. Pike, K. L. (1959) ‘Language as particle, wave and field’, Texas Quarterly 2. Quirk, R. (1959) ‘Review of W. Nelson Francis: “The structure of American English” ’, Archivum Linguisticum 11. Reddy, M. J. (1979) ‘The conduit metaphor: a case of frame conflict in our language about language’, in A. Ortony (ed.), Metaphor and Thought. Cambridge: Cambridge University Press. Robins, R. H. (1953) ‘Formal divisions in Sundanese’, Transactions of the Philological Society. Robins, R. H. (1957a) ‘Vowel nasality in Sundanese’, in J. R. Firth et al. (eds), Studies in Linguistic Analysis. Oxford: Blackwell (Special Volume of the Philological Society). Robins, R. H. (1957b) ‘Aspects of Prosodic Analysis’. Vol. 1, Proceedings of the University of Durham Philosophical Society 1. Durham: Durham University Philosophical Society (series B). Robins, R. H. (1959) ‘Status of grammar’, Lingua. Robins, R. H. (1961) ‘Syntactic analysis’, Archivum Linguisticum 13. Roget, P. M. (1960) Thesaurus of English Words and Phrases. London: Longman. Russell, B. (1925) ABC of Relativity. London: Allen & Unwin (3rd revised edn 1977). Sadovnik, A. R. (ed.) (1995) Knowledge and Pedagogy: The Sociology of Basil Bernstein. Norwood, N.J.: Ablex. Schapiro, L. (1972) Totalitarianism. London: Pall Mall Press. Sinclair, J. McH. (1966) ‘Beginning the study of lexis’, in C. E. Bazell et al. (eds), In Memory of J. R. Firth. London: Longman (Longman’s Linguistic Library). Sinclair, J. McH. (1991) Corpus, Concordance, Collocation. Oxford: Oxford University Press. Sinclair, J. McH. (1992) ‘Trust the text: the implications are daunting’, in M. Davies and L. Ravelli (eds), Advances in Systemic Linguistics: Recent Theory and Practice. London: Pinter.

431

bibliography Sinclair, J. McH. and Coulthard, R. M. (1975) Towards an Analysis of Discourse: The English Used by Teachers and Pupils. London: Oxford University Press. Sinclair, J. McH., Jones, S. and Daley, R. (1970) English Lexical Studies. Birmingham: University of Birmingham, School of English (Report to the office of Scientific and Technical Information). Smart, N. (1968) Secular Education and the Logic of Religion. London: Faber & Faber. Sugeno, M. (1995) ‘Intelligent fuzzy computing’. Paper presented to the second conference of the Pacific Association of Computational Linguistics (PACLING II), University of Queensland, Brisbane. Svartvik, J. and Quirk, R. (eds) (1980) A Corpus of English Conversation. Lund: C. W. K. Gleerup (Lund Studies in English 56). Svoboda, A. (1968) ‘The hierarchy of communication units and fields as illustrated by English attributive constructions’, Brno Studies in English 7. 49–101. Sweet, H. (1891) A New English Grammar: Logical and Historical. Oxford: Clarendon. Tannen, D. (1982) ‘Oral and literate strategies in spoken and written narrative’, Language 58. Taylor, C. V. (1979) English for Migrant Education Curriculum. Canberra: Education Development and Research Committee (Report 18). Thibault, P. J. (1992) ‘Grammar, ethics and understanding: functionalist reason and clause as exchange’, Social Semiotics 2.1. Turner, G. J. (1973) ‘Social class and children’s language of control at age five and age seven’, in B. Bernstein (ed.). Ure, J. (1971) ‘Lexical density and register differentiation’, in G. E. Perren and J. L. M. Trim (eds), Applications of Linguistics: Selected Papers of the Second International Congress of Applied Linguistics, Cambridge 1969. London: Cambridge University Press. Ure, J. N. and Ellis, J. O. (1979) ‘Register in descriptive linguistics and linguistic sociology’, in O. U. Villegas (ed.), Issues in Sociolinguistics. The Hague: Mouton. Vachek, J. (1966) The Linguistic School of Prague. Bloomington: Indiana University Press. van Leeuwen, T. (1988) ‘Music and ideology: notes towards a socio-semiotics of mass media music’, Sydney Association for Studies in Society and Culture Working Papers 2:1/2. Winter, E. (1977) ‘A clause–relational approach to English texts: a study of some predictive lexical items in written discourse’, Instructional Science 6. Yngve, V. H. (1960) ‘A model and an hypothesis for language structure’, Proceedings of the American Philosophical Society 104.5.

432

INDEX

abstraction 38, 41, 52–69, 185, 296, 351, 411 degree of 46–7, 53, 114, 158 higher order of 185, 296 level of 220–1 accusative 293 active 28–9, 400 and passive 115, 182–3 Malinowski’s 227, 236 actor 113, 176, 179–94, 204, 212, 224, 299 additive 223 adjective 55, 108, 187, 302, 336, 342, 384 adjectival 343 adjunct 47, 56, 96, 98–9, 104, 225, 238 adverb 28–9, 95, 180 adverbial 28, 32, 343 group 96, 103, 180, 263 adversative 225 Affected 187–8 Agent 238, 244, 277, 298, 300, 381 allophone 68 alternative 223 ambiguity 104, 159, 163–4, 185–6, 343, 385, 399–400, 410, 416 ambiguous 298, 346–8, 385, 399–400 apposition 215 appositive 225 Arabic 223, 292

arbitrary 3, 47, 114, 294, 300, 378–9, 397 art 5, 357, 389, 402 article 98, 299 articulation 354, 357, 364, 366 articulatory 7, 324 attributive 185, 229, 243, 276, 336, 413 autostable 358, 362 auxiliary 28, 268 noun 32–4 axis 23, 95, 97–8, 100, 109–10, 120, 163–4, 221, 402–3 benefactive 181 Beneficiary 178–81, 193, 238 Bernstein, B. 7–8, 175, 403, 412 Bloomfield, L. 65, 219 Bloomfieldian 4, 65 Boas, F. 246, 262, 302, 311, 325 bonobo 3 branching singulary 122–5 Bu¨hler, K. 173–4, 226–7, 234 Cantonese 32–4, 202 Carrier 244, 413 category 12–13, 21–34, 37–72, 95, 97–9, 106, 160–4, 167, 170, 177–9, 186–7, 200–2, 209, 215, 223–33, 242–3, 284–5, 291–313,

433

index category (cont.) 335, 343, 351, 354–6, 385, 397, 399–400, 408, 410, 412 causal consequential 225 external 233 relation 357 causality vs. redundancy 357 causation 400 causative 186, 225 Cause 238, 305 chain 45, 67, 95, 97–100, 208, 224, 232–5, 244–6, 292, 352, 357, 360 chain-exhausting 166 chain-exhaustive 120 chaos 353, 365–6 Chinese 25–33, 167, 202, 294–301, 372–3, 385, 415–16 box 43 choice 33, 51–3, 95–9, 100, 116–18, 163, 174, 182, 192, 198–200, 228, 239–40, 262, 268, 279, 283, 301, 304–7, 310, 328, 348, 356, 364, 377, 380, 395, 404 chooser 301 points 395 Chomsky, N. 4, 6, 72, 106, 112, 304 class(es) 24–34, 41, 45, 49–68, 95–104, 106–13, 123–5, 159–68, 184–5, 212, 222, 234, 294, 307, 371, 374–5, 397, 412–14 class-defining 101 class-sequence 109 class-structure 50 class-type 108 natural 391 clause 26–32, 45–58, 68–70, 95–104, 109–15, 121–4, 175–94, 205–17, 219–22, 228–47, 262–83, 297–8, 305, 308–9, 329–48, 352, 371–7, 394–5, 414 clause-classes 26 clause-final 238, 252 clause-initial 238 clause-like entities 221 clause-to-text analogy 241

complex(es) 215, 217, 232, 242, 262–70, 281–2, 332–6, 341, 343, 371 type(s) 181–7, 233, 395, 414 English 47, 70, 97–9, 186, 190, 239, 297–8, 394 code 8, 196, 226, 233, 296, 309, 350, 412 cognitive 116, 197, 231, 276–7, 282 coherence 222–6, 232–5, 244–5 coherent 41, 216, 340 cohesion 223–44, 263, 271, 280–5 cohesive 170, 175, 342 cohesive harmony 224, 232, 235 collocation(al) 33, 60–1, 158–70, 201, 282 communication 175, 189–200, 193, 199–201, 294, 351 communicative 236, 283 complement 47, 96–102, 113, 238, 374 clauses 336 conative 226–7, 236 conceptual 173, 216, 245, 301, 363, 378 concord 55, 70–1 tone 266 constituent structure 113–16, 121–6, 204–6, 216, 238 construe 3, 12–13, 306, 310, 353, 357, 362–6, 369–77, 382, 390–1, 393, 396, 398–9, 402, 407–14 de-construe 373, 355, 363 context 23–33, 39, 56, 70, 96, 99, 174, 190, 198–202, 220–1, 225–9, 243–4, 293, 300–1, 306–11, 324–5, 336, 344, 349–51, 400–5 function in 398 of situation 29, 201, 221–9, 243, 263, 283, 311, 357–9, 405 social 10, 201–2 conversation 7–8, 169, 326, 346, 351, 371 coordination 215 coreferential 224 corpus 8, 9, 38, 466–8, 412 corpus-based 159–60, 362, 407 grammar 412 cortex 272

434

index creolized 362 crescendo 228, 240, 271 criteria 24–34, 40, 42, 45, 48, 55–6, 58, 61, 67–72, 96–7, 107, 114, 119, 161, 166–8, 170, 178–9, 341–2, 398–401, 410 (situational-) contextual 32–3 cryptogrammatic 407 crypotype 302 crypotypic 360 crystalline 303, 336, 350 Davidse, K. 380, 413 declarative 109, 111, 189, 233, 268, 273, 305 deep grammar 116 structure 106, 116 definite 98, 185, 299–300, 307 deictic 55, 99–101, 243 deicticity 272 postdeictic 100 predeictic 100 delicacy 40–1, 48–70, 98–9, 114–15, 158–9, 165, 223, 285, 293, 396, 402–6, 415 degree of 48, 54, 57, 69, 99, 159, 223, 404 most delicate grammar 49, 54, 59, 405 depth 48, 58, 101–4, 107, 120, 285, 327, 333 depth-ordered 97 determiner 99, 104 diachronic 23, 324 synchronic-diachronic 22 dialect 7, 32–4, 202, 377 British rural dialects 377 Wu and Yueh dialect groups 34 dialogue 225–33, 239–40, 271, 283, 308, 325, 335, 341 dictionary 54, 158, 160, 165, 186, 392–3, 400, 407 diglossia 296 dimension(s) 22, 42–3, 100–1, 159, 161, 164, 188, 194, 222, 225, 232, 241, 328, 355–62, 391 of abstraction 59, 66, 96

of choice 356 of classes 26–9 of realization 365 diminuendo 228, 240, 270 diminuendo-crescendo 233, 243 discourse 7–8, 10–11, 175, 189, 193–4, 199–209, 225–8, 239, 242–3, 245–7, 261, 268, 270–3, 282, 285, 292, 294, 296, 302–3, 311, 324–5, 329–31, 335–50, 365–6, 369, 371, 377, 386–7, 391, 396–7, 407, 412, 415 spoken 270, 324, 331, 335–7, 340, 403 spontaneous 325, 337–40 written 331, 335–6, 340–6, 348–50 ecolinguistics 9 Eggins, S. 411, 413 eidological 232 eidon 231 eidos 231 ellipsis 181, 225, 232, 237, 281–2 Ellis, J. 174, 227, 391 embedding 126, 343–4 embedded 227, 270, 280, 329, 341–2 empirical(ly) 119, 242 enact 3, 5, 356, 382, 390, 392, 407 encode 202 encoded 202, 204, 220, 240–1, 307, 342, 346 encoding 202, 235, 292–3 endocentric word groups 243 English 33, 44–8, 54–5, 58, 60, 70, 96–104, 113–15, 120, 160, 167, 175–92, 202, 206, 209–10, 214–17, 228–46, 266, 270, 276, 279, 282, 297–312, 323–30, 343, 347–50, 362, 369–82, 385, 392–400, 405–7, 410, 413–16 modern 187–8, 191, 261, 375–7 spoken 70, 101, 326, 343, 350 written 101, 261, 330, 343, 350 equative 183, 185, 243, 300 ergative 28–9, 186–8 and transitive 380–1, 400 see also voice

435

index ergativity 312 Eskimo folk tales 231 ethnographic 173, 226, 231, 236, 386 ethnographer 230 ethological 227, 232, 236 ethos 4, 231 evolution 65, 303, 349, 355–62, 370, 377–8, 390, 392, 413 evolutionary semogenesis 362 exemplificatory 45, 70, 72 existential 243, 275 exophoric 201 experiential 198–217, 224, 276, 284, 377, 379, 390, 400, 405 exponence 41, 45, 53–72, 97, 221–2, 293 Firth’s concept of 352 scale 57–61, 66 see also rank exponent(s) 23–7, 45–72, 97, 116, 221 expressive 227, 236 Extent 204, 212, 238 extralinguistic 31, 236, 295 extratextual 39–40

linguistic 111 rhetorical 226–34 speech 233–9, 268, 273, 307 syntactic 107 textual 175–6, 182, 193–5, 237, 273 functional categories 209, 300 component(s) 200–1, 211, 215, 241 element(s) 175, 225, 242, 262 environment 110, 122 interpretation 200, 235 labels 107, 203 schemata 236 semantic(s) 209, 237, 311 tenor 227, 231 theories 226, 235–6 variety 301, 307 variational 403 fuzzy 3, 210, 417 computing 402, 416 set(s) 409–10, 417

Field field, tenor, mode 201, 217, 221, 227–31, 243, 283–4, 364 particle, field and wave 209–11, 241 Fillmore, C. 178–9, 181, 187, 194 Firth, J. R. 7, 12, 21, 24, 25, 38, 67, 106, 109, 110, 158, 170–1, 174–5, 210, 219, 262, 296, 301, 307, 311, 384, 387, 398, 403, 414 foreground 235 foregrounded 234 foregrounding 230, 349, 379, 409 free/bound 27–8 French 372, 373, 381, 415 Fries, Peter 228, 233, 245–6, 371 function 12, 107–16, 122, 173, 202–4, 212–47, 262, 281, 292–4, 299–308, 329–50, 376–82, 389–416 experiential 202, 224 ideational 175–93 imaginative 227 interpersonal 272, 382

gender 302 generative 37 genetic 22, 32–4 genitive 293, 300, 336 genre(s) 209, 222, 230, 234, 242, 312, 357, 413 gesture 354, 356, 366 given and new 29, 190–1 Given 192–4, 207, 209, 238, 270–1, 376 Gleason, H. A. 222, 244 Goal 113, 176, 178–83, 186–8, 203–4, 212, 300, 312, 376, 379, 381 Goal/Medium 305 goods and services 199, 273 grammar descriptive 30, 163, 415 lexis 37–67, 165, 379, 404, 410 phonology 56–68, 220, 239 protocol version 415 rank 121–7 semantics 220, 239, 306, 324 systemic(-functional) 261–2, 332 theory 41, 44, 67, 370, 401–2, 416 see also grammatics

436

index grammatics 11, 365–6, 369–78, 384–417 trinocular 402–9 graphology 39 Greek 294–6 ancient 292, 365 ancient Greece 371 alphabet 104 Hasan, R. 11, 175, 222–5, 229, 231, 242, 244, 261, 285, 351, 404, 406, 411–13 hearer 199, 205, 207, 301 hearer-oriented 199, 207, 240 hesitation 205, 337–8, 340 hierarchy 25–6, 42–4, 56, 59–60, 110–11, 115, 119–24, 166, 213, 228, 242 Hjemslev, L. 4, 5, 12, 106, 109, 110, 112, 219, 236, 262, 301, 312, 354, 401 Hockett, C. 106, 112, 219, 221 Huddleston, R. D. 120, 125, 215, 349 Hudson, R. 204, 349 hyponomy 12, 226 hyponym 226, 282 hyponymic 396 hypotaxis 107, 266, 302, 327, 332–3, 333, 343–4, 362, 410 hypotactic 213–17, 242, 266, 266, 282, 333 see also parataxis iconicity 312 ideational 210–11, 216–17, 227–44, 268, 298, 308, 311, 348–9, 356, 364, 382, 384, 392, 401 component 186, 198–200, 208, 237, 242 features 243 function 175, 177, 189, 193 meaning 177, 188, 193, 199, 201, 217, 229, 295 semantics 13 structure 231, 241–2, 244 voice 230 indefinite 185, 299

indeteminacy 179, 239, 310, 335, 338, 365–6, 392, 399–402, 409–11 indicative 111 Indo-European 35, 236 infant 3–4, 304, 323, 388–9 inferential relations 11 information flow 184 focus 207–8, 233, 270 prominence 271 retrieval 170 structure 192, 216, 233, 266, 269–70, 305, 312 theory 40, 42, 70, 72 unit 192, 207, 215, 262, 266, 270–2 inherence 302 instantial 10, 279, 405 instantiation 12, 262, 352, 359, 362, 387, 410, 412–13 interaction 175, 189, 199, 201, 210, 216, 227, 230, 310, 377, 382, 390, 392 interactional 353, 363 interlevel 39, 56, 67 interpersonal 175, 182, 189, 199–217, 227–45, 268, 273, 284, 307–8, 311, 337, 356, 364, 382, 384, 390, 392, 401, 413 interrogative 111–13, 189, 191, 200, 233, 268, 274 intersubjective 245, 354–5 intonation 55, 114, 123–5, 192–3 contour(s) 205, 217, 238, 270, 399 pattern 114, 123, 193 of pause 26 and rhythm 114, 262–3, 405 intransitive 28, 164, 187 intricacy 331–7, 341, 343 Jacobson, R. 189, 226, 301 Jesperson, O. 298–9 juncture 55, 68 see also tone Labov, W. 222, 350 Lamb, S. 5–6, 108, 112, 198, 355, 378, 383, 389

437

index language 1–13, 21–30, 37–72, 95–104, 106–17, 119–25, 158–71, 173–95, 197–217, 219–20, 226–46, 261, 282–5, 291–313, 323–51, 353–66, 369–83, 384–417 child language development 3–4, 310, 411 “language-in-action” 229, 329 teaching 7, 170, 403 natural 175, 198, 237, 296–309, 364, 369, 389–417 spoken 4, 7, 10, 261, 302–6, 323–48, 366, 406 written 43, 227, 323–51 langue 44, 236–7, 403, 412 Latin 25, 95 Lemke, J. L. 306–7, 312, 352, 356, 358, 370, 382, 387, 391, 412–13 lexical density 229, 327–32, 335, 341–2 lexicogrammar 3, 6, 8–12, 163, 168–9, 185, 217–21, 231, 233, 239, 241, 243, 246, 261–2, 281–5, 294, 298, 341, 345, 357–8, 369, 378–9, 387, 389, 391, 393, 396–8, 401, 403–4, 408, 411–12 lexicoreferential 224, 232, 244 lexico semantic 246 lexico syntactic 328 lexicogrammatical entities 241 lexicogrammatical system 345 see also wording(s) lexis 37, 39–43, 49, 54–61, 67, 98, 104, 158, 162–70, 379, 404, 410 lexis-based 407 lexis-driven 407 linguistics 1–12, 21–30, 37–40, 42, 46, 51, 56, 60, 68, 72, 95, 98, 105, 118–19, 123, 158–60, 170, 174–7, 219, 236–7, 243, 292, 294–8, 307, 311–13, 324–5, 351, 365, 384–6, 406, 411, 415–17 listener 175, 189–90, 192–3, 209, 227, 237, 240, 243, 270–3, 338, 344, 350 listener-prominence 240 literate 323, 325, 405 literature 2, 5, 37, 230, 313, 323

Location 238 logic 3, 6, 8, 198–9, 212, 399, 402, 416 logical-semantic 262, 264, 266, 281–2 logo-centric 414 logo-genetic 360 Longacre, R. E. 106, 222, 242 Lyons, J. 114, 178, 194 McIntosh, A. 227, 396 Malinowski, B. (Kasper) 173, 226–7, 236–7, 262, 311 Mann, W. C. 301, 394–5 Manner 204, 212, 238 Martin, J. 13, 223, 225–7, 231–3, 242, 285, 310, 312, 344, 357, 373–4, 389, 392, 395, 397, 412–13 Mathesius, V. 190, 202, 311, 324–5, 371 Matthiessen, C. 3, 13, 301, 365, 370, 373, 390, 397, 407, 411, 413, 416–17 meaning 3, 9–13, 23, 39–41, 45, 61, 66, 71, 95–8, 121, 170, 174–93, 196–207, 221–4, 227, 231, 237–47, 262, 276, 294–5, 297–312, 323, 325, 346, 348, 350–66, 369–74, 386–93, 397–414 act of 201, 348, 354, 356, 391, 414 choices in 307, 310 contextual 40, 61 formal 40, 71 meaning-creating 8, 354 meaning-making 397 meaning-oriented 350 Medium 238, 274, 277, 380–1 Melrose, R. 227–8, 242 metafunction 12–13, 268, 284, 310–11, 356–8, 364–5, 382, 390–2, 411, 413 metalanguage 30, 293–300, 309–13, 354, 366, 369, 378, 384, 415 metaphor clause as 222, 234, 245 conduit 293–4, 297 grammatical 12, 280, 282, 345–51, 358–60, 397, 400–1 rhetorical 328 metaredundancy 352, 356–8, 362

438

index microclass 101, 159 Mitchell, J. F. 205, 223 modal 216, 242, 245–6, 300, 340, 399 subject 190–1, 194 modality 200, 205, 215, 230, 234, 237–8, 242, 245, 268, 271–4, 278–82, 399 visual 312 mode 221, 227, 229, 283 active 199, 237 experiential 202 interpersonal 205 logical 211–15 metaphoric 397 reflective 199, 237 textual 206 model 159–67 bricks-&-mortar 404 modifier 47, 108 modifier-head 108–9 mood 189–90, 194, 199–200, 205, 215, 230, 233, 237, 242, 262, 268, 271–3, 364 morpheme 45, 53–69, 71, 95–6, 103, 124–5, 164, 219–20, 295, 298 morphemics 69

narrative 222–4, 227, 230–1, 237, 242, 245, 309, 327, 375 Navaho 312 Nesbitt, C. N. 362, 407 network 12, 40, 48, 109, 111–16, 174, 200–1, 212–13, 225–30, 242, 301, 310, 363, 390, 393–410 neural 5, 370, 395 New 192–3, 207, 238, 240, 243, 270–1, 300, 305, 307, 326 news 231, 238, 244, 269–70, 376 Nigel 301, 353–4, 362–4 nominal 32–4, 54, 70, 104, 190, 309, 342, 348, 371 group 51, 55, 70, 96–9, 102–4, 180, 204, 212, 243, 263, 279, 342–3, 396, 410 nominative 189, 293, 300 non-finite 40, 278, 326 noun 28–34, 51, 55, 95, 108, 167, 180,

185, 190, 280, 292–3, 295, 300, 307, 342–3, 397, 410 order 43, 46, 116, 213–14, 269, 362, 364–6, 390 higher order consciousness 388–9 higher order semiotic 313, 388, 410–11, 409 social 356, 390 word 29 orthography 39, 261, 400 orthographic 46, 124, 406 orthographies 120 O’Toole, M. 389, 402 Painter, C. 363, 397 Palmer, F. R. 106, 108, 116 paradigmatic 49, 61, 106, 109–16, 120–3, 160–3, 216, 262, 378, 402–5, 411 paragraph 211, 228, 234, 243, 246 paraphrase 114, 325, 346 parataxis 266, 327, 333, 343, 362 paratactic 213, 216–17, 242, 266, 282, 333 see also hypotaxis parole 236, 324–5, 365, 403, 412 see also langue parser 407 parsing 246, 395, 407 participant 23, 26, 31, 178–93, 200–2, 215, 224, 227, 230, 237, 244, 263, 268, 273–4, 277, 303, 307, 374–5, 377, 380–1 participial 326 participle 336, 381 particle 204, 209, 211, 239 particulate 211, 215, 232, 239, 242 passive 28–9, 35–7, 182, 190–3, 298–300, 400–1 imperative 190 tenses 375 verbs 375 past 102, 273, 279, 326, 379, 400 pedagogical 32, 300 Pekingese 27, 29, 31–4 Penman 301, 394, 417

439

index person 70, 230, 236–7, 268, 337, 353, 378 phatic 226 phoneme 65–6, 68, 71, 96, 219–20, 222 supra-segmental 68 phonetic 26, 71 phonetics 39, 324 phonetic/kinetic 354 phonic 25, 39, 67 phonology 3, 6, 33, 37–9, 56, 60, 65–71, 104, 120, 206, 219–20, 239, 262, 297, 324, 357, 387, 389, 408 Chinese 292 diachronic 324 phonological 32–3, 39, 55, 66–71, 114–15, 208, 213, 231, 239, 293, 324, 338, 345, 357 phonological-lexical 22 phonological-morphological 22 prosodic 67, 71 see also grammar phrasal verb 101, 376 phrase 45, 68, 121, 125–6, 180, 206, 263, 278, 298, 336, 342, 410 phylogenetic 355, 360 Pike, K. 106, 174, 211, 239, 242 Pinyin 33 plosive 34 Plum, G. 327, 362 plural 295, 307 polarity 12, 61, 230, 279, 363, 398 polarity carrying element 189, 266 polysemy 294, 297, 386, 405 possession 327, 371, 373 possessive 100, 275 postmodifier 342 pragmatic 10, 11, 412 Prague School 174, 190, 262, 292, 299, 311, 371 predicate 102, 188, 194, 298–9 predicator 47, 99, 101–2, 113, 238 preposition 193, 376–7 prepositional 178, 180, 193, 263, 298, 342 present 70, 102, 110, 274–5, 326, 379 Priestley, J. B. 234, 245 process 187–8, 203–5, 212, 224, 238, 244, 274, 305, 346

mental 183–8, 282, 398 relational 229, 234, 364, 413 semogenic 227, 355, 364 social 201, 227, 386, 407 type 177, 235, 243, 263, 274, 277, 282, 364, 399 pronominal 29, 31, 305 pronoun 24, 29, 31, 326, 372 prosodic 67, 71, 115, 205–17, 238, 364 prosodies 206, 232 protolanguage 4, 304, 349–51, 353–65, 378, 389 protolinguistic 354–64 question 174–6, 189, 191–3, 233, 268 question-and-answer 283 Quirk, P. 9, 371, 407 Range 238, 300, 348 rank 41–69, 95–103, 115–16, 118–26, 159–66, 212–15, 221–2, 242, 268, 292, 343–4, 415 scale 51–69, 123 see also exponence rankshift(ed) 103–4, 121–4 upward rankshift 122–4 realization 44, 46, 52, 108, 112–16, 124, 175, 194, 196–212, 217, 220–1, 231, 236–46, 273, 285, 307, 310, 352, 357–9, 362–5, 404–13 realizational chain 292, 352 realizational cycles 196–7, 204 Recipient 305 recursion 101, 213 recursive 45, 102–4, 121, 212–17, 242, 279, 343 Reddy, M. 293–4, 297 redundancy 65, 70–2, 295, 356–7, 387 reflection 197, 227, 236, 353, 355–6, 362, 382, 389 reflective 197, 237, 329, 356 relational 106–7, 227, 241, 277–8 clause(s) 181, 185–6 see also process representational, conative and 226–7, 236

440

index rhema 292, 294 rheme 188–91, 205, 226, 231, 236, 264, 266, 269, 300, 371, 376 rhetorical 191, 199, 221, 224–6, 231–3, 240–4, 283, 328, 371, 411 rhythm 114, 216, 262–3, 338, 348, 405 rhythmic 207, 269 Rumbaugh and Rumbaugh 3, 4 Russell, Bertrand 349 Russian 372–3, 392, 400, 416 Sanskrit 296 Sapir, E. 219, 246, 262, 311, 377 Saussure, F. 4, 220, 262, 325, 403, 412 science 5–6, 9, 296, 396 language(s) of 12, 365–6 linguistic 21–2, 65 technology and 291, 392, 395 semantic 11–12, 32, 114, 159, 170, 175, 185, 196–216, 220–46, 283–7, 294–312, 324–50, 354, 373–8, 389–416 component(s) 200, 234–41, 389 system 197–216, 237, 310–11, 345 semiotic 1–12, 109, 196–247, 296–313, 325, 353–66, 370–82, 387–414 higher-order semiotics 3, 389, 392 space 353–65 system(s) 8, 12, 109, 196–7, 298, 301, 311, 313, 325, 356–7, 370–3, 387–8, 392, 397, 402, 406, 414 transformation 390, 408 semogenic 8, 227, 303–6, 354–64, 392–7 semogenesis 357–62 Sinclair, J. McH. 168, 170, 242, 407 Slav 2 sociolinguistics 10 Soviet 2 speaker 31, 174–6, 188–92, 199–200, 205–7, 226–7, 229, 236, 240, 243, 270–4, 301–11, 325–6, 337–8, 344, 371, 376 speaker-now 273 speaker-oriented 207, 240 speaker-prominent 240–3

speaker/listener 395, 405 speech 6–8, 174, 176–7, 192, 195, 197–9, 206–10, 215, 219, 229–35, 239–43, 266, 268, 273, 282–3, 302, 305–8, 323–31, 335–41, 345, 348–51, 362 see also language, spoken speech-functional 233 speech-like 345 spontaneous 302, 310, 323–41, 369, 377, 382, 406 see also discourse strata/stratal/stratum 3, 115, 196–7, 222, 262, 294, 296–7, 301, 308, 352–65, 369–70, 378–9, 387–90, 397–8, 402–16 stratified 196, 356, 364, 389, 397–8, 409 stratification(al) 112, 115, 222 Subject 96–101, 113, 176, 183, 191, 194, 238, 262, 268, 271–4, 298–310, 372–5, 410, 416 psychological 189–94 Sugeno, M. 3, 12, 402, 416–17 Svartvik, J. 181, 371 Sweet, H. 179, 189, 299 syllable 33, 219–22, 231, 234, 292–3 syntagmatic 27, 46, 61, 106–16, 120–5, 160–4, 216, 262, 305, 378 syntactic 56, 71, 95–7, 106–7, 109, 115, 166–7, 182, 185, 280, 335 syntax 12, 51, 369, 386 systemic description 110–12 feature(s) 112–15, 362, 404–5, 411 grammar(s) 301, 332, 406 grammatics 402, 417 meaning potential 414 network 115 theory 215, 348, 415–16 typology 33 Tagalog 374 tagmemic 107, 110, 121, 222 taxonomic 26, 279, 404 taxonomies 291, 379, 396 taxonomy 42–3, 199, 264, 371

441

index tense 25, 112, 215, 273–4, 279–80, 300, 310, 326, 365, 375, 379, 393–4, 400 text 7–12, 22–30, 38, 45, 49, 109, 166, 169, 175, 184, 192, 199–201, 207–11, 217–47, 261–87, 294–5, 300–11, 324, 328–35, 338–47, 350, 360–3, 386–7, 394, 412 text-forming 206–7 text-generation 284 text-like 234 text-linguistic 222 texture 207–8, 211, 224–5, 233, 271, 281 Theme 111, 113, 190–4, 206–9, 216, 228, 230, 233–5, 238–46, 266–8, 270–3, 297–300, 303–9, 347, 357, 371–8, 416 Theme-Rheme 209, 228, 233, 262, 271 (un)predicated 111 Token 292–3, 296–8, 364, 366, 413 tone 67–8, 164, 192, 205, 270 concord 266 contour 269, 399 group(s) 55, 192, 207–8, 266, 269 key 203 tonic 113–14, 126, 192, 270 accent 206, 208, 270 prominence 206–7, 270 segment 270 topic 266 and comment 299 sentence 211, 228, 234, 243 Topic 300 transcription 262–3, 345 transitivity 176–93, 200, 215, 224, 229, 233–7, 243, 246, 263, 268, 274, 276, 302, 305, 312, 364, 382 typological 13, 120, 397, 402 typology 9, 33–4, 167, 388, 399

universal 21–31, 95, 120, 301 universals 12, 209, 414 unmarked 33, 164, 167, 207, 233, 240–1, 264–9, 279, 305 Urdu 374 Ure, J. 227, 229, 327 verb 25, 28–30, 51, 95, 167, 178, 180–3, 186–7, 190, 236, 297–302, 312, 326, 346–8, 376, 394, 410–11 postpositive 28 prepositive 28 pro-verb 28 verbal group 47, 51, 70, 102, 243, 263, 280, 325, 393–5, 400 Vietnamese 34 voice 28–9, 182–4 active 29 ergative 28–9 WH- 189, 206, 268, 275 Whorf, B. 219, 262, 293, 302–3, 306, 311, 312, 373, 397 word 26–34, 44–5, 51, 55, 58–60, 69, 95–104, 106, 116, 121–6, 164, 180, 190, 205–6, 217, 221–2, 239, 243, 280, 296, 303, 371 class(es) 28, 31, 34, 95–6, 104 wording(s) 3, 222, 345 meaning 197, 231, 293–4, 310, 356–7, 402 patterns of 219–20, 241 see also lexicogrammar writing 227–9, 302, 405 expository 225 scientific 349 speech vs. 323–51 theory 227 see also language, written Zadeh, L. A. 399, 416

442