M. a. K. Halliday, Jonathan J. Webster] on Grammar

7/21/2019 M. a. K. Halliday, Jonathan J. Webster] on Grammar

http://slidepdf.com/reader/full/m-a-k-halliday-jonathan-j-webster-on-grammar 1/452

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

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

Reprinted 2003, 2005



CONTENTS

Preface vii

Acknowledgements ix

Introduction: a personal perspective 1

Professor M. A. K. Halliday

S ECTIONONE: EARLY PAPERS ONBASIC CONCEPTS

Editor’s Introduction 17

1 Some aspects of systematic description andcomparison in grammatical analysis 21

2 Categories of the theory of grammar 37

3 Class in relation to the axes of chain andchoice in language 95

4 Some notes on ‘deep’ grammar 106

5 The concept of rank: a reply 118

Appendix to Section One 127

S ECTION TWO: WORD– CLAUSE– T EXT


6 Lexis as a linguistic level 158

7 Language structure and language function 173



contents

vi

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 atext like a clause? 219

10 Dimensions of discourse analysis: grammar 261

S ECTION THREE: CONSTRUING AND ENACTING


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 andcomplementarity 369

15 On grammar and grammatics 384

Bibliography 419

Index 433



vii

PREFACE

For nearly half a century, Professor M. A. K. Halliday has beenenriching the discipline of linguistics with his keen insight into thissocial semiotic phenomenon we call language. His scholarship hasadvanced 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, Hallidayapproaches language from the vantage point of meaning and purpose,and provides a sound theoretical framework for dealing with questionsabout 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 manyfind appealing in the man and his scholarship is his rejection, on theone hand, of the elitism typical of certain other schools of linguistics,while on the other hand embracing the study of that which powerslanguage 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 tolanguage and linguistics, these are:

1 On Grammar

2 Linguistic Studies of Text and Discourse

3 On Language and Linguistics

4 The Language of Early Childhood 5 The Language of Science 6 Computational and Quantitative Studies7 Studies in English Language



preface

viii

8 Studies in Chinese Language 9 Language and Education

10 Language and Society

Halliday approaches language from above, from below and fromroundabout (see Chapter 15, Section 15), but not from a distance. Hiscollected works, as presented in these ten volumes, reflect his charac-teristic balance between formulating and applying linguistic theory.The depth of his insight into language as system is highlighted in suchvolumes as the present one On Grammar and the third volume OnLanguage and Linguistics. The strength of his commitment to the study

of language as it is actually used is demonstrated in subsequent volumesdealing with Linguistic Studies of Text and Discourse , The Language of

Early Childhood and The Language of Science . The breadth of Halliday’sinterest in all things Language is glimpsed in the volumes Studies inEnglish Language and Studies in Chinese Language . The application of hisknowledge and experience as linguist and social scientist is visited involumes Computational and Quantitative Studies, Language and Educationand Language and Society.

The first volume contains fifteen papers, with the addition of a newpiece entitled ‘A personal perspective’, in which Professor Hallidayoffers his own perspective on language and linguistic theory as coveredin his collected works. The papers are divided into three sectionsaccording to topic, and within each section the papers are orderedaccording 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 andrank. Interestingly, the second section highlights how over the span of two decades (mid-1960s to mid-1980s) Halliday developed systemictheory to account for linguistic phenomena extending upward throughthe ranks from word to clause to text. The third section includes morerecent work in which Halliday discusses the issues confronting thosewho would study linguistics, or as Firth described it ‘language turnedback on itself’.



ix

ACKNOWLEDGEMENTS

We are grateful to the original publishers for permission to reprint thearticles 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 grammaticalanalysis’ from Studies in Linguistic Analysis, published by BlackwellPublishers, 1957, pages 54–67. Reprinted by permission of BlackwellPublishers.

‘Categories of the theory of grammar’ from Word , 17(3), 1961, pub-lished by the Linguistic Circle of New York (now the InternationalLinguistic Association), pages 241–92. Reprinted by permission of theauthor and the International Linguistic Association.

‘Class in relation to the axes of chain and choice in language’ fromLinguistics, vol. 2, 1963, published by Mouton (now Mouton de Gruy-ter), 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 bypermission 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 bypermission of Cambridge University Press.

‘Lexis as a linguistic level’ from In Memory of J. R. Firth, published byLongman, 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 inLinguistics, published by Penguin Ltd, 1970, pages 140–65. Edited by

John Lyons. M. A. K. Halliday, 1971, collection John Lyons,1970. Reprinted by permission of the Penguin Group (UK).



acknowledgements

x

‘Modes of meaning and modes of expression: types of grammaticalstructure and their determination by different semantic functions’ fromFunction 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’ fromThe Seventh LACUS Forum, edited by James E. Copeland and PhilipW. Davies, published by LACUS, 1981, pages 31–59. Reprinted bypermission of LACUS.

‘How is a text like a clause?’ from Text Processing: Text Analysis and Generation, Text Typology and Attrition (Proceedings of Nobel Symposium51), edited by Sture Allen, published by Almqvist and Wiksell Inter-national, 1982, pages 209– 47. Reprinted by permission of Almqvistand Wiksell International.

‘Dimensions of discourse analysis: grammar’ from The Handbook of Discourse Analysis, Vol. 2: Dimensions of Discourse , published by Aca-demic 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 Theoryand Practice , edited by Martin Davies and Louise Ravelli, published byPinter, 1992, pages 20–35.

‘Grammar and daily life: concurrence and complementarity’ fromFunctional Approaches to Language, Culture and Cognition, edited by Teun

A. van Dijk, published by John Benjamins Publishing Co., 2000, pages221–37. Reprinted by permission of John Benjamins Publishing Co.

‘On grammar and grammatics’ from Functional Descriptions: Theory inPractice , 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.



1

INTRODUCTION:

A P ERSONAL P ERSPECTIVE

The volumes in this series will contain a selection of my writings onlanguage, extending over the half century beginning in 1951. A few,including one or two items written specially for the series, have notbeen 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 thetopical arrangement will tend to be fairly loose, partly because mywriting has always been inclined to drift, and partly because both the

editor and I prefer it to be that way – weak boundaries have alwaysbeen characteristic of my approach.

I have never really thriven in a discipline-based structure of know-ledge. It was a feature of my century – the late and rather unlamentedtwentieth, perhaps mercifully short in Eric Hobsbawm’s conception of it – that it began by erecting walls between the disciplines, and it isproving difficult to demolish these walls now that they have come to beconstraining 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 inde-pendence and autonomy of the discipline of linguistics, and one couldsympathize 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 oncemade some reference to this; if so, it will turn up in Volume 3. But



o n g r am m ar

2

what I did stress was how much linguistics had in common with other scholarly pursuits and, when it came to asking questions aboutlanguage, 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’swhat we were there for. But it did not seem fine, to me at least, for usto determine what questions should be asked about that domain. I wasinterested 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-rewardedgroup who have to be both scholars and practical problem-solvers,namely teachers.

There was nothing surprising about this last perspective: not onlyhad my parents both been teachers, but I myself had taught languagesfor thirteen years before transferring myself into a teacher of linguistics.

But even before that, while still at school, I had been trying to find outabout language – because I was keen on literature and wanted to

understand why its language was so effective, what was special aboutit. There is no separating one’s personal history from the academicpaths one pursues, nor any way of detaching cause from effect inexplaining one’s chosen approach to a field of study. One way or another, I have always found myself asking the kinds of questions aboutlanguage 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. Butat any given “moment” in space-time, there are likely to be only a fewpredominant motifs by which the context of linguistic scholarship isdefined. This may even be legislated from on high, as when Stalinwriting 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 justtaken them all over. Usually it is determined by less overtly political

factors: by particular social movements and demands, or notableadvances 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 neuro-science; the combination of new technology – positron-emissiontomography, magnetic resonance imaging and its derivatives – withnew insights in evolutionary theory and its contributing disciplines hastransformed the way we understand the human brain, how it has



a personal perspective

3

evolved in the species and how it develops in the individual from birth(and before) to maturity. And this new understanding has radicallyredefined the place of language. It is now clear that language and the

brain evolve together, and that these develop together in infancy andchildhood. The development of the brain is the development of theability to mean; as in every aspect of human history, so in the ontogeny

of the individual human being the material and the semiotic inter-penetrate, 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, asimilar type of higher-order semiotic. On the contrary. The work of Duane Rumbaugh and Susan Savage-Rumbaugh has brought out thepoint that the bonobo chimpanzees can operate with sets of arbitrarysymbols in a way that is analogous to our own system of wording(lexicogrammar). They lack an analogous vocal apparatus, but that isbeside the point. It is tempting to assume that they have been followingthe same evolutionary path and are simply less far advanced along the

way – this is the assumption that prompts questions like ‘what age havethey reached, in terms of a human child?’ But this assumption is

probably wrong, or at least misleading, if it is used to describe an adultchimpanzee in terms of an immature human; the adult bonobo’s brainis fully wired up in terms of the construing of experience, and enactingof social relations, that constitute bonobo culture. The question canfairly be asked about bonobos brought up from birth in a human-likesemiotic environment, like Kanzi, and it is too early to say yet whether Kanzi and the other youngsters’ development of the power of meaningtracked 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 brainfunctions, and how it evolves and develops, has no significance for theway linguists describe and explain language, especially at the ‘inner’strata of lexicogrammar and phonology (wording and sounding). Pos-

sibly; although even here it seems to me to set certain constraints andmore importantly perhaps to favour certain explanations over others. Itsuggests “systems thinking” rather than compositional thinking (Mat-thiessen), grammatical logic rather than formal logic (Sugeno 1995),fuzzy and probabilistic categories rather than clearly bounded anddeterministic ones. Since the brain is more like a jungle than like acomputer (Edelman 1992), it disfavours representations of grammar andphonology that are influenced, however indirectly or subconsciously,



o n g r am m ar

4

by the way that computers happen to be being designed and operatedat 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 thatof those branches or special fields within linguistics that relate closelyto 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 contributeto human well-being. I have not worked in the latter area. But I have

written a certain amount on child language development, based onintensive research of the ‘diary’ type (Bates, in Rumbaugh and Rum-baugh) and it was clear already when I started that this must havemutual implications with the study of the brain. This was in the early1970s, before the explosion of knowledge in neuroscience; I was notable to make any sensible use of what little I had read then about brainfunctioning. I simply drew attention to two factors that had emergedvery clearly from my own researches: first, that the language surround-

ing a child was rich and highly structured, very different from theformless and impoverished quality that was being asserted about spoken

language at the time; secondly, that before mother tongue came childtongue (I called it “protolanguage”), which had a different structurefrom “adult” (post-infancy) language – so if children were bornendowed with a grammar-shapen brain, why did they first construct alanguage of quite a different type, which had no grammar in it at all?2

Today matters have changed, and students of child language develop-ment can hardly avoid taking note of what has been found out aboutthe development of the infant brain.

The other aspect of the work of linguists that is impacted byneuroscience is a more macro one: the modelling of the system of language as a whole. The overall construction of language as systemwas very much part of the enterprise of twentieth-century linguistsfrom Saussure onwards and reached a high point with Hjelmslev’s

Prolegomena (1961), first written in 1942. Since then it has becomebackgrounded, for various reasons: the subject expanded into a colonyof subdisciplines, or branches, which seemed not to need any generalperspective; in the west, at least, including the USA where most of theresearch was being done, Chomsky’s post-Bloomfieldian model becamedominant and was not open to challenge; and the general post-modernethos was in any case hostile to comprehensive accounts (they wereseen as “totalizing”), often in fact to theorizing of any kind.





o n g r am m ar

6

but you have to know language in order to understand them. Com-puters are built to a logic derived by design from grammar; they willhave 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 aspecially 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 andexplained. It is one that happens to be particularly favourable just at

present, because of the success in brain science. But any other perspec-tive – literary, social, physical, logico-philosophical or whatever – isequally valid and language will look somewhat different from each of these different vantage points. Some will obviously be more relevantthan 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 againthere is a special branch of linguistics, speech science, where knowledge

about language as a physical system is one of the central concerns. Thefragmentation 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 withdifferent aspects of one of the same system-&-process, then it is inlinguistics 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 anddescribed differently in all these different contexts, at least for purposes

of different applications or different research goals. This was the viewexpressed in ‘Syntax and the consumer’ (Halliday 1964). This approachwas partly taken as a defence against the dominant elite, for whomlinguistics was “a branch of theoretical psychology” (Chomsky) – inthe 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 morethan 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 stuckto it, I was constantly jumping around to see what language lookedlike when viewed from the other side. To the extent that I favouredany one angle, it was the social: language as the creature and creator of




7

human society, as expounded by my teacher J. R. Firth and by myfriend and colleague Basil Bernstein. But by nature, and also byexperience, I was (and have always remained) a generalist. So while

consciously I was trying to model language as a social phenomenon, infact I was acting against my own advice and trying to look at languagefrom every possible vantage point in turn.4

Most linguists, it seemed to me, looked at language only from theinside, 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 greatteacher Wang Li (then Professor of Linguistics at Lingnan University,Canton), there were still no tape recorders. We had to transcriberesponses directly into IPA script, which was excellent training for my

later investigation of child language. Professor Wang was able to acquirea primitive version of the same thing – a wire recorder, but it was notmuch 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 investigat-ing the articulatory mechanism of speech. Gramophone records werewidely used in language teaching: when I was taught Chinese for thearmed services at the University of London in 1942–43, the Depart-ment had its own recording equipment on which students couldregister their own performance and compare it with the recordedmodel. There were archives of spoken language on disk and even oncylinder, 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 andto conduct experiments. Linguists either relied on the kind of mani-cured discourse that is produced in writing and in prepared and self-monitored speech, or else invented data for themselves from their ownintuition of the language, and they had no way of processing largequantities even of that.

For linguistics, the two most important advances in the latter half of



o n g r am m ar

8

the century were technological ones: the invention of the tape recorder and the evolution of the computer. The tape recorder made it possibleto record natural speech. The computer made it possible to process

large quantities of data. The two together have given us the moderncomputerized 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 (“pars-ing”), 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 arebeing revealed that we have known must be there, because there was agap where the approaches from the lexical and the grammatical polesof the lexicogrammar converged, but which we could not see: thenature of grammatical logic is beginning to be understood; the semo-genic (meaning-creating) power of discourse is coming into view, bothin monologic and in dialogic mode; quantitative mechanisms of lin-

guistic change are beginning to appear on the agenda. From all this itshould be possible in the next decade or two to crack the semiotic

code, in the sense of coming fully to understand the relationshipbetween observed instances of language behaviour and the underlyingsystem of language – something that has eluded us up till now, so thatwe have even turned the two into different disciplines, calling only oneof them “linguistics” and labelling the other “pragmatics”.

Some people will feel threatened by this new understanding. Weknow this because there are those who already do. To bring to lightthe systems and processes of society is already threatening enough, aswitness the panic reactions to Bernstein thirty years ago when hedemonstrated how social class structures are transmitted, but semioticsystems and processes are even nearer the bone. As long as linguistsconfined their attention to dead languages, codified texts or sanitizedexamples 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 meaning-creating power of everyday real life talk, it starts to become dangerous.Some people feel worried that the grammarian is someone who knowswhat they are going to say next and even if they can be reassured thatthat is not what theory is about, it is scarcely less threatening (appar-ently) to be told what proportion of positive to negative clauses theyare going to use in their speech. And for others, just to be faced with arecord of real life conversation can be unnerving; they feel embarrassed




9

and ill-at-ease at what seems to invade the interactants’ privacy andstrip 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 extradimension into our understanding of language? Isn’t it enough to playthe traditional part of a grammarian or a phonologist and join in the

endeavour – itself an enterprise that has notched up considerablesuccesses – of broadening our knowledge of the history, typology andstructure of the world’s many languages? After all, there is more thanenough work here to occupy the community of linguists, even if it was

enlarged many times over, in meeting all the theoretical and practicaldemands in education, multilingualism and multi-culturalism, ecolin-guistics, language maintenance, translating and interpreting, forensiclinguistic work and so on. Why do we need a huge computerizedcorpus of authentic data, which in any case will be available only for asmall number of the world’s major languages?

There are it seems to me two answers to such questions – or twoparts 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 makeit more comprehensive and more accurate. This is what Quirk had in

mind when, in launching the first systematic modern corpus, theSurvey of English Usage at University College London, he described itas 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 thelexicogrammatical 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 describethemselves as “mere data-gatherers”: to upgrade our theory – toimprove 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 tothe semiotic domain, leaving machines and robots to do the material

business, then the demands on language and its satellite systems aregoing to go on increasing and hence, inevitably, the demands on theoryof language. Our world consists of these two grand phenomenaldomains, matter and meaning. The science of matter is physics; thescience of meaning is linguistics.5

There are fads and fashions in every field of study and linguistics isno exception. In the 1960s, it was almost impossible to get publishedany analysis of a text. The worst insult that could be paid to a linguist



o n g r am m ar

10

was to say that he or she was “data-oriented”. Data were said to beirrelevant to the serious study of language; the actual language used byreal people, especially spoken language, was dismissed as impoverished

and unstructured, a mere matter of performance that could tell usnothing about the true object of description, which was linguisticstructure – the rules generating the set of idealized sentences that

constituted the ideal speaker’s competence or knowledge of thelanguage.

This monolithic Cartesian culture maintained its stability by con-stantly re-examining its own foundations, finding newer and more

elegant ways of going over the same ground. The idealogy that pervadedit and the conditions it brought about have been well described byde Beaugrande. Since it excluded any reference to the social context of language, it was necessary to invent a new field called sociolinguisticsand a new kind of competence called “communicative competence” togo with it (Hymes 1971). And when a change of fashion broughtdiscourse 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 independentdisciplinary base (and channel for getting things published) and suddenly

everybody was “into discourse”. A number of factors came together toensure the success of the pragmatics enterprise, which has released anenormous amount of energy and raised to theoretical status discursiveissues such as implication, relevance and politeness.

Having grown up in opposition to linguistics, pragmatics has largelydispensed with grammar; what theoretical input it has had has beendrawn 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 betweentwo aspects of what to me is a single enterprise: that of trying to explainlanguage. It seems to me, however, that both parts of the project areweakened 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 systemsare instantiated. From this point of view, pragmatics is the instantialaspect of semantics: the semantics of the instance, in other words. Toput this in the opposite perspective, the system is the meaning potentialthat lies behind every instance of discourse. Children construct thesystem, in fact, from very large numbers of discourse instances, whichin the typical case are fluent, well organized and related (and hencerelatable) to their instantial situational contexts.




11

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 influentialexponent of this position has been Bourdieu (Hasan 1999). Bourdieu isan expert in exploiting the power of language to proclaim that languagehas no power. This carries the comforting message that therefore youneedn’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 begratefully received. The argument being offered is that if you takeaccount of the centrality of language you are being “logocentric”:anything “centric” is condemned without being tried. In practice theproblem is exactly the opposite – we might call it ‘centrifugal’:

whenever in dealing with any issue people are brought face to facewith 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 incorporatethese 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 issensible 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 tothe 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 thepoint where it can account for inferential relations like these: explainingwhy Johnny has not yet tried the dynamite, and how the properties of silk purses and sows’ ears are analogous to those of the imagined Harryafter Maggie’s makeover and Harry as he now is. But the relations tobe accounted for are relations of meaning; they are rather more

complex than the singular semantic relations that have always been



o n g r am m ar

12

familiar in the lexicogrammar, like hyponymy and polarity and voice,but still ultimately of the same kind. It should be possible to extend thepower 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 todo this, I think, in order to achieve the kind of “intelligent computing”

that is envisaged by Michio Sugeno in his work at the Brain SciencesInstitute in Tokyo.

What the following chapters do is to illustrate some of the steps inmy own thinking that have led me in this general direction. The steps

that have seemed to me perhaps most critical in this endeavour mightbe 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 imposedby structure; the metafunctional foundation, disentangling the strandsof meaning that are woven together in the syntax; the construction of language by children, from protolanguage to mother tongue; thedecoupling and recoupling of lexicogrammar and semantics – the

phenomenon of grammatical metaphor; the conceptualizing of therelation between system and text (instantiation) and the probabilistic

nature of linguistic systems. Some of these will be treated separately inlater volumes: in particular Volume 4 on child language and Volume 5on grammatical metaphor and the language of science. Others willappear in various contexts and under a variety of headings.

I am often asked about my views on “linguistic universals”. Theanswer is that I follow Hjelmslev and Firth in distinguishing theoreticalfrom descriptive categories. The theoretical categories, and their inter-relations, construe an abstract model of language (and other semioticsystems); they are interlocking and mutually defining.8 The theory thatis 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 linguis-tics” – except institutionally where, for example, an education authority

will give teachers release time and professional credit for a degreecourse 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 usuallymean descriptive categories that are assumed to be found in alllanguages. The problem is that there is no mechanism for decidinghow much alike descriptive categories from different languages have to

be before they are said to be “the same thing”. There is a method,




13

based on the (theoretical) category of system, for matching up descrip-tive 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, isproclaimed 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 (Caf-farel, Martin and Matthiessen 2002). My own interpretation of thegrammar of modern English will be found in An Introduction toFunctional Grammar (Halliday 1985 and later editions). Other descriptive

papers on English and on Chinese will be presented in Volumes 7 and8. A theory-based account of the ideational semantics of English is inConstruing Experience through Meaning (Halliday and Matthiessen 1999).

I doubt whether any of the present volumes would have appearedwithout the enthusiasm, energy and efficiency of “my” editor, Dr

Jonathan Webster, of the City University of Hong Kong. He broughtthe whole project to life, convincing me that it was worthwhile andconvincing 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 patiencewith me or faith in the enterprise, and to Robin Fawcett, who set thewhole thing going and provided many rounds of valuable suggestionsand 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 semioticsystems are of the fourth order of complexity, being also physical and

biological and social. This means that one and the same linguistic phenom-

enon (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



o n g r am m ar

14

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 exploreit?’

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 (thoughsystemic 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 your-

self 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 appli-

cable to particular linguistic descriptions, not a theory of universals for

general linguistic description” (Firth 1957: 21).



S ECTIONONE

EARLY PAPERS ON BASIC CONCEPTS



This page intentionally left blank



17

EDITOR’S INTRODUCTION

In this first section we look at five papers written and published byProfessor Halliday over a ten-year period from 1957 to 1966. The basicconcepts, which form the foundation of Halliday’s systemic theory, areelaborated in these papers. These include such fundamental categories

for a theory of grammar as unit , structure , class and system. He alsoaddresses the relations of these categories to each other and to the datain 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 invarieties of a language, Hjelmslev’s focus on language as a whole and

Jakobson’s search for universals across all languages. Already in his earlywritings, Halliday draws on the insights of these and others to constructa theory of grammar grounded in the linguistic analysis and descriptionof particular languages, which acknowledges the primacy of meaningand the need for systematicity.

Published in 1957, the first paper in this section, ‘Some aspects of systematic description and comparison in grammatical analysis’, discussestheoretical considerations which developed out of the body of ideas that

went into his doctoral dissertation. Building on and extending the generallinguistic principles established by Firth and other scholars, Hallidaydemonstrates the application of formal methods of linguistic descriptionto New Chinese (Modern Pekingese). Identifying formal linguistic

methods as being derived from structural linguistic theory, Hallidaymaintains that ‘a complete analysis at the grammatical level, in a particular description in which all forms of the language are related to systems setup within the language itself, requires establishment of grammatical



e a rl y p ap e rs on ba sic c onc e pts

18

categories, ordered as terms in interrelated systems and having asexponents 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 the-oretical approach, as outlined in ‘Categories of the theory of grammar’(Chapter 2), diverged from that of his mentor, Halliday sought to havethe opportunity to discuss the paper with Firth. Firth’s sudden passing,

however, prevented this from happening. In this paper, Halliday setsout the following fundamental categories for the theory of grammar:unit , structure , class and system, which relate to one another and to thedata along three distinct scales of abstraction, including rank, exponence and delicacy. Halliday prefaces his discussion by stating what he regardsas ‘given’, among which he includes the following:

a. Texts, or observed language events, are the data to be accountedfor, whether spoken or codified in writing.

b. Description consists in relating the text to the categories of thetheory. Description is not theory; rather it is a body of methodderived from and answerable to the theory.

c. Linguistic events should be accounted for at a number of differentlevels. 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 relatedbody of theory, namely General Phonetics. Phonology, on theother hand, relates form and phonic substance, i.e. where linguis-tics and phonetics interpenetrate.

e. Language has both formal meaning and contextual meaning.Formal meaning is the information of information theory; con-

textual meaning relates to extralinguistic features.f. We must distinguish not only between theory and description,

but also between description and presentation, being the way thelinguist expounds the description.

Elaborating on each of the fundamental categories for the theory of grammar, Halliday describes units as pattern carriers. The scale onwhich units are ranged in the theory is called rank. Structures are ‘the



editor’s introduction

19

ordered repetition of like events that make up the patterns’. There areboth 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 onerather 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 veryfamiliar kind of patterned activity, namely, eating a meal. Looking backon this chapter after forty years, Professor Halliday provides somebackground 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 theaxes of chain and choice in language’ (Chapter 3), Halliday discussesthe relation of class to structure , the chain axis and class in relation tosystem, the choice axis. Class is related to two kinds of structure found inlanguage: the place-ordered, in which a limited number of differentelements occur non-recursively, and the depth-ordered or recursivestructure. Rankshift , for example, refers to a type of recursive structurewhich cuts across the scale of rank. It is in the fourth paper in thissection, ‘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 andsystemic description may be understood in terms of syntagmatic andparadigmatic relations.

Appearing in the same issue of the Journal of Linguistics is the lastpaper in this section, ‘The concept of rank: a reply’, in which Hallidayreplies to arguments against rank grammar put forward by P. H.Matthews. As Halliday explains, a rank grammar ‘specifies and labels afixed number of layers in the hierarchy of constituents such that any




20

constituent can be assigned to one or other of the specified layers, or ranks’. On one point both Halliday and Matthews agree, namely, thatrank grammar is a hypothesis about the nature of language. As Halliday

argues, it is a hypothesis worth making both for its descriptive advan-tages 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 appearedin Kress (1976).



21

Chapter One

SOME ASPECTS OF SYSTEMATIC

D ESCRIPTION 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 allother levels, systems of related categories. Such categories as are estab-lished 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 pointsof reference is clearly of a different order from the other two: there canbe no universal formal-linguistic categories (there might theoretically becategories formally identified as common to all languages studied here-tofore, but such identification is not yet a practical possibility), whilenon-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 thestudy of the evolution of linguistic forms, the improvement of themethods 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 con-tribution between descriptive and historical studies in linguistics; and it isnot surprising that many recent developments in the former have beenfounded on work done in languages where it has not yet been possible toestablish 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.




22

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 theneed for a general theory of description, as opposed to a universal schemeof descriptive categories, has long been apparent, if often unformulated,

in the description of all languages. A distinctive contribution of thetwentieth century has been the progress towards its achievement.

The sort of descriptive statement which has been the fruit of theseachievements is too often characterized negatively by opposition to a

historical (the synchronic–diachronic dichotomy) or to a comparativestatement. This is probably due in part to the very fact that the techniqueshave been applied to languages which have no ‘history’ (that is, nowritten document of the past) and even no script, and which have notbeen satisfactorily organized into families by the comparative historicalmethod. It is not unnatural that what is new in descriptive techniquesshould have been emphasized by its being contrasted with the historicalmethods which, in the modern period at least, developed earlier; but,

while the comparative historical, like all other scientific methods, willbenefit 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 whichguides and controls the procedures of the various branches of linguisticscience, then any linguistic study, historical or descriptive, particular or comparative, draws on and contributes to the principles of generallinguistics. A simple scheme for ordering the branches of linguisticscience controlled by general linguistics might recognize two dimen-sions: 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 textsgreater than one). Thus:

Descriptive Historical

Particular —

—

—

—

—

— —

Comparative — — — —

— —

— —



systematic description and comparison

23

Any of these types of study may be undertaken with the use of formal linguistic methods: that is, by the methods of what is sometimescalled ‘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 schoolof linguistics but to that body of general linguistic theory whichcontrols the application of formal linguistic techniques.) This inclusion

of historical studies in the field of application of formal linguisticmethods rests on the acceptance of the possibility of arranging languagetexts 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 thetype of comparatism envisaged by Allen, in which ‘time has nodirection and there is no becoming’ (Allen 1953: 106), diachronic hasof course no meaning. In this scheme it is envisaged that there mightbe a difference in the treatment of material consisting of more than onelanguage text according to whether or not the texts are arranged on atime-scale and treated as exponents of the same language at differentperiods (for example, by the modification of the description in such a

way as to present a continuum in which the systematic ordering of thetexts corresponds to their ordering in time).

A historical study is then formed out of material provided by a seriesof descriptive studies. The distinction between this method, in whichany linguistic form is placed in its descriptive context (systematized)and the systems in which it operates are treated historically, and thetype of historical study which ‘structural’ linguistics has excluded – inwhich is traced the evolution of particular forms without descriptivesystematization – might be reflected in the addition to the diagram of athird 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 theacceptance or non-acceptance of history but on the type of historicalapproach. So may the social anthropologist study either the evolutionof 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 differentperiods 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 lessinaccessible 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 mightbe the “universal”: the question is whether this is at present, to use






25

have earned the name of “structural” linguistics. The term is perhapsmost favoured, in Britain at least, by those who do not apply thesemethods. One sometimes comes across an expression of “disagreement

with” “structural” linguistics. But one might as well disagree withnuclear 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 ita 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 differenttype, 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 askwhether an analysis of Chinese made on the same principles as thoseapplied to Latin, with reference to the forms of Chinese alone, mightnot be equally valid. Likewise the belief in the impossibility of formalgrammatical analysis of some languages has little currency today, though

a question concerning the methods to be applied, which was a subjectfor 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 relatedto systems set up within the language itself, requires the establishment

of grammatical categories, ordered as terms in interrelated systems andhaving as exponents the substantial (phonic or graphic) segments of thetext. Such categories are of two types, which we may call units andclasses. The units are defined by interrelation in terms of extent: unlikea system of classes, whose terms are both collectively exhaustive and

mutually exclusive, the single system of units forms a hierarchy indescending progression, such that each term is defined as n times thesucceeding term, that is, as consisting of one or more members of thesucceeding term (exponentially, every exponent of a given unit isstatable 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 grammaticalcriteria, though in the delimitation of exponents of each unit within




26

the text other criteria, phonetic or graphic, may contribute and may betaken, where to do so is compatible with the general aims of simplicityand comprehensiveness, as the primary or even sole criteria (for

example, punctuation or spacing in a written text, features of intonationof pause in a spoken text, all of which then enter into the grammaticaldescription). It is probable that in the description of any language at

least two units will be required: these would be such as could be namedthe sentence and the word .3

A descending order of procedure seems preferable not only for thepresentation (where indeed it may be varied for a particular purpose),

but also for the analysis of the grammar, where in such a hierarchicprogression 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 (whichwe may call the “sentence”) which, while within the scope of gram-matical statement – not so extensive as to be incapable of systematicanalysis – is yet enabled to operate as the linguistic action of participantsin a situation: which is, in fact, “living language” and constitutes the

unit of analysis at the contextual level. In the subsequent establishmentand classification of the lower units, the statements made about eachunit 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, formallyestablished in the grammar and exhaustive, such that statements may bemade which are valid for all exponents of a given unit. These classesare set up independently of structure: that is to say, a unit (for exampleclause) having been established, it is then classified by reference tovarious sets of formal criteria (for example presence of, or ordering of,certain formally defined elements); each set of criteria permits theestablishment of one system of classes (“clause–classes”). Such mutuallyindependent systems of classes of any unit are referred to as ‘dimen-sions’: 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 placedwithin 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 thereappears the dimension compound / simple, while for each unit except




27

the highest there may be set up the complementary system of free /bound; the free member is exponentially identified with the simplemember of the unit next above. How far these dimensions of classes

enter into the description will depend on the extent of their deter-mination of other features.

Each unit is characterized by certain structures. The structure is a

syntagmatic framework of interrelated elements, which are paradigmat-ically established in the systems of classes and stated as values in thestructure.5 The terms in the system of any one dimension may operateas 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 arestated 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 thesentence might also be stated in terms of the values F and B theexponents of which are segments of the text as systematized in a free /bound dimension of clause–classes. Similarly if a unit “word” isestablished there will be dimensions of word–classes the terms in which

operate as values in clause structures: given a verb / noun / adverbsystem of word–classes, it might be that the structure ANV and NAVwere 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 isidentified as “bound” through the presence within it of a member of acertain word–class) or exterior to it (as when a clause is identified asnon-sentential in virtue of its being internal to another clause (in thesentence 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 necessarilyincomplete: it is, however, abstracted from a descriptive statement of Modern Pekingese grammar, one among many possible such state-

ments, 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:




28

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 ba , rarely jiang .

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 athe 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 pre-

positive / 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 v N). 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 alanguage in terms of categories set up within the language itself, I might




29

mention one further instance of the application here of formal methodsof analysis. This concerns the determination of the occurrence of amember of a particular grammatical class by formal linguistic but non-

grammatical features. The presence of a certain form in a given unit inthe syntagm may render probable the occurrence in a subsequent unitof 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 onlythe linguistic (verbal) action in a context of situation – the source of the determination may be found to be in what might be called ‘contextof mention’. This requires the two-term system “given” / “new”, the

given being that which has been mentioned in the preceding linguisticcontext.Modern Pekingese shows some correlation between this dimension

and word order in the clause, the position of the given being regularlyprecedent to that of the new. This may not only determine the relativeposition of words where the basic clause structure is unaffected (for example the relative position of preverbal free noun or pronoun andpreverbal adverb) but also permit the prediction of the occurrence of a

particular class of clause. A correlation is observable between theoccurrence 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 anergative clause, both nouns are given and the verb is new; while in apassive clause, the verb is given and the (directly) preverbal noun isnew.

The correlation is more clearly observable if context of mention istaken to include not only repetition of the term mentioned but alsoreference (for example, pronominal or synonymic). With such “contextof reference” we are well on the way towards context of situation;7 ina spoken text, categories of given and new established in the contextof situation, though further removed from the level of grammaticaldescription as usually envisaged, may be expected to display suchcorrelation with grammatical categories to at least as great an extent as

in a written text, where contextual reference is restricted to thelinguistic context (since there is no independent context of situationfor any unit less than the whole text). Indeed the contextual categoriesof given and new may aid in the identification of grammatical categor-ies: a certain category might be identified as the form taken by thegiven or that taken by the new.




30

3 Language under description, language of description

While perhaps modern general linguistics would recognize the estab-

lishment of categories within the language under description itself asthe basis of a particular description, reference to the forms of another language, including the language of description, may be made withoutinfringing the requirements of formal analysis. In any other than amonolingual description of a language text there arises in any case thespecific 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 desiredto exclude from consideration as far as possible all forms that donot belong to the l.u.d., the nearest approach is that outlined above,where the aim is achieved through the creation of a metalanguagewhose terms, whatever the context of their previous usage, are to betaken as defined only with reference to the text under description. Allidentification of categories either comparative or universal is therebyexcluded.

It may, however, be desirable in a given instance or for a givenpurpose 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 atvarious points. At one extreme, it is possible to make a descriptivegrammar of the l.o.d. using the same procedures as are applied to thel.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; thelatter 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 maybe possible to make a complete translation of the text under description,a translation of the type that may be characterized as contextual – onesuch 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 withinhis terms of reference to make a description such as would permit asystematic comparison. It may then be found that certain categories of the l.u.d. show a regular translation equivalence to certain unsystem-atized 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 some-




31

times be found preferable to reference to undefined categories of thel.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 yetbe identified as comparable on the basis of non-grammatical criteria.The category of personal pronouns is perhaps the most readily suscep-tible of such treatment, since, even if the category is not grammaticallyidentified in its place in the system of word–classes in the twolanguages, the terms of the pronominal system may be contextually

identified with reference to persons participating in a situation asspeaker, addressee and other participant. In a descriptive grammar of Modern Pekingese written in English, one might compare the personalpronoun systems of the two languages, Chinese being taken as thepoint of reference, as follows:

Reference Chinese English

1 wo I

33 tamen they

2 nı } you22,2(2)3(3) nımen

3 ta { he

she

12(2), 13(3), 12(2)3(3) women } we12(2), 12(2)3(3) zamen

Contextual reference is as follows: 1, speaker; 2, addressee; 22, address-ees; 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 calcu-lation of degree of relationship is necessary. Identification has beenpermitted between terms operating in systems with different numbersof terms: in any calculation of degree of relationship this point wouldof course be crucial. Prima facie , such identification would seem justifiablewhere the criteria for the identification were contextual. The identifi-cation of terms by grammatical criteria in a comparative study posesseparate problems which are touched upon briefly in the next section.




32

As a conclusion to the foregoing it may be remarked that referenceto the forms of the l.o.d. is invaluable for pedagogical purposes, whereits further systematization might prove helpful, as well as in those

instances where a single l.o.d. is common to linguists working in widelydivergent fields.8

4 Descriptive comparison

In a descriptive comparison there is no implication of genetic relation-ship; but beyond the application to particular description permitted by

comparison of this type (since any l.u.d. may be compared with anyl.o.d.) systematic comparison itself connotes a wider purpose, seeninitially in the establishment of degrees of relationship. Here the gram-

matical identification of terms across languages is an essential basis of comparison; and while much can be done on contextual criteria (thoughnot as much as in the establishment of lexical systems for comparativepurposes) some formal procedure must be found in order that statementof degrees of relationship may assume a general significance.

Possibly, however, we should not impose a strict demarcationbetween identification of grammatical terms on grammatical and onnon-grammatical (e.g. contextual) criteria. Allen, rightly rejecting bothterminological identification and identification by translation, considersthat identification by formal grammatical criteria (as opposed to thesituational–contextual criteria which are available for the semanticidentification of lexical items) seems not to be possible (Allen 1953: 99,100). Some attempt at identification may be made with the Chinesedialects, 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 equatesnei with Pekingese nı and, as readily, keude

¯i with Pekingese tamen. This

identification in practice demonstrates the contextual basis of the iden-tification; its validity on grammatical criteria may be tested by referenceto the place of the class “pronoun” in the system of word–classes.

When we find that it is possible to describe both Modern Pekingeseand 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 athree-term system of classes, verbal, nominal and adverbial, and thatone term in the system of nominal word–classes, the auxiliary noun,enters in both into the noun group with identical position and value inthe structure, we can regard the class of auxiliary noun in the twodialects as comparable, and are then justified in seeking to identify




33

certain terms in the two systems of auxiliary nouns.9 In this comparison,both grammatical and contextual criteria must be invoked: the Pekingesesystem has an unmarked term ( ge ) which is absent from the Cantonese

system, which would exclude the grammatical identification of anyterms whatsoever, since each marked term in the Pekingese system mustbe grammatically identified as excluding not only the other marked

terms but also the unmarked term, to which latter the Cantonese hasno parallel. On contextual criteria, however (including linguistic– contextual, i.e. collocational), the marked terms of the Pekingese systemmay be regarded as forming a distinct sub-system comparable with the

Cantonese system, so that a term can be identified in the use of theauxiliary noun Pekingese zhı , Cantonese (graphically and historicallyidentical) zek, we find the following instances (with specimen Englishequivalents):

Pekingese Cantonese yı ge gou yat zek gau ‘a dog’ (in general context or ‘given’) yı zhı gou yat zek gau ‘one dog’ (in specific context or ‘new’)(na ge) gou zek gau ‘the dog’

The non-identification of the first and third instances will not preventthe contextual identification of the second if the total spread within thatsub-system can be shown to be the same.10

Some such procedure, for the identification of terms in grammaticalsystems, once established, the field of application of descriptive com-parison is seen to be very wide. A comparison of systems in twolanguages such as Chinese and English, which at first sight seems to beof little purpose beyond its application to the particular description of one of the two, provides material for a formal systemic typology whencompared with other such comparisons – for, if with Allen ‘relationshipis not of languages but of systems’, the significance of such relationshipis that typology is not of languages but of systems. But without going

so far from the scope of comparative studies as usually envisaged, onemay seek a field for the application of systematic comparison both in

the comparison of languages considered to be genetically related and inthe study of languages in geographical proximity where there is nomaterial for genetic groupings.

With regard to the former, any discussion of the nature and impli-cations of genetic relationship may be avoided by the choice of a simpleinstance of systematic comparison in the phonology of the dialect areasof Chinese. In the Modern Chinese dialects, in the system of nasal finalsin the phonological unit the syllable may have two or three terms, or




34

there may be only one nasal final. If we compare these systems in certaindialects of the Mandarin, Wu and Yueh dialect groups, we find that itis 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 prosodicallyidentical 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, sothat 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 comparisonimmediately suggests the historical interconnection between the disap-

pearance of the final plosive system and the elimination of one termfrom the final nasal system in the dialects where these have occurred.

With regard to the study of languages in geographical proximity inan area where no, or only partial, genetic groupings have been estab-lished, systematic comparison can determine whether the question of

grammatical affinity is to be posed at all, and if so in what form. Thesuperficially apparent affinity among the languages of certain areas haslong been the basis of traditional typology. An instance of how it maybe demonstrated or disproved, initially as a function of systems, mightbe found in the grammar of East Asian languages. It seems possible toset up in, for example, Pekingese, Cantonese, Vietnamese, Siamese andMalay systems of nominal word–classes which would permit the iden-tification as comparable of certain terms, including the category discussedwith reference to Pekingese and Cantonese above and which may be

called the auxiliary noun. If the place of the auxiliary noun in the systemof word–classes, and the terms in the various systems of the auxiliarynoun, can be compared as between these languages on the basis of adequate criteria, we may determine whether or not a formal descriptionof these languages would reveal for these and other systems anythingthat could be regarded as grammatical affinity – such as, that is, to

exclude from such affinity any systems set up for other languages inother areas. Only if such affinity could be established among a significantnumber of systems in the grammar of these languages would it bepossible 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 grammat-ical similarity among the East Asian languages as contrasted, for example,




35

with the Indo-European languages, while on the other hand there existsthe quite justifiable scepticism among linguists either as to how this isto 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 inadequatelydefined, 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, thechoice 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 anexample 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 linguisticand the graphic units by the single name tzu (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




36

‘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 mustdepend 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 / zek 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 contextualcriteria 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 labio-

velarization, symbolized by y and w”.



37

Chapter Two

CATEGORIES OF THE THEORY

OF GRAMMAR (1961)

There have been in the main two approaches to description in modernlinguistics: the “textual” and the non-textual or, for want of a better word, “exemplificatory”. More recently a third has been added, pri-marily in grammar but lately also in phonology, the “generative”

(strictly “transformative–generative”, since generation does not presup-pose transformation). Some linguists have gone so far as to suggest thattransformative generation should replace other types of description1 asa linguistic method of making statements about language.2 Others,myself included, feel that all three approaches have a fundamental placein linguistics; that they do different things, and that the third is avaluable supplement to the first two.3

Description is, however, not theory. All description, whether gen-erative or not, is related to General Linguistic theory; specifically, tothat part of General Linguistic theory which accounts for how language

works. The different types of description are bodies of method whichderive from, and are answerable to, that theory. Each has its place inlinguistics, and it is a pity to deny the value of textual description(which is appropriate, for example, in “stylistics”, the linguistic studyof 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 tobe the fundamental categories of that part of General Linguistic theorywhich is concerned with how language works at the level of grammar,with brief reference to the relations between grammar and lexis andbetween grammar and phonology. The theory sketched out here

First published in Word , 1961, 17(3), pp. 241–92.




38

derives most of all from the work of J. R. Firth.4 At the same time Ido not of course imply that I think Professor Firth would necessarilyhave found himself in accord with all the views expressed, which in

some places depart from his own; nor do I underestimate the debt tomy 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 Linguistictheory. While what has made linguistics fashionable has been, as withother subjects, the discovery that it has applications, these applicationsrest on many years of work by people who were simply seekers after

knowledge. It would not help the subject if the success of theseapplications led us into thinking that the theoretical problems weresolved 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 arederived.

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 asmaterial for linguistic description, is a “text”.6

1.3 Description consists in relating the text to the categories of thetheory. The methods by which this is done involve a number of processes of abstraction, varying in kind and variable in degree. It is thetheory that determines the relation of these processes of abstraction toeach other and to the theory.7

1.4 The theory requires that linguistic events should be accountedfor at a number of different levels: this is found to be necessary becauseof the difference in kind of the processes of abstraction involved.8



c a te g ories of the theory of g ram ma r

39

1.5 The primary levels are form, substance and context . The substanceis the material of language: phonic (audible noises) or graphic (visiblemarks). The form is the organization of the substance into meaningful

events: meaning is a concept, and a technical term, of the theory (seebelow, 1.8). The context is the relation of the form to non-linguisticfeatures of the situations in which language operates, and to linguistic

features other than those of the item under attention: these beingtogether “extratextual” features.

1.6 The complete framework of levels requires certain further subdi-visions 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 relatedbody of theory, that of General Phonetics. Since phonology relates

form and phonic substance, it is the place where linguistics andphonetics interpenetrate. Linguistics and phonetics together make up“the linguistic sciences”.10

Figure 1 Levels of language




40

1.8 Language has formal meaning and contextual meaning . Formal mean-ing is the “information” of information theory, though (i) it can bestated 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 quantifieduntil 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 itemis 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 formalmeaning and has nothing whatever to do with ‘information’.12 Thecontextual 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 inits place in linguistic form: contextual meaning is therefore logicallydependent on formal meaning.13

1.10 It follows from 1.8 and 1.9 that, in description, formal criteriaare crucial, taking precedence over contextual criteria; and that thestatement of formal meaning logically precedes the statement of contex-tual meaning.14

1.11 Finally, it is necessary to distinguish not only between theoryand description but also between description and presentation. Presen-tation, the way the linguist expounds the description, varies with

purpose, and relative merit is judged by reference to the specificpurpose intended. Description depends on the theory; theoreticalvalidity is demanded, and relative merit is judged by reference tocomprehensiveness and delicacy.15

2 Grammar

2.1 Grammar is that level of linguistic form at which operate closedsystems.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 beused as the name of one of the four fundamental grammatical categories(see below, 6) it is useful to retain “closed system” when referring tothe system as the crucial criterion for distinguishing grammar fromlexis.

A closed system is a set of terms with these characteristics:




41

(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 theoperation of closed systems belongs to the level of lexis. The distinctionbetween closed system patterns and open set patterns in language is infact a cline; but the theory has to treat them as two distinct types of

pattern requiring different categories. For this reason General Linguistictheory must here provide both a theory of grammar and a theory of lexis, and also a means of relating the two. A description depending onGeneral Linguistic theory will need to separate the descriptions of thetwo 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 highestorder 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 accountfor all grammatical patterns that emerge by generalization from thedata. As the primary categories of the theory, they make possible acoherent 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 derivablefrom, each of the others. There is no relation of precedence or logicalpriority among them. They are all mutually defining: as with theoretical

categories in general, “definition” in the lexicographical sense is impos-sible, since no one category is defined until all the others are, in thetotality of the theory.19 The order chosen here for exposition istherefore simply that which seemed the easiest: namely the order inwhich they are listed above.

The relation of these categories to each other and to the data involvethree distinct scales of abstraction, those of rank, exponence and delicacy;these are considered separately (see below, 7) but have also to be




42

referred to in connection with the categories. In discussing these I haveused 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 logicalprecedence (such as inclusion).20 A taxonomy is taken to mean a specialtype of hierarchy, one with two additional characteristics: (i) there is aconstant relation of each term to the term immediately following it,and a constant reciprocal relation of each to that immediately precedingit; and (ii) degree is significant, so that the place in order of each oneof the terms, statable as the distance in number of steps from either

end, is a defining characteristic of that term.21

A cline resembles ahierarchy in that it involves relation along a single dimension; but

instead of being made up of a number of discrete terms a cline is acontinuum carrying potentially infinite gradation.

2.3 In this view of linguistics description is, as already emphasized, abody of method derived from theory, and not a set of procedures. Thishas one important consequence. If description is procedural, the only

way of evaluating a given description is by reference to the proceduresthemselves: a good description is one that has carried out the rightprocedures in the right order, but for any more delicate evaluationexternal criteria have to be invoked. Moreover, every language has tobe treated as if it was unknown, otherwise procedural rules will beviolated; 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 areaccounted for. The linguist makes use of all he knows and there is nopriority of dependence among the various parts of the description. Thebest description is then that which, comprehensiveness presupposed, ismaximally grammatical: that is, makes maximum use of the theory toaccount for a maximum amount of the data. Simplicity has then to beinvoked only when it is necessary to decide between fewer systems

with more terms and more systems with fewer terms; and since bothinformation theory and linguistic intuition favour the latter even thispreference might be built in to the theory.22

3 Unit

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




43

exhibited over certain stretches of language activity. An essential featureof 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 areassociated with stretches that not only are of differing extent but alsoappear 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 begeneralized 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 measurementsof linear progression belong to language substance (where one maybe interested in the number of seconds, or possibly even the number of inches, occupied by an utterance). What is less obvious is that thewhole dimension of progression in fact belongs to substance, and thatthe stretches which carry grammatical pattern – or rather the membersof 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 amanifestation in substance: a dimension we may call “order”.23 Byimplication, 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 isshown that the theory does not need to make provision for itshappening. In fact it does happen: sequence is a variable, and must bereplaced in the theory by the more abstract dimension of order.24

3.2 The category set up to account for the stretches that carrygrammatical patterns is the unit . The units of grammar form a hierarchythat is a taxonomy. To talk about any hierarchy, we need a conver-sational scale; the most appropriate here might seem that of size, goingfrom “largest” to “smallest”; on the other hand size is difficult torepresent 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, eventuallypreferring the latter. The relation among the units, then, is that, goingfrom top (largest) to bottom (smallest), each consists of one, or of more than one, of the unit next below (next smaller). The scale onwhich the units are in fact ranged in the theory needs a name, and maybe called rank.

“Consists of”, like “unit” and “rank”, also belongs to the theory: its




44

realization in form varies between and within languages, and is statedof course in description.25 The possibilities are sequence, inclusion andconflation. 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 onefollowing, 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 notallow for upward rankshift. Third, only whole units can enter intohigher 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 butnot a unit of rank more than one degree lower than itself; and not, inany case, a part of any unit.26

3.3 The number of units in the hierarchy is a feature of the descrip-tion. It varies from language to language, but is fixed by the descriptionfor each language, or rather for each describendum or “etat 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, andan artificial language could be constructed on these lines (whereas itwould not be possible to construct an artificial language having onlyone 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 assignedby grammatical theory to any one unit. Since in any case only two, asa minimum, are required, only two would be available for special

status. As it happens we can assign special status to two grammaticalunits by reference to other levels on a “more / less” basis. There willalways be one unit which, more than any other, offers itself as an itemfor contextual statement because it does the language work in situ-ations: 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 andthus offers itself as an item for lexical statement;28 this we may as wellalways call the word . So, in grammatical theory, all languages have atleast two units; in description, all languages have sentences and alllanguages 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




45

when they turn up. For English, for the two units between sentenceand word the terms clause and phrase are generally used. It is at therank of the phrase that there is most confusion – because there are here

the greatest difficulties – in the description of English; one reason isthat 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 tomake a class distinction within it between group and phrase. Below theword, 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 morecomplete morphemes. The descriptive meaning of “consists of”, andthe possibilities of rankshift (including recursive rankshift), are stated asand where applicable. One distinction that is often useful is between amember of a unit that consists of only one member of the unit nextbelow 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 mustbe kept rigorously to this, and no other, use.31

3.4 The theory requires that each unit should be fully identifiable indescription. This means that, if the description is textual, every item of the text is accounted for at all ranks, through the various links of theexponence chain which involve, of course, the remaining theoreticalcategories. If the description is exemplificatory, exactly the same isimplied, except that the description proceeds from category to expo-nent instead of from exponent to category.

It will be clear from the discussion in the next sections that therecan 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 clausecan 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 isessential to “shunt”, and “shunting” is a descriptive method that isimposed on description by theory.




46

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 therepetition of like events. Likeness, at whatever degree of abstraction, isof course a cline, ranging from “having everything in common” to“having nothing in common”. The commonplace that no two eventsare ever identical, that the same thing can never happen twice, is of norelevance whatever to linguistics; as soon as description starts, however

little the generalization involved, absolute identity is a necessaryhypothesis, which is then built into the theory, as one endpoint of thelikeness cline. Likeness, including absolute identity, is of course re-defined for each level and each category.

In grammar the category set up to account for likeness betweenevents in successivity is the structure.32 If the relation between eventsin 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 orthographicscale: to those of us brought up on the roman alphabet this happens torun horizontally from left to right, which is enough reason for adoptingthis 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 structureis made up of elements which are graphically represented as being inlinear progression; but the theoretical relation among them is one of order . Order may, but does not necessarily, have as its realizationsequence, the formal relation carried by linear progression; sequence isat a lower degree of abstraction than order and is one possible formalexponent 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) place-ordered, though they can be listed (X, Y, Z) as an inventory of elements making up the particular structure.34 A structure is always astructure of a given unit.

Each unit may display a range of possible structures, and the onlytheoretical restriction is that each unit must carry at least one structurethat consists of more than one place.35 Each place and each element in




47

the structure of a given unit is defined with reference to the unit nextbelow. Each place is the place of operation of one member of the unitnext 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 followsfrom 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 “standsfor” means simply “is shorthand for”, like that of an initial: “ ‘U’ standsfor ‘United’.” In a few cases traditional names exist which can usefully

serve as names for elements of structure, with the initial letter as thedescriptive symbol. In the statement of English clause structure, for example, four elements are needed, for which the widely acceptedterms subject , predicator , complement and adjunct are appropriate.38

These yield four distinct symbols, so that S, P, C, A would be theinventory 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 thenames modifier , head , qualifier , giving an inventory M, H, Q: here, if the total range of possible structures is H, MH, HQ, MHQ, thesepossibilities 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, haveadvantages over letters in presentation, though there are not enough of them and they have to be redefined in description for each unit. It istempting sometimes to derive the symbols from the name given to thegrouping of members of the unit next below which operates at thegiven 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 avoididentifying this grouping, which belongs to a different category as wellas a different rank, with the element itself – therefore if this method isto be used at all it must be used all the time and a statement made tocover it.40

There are some instances where an element of structure is identifiedas such solely by reference to formal sequence: where the element isdefined by place stated as absolute or relative position in sequence. It




48

is useful to indicate that here sequence is so to speak built in tostructure, and this can be shown by an arrow placed over the symbolsfor the elements concerned. For example, in English clause structure it

is a crucial criterion of the element S that it precedes P in sequence:structures can be stated as SP

CA, SAP

A, ASP

, etc.41 This displays thecontrast between this situation, where S is crucially defined by positionrelative to P, and realized sequences of elements which are not,however, defined by sequence, which may be indicated by simplelinearity 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 (seebelow, 7.4). This is depth of detail, and is a cline running from a fixedpoint at one end (least delicate, or primary) to that undefined buttheoretically crucial point (probably statistically definable) where dis-tinctions 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 theminimum number of elements necessary to account comprehensivelyfor the operation in the structure of the given unit of members of theunit next below: necessary, that is, for the identification of every itemat 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 groupstructure, or all groups in clause structure, with fewer than theseelements or places.

Subsequent more delicate differentiations are then stated as secondarystructures. These are still structures of the same unit, not of the unitnext below; they take account of finer distinctions recognizable at thesame 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 secondaryclause structures differ in delicacy but not in rank.

As the description increases in delicacy the network of grammaticalrelations becomes more complex. The interaction of criteria makes therelation between categories, and between category and exponent,increasingly one of “more / less” rather than “either / or”. It becomesnecessary to weight criteria and to make statements in terms of probabilities. With more delicate secondary structures, different com-

binations of elements, and their relation to groupings of the unit nextbelow, have to be stated as more and less probable.44 The concept of




49

“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 beingan 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 allgrammar can be stated in probability terms, based on frequency countsin 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 isa scale of delicacy at all – shows that the nature of language is not tooperate with relations of “always this and never that”. Grammaticaltheory 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 thisfeature of language. The category of structure, for example, is the morepowerful because it can be used to state the patterns of a given unitcomprehensively at the primary degree without the assumption that ithas 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 nextbelow, and at each element operates one grouping of members of theunit next below. This means that there will be certain groupings of members of each unit identified by restriction on their operation instructure. The fact that it is not true that anything can go anywhere inthe structure of the unit above itself is another aspect of linguisticpatterning, and the category set up to account for it is the class.

The class is that grouping of members of a given unit which isdefined by operation in the structure of the unit next above. It accountsfor a paradigmatic relation, being a grouping of items “at risk” under certain conditions. It is related primarily to elements of structure: thefirst 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 firstplace, 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 yielddistinct secondary classes. If a given unit has primary structures XY,




50

XYZ, YZ and XYZY, the primary classes of the unit next below are“class operating at X”, “class operating at Y” and “class operating atZ”. 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 Y

a and only another

section (not necessarily mutually exclusive) at Yb, this yields as second-

ary classes “class operating at Ya” and “class operating at Y

b”. Secondly,

with increased delicacy the elements of primary structure will bedifferentiated into secondary elements. A primary structure generalizedas 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 secondaryclasses “class operating at p”, “class operating at q”, etc.

In the second place, more delicate classes appear whenever arestriction 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 suchthat 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 atX is 1 and that at Y is 2, a structure XY must have as its exponenteither 1.1+2.1 or 1.2+2.2. Secondary classes arrived at in this way indescription may be referred to distinctively as sub-classes, to indicatethat they are derived by differentiation from primary classes withoutreference to secondary structures; but it is important to state that thereis no theoretical difference here. The relation between structure andclass is a two-way relation, and there is no question of “discovering”one “before” the other. In any given instance there may be descriptivereasons for stating the one without the other; but all structurespresuppose classes and all classes presuppose structures.

5.3 What is theoretically determined is the relation between struc-ture and class on the one hand and unit on the other. Class, likestructure, 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 alwaysdefined with reference to the structure of the unit next above, andstructure 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 arederived “from above” (or “downwards”) and not “from below” (or “upwards”).

The distinction between downward and upward movement on the




51

rank scale is important in grammar, but it is a mistake to raise it to thestatus 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 functionallydetermined, since they are set up with reference to the form of theunit next above – the whole description is both formal and functionalat the same time, and “function” is merely an aspect of form. Thedistinction does, however, need a name, and this seems the best use for the terms “syntax” and “morphology”. Traditionally these terms haveusually referred to “grammar above the word” (syntax) and “grammar

below the word” (morphology); but this distinction has no theoreticalstatus.46 It has a place in the description of certain languages, “inflex-ional” languages47 which tend to display one kind of grammaticalrelation above the word (“free” items predominating) and another below the word (“bound” items predominating). But it seems worth-while making use of “syntax” and “morphology” in the theory, torefer to direction on the rank scale. “Syntax” is then the downwardrelation, “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 destroysthe consistency in the use of the term “class”, no longer restricting itto a category of unique theoretical status. When a name is needed for morphological groupings, groupings of items on the basis of likeness intheir 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 primaryclasses first, since these form the link between elements of structureand more delicate classes.50 For the different classes of each unit, manynames are available, especially at word rank – though it is precisely thelong 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 classidentification between units: if certain word–classes only (or predomi-nantly) operate in one class of group and others only in another, onecan talk of verbal group and verb (= “verb word”), nominal group andnoun (word) and so on. At the same time it is safer not to allowcomplete 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




52

avoiding confusion with elements of structure. This is not only the-oretically desirable, because of their different status; it has descriptivevalue 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 primaryclasses 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 useof 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 orderedrepetition of like events that makes up the patterns and the groupingof like events by their occurrence in patterns. What remains to beaccounted 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 theitem itself considered as a formal realization53 but the class, which isnot a list of formal items but an abstraction from them. By increasein delicacy, the primary class is broken down into secondary classes of

the same rank. This set of secondary classes now stands in the relationof 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 secondaryclasses 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 thatis implied by the nature of the class (as a grouping of items) but that isdisplayed first still in abstraction, by reference to the category of classitself.54

6.2 Systems of secondary classes thus allow the description to remainat a high degree of abstraction while displaying at each step, eachincrease in delicacy, a more finely differentiated range of choice. This




53

is the value of the concept of “sub-class” (above, 5.2), since wherever a choice among a finite number of mutually exclusive possibilitiesis 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 bestated.

Thus the system provides what is, in the order in which thecategories are presented here, the final requisite for the linking of thecategories to the data.55 Through the system, in one of two ways(below, 6.3), the description can now account for the formal expo-

nents, the items identified in linguistic form – and through them islinked, by description at other levels, to the ultimate exponents insubstance. 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 thesame 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 sametime the aim of grammar is to stay in grammar: to account for as muchas can possibly be accounted for grammatically, by reference to the

categories of grammatical theory. This, since it implies maximumgeneralization and abstraction, means that one proceeds from categoryto exponent by the longest route that is compatible with never goingover the same step twice.

Here it is important to avoid theoretical confusion between thescales of exponence and rank. Since the relation of class to structure issyntactical (as defined above, 5.3), if one derives classes from structuresthen “remaining in grammar” means moving step by step down therank 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 theexponence scale. But it is not. The sentence stands in exactly the samerelation to its exponents as does the morpheme; one can move over (atright 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 couldequally 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 thatone reached the exponents.

6.3 There comes a point, however, when one is forced out to theexponents; and this happens in one of two ways. In the first case the




54

description yields a system in which the formal exponents themselvesoperate as terms. Here we have gone all the way in grammar; theformal items are grammatically contrastive (and do not belong in the

dictionary60). In the second case the description yields a class where nofurther breakdown by grammatical categories is possible, a class whoseexponents make up an open set. Here we must leave grammar; therelations between the exponents must be accounted for as lexicalrelations.

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 iscalled in description the “word” (above, 3.3). The system of exponentsalso tends to operate at the lower end of the rank scale; but this,although predictable as economy of resources, is not a theoreticalrestriction: the rank distribution of formal exponents in systems is adescriptive feature which, in a five-unit description, may be expectedto 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 asthe 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 thewhole of linguistic form into grammar, hoping to show that lexis canbe defined as “most delicate grammar”. The exit to lexis would thenbe closed, and all exponents ranged in systems. No description has yetbeen made so delicate that we can test whether there really comes aplace where increased delicacy yields no further systems: relations atthis degree of delicacy can only be stated statistically, and seriousstatistical work in grammar has hardly begun. It may well be that thenature of language is such that this “most delicate grammar” willevaporate in distinctions which are so slenderly statistical that thesystem 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, ason a different level, and to require a different theory to account for them.

6.4 A brief illustration, from English grammar, of the four fundamen-tal categories. The exponent of the element S in primary clausestructure is the primary class nominal of the unit group. The primary




55

structure of the nominal group is (M . . .n)H(Q . . .n). The primaryclasses of the unit word operating at H and M are respectively the nounand the pre-noun. The element M can be broken down into secondary

structures composed of any combination of the elements D, O, E inthat order, whose exponents are the secondary classes deictic , numera-tive and adjective of the word. Deictics include a number of systems

whose terms are the formal exponents themselves, with further second-ary classes separating, for example, “all / both / such / half ” (class 1, atD

a) from “a / the / some , etc.” (class 2, at D

b) 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 scalesof rank, exponence and delicacy. These need to be kept apart from

each other, and also from two other things: regressive structures,63which are handled in the description by rankshift, and the range of levels by which grammar is distinguished from other aspects of languagein the first place.

The separation of levels has been taken as a starting-point, and it isnot the purpose of this paper to explore this part of the theory. Myown view is that, whatever is decided for presentation, which will varywith purpose, both in theory and in description it is essential to separatethe levels first and then relate them. The theoretical reason is thatdifferent kinds of abstraction are involved, and therefore differentcategories. In description, the attempt to account for the data at alllevels 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, tocombine 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 hasrecourse to such things as a system of juncture in which any of theexponents in substance can be an exponent of any of the terms in thesystem. The exponents of the phonological unit which carries contrastsof intonation in English, the tone group, are in fact not coextensivewith the exponents of any one or even any two of the grammaticalunits: if one describes English grammar with the tone group as a unit,when it can be exponentially coextensive with every one of what a




56

“monolevel” grammar would regard as five different units, one fails toshow what is in fact the relation between the grammar and thephonology of the language.64

A meaningful account of how grammar and phonology are relatedin English must be based on a prior separate statement of the two.65 Assaid above (1.4), it is misleading to think of the levels as forming ahierarchy. They represent different aspects of the “patternedness” of linguistic activity. In this respect phonology is an “interlevel”, sincethere 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 bestated separately because of the distinct nature of the abstractionsinvolved, and the fact that (not surprisingly) the patterns are carried bydifferent stretches of exponents, are only significant with reference tothe two levels of form and substance. Context is an interlevel in adifferent 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 thisto language form. Though there is no precedence or priority, there isof course order among the levels, as determined by their specificinterrelations; and in the study of language as a whole form is pivotal,since it is through grammar and lexis that language activity is – and isshown to be – meaningful.

7.2 The scale of rank has been discussed (above, 3) with reference tothe unit, the basic category which operates on this scale. The syntactic(‘downward’) determination of classes is a feature of the theory, so thatwith respect to the category of class the rank scale appears as one

of logical precedence running from highest to lowest unit. But thisprecedence applies to class criteria only, so that even in the theory it isnot a one-way relation: the theory itself embodies “shunting” (movingup and down the rank scale) as crucial to the interrelation of the

categories. In description, all statements presuppose shunting; thedescription 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 presenta-tion seems easier to make clear, but this may well be overridden – for example if grammar is an adjunct to lexical description, as sometimesin the statistical study of lexis.

Rank is distinct both from exponence and from delicacy. A shift in




57

one never by itself entails a shift in either of the others. The reasonwhy rank is often confused with other scales is that there are caseswhere a shift in rank does accompany a shift in something else; but this

is always by virtue of the logical relations among the categoriesinvolved. 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 betweenexponence and rank. The descriptive relevance of keeping the scalesdistinct 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 seemnecessary to recognize four different scales of exponence, one leadingfrom each category. In fact, however, exponence can be regarded as asingle scale.

In the first place, each category can be linked directly by exponenceto the formal item: it is in fact a requirement of the theory that anydescriptive category should be able to be so linked. This may be statedby way of exemplification, as when we say “the old man is (an exampleof) 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 astage 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 andmoving out to lexis while this is still avoidable, the description takessuccessive steps down the exponence scale, changing rank wherenecessary, until (at the degree of delicacy chosen) it is broughtunavoidably face to face with the formal item.

In the second place, therefore, the step by step move from any onecategory 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 isreplaced by which). While therefore the categories are distinct, theyare 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 itmay also be called “exponence”, entails a new scale, in which the




58

nature of the abstraction is different and the formal item is now at theother end – it is itself the abstraction, and is not in any way delimitedor categorized by grammar.68

When grammar reaches the formal item, either it has said all there isformally to be said about it or it hands over to lexis. The formal itemis the boundary of grammar on the exponence scale. It is not of coursethe boundary on the rank scale: whenever the formal item is anythingother 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) entersinto a system at group rank, while pickup emerges from the grammar asa word, though being a lexical item it would not necessarily be anexponent of any whole grammatical unit. Once it has been taken over by lexis, the grammatical categories, and the grammatical exponencescale, 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 beyondwhich no further grammatical relations obtain: where there are nocriteria for further secondary structures, or systems of secondary classesor formal items. In description, delicacy is a variable: one may chooseto describe a language without going beyond the primary degree, stillbeing comprehensive in rank and exponence and making use of all thecategories of the theory. Each subsequent increase in delicacy delaysthe move to the exponents (cf. above, 7.3) and thus increases thegrammaticalness of the description. The limit of delicacy is set by themeans 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 large-scale textual studies with frequency counts and complex secondaryclassification based on multiple criteria, criteria which often cut across

each other and may have to be variably weighted. And, as suggestedabove (6.3), a point will perhaps be reached where probabilities are soeven as to cease to be significant69 and classes so delicately differentiatedthat the description will have to decide on crucial criteria and ignorethe others,70 thus setting its own limits.

Delicacy is distinct from rank and the limit of delicacy applies at therank 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




59

grammatical differentiation of items which then remain to be lexicallydifferentiated: 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 hasto move from abstract category to exponential item. That set on thedelicacy scale is where differentiation ceases: the set of exponents of

each class, and of each element of structure, permits no further, moredelicate groupings. If the formal items are still not ranged in systems,the implication in either case is that further relations among them arelexical.

Whether or not, as discussed above (6.3), grammatical delicacy canreach a point where there is a one / one category–exponent relation(where each element of structure, and each class, has only one formalitem as exponent), when all formal relations, including those amongwhat are now treated as lexical items, can be accounted for by thegrammatical categories and stated grammatically – in other words,whether or not, ultimately, all linguistic form is grammar, we do notknow. At present, lexical items must be treated separately, and lexical

relations established in their own right. These lexical relations do notdepend on grammatical categories71 (so they are not yet “most delicategrammar”) 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 cannothandle.

8 Lexis

8.1 This section is intended merely to bring lexis into relation withgrammar, not to discuss the theory of lexis as such. As has been pointedout (above, 3.3. and 6.3), there is no one / one correspondence inexponence between the item which enters into lexical relations andany one of the grammatical units. It is for this reason that the termlexical item is used in preference to word , “word” being reserved as the

name for a grammatical unit, that unit whose exponents, more thanthose 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 gram-matical 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 sameformal item) may be a number of different grammatical items, so it is




60

not true that one lexical item always has the same relation to the rankhierarchy. 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 thewhole of another, sometimes discontinuously; and (iii) one and the

same lexical item may in different occurrences cover any range of thepossibilities under (i) and (ii).72

This does not mean that lexical items cannot be identified ingrammar; it means that they are not identified by rank. They are

identified, as has been suggested (above, 6.3), by their being unac-counted 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 nextstage is to relate them; and it is here that the complex relation betweenlexical item and grammatical unit must be accounted for. This is exactlyparallel 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 lexicalitem and the categories of phonology.

8.2 The task of lexis can be summed up, by illustration, as that it hasto account for the likelihood of wingless green insects and for the, bycontrast, unlikelihood of colourless green ideas.73 As in grammar, we shallexpect language to work by contrasting “more likely” with “less likely”rather than “possible” with “impossible”; but, as has often been pointedout,74 this particular type of likelihood is not accounted for bygrammar, at least not by grammar of the delicacy it has yet attained. Itis, however, too often assumed that what cannot be stated grammat-ically cannot be stated formally: that what is not grammar is seman-tics, and here, some would add, linguistics gives up.75 But the viewthat 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 areneeded are theoretical categories for the formal description of lexis.

It seems that two fundamental categories are needed, which we maycall collocation and set .76 The first basic distinction between these andthe categories of grammar is that in lexis there are no scales of rank andexponence. 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




61

and a collocation is a collocation of formal items; therefore noexponence scale (exponence there is, of course, but it is a simplepolarity). 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 aparadigmatic one. There the resemblance ends.

Collocation is the syntagmatic association of lexical items, quantifi-able, 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 withwhich it is collocated being ranged from more to less probable; anddelicacy is increased by the raising of the value of n and by the takingaccount of the collocation of an item not only with one other butwith two, three or more other items. Items can then be groupedtogether by range of collocation, according to their overlap of, so tospeak, 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 progressivelydifferentiated as the degree of collocational likeness set as definingcriterion increases.77

In lexis, as in grammar, it is essential to distinguish between formaland contextual meaning.78 Once the formal description has identifiedthe categories and the items, these can and must be treated context-ually. The formal item of lexis, the lexical item, is unrestricted gram-matically; grammatical categories do not apply to it,79 and theabstraction of the item itself from a number of occurrences (including,for example, the answer to the question whether one is to recognizeone lexical item or more than one) depends on the formal, lexicalrelations into which it enters. The nature of these relations is suchthat formal statements in lexis require textual studies involving large-

scale frequency counts: not of course of the frequency of single items,but of items in collocation. Since these are no longer difficult toundertake, it should not be long before we find out much more abouthow language works at this level.




62

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, itis limited in relevance; its purpose is to throw light on, and suggestproblems of, the categories of grammar by relating these to an activitywhich is familiar and for much of which a terminology is ready tohand.

The presentation of a framework of categories for the description of eating might proceed as follows:

Units:

Daily menu

MealCourseHelpingMouthful

Unit: Daily menu

Elements of primary structure E, M, L, S (“early”, “main”,“light”, “snack”)

Primary structures EML EMLS (conflated asEML(S))

Exponents of these elements

(primary classes of unit‘meal’)

E: 1 (breakfast)

M: 2 (dinner)L: 3 (no names available; see

secondary classes)S: 4

Secondary structures ELaSaM EL

aM EML

bSb

EMSaLc

Exponents of secondaryelements (systems of

secondary classes of unit‘meal’)

La: 3.1 (lunch)

Lb: 3.2 (high tea)

Lc: 3.3 (supper)Sa: 4.1 (afternoon tea)

Sb: 4.2 (nightcap)

System of sub-classes of unit‘meal’

E: 1.1 (English breakfast)1.2 (continental breakfast)

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




63

Unit: Meal, Class: dinner

Elements of primary structure F

, S, M, W, Z (“first”,

“second”, “main”,“sweet”, “savoury”)

Primary structures MW MWZ MZW FMWFMWZ FMZW FSMWFSMWZ FSMZW(conflated as (F(S)MW(Z))

Exponents of these elements(primary classes of unit

“course”)

F: 1 (antipasta)S: 2 (fish)

M: 3 (entree)W: 4 (dessert)Z: 5 (cheese*)

Secondary structures (various, involving secondaryelements

Fa–d, Ma,b, Wa–c)Exponents of secondary

elements (systems of secondary classes of unit“course”)

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*)

Systems of sub-classes of unit“course”

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 operatingin meal structure

Fc: grapefruit / melonFd: grapefruit juice / pineapple

juice / tomato juiceMa: beef / mutton / porkMb: chicken / turkey / duck /

goose




64

At the rank of the “course”, the primary class “entree” has secondaryclasses “meat dish” and “poultry dish”. Each of these two secondary

classes carries a grammatical system whose terms are formal items. Butthis system accounts only for simple structures of the class “entree”,

those made up of only one member of the unit “helping”. The class“entree” also displays compound structures, whose additional elementshave as exponents the (various secondary classes of the) classes “cereal”and “vegetable”. We will glance briefly at these:

Unit: Course, Class: entree Elements of primary structure J, T, A (“joint”, “staple”,

“adjunct”)Primary structures J JT JA JTA (conflated as

J((T) (A)))Exponents of these elements

(primary classes of unit“helping”)

J: 1 (flesh)T: 2 (cereal)A: 3 (vegetable)

Secondary structures (various, involving – amongothers – secondaryelements Ja,b, Ta,b, Aa,b)

Exponents of secondaryelements (systems of secondary classes of unit

‘helping’)

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 grammaticalsystems or in classes made up of lexical items (marked *). Thepresentation 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 isneither more nor less fundamental than any other unit. But how little

it reveals, if description proceeds in one direction from it, of thecomplexity of the whole activity!

Our predecessors thought of language as an organism, and drew




65

their analogies from evolution. We reject this as misleading; but no lessmisleading is its familiar substitute, according to which language is anedifice and the morphemes are the bricks. Perhaps if language had been

thought of as activity we should never have heard of “morphemics”. Itis unfair on one who was among the greatest figures in linguisticscience to call the “bricks and edifice” view of language “Bloomfield-

ian” – 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 withmany 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 questionablefeatures of “Bloomfieldian” linguistics. What is being called in questionis 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 wouldperhaps justify critical comment.81 These are:

(1) confusion of “level” with “rank”(2) confusion of “rank” with “exponence”(3) conflation of different level levels(4) overemphasis on lowest unit

(5) reluctance to shunt

(6) reluctance to shade(7) distribution of redundancy

The confusion of level with rank takes a specific form: the relation

between different units at one level (morpheme . . . sentence) isconflated with the relation between two different levels, grammar andphonology. 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 thelowest to the highest; cf. below, 10.5) in (a) phonology and (b)grammar. In description, this leads to unwanted complexity and thusweakens the power of the grammar: one must be free to recognizegrammatical units whose exponents in substance both overlap with andcompletely fail to coincide with the units carrying phonological con-trasts.82 In the theory, there is a confusion of abstractions: the abstrac-




66

tion involved in relating one unit to another at the same level is quitedifferent from that involved in relating one level to another. The sensein which a sentence “consists of” morphemes is stated in description

with reference to its definition in the theory, and is totally differentfrom the sense in which a morpheme “consists of ” phonemes – indeedit is doubtful if there is any meaningful sense in which a morpheme

consists of phonemes. (The contrastive use of “morph” and “mor-pheme” is designed to build in an extra stage to account for the twokinds 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 under-lying their phonological abstraction.83)

10.2 The relation among the units also tends to be confused withthe relation between a category and its exponent(s). It is assumed thatin moving up the rank scale, from morpheme to sentence, one is alsomoving 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 gram-

maticalness of language, one takes the final step on the exponencescale at the lowest rank possible (though this, as already shown, isby no means always the morpheme),84 and this is probably the reasonfor the confusion of rank with exponence. But the scales of rankand exponence are again different dimensions of abstraction, andone can link any unit directly to its formal exponent (and throughthis to its exponent in substance): the relation of an exponent of theunit “sentence” to the category of sentence is exactly the same asthat 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 confusionof level with rank (above, 10.1). The theoretical basis of this criticismis 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 adistinction in form, with meaning either at the grammatical or at thelexical level.85 All formal distinctions presuppose some distinction insubstance,86 and once a distinction in substance is shown to carry aformal distinction it must be accounted for in phonology.87 But norelation whatever is presupposed between the categories required tostate the distinction in form (grammar or lexis) and the categories

required to state phonologically the distinction in substance which




67

carries it.88 For example: the units may not be coextensive, and avariable relation of phonological unit to grammatical unit may be thevery thing that carries a formal system (cf. above, 10.1 n. 82); or a

single system in grammar may be carried by different phonologicaldistinctions, say two of its terms by tone and a third by addition of asegment; or a phonological system, such as tone associated with a given

unit (recognized as phonological because it carries some formal distinc-tion), may carry different formal distinctions, part grammatical and partlexical, 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 adescription 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 thedescription of the sentence, the group and so on in terms of their structures and classes: hence the mutual definition of all units and of allgrammatical categories (and, procedurally, until a description is “com-prehensive” (primary delicacy at all ranks) all parts of it remain subject

to revision). A grammatical category is not required to be identifiableby reference to a particular feature of substance stated phonologically:it merely carries the potentiality of being stated in phonological termsthrough a long chain of exponence.90

(The starting point here of course is the theory of grammar, so thatwhat is being considered, and objected to, is the identification of grammatical categories on phonological criteria. For linguistictheory as a whole, the question must also be formulated the other wayround: 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 categoriesshould be derived from those of form; and this is usually done in

prosodic phonology, as developed by Firth and others, though it is notan inherent requirement of prosodic method.91 Inter-level dependence

in this direction is theoretically justifiable, since the role of phonologyis to account for the formally meaningful organization of phonicsubstance. But it has descriptive dangers: first, that a system carriedonly by variable relation between grammatical and phonological unit

may be missed, and second, that unless a comprehensive formaldescription has first been made, formal features may be distorted into aphonological mould – the phonologist may take the “word” class“verb” as a phonological unit, but he (or someone) must have described




68

the grammar first, or his “word” and “verb” may not turn out to bethe 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 grammaticalcategories, as when supra-segmental phonemes are used as criteria for

the category of “phrase”.92 A phrase is a phrase because it operates inthe 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 phonologicalcategories which are arrived at by a different process of abstraction on

the basis of the minimal requirement that some formal distinction isalways involved. If at any point this yields a one / one relation of categories, so much the better: if the phrase turns out to be exponen-tially coextensive with the tone group, the latter can be used as arecognition signal for the former. But it remains a signal, not a criterion.

In description, the trouble arises when phonological features are usedas grammatical criteria even when they clash. If, for example, thesegment which carries tone contrasts, or is bounded by juncture

features, will not work as a grammatical unit, then tone and junctureare no use even as recognition signals; so to “define”, say, the clauseby reference to tone or juncture one has to set up a phonologicalsystem in which any feature in substance can be an allophone of anyterm in the system. There will be clash; and if it is recognized in thefirst place that there is nothing at all surprising when, say, units carryingformal patterns do not coincide with those carrying patterns of theorganization of substance, then the search for one / one phonologicalidentification signals of grammatical categories, such as a phonologicalstatement of clause boundary, can be abandoned as being withoutprofit.93

10.4 In both grammar and phonology the smallest units, morphemeand phoneme, are often assigned a special status distinct from that of any of the other units. Since the description usually proceeds unidirec-

tionally upwards (consistently in phonology, and in grammar up toabout group rank; cf. below, 10.5) these units are treated first: theyare then rated (uniquely) fundamental, and in phonology “supra-segmental” features are given phoneme status. The morphemebecomes a grammatical brick (here the analogy does yield an inter-

esting 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”.




69

The result is sometimes referred to as excessive segmentation, thoughthis is not really the essence. The objection is not that structural analysisis carried through to the morpheme; it is clearly necessary that it should

be, though not to the extent of requiring one morpheme with distinctsegmental exponent for each system in which the item in question is aterm.94 The objection is that the smallest unit, being mistaken asfundamental by contrast to the others, is made to do much too muchwork: to carry features and contrasts which properly belong to larger units. The morpheme is no more and no less “fundamental”, being nomore and no less an abstraction, than the sentence. It has enough

patterns of its own to carry without having foisted on it those that thelanguage 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 mor-pheme, through the word to somewhere around the group.96 Notsurprisingly, since there is no shunting, it proves extremely difficult totake 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 toshow the relation among the different units: to permit a unifieddescription with links, through all categories, all the way from mor-pheme to sentence. In the absence of shunting, the description has to

jump to the top end of the rank scale and proceed downwards fromthe 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 twowill meet in the middle. With the aid of a good homogeneousmathematics they might; but two totally different bodies of method areinvolved, 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 toavoid leaving the middle ragged. Presentation can of course proceed in

any direction that is desired; but it needs to be based on a descriptionthat permits – indeed presupposes – constant shunting.

10.6 It has been suggested at various places above that the theorycannot validly regard constrasts and relations that are clearcut, andstatable in “yes / no” terms, as the norm in language. Even at theprimary degree of delicacy the description will encounter featureswhere “shading” is necessary: where a feature is better stated in terms




70

of “more likely / less likely’ and a statement is more powerful when itaccounts for only 90 per cent of occurrences; and this becomes moretrue 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 turnall constrasts and relations into “yes / no” ones, an attempt which mayslant the choice of criteria. Criteria are sometimes chosen for thispurpose which make it necessary to account for a large number of occurrences by changing them into something else, because they donot display the contrast chosen as crucial. (This device is more readilytolerated by those who give relatively little weight to textual as opposed

to exemplificatory description.)For example, the exponent of the element “subject” in Englishclause structure is sometimes identified as being that nominal groupwhich is in person / number concord with the verbal group. But sincea large number of English clauses do not display this concord contras-tively99 (quite apart from those which violate it, as does the clause Ihave just written), the textual application of this criterion has to relyon 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 innumber, on the dissimilation of one of them. If instead one says thatthe exponent of “subject” is that nominal group which precedes theverbal group with no other nominal group in between, this can be

stated as a high probability relation in spoken English and is applicableto 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 begrammaticizable and others, for the moment, not. And it ceases to bepuzzling why, if a foreigner (or native) says my friends is inviting the neighbour , every English speaker knows that it is the friends who aredoing the inviting and not the neighbour.

10.7 The problem of “redundancy” is complex, and needs treatingseparately and at length; but the term has become a commonplace in

description and a brief reference may be made here. The name isassigned to a number of varied phenomena, none of which is related inany 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 hasno relevance to the language at all.




71

“Redundancy” is assigned to what is displayed as multiple expon-ence: either in form, where more than one formal item is said to bethe exponent of one grammatical category, or in substance, where a

distinction is carried by what is said to be more than one phoneticfeature. 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 thanone” exponent as opposed to “one”, and that these are exponents of “the same” category.101 Discontinuous morphemes, for example, maysometimes 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 differentfrom the category of which the form is exponent, and that has its ownformal meaning.103 Redundancy in substance appears when formal or phonological distinctions are related to contrastive features.104 Hereprecisely the same problem arises, since it is not possible to giverigorous criteria for deciding what is “one” phonetic feature and what

is “more than one”.105 Each time a new parameter or a further degreeof 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 acontrast (or system) is assigned to a unit lower than that to which it isappropriate, and may result therefore from overemphasis on the lowestunit. This tends to happen more in phonology, when the phoneme ismade to carry contrasts appropriate to a higher unit; one of the meritsof prosodic phonology is that it avoids this error.106 But it is notunknown 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 theorythat the “appropriate” unit for the assignment of any feature is the

highest unit that can carry it without requiring the statement to bemade twice. The best description therefore can be thought of as thatwhich minimalizes artificial “redundancy”. But at the same time thoseinstances where what is called “redundancy” is an artificial product of the description are not essentially different from, but are merelyextreme 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




72

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 “redun-dancy” is unfortunate for two reasons. On the one hand it implies that

some features, by contrast to others,109 can be recognized as carried bysomething more than what “would be enough” – to the extent evenof 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 thiskind, its relation to the redundancy of information theory is extremelycomplex and it would be better not to call it by the same name. The

redundancy of information theory is of considerable interest to linguis-tics in the study of the information carried by grammatical systems; butthis, as far as I know, has not yet been seriously attempted. Thequantification of systems, rather than the appraisal of features ascontrastive or idle, which rests on a very partial interpretation of the

redundancy of information theory, seems the more useful role for theconcept of redundancy in linguistics.

Most, if not all, of the points made in this section can be broughttogether under Chomsky’s observation that “a linguistic theory shouldnot be identified with a manual of useful procedures, nor should it beexpected to provide mechanical procedures for the discovery of gram-mar”.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 apractical evaluation procedure for grammars”?113 This it must do. Butit can be asked to do more: to provide a framework of logicallyinterrelated categories (so that it can be evaluated as a theory, andcompared with other theories) from which can be derived methodsof 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




73

theoretical foundation. The point of view adopted here is that transfor-

mation-generation is a type of description which, like other types,

depends on but does not replace a theory.

2. Even Chomsky (1961) seems to imply that a textual study cannot betheoretical. 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.

3. 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 state-ments of meaning so that we may see how we use language to live”

(p. 23; cf. also p. 11).

4. 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).5. 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).

6. 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).

7. 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 gener-

alizations and hypotheses by which the theory was arrived at in the first

place. The latter precede the theory and are not susceptible of “rigori-

zation”; 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.

8. 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 linguis-

tics (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




74

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).9. Cf. McIntosh (1956). Professor McIntosh followed this up subsequently

in a study of the underlying theoretical problems.

10. This figure is a schematic representation of §§1.5–1.7.

11. Cf. Firth (1957b): “A nominative in a four-case system would in this

sense necessarily have a different ‘meaning’ from a nominative in a two-

case or in a fourteen-case system, for example” (p. 227). The article

from which this is quoted, “General linguistics and descriptive gram-

mar”, 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).

12. 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 quanti-fied, 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.

13. 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.

14. 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), aswhat 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 auto-

matically divide, on the preverbal level, all events into three categories:

those anterior to PP (point present) . . . , those simultaneous with PP




75

. . . , 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.

15. 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 thesteps 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”.

16. Cf. Firth (1957a: 22), and above, 1.8 n. 11; Garvin (1957); see below,

6.3 n. 60; Robins (1959).

17. “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.

18. 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 contextualdistinctions; 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 . . .”).




76

1 me me

1+ noi noi

2I te te

2I+ voi

2E lei voi

2E+ loro }

(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.19. 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”.

20. 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 phonet-

ics 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).

21. 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).)22. The theory thus leads to “polysystemic”-ness in description – both

syntagmatically and paradigmatically. Syntagmatic polysystemic state-

ment 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:




77

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).23. 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 canthen be distinguished as “realization” and “manifestation”.

24. 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).

25. Cf. above, 3.1 n. 23.

26. 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.

27. Such as the status of “being the smallest”.

28. 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




78

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

rank

(

sentence

clause

group (phrase)

word

morpheme

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 thelatter: orthographically, The man came(,) from the Gas Board . Phonolo-

gically (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




79

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 //

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 distinc-

tion 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 constructionof 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 aslinear 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.

34. 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.

35. 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 “mor-

pheme” would be one and the same unit.

36. For the name and nature of this grouping, see below, 5.

37. 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 phono-logic 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




80

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 accidentfor 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 (systemi-

cally 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 primitiveterms with the same interrelations.

38. 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.

39. 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 formulawould then read (M . . .n)H(Q . . .n), where . . .n allows infinite progres-

sion (not regression).

40. 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.

41. If, instead, an inventory of elements is stated first, 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.

42. 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 distinctions beween 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.




81

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 distinctsecondary structures, the first example having D

2D

bO, the second

DbOE.

44. When Hockett writes (1955: 17) “In general, then, if we find continu-

ous-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 cannotbe a cline. But units and classes are not crucially discrete: in exponence,

units display syntagmatic non-discreteness (syncretism); classes, paradig-

matic non-discreteness (statable in various ways, such as multidimen-

sional 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 objectionto 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




82

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 theydo, 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 sameword–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 mem-

bers 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

Highest

abstraction

Scale of

Exponence

Data

Primary

structure

Primary

class

Secondary

structure

Secondary

class

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 highdegree 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.




83

52. Cf. Firth (1957b): “Various systems are to be found in speech activityand when stated must adequately account for such activity. Scienceshould not impose systems on languages, it should look for systems in

speech activity, and, having found them, state the facts in a suitablelanguage” (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 theplace of) an element of structure (S or C, of the unit “clause”). Theformal item itself, of course, has its own (and ultimate) exponents inphonic or graphic substance.

54. Diagrammatically (axes as in 5.4 n. 50, above):Primary structure

X Y Z Y

1 (at X)

2 (at Y)

3 (at Z)

Primary classes

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

S Y S T E M

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): “Atheory derives its usefulness and validity from the aggregate of experienceto which it must continually refer in renewal of connexion” (p. 1). Therelation of category to exponent can be generalized as one of abstraction;one endpoint of this relation may be any one of four categories, butthere is no scale of abstraction among the categories – their relation toeach other is such that the move from any category to its exponent maybe made either directly or via any or all of the other categories. As saidbelow (6.2), the route may involve rankshift; but this does not meanthat rank is to be equated with exponence or that there is any distinctionbetween different units as regards the kind or degree of their relation totheir exponents. (That is to say, even if one chooses to move from“clause” to exponent via “group”, this does not mean that the group isin any sense “nearer” to the data than the clause; indeed, the move from




84

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 structures . . .

clause (system of ) classes . . .

structures

group (system of) classes . . .} exponents

etc. etc.

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 scaleinto 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 alterna-

tive is to be chosen which is most congruent with the grammatical




85

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.

65. 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 whichsimplifies the statement of the relations between phonology and gram-

mar. 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?”

66. 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 compareclasses 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.

67. 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’smajor contributions to linguistic theory. References will be found

throughout the writings of Allen and of Carnochan, Eugenie Hender-

son, 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.

68. Strictly speaking the relation of the formal item to its exponent in

substance entails a two-fold relation of abstraction, one of whose




86

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 compre-

hending 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).

69. 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, andcline, of the determination of probabilities by the occurrence of a

member of each class. (Cf. my review of Whatmough (Halliday 1958).)

70. 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 Eng-

lish, shows that fifteen different formal criteria yield fifteen different sets

of classes.

71. 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 hetero-

geneity 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.

72. 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) (ii)Primary structure: SPA SPC

Exponent of P: class: verbal verbalof unit: group Group

formal item: ran ran up

Down the rank scale, verbal groups ran and ran up:

(i) (ii)Primary structure: F Fpo

Exponent of F: class: verb finite verb finiteof unit: word Word

formal item: ran 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




87

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 andthen 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 compli-

cates the description of the verbal group.) If the analysis is taken to themorpheme, 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 lexicalitem, 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 coun-

tenance 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 “grammat-

ical 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 contextrelates 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).




88

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) – thoughFirth’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 theoret-

ical 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: “Accord-

ingly, the signals (linguistic forms, with morphemes as the smallest

signals) consist of different combinations of the signalling-units (pho-

nemes), 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 calledthe “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 comprehen-

sive account of English grammar yet published and is an example of the




89

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, inthe 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:

Substance Form

L1 [a] / [c] - (free variant)

L2 ’ singular / plural

(grammar)

L3 ’ ‘cat’ / ‘dog’ (lexis)

(Phonologically restricted variants may be of either type: the languagemight 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 [pat]

and [pat], or [’pata] and [pa’ta], etc.

87. Again it does not matter where: the following grammatical contrasts in




90

English are carried by substantial features requiring very different

phonological statements:

Form Phonology

(Unit) (Contrast)

I work : I worked phoneme: addition of segment /t/

they’ve worked: they’d worked ” : replacement of /v/ by

/d/

he was working: was he working syllable: change in sequence

he was / working : he / was

working

foot: shift of unit boundary (so

re-distribution of strong

and weak syllables –

change of tonicity)

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

he was / working //

4 tone

group:

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 saidbelow, it must carry the potentiality of being so stated, since it presup-

poses 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

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 thephonic 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




91

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 theclass 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 objec-

tionable – and leads, not surprisingly, to a phonology in which any

substantial feature is a possible exponent of any term in the phonological

system!

91. A phonological description will, in this view, be prosodic if (i) itincorporates 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).

92. Cf. Haas’ reference to “a structure of a number of pyramids, all inverted”

(1960: 267).

93. I personally feel that English requires a totally different set of phonolog-

ical 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”.

94. 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).

95. I should thus agree with Robins (1959): “The morpheme must berecognized 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, how-

ever, 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




92

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.

96. On the implications of upward description cf. Haas (1960: 263–9).97. 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 ranktransformation).

98. 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 theseare 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.

99. 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.

100. 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 theoret-

ically 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.

101. Harris recognizes this in his use of the “broken morpheme” (1951:

165–7).

102. 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 anoun 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?

103. 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




93

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 (sincenot all contiguous items are members of the same sentence element).

But even if concord and contiguity were completely mutually deter-

mined, the problem would still not arise, since there would then be no

valid reason for not treating the whole complex as a single exponent.

104. 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”.

105. They might perhaps be acoustic, so that all but one of the formants

which distinguish [a] from [i] would be redundant?

106. 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).107. 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.

108. Another use of this same “redundancy” which has not been mentionedhere 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




94

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.

109. In fact, all description of language is the description of this “redun-dancy”. 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”.110. 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).

111. Chomsky (1957: 55n.).

112. Cf. Firth (1955: 93, 99; 1957a: 1).

113. Chomsky (1957: 52).



95

Chapter Three

CLASS IN R ELATION TO THE AXES OF

CHAIN AND CHOICE IN L ANGUAGE (1963)

This paper is concerned with the nature of the class as a theoreticalterm in Descriptive Linguistics: that is, with the meaning of “class” instatements which are made in order to explain how a given languageworks. 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 thepaper, that this category of “class” is to be defined syntactically. By thisI mean that the concept is introduced into the description of a language

in order to bring together those sets of items that have the samepotentiality of occurrence; in other words, sets of items which are alikein the way they pattern in the structure of items of higher rank. Thus,to take a typical instance from grammar, we may have morphemeclasses 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 wemight have word–classes defined by group structure, or clause–classesby 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 universalin linguistics; but it would probably be granted that some such categoryis necessary to linguistic description whatever name we choose to adoptfor 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.




96

“form” and “function”) is a central feature of linguistic method, andone which it seems appropriate to discuss in the present context.

The alternative to this use of the term “class” is to consider morpho-

logical 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 thatthey 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 formationout of morphemes. In this usage there are no morpheme classes, since“morpheme” is the name given to the smallest unit in grammar, whichby definition has no structure: its relation to items abstracted at other

levels, such as phonemes, is not one of structure, but involves theinterrelation of different dimensions of abstraction.It is important to notice that this is in the first instance a termino-

logical alternative, not necessarily implying a different theory. It is notthe case that the linguist has to choose between two different kinds of classification, the syntactic and the morphological; he has in fact torecognize both kinds of likeness. Moreover, the sets of items identifiedon these two criteria often coincide: we may recognize a syntactic class

“noun”, for example, defined as “that class of word which operates ashead 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 amorphological criterion such as “that type of word which is made upof a stem morpheme followed by a morpheme of case and a morphemeof number”.1 Indeed other things being equal, it is usually accepted asdesirable that the two should coincide: when the linguist is faced, ashe often is, with a choice between two descriptions, both theoreticallyvalid and both accounting for the facts, one in which the twoassignments coincide and one in which they do not, he will normally,and “intuitively”, choose the former. For example, groups in Englishsuch 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 syntacticclass 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 bothof these classes. Morphologically, however, it clearly resembles other nominal groups (the morning , this man, etc.) rather than other adverbialgroups (quickly, on the floor , etc.), and this can be allowed to determineits primary syntactic assignment.

There are, however, clear instances where syntactically defined setsdo not coincide with morphologically defined sets; and it would



c l as s in re la tion to c hain a nd c hoice

97

probably be generally agreed that, whatever the status accorded to thelatter, the former cannot be ignored. Syntactic likeness must beaccounted 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 becalled by the same name. It seems to me appropriate that the term

“class” should be reserved for the syntactic set (the morphological setmay then be referred to as a “type”), and I propose to adopt this usagehere. It is also true, in my opinion, that the class thus defined, thesyntactic 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 descriptiveconvenience whose theoretical implications are largely internal to itself.

In the remaining sections of this paper, therefore, I should like todiscuss two aspects of the syntactically defined set, which I shall refer to henceforward simply as the “class”. I shall be concerned with itstheoretical status in linguistic description, and shall confine myself tothe 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 chainrelation, such as the English clause structure “predicator + comple-ment” (for example fetch the ink). While (in this instance) the ultimateexponent of the element “predicator” is fetch3 and that of the element“complement” is the ink, the direct exponents of these elements arerespectively 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 areboy 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 bereferred 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 one-member classes. These do not really constitute a problem, being nodifferent, qua classes, from the others; but the tendency to refer tothem sometimes as category (by class name) and sometimes as item (bynaming the unique member) may obscure the fact that the exponence




98

relation (that is, the relation of item to category) is unaffected by thefact that class membership is limited to one item. For example, inEnglish the definite article the forms a one-member class; if we are

describing the particular choice, in English grammar, that is exemplifiedin the man and a man it does not matter whether we state the terms inthe system as the / a or as “definite article” / “indefinite article” (or

other more appropriate class name): it is the class that enters into therelation of choice. It may be noted in passing that the one-member class has a particular significance in linguistic theory: if grammar istaken to be that part of linguistic form in which choices are “closed”,

by contrast with lexis (vocabulary) in which they are “open”, then anyitem which can be shown to be the unique member of its class is fullyand unambiguously identified in the grammar.4 Thus the can be shownto be grammatically distinct from all other items in the Englishlanguage; whereas the grammar has no means of distinguishing, say,haddock from halibut , and this distinction must be accounted for in someform 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. Thatis to say, the sequence in which items occur may or may not be acrucial property of the structure in question. It is important to realize,however, that this “may or may not be” is something of an oversimpli-fication. To take an example, in the English clause John saw Maryyesterday the sequence is clearly crucial in one respect: John is Subject

and Mary Complement; whereas in the clause Mary saw John yesterdayMary 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 Maryyesterday and yesterday John saw Mary, we should be justified in sayingthat this particular feature of the sequence had no significance: it madeno difference whatever to the structure. This, however, is manifestlynot true: there is a difference in meaning, and although it does notseem 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. Thedifference between “John saw Mary yesterday” and “yesterday Johnsaw 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 languagewhere a difference in sequence makes no difference whatever to themeaning, and therefore does not need to be recognized as expounding




99

a distinct structure, though we should allow for such cases in thetheory. 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 distinctionbecomes so delicate that we do not know what to say about it; in suchcases 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 wecan 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 primarystructures with the value of the elements concerned. Thus the classcorresponding to predicator is “verbal group”, that to both subject andcomplement is “nominal group”, and that corresponding to adjunct is“adverbial group”. Where the sets of items operating as two or moreelements of structure show more than a certain degree of overlap, as in

the case of subject and complement – most items that can be subjectcan also be complement, and vice versa – these are conflated into a

single primary class: thus the “nominal group” is the primary classexpounding 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 primaryelements of structure. When we take the class further in delicacy,however, and recognize “secondary classes”, some of these moredelicate classes are chain classes and others are choice classes. It is clear that primary classes cannot be choice classes, since we cannot accountfor a choice until we have established that place in structure where thechoice is made: for example, the choice classes “singular nominalgroup” and “plural nominal group” are meaningful only in relation tothe primary (chain) class “nominal group” which defines the contextof the choice.

For examples of secondary classes of both types we may take anelement of structure of the English nominal group, the element“Determiner”. The class of word defined as operating with this valueis the class sometimes known as “deictic”. [Note: Halliday later reversedthese terms, using Deictic as the function (element of structure) anddeterminer as the class. (Ed.)] This includes some forty items that arefully grammatical (i.e. reducible to one-member classes by successivesteps in delicacy: the this that which whose my its a any some either each




100

other same certain, etc.), together with other items forming an open set(i.e. that cannot be so reduced: John’s, etc., including compound onesas in the railway company’s property).

This class of deictic may be variously subdivided along both axes. Onthe one hand, there are certain sets whose members can occur incombination, as in all my other friends; there are in fact three such

secondary groupings, the members occurring respectively in first, secondand third place in a maximum sequence. This gives three secondarychain classes which may be called “predeictic” (for example all), “deic-tic” (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 principaldimensions. The deictic, for example, may be “specific” / “non-specific” (my / every); “selective” / “non-selective” (my / the ); and, as afurther 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 subdivisionsand 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 Non-specific

this / thesethat / those

whichwhat

both allevery each

no neither

Selective my your our

their his / her

its John’s (etc.)

whose

Non-selectivethe a

some anyeither another

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




101

deictics. But the patterns they display are typical in their complexity: agiven class breaks down by simple subdivision into a system of moredelicate classes, but the same original class will also subdivide in a

number of different ways, so that many dimensions of classificationintersect with one another. Any given item, to be fully identified, mayrequire to be simultaneously classified on all such dimensions. In this

way it can be assigned to a “microclass”, this representing its value inrespect of all the properties which have been found relevant to the wayit patterns in the language. There will be, of course, a very largenumber of such microclasses: for example, in a computational study of

English “phrasal verbs” (items like take up, put on) which is beingcarried out at the moment, 557 such items were found to yield 125microclasses.

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 composedof 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 theclause structure “Subject + Predicator”, with classes respectively “nom-

inal group” and “verbal group”, as in “my friends have arrived ”; or itmay be only partially class-defining, where two or more elements areexpounded by the same class but differentiated in sequence. In thistype 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 toComplement.

This is not the only type of structure found in language, and thereseems no particular reason for assuming it to be the norm, especially inits 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 aprogression. It is doubtful whether one should set a theoretical limit tothe degree of depth in recursion; rather there appears to be somelogarithmic scale of diminishing frequency, so that the number of observations one would expect to have to make before recording adepth of, say, ten would be extremely high. Spoken English seems totolerate greater depth in recursion, or at least to tolerate it more readily,than written English; and this may be true of language generally. The




102

following are some examples of recursive structures in English, withdepth 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 stop him asking people to help him do

Pa Ca Pb Cb Pg Cg Pd Cd 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 doingpresent in future in past in future

4) Nominal group structure:

flue pipe support strap

d g b 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 maybe distinguished by class (as in example 1, above, where the alpha-element is expounded by the class “independent clause”), each elementof 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 atevery place. (The items themselves are of course not necessarilymutually substitutable in every instance, since there may be more




103

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 involvingrankshift are:

5) Nominal group structure and adverbial group structure, bothrankshifted (q = “qualifier” in nominal group; c = “prepositional

complement” in adverbial group; [ ] = boundary of rankshiftedgroup):

the peartree in the garden in front of the house near [qa [ca [qb [cb [qg

the bridge over the river

[cg [qd [cd]

8

6) Nominal group structure and clause structure, clause only rank-

shifted ([[ ]] = boundary of rankshifted clause):this is the farmer sowing his corn that kept the cock

[[qa1 [[qa2

that crew in the morn that waked the priest all shaven and shorn

[[qb [[qg

. . . that lay in the house that Jack built5

[[ql [[qm ]]

10

“Rankshift” is in fact merely a name for that type of recursive structurewhich cuts across the scale of rank. That is to say: in non-rankshiftedstructures, whether recursive or not, classes of each rank enter into astructure of the rank immediately above: in English, morpheme classesin word structure, word–classes in group structure, group classes inclause structure and clause–classes in sentence structure.6 In rankshift,

this relation is broken and the classes enter into a structure of their ownrank 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 secondshows 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 bea distinct class which occurs only with rankshifted status; but this is notnecessarily so. In example 5, the adverbial group in the garden isrankshifted to the status of Qualifier in the nominal group the peartree




104

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 mayoccur either rankshifted or not, often ambiguity may arise, as in take

that chair in the garden: is the chair already in the garden (= ‘take thatchair which is in the garden’; rankshift) or not (= ‘take that chair andput it in the garden’; no rankshift)?

Some of the most complex problems in the description of alanguage arise where the same structure combines place-ordered withrecursive elements. Perhaps the most striking example of this in Eng-lish is the nominal group, which is a troublesome mixture of the two

types; the earlier elements are largely place-ordered, recursive elementsbeing increasingly tolerated as one approaches the Head (the headbeing, e.g., houses in the two old stone houses by the river ) and continuing,by rankshift, thereafter. The element immediately following theDeterminer, which may be called Ordinative and defines the classesof cardinal and ordinal numeral, together with superlatives, is alreadymarginally 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 inthe language of headlines: holiday coach death crash inquiry verdict. More-

over, the linear succession of the items does not act as a constant inshowing the depth relation: compare 5-millimetre perspex boxes (gammabeta alpha) with 6-inch perspex boxes (beta beta alpha). The descriptionof the word–classes entering into the structure of the English nominalgroup is extremely complicated if one treats it as a simple place-ordered structure, with classes defined for each possible position, asthe various attempts to do so have already shown. On the other handit is unsatisfactory to treat the whole thing as recursive in structureand to recognize no secondary classes beyond the primary word–class“noun”. The facts of the language here lie in between the extremesof these two types of structure, and the best description seems to beone which takes this into account.

These seem to me to be some of the problems that arise in intra-

linguistic classification at the level of grammar. Such problems areprobably most acute at this level, but similar ones arise also at other levels, notably phonology, which has other additional classificationproblems of its own. Lexical classification is rather a different matter,and there are reasons for preferring not to use the term “class” intalking about lexis; but this subject would require a separate paper. Thefield of classification as a whole is one where linguists can learn muchfrom disciplines faced with similar or related problems. At the same




105

time perhaps the experience gained within linguistics may not bewholly 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 levelof 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 “prepositionalcomplement” 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”.



106

Chapter Four

SOME NOTES ON ‘D EEP’ GRAMMAR (1966)

In the representation of syntagmatic relations in language, we maydistinguish between a linear sequence of classes, such as “adjectivefollowed by noun”, and a non-linear configuration of functions, suchas “modifier-head relation” or simply “modification”.1 Both of these

have been referred to as “structure”, although this term has also beenextended 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 apurely relational approach to the language pattern” (1948: quoted inFirth 1951: 220); among others who have emphasized the relationalaspect of such studies are Firth2 (1951: 227–8; 1957: 17ff.; cf. Robins1953; Palmer 1964a), Tesniere (cf. Robins 1961: 81ff.) and Pike (cf.Longacre 1964: 16). Chomsky’s (1964: 32) distinction, using Hockett’sterms, 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 “aproper bracketing of the linear, temporally given sequence of elements,with the paired brackets labelled by category names”, while the deepstructure, which is “in general not identical with its surface structure”,

is “a much more abstract representation of grammatical relations andsyntactic organization”.A representation involving the concepts of class and sequence may

thus be said to be a representation of surface structure. Here theordering, 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. Firstpublished in Journal of Linguistics, 1966, 2(1), pp. 57–67.



s ome note s on ‘deep’ grammar

107

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 representationsand the much sparser bracketing of, for example, a tagmemic analysiscan adequately specify the relation of sequence in a surface structure.

The labelling attached to the entities specified as entering into thisrelation of sequence may then be “class”-type labelling and interpretedas 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 ashypotaxis, 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 determinablerelation. A considerable amount of bracketing may be introduced inorder to give as much information as is possible, within this limitation,about the syntactic relations involved. Class labels do not by themselvesreduce the bracketing required, since classes do not fully specify

syntactic function. Such labels may be conventionally interpreted asfunctional, 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 nodein 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 sequencealone are inadequate for the representation of syntagmatic relations inlanguage. Indeed the development of modern structuralism may beseen as having taken place in the context of a tradition in which it wasthe more surface elements that had remained least explicit. Relationalterms like “subject” and “predicate” have always co-existed with classnames such as “noun” and “verb”; and the definition of the classes hasrested at least in part on syntactic criteria, so that the designation of an

item by its class name indicates something of its potentiality of syntacticfunction. Classes were not thought of as specifying actual syntacticfunction within a given sentence, since the theory also recognized thedeeper syntactic relations into which the classes entered; the attempt tocombine morphological with syntactic criteria in the definition of theclasses (since morphological “types” have to be accounted for some-where), while it may lead to difficulties, is entirely explicable withinsuch a framework.




108

While the underlying syntagmatic relations have been recognized asnon-linear, or at least as not manifested in the linear sequence of thelinguistic items, their representation, as Palmer (1964a: 125) points out,

has usually involved some form of linear notation. Since there is also alevel 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 areinherently surface concepts, specifying the items of the language andtheir 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 terminologicalsimplicity we might perhaps here follow one tradition in referring toan arrangement of classes in sequence as a syntagm, reserving the termstructure for a configuration of functions. If then function-type labelssuch as “modifier” are introduced, whether as such or as conventionalinterpretations of class-type labels, they will not be located in thesyntagm, since their defining environment is not stated in terms of (its)sequence. This holds true even if in a given language (say) a modifier-

head structure is always realized as a syntagm of adjective followed bynoun; a structure is not defined by its realizations.

The ordering that is ascribed to structure may be thought of independency terms, or in constituency terms as an underlying sequencewhich does not (necessarily) correspond to syntagmic sequence, or asmere co-occurrence or absence of ordering.3 In all cases it is of adifferent nature from syntagmic sequence, in that the components arefunctions, not sets of items. If (with Lamb) we use to representconfiguration, this being interpreted as “unordered with respect tosyntagmic sequence, whether or not any other form of ordering isconsidered to be present”, then a structural representation may take theform m:h, or interchangeably h:m (modifier–head); this contrasts with

a syntagmic representation of the form adj^n (adjective followed bynoun). Representation such as m^h and adj:n would then appear asmixed types, where deep (structural) labels are attached to surface

(syntagmic) relations, or vice versa. These might be given conventionalinterpretations, perhaps for example m^h as “modifier–head structurewith realization by sequence alone” (i.e. where modifier and head arerealized by the same class), adj:n as “modifier–head structure withrealization by class alone” (i.e. where the classes may occur in either sequence); but these would be merely a shorthand for combining twotypes of representation.

While many other formulations are possible, the recognition in some




109

form or other of two distinct types of representation, linked by someform 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 bothrelation and correlation; relation is syntagmatic, within the semioticprocess, or the text, while correlation is paradigmatic, within thesemiotic 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 paradigmaticrelations as relations of system.

Provided there is at least some syntactic element in the definition of the classes concerned, even a syntagmic, class-sequence representationalready gives some information about the paradigmatic relations intowhich an item enters, but to a very limited extent. This limitation is

again inherent in the nature of the class–sequence concept: paradig-matic “relatedness” depends on a functionally defined environment.Two entities can only be said to contrast if they have a functionalenvironment in common,5 and this environment is generally specifiedin terms of syntagmatic function; it presupposes therefore a representa-tion of structure – i.e., of “deep” syntagmatic relations. The structuralrepresentation thus specifies the environment both for sets of paradig-matic relations and for further networks of syntagmatic relations, thosewithin the lower-order constituents: for example the function of “subject”, itself specified syntagmatically in clause structure, defines anenvironment both for the syntagmatic relation of modifier–head andfor the paradigmatic relation of singular / plural. For paradigmaticrelations in the highest unit there is no functionally defined environ-

ment in this sense, so that if the sentence is said to be either declarativeor interrogative this either / or relation rests on other grounds: the

sentence as a primitive term, or some postulated higher unit not yetstructurally described; the appeal to contextually-defined sentencefunctions such as “statement” and “question” is not one of these, thisbeing rather a way of saying that declarative and interrogative have noenvironment in common.

The paradigmatic contrasts associated with a given, defined environ-ment may be thought of as being accounted for either in a singlerepresentation of “deep” grammar, in which are incorporated both




110

syntagmatic and paradigmatic function, or in a separate form of statement, distinct from, but related via the specification of the environ-ment to, the statement of syntagmatic relations. Firth’s concept of the

system embodies the second approach. The system may be glossedinformally as a “deep paradigm”, a paradigm dependent on functionalenvironment; in a sense, and mutatis mutandis, the relation of system to

paradigm is analogous to that of structure to syntagm as these termswere used above. One could think of “paradigmeme” as a possibletagmemic term. In Hjelmslev the “system”, likewise a paradigmaticconcept, is defined as a “correlational hierarchy”, the underlying notion

being that of commutation (Hjelmslev 1961: 73–4). A system is thus arepresentation of relations on the paradigmatic axis, a set of featurescontrastive in a given environment. Function in the system is definedby the total configuration: for example “past” by reference to “present”and “future” in a three-term tense system, as structural function isdefined 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 shouldcontain both a structural and a systemic component. It may be useful

therefore to consider the notion of a systemic description as one formof representation of a linguistic item, the assumption being that itcomplements but does not replace its structural description. Thesystemic 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 oneor 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 along time in the form of “this clause is interrogative, finite, presenttense, . . .”, given that we are told somewhere in the grammar notmerely what other tenses, moods, etc. are found in the language butalso which of them could have occurred in this particular clause allother features being kept constant.

There is, however, one modification of a traditional “systemic

description” of this kind which may need to be considered. Thisconcerns the ordering of the features listed. In the traditional versionthey are unordered; but if the grammar specifies not only relevantsystems 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 afeature in the other in a systemic description, may be hierarchical or




111

simultaneous; if two systems are hierarchically ordered, features assignedto these systems are ordered likewise. So for example the system whoseterms 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 theitem John has seen the play may be represented in respect of thesefeatures as:

(indicative : (declarative / unpredicated theme))

where : indicates hierarchy and / simultaneity. Then it’s John who hasseen 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 thepossibilities recognized by the grammar. As far as these examples areconcerned, the grammar would show that in a given environmentselection is made between indicative and imperative; and if indicativeis 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 itemthus contrasts with others in respect of such features and combinationsof features as the ordering of the system permits.

For any set of systems associated with a given environment it ispossible to construct a system network in which each system, other than those simultaneous at the point of origin, is hierarchically orderedwith respect to at least one other system. The point of origin isspecified syntagmatically, so that all features are associated with asyntagmatic environment; at the same time the system network pro-

vides a paradigmatic environment for each one of the features, specify-ing 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 descrip-

tion may be thought of as complementary to structural description, theone concerned with paradigmatic and the other with syntagmaticrelations. On the other hand it might be useful to consider somepossible consequences of regarding systemic description as the under-




112

lying form of representation, if it turned out that the structuraldescription could be shown to be derivable from it. In that casestructure would be fully predictable, and the form of a structural

representation could be considered in the light of this. It goes withoutsaying that the concept of an explicit grammar implied by this formu-lation derives primarily from the work of Chomsky, and that steps

taken in this direction on the basis of any grammatical notions aremade possible by his fundamental contribution. My own more specificdebt here is to Lamb, whose formalization of stratification theory isbased on general notions closely akin to those which I had adopted (cf.

Hockett 1965: 198). The present paper, however, attempts no morethan an informal discussion of the question of a grammatical descriptionin terms of features, here based on the notion of a feature as one of aset of contrastive “terms in system”.

Presenting the systemic description of a linguistic item as theunderlying grammatical representation of that item would seem toimply that its paradigmatic relation to other items of the language wasin some way its more fundamental property, from which its internal

(syntagmatic) structure is considered to be derived. This would seemto be Hjelmslev’s view, in his discussion of system and process (1961:

39–40). But the priority which is implied is not one between paradig-matic and syntagmatic relations as such, but rather between the externalrelations, both paradigmatic and syntagmatic, into which an item enters(the point of origin of a system network being defined in syntagmaticfunction) 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 isbeing considered therefore is that that part of the grammar which is asit 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 operatingin the environment of other contrasts. Structure would then appear as

the realization of complexes of systemic features, involving in placesboth neutralization and diversification as defined in Lamb’s terms(Lamb 1964a: 64).

If the structural representation is not required to account for paradig-matic relations, the question of how “deep” it needs to be is determin-able by reference to other considerations: it should give an adequateaccount of syntagmatic relations, and permit the explicit realization of the systemic description in terms, ultimately, of a sequence of classes.




113

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 wouldseem 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 simplestsequence of classes in the syntagm, although it could be shown to be

unsatisfactory even on class-distributional grounds. (2) is distributionallyacceptable and would account adequately for the syntagmatic relations;but it fails to account for the paradigmatic relations in that it does notshow the “relatedness” of this clause to John has seen the play, etc. If thestructural description is required to show the paradigmatic as well asthe syntagmatic relations of the grammar we need some representation

such as (3) in which John is the Subject. This leads to complexity inthe realization, since a nominalization of the form it’s John who seems

to add no new insight elsewhere in the grammar. A more seriousdifficulty arises in relation to the element “Subject” in English, whichis a complex element within which it is possible to distinguish threecomponents, or features; each of these may contrast independently of the other two, although there is a general, and generalizable, tendencyto 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 Subjectnonthematic); (iv) John has seen the play (with John tonic; Subject as“given” versus Subject as “new”). Each of these three is related paradig-

matically to the original item, and each of them contrasts with it inrespect of one feature only. By a further contrast, that of “unpredicatedtheme” 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 constituentstructure, (v) differs from (iv) in respect of only one feature. Suchpatterns, 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 structuraldescription; if they were handled in systemic terms, the structure needrepresent only their realization in syntagmatic relations. We could thenadopt a form of structural representation such as (2) above.




114

The examples cited might be regarded as irrelevant on the groundsthat they do not involve cognitive distinctions and therefore belong tothe 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 withinthe domain of a semantic theory, where the selection of a particular

variable, such as paraphrase, as a basis for the classification of distinctionsis not arbitrary as it seems to be in the grammar. This is not to denythat the speakers of a language recognize some distinctions as “moreimportant” than others, and that this may depend at least in part on a

concept of paraphrase. The hierarchization of systems in delicacy, in asystem network, does seem to reflect some notion of the relativeimportance of the systems involved; this is an instance of the conver-gence of semantic and distributional criteria referred to by Lyons in hisimportant discussion of semantics and grammar (1963: Chapter 2).Even if a clear answer can always be given to the question “is a aparaphrase 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 placeof a given distinction in the grammar would, as I see it, depend on its

environment in terms of other distinctions, this presupposing also itssyntagmatic environment, rather than on a classification of its semanticfunction.

To return to the discussion, another relevant factor here would bethe desire to incorporate into the grammar phonological realizations of grammatical features, particularly (in English) those of intonation andrhythm. Such features may be assigned a place in a syntagmaticrepresentation, either as superfixes in a syntagm or as elements in astructure; but the assignment of, say, a pitch contour as a constituentto a specified place in a structural representation, while it may benecessitated by the realization requirements, seems in other respectsrather arbitrary. Intonation, in English, provides instances of bothneutralization and diversification; one and the same feature may be

realized in some environments by a structural pattern and in others byintonation, and a given intonation pattern may realize different featuresin different environments. In other words, intonation is not predictablefrom its structural environment. It can, however, be shown to bepredictable in the grammar if it is regarded as a form of the realizationof 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




115

general point: namely that if a representation in terms other than of constituent structure is adopted for the statement of paradigmaticrelations, and is then made to determine the constituent structure, then

provided the structural description adequately handles the syntagmaticrelations there is no need for everything to be accounted for at aconstituent stage of representation. This is most obviously relevant to

phonological features of the prosodic type, but could be extended alsoto 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 systemicnetwork. A putative feature which could not be shown to contrastindependently with one or more others at some point would not be adistinct feature; each feature that is recognized is thus a term in asystem, which system is located in hierarchical and simultaneousrelation to other systems. The location is “polysystemic”: the recog-nition of a system, and the assignment of a feature to it, depends onthe potentiality of contrast in the stated environment. For example,

there might seem to be a proportionality in English such that he can gois 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 twosyntagmatic environments determine different sets of paradigmaticrelations. The ordering of the systems in delicacy would thus beimportant 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 bedone in terms of the notion of rank, where the initial identificationand labelling of certain stages in a constituent hierarchy in such generalterms would provide a starting point for the delimitation of differentmore 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 theassignment of a system to a place determined solely by constituent

status (for example all clauses) and allows further specification of theenvironment to be in terms of features: a feature x may be associatedwith constituents having the features y and z rather than with constitu-ents having a given syntagmatic function. The possibility of contrastbetween 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 sub-strata, within the outer grammatical stratum, with each rank character-




116

ized by a different network of systems. While many, though by nomeans all, features would be present at more than one rank, withconsequent preselection at certain points, an important distinction is to

be made here between preselection, where the choice of a feature atone rank determines the choice of a feature at a lower rank but thetwo 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 thestructure of the word of a choice, such as that of number, associatedwith the group. There seems no reason to assume a necessary relation-

ship between the rank at which a feature has its environment and thatin 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. Whileclearly a constituency notion, reflecting here the speaker’s awareness of the hierarchical organization of linguistic items, it imposes a minimumof bracketing and in this way facilitates the interrelating of paradigmaticand 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 givenrank as its most generalized functional environment. It is not impliedthat a description in terms of features would necessitate a rank-typeconstituent structure, but rather that the status of constituents in thegrammar would need to be brought into the discussion.

Palmer (1964: 130) wrote: “Perhaps we need a pre-grammaticalstatement in which order is utterly divorced from sequence”. In thispaper I am following up Palmer’s conclusion by asking whether such astatement could be thought of as a representation of the “deep”grammar. If deep grammar is equated with deep structure, in the senseof being thought of as relations of the constituency type, it may bedifficult to avoid connotations of sequence and to solve some of theproblems Palmer raised. If the underlying “order” is thought of as

systemic, the more abstract representation of grammatical relationscarries no implication of sequence. Sequence can be stated by referenceto these, a “linguistic element whose exponent is sequence” having astatus no different from that of others. Such a description is in a senseof the WP type, with word replaced by unit, or constituent, andparadigm by system. It is not suggested that paradigmatic relations aresomehow “more important” than syntagmatic ones; but merely that adescription in terms of features, if it can be made explicit, may help in




117

bringing the “unidimensional time sequence” of language into relationwith 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 misinter-

preted; 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” preciselyfor 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 givenset 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.



118

Chapter Five

THE CONCEPT OF R ANK: A R EPLY (1966)

I am most grateful to Peter Matthews (who I am pleased to note seemsto deplore, as I do, “missionary fervour” in linguistics1) for allowingme the opportunity of replying in the same issue to his criticisms of theconcept 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’ strictureson 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 toterminology is in no way different from that of other linguists. YetMatthews is well aware that I could cite numerous examples of theredefinition 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 oneanswer a charge such as ‘If Halliday does not mean by “word”, inparticular, what the ordinary linguist on the Clapham omnibus meansby “word”, why has he never said so?’ other than by pointing out,

what is surely obvious, that linguists do use terms in different ways andthat 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.



t he c on c ep t o f r an k: a r e pl y

119

that is my fault rather than his. I plead guilty unreservedly to the chargeof inadequate documentation of my own work, although the relevantbibliography 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 totake up with reasonable brevity some of the issues as Matthews himself sees them.

Two assertions seem crucial to Matthews’ argument. (1) A rank-freeconstituency grammar is to be preferred to a rank constituency gram-mar. (2) All formulations of rank grammars are either theoreticallyinsignificant or empirically unsound: unsound if they differ materiallyfrom rank-free grammars and insignificant if they do not. (1) followsfrom (2), except that, if they do not differ materially, this by itself givesus 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 anyconstituent, and any constitute, can be assigned to one or other of the

specified layers, or ranks. The European linguistic tradition, by its useof terms such as “sentence” and “clause”, has always implied thepossibility of such a grammar, although, as is well known, the absenceof criteria regulating the necessary modifications of the simple con-stituency relation led to various difficulties. I return below to theconcept of rank as suggested in such formulations as “clause used as aword” and the like.

A rank grammar is, as Matthews observes, a hypothesis about thenature 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 manyother hypotheses, both in linguistics and in other disciplines, itsempirical falsification, given that it cannot be logically disproved, isunlikely 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 rankconstituent terms. It would be likely to take the form rather of demonstrating that the hypothesis prompts no interesting new questionsand leads to unnecessarily complex or otherwise unsatisfactory accountsof the facts. Neither this limitation on the conditions of its falsificationnor 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




120

which is capable of being falsified in these terms – that there are, inother words, conditions under which I would regard it as having beenshown to be inadequate; and that such conditions do not obtain.

We must then ask whether it is worth making in the first place, andbasically there are two grounds for thinking that it is: its descriptiveadvantages, 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 thestructures and systems themselves, becomes an important step in

generalization. To show that a system operates at a given rank is thefirst step in stating its relationship to other systems; likewise to assignan item to a given rank is the first step in stating the systemic andstructural relations into which it may enter and those which it mayembody within itself. On the structure axis, rank is a form of generali-zation about bracketing, and makes it easier to avoid the imposition of unnecessary structure. It also serves frequently to distinguish betweensimilar structures, for example between defining and non-defining

relative clauses in English. It may contribute towards a significantmeasure 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 toa 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 itis of interest)? Are certain paradigmatic or syntagmatic relations univer-sally associated with specific ranks? Is there any statistical associationamong the relative frequencies of items of different rank? Is there anytype of aphasia characterized by progressive, rank-by-rank loss of grammatical structures? Is there any reason why different languageshave institutionalized different grammatical units in their orthographiesor the same unit in different ways?

Since Matthews considers the rank hypothesis to be of no signifi-cance unless accompanied by a requirement of total accountability(chain-exhaustive assignment to constituents) at each rank, we shouldperhaps ask what the alternative hypothesis is. A rank-free grammar isalso a hypothesis about the nature or language, one which would seemto hold that the number of pairs of brackets in a given constituenthierarchy would enable us to make no predictions about the syntag-matic and paradigmatic relations to be encountered in association with




121

a given pair of brackets. In other words, in a labelled constituentstructure there will be no necessary association between the labels andthe distance, in number of nodes, from the top of the tree, although it

would still be possible to predict that certain syntagmatic and paradig-matic 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 stageswhere new paradigmatic and / or syntagmatic relations can be intro-duced; they must be new, since for the mere repetition of existingrelations grammars of both kinds require some form of recursive

mechanism. The rank-free constituency hypothesis is thus harder tofalsify; to me it seems less satisfactory, but all I wish to suggest here isthat some form of fixed constituency hypothesis about language doesseem 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 thisrequirement (sc. of total accountability at each rank) by permitting bothdownward and upward rankshift, the concept of rank is at once stripped

of its theoretical significance”. Downward rankshift does not affecttotal accountability, and is in any case merely a formulation in rank-

grammatical terms of something that is explicitly provided for in alllabelled grammars. It is therefore upward rankshift; which we mustconsider. It may be worth noting here the contrast between thetagmemic 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 alternativepossibility 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 thatthere are “no such things” (to put it crudely) as words, phrases andclauses. On the contrary, . . . these constructs are useful’ (p. 102).Matthews’ reference to “the sense which practising linguists haveadopted (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. Whatthis 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 isthat it uses formulations like that mentioned above, “a clause used as a




122

word”, which suggest a clear understanding that constituents of acertain type have as it were an “unmarked” place of operation in thelanguage relative to constituents of other types. In other words it may

happen that, while the “normal” rank-value of the item operating as aconstituent at a given place in structure is rank x, we also find instanceswhere an item of rank y occurs instead. This intuition, though

admittedly inadequate as formulated, seems to me basically sound; thequestions are how it can be made explicit, and in particular, at thispoint 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 morphologicaltypes but have certain characteristic functions associated with them. Incases of (downward) rankshift, an item normally having the function of (entering into the paradigmatic and syntagmatic relations associatedwith) rank x characteristically “loses” these functions on taking over those of rank y: a clause operating in group structure cannot enter intodirect syntagmatic relations with clauses outside the structure of thatgroup. There are good reasons, in other words, for saying that the

relevant functional environment for who came to dinner , in the man whocame to dinner , is that of group structure; hence the traditional label

‘clause used as a word’. I do not see any reason, however, for sayingthat therefore it ‘is’ a word; I did not know that this had beensuggested, although it would not be so far from traditional usage asMatthews seems to imply.2 Since, however, Matthews is not question-ing the notion of (downward) rankshift as such, but merely objectingto a transfer of rank labels which has not in fact been proposed, we canignore this as common to all grammars and concentrate on the questionof 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 onerank. The two cases are not parallel. While the consistency of theformulation “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 alsosometimes 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 thetendency in constituency grammars to introduce singulary branching;and total accountability in a rank grammar is one way of determiningthe amount of singulary branching that is required.




123

Of course total accountability of constituents at all ranks is open toargument, like many other matters in linguistics; one would expect itto be taken for granted that such things were extensively discussed,

although Matthews’ crusading zeal on the part of those he considersless able to look after themselves seems to have led him to think it hasnever been questioned. As a hypothesis it has certain things to recom-

mend it. Matthews’ formulation “extra turns round the mulberry bush”suggests that he finds it counter-intuitive; if so, this shows that we havedifferent intuitions, since to me the notion that an item can be aconstituent at more than one defined layer is, while not new, highly

illuminating. Indeed it seems to me to be one of the requirements of aconstituency grammar that it should allow some singulary branching of constituents. But once singulary branching is allowed, the grammar isno 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 leastsome conditions under which singulary branching occurs (i.e. upwardrankshift 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 clausestructure 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 besaid about such items at more than one rank. If for example yes cannot“be” a clause, then the range of intonation patterns from which itselects in the sentence Yes! , which is that characteristic of clauses,and, moreover, of clauses of a particular class, already specified, towhich it can be shown to belong on other grounds, has to be statedover 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. Itis much simpler to let rank scale define the environment. The reasonfor recognizing singulary branching is basically the same as that for recognizing a scale of rank, which in principle is what singularybranching 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 morecomplicated. (This is not to say that, in the examples cited byMatthews, and must be a clause, which I did not know to have beensuggested or implied.) Multiple rank assignment, in fact, is merely aformulation in rank-grammatical terms of the notion that a constitutemay have only one constituent; if systems and structures are stated for




124

each rank, constituents being assigned on this basis to the rankappropriate to them, the multiple assignment of constituents is not onlysimpler than the restatement of relations but also avoids making the

relatively surface notion of constituency as the basis of grammaticalorganization.

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 elementof) group structure, in the case of proposed “upward rankshift” the

item may be entering simultaneously into more than one set of relationsand it is this that is brought out by its assignment of an item to morethan one rank. This seems no more absurd than the assignment of anitem to more one class, rather than adjusting classes so that each itemfigures only once or denying the relevance of classes altogether. Thereis an analogy in the orthographic hierarchy. In the orthographicsentence I . it does not seem strange to say that this is a sentenceconsisting of one orthographic word and that this word consists of one

letter. (Note that I have to specify “orthographic word” here, becauseof the different ways in which the term “word” has traditionally been

used.) If we say that the sentence consists of one letter, complicationsarise: we have to restate the structure of the orthographic sentence interms of letters as well as in terms of orthographic words, thus requiringamong other things a second and much more complex statement of the distribution of punctuation marks; the form I will not appear at allin the set of orthographic words, but only in that of letters, of whichwe must then define a subset consisting of those that can operate in thestructure of the sentence; and so on.

How far Matthews is objecting to all singulary branching I do notknow; he may want to say that a word can consist of one morphemebut 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 thatof (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 differentfrom the question whether of the girl I saw last night is assigned to therank of word (which for me it is not). The important point is notwhether the grammar will make it possible for after to be assigned tothe rank of group, since some degree of singulary branching is usuallyconsidered desirable, but whether it will require this assignment; and if it is not required, then under what conditions it will and will not takeplace. To require it, by demanding total accountability at all ranks, has




125

the merit of underlying simplicity and, for some people at least,intuitive adequacy, but at the cost of some surface complexity; this issomething to be avoided if possible, and Huddleston (1965) has

introduced an important modification into the definition of rank whichgoes 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 constituentstatus at all; it is quite usual not to recognize intonation features asconstituents, and the same considerations could apply, as Matthewspoints out, provided limitations were stated, to markers such as and and

or . I do not know how to specify in a general formation the conditionsunder which accountability in constituent terms would not be required.But this problem is no more difficult for a rank grammar, which has atleast 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 bemade, once and for all, by specifying a single element of structure atwhich all the relevant classes or systems may operate: it does not matter

if some of these classes are classes of phrases, words or morphemesrather than clauses”, is not a solution, since in many cases it does

matter; otherwise no constituency grammar would have introducedsingulary branching. Matthews has shown no convincing reasons for abandoning the generalizations which the notion of rank facilitates; tothe 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 befailing to exploit the concept of rank it would not render it insignifi-cant. It would remain a form of labelling of greater generality thanstructure labelling; and it would embody the hypothesis that there area finite number of points of origin of distinct systemic choices inlanguage and a finite number of layers of constituent structure at whichdistinct 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 isimplied. I am not sure whether Matthews wishes to suggest, as at onepoint he appears to do, that there is no problem here except the one heascribes to me; to many linguists it has seemed that there is. It is beyondthe scope of a discussion of this nature to attempt to do it justice, and Ido not want to argue here in favour of any one approach; but a brief comment on Matthews’ own observations seems to be called for.




126

It is perfectly true that the fact that something is “in some sense” aword does not make it an immediate constituent of anything. But it issurprising 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 premisethe requirement that “each unit should be fully identifiable in descrip-tion”’. 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 restedon such grounds; in any case we may reasonably ask for more than themere 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 thegrammar take account of the fact that the placement of the ’s isparalleled by the placement of the tonic (in the king of England) whichlikewise depends on location in the syntagm and not on the constituentstructure? and so on. If Matthews is suggesting that the problem iscreated, or even that it is complicated, by the introduction of rank intothe 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 notmake 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 ownlinguistic 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 termino-

logical 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.



127

APPENDIX TO S ECTION ONE

TABLE 1

Tonality: Distribution of utterance into tone groups (location of tonegroup boundaries)

Tonicity: Distribution of tone group into tonic and pretonic (locationof tonic foot)

Tone (primary; pitch movement on tonic)

Tone (secondary)

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




128

Symbols

//

/

tone group boundary (always

also foot boundary)foot boundary

—

.

. . .

tonic syllable

silent ictuspause

Tone, primary and secondary, is shown by Arabic figures, alone or

with diacritics, placed immediately after the tone group boundarymarker.



appendix

129

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 clausesreservation: 1 unreserved, 4 reserved

involvement: 1 neutral, 3 uninvolved, 5 involvedagreement: 1 neutral, 3 confirmatory, 2 contradictoryinformation: 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) mildrelation to previous utterance: 1 unrelated, 2 (with WH-tonic) echo

Interrogative clauses, yes/no type‘key’: 2 neutral, 1 stronginvolvement: 2 neutral, 3 uninvolved, 5 involvedplace in alternative question: 2 first alternative, 1 second alternativespecification 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 mildforce: 1 neutral, 4 compromising, 5 insistentfunction: 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 representingthe terms in the choice with their appropriate tone. Thus ‘1 unreserved,4 reserved’ means ‘in this choice tone 4 indicates reservation, bycontrast with one 1 which indicates no reservation’. Secondary tonesare indicated where relevant.



1 Systems of tone



appendix

131

2 Systems of the clause

2.1 Dependent clause




132

2.2 Declarative clause



appendix

133

2.3 Interrogative 1




134

2.4 Interrogative 2






136

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 dogextensive: (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: ( 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



appendix

137

2.7 Theme

Examples of tone-expounded systems in declarative clause (from spon-taneous 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/nor mously / 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 behindthe / 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 //




138

//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

Examples1 2 3 4 5 6 7 8 9 10

//1 who said / that // 0 0 1 - 0 - 1 - 0 -

//2 who said / that // 0 0 1 - 1 0 1 - 0 -

//1 who was it / said that // 0 0 1 - 0 - 1 - 1 0

//2 who was it / said that // 0 0 1 - 1 0 1 - 1 0

//2 it was / who said / that // 0 0 1 - 1 0 1 - 1 1

//1 who did you / see // 0 0 1 - 0 - 0 0 0 -

//2 who did you / see // 0 0 1 - 1 1 0 0 0 -



appendix

139

//2 who was it you / saw // 0 0 1 - 1 0 0 0 1 0

//2 where did he / go // 0 0 1 - 1 0 0 1 0 -

//2 where was it he / went // 0 0 1 - 1 1 0 1 1 0

//1 who said / that // 0 0 0 0 - - 0 1 0 -

//2 who said / that // 0 0 0 1 - 0 1 - 0 -

//1 who was it / said / that // 0 0 0 0 - - 1 - 1 0

//2 who was it / said / that // 0 0 0 1 - 1 1 - 1 0

//1 who did you / see // 0 0 0 0 - - 1 - 0 -

//1 who was it you / saw // 0 0 0 0 - - 0 0 1 0

//2 where did he / go // 0 0 0 1 - 1 0 1 0 -

//1 where / was it he / went // 0 0 0 0 - - 0 1 1 0//1 who wants / what // 0 1 (1) 0 - - 1 - 0 -

//2 who’s / going to sit / where // 0 1 (1) 1 - 0 1 - 0 -

//1 which shall I / put / where // 0 1 (1) 0 - - 0 - 0 -

//1 John said / what // 1 0 (1) - 0 - (0) 0 0 -

//2 John said / what // 1 0 (1) - 1 0 (0) 0 0 -

//1 put it / where // 1 0 (1) - 0 - (0) 1 0 -

//2 John put / what there // 1 0 (1) - 1 1 (0) 0 0 -

2.9 Second order WH clause systems

Key to examples

Examples1 2 3 4 5 6

//2 . did / who do it // 0 1 0 0 - 0

//1 . did / who do it // 0 1 1 0 - -

//2 did he do / what // 0 0 0 0 - 1//1 did he go / where // 0 0 1 0 - -

//1 who said / what // 1 - - 0 0 -

//1 who was it / said / what // 1 - - 1 0 -

//1 what did / who say // 1 - - 0 0 -

//2 where did / who go // 1 - - 0 1 1

//2 who went / where // 1 - - 0 1 0

//2 where was it / who went // 1 - - 1 1 0




140

3 Imperative systems: jussive

Key to examples

Examples1 2 3 4 5 6 7 8 9

//1 ask / John // 0 - 0 0 0 0 - 0 0

//1 you ask / John // 0 - 0 0 0 1 - 0 0

//1 you ask / John // 1 0 0 0 0 1 - 0 0

//1 let’s ask / John // 0 - 0 0 1 - - 0 0

//1 let’s ask / John // 1 0 0 0 1 - - 0 0

//1 do ask / John // 0 - 0 1 0 0 - 0 0

//13 do ask / John // 1 0 0 1 0 0 - 0 0

//1 you / do ask / John // 1 0 0 1 0 1 0 0 0

//1 do let’s ask / John // 0 - 0 1 1 - - 0 0//13 do let’s ask / John // 1 0 0 1 1 - 0 0 0

//3 don’t ask / John // 0 - 1 - 0 0 - 0 0

//13 don’t ask / John // 1 0 1 - 0 0 - 0 0

//4 don’t / you ask / John // 1 0 1 - 0 1 1 1 0

//1 don’t / let’s ask / John // 1 0 1 - 1 - 1 0 0

// 1 let’s / not ask / John // 1 0 1 - 1 - 0 0 0

//13 don’t let’s ask / John // 1 0 1 - 1 - 0 0 0

//13 let’s not ask / John // 1 0 1 - 1 - 1 0 0//1 you / ask / John // 1 1 0 0 0 1 - 0 0

//5 don’t let’s / ask John // 1 1 1 - 1 - - 1 0

//53 . let’s / ask / John // 0 - 0 0 1 - - - 1

//1 you ask //4 John - - 0 0 0 1 - - 1



appendix

141

4 Theme systems: examples

Key to examples

Examples1 2 3 4 5 6 7 8 9

//1 John / saw the / play / yesterday // 0 - 0 0 - 0 0 0 00


//4 John //1 saw the / play / yesterday // 0 - 0 0 - 0 0 0 10



//13 John / saw it / yesterday the / play // 0 - 0 0 - 0 1 0 01

//13 . he / saw the / play / yesterday / John // 0 - 0 0 - 1 0 0 01

//4 . it / wasn’t / John that / saw the / play /

yesterday // 0 - 0 1 1 0 0 1 00

//13 . it was / John that / saw the / play /

yesterday // 0 - 0 1 0 0 0 1 01

//4 . it / wasn’t the / play John / saw /

yesterday // 1 1 0 1 1 0 0 1 00

//13 . it was the / play John / saw / yesterday // 1 1 0 1 0 0 0 1 01

//4 John he //1 saw the / play / yesterday // 0 - 1 0 - 0 0 0 10

//1 yesterday / John / saw the / play // 1 0 0 0 - 0 0 0 00

//4 yesterday //1 John / saw the / play // 1 0 0 0 - 0 0 0 10

//1 . the / play / John saw / yesterday // 1 1 0 0 - 0 0 0 00

//4 . the / play //1 John saw / yesterday // 1 1 0 0 - 0 0 0 10

//4 . it / wasn’t / yesterday John / saw the /

play // 1 0 0 1 1 0 0 1 00

Examples of theme and information systems (from spontaneousconversation)

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




142

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 //



appendix

143

5 Systems of the verbal group 1

6 Systems of the verbal group 2




144

7 Verbal group

7.1 Verbal group 1

te td tg tb ta past 1

present 2

future 3

past in past 4

present 5

future 6present in past 7

present 8

future 9

future in past 10

present 11

future 12

past in future in past 13



appendix

145

present 14

future 15

present in past in past 16

present 17future 18

present in future in past 19

present 20

future 21

future in past in past 22

present 23

future 24

past in future in past in past 25present 26

future 27

present in past in future in past 28

present 29

future 30

present in future in past in past 31

present 32

future 33present in past in future in past in past 34

present 35

future 36

7.2 Verbal group 2

Tense (major system)

1 took/did take

2 takes/does take

3 will take

4 had taken

5 has taken

6 will have taken

7 was taking

8 is taking9 will be taking

10 was taking

11 is going to take

12 will be going to have taken

13 was going to have taken

14 is going to have taken

15 will be going to have taken




146

16 had been taking

17 has been taking

18 will have been taking

19 was going to be taking20 is going to be taking

21 will be going to be taking

22 had been going to take

23 has been going to take

24 will have been going to take

25 had been going to have taken

26 has been going to have taken

27 will have been going to have taken28 was going to have been taking

29 is going to have been taking

30 will be going to have been taking

31 had been going to be taking

32 has been going to be taking

33 will have been going to be taking

34 had been going to have been taking

35 has been going to have been taking36 will have been going to have been taking

7.3 Verbal group 3

Tense (minor system: modal / non-finite)

I to take, taking; can take (= 2)

II to have taken, having taken; can have taken (= 1, 4, 5)III to be taking, being taking;* can be taking (8 =)

IV to be, being; can be + going / about to take (= 3, 11, 12)

V to be, being; can be + going / about to have taken (= 6, 14, 15)

VI to have, having; can have + been taking (= 7, 16, 17)

VII to be, being; can be + going / about to be taking (= 9, 20, 21)

VIII to have, having; can have + been going / about to take (=10, 12, 23)

IX to have, having; can have + been going / about to have taken (=13, 25, 26)

X to be, being; can be + going / about to have been taking (= 18, 29, 30)XI to have, having; can have + been going / about to be taking (= 19, 31, 32)

XII 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.



appendix

147

7.4 Verbal group 4

Tense (minor system: sequent)

1 had taken

2 took

3 would like

4 (1)

5 (1)

6 would have taken

7 had been taking

8 was taking9 would be taking

10 had been going to take

11 was going to take

12 would be going to take

13 had been going to have taken

14 was going to have taken

15 would be going to have taken

16 (7)

17 (7)

18 would have been taking

19 had been going to be taking

20 was going to be taking

21 would be going to be taking

22 (10)

23 (10)24 would have been going to take

25 (13)

26 (13)

27 would have been going to have taken

28 had been going to have been taking

29 was going to have been taking

30 would be going to have been taking

31 (19)32 (19)

33 would have been going to be taking

34 (28)

35 (28)

36 would have been going to have taking




148

8 Nominal group 1

8.1 Principal systems



appendix

149

8.2 Determiners




150

8.3 Determiners



appendix

151

8.4 Quantifiers






S ECTION TWO

WORD– CLAUSE– T EXT






155


In the second section, which includes works spanning two decadesfrom the mid-1960s to the mid-1980s, we observe how Halliday’sdeveloping theoretical framework is applied in the analysis and descrip-tion of patterns at various linguistic levels ranging from lexical item to

clause to text. The clause as lexico grammatical construct is different inkind from both lexis and text. While each is defined differently – lexical item defined by reference to collocation, clause by reference tostructure, and text by reference to context of situation – they arenevertheless analogous in nature and systemic in orientation.

‘Lexis as a linguistic level’ (Chapter 6), published in 1966, firstappeared in a volume dedicated to Halliday’s mentor, Professor J. R.Firth. Noting the importance Firth gave to lexical studies in descriptivelinguistics, Halliday explores some of the issues involved in handlinglexical patterns in language, which he regards as being different in kind,not just in degree of delicacy, from grammatical patterns. WhatHalliday is proposing is a lexical model “with distinct, though analo-gous, categories and forms of statement”. In lexical analysis, theoccurrence of an item appears to have more to do with collocationalrestrictions than a “known and stated set of terms in choice relation”.

This being the case, Halliday concludes, “even such a thing as a tableof most frequent collocates of specific items, with information abouttheir probabilities, unconditioned and lexically and grammatically con-ditioned, 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’



word – clause – text

156

(Chapter 7) introduces the three grammatically relevant ‘languagefunctions’: ideational, interpersonal and textual. Similar to the Praguelinguists, Halliday’s study of grammar represents a synthesis of both

structural and functional approaches. Grammar is described by Hallidayas 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 relativelyindependent ‘networks’; and these networks of options correspond to

certain basic functions of language”. Thus the functions of language arereflected in the structure of the clause. Halliday proceeds to show howeach 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 interpersonalmeaning is captured in the mood structure of the clause, and how thetextual 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 WilliamHaas, Halliday borrows Pike’s insight into language as particle, waveand field to distinguish between experiential structures which areconstituency-based (particle-like), interpersonal structures which areprosodic (field-like) and textual structures which are periodic (wave-like). Halliday also recognizes, as a distinct component, the logicalmode, in which “reality is represented in more abstract terms, in theform of abstract relations which are independent of and make noreference to things”.

Chapter 9 is based on what originally appeared as two separatepapers ‘How is a text like a clause?’ and ‘Text semantics and clausegrammar: some patterns of realization’. Both were written in 1980.

Making it clear that he considers a text to be a semantic rather than aformal lexicogrammatical entity, Halliday argues that ‘the elements of structure of the text are more abstract; they are functional entitiesrelating to the context of situation of the text, to its generic propertiesin terms of field, tenor and mode’. But having noted that texts andclauses have two distinct natures, texts being semantic and clauses beinglexicogrammatical, Halliday proceeds to point out how they are alike,metaphorically speaking. While on the one hand, clauses are the



editor’s introducton

157

constituents, or building blocks, of the text, they also have ‘evolved byanalogy 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 analogouscounterparts 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 spokenlanguage, i.e. a discussion between an adult and three nine-year-old

schoolgirls. The analysis is presented in ten steps, ranging from tran-scription 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 rhythm2. analysis into clauses and clause complexes, showing interdepen-

dencies and logical semantic relations3. analysis of clauses and clause complexes, for thematic (Theme–

Rheme) structure4. comparison of clauses and information units, and analysis of the

latter for information (Given–New) structure5. analysis of finite clauses for mood, showing Subject and Finite6. analysis of all clauses for transitivity, showing process type and

participant and circumstantial functions7. analysis of groups and phrases (verbal group, nominal group,

adverbial group, prepositional phrase)

8. analysis of grammatical and lexical cohesion9. identification, rewording and re-analysis of grammatical

metaphors10. 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 derivesfrom the linguistic system and how it comes to mean what it does.



158

Chapter Six

L EXIS AS A L INGUISTIC L EVEL (1966)

At a time when few linguists, other than lexicographers themselves,devoted much attention to the study of lexis, and outlines of linguisticsoften contained little reference to dictionaries or to other methods inlexicology, 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 possibleand useful to make formal statements about lexical items and their relations. For this purpose Firth regarded the statement of collocationas the most fruitful approach, and he sometimes referred, within theframework of his general views on the levels of linguistic analysis, tothe “collocational level”.3

The aim of this paper is to consider briefly the nature of lexicalpatterns in language, and to suggest that it may be helpful to devisemethods 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 beusefully thought of (a) as within linguistic form, and thus standing inthe same relation to (lexical) semantics as does grammar to (grammat-ical) semantics, and (b) as not within grammar, lexical patterns thus

being treated as different in kind, and not merely in delicacy, fromgrammatical patterns. This view is perhaps implicit in Firth’s recog-nition of a “collocational level”.4

One of the major preoccupations of grammatical theory in present-day 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.



l e xis a s a l ing uistic l e ve l

159

of delicacy greater than has hitherto been attained, and it is rightlyclaimed as a virtue of contemporary models that they permit moredelicate statements to be made without excessive increase in complex-

ity. A grammar is expected to explain, for example, the likeness andunlikeness between this brush won’t polish and this floor won’t polish, thethree-way ambiguity of John made Mary a good friend , and the non-

acceptability of beautiful hair was had by Mary beside the acceptability of the last word was had by Mary. Such explanations require the recognitionof distinctions which, as is well known, begin to cut across each other at a relatively early stage in delicacy, and the model has to accommo-

date cross-classification of this kind. The form of statement adopted(and the terminology) will of course depend on the model; but it isgenerally agreed that all such patterns need to be accounted for.5

As part of the process of accounting for these distinctions thegrammar attempts, both progressively and simultaneously, to reducethe very large classes of formal items, at the rank at which they can bemost usefully abstracted (for the most part generally as words, but thisis merely a definition truth from which we learn what “word” means),

into very small sub-classes. No grammar has, it is believed, achievedthe degree of delicacy required for the reduction of all such items toone-member classes, although provided the model can effectivelyhandle cross-classification it is by no means absurd to set this as theeventual aim: that is, a unique description for each item by itsassignment 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, contemporarygrammatical models can reasonably claim to aim at providing, therewould seem to be two possible evaluations of it. One is that, when themost delicate distinctions and restrictions in grammar have beenexplained, all formal linguistic patterns will have been accounted for;what is left can only be accounted for in semantic terms. The second isthat there will still remain patterns which can be accounted for in

formal linguistic terms but whose nature is such that they are bestregarded as non-grammatical, in that they cut across the type of relationthat is characteristic of grammatical patterning.

The particular model of grammar that is selected may suggest, butdoes not fully determine, which of these two views is adopted

operationally. For example, a model which distinguishes sharplybetween the grammar of a language and the use of the grammar,regarding corpus-based statistical statements as proper only to the latter,




160

and therefore as outside the range of validity of a descriptive statement,is less easily compatible with the second view than is a model whichdoes not make this distinction and which allows statistical statements a

place in linguistic description; nevertheless it is not wholly incompatiblewith it.6 Lexical statements, or “rules”, need not be statistical, or evencorpus-based, provided that their range of validity is defined in someother way, as by the introduction of a category of “lexicalness” toparallel that of grammaticalness.

One may validly ask whether there are general grounds, independentof any given model, for supplementing the grammar by formal state-

ments of lexical relations, at least (given that the aim of linguistics is toaccount for as much of language as possible) until these are shown tobe 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 isexplained lexically could also have been incorporated in the grammar;there still remains the question whether or not it could have beenexplained 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 patternsrepresented by such citations are to be stated with a sufficient degreeof abstraction, and whether this can best be achieved within or outsidethe 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 somesort of marked contrast with a powerful car and strong tea; in either case

the paradigmatic relation of strong to powerful is not a constant butdepends 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, willamount to saying that, first, strong and powerful are members of a classthat enters into a certain structural relation with a class of whichargument is a member; second, powerful (but not strong ) is a member of aclass entering into this relation with a class of which car is a member;




161

and third, strong (but not powerful ) is a member of a class entering intothis relation with a class of which tea is a member. It would be hopedthat such classes would reappear elsewhere in the grammar defined on

other criteria. Argument, car and tea will, for example, already have beendistinguished on other grounds on the lines of “abstract”, “concreteinanimate” 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, toaccount for the power (but not the strength) of his car and the strength (butnot the power) of her tea. But these involve different structures; elsewherein the grammar strong , strongly, strength and strengthened have beenrecognized as different items and assigned to different classes, so thatthe 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 dimensionof grammatical classification or will need to be restated for each primaryclass.

But the added complexity involved in either of these solutions doesnot seem to be matched by a gain in descriptive power, since for thepatterns in question the differences of (primary) class and of structureare irrelevant. Strong , strongly, strength and strengthened can all beregarded for this present purpose as the same item; and a strong argument ,he argued strongly, the strength of his argument and his argument wasstrengthened 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) andtea; and an item power (powerful , powerfully) which collocates with argue

and car. It can be predicted that, if a high-powered car is acceptable, thiswill 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 itemsenter that is irrelevant; it may also be irrelevant whether they enter into any grammatical relation with each other or not. They may be in




162

different sentences, for example: I wasn’t altogether convinced by hisargument. He had some strong points but they could all be met. Clearly thereare limits of relevance to be set to a collocational span of this kind; but

the question here is whether such limits can usefully be definedgrammatically, 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 courseno procedural priority as between the identification of the items andthe identification of the paradigmatic and syntagmatic relations into

which they enter: item, set and collocation are mutually defining. Butthey are definable without reference to grammatical restrictions; or, if that is begging the question, without reference to restrictions statedelsewhere in the grammar. This is not to say that there is no inter-relation between structural and collocational patterns, as indeed therecertainly is; but if, as is suggested, their interdependence can beregarded as mutual rather than as one-way, it will be more clearlydisplayed by a form of statement which first shows grammatical and

lexical restrictions separately and then brings them together.7 If there-fore one speaks of a lexical level, there is no question of asserting the

“independence” of such a level, whatever this might mean; what isimplied is the internal consistency of the statements and their refer-ability to a stated model.

Possible methods of lexical analysis, and the form likely to be takenby 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 thistype of pattern in language, and some of the problems of accountingfor it. Clearly lexical patterns are referable in the first place to the two

basic axes, the syntagmatic and the paradigmatic. One way of handlinggrammatical relations on these two axes is by reference to the theoret-ical categories of structure and system, with the class definable as thatwhich enters into the relations so defined. In lexis these concepts needto 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 whichthe value of an element depends on complex factors in no sensereducible to simple sequence, lexis seems to require the recognitionmerely of linear co-occurrence together with some measure of signifi-cant proximity, either a scale or at least a cut-off point. It is thissyntagmatic relation which is referred to as “collocation”. The impli-cation that degree of proximity is here the only variable does not of




163

course imply how this is to be measured; moreover, it clearly relatesonly to statements internal to the lexical level: in lexicogrammaticalstatements collocational restrictions intersect with structural ones. Simi-

larly in place of the “system” which, with its known and stated set of terms in choice relation, lends itself to a deterministic model, lexisrequires the open-ended “set”, assignment to which is best regarded as

probabilistic. Thus while a model which is only deterministic canexplain so much of the grammar of a language that its added power makes it entirely appropriate for certain of the purposes of a descriptivegrammar, it is doubtful whether such a model would give any real

insight into lexis. Collocational and lexical set are mutually defining asare structure and system: the set is the grouping of members with likeprivilege of occurrence in collocation.

Second, in grammar a “bridge” category is required between ele-ment of structure and term in system on the one hand and formal itemon the other; this is the class. (This specific formulation refers to the“scale-and-category” version of a system-structure model; but it isprobably true that all models make use of a category analogous to what

I am here calling the “class”.) In lexis no such intermediate category isrequired: the item is directly referable to the categories of collocation

and set. This is simply another way of saying that in lexis we areconcerned with a very simple set of relations into which enter a largenumber of items, which must therefore be differentiated qua items,whereas in grammar we are concerned with very complex and variablerelations in which the primary differentiation is among the relationsthemselves: it is only secondarily that we differentiate among the items,and we begin by “abstracting out” this difference. In other words thereis 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 axiswith the item, or rather with any of the items, identified and accountedfor in the grammar. For example, on the paradigmatic axis, in she made up her face one can identify a lexical item make up

1 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 isnecessary is shown by the ambiguity of she made up the cast , she was

responsible for the makeup of the cast. Grammatically, the primary distinc-tion is that between made up and makeup; this distinction of courseinvolves a great many factors, but it also relates to many other itemswhich are distinguishable, by class membership, in the same way. If the




164

grammar is at the same time to handle the distinction between make up

1 and make up

2 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 byad hoc grammatical devices: here for instance the potentialities of make up

2 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 isof 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 candistinguish grammatically, but not lexically, between they want the pilot to take off (= “so that they can take off” and = “they desire him to doso”): these are not necessarily distinguished by intonation, although theunmarked 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 itemwhich has no defined status in the grammar and is not identified asmorpheme, 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 discontinuousitem in the grammar; but this complexity is avoided if one is preparedto recognize a lexical item let in for without demanding that it shouldcarry any grammatical status. Similarly the ambiguity in he came out witha beautiful model may be explained, instead of by giving two differentgrammatical 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 betweenthe items of grammar and those of lexis is the norm, but merely thatfor certain purposes it is useful to have a descriptive model of language

that allows for it. At the same time the above considerations suggestthat the lexical component requires not, as it were, a second ‘run-through’ of the model designed for the grammar but rather a specificallylexical model with distinct, though analogous, categories and forms of statement.

Nor is it suggested that the set of patterns recognized as languageform is neatly divided into two types, the grammatical and the lexical.A model for the description of language form may recognize only one




165

kind of pattern and attempt to subsume all formal relations within it:some grammatical models, as has been noted, envisage that it is thegrammar’s task to distinguish strong from powerful as well as to dis-

tinguish a from the and “past” from “present”; while a lexicographicalmodel in which a and the , as well as strong and powerful , are entered inthe dictionary and described by means of citations could be regarded as

in a similar way attempting to subsume grammar under lexis. Evenwhere the model recognizes two distinct kinds of pattern, these stillrepresent different properties of the total phenomenon of language, notproperties of different parts of the phenomenon; all formal items enter

into patterns of both kinds. They are grammatical items when describedgrammatically, as entering (via classes) into closed systems and orderedstructures, and lexical items when described lexically, as entering intoopen sets and linear collocations. So in a strong cup of tea the grammar recognizes (leaving aside its higher rank status, for example as a singleformal item expounding the unit “group”) five items of rank “word”assignable to classes, which in turn expound elements in structures andterms 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 thelexical, is more significant for the explanation of restrictions on their occurrence qua items. The items a and of are structurally restricted, andare uniquely specified by the grammar in a very few steps in delicacy;collocationally on the other hand they are largely unrestricted. For theitem strong , however, the grammar can specify uniquely a class (sub-class 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 fromother members of the class (and if the members of its ‘scatter’ strong ,strength, etc., turn out to operate collocationally as a single item thenthis conflated item is not even specifiable qua class member); colloca-tionally, however, it is restricted, and it is this which allows itsspecification as a unique item. There might then appear to be a scale

on which items could be ranged from “most grammatical” to “mostlexical”, the position of an item on the scale correlating with its overallfrequency ranking. But these are three distinct variables, and there isno reason to assume a correlation of “most grammatical” with either “least lexical” or “most frequent”. The “most grammatical” item is onewhich is optimally specifiable grammatically: this can be thought of as“reducible to a one-member class by the minimum number of steps indelicacy”. Such an item may or may not be “least lexical” in the sense




166

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 environ-ment. This therefore takes into account the frequency of the item in astated environment relative to its total frequency of occurrence. Whilea and of are unlikely to occur in any collocationally generalizableenvironment with a probability significantly different from their overallunconditioned 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. Asextreme cases, fro and spick may never occur except in environmentsincluding respectively to and span (the fact that to and fro accounts for only a tiny proportion of the occurrences of to, while spick and spanmay account for all occurrences of this item span, is immaterial to thespecification 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 thedefinition 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 thelexical set (of items) can be defined by reference to collocation. Sinceall items can be described lexically, the relation of collocation could beregarded as being, like that of structure, chain-exhausting, and a lexicalanalysis programme might well begin by treating it in this way; but thisis not a necessary condition of collocation, and if closed-system itemsturn out, as may be predicted, to be collocationally neutral these itemscould at some stage be eliminated by a “deletion-list” provided either by cross-reference to the grammar or, better, as a result of the lexicalanalysis itself. Once such “fully grammatical” items are deleted, collo-cation 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 isa chain-exhausting unit (text items being then fully accounted for insentence structure and again in clause structure and so on), it does notseem useful to postulate such an ordered hierarchy for lexis. Lexicalitems 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, andthough triads of this kind may be rare it looks as though we need the




167

categories of simple and compound , and perhaps also phrasal, lexicalitems, in addition to collocational span, as units for a lexical description.Since the only “structural” relation in lexis is one of simple co-

occurrence, these represent a single serial relation: the item stone enters(say) into the collocation grindstone , which then does not itself collocateexactly like the sum of its parts but enters as an item into (say) the

collocation nose to the grindstone , which likewise does not collocate likethe 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 suchcompounding yields a morphological (upward) grouping of items, the

lexical series which, like its analogue in grammar, may or may notcoincide 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 isformed 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. Equiva-

lence or non-equivalence between series and set is an interesting featureof 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 specificto lexis. The type–token relation can be made dependent on classmembership: just as in grammar two occurrences assigned to differentprimary classes, such as ride (verb) and ride (noun), can be regarded asdifferent (grammatical) items, so in lexis two occurrences assigned todifferent primary sets can be regarded as different lexical items. Thiscan then be used to define homonymity: if the two occurrences of model in the example above are shown to differ according to criteriawhich would assign them to different sets then they represent twohomonymous items. It is not to be assumed, of course, that grammat-

ically distinguished items such as ride (verb) and ride (noun) may notalso 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 besaid to constitute a single lexical item, as also will strong , strength, stronglyand strengthen, and perhaps also (if they can be suitably delimited) non-cognate “scatters” such as town and urban. This would provide a basisfor deciding how many lexical items are represented by “expressions”such as form, stand and term.




168

If we say that the criterion for the assignment of items to sets iscollocational, 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 weconsider n occurrences of a given (potential) item, calling this item thenode, and examine its collocates up to m places on either side, giving a

span of 2m, the 2mn occurrences of collocates will show a certainfrequency distribution. For example, if for 2000 occurrences of sun welist the three preceding and three following lexical items, the 12,000occurrences 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 frequentcollocates.

On the basis of their high probability of occurrence (relative to their overall frequency) in collocation with the single item sun, the itemsbright , hot , shine , light , lie , come out constitute a weak provisional set; thisresembles the weak provisional class recognizable in the grammar on

the basis of a single “item-bound” substitution frame – although inlexis it is relatively less weak because of the lower ceiling of generality:

lexis is more item-bound than grammar. If we intersect these with thehigh frequency collocates of moon we get a set, whose members includebright , shine and light , with slightly greater generality. That is to say,bright , shine and light are being grouped together because they display asimilar potentiality of occurrence, this being now defined as potentialityof occurrence in the environment of sun and in that of moon. Theprocess can be repeated with each item in turn taken as the node; thatis, as the environment for the occurrence of other items. The set willfinally be delimited, on the basis of an appropriate measure of likeness,in such a way that its members are those items showing likeness intheir total patterning in respect of all those environments in which theyoccur 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. Manyother factors are involved, such as the length of the span, the signifi-cance of distance from the node and of relative position in sequence,the possibility of multiple nodes and the like. One point should bementioned here: this is the importance of undertaking lexicogrammat-ical as well as lexical analysis. It is not known how far collocationalpatterns are dependent on the structural relations into which the items




169

enter. For example, if a cosy discussion is unlikely, by comparison with acosy chat and a friendly discussion, is it the simple co-occurrence of thetwo items that is unlikely, or their occurrence in this particular

structure? All that has been said above has implied an approach inwhich grammatical relations are not taken into account, and reasonshave been given for the suggestion that certain aspects of linguistic

patterning will only emerge from a study of this kind. But it is essentialalso to examine collocational patterns in their grammatical environ-ments, 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 meaningfulone.9

An investigation on the lines suggested requires the study of verylarge samples of text. The occurrence of an item in a collocationalenvironment 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 clearlyregarded as “lexical”. Moreover the native speaker’s knowledge of hislanguage will not take the form of his accepting or rejecting a givencollocation: he will react to something as more acceptable or lessacceptable on a scale of acceptability. Likely collocations could beelicited 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 agiven node;10 but the number of such studies that would be requiredto cover even the most frequent lexical items in the language is verylarge indeed. Textually, some twenty million running words, or

1500–2000 hours of conversation, would perhaps provide enoughoccurrences to yield interesting results. The difficulty is that, sincelexical patterns are of low generality, they appear only as properties of very large samples; and small-scale studies, though useful for testingmethods, give little indication of the nature of the final results.

It is hard to see, however, how the results could fail to be of interestand significance for linguistic studies. Their contribution to our know-ledge of language in general, and of one language in particular, mayperhaps be discussed in relation to the use of the term “semantics”. If lexis is equated with semantics, the implication is that lexical patternscan only be described either externally (that is, as relations betweenlanguage and non-language, whether approached denotatively or con-textually) or lexicogrammatically (that is, in dependence on grammati-




170

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 withtea. The fact that the labels for grammatical categories are chosen onsemantic grounds should not be taken to imply that they represent anadequate substitute for grammatical semantics; but equally the existenceof traditional methods in lexical semantics does not mean that lexicalitems display no internal, formal patterns of their own.

A thesaurus of English based on formal criteria, giving collocationallydefined lexical sets with citations to indicate the defining environments,

would be a valuable complement to Roget’s brilliant work of intuitivesemantic 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 abouttheir probabilities, unconditioned and lexically and grammatically con-ditioned, 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. Theseinclude 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 ownenvironment, and the cohesive power of lexical relations, are of greatpotential interest.12 Lexical information is also relevant to foreignlanguage teaching; many errors are best explained collocationally, anditems can be first introduced in their habitual environments.13 A further

possible field of application is information retrieval: one research groupin this field is at present undertaking a collocational analysis of thelanguage of scientific abstracts. 14

Only a detailed study of the facts, such as that now being undertakenby Sinclair,15 can show in what ways and to what extent the introduc-tion of formal criteria into the study of lexis, as implied by therecognition of a “lexical level”, are of value to any particular applica-tions of linguistics. But there seem to be adequate reasons for expecting




171

the results to be interesting; and if they are, this is yet another indicationof the great insight into the nature of language that is so characteristicof 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 thesyntagmatic 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 givenword 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 ele-

ments 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 thesein 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 maybe 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 cliches: in shabby treatment the

mutual expectancy is purely lexical, and is paralleled in they treated him




172

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.

10. Compare the methods used to assess the disponibilite of lexical items in thedevelopment of “Francais fondamental” (Gougenheim, Michea, Rivenc

and Sauvageot 1956).

11. Roget (1960: 8).

12. Cf. Firth (1951).

13. 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, fashionableairliner, 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.

14. This research is being undertaken by Dr. A. R. Meetham and Dr. P. K.

T. Vaswani at the National Physical Laboratory, Teddington, Middlesex.

15. Sinclair, Jones and Daley (1970).



173

Chapter Seven

L ANGUAGE STRUCTURE AND

L ANGUAGE FUNCTION (1970)

1 The functions of language

Why is language as it is? The nature of language is closely related tothe 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 useof 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 moregeneral functions which are common to all cultures. We do not all goon 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 notbased on an analysis of linguistic structure, will not answer the question;we cannot explain language by simply listing its uses, and such a listcould in any case be prolonged indefinitely. Malinowski’s ethnographicaccount of the functions of language, based on the distinction between“pragmatic” and “magical”, or Buhler’s well-known tripartite divisioninto 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 notintended primarily to throw light on the nature of linguistic structure.

At the same time, an account of linguistic structure that pays noattention 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.




174

perspicacity, since it offers no principles for explaining why thestructure 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 sometime; but the usefulness of a synthesis of structural and functionalapproaches has long been apparent from the work of the Prague

linguists (Vachek 1966) who developed Buhler’s ideas, especially in thestudy of grammar. The particular form taken by the grammatical systemof language is closely related to the social and personal needs thatlanguage 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 atthe same time; otherwise we will lack any theoretical basis for general-izations 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 addsomething 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 areconcerned not with deliberate acts of choice but with symbolic

behaviour, in which the options may express our meanings only veryindirectly: in the same sense we may be said to “opt” between a longvowel and a short one, or between a straight arm and a bent one(where the meaning is likewise mediated through the symbolic signifi-cance of the distinction between a handshake and a salute). The systemof 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 repetitiveexercise of options in social and personal situations and settings (Ellis1966; Pike 1967; Firth 1968).

It is fairly obvious that language is used to serve a variety of differentneeds, 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 numbersof options embodied in it combine into a very few relatively indepen-dent “networks”; and these networks of options correspond to certainbasic functions of language. This enables us to give an account of thedifferent functions of language that is relevant to the general under-standing of linguistic structure rather than to any particular psycho-logical or sociological investigation.

1. Language serves for the expression of “content”: that is, of the



language structure and language function

175

speaker’s experience of the real world, including the inner world of hisown consciousness. We may call this the ideational function, though itmay be understood as easily in behavioural as in conceptual terms

(Firth 1968: 91). In serving this function, language also gives structureto experience, and helps to determine our way of looking at things, sothat 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 rolescreated by language itself – for example the roles of questioner or

respondent, which we take on by asking or answering a question; andalso for getting things done, by means of the interaction between oneperson and another. Through this function, which we may refer to asinterpersonal, social groups are delimited, and the individual is identi-fied and reinforced, since by enabling him to interact with otherslanguage also serves in the expression and development of his ownpersonality.

These two basic functions, to each of which corresponds one broad

division in the grammar of a natural language, are also reflected inBernstein’s studies of educational failure (e.g. Bernstein 1970). Bern-

stein’s work suggests that in order to succeed in the educational systema child must know how to use language as a means of learning, andhow to use it in personal interaction; these can be seen as specificrequirements on his control of the ideational and interpersonal func-tions of language.

3. Finally, language has to provide for making links with itself andwith features of the situation in which it is used. We may call this thetextual function, since this is what enables the speaker or writer toconstruct “texts”, or connected passages of discourse that is situationallyrelevant; and enables the listener or reader to distinguish a text from arandom set of sentences. One aspect of the textual function is theestablishment of cohesive relations from one sentence to another in adiscourse (Hasan 1968).

All these functions are reflected in the structure of the clause. Herewe attempt to show, by reference to English, what a clause is: how itserves for the realization of a number of very general meanings, or semantic options, relating to the interpersonal, ideational and textualfunctions of language; and how these are expressed through variousconfigurations of structural “roles” – functional elements such as“process” and “actor” that derive from these basic functions. For amore detailed exemplification we shall consider an aspect of ideational




176

meaning, the system of transitivity; the remaining areas, which havethe same formal properties, will be referred to only briefly. Any oneclause is built up of a combination of structures deriving from these

three functions (for the sake of brevity we shall leave out the logicalcomponent in linguistic structure, which is somewhat different in itsrealizations).

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 optionssimultaneously. Hence the various sets of structural “roles” are mappedonto 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, isat once Actor and Subject and theme (see, 13 below). Each of thesethree represents a value in some configuration – some melodic line, soto speak – such as “Process plus Actor plus Goal”. And all such

configurations are meaningful, since what we have called the basicfunctions of language, looked at from another point of view, are simplydifferent kinds of meaning. So for example there is a difference inmeaning 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 apiece of information, and it derives from the textual function. There isalso 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. Thespeaker 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 intoa 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 discreteoperation and it would be impossible ever to answer a question that hadactually been asked. Speech acts involve planning that is continuous andsimultaneous 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




177

to write a theoretical grammar of just one instance if the need arises; itusually does not). It is concerned rather with the description of speechacts, 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 distinctionbetween an idealized knowledge of a language and its actualized use:

between “the code” and “the use of the code”, or between “compe-tence” and “performance”. Such a dichotomy runs the risk of beingeither unnecessary or misleading: unnecessary if it is just another namefor the distinction between what we have been able to describe in the

grammar and what we have not, and misleading in any other interpre-tation. The study of language in relation to the situations in which it isused – to situation types, i.e. the study of language as ‘text’ – is atheoretical pursuit, no less interesting and central to linguistics thanpsycholinguistic investigations relating the structure of language to thestructure of the human brain.

We shall consider each of the functions in turn as it is reflected in thestructure of the English clause, beginning with what we have called the

“ideational”. To the adult – though not, be it noted, to the child – thepredominant demand that we make on our language (predominant, at

least, in our thinking about language; perhaps that is all) is that it allowsus to communicate about something. We use language to represent our experience of the processes, persons, objects, abstractions, qualities, statesand relations of the world around us and inside us. Since this is not theonly demand we make on language it is useful to refer to it specifically;hence ideational function, ideational meaning, etc. (Other terms thathave been used in a similar sense are “representational”, “cognitive”,“semantic”, “factual-notional” and “experiential”.)

Let us consider the expression of processes: of actions, events, statesand relations, and the persons, objects and abstractions that are associ-ated with them. For this purpose we will focus our attention on oneunit of linguistic structure, namely the clause. In any language, a vastnumber of different processes can be distinguished; but these are

reducible to a small number of process types, and the grammar of everylanguage comprises sets of options representing broad categories of thiskind. The most familiar, and simplest, model is that which groups allprocesses 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, objectsor other classes of phenomena may play in the process concerned. For example, in:




178

(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 justthese 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”; thesebeing realized (iii) by built , Sir Christopher Wren and this gazeborespectively.)

3 Transitivity functions: process and participant rolesThe roles which appear in the expression of processes are of differentkinds. First there is the process itself, usually represented by a verb, for example built in (2). Then there are the participant functions, thespecific roles that are taken on by persons and objects, for exampleWren and gazebo; and finally there are what we may call the circumstan-tial functions, the associated conditions and constraints such as those of

time, place and manner (Fillmore 1968, where the two together arereferred 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 thedistinction 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) thegazebo 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 prepositionalform 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 ). Butthere may be many, often contradictory, criteria to choose from (seebelow, Section 4); moreover, the more categories one sets up, the




179

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 “grammat-

ical” categories (Sweet 1891: 10ff., 89ff.). In Sweet’s terms, General Leathwall is the logical subject in (5i–iii), though it is the grammaticalsubject only in (5i). Conversely in the book sells well , the book is gram-matical subject but “logical direct object”. The concepts of actor, goaland 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 thegeneral term transitivity. Transitivity comes under what we have calledthe ‘ideational’ function of language. Actor, Goal and Beneficiary are

structural functions, or roles, in transitivity; and just as the sametransitivity function may be realized in more than one way, as in (5),so also the same constructional form may express different transitivityfunctions. 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 beconsidered instrument rather than actor, on the grounds that it isinanimate. 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 grammat-ical 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).




180

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) Actor (“logical subject”): prepositionally by(b) Goal (“logical direct object”)(c) Beneficiary (“logical indirect object”): prepositionally to / for (d) 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, circum-stance – 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 bynominal groups and circumstances by adverbial groups – the last often

in the form of prepositional phrases. There are also incongruent formsof expression, with functions of one type expressed by classes primarilyassociated 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 twoparticipants, one is omitted and the other (roast beef ) is made to qualifydinner (contrast in the evening they ate roast beef and then swam).

The circumstantial functions seem less central to the process than dothe 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 thanon the bridge : we can say what was he throwing stones at? and not (in thissense) what was he doing at the bridge? (On the other hand, we can say




181

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 variousfactors 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, fromwhom (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 relatetransitivity functions to a system of clause types. As, however, thisinvolves recognizing that an “obligatory” element may in fact beabsent, we shall use the term “inherent” rather than “obligatory”. Aninherent function is one that is always associated with a given clausetype even if it is not necessarily expressed in the structure of all clausesof 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 men-tioned the receiver interprets the process as having an instrumental roleassociated with it.

The same verb may occur in clauses of more than one type. But




182

within one type there may be different sets, and different alignments,of participants; this is the function of the system of voice – of the choicebetween active and passive, though the actual patterns are more

elaborate than this. The options in the voice system (simplifyingsomewhat) are (a) middle / non-middle (see next paragraph); if non-middle, then (b) “active” / “passive” (not exactly equivalent to active

and passive in the verb; see Halliday, 1967–68: § I 39ff., where theyare referred to as “operative” and “receptive”); if “active”, then (c)plus / minus goal; if ‘passive’, then (d) plus / minus actor. The reasonfor choosing one rather than another of these options lies in the textual

function of language (see Sections 11 and 12 below); but which optionsare 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” clauseis one which has only one inherent participant, which for the momentwe 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”), andif ‘passive’, no clear Actor, e.g. the clothes have been washed (“by Mary”).

All actions are classified into those involving one participant role andthose involving two; there are then different ways of presenting thesituation in those cases where there are two.

The point was made earlier that the notion of “participant” derivesfrom the more fundamental concept of syntactic function, or “role”.The basic elements of transitivity structure are the various roles associ-ated with processes; and two or more such roles may be combinedin one participant, as in a reflexive clause such as John is washing (“himself”) where John is both Actor and Goal at the same time. Theelements that operate as Actor, Goal, etc. also play a part, simul-taneously, in other structures of the clause, expressing aspects of theinterpersonal 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 thatmakes it possible for the various functions of language to be integratedin 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




183

this type as action clauses. Action clauses all have correspondingequative 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 inaction clauses, together with the roles associated with each of them:

voice roles voice example

(clause ) (verb)middle Actor active the gazebo has collapsed

{

non-

mid

dle

active

active

passive

passive

passive

Actor, Goal

Actor, (Goal)

Goal

Goal, Actor

Goal, (Actor)

active

active

active

passive

passive

the Council are selling the gazebo

the Council won’t sell

the gazebo won’t sell

the gazebo has been sold by the Council

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. Mentalprocess clauses, and clauses of relation, are associated with what are atfirst sight rather different sets of participant roles.

7 Transitivity clause types: mental process clauses, relationclauses

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 anyrate animate, being whose consciousness is impinged upon, and somephenomenon which impinges upon it. Let us refer to these as the“Processer” and the “Phenomenon”. The voice potentialities are nowsomewhat different; among the non-middle (two participant) clausesthere are two types: those having the Phenomenon as Subject in active




184

voice (15i), and those having the Processer (15ii). In the first types, thepassive 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 governedby by into prominence as the focus of information (see below, Section12); in (15ii) the by element, i.e. her , is the Processer, and in Englishthis tends to be the “given” element in the situation (she must have

been referred to already in the text), and thus does not appropriatelycarry 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) verbali-zation, e.g. say, speak. They are distinct in that the Phenomenon – thatwhich 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 phenom-

enon of this kind, but it may also be what we might call a metaphe-nomenon: a fact or a report – a phenomenon that has already as it werebeen filtered through the medium of language. Here words as well asthings may participate in the process.

For example, in (16) all the “processed” entities are simple phenom-ena, or “things”:

(16i) I noticed Helen over there [person](16ii) I noticed a discrepancy [abstraction]

(16iii) I noticed a quarrel (going on) / them quarrelling [event](16iv) 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) I noticed what (the fact of what) she was wearing(17ii) it worries me that you look so tired(18i) I notice the bank rate’s going up again

(18ii) 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 clausesaccept only reports, and “reported speech” is the meaning of clauses of




185

this type. The difference between fact and report is that a “fact” is arepresentation at the semantic level, where the truth lies in the meaning

– (she regretted ) that he had gone away; whereas a “report” is a represen-

tation at the lexicogrammatical, or syntactic, level, where the truth liesin 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 tothe class of poets,” “. . . the class of people who look desperate”. Thisis a relation between entities of the same order of abstraction butdiffering 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 theone serves to identify the other, as in x = 2. Here the two entities arealike in generality but differ in abstraction: the identifying element maybe 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 berecognized:

(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 werein view on the platform; there is in fact partial ambiguity betweenthese two sub-types.

These two major types of relational clause, the attributive and theequative, 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 therole Attribute which may be an adjective and is usually indefinite,express class inclusion, are usually questioned by what? or how? and areexpressed by the verbs be , get , turn, keep, remain, seem, sound , look, etc.

The equative are reversible (i.e. have a “voice” system), have the roleIdentifier which must be as noun and is usually definite, express classidentity, are usually questioned by who? or which? and are expressed bythe verbs be , equal , represent , resemble , stand for , etc.




186

It is interesting to note that, in relational clauses, quite unlike clausesof action or mental process, the verb is regularly unstressed. This is asymptom 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, theEnglish 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 frame-work for the representation of processes in the grammar; possibly alllanguages distinguish three such categories. We need to ask, at thispoint, whether the structural functions can be generalized across clausetypes; 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 amental process clause. This may be approached through a reconsidera-tion of the functions in action clauses, a reconsideration which suchclauses 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 andGoal, (ii) causative, with Initiator and Actor. However, there is aproblem 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 madebetween doing something to someone and making someone do some-thing, so that (23iii) can be interpreted in either way without any senseof 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 thissituation, since with these we are forced to describe (23i) and (23ii)differently. The distinction between them is by no means entirely




187

unreal, since there are verbs like lead which are normally transitive (twoinherent participants) and others like march which are normally intran-sitive (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 verbbounce , 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 thatMary 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 tobe organized on an ergative rather than on a transitive (or “nomina-tive”) basis. This means that, with any action clause, there is associatedone inherent role which is that of the participant affected by the processin 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 equallynormal; it may be considered the “favourite” clause type of ModernEnglish. The transitive and intransitive types – those with non-middleas 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 aProcess and an Affected. Those in (24ii) also have a Causer (Fillmore’s“agentive”):

(24i) they’re being ledthey’re being trained / they’re training

they’re being marched / they’re marching(24ii) he’s leading them

he’s training themhe’s marching them

These two ways of representing processes, the transitive and theergative, are very widely distributed; possibly all languages display oneor the other, or (perhaps always) both, in different mixtures. In English,




188

the two occur side by side. The transitive system asks “does the actionextend beyond the active participant or not?”, the ergative, “is theaction caused by the affected participant or not?” The ergative com-

ponent is more prominent now than it was in Middle English, and thisappears in various ways, for example, the change from impersonal topersonal forms in mental process clauses (formerly methinks, it likes me ).

In the modern form I like , I cannot be explained as an Actor (amongother things we cannot say what he does to jam is like it ); but it can beshown on various grounds to have the function Affected.

As this suggests, the ergative pattern, whereby a Process is accom-

panied by an obligatory Affected participant and an optional Causer, ismore readily generalizable than that of Actor and Goal. It extendsbeyond action clauses to those of mental process, and perhaps even toclauses of relation as well. We want to say that Paul has the samefunction in both (25i) and (25ii):

(25i) Paul fears ghosts(25ii) ghosts scare Paul

– not that they are identical in meaning, but that the transitivity rolesare the same. This is not possible in Actor–Goal terms. But in anergative system there is considerable evidence for regarding Paul as the‘Affected’ participant in both cases. The ergative, therefore, representsthe more general model of the transitivity patterns of Modern English

– that is, of the options available to the speaker of English for talkingabout processes of all kinds.

9 Other dimensions of clause structure

So far the discussion has been confined to the expression of ideationalmeanings. We have not yet considered the structure of language in itsother functions, the “interpersonal” and the “textual”. Both thesefunctions are manifested in the structure of the clause.

Certain problems that have arisen in the history of the investigationof subject and predicate provide an insight here. A sentence such as(26i) presents no problem in this respect: my mother is clearly subjectand the rest predicate. But in (26ii) there seem to be three candidatesfor 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




189

The solution was to recognize different kinds of subject. For Sweet,my mother was “logical subject”, I was “grammatical subject”; these beadscame to be known as ‘psychological subject’. In (26i), all these coincide.

The notion of subject conflates three distinct roles which, althoughthey are typically combined into one element, are nevertheless inde-pendent 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 selectedunless there is good reason for selecting otherwise (cf. Jakobson 1963:268ff.).

These three “kinds of subject” relate to the functions of language asdescribed above. The logical subject is the actor; this is a transitivityrole, deriving from the ideational function. The other two havedifferent sources, though they are no less meaningful. The grammaticalsubject derives from the interpersonal component in language function:specifically, it has to do with the roles taken on by the performer andreceiver in a communication situation. The psychological subjectbelongs to the textual component; it is concerned with the organization

of the clause as a message, within a larger piece of discourse. The nexttwo sections will examine these in turn.

10 Mood

As we have said, one function of language is to provide for interactionbetween people, by allowing the expression of statuses, social andindividual attitudes, assessments, judgements and the like; and thisincludes participation in linguistic interaction. Language itself definesthe roles which people may take in situations in which they arecommunicating with one another; and every language incorporatesoptions whereby the speaker can vary his own communication role,

making assertions, asking questions, giving orders, expressing doubtsand 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 thesystem of mood (cf. Sweet 1891: 105), in which the principal optionsare declarative, interrogative (yes / no and wh- types) and imperative.The difference between he can and can he? is a difference in thecommunication role adopted by the speaker in his interaction with alistener.

The notion “grammatical subject” by itself is strange, since it impliesa structural function whose only purpose is to define a structural




190

function. Actually, just as the “logical subject” is a function defined bytransitivity, so the “grammatical subject” is a function defined bymood. If we consider an example such as (27):

(27) Tigers can climb trees. – Can tigers climb trees? – They canclimb trees, can’t they? – No they can’t.

we find that one part, tigers can, has the function of expressing moodthroughout; it also typically carries the positive / negative option. Itconsists 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 meaningfulfunction in the clause, since it defines the communication role adoptedby the speaker. It is present in clauses of all moods, but its significancecan perhaps be seen most clearly in the imperative, where the meaningis “request you to . . .”; here the speaker is requiring some action onthe 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 onlyoption is plus or minus the speaker himself, as in let’s go home asopposed to (you) go home. Hence, in a passive imperative such as be

guided by your elders, although the actor is your elders, the modal subjectis “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 optionsby means of which a speaker or writer is enabled to create texts – touse language in a way that is relevant to the context. The clause, inthis function, is organized as a message; so in addition to its structurein transitivity and in mood, it also has structure as a message, what is

known as a “thematic” structure. (It was linguists of the Prague schoolwho first studied this aspect of language, cf. Mathesius 1928; Firbas1959; 1964; Svoboda 1968 and references therein.)

The English clause consists of a Theme and a Rheme. The Themeis another component in the complex notion of subject, namely the“psychological subject”; it is as it were the peg on which the messageis hung, the theme being the body of the message. The Theme of aclause is the element which, in English, is put in first position; in (28




191

i–v) the Theme is the item outside the brackets, what is inside beingthe Rheme:

(28i) I (don’t know)(28ii) yesterday (we discussed the financial arrangements)

(28iii) his spirit (they could not kill)(28iv) suddenly (the rope gave way)(28v) people who live in glasshouses (shouldn’t throw stones)

As we have seen, Theme, Actor and modal Subject are identical unlessthere is good reason for them not to be (cf. (26) above). Where they

are not, the tendency in Modern English is to associate Theme andmodal Subject; and this is the main reason for using the passive. Thepassive has precisely the function of dissociating the Actor from thiscomplex, 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 aninterrogative is a request for information. Hence we put first, in aninterrogative clause, the element that contains this request for infor-mation: 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 byputting something in first position in the clause. The Theme is thepoint of departure for the message; a paradigm form of it is theheadword in a definition, for example a gazebo in (31):

(31) a gazebo is a pavilion or summerhouse on an eminence, openfor the view

In addition to the selection of a particular element as the theme, thespeaker has other options in thematic structure open to him (Halliday1967–68: § 2); for example, any clause can be split into two parts bythe use of nominalization, as in




192

(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 textualorganization of language, which we may call information structure. Thisrefers to the organization of a text in terms of the functions Given andNew. These are often conflated with Theme and Rheme under thesingle heading “topic and comment”; the latter, however, is (like thetraditional notion of “subject”) a complex notion, and the association of Theme with Given, Rheme with New, is subject to the usual “goodreason” principle already referred to – there is freedom of choice, butthe Theme will be associated with the Given and the Rheme with theNew unless there is good reason for choosing some other alignment.

In English, information structure is expressed by intonation. Con-

nected speech takes the form of an unbroken succession of distinctivepitch contours, or tone groups; each tone group represents what the

speaker decides to make into one unit of information. This is notnecessarily the same length as a clause, though it often is so. Theinformation unit consists of an obligatory New element – there mustbe something new, otherwise there would be no information – and anoptional Given element; the main stress (tonic nucleus) marks the endof the New element, and anything that is Given precedes it, unless

with good reason – which means, here, unless it is a response to aspecific question, either asked or implied. The function Given means‘treated by the speaker as non-recoverable information’: informationthat the listener is not being expected to derive for himself from thetext or the situation.

(33) illustrates the interaction of information structure with thematicstructure (information unit (“i.u.”) boundaries are marked by //; mainstress 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




193

actually (New) quite impossible’. No such suggestion need actuallyhave been made for this clause to occur; one of the features of theGiven–New structure is its use for various rhetorical purposes, such as

bullying the listener. Given and New thus differ from Theme andRheme, though both are textual functions, in that Given means ‘hereis a point of contact with what you know’ (and thus is not tied to

elements in clause structure), whereas Theme means ‘here is theheading to what I am saying’.

The functions of given and new link up in turn with the functionsin transitivity. It was noted earlier (see 3 above) that a number of

participant roles may be expressed in either of two ways, either directlyor through the mediation of a preposition, for example the Beneficiaryin:

(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 butdiffer in information. Typically, the prepositional form of the Benefici-

ary is associated with the function New, the other form with thefunction Given; and if we assume here the expected intonation pattern,then in (34ii) Oliver is New and the tie is Given, the implied questionbeing ‘who did you offer the tie to?’, while in (34i) a tie is New andOliver is Given, answering ‘what have you offered to Oliver?’ (notethat 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 information-ally distinct forms, one with a preposition and one without, for expressing participant roles. The textual function of language requiresthat, for effective communication, new information should be madegrammatically explicit. New lexical content has to be backed up, as it

were, by adequate quanta of grammar; specifically, it has to be madeclear what is the ideational function of any new material in thediscourse, and here it is the preposition that indicates the role of the

unfamiliar element. The use of a preposition to specify function in theclause in just those cases where the element in question is typicallyNew (compare the use of by with the Actor in a passive construction)illustrates how the “texture” of discourse is achieved through theinterplay of varied grammatical resources expressing different facets of the total meaning.




194

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”): ideational2. modal Subject (“grammatical subject”): interpersonal3. Theme (“psychological subject

1”): 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 modalSubject 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 inpredication” and the rest predicate; but the subject in this sense wouldbe identical with the modal subject. The subject-predicate structure isentirely derivable from mood, and has no independent significance (cf.Fillmore 1968: 17; and Fillmore’s reference, ibid., to Tesniere 1959:103–5). As a form of generalization, it may be useful in that it expressesthe fact that Actor, modal Subject and Theme are regularly associated;but it obscures the equally important fact that they are distinct andindependent 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 functionsenables us to relate the internal patterns of language – its underlyingoptions, and their realization in structure – to the demands that aremade on language in the actual situations in which it is used. Asperformers and receivers, we simultaneously both communicatethrough language and interact through language; and, as a necessarycondition for both of these, we create and recognize discourse (the




195

textual function is thus instrumental to the other two). A speech act isessentially a complex behaviour pattern which in most instances com-bines the ideational and interpersonal functions, in varying degrees of

prominence. These very general notions in turn encompass a broadrange of more specific patterns relating to the creative and the repetitiveaspects of language in use.



196

Chapter Eight

MODES OF M EANING AND MODES

OF EXPRESSION: TYPES OF

GRAMMATICAL STRUCTURE AND THEIR

D ETERMINATION BY DIFFERENT

S EMANTIC FUNCTIONS (1979)

1 Preamble

Let us say that a code is a system of signs having a content and anexpression: 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 twoor more realizational cycles in it, so that the expression of content

1 (call

it expression1) is itself a content (content

2) 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 = content

2

expression2 . . .

A semiotic, in other words, is a stratified, or stratal, system, in whichthe 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.






198

Hence when one organizational system is represented in terms of another there will be mismatches of various kinds: what Sydney Lambcalled “interlocking diversifications” in the realizational process. This

means that we cannot simply operate with a schema of definitions andsay, for example, that the elements of this system are defined “fromabove”; it is not possible ever to derive the structure fully from

statements of this kind. The description of any part of the systeminvolves an interpretation of all three sets of relations into which itenters, “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 generaliza-tions. Whatever is said in interpretation of one level has implicationsnot 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 anatural language is organized into a small number of distinct compo-

nents, different kinds of meaning potential that relate to the mostgeneral functions that language has evolved to serve. Here are theheadings we shall use:

IDEATIONAL INTERPERSONAL TEXTUAL

EXPERIENTIAL LOGICAL

The first of these is language as representation: the semantic systemas expression of experience, including both experience of what is roundabout us in the outside world and experience of the world of conscious-ness 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, animateand inanimate objects, institutions, abstractions) and circumstantialfeatures (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



m ode s of m ea ning a nd m ode s of e xpre ss ion

199

formal logic are ultimately derived. These two, the logical and theexperiential, together make up the ideational component in the seman-tic system: that of meaning in the reflective mode.

The second main component, the interpersonal, is language asinteraction: it is meaning in the active mode. Here the semantic systemexpresses the speaker’s intrusion in the speech event: his attitudes,

evaluations and judgements; his expectations and demands; and thenature of the exchange as he is setting it up – the role that he is takingon himself in the communication process, and the role, or rather therole choice, that he is assigning to the hearer. This component is

therefore both speaker- and hearer-oriented; it is interpersonal – whatHymes called “socio-expressive” – and represents the speaker’s ownintrusion into the speech situation.

All discourse involves an ongoing simultaneous selection of meaningsfrom both these components, which are mapped into a single outputin the realization process. But there is also a third component, whichwe are calling the textual, whereby the meanings of the other twokinds take on relevance to some real context. Here the semantic system

enables the speaker to structure meaning as text, organizing eachelement 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 islanguage as action (the speaker as intruder in reality), the textualcomponent is language as relevance (the speaker as relating to theportion of reality that constitutes the speech situation, the contextwithin which meanings are being exchanged). The textual componentprovides 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 thingthat is cut and the one that the thing is cut for; also the place of cut inthe taxonomy of actions and of slice in the taxonomy of things. Theinterpersonal meaning is a demand for goods-and-services, “I want youto do something for me”, embodied in the selection of the imperativemood, direct, explicit and without any special modulation. The textualmeaning is the internal organization of this as a message with the focuson what is demanded, together with its relation to the preceding text




200

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

3 Above and below the semantic system

Since this functional interpretation is a generalization about language itcan be examined from all three angles of approach: from above, frombelow and from its own level. Presumably, if it is valid, it has someimplications 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 semanticcomponents typically generates a different kind of structuralmechanism as its output, or realization; and that (ii) thesedifferent types of structure are non-arbitrarily related to thekinds 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 withits condition of entry, these functional components appear as relativelyindependent sets of semantic options.

Within each component, the networks show a high degree of internal constraint: that is, of interdependence among the variousoptions involved. The selections made by the speaker at one point tendto determine, and be determined by, the selections he makes at another.For example, within the interpersonal component, there is a highdegree of interdependence of this kind between the systems of moodand 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 isin 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, andlittle 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 alwayscontribute his own judgement of probability no matter what the natureof the process he is talking about or what participants are associatedwith it.

Hence the categories of ideational, interpersonal and textual appear




201

clearly in the semantic system itself, as system networks each having ahigh degree of internal dependence but a very low degree of externaldependence. Choices made within one component have a great deal of

effect on other choices within the same component but hardly anyeffect 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 thesefunctional components have considerable significance for the way inwhich 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 totake place at all, it is necessary for those who are interacting to be ableto make intelligent and informed guesses about what kinds of meaningsare likely to be exchanged. They do this on the basis of their interpretation of the significance – the semiotic structure – of thesituation.

Let us postulate that the relevant features of a situation in whichlanguage has some place are the field of social process, the tenor of

social relationships and the mode of discourse itself: that is, (1) what isgoing 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 compo-

nents of the situation and the functional components of the semanticsystem. It appears that, by and large, the field – the nature of the socialactivity – determines the ideational meanings; the tenor – the socialstatuses and roles of the participants in the situation – determines theinterpersonal meanings; while the mode – the part assigned to thelinguistic interaction in the total situation – determines the textualmeanings.

In the example in Section 2, the activity going on is that of having ameal; the Walrus and the Carpenter are dining off the oysters, accom-panied 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 instructionfrom one to the other. The text is language-in-action, directed towardsfurthering the activity in question; hence selection of the textualmeanings makes exophoric (situational) reference to the processes andthe objects involved, as well as internal reference to an earlier occur-rence a loaf of bread through the collocational potential of slice .

Hence the categories of ideational, interpersonal and textual appear




202

to have implications for what is ‘above’, in that they represent differentcomponents of the meaning potential which are activated by differentcomponents of the social context (cf. Halliday 1977). This appears to

be the basis on which interactants make predictions about the meaningsthat will typically be exchanged in any particular situation with whichthey 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 structuralmechanism are typically involved in the realization of these variouscomponents of meaning. The suggestion will be that here too the same

categories are relevant, since they tend to be expressed throughfundamentally different types of structural organization (cf. Mathesius1964: 24–5 of Czech original).2

Let us consider the experiential function first. Here we are con-cerned with the semantic (linguistic) encoding of experience; particu-

larly our experience of the processes of the external world and of theinternal world of our own consciousnesses. We tend to encode suchexperience 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 specializedparticipants and circumstantial elements.

For example, suppose there is a flock of birds flying overhead. Werepresent this in language as something like There are birds flying : that is,a process of “flying” and, separated out from this, an entity that isdoing 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 thephenomenon as a single unanalysed process, not as a process plus aparticipant; 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 flyingbeing done by other things than birds. Some languages feel the same

about rain: in Chinese one says, liberally translated, “There’s rainfalling”; and in one south Chinese dialect, a variety of Cantonese, thereare 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 willalways be encoded as just this or that particular configuration of




203

elements. But we can formulate the general principle that this is howprocesses 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 func-

tionally labelled constituent structure as shown, for example, in Figure3.

Figure 3

There is no particular reason at this stage why the representation shouldhave 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 representthis as in Figure 4. If ordering is to be introduced into the representa-tion, there is the possibility of using

Figure 4

a dependency construct having the Process at the centre, as in Figure5. But a more appropriate ordering would have a nucleus consisting of Process plus Goal, with the other elements clustering around it, as inFigure 6. (We will leave aside here the question of the appropriateness

of the functional labels themselves.)

Figure 5 Figure 6




204

Each one of these elements may have a substructure of the samegeneral type, with different labels as, for example, in Figure 7. Thisagain 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’sclassic mode of representation of constituent structure in the form of

trees, or labelled bracketings. But the appropriate bracketing for func-tionally labelled structures is minimal, not maximal (“many ICs” rather than “few ICs”, in the terms of Hudson 1967). Maximal bracketingimposes too much structure for a functional grammar: for example, in

four young oysters it is reasonable to recognize young oysters as aconstituent provided the labelled elements are classes, since it is a

nominal group, but in no way does this correspond to any meaningfulfunctional constituent. The nonlinear representation implies more of amolecular model of structure, with a taxonomy similar to cell: mol-ecule: atom: subatomic particle.

Experiential meanings are typically realized as elemental structures of this type. The basic structural mechanism is that of constituency, withlarger 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 finaloutput – the syntagm – that serves as input to the next realizationalcycle. So let us say that experiential meanings are realized throughsome kind of constituent structure.

This expresses the particular way in which we order our experienceof reality when we want to turn it into meaning. The bounded entities

that enter into constituent structures with specific functions like Pro-cess, Actor, Goal, Extent or Manner offer a presentation of reality interms of “things” – doings by, and happenings to, persons and objects,in the environment of other persons and objects, with yet other personsand objects, and also times and places and so on, as attendant circum-stances; and including various “metathings” (facts and reports), whichare complex things that have already been encoded in language and so

acquired a status which enables them to participate in certain types of




205

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

5 The interpersonal mode

Now let us consider interpersonal meanings. These are expressedthrough very different structural devices.

Think of an example such as the following, uttered when somebodyis sick: I wonder if perhaps it might be measles, might it d’you think? The

experiential meaning is: Attribute “measles” plus Time “now”. Withthis the speaker has combined an interpersonal meaning of “I consider it possible”, together with an invitation to the hearer to confirm theassessment.

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 theproposition.

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, wonder-ment, 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 mappedon to the clause is a whole, running through from beginning to end.

We shall refer to this mode of realization as “prosodic”, since themeaning is distributed like a prosody throughout a continuous stretch

of discourse (cf. Mitchell 1958). It is characteristic of interpersonalmeanings that they are expressed in this prosodic fashion. Mood andmodality, tone and key, intensity and other attitudinal meanings aretypically realized through this kind of structural pattern. Swearwords

and obscenities, also, may occur at any or all points in the clause; itdoes not matter what segments they are attached to – many writershave noted that such elements readily occur even in the middle of aword. 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.




206

It is not difficult to see the rationale behind this mode of realization.The interpersonal component of meaning is the speaker’s ongoingintrusion into the speech situation. It is his perspective on the exchange,

his assigning and acting out of roles. Interpersonal meanings cannoteasily be expressed as configurations of discrete elements. They may beattached, as connotations, to particular lexical items, like bastards above

meaning “people” plus “I’m worked up”; but connotations do notenter into constituent-like structural relations. The essence of themeaning potential of this part of the semantic system is that most of theoptions 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 beinterpolated at more or less any point in the clause; and even whentwo or more such elements are present at the same time, they still donot go together to form constructions.

It is much more difficult to know how to represent prosodicstructures in a description of language. Usually they are either ignoredor forced to fit into the constituency mould. It may be more effectiveto treat them like prosodies in phonology, that is as contrasting features

having no place in the constituent structure (which is, after all, anexperiential structure) but which are specified separately and then

mapped on to the constituent structure as a distinct step in therealizational process.

6 The textual mode

The “textual”, or text-forming, resources in the semantic systemgenerate 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 WH-question, 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 associatedwith 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. Tobe 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 thetonic accent, which is the dynamic centre of the pitch contour, the




207

place where the greatest pitch movement takes place. (This may be afalling movement or a rising movement or some kind of complexmovement, 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 likelyform 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; itsignals the climax of what is new in the message. This kind of focalprominence can be assigned at any point in the clause; it is not realizedby final position, in the way that thematic prominence is realized byinitial 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 theTheme; and another peak of prominence, usually at the end, which isthe focus of information or, simply, the New. The two are different inmeaning. 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 thisexplicit, by starting ‘as far as . . . is concerned’. The New is hearer-oriented (though still, of course, selected by the speaker); it is thespeaker’s presentation of information as in part already recoverable tothe hearer (the Given) and in part not recoverable (the New). Thesetwo types of prominence are independent of each other. But bothcontribute 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 hasits 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 torepresent both thematic and focal prominence as constituent-likestructures of the clause, by recognizing the functional significance of the non-prominent part. So Theme contrasts with Rheme, and Newcontrasts with Given, as in Figure 9. In fact, the information unit is notalways coextensive with the clause; to return to the Walrus and theCarpenter, in The moon was shining sulkily (following the earlier The sunwas shining on the sea, shining with all his might )




208

Figure 9

we should presumably have the structure shown in Figure 10. Butbecause a clause is typically one information (focal) unit, and the focusin the information unit is typically at the end, final position in the

clause carries a potential for prominence which highlights it in thesame way that initial position is highlighted by prominence of athematic kind.

Figure 10

The structures that realize options in the textual component arewhat we may call “culminative” structures. They are not configura-tions or clusters of elements such as we find in the ideational compo-nent; nor are they prosodic chains of the interpersonal kind. What thetextual component does is to express the particular semantic status of elements in the discourse by assigning them to the boundaries; thisgives special significance to “coming first” and “coming last”, and somarks off units of the message as extending from one peak of promi-nence 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, thematicprominence, and ends with prominence of another kind, prominence

due to information focus. The latter is expressed through the assign-ment of the tonic accent to a particular place in the tone group; so theprominence is also in part phonological – and can be heard. Theperiodicity is further reinforced by the use of conjunctives to link onesentence with another; these contribute to the texture by relating aclause cohesively to what went before it, and they also occur at theboundaries – usually the beginning, but sometimes, especially in casualspeech, at the end.




209

Perhaps the clearest instance of the periodicity of texture is to befound in poetic forms. The metric regularity of the structures that haveevolved in poetry – lines with a fixed number of feet, and stanzas or

genres with a fixed number of lines – expresses in symbolic fashion theregularity of the structure of discourse as an exchange of messages. Thisis not to suggest, of course, that the structural unit of any particular

genre, such as the iambic pentameter line, functions directly as therealization 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), cuttingacross that of the metric form. But the impact of this tension on thereader, and especially on the listener, is one of the clearest indicationsof 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 thesestructural types with one of the functional semantic components, weare talking of a tendency not a rule. Experiential structures tend to bemore elemental in character, interpersonal structures tend to be pro-sodic 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 thateach should take this form – may differ very considerably from onelanguage to another.

Figure 11

Given structures of these very general kinds, it is clear that each canbe reduced to some form of constituency; but not all with the samesuccess. Experiential structures are quite constituent-like; whereas inter-personal ones are not, and the attempt to represent them in constitu-ency terms involves idealizing them to an extent that is tantamount to




210

a form of reductionism. An example of this is the attempt to reduce anintonation contour to a sequence of pitch phonemes (which are thenattached to specific places in a string).

Figures 12 and 13 represent an English clause first in non-constituencyterms 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-philosophicaltradition, 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) inter-personal : clause as interaction; (3) textual : clause as message

Figure 12




211

Figure 13

American structuralist and transformationalist theories, for example. Bycontrast, linguists in the socio-anthropological tradition, like Firth, whoare 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, havedeveloped models of a periodic kind: the structure of the paragraph(topic sentences, etc.), generic structures of various kinds and of coursethe whole theory of metrics.

It is interesting to recall here Pike’s (1959) important insight intolanguage as particle, wave and field. Although Pike did not conceiveof these in quite the same way, it seems very clear that this is what wehave here:

constituent (experiential) structures are particulateprosodic (interpersonal) structures are field-likeperiodic (textual) structures are wave-like

8 The logical mode

There is one functional component that we have omitted to take intoconsideration so far, and that is the logical. This is perhaps the mostdifficult to interpret.

As far as its origin at a higher level is concerned, the logical is asubcategory of the ideational, since it is language in the representationof reality. But there are two distinct modes of representing reality insemantic terms. In the experiential mode, reality is represented moreconcretely, in the form of constructs whose elements make some




212

reference to things. The linguistic structures actually stand as meta-phors for the relations between things; and the elements that enter intoand 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 andsulky.

In the logical mode, reality is represented in more abstract terms, inthe form of abstract relations which are independent of and make noreference to things. No doubt these relations, which taken all together

constitute what we might call the logic of natural languages, haveevolved by a process of generalization out of relations between things;and some of them, for instance “and”, are not hard to interpret inconcrete terms (one can lay a set of objects side by side). But unlikeexperiential structures, logical structures present themselves in thesemantic 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, isthe question whether languages differ as to what relations they are

going to treat as logical. It seems to me that they do, although thisargument will depend on our being willing to accept evidence “frombelow” – 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 fromthe logical component. Since we are claiming these structural manifes-tations are only tendencies, such an argument is only tenable on thegrounds that the type of structure that is generated by the logicalcomponent is in fact significantly different from all the other three.

The principle is easy to state: logical structures are recursive. But weimmediately encounter a difficulty here, a difficulty that is associatedwith 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 anelement of structure which is a point of entry into the same (or somehigher rank) network. A typical instance of rankshift is nominalization,where a function in the structure of a clause may be filled, not by anominal group (the congruent form) but by something that itself hasthe structure of a clause, for example to come and spoil the fun and That you have wronged me in:




213

It’s very rude of him to come and spoil the funThat you have wronged me doth appear in this

Another example of rankshift is a restrictive relative clause, for examplethe lights went out in:

The day the lights went out

These are not true recursive structures. The recursion-like effect thatis 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 rankin the constituent hierarchy. Clearly, this effect may appear more than

once, as in a “house that Jack built” routine; but it is strictly a non-event – there is no function involved that we could identify as arecursive function.

True recursion arises when there is a recursive option in thenetwork, of the form shown in Figure 14, where A:x,y,z . . . n is anysystem and B is the option ‘stop or go round again’. This I have called“linear recursion”; it generates lineally recursive structures of the forma1

+ a2

+ a3

+ . . . (not necessarily sequential):

Figure 14

These are of two kinds: paratactic and hypotactic. The paratacticinvolve relations like “and” and “equals”, which are logically transitive(A ‘&’ B, and B ‘&’ C, implies A ‘&’ C; A ‘=’ B, and B ‘=’ C, impliesA ‘=’ C). The hypotactic are logically non-transitive; these includerelations 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 inwhich it can be expressed – no way, that is, in which a sequence Aand B and C could realize an order “A & C & B”. The only departurefrom this strict sequence is nesting, where a sequence A and B and C and D represents an order, say, “A & (B & C) & D”. These aretypically signalled in the phonological structure, e.g.:




214

Table 1

// 13 soup // 3ˆ/ ham and / eggs // 3^/ apple/pie // 1^ and / teaor / coffee //i.e. A ‘and’ (B ‘and’ C) ‘and’ D ‘and’ (E ‘and’ F)

In hypotactic structures there is no such restriction; recursive order isnot always signalled by the sequence, which may reverse it ( ß beforea) or modify it in other ways ( ß inside a, for example).

In English the principal instances of ‘recursive structures’ (i.e. struc-tures deriving from recursive options) are as set out in Table 1.

We should distinguish structures of this kind from those whichhappen to be realized as lists, like counting. Counting 1 2 3 4 5 6 . . .is not a recursive operation, linguistically; it is simply the enumerationof a list, like Sunday Monday Tuesday Wednesday Thursday FridaySaturday, but one which happens to be endless because there is a (non-linguistic) recursive mechanism for generating the items in the list. In




215

the same way, the days of the week do not form a closed system, inthe linguistic sense; they form a lexical set whose members happen tobe 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 areproperties 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 generatecomplexes (clause complexes, group complexes, etc.) (Huddleston

1965). The apparent exception is the sentence, which is generated bylogical structures; but this is merely a terminological exception – thesentence is, in fact, simply a name for the clause complex. While thepoint of origin of a non-recursive structure is a particular rank – eachone is a structure “of” the clause, or of the group, etc. – recursivestructures are in principle rank-free: coordination, apposition, subcate-gorization are possible at all ranks. The more restricted ones, like tenseand report, are also the ones that are nearer the borderline; they are

only just logical structures. Tense is particularly interesting because ithas 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 andthe same kind of relation may be coded in the semantic system in morethan one way, i.e. may be realized through more than one of thefunctional components. The “and” relation, for example, may be codedin 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 onthe other hand they can also be coded experientially. Table 2 showsdifferent 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 rankof the clause that they are most clearly in evidence.

As an experiential construct, the clause is the locus of transitivity: itis the representation of the processes, participants and circumstancesthat constitute our experience of the real world. As an interpersonalconstruct it is the locus of mood and modality: the speaker’s adoptionand assignment of speech roles and his judgement of the validity of the




216

Table 2

Functional mode Example

logical : paratactic He sang, then people applauded, then . . .logical : hypotactic After he had sung, people applauded

Textual (First) he sang. Afterwards, people applauded

Experiential 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 thespeaker’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, characteristicin terms of the kinds of structure we have been talking about. Theclause is orchestrated as melody (the experiential component, constel-lations 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 isstructurally neutral, not based on the concept of constituent structureas the norm to which all grammatical patterns are expected to conform.In a systemic interpretation, language is treated as a resource, apotential; the various kinds of structure are the different means of expression of this potential. “Structure” is then no longer the basicorganizing concept; instead, structural representations are derived from

a more abstract conceptual framework that is paradigmatic rather thansyntagmatic, according to which the semantic system of a language iscomprised of sets of options in meaning, each of which makes somecontribution 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 interpersonaland a textual; with the ideational further subdivided into an experientialand a logical.

This pattern appears clearly at the semantic level itself: within eachcomponent, the networks of systemic options are closely inter-connected, whereas between one component and another there are




217

relatively few connections. In other words, choices made in onecomponent affect other choices within the same component but hardlyat all affect the choices in other components.

It also appears clearly at a higher level, in the relation betweenlanguage and situation. Broadly speaking, ideational meanings reflectthe field of social action, interpersonal meanings reflect the tenor of

social relationships and textual meanings reflect the mode of operationof 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 constituent-like structures, actually constellations of elements such as can be fairlyeasily represented in constituency terms. Interpersonal options generateprosodic structures, extending over long stretches (for example inton-ation contours), which are much less constituent-like. Textual optionsgenerate culminative structures, elements occurring at the boundariesof significant units, and give a kind of periodicity to the text, which ispart of what we recognize as “texture”. Logical options generate

recursive structures, paratactic and hypotactic, which differ from all theother 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 typesof structure function as alternative modes of the realization of systemicoptions. 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 derivedcomponents 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 Grober, . . . [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




218

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.)



219

Chapter Nine

T EXT S EMANTICS AND CLAUSE GRAMMAR:

HOW IS A T EXT L IKE A CLAUSE? (1981)

1 Patterns of wording in the clause

Thanks to the work of our predecessors, and especially perhaps to thatof outstanding figures of mid-century linguistics such as Sapir and

Whorf and Bloomfield and Firth and Hjelmslev, linguists of subsequentdecades 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 thatwe 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 meanskeeping all these things in focus at once: we are trained to do this bothas observers, when we listen simultaneously to the meanings, thewordings and the sounds of speech, and as theorists, when we constructrepresentations of language as simultaneously semantics, lexicogrammar and phonology. But it is not always easy to maintain this multiplefocus, because each shift of attention involves a shift in two directions,a knight’s move that is a move both upwards and outwards. This sametwo-dimensional relationship was described by Hockett many years agoin 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.




220

1961), in which he discussed the different statuses of morpheme andphoneme 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 secondrelationship, that of realization, or coding, that is the critical one; thesize distinction is typically associated with it, but not obligatorily – there

can be a clause encoded as a single syllable. It took a surprisingly longtime to clarify this two-dimensional relationship, and to accept that therelation of syllable to clause (or of phoneme to morpheme) was notsimply 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 dimensionsseparates the clause from the text. A text is likewise – typically but notnecessarily – bigger than a clause. But it is also, and more importantly,of a different level of abstraction. A text is a semantic entity rather thana 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 besuch 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 actuallyto generate any. Another is that once suggested by Peter Wexler whenhe proposed to introduce a talk with the words “This paper is aboutthe language of this paper”, so making the same entity serve both astext and as metatext. I have usually approached the problem in adifferent way, by using text that is so familiar that the audience cansupply the missing parts for themselves, like Mother Goose or Alice in

Wonderland . We need to face up to it somehow or other; there issomething discouraging about a publication where the author is insist-ing on the importance of context but cites nothing longer than adecontextualized clause.

When text is discussed in this way, with reference made only toisolated 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 thingas a clause only bigger, we can reasonably use clauses as instances whilemaking observations about text. And perhaps this approach to text inturn reflects a presemantic view of language, in which it is assumedthat the linguistic system is no more than grammar and phonology; soa text must be a grammatical unit, something that consists of clauses inthe same way that a clause consists of words.

Since there is no line of arbitrariness between semantics and gram-



text semantics and clause grammar

221

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. Therelations between the parts of a text are not such that we can set up

structures whose exponents will be clause-like entities. The elementsof structure of the text are more abstract; they are functional entitiesrelating 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 natureof a text if we treat a text as if it was a macrosentence, just as it wasnot easy to explain the nature of a sentence when a sentence wastreated as if it was a macrophoneme.

I am saying this at some length (despite the fact that I have said itoften 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 insistedthat a text is not like a clause, I now intend to suggest that it is. It is notthat I have changed my mind on the issue. The point is rather that, oncewe have established that texts and clauses are of different natures, theone 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. Startingfrom Hockett’s diagram, where one axis stands for constituency and theother for realization, we can link text to clause along the diagonal; therelationship between them is there but it is an oblique one, and thisdetermines 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 theclauses are simply parts. So when we speak of the problem of relatingclause 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 onelevel of abstraction beyond the clause. The relationship is not so muchone 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 textare made manifest. For example, the notion that a text has “structure”




222

would imply, if a text was a lexicogrammatical entity, that the elementsof structure would be “filled” by classes of the clause (perhaps with someintermediate 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 textstructures in a way which represents the text simply as a higher-rankgrammatical constituent; the configurations are different in kind, and

the relationship to the wording is both indirect and complex. Functionalelements 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 likea 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 relationof another kind: an analogic or metaphorical similarity. A clause standsas a kind of metaphor for a text. In this paper I shall refer to somewell-known properties of a text, and then, drawing on some recenttext-linguistic studies in a systemic-functional framework, try to showthat these are paralleled in significant ways by properties of a clausethat are in some sense (not always the same sense) analogous. The

textual properties to be considered are the following:

1. A text has structure.2. A text has coherence.3. A text has function.4. A text has development.5. A text has character.

1.1.1 A text has structure

For at least some registers, perhaps all, it is possible to state the structureof a text as a configuration of functions (Hasan 1979). A generalizedstructural representation is likely to include some elements that areobligatory and others that are optional; and the sequence in which theelements occur is likely to be partly determined and partly variable.

Most of the actual formulations of text structure that have been putforward seem to relate to one broad genre, that of narrative. Theoriginal source of inspiration for these was Propp’s theory of the folktale. The structure of traditional oral narrative has been investigated indetail within tagmemic and stratificational frameworks, on foundationsprovided by Longacre and Gleason. A well-known representation of another kind of narrative is Labov and Waletsky’s structural formulafor narratives of personal experience:




223

Abstract ^ Orientation ^ Complication • Evaluation • Resolution ^Coda

Outside narrative registers, Mitchell (1957) set up structures for the“language of buying and selling” in Cyrenaican Arabic, recognizingthree 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 ^ ConclusionAuctions: Opening ^ Investigation of object of sale ^ Bidding ^Conclusion

Mitchell refers to these as “stages” and comments that “stage is anabstract category and the numbering of stages does not necessarilyimply sequence in time”. Hasan considers that structure is a propertyof texts in all registers. For any register, specified at any appropriatedegree of delicacy, it should be possible to state a generalized structureby reference to which any actual text can be interpreted. Her suggestedformula for a particular class of transactions, retail sale in a personal

service food store, is as follows:

((Greeting •) Sale initiation ^) ((Sale inquiry •) (Sale request ^ Salecompliance) ^) Sale ^ Purchase ^ Purchase closure (^ Finis)

Martin (1980), who uses “functional tenor”, or rhetorical purpose, asthe superordinate concept for characterizing registers, gives the follow-ing 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 tobe defective in some way is that they do not “hang together”: they

“lack coherence”. A text has coherence; it forms a unity, a whole thatis 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 definedby Halliday and Hasan (1976). Cohesion is a necessary but not asufficient condition of coherence. The different types of cohesiverelation 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




224

coherent texture. These resources have to be organized and deployedin patterned ways.

(a) Ruqaiya Hasan identifies a feature which she calls “cohesive

harmony” (Hasan 1984). This is based on the recognition of cohesivechains 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 arecoreferential to a narrative participant; but Hasan points out that theyare 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 circumstantialelement; event, action or relation; fact or report; or any recoverableportion of text.

What makes a text coherent is not merely the presence of suchchains but their interaction one with another. In comparing textswhich were judged coherent, by herself and others, with those whichwere not, Hasan found that in the former it was always the case that a

significant majority of the tokens in each chain were functionallyrelated with tokens in some other chain; while in the latter the ones

that were related in this way were only a minority. Specifically, theywere related in some experiential function – transitivity, or an extensionthereof – 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 someother element as Process. In other words, in order to achieve coherencethere had to be not merely parallel currents of meaning runningthrough the text, but currents of meaning intermingling in a generalflow, 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 thedifference 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 theydo differ is in the proportion of such tokens that occur in interactionwith others from other chains; in other words, in the extent of cohesiveharmony displayed. This has proved to be a significant element indiscrimination between passages perceived as coherent and those wherecoherence is felt to be lacking.

(b) Another type of cohesive relationship identified by Halliday and




225

Hasan is that of conjunction, the linking of successive elements of atext 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 sen-tences. 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 inHalliday and Hasan, eliminating the category of “adversative” andgrouping “as against” with additive and “contrary to expectation” withconsequential (causal), is expressed in the network in Appendix 2,pp. 249–51. The category of “implicit”, also not in Halliday andHasan, accounts for those instances where the semantic relationship ispresent but there is no conjunction or textual (discourse) adjunctmaking it explicit. Martin’s analysis of a short passage of dialogue

includes instances of conjunctive relations bridging a number of inter-mediate turns (Appendix 2).

2. Conjunctive relations may be set up between passages of anyextent. Not only a turn in a dialogue, but an episode, argument, sceneor any other functional element may be “picked up” conjunctively ina succeeding portion of the text. In this way the presence of conjunc-tive relations creates coherence over extended passages of discourse. Anexample of this would be a section beginning Because of this, where thisrefers 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 sentenceperspective’ – tend to be determined by the “mode”, the functionascribed to the text in the given context of situation, the purpose it isintended to achieve. Thus the mode would determine the balanceamong the different types of cohesive resource – reference, ellipsis,

conjunction, lexical cohesion; and within conjunction, the relativeweight accorded to internal and external conjunctive relations and tothe various semantic alternatives within each. In this way the kind of conjunctive relations found in the text will be characteristic of theregister (as defined on the dimension of ‘mode’) to which the textbelongs.

An illustration of this principle is provided by Mary Ann Eiler (1979)in her study of expository writing by ninth graders in an American




226

high school. She has shown how the conjunctive relation of specificinstance to general principle, coded in Martin’s network as:

(internal / comparative) : similarity : nonexhaustiveis 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 cohesionbetween individual items – the repetition of respect . There is cohesionbetween Odysseus and heroic (men), with ‘Odysseus’ being a hyponymof ‘hero’. But between the two sentences as wholes there is a conjunct-ive relation – itself an extension of hyponymy – such that the secondone stands as a general principle of which the first one offers a specificinstance; 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 identifiablerhetorical function with reference to that context. This is the domainof functional theories of language, insofar as these are concerned withthe 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 textcan be interpreted as having one or other of a small set of “rhetorical”functions – exclusively, or predominantly, or in some recognizablecombination. 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 towardsspeaker, addressee and the rest of reality; to these Jakobson later addedthree more, having orientation towards the channel (“phatic” – inap-propriately), the message (“poetic”) and the code (“metalinguistic”).Britton, as an educational theorist, realigned Buhler’s categories so asto group conative with representational (both being “transactional”),and added the poetic as a fourth. In the work of Desmond Morris we




227

can even find an ethological categorization: “mood talking / groomingtalking / information talking / exploratory talking”.

All these apparently very divergent interpretations have in common

one basic distinction: that between language as reflection and languageas 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 discoursethat is oriented towards the function of interpersonal behavior (Mali-nowski’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 compo-

nent in the context of situation of the text. Halliday, McIntosh andStrevens (1964) proposed a tripartite framework for interpreting theregister: (i) the nature of the social process in which the text is embedded

– ‘what is going on’ (field); (ii) the interpersonal relationships amongthe participants – ‘who are taking part’ (tenor); (iii) the role assigned to

the text, including both medium and rhetorical function – ‘what partthe language is playing’ (mode). Gregory (1967) separated the rhetorical

function from the medium and associated it more closely with theparticipant relations, referring to the latter as “personal tenor” and torhetorical function as “functional tenor”. Ure and Ellis (1979) take thisone 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 functionaltenor, which he defines as “the purpose of the text”, as superordinateto field, mode and personal tenor. His argument is that “it is thefunctional 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 iscarried forward by the interaction of speaker and listener; obviously soin the case of dialogue, but so also in monologic modes where the activeparticipation of a listener still contributes to the construction of meaning.Even in written language the semogenic process is essentially of the samenature; researchers in writing theory now strongly insist on the partplayed by the imagined audience in the process of written composition.




228

Peter Fries refers (1981) to the “method of development” of a text.Below is one of his examples, with the relevant sections of the comment-ary (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 tothe 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 themethod of development of the text; (3) “theme–rheme” is clearly distinctfrom, but also clearly related to, “given–new”; (4) the theme–rhemeorganization of the clause “fits into a larger pattern governing the informa-

tion flow in sequences of sentences in English discourse in general”.Fries regards the theme as a “ground”: “In English discourse at leastthere seems to be a strong tendency to set up certain information as aground first, and then to introduce later information using that groundas a basis for evaluation and comparison”. (It is reasonable to understand“information” rather broadly here; presumably the ground may be anyconfiguration of ideational and interpersonal meanings.) So there is amovement from the ground, to something that is defined by it as “not

ground”. There is also, we may add, a complementary movement in atext, which is a movement towards rather than away from: a movement

to what we may call the “point” (generalizing from Fries’ “mainpoint”), from something that is defined by it as “not point”. Rhetoricaltheory 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 isnot a single movement. A text is a kind of diminuendo – crescendo,beginning and ending with prominence; but the prominence is of twodifferent kinds. Rather than thinking this time of a flow, a unidirec-tional current with a set of rapids at each end, we should perhapschange the metaphor to that of a gift, or rather an exchange, in whichthere is a shift of focus from donor to recipient in the course of theexchange, or rather from giving to receiving. The process begins asgiving and ends as receiving; but “giving – not giving” is not the same

movement as “not receiving – receiving”. Moreover although theprocess must start with the giving – until then there is no exchange – it need not necessarily end with the receiving, which may occur quiteearly and be followed up in various ways.

In the development of a text, phasing out the “ground” goes alongwith phasing in the “point”. This pattern is one that can be repeatedover as many levels in the hierarchy of constituents as the text has in it,from the entire text down to the individual clause.




229

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 configura-tion”, in Hasan’s terms. A text is also an individual entity, having aspecific character of its own distinct from that of other texts within thesame register. Some texts are highly valued as individual texts, and oneof the interests of text studies is the stylistic one, the attempt tounderstand 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 fromits context of situation. Taking the categories of field, tenor and modeas a predictive framework, Halliday proposed that the ideational mean-ings of a text tend to be determined by the field, the interpersonalmeanings by the tenor, and the textual meanings by the mode,suggesting that this was how listeners and readers make predictionsabout what is coming next – predictions that they must make if

meanings are to be successfully exchanged. For illustrations of this seeHalliday (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 afunction of its level of formality, the amount of self-monitoring doneby the speaker or writer; writing has a higher density than speech, withwhat she calls “language-in-action” having the lowest density of all.Lexical density can be defined as the number of lexical items per unitgrammar (per clause, as the most natural measure), though Ure meas-ures it as a percentage of running words; in her sample of 68 texts,comprising about 21,000 words each of speech and writing, the valuesrange 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 definesthe 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 aparticular “message type”. Melrose distinguishes factual, phenomenaland relational message, with various subcategories; deriving these fromthe 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.




230

According to Melrose, certain other features are associated withthese message types: different patterns of theme, conjunction and lexicalcohesion, and also different kinds of “message superordinate”, the

summative expression described by Winter (1977); the table in Appen-dix 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 predictablefrom the contextual configuration. This has sometimes been character-ized as deviation, the text creating its own rules; but “breaking therules” is a minor and relatively insignificant form of uniqueness. What

a text does is to create its own norms, its own unique selection fromthe 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 wethink of as literature. But any text can be described and interpreted asan event that is sui generis. If the qualities that we perceive as specific to

a text reside not merely in the particular combination of featuresselected but also in a special highlighting of some aspect of these, we

usually try to relate this highlighting to our interpretation of theunderlying theme, seeking the kind of semiotic convergence thatwould explain the particular impact that the text has on us.

A text is a polyphonic composition of ideational, interpersonal andtextual “voices”. The ideational voice provides the content: the things,facts and reports; processes, participants and circumstances; the logicalrelations of different kinds. The interpersonal voice provides theinteraction: mood, modality, person, polarity, attitude, comment, key.The textual voice provides the organization: thematic and informationalprominence; 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 dis-tinctively characterized by the foregrounding of modality. The firstpart is dominated by modalized assertions, which move from probabil-ity to obligation, the second part by assertions about obligations; thefinal speech shifts into narrative, returning at the end to the assertivemode 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




231

modalities determines a strong narrative line leading up to a no-nonsense 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 lexicogrammaticalobject, a structure of wording; whereas a text is a semantic object, astructure 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 theline of arbitrariness in language. In the realization of wording in sound,natural symbols are the exception. But there is no such line of arbitrar-

iness separating the clause from the text. The realization of meaning inwording is largely “natural”, non-arbitrary. This leads us to speculate

whether the text may display the same kind of multiple structuring thatis found in the clause, ideational, interpersonal and textual. The repres-entations of text structure proposed by Hasan and others suggest thatwe might want to interpret a text as having, potentially at least, anideational structure relating to its field and an interpersonal structurerelating to its tenor, rather than (or as well as) a single structure derivingfrom the mode (functional tenor) as proposed by Martin.

Benjamin N. Colby and Lore M. Colby (1980) analyse traditionaland other oral narratives in terms of “eidons”, which are ideationalelements of text structure set up to allow for the interpretation of thetext as an ethnographic document, as a window on the culture. Thetheory and method are set out in Colby’s study (1973) of Eskimo folktales. 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 eidosand ethos in cognitive and affective aspects of cultural behavior. Colby’sinterpretation of text structure is “eidological”, corresponding to theideational component in the structure of the clause; he suggests thepossibility of a parallel “ethological” interpretation corresponding tothe interpersonal component in the structure of the clause.




232

We should not press the analogy too far. But if it seemed useful toset up simultaneous structures in a text along these lines we might askwhether there is the same kind of structural variation as we find in the

clause, with the eidological structure being “particulate” (representedby definable segments of the text) and the ethological being “field-like” (represented by overlapping prosodies in the text). (The “wave-

like” 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 itsstructure, whereas coherence in a text largely depends on cohesion.Cohesion is the resource that takes over, as it were, when grammaticalstructure no longer holds (i.e. above the clause complex). We couldpoint out that cohesion also obtains within clauses; we find reference,substitution, ellipsis, conjunction and lexical cohesion all operatingbetween 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 befound with the notions of cohesive harmony and conjunctive relationsdiscussed 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 somekind of regular functional relationship with each other. But thesefunctional relationships are relationships within the clause; and thisreflects the fact that the elements in these chains themselves cannotoccur as isolated entities. Names have no place in language except in

function with other names; and the functions are defined within thestructure 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 thefollowing:

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.




233

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 invery 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 thatis achieved by the presence in the text of semantic relationships of theconjunctive kind is essentially a clause-type coherence, one that isbased on relations defined systematically within the transitivity system

of the clause.The notion that a text has function is again closely related to ananalogous feature of the clause, also one that is coded in the lexico-grammatical system: that a clause has a speech function, realized by themood system. The speech function of the clause – in simplest terms, asstatement, question, command, offer, or a minor speech function – isrepresented by the grammatical categories of declarative, interrogativeand 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 analysisof dialogue.)

The development of the text again has its analogy in the clause. Thishas already been made clear from the example cited in 1.1.4, sincePeter Fries used the theme–rheme structure of the clause as the sourcefrom which to derive the method of development of the text. We cangeneralize this still further by bringing in the notion of informationstructure, the given–new movement within the clause.

In its “textual” aspect, a clause has a wave-like periodic structurecreated by the tension between theme–rheme (where theme is theprominent element) and given–new (where new is the prominentelement); the result is a pattern of diminuendo–crescendo, with a peakof 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, wherethe theme is selected from what is given, and the news is put into therheme. But they can also be out of phase, and this gives an alternativepattern of texture to the clause. Putting the two out of phase meanslocating the new (the focus of information) somewhere other than atthe end of the rheme; this as it were changes the wave shape but doesnot disturb the essential periodicity.

This pattern is the “method of development” of the clause. It is




234

closely analogous to what takes place in a text; not only over the wholetext 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 tomain point), is one of the clearest manifestations of the analogy between

clause and text. It is in the clause that this movement is displayed inthe most systematic and clearly motivated form.

Finally a clause can be said to have “character” in both the genericand the specific sense. If a text is typified by virtue of its being

organized around the expression of processes of a particular type, theclause 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 subcat-egories. Thus analogous to the major types or genres of text are themajor types or classes of the clause, each being characterized by theselection 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 orientationtowards 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 orientedtowards a certain type of modality: interpersonal meanings are high-lighted, with the speaker’s skeptical doubting as the predominantrhetorical function. The passage cited is a unique interplay between theexploration of probabilities and the assertion of obligations, and so isthe entire text. No one clause can recapitulate the whole; but allcontribute, and some achieve a remarkable likeness – a likeness that ispossible because the systems of the clause not only embody all thesemantic components from which the text is built but do so in a waythat allows an infinitely varied, almost text-like balance among them.

Thus the properties that we recognize in a text are also, in atransformed way, properties that we recognize in a clause. A clause is akind of metaphor for a text – and a text for a clause. That this ispossible 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 isbetween clause and syllable. Hence it is not only in a formal sense thata text is like a clause. It is no accident that it is possible to illustrate so




235

many of the relations in a text by reference to relations in a clause. Theillustrations given here already contain within themselves a demon-stration of this conclusion. In showing that a text has structure,

coherence, function, development and character, we cannot help at thesame time showing that a clause has all these things too, though in aninteresting variety of different ways. Presumably this is how clauses

evolved – as the most efficient means of encoding text.

SUMMARY

TEXT CLAUSE

Structure: configuration of contextually motivatedsemantic functions

configuration of semantically grammaticalfunctions

Coherence: by cohesion(i) cohesive harmony:chains interrelate byfunction 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, asmajor or minor process

Function: has “functional tenor”(rhetorical function astext)

has “speech function”

(rhetorical function asspeech act)

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

has “informationstructure”:theme - - - - >- - - - > focus / new

{ }{ }

Character: generic: selects“favourite” process typeas message type

specific: foregrounds oneor more systems

generic: selects processtypespecific: foregrounds

selections from one or more systems

1.2 A functional interpretation

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




236

language. Functional theories of language came originally from outsidelinguistics; the consequence was that they were only theories of thetext – they had nothing to say about the system. According to such a

theory, any piece of text can be assigned a particular function, in thesense that it is oriented, exclusively or at least predominantly, towardssome communicative purpose. The unit that is described in this way

may be a very small piece of text, realized as one clause (the functionsare then “functions of the utterance”); or it may be a larger piececonstituting a recognizable semiotic event. The best-known functionalschemata are two dating from around 1930, the ethnographic one of

Malinowski (1935) and the psychological one of Karl Buhler (1934).Buhler’s scheme is interesting because although extralinguistic in intentit is one that is explicitly derived from language – that is, from thelinguistic system – in the first place: his tripartite framework of expressive, conative and representational functions denotes text that isoriented, respectively, towards speaker, addressee, and the rest of theuniverse – in other words the first person, second person and thirdperson categories of the Indo-European verb. This is similar to the way

in which various logical relations originally derived from naturallanguage 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 theywere, they would be of limited concern; but they are more than this – they are functions that are built in to language as the fundamentalorganizing principle of the linguistic system. We shall not be surprisedat this, if we take a Hjelmslevian view of language as system andprocess: if we accept that language and text are one and the same thing,and that the system evolved as a means of serving human intentionsthrough the creating of text. It is only if we set up artificial dichotomieslike langue and parole, or competence and performance, that we aresurprised when a system displays properties relating it to its use. Now,despite the divergencies that separate Buhler’s and Malinowski’s func-

tional theories, from each other and from various subsequent schemata,divergencies that are a natural consequence of the different purposesfor which they were devised (ethnographic, psychological, ethological,educational, etc.), there is one feature that strikes us as common to allof them. They all share in the fundamental opposition of action andreflection, the distinction between language as a means of doing andlanguage as a means of thinking. The former is Buhler’s first and secondperson function, Malinowski’s active function; the latter is Buhler’s




237

third person function, Malinowski’s narrative function. And the opposi-tion is incorporated into the semantics of natural languages, in the formof what I have referred to as the “interpersonal” and “ideational”

components. (The distinction between first and second personlanguage, however, is not a systematic one; the two are simply differentangles 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 andunderstood. Language has evolved to serve both these elementary

functions. The reflective mode is coded directly as the ideationalelement in the semantic system. But since language is symbolic, onewho speaks does not act on reality directly but only through theintermediary of a listener. Hence the active mode, when translated intoa 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 alanguage has to have a third semantic component, whereby it is enabled

to latch on to those situations in a systematic way. There must be arelevance function, a system of meaning potential which allows a text

to cohere with its environment, both the non-linguistic environmentand that part of the environment which consists of what has been saidbefore. So there is a third component in the semantics of naturallanguage which only an immanent linguistics will discover, since it hasno transcendent motivation; this is the contextualizing function – or the “textual” function, as I have called it, because it is what makes texttext, what enables language to be operational in culturally meaningfulenvironments.

Now a clause is a complex realization of all these three semanticfunctions. It has an ideational component, based on transitivity, theprocesses, participants and circumstantial elements that make up thesemantics of the real world, and including the onomastic systems thatclassify these into nameables of various kinds. It has an interpersonal

component, consisting of mood, modality, person, key and all thevarious attitudinal motifs that come to be organized as meaningfulalternatives. And it has a textual component, the “functional sentenceperspective” (thematic and news-giving systems) and the cohesiveresources of reference, ellipsis and conjunction. Each of these compon-ents makes its contribution to the total make-up of the clause. Whatwe identify as a clause is the joint product of functional-semanticprocesses of these three kinds.




238

But what is the nature of the contribution, in each case? We areaccustomed to thinking of this in structural terms: that each semanticcomponent generates its own particular tree, a configuration of parts

each having a distinct function with respect to the organic whole. Theideational component generates “actor–action–goal”-type structures:configurations of Process, Medium, Agent, Beneficiary, Range, Extent,

Location, Manner, Cause and so on. The interpersonal componentgenerates so-called “modal-propositional” structures: configurations of Subject, Finite, Modality, Predicator, Complement and Adjunct. Andthe textual component generates thematic and informational structures,

configurations of Theme and Rheme, and Given and New; as well ascohesive 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 constitu-ent structure is so simple, it is natural that a linguist should want to doas much as possible with it. And it can be made to do quite a lot. Butthere comes a point where it ceases to be appropriate; where moulding

the facts so that they fit the notion of constituency will distort themrather 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 con-cerned, these do tend to generate part–whole structures; they are realizedby organic configurations which themselves, and whose constituents,are reasonably clearly bounded, such that it can be specified where oneclause element leaves off and the next one starts. But this is not nearlyso true of interpersonal systems. Interpersonal systems tend to generateprosodic patterns that run all the way through the clause: not onlyintonation contours, though these provide a clear instance, but alsoreiterations of various kinds like those that are typical of modality inEnglish, 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 thesepeaks typically appear at the beginning or the end of the clause, wherethere is a sequence of clauses they result in a kind of periodicity, amovement from a clause-initial peak via an off-peak medial state to aclause-final peak which is then sustained to form the initial peak of the




239

succeeding clause. Thus a clause is at one and the same time particle,field and wave, as Pike suggested more than twenty years ago (Pike1959), 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 benon-arbitrary; this is clearly important when we come to ask whether such tendencies are found in every language. The grammar of languagesis a natural grammar; as I expressed it earlier, there is no line of

arbitrariness between semantics and grammar as there is betweengrammar and phonology. If the clause is at once particle, field andwave this is because the meanings it has to express have differentsemiotic contours, to which these three realizational forms correspondin a natural, non-arbitrary way.

The particular nature of ideational structures reflects the relativediscreteness 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 wordingreflects this fact: the word cow has an outline because the object cowhas an outline. Of course not all experience is like this; indeed I havealways tended to emphasize the unboundedness of many phenomena,the indeterminacy and the flux; and I share Whorf’s view that languageitself, once it has been constituted in this way, strongly influences theforms that our perceptions take. Nevertheless there is a basic fit betweenthe discreteness of words and the discreteness of things; otherwise weshould not be able to talk about the things at all, or explain contrastivelythose instances where the fit of words to things is less than perfect.

By contrast, the interpersonal kind of meaning is a motif that runsthroughout 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 andprobability; his choice of speech function, the mode of exchange indialogue – such things are not discrete elements that belong at someparticular juncture, but semantic features that inform continuousstretches of discourse. It is natural that they should be realized notsegmentally but prosodically, by structures (if that term is still appropri-ate) that are not particulate but field-like. The linguist’s tree is aninappropriate construct for representing structures of this kind.




240

Thirdly, the undulatory movement by which textual meanings areencoded in the English clause may again be in some sense a naturalform for their representation. All the patterns I have been discussing

vary from language to language, as is very obvious; those of Englishmerely provide one specific instance of something that seems to be ageneral tendency in the expression of meanings of each kind. The

English clause, as a message, is a movement from prominence toprominence, a diminuendo that is then picked up and becomes acrescendo; but the prominence is of two different kinds. It is amovement away from a Theme, to something that we can characterize

as non-Theme; that is the diminuendo aspect. It is a movement towardsa New, from something that we can characterize as non-New; that isthe crescendo aspect. But Theme is not the same as non-New, nor isNew the same as non-Theme; there are two movements here, notone. Their relationship is less automatic than the above formulationimplies, and they can be combined in other ways besides; what isdescribed here is just the unmarked, typical form. The essential point isthat the two types of prominence differ; and that they differ as speaker

to listener. The Theme is speaker-oriented prominence: it is “what Iam on about” (grammarians used to call it the psychological subject).

The New is hearer-oriented prominence: it is “what I present as newsto you”. The English clause is textured by this shift in its orientation,from speaker-prominence to listener-prominence. Each clause is in thissense a kind of gift, one move in an exchange, symbolized by thechange 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?, shebegins 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 hadasked 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 isbeing helpful, keeping the wave pattern of the dialogue in phase. Butshe need not do this; the Theme is the speaker’s choice, and in anycase 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 – ”




241

The Footman, just like the rest of us, favours himself as an unmarkedTheme.

So the patterns of wording in the clause, which is the basic unit of

lexicogrammatical organization, display a variation that derives fromthe different kinds of meaning they express; and the structural shape isin each case a natural product of the semantic functions. A functional

grammar is an interpretation of the primary semiotic purposes thatlanguage has evolved to serve, and of the different ways in whichmeanings relating to these different purposes tend to be encoded (andthe patterns just described are only tendencies). When we go on to

observe the developmental processes whereby young children constructtheir language, we gain a further insight into the steps by whichgrammar may have evolved on the way towards its present form.

2 From clause to text

Since the functions that we have called ideational, interpersonal andtextual are components of the semantic system, and since a text is a

semantic unit, it follows that these components will be present in thetext just as they are in the lexicogrammatical entities, the wordings, bywhich 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 thisis to say no more than that both derive from the linguistic system – apoint that is perhaps worth making, since text is still sometimes treatedas if it had no source of its existence in language, but is neverthelessnot saying a very great deal. The problem to be solved is how featuresfrom these semantic components are represented, on the one hand inclauses and on the other hand in texts, and with what kind of systematicrelationship between the two, such that the clause can function as theprincipal medium through which meanings of such different kinds, anddiffering 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 relationshipin size, and their relationship in abstraction. To go from text to clause

involves a move along the axes both of composition (constituency) andof realization. I shall consider the size dimension first.

(A) Do we find, extending over a whole text, patterns that are likethose we find in the clause? Let us take each of the three functionalcomponents in turn.

1. Ideational. Like a clause, a text has an ideational structure, with




242

something of the same particulate kind of organization: it is possible torecognize functional constituents of a text, always allowing (as in theclause) for some variation in sequence and a certain amount of

overlapping. These structural elements have been identified mostclearly, perhaps, in narrative; and the researches of Pike, Longacre andtheir co-workers on the one hand, and of Gleason and his colleagues

on the other, have provided a rich body of empirical findings aboutthe structure of narrative in languages and cultures from every contin-ent. 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 eachwith 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 recognizesoptional and obligatory elements, variations in sequence, recursiveoptions and the like, all of which make the text structure look rather similar to the ideational structure that is characteristic of the clause. Wecan sum this up by saying that, in at least some genres, and perhaps in

all, a text is a configuration of functional elements, collectively repres-enting 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 becoded in a great variety of different ways. They appear in many formswithin 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 generalizedsystem network and suggested how they may be accounted for as anaspect of the ideational structure of a text.

2. Interpersonal. A clause has an interpersonal pattern of organiza-tion, including a modal structure (mood, modality and key) whichexpresses 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 predominantin it: material including action, event, behaviour; mental, includingperception, reaction, cognition; verbal; or relational, including attribu-tive, equative, existential. This “type of process” is of course anideational category; but a clustering of processes of the same kindexpresses the rhetorical design of the text rather in the same way that a




243

particular complex of ideational features such as “I am certain” or “Iwant to know” functions as the interpersonal motif of a clause. Thebridging 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 selectionsin it tend to be mainly determined (Halliday 1977); the “field” isdefined 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 areopen to it to play (Appendix 4, pp. 255–7).

3. Textual. That discourse displays some kind of a wave form, withpeaks of prominence at both ends, has been a commonplace of

rhetorical theory ever since it was first hypothesized that a text has abeginning, a middle and an end. (This is perhaps one of the fewexamples of a verifiable hypothesis in linguistics, though characteristic-ally 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 suchas initiating, introductory or topic sentence referring to movementdownwards from a beginning, and terms such as culminating, summat-ive or focal sentence referring to movement upwards to an end. The

diminuendo–crescendo pattern we find in the clause is thus also presentin 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 thisnesting of wave-like structures one inside another is characteristic alsoof lexicogrammatical organization: among the constituents of the clausein English, endocentric word groups (verbal groups and nominalgroups) display this same kind of movement from speaker prominenceto 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 speaker-prominent deictic have , locating the process in speech time, to thelistener-prominent lexical item finish, saying what the process actuallyis; and the nominal group my three brilliant articles likewise goes from adeictic 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 non-New to New in the clause. So both text and clause can be seen toparticipate in a multilayered pattern of organization in which themovement has this same underlying periodicity repeated over structuresof differing extent.

So much for what we might call the metonymic aspect of therelation of clause to text. Now we turn to the metaphorical: where thefeature that we have identified in the clause is not being repeated on a




244

larger canvas, as in the instances just considered, but rather is standingas 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 thefeature which it serves to realize.

(B)1. Ideational. Cohesion, as defined by Halliday and Hasan(1976), is the semantic resource through which textual coherence isrealized. A text displays cohesion; and this cohesion is achieved bymeans of a variety of features of the clause, which serve to relate one

clause to others that constitute its context. However, while cohesion isa necessary condition of textual coherence, it is not by itself sufficientto guarantee it; and in her subsequent studies Ruqaiya Hasan (1984)has been comparing pairs of texts, of similar nature and origin, whereone is judged coherent and the other not, in order to establish whatare the differences between them. She has one set of texts which arestories told by children; she has also examined texts from schizophrenicpatients, including a pair of texts from one particular patient, one when

undergoing treatment and the other when the same patient was judgedas 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 systematicallyinterrelated: a majority of the occurrences in any one chain wererelated to occurrences in some other chain. They were systematicallyrelated, that is to say, in the ideational structure of the clause; for example as Agent to Process, or Attribute to Carrier, or by their bothhaving the same role with reference to some other element, such asboth being carriers of the same attribute. In the texts that were judgedto be non-coherent, on the other hand, although the proportion of lexicoreferential occurrences entering into cohesive chains was no lessthan in the coherent texts, only a minority of these occurrences were

cross-related in this way; in general, the recurring elements ran along-side each other through the text but without intermingling to anyextent. The coherence of the text appears to be the product of this“interchaining”. If a text is coherent there is a movement of relatedparticles through a succession of clauses, so that not only do theindividual particles persist from one clause to another, but the structuralconfigurations, though not remaining static, also preserve a recogniz-able continuity. Just as individual elements form a clause not as isolated




245

entities but as roles in a structural configuration, so chains of elementsform a text not as isolated chains but as role-chains in an ongoingconfigurational movement (Appendix 1, pp. 247–50).

2. Interpersonal. How does one recognize the unique rhetoricalflavour of a text? Partly at least from the overall pattern of interpersonalfeatures of the individual clauses. A text has its own character as an

intersubjective event, and this tenor of discourse is manifested primarilythrough the cumulative force of the options taken up in the interper-sonal systems of meaning. In Priestley’s play An Inspector Calls, theunderlying theme, or rather one of the underlying themes, is that of

social responsibility: we are all members of one body. This confersobligations on all of us, each one towards others. In the course of theplay these obligations are acted out – or rather the consequences of their not being fulfilled are acted out – through the step-by-stepuncovering of a chain of irresponsibility, compounded by sheer chanceand observed through a confusion of prejudices and doubts. Now, thethree 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 theunderlying semiotic of the play is worked out metaphorically, at a

critical point in the action, through the highlighting of modal selectionswithin the clause, backed up by lexical choices from the same semanticfields. The clauses in this key passage, each with its own smallmomentum, combine to produce a powerful semantic movement, amotif 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 sweptaway. As audience we respond to this movement even though theevents which call it forth are in themselves trivial, no more than anargument over the identification of a photograph. Here the interper-sonal features of the clause stand as a metaphor for the social semioticof the text, as an exploration of the complex symbolic structuresbinding 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 thetransitivity and modal features that predominate in the individualclauses. When we come to consider the rhetorical organization of atext, 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 whatare the elements that function predominantly as news. In his study of




246

the thematic organization of discourse, Fries (1981) has shown howthese patterns realize the development of the paragraph. Examining atightly constructed paragraph by Lytton Strachey, Fries found three

lexicosemantic chains, one having to do with the opposition of wisdomand chance, one with the English constitution, and one with politicalapparatus in general; of the three, the former was overwhelminglyassociated with initial position in the clause, the second with finalposition, while the third showed no particular pattern of distribution.Fries points out that this reflects the rhetorical interpretation of theparagraph as having the “wisdom versus chance” motif as its method

of development and the English constitution as its main point. Thusthe mode of discourse is manifested in the same cumulative manner bythe ongoing selections, in each clause, from the thematic and informa-tional systems, those comprising the “textual” element in the meaningpotential of the clause (Appendix 3, pp. 254–5).

I am not suggesting, of course, that listeners and readers process textin a conscious manner, parsing each clause as they go along. On thecontrary, speaking and understanding are, as Boas and Sapir always

insisted, among the most unconscious of all the processes of humanculture. The conscious task is that which falls to the linguist, when hetries to find out how text is organized. Listeners and readers makepredictions – they have a good idea of what to expect; if they did notmake these predictions, with a greater than chance probability of beingright, they would not be able to understand each other. It is theorganization of a text, and in particular the relation of a text, as asemantic unit, to a clause as the primary lexicogrammatical unit throughwhich it is realized, that makes such prediction possible. The linguisticanalysis of text is a necessary step in the interpretation of how meaningsare 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 realizationof a text as a whole, or some structurally significant portion of it, inthe indirect, metaphorical sense that these examples suggest. The

former is its automatic function, as determined by the system of thelanguage. The latter is what Mukar ˇovsky (1977) recognized as “deau-tomatization”: still, of course, part of the potential of the linguisticsystem but deployed in a metagrammatical way, conveying meaning bythe act of systemic choice instead of (in fact always as well as) by theact 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




247

does. What are clauses for? – to make it possible to create text. Aclause does this effectively because it has itself evolved by analogy withthe text as a model, and can thus represent the meanings of a text in a

rich variety of different ways.

Appendices

Appendix 1from Hasan (1980).

Text A

1. once upon a time there was a little girl2. and she went out for a walk3. and she saw a lovely little teddybear 4. and so she took it home5. and when she got home she washed it6. and when she took it to bed with her she cuddled it

7. and she fell straight to sleep8. and when she got up and ( ) combed it with a little wirebrush

the teddybear opened his eyes9. and ( ) started to speak to her

10. and she had the teddybear for many many weeks and years11. and so when the teddybear got dirty she used to wash it12. and every time she brushed it it used to say some new words

from a different country13. and that’s how she used to know how to speak English Scottish

and all the rest

Text B

1. the sailor goes on the ship2. and he’s coming home with a dog

3. and the dog wants the boy and the girl4. and they----- don’t know the bear’s in the chair

5. and the bear’s coming to go to sleep in it6. and they----- find the bear in the chair 7. they----- wake him up8. and (-----) chuck him out the room9. and (-----) take it to the zoo

10. the sailor takes his hat off




248

11. and the dog’s chased the bear out the room12. 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 onetoken, for example they----- referring to the girl, the boy, the sailor andthe 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 considerablygreater cohesive harmony.

When subjects were asked to judge the coherence of the two texts,Text A was consistently rated “more coherent” than Text B.



Chain Interaction – Text A

home

home

girlgirl

girlgirlgirl

girlgirlgirl

girlgirlgirlgirl

wentgot

tookhad

took-to-bedfell-to-sleepgot-up

washedcombedwashedbrushed

lovelydirty

teddybear teddybear

teddybear teddybear teddybear teddybear

teddybear teddybear

teddybear teddybear



Chain Interaction – Text B

gocome

go-to-sleepsleep

chuckchase

sailor sailor

bear bear

bear bear

bear bear

Each rectangle corresponds to one chain; subdivisions in the chain are inditems which are in a constant functional relation (shown by a double-hdifferent chain; for example in Text A, between girl (4) and the box contaaction relation; between the latter and teddybear (4), a relation of action–g




251

Appendix 2from James R. Martin (forthcoming)

1. B: Lips are a must.2. They’re in fashion.3. So . . . what are you using in your skin care?4. A: Oh I just – I don’t know.5. Something my mom gets: Ponds or something.6. B: Yes.7. A: I don’t know.

8. B: Well really uh that’s not good enough really.9. You want something that’s going to treat the skin.10. You need to cleanse your skin well11. uh to use a good toner 12. A: Hmm.13. B: and moisturiser is a must14. and of course then you can go into the make-up.

15. But if you do all these things

16. your skin will start to improve.17. A: Yeah.18. B: You’re finding a few little spots under your skin, aren’t you?

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




252

Figure 1 Conjunctive relations in Talking Shop (scene 21)




253

Figure 2




254

Appendix 3from Peter Fries (1981).

1. A. The English Constitution – that indescribable entity – is aliving thing, growing with the growth of men, and assumingever-varying forms in accordance with the subtle and com-plex 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 Englishgave it one of its essential peculiarities – the system of acabinet independent of the Crown and subordinate tothe Prime Minister.

5. C The wisdom of Lord Grey saved it from petrification andset 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 irre-sponsible administrative power – was about tobecome its predominant characteristic and changecompletely 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 itthree lexical systems; the first concerns living, growing, changing, the

second system concerns wisdom versus chance and the third systemconcerns concepts having to do with government. From reading theparagraph it is clear that the main point of the paragraph is that theEnglish constitution is living, growing and changing, that the paragraphis developed via the opposition between wisdom and chance and thatthe 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




255

changing typically occur within the rhemes of the component sentencesof 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 theform of government occur more or less equally within the themes andrhemes of the component sentences of the paragraph. Thus the

consistent placement of the terms of a lexical system inside or outsidethe themes of the component sentences of the paragraph affect theperceived 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) indicatesthe point of departure of the message expressed by that sentence, andb) the information contained within the themes of all the sentences of a paragraph creates the method of development of that paragraph. Thusif the themes of most of the sentences of a paragraph refer to onesemantic 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 simplemethod of development will be perceived.

Appendix 4From 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 strictlycontrolled, 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 profes-

sions, 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. Subordinatesenates of cities are controlled by the Supreme Senate. At the head of

the state is the General.

Totalitarianism, Leonard Schapiro (pp. 87–8)




256

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 andpyramidal

This is a particularly clear example of an attributive message. Itbegins with a summation, “every aspect of life”, with the general noun“aspect” acting as Head of a nominal group. It is this summation whichdetermines the Theme of the clauses that follow: thematic prominenceis 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 arelationship of hyponym to “machinery of government”, itself an aspectof life.

There is no explicit conjunction of interest : the chief conjunctiverelationship is an implicit one, of the internal additive type. Moreworthy of attention is the lexical string of Group A. Just as Theme was

determined by the summation, so the lexical string of Group A isdetermined by the non-finite clauses dependent on the clause of whichthe summation is an element. Together with the three verbs in thesenon-finite clauses, nine lexical items of the message proper constitute astring of synonyms, near-synonyms, and collocates – and of these nine,six function as Complement, and so form a kind of pattern in theRheme. Thus it may be seen that in this attributive message thesummation and the clause complex of which it is an element are both

closely related to the message proper that follows: the summation ishyperonymously 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 ismost often the case that “aspects of life” are encoded in the Subject,while “regulation” is realised in the Complement.




257

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 ismeronym

Appendix 5from M. A. K. Halliday (1982).

Mrs. Birling: I think we’ve just about come to an end of thiswretched 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 thatphotograph of her.

Inspector: No. It wasn’t necessary. And I thought it better notto.

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 atit.)

Inspector: (taking back the photograph) You recognize her?




258

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 publicman –

Inspector: (massively) Public men, Mr. Birling, haveresponsibilities as well as privileges.

Birling: Possibly. But you weren’t asked to come here to talkto 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 decentwages. I went and pushed her further out, right intothe 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 youdon’t recognize her from that photograph. I admit Idon’t know why you should, but I know jolly well

you did in fact recognize her, from the way youlooked. And if you’re not telling the truth, whyshould 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 areopinions relating to duties, and, as a minor motif, duties relating toopinions. The two themes are closely interwoven. We have alreadyseen that this is a projection into the text of a relation that existsbetween them in the system. The scales of “possible–certain” and“allowed–required” both typically combine with a common semantic




259

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 formof the realization of both.

The significance of the lexicogrammatical selections in the text canonly be fully revealed by a consideration of their value in the semanticsystem. Textually, the passage under discussion centres around the

scrutiny and recognition of a photograph. The words and structureswhich, in their automatic function as the “output” of semantic choices,carry forward the movement of the text, also become de-automatizedand 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’tbelieve she could have changed so much.

Clause 1 polarity positivemodality low / (indicative : probability) / (subjective :

congruent)

Clause 2 polarity negative : transferredmodality high /

(a) can (imperative : inclination) / (subjective :congruent)

(b) I . . . not believe (indicative / probability) /(subjective : explicit)

(c) could (indicative / probability) /

(subjective : congruent)



modal

indicative(MODALITY)

‘it is’probability‘it either is

or isn’t’

temporality‘it both isand isn’t’

categorical

categorical

positive

high

median

low

negative

certain

probable

possible

is

must be

will be

can be

isn’t

always

usually

sometimes

high: must ought to need has/had to subjective, congmedian: will would shall should subjective, expli

low: can could may might objective:



261

Chapter Ten

DIMENSIONS OF DISCOURSE ANALYSIS:

GRAMMAR (1985)

This chapter presents a brief sketch of a lexicogrammatical text inter-preter 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 isthe text in standard orthography and punctuation (Hasan 1965: 65):

Adult: Do you – when you have a small baby in the house, do youcall 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 youought to call it ‘it’, because you don’t know whether you’recalling 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’tstop 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 heor a she, if you refer to it as ‘it’ well then the neighbour willbe 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’sa boy or a girl?

First published in The Handbook of Discourse Analysis, Vol. 2: Dimensions of Discourse , 1985.London: Academic Press, pp. 29–56.




262

Tilly: She didn’t. I don’t suppose she knew.Elsie: Call it ‘the’.Lacey: Hello, The!

Elsie: Oh, I know. Call it ‘baby’.

Systemic grammar is an analysis–synthesis grammar based on theparadigmatic notion of choice. It is built on the work of Saussure,Malinowski and Firth, Hjelmslev, the Prague school, and the Americananthropological linguists Boas, Sapir, and Whorf; the main inspirationbeing 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 bechosen if the entry condition is satisfied. Options are realized assyntagmatic constructs or structures; a structure is a configuration of functional elements – functions or function bundles. The functions aremotivated (nonarbitrary) with respect to the options they realize; thegrammar as a whole is motivated with respect to the semantics. Theonly line of (relative) arbitrariness is that between content andexpression (between the lexicogrammar and the phonology).

A text in systemic-functional grammar is an instantiation of thesystem (in the Hjelmslevian sense of “the linguistic system”). (Notethat “instantiation” is not the same thing as “realization”; the twoconcepts seem to be confused in Saussure.) Text may be studied asprocess 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 fromwhich the text is derived, since that would involve representing largeportions of the grammar. Instead we employ structural notations, withbrief discussion of some of the options from which the structuralfunctions 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 rhythm2. analysis into clauses and clause complexes, showing interdepend-encies 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 thelatter for information (Given–New) structure

5. analysis of finite clauses for mood, showing Subject and Finite



dimensions of discourse analysis

263

6. analysis of all clauses for transitivity, showing process type andparticipant and circumstantial functions

7. analysis of groups and phrases (verbal group, nominal group,

adverbial group, prepositional phrase)8. analysis of grammatical and lexical cohesion9. identification, rewording and reanalysis of grammatical

metaphors10. 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 intonationand rhythm:

Figure 1




264

Figure 2 Systems and Notation for Intonation and Rhythm

2 Clauses and clause complexes

The text is analysed into clause complexes, showing the interdependen-cies and logical-semantic relations among their constituent (ranking,

nonembedded) clauses.

Clause complex 1 b ^ a ( 1 ^ +2 )

1.1 b ||| when you have a small baby in the house ||

1.2 1 || do you call it ||

1.3 +2 || or do you call it she or he ||

Figure 3a




265

Clause complex 2 a ( b ( a ^ ‘b ) ^ a ) ^ b ( 1 (a ^ ‘b ) ^ +2 ( b ( 1 ^ +2 ( 1 ^ 2 (a ^ ‘b ) ) ) ^ a ( a ^ b ) ) )

2.1 a b a ||| well if you don’t know ||

2.2 a b ‘b || what it is ||2.3 a a || I think you ought to call it it ||

2.4 b 1 a || because you don’t know ||

2.5 b 1 ‘b || whether you’re calling it a boy or a girl ||

2.6 b +2 b 1 || and if it gets on ||

2.7 b 2 b +2 1 || and if you start calling it she ||

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 ||

2.10 b 2 a a || you can’t stop yourself ||

2.11 b 2 a

b || ’cause you’ve got so used to calling it she |||Clause complex 3 a ^ ‘‘b ( b ( 1 ^ + ( a ^ ‘b ) ) ^ a ( b ^ a ) )

3.1 a ||| Mrs. Siddons says ||

3.2 ‘‘b b 1 || that if some neighbour has a new baby next door ||

3.3 b b +2 a || || and you don’t know ||

3.4 b b 2 1b || whether it’s a he or a she ||

3.5 b a b || if you refer to it as it ||

3.6 b a a || well then the neighbour will be very offended |||

Clause complex 4 1 ( b ^ a ) ^ 2 ( a ( a b ) ^ 1b )

4.1 1 b ||| well if it’s in your family ||4.2 1 a || I think you should call it either he or she ||

4.3 2 a a || or else the poor thing won’t know ||

4.4 2 b when it grows up 4.5 2 a ‘b || what it is |||

Clause complex 5 ‘‘b a ) a ( ^ b (1 ^ +2 ( a ^ 1b ) ) )

5.1 ‘‘b a ||| well what you should do ||

5.2 a did Mrs. Siddons suggest 5.3 b b 1 || if your neighbour has a baby ||

5.4 b b +2 a || and you don’t know ||

5.5 b b 2 ‘b || whether it’s a boy or a girl |||

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 |||

Figure 3b Clauses and Clause Complexes




266

The major portion of the text consists of five turns, each made upgrammatically 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-semanticrelation 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”; theremainder (8) are projection, of which 2 are saying (the teacher asSayer) and 6 are knowing, mainly negative and with a generalizedSenser “(if) you don’t know” – this being an aspect of the problem

under discussion. (The two instances of I think [new addition] aremetaphorical modalities, not projections, as can be shown by adding atag: 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 impeccablywell formed, as is typical of casual spontaneous speech (including thatof 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 paratacticelaboration, 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, acomparison 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 followingsystematic exceptions: (1) seven out of the eight projections have bothprojecting 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 unitsare mapped on to one clause. For the analysis of theme and informationstructure see the next two sections.




267

3 Theme

Each clause, and each clause complex characterized by rising depend-

ency (b ^ a ), is analysed for thematic structure:

Clause Theme

Markedness of Clause Textual Interpersonal Topical topical theme

Figure 4 Clause Theme




268

The Theme is the (speaker’s) point of departure for the clause. It isrealized in English by position in the sequence: thematic elements areput 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 exampleoh, 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 thetopical element: in other words, the Theme of a clause extends up tothe 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 unmarkedTheme 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 unmarkedtopical Theme. At first, the impersonal you predominates, followedlater 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 adiscussion of a general topic with particular personalities brought in tocarry it forward, the whole being linked together both dialogically andlogically. 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 hypotheticalcases, without much concern for attitudinal rhetoric (the only interper-sonal Themes are questions and uncertainties), and with some concret-ization towards the end.




269

4 Information structure

Each information unit is analysed for information structure (numbers

refer to clause complexes, letters to information units):

Figure 5




270

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 con-

tour 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 tocarry 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




271

attention are the baby’s growing up, the children’s understanding andobligations, 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 ismade prominent (by the speaker) to the listener; it is “what you arebeing invited to attend to”. When clause and information unit are

mapped on to each other, the result is a wavelike movement fromspeaker to listener, the diminuendo of the speaker’s part being as itwere 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 Themesrepresents the “method of development” of the dialogue, while thesequence of News represents the “main point” of the discussion, witheach speaker contributing her part to the construction of the overallpattern – all unconsciously, of course.

This interplay of two distinct waves of prominence is possiblebecause 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 combina-tions of the two kinds of texture. In unit 2h, for example, Elsie might

have chosen a different distribution by combining thematic and infor-mation 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, andmarked calling it she as Given; the effect of the latter would have beento bring out the repetitive facet of calling it she , thus reinforcing itscohesion with 2e, but by the same token to deprive it of its status as amain point for attention. The interaction of the thematic and informa-tional 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:




272

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’).






274

This is typical of logical argument; and it is interspersed with interrog-atives 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 justthree 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 twoinstances of I think). As it happens, in two paired instances the twokinds of modality are associated: where the speaker expresses a judge-ment 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:




275

Figure 8

Transitivity is the representation, in the clause, of the experientialcomponent of meaning: specifically the processes, the participants inthem, and the attendant circumstances.




276

This text is a discussion of a problem: what to do if a certainsituation arises. The majority of the clauses in it relate to the situation,the problem, its solution, and the process of problem-solving:

(i) possessive / attributive: have + baby ( 3)(ii) intensive / attributive: be + male / female ( 5)

(iii) intensive / identifying: call+ baby+ he / she ( 11)(iv) cognitive: know, find out ( 8)(v) verbal: say, suggest ( 3)

The remaining six clauses include three characterizing the baby (twomaterial: grow, get on; one circumstantial: be in + family), one character-izing 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 andverbal); Type III, being (relational and existential). They are distin-guished in the grammar in various ways; the principal distinctions are

as follows:

Figure 9




277

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 noAgent; (iii) however is effective – there is an Agent in the identification

process, always represented as you, but moving from personal “you” inclause 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 reachednow; and the teacher has put the problem into words.

A summary of process types and the relevant participant functionsfollows:

Figure 10




278

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




279

possible combinations. In the “full” (finite) system of tenses the num-ber is 36; from this is derived, by a neutralizing of certain tenses inthe “past” series (he said she had arrived corresponds to she arrived, she

has arrived , and she had arrived ), the 24-tense “sequent” system; andfrom this in turn, by a parallel neutralization in the “future” series (tobe 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 just12. This last is also the system that applies if the Finite verbal elementis a modality (e.g., should, must ), since that eliminates the primarytense choice.

Most of the verbal groups in this text are simple present tense; notonly because of the general nature of many of the propositions, butbecause most of the processes are other than “doing” ones, andtherefore have simple present, not present in present, as their unmarkedchoice. Furthermore, even the material ones are dependent (if it getson, when it grows up), which again requires simple present.

Nominal Group. The only nominal groups with structure other thansimply 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 theentity represented by the Head noun. There are similarities betweenthe two types of word group; but the nominal group has much morelexical material, since entities have a more developed taxonomic organi-zation. Hence the nominal group has a functional structure Deictic– Numerative–Epithet–Classifier–Thing–Qualifier, with left–right ordering




280

from the most instantial, situation-bound to the most permanentcharacteristic, modified by a purely syntactic principle which putsanything 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 thathave 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 followingthe Attribute used ; and this is a metaphor for a modality “have sousually been calling it she”.

Figure 13

8 Grammatical and lexical cohesion

The text is analysed for grammatical and lexical cohesion:




281

Figure 14

The headings given in the key are the basic lexicogrammatical

resources for creating texture between clause complexes. In fact theyfunction 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




282

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 textualretrieval and rarely extends beyond one clause complex. Conjunctionrefers to the nonstructural representation of logical-semantic relationsthat may also be expressed structurally; for example, on the other hand ,

semantically related to paratactic but and hypotactic although. Lexicalcohesion 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-frequencycollocate (e.g., house . . . family); or the use of a hyponym or superor-dinate – 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 andpersonal reference; the discourse unfolds around a small number of

entities that are constantly being referred to. When a new instance isbrought in, the link is achieved by collocation: in the house . . . next door . There is very little conjunction, because the logical-semanticrelations are realized by the hypotactic organization of the clausecomplex: The reasoning is systematic and explicit. There is also littleellipsis, which comes in only when the reasoning gives way to a moredialogic pattern with shorter exchanges.

9 Grammatical metaphor

Grammatical metaphors are identified, reworded, and reanalysed.Most examples of adult English contain some instances of grammat-

ical metaphor: clauses in which one type of process is represented inthe 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 pricefor 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 adultlanguage. There are no examples of it in the present text.

There are however certain grammatical metaphors that have beenbuilt 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




283

“his nose is long”. One very common type of these is the use of amental 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 oneof these as an example:

|| I don’t suppose mental: cognitivenegative

|| she came materialpositive

||| a ˆ ´ b

reworded as:

||| she probably didn’t come material

modalized:probability/mediannegative

||| (single clause)

In cases like this it saves time if the analysis moves directly to the

nonmetaphorical version, since the rewording is quite automatic. Inother 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 severalsteps in rewording to reach a nonmetaphorical version, and there maybe more than one possible route; all are potentially relevant to anunderstanding 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 social-

semiotic activity. The “tenor” is who are taking part: the statuses andmutual roles of the interactants. The “mode” is what part the languageis playing: the rhetorical and communicative channels. This descriptionis then used to interpret the lexicogrammatical features of the text.

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




284

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-and-answer exchanges. Exploratory; hypothetical and logical inorientation, 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 linguisticsystem, the patterns are probabilistic: given these features of the con-text of situation, we can make semantic (and therefore lexicogram-

matical) 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




285

learn the language in the first place: the systematic relationship betweenthese categories of the situation and the metafunctions of the contentsystem. By and large, characteristics of the field predict experiential

meanings, those of the tenor predict interpersonal meanings, and thoseof 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 tobe called upon if the situation was represented in this way.

11 Conclusion

Three final points should be made about an outline of this type. Oneis that it is just an outline: Obviously, the analysis under every headingcould be developed much further in delicacy, and other headings couldbe added. The guiding principle is to select and develop whatever isneeded for the particular purpose in hand. There are many differentpurposes for analysing a text, and the scope and direction of the analysiswill 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 aremany tendencies, continuities, and overlaps. Many actual instances canbe analysed in two or more different ways, none of which can be ruledout as impossible; some may be less sensible than others, and so can bediscarded, 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 multiplegrammars; but this is not a distortion of the system – it is a richer useof its natural resources. All analyses may need to figure in theinterpretation.

Thirdly, the lexicogrammatical analysis is only a part of the task. It isan essential part; all text is made of language, and the central processingunit of the linguistic system is the lexicogrammar. But just as thegrammatical system does not itself create text – text is a semantic

creation, with the grammar functioning largely (though not entirely) asthe automatic realization of the semantic choices – so the analysis of the grammar does not constitute the interpretation of a text. (Therehas been some misunderstanding on this point, for example in the useof cohesion as a method of text analysis. Cohesion is an essentialproperty of texts, but it is the way the cohesive resources are deployed

that makes the difference between text and non-text, and between onetext and another.)




286

Editor’s note: Examples of text analysis based on the grammaticalprinciples 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 theanalysis of discourse.



S ECTION THREE

CONSTRUING AND ENACTING






289


Linguistics like other sciences requires a metalanguage for representingits object of study, which in the case of linguistics is language itself.However, as Halliday points out in ‘On the ineffability of grammaticalcategories’ (Chapter 11), unlike other sciences, linguistics is “languageturned back on itself”, to use Firth’s expression. The problem lies,

Halliday argues, in the nature of language as object. Because languageis an evolved system, not a designed system, it rests on principles thatare ineffable. Its very richness, “its power of distilling the entire collectiveexperience of the culture into a single manageable, and learnable, code. . . puts its categories beyond the reach of our conscious attempts atexegesis”. This richness is most apparent in unconscious, spontaneous,

un-self-monitored language, or as Halliday notes, “our ability to uselanguage depends critically on our not being conscious of doing so”.

In ‘Spoken and written modes of meaning’ (Chapter 12), originallypublished as a chapter in Comprehending Oral and Written Language (1987), Halliday elaborates further on the differences between uncon-scious and spontaneous spoken discourse, and its more conscious andself-monitored counterpart, written language. Arguing against assump-tions that written language is syntactically more complex or more richlyendowed than spoken, Halliday maintains that each is highly organized

and complex in its own way: “Written language tends to be lexicallydense, but grammatically simple; spoken language tends to be gram-matically intricate, but lexically sparse”. Halliday describes writtenlanguage 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 languageobjectifies. “A written text is an object, so what is represented inwriting tends to be given the form of an object. But when one talks,



construing and enacting

290

one is doing; so when one talks about something, one tends to say thatit happened or was done.” By means of grammatical metaphor, writtenlanguage symbolically distances “the act of meaning and its counterpart

in the real world”. Each plays complementary roles when it comes tousing 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 takenas a mode of action occurring at the intersection of the conscious and

material modes of experience. In particular, Halliday examines theevolution of protolanguage into language, or how the two-dimensionalsemiotic that defines the mammalian experience evolves into ‘a semioticof a new kind: a stratified, tri-stratal system in which meaning is “twice-cooked”, thus incorporating a stratum of “pure” content form. Thetwo 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 introduc-tion of grammar, or as Halliday describes it: ‘a purely symbolic mode

of being between these two interfaces’. The processes of instantiationand 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 andgrammatics’, published as Volume 121 of the series Current Issues inLinguistic 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 howthis theory of our human experience works. Halliday introduced theterm “grammatics”, in contradistinction to grammar, to distinguishbetween a particular stratum of a natural language and the study of thisstratum. “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 powerslanguage and also conditions our attitudes to each other and to theworld around us. “To be a linguist”, Halliday writes, “is inevitably tobe concerned with the human condition”, and those who ‘thinkgrammatically’ will be better prepared not only to address issues of social injustice and inequality, but also to contribute to the develop-ment of new applications of linguistics such as intelligent computing.



291

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 morethings come into our lives; and things, and all their parts, need names.

So more and more words come in with them – new words, or newways of exploiting, embellishing and combining the old ones; and inthis way the balance is maintained. There is no sign that our onomastic

resources are drying up; indeed we are likely to run out of naturalresources long before we run out of names for the things we make outof them. But there are signs that the things are becoming more resistantto being named. There is no natural way of referring to the varioussmall plastic objects that lie around the house, or the toys we give our children, or the furniture we now have to assemble for ourselves. Theyno longer fit our taxonomies. We live in modules, sit on units andentertain ourselves with systems.

Behind these nameless objects is a technology and a science thatproduce them; and there, less visible to us, is another realm of thingsthat have to be named. Many of them are abstract things, the categor-ies and concepts of a theory; and some of these also prove recalcitrant

to ordinary onomastic processes – they only come to be ‘named’ bysome mathematical formula, like a function of the co-ordinates of x and y,the integer over psi

1 and psi

2, 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 Martinand Kim McCalla, pp. 3–18. Amsterdam: John Benjamins.




292

In this respect linguistic categories are no different from those of other theories; they too need names. It is true that in one instance atheory evolved without them: Chinese phonology, in its first few

centuries, had neither terms nor definitions – and still managed to givean 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 itfrom some other field. The earliest known Arabic grammarian, Al-Khalil ibn Ahmad, took various terms over from architecture; his greatpupil Sibawayhi, who had been trained as a lawyer and so placed a

high value on names and definitions, replaced them with a moresystematic 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, extendingover three or four centuries; and in the course of this time the originalterms had moved some distance from their non-technical meanings,evolving as the theory evolved. Thus, in everyday language, onomameant ‘a name’, rhema meant ‘a saying’; logos meant ‘speech, discourse’

and grammatike meant ‘writing’. As grammatike evolved into ‘grammar’,onoma came to mean ‘a noun’; rhema 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 whichbecame the source of technical nomenclature in grammar.

Once having reified these abstract categories by naming them, theGreek grammarians went on to ask what the names meant. What ‘is’ anoun? they wondered. At first, this was a question of: how do Irecognize a noun when I see it? how do I know whether something isa noun or not? But before long the questions came to be asked in theother direction: what ‘is’ a noun, in the sense of what function does itserve? At this second stage, instead of treating ‘noun’ as the Value andthen supplying a Token for it, the definition now treats ‘noun’ as theToken and seeks to supply a Value for it (for the terms Token andValue 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 adecoding definition, one which embodies a notion of ‘what thecategory means’.

To define a linguistic term by encoding is relatively simple: onehops along the realizational chain of grammatical categories untilreaching some form of output. Defining a noun in this way wouldinvolve, altogether, three steps: (1) a move in rank – a noun inflects



on the ineffability of grammatical categories

293

for case; (2) a move in delicacy – case is nominative, genitive, dativeor 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 themetalanguage in which the category is ultimately represented is of adifferent 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 twoworlds, as happens in a language where every morpheme is mappedinto a syllable; but this, while it is a bonus to the linguist, definitelydoes not lead him to say that a morpheme and a syllable are ‘the same

thing’. Hence in an encoding definition the category and its interpreta-tion 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 firstplace. 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 thename of something (and the gloss ‘of a person, creature or thing’ addsnothing; it simply lists everything nameable) is, at first sight, merelytautologous.

This apparent tautology is one that is discussed by Michael Reddyin his paper ‘The conduit metaphor’ (1979), in which he develops

Whorf’s theory of the metaphor of the container in languages of theStandard Average European mould. According to Whorf, westernlanguages characteristically employ an extended spatial metaphor whereby mental processes and relations become ‘objectified’, as illus-trated in his famous passage (1956: 146):

I grasp the thread of someone’s argument, but if its level is over myhead my attention may wander and lose touch with the drift of it.




294

This is accompanied by an antinomy of count / mass, which producesexpressions such as a pane of glass, conceptualized as ‘a pane with glassin 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 andshows how the metaphor determines the way we talk about communi-cation, with ideas, meanings and emotions being packaged inside words

and sentences and piped along a conduit. He lists about 150 suchexpressions, 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 termino-logy of linguistics, because the effect of this metaphor is that all wordsfor kinds and quantities of discourse, like poem, message, text , are splitinto 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 (lexicogrammaticalstratum). Reddy’s view is that as long as one remains within theconduit metaphor the effect is still benign, because the one sense is

then metonymic to the other: that is, the meaning is ‘contained in’ thewording. So text

2 ‘wording’ ‘contains’ text

1 ‘meaning’, and no great

harm is done. It is if one tries to escape from the metaphor that thiskind of polysemy becomes pathological.

It will have been noticed that this stratal polysemy is precisely theprocess by which the original Greek terms had come to be extendedso as to serve in the grammar: starting as names of semantic (or pre-semantic) categories, they were then transferred to become names for the lexicogrammatical categories by which the former were (typically)expressed. So logos

2 ‘sentence’ is that by which logos

1 ‘discourse’ is

realized. Likewise with onoma and rhe ¯ma: onoma2

‘noun’ is what realizesonoma

1 ‘name’ – while with rhema there is an additional step: rhema

1

‘saying’ is realized by rhema2

‘Rheme’, which is in turn realized byrhema

3 ‘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 toinvestigate its central processing unit, that the categories of this codingsystem are not arbitrary, but relate systematically to the meanings. It isnatural, therefore, to name them by reference to their semanticfunction.

With terms for classes and units, like those above, the polysemy isconfined within language. Both ‘noun’ and ‘name’ are linguisticentities; the difference is simply that one is grammatical, the other




295

semantic. The same is true of the pair ‘sentence’ and ‘discourse’. Butwith the majority of grammatical terms – those for functions, systemsand terms in systems (features) – this does not hold. These too have to

be imported into grammar from outside; but in this case they comenot from folk semantics but from outside language altogether. Whencategories 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 arelation between that category and the speaker’s experience of theworld. The term ‘plural’ is the name of this relationship. But the sameterm is also used as the name of the grammatical category which realizesthis relationship; a noun will be said to be ‘plural (in number)’. Andthis can cause problems.

Typically in the history of western linguistics the reasoning hasproceeded 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 redun-dancy, such that x redounds with one of a set of countable things andy redounds with two or more. The abstract labels ‘a : x/y’ are thenreplaced by the substantive labels ‘number : singular / plural’. This isof course an idealized model of the process; such abstract labels havenever been used, as far as I know, at least until modern times. But ithelps to bring the issue into focus.

What happens next? The categories that have been labelled in thisway 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 morethan one’. In fact, these are definitions of the words singular and plural ;but they are made to serve as definitions of the metawords, the termsof the metalanguage. Next, the terms are evaluated for their predictive

power: will they correctly predict text from situation, or situation fromtext? 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 disturb-ing for those who like their categories pure. This then gives rise to atheory of ‘core meanings’: a term of the metalanguage is said torepresent the ‘core meaning’ of the category. With this defence, in




296

spite of fruit and furniture, scissors and oats (or their counterparts in thelanguage 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 metalan-guage 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 swellingsomewhere on my body. He looks at it, and pronounces ‘You’ve gotan oedema.’ What is oedema? – Greek for ‘swelling’. But what he isdiagnosing is a more abstract swelling – it is that which is manifested

by the swelling on my body. An oedema is the Value of which aswelling is the Token. The use of a different term in this way allowsfor stratal diversification: not all swellings realize oedemata, and not alloedemata are realized as swellings. The relationship is a probabilisticone – and hence invites further, more delicate investigation.

Where do such terms come from? Ideally, they come from a parallelbut distinct semiotic. It should be a natural language, in order tomaintain the non-arbitrariness of the relation between the symptom

and the underlying condition – given swelling and oedema we wouldpredict that, in default of any special circumstances, they will refer to

the same thing. But it should be a different language, or at least adifferent sub-language, in order to allow for instances where they donot. And it should be a higher status code, in order to symbolize ahigher order of abstraction (and also the social value of abstractknowledge). The ideal source of a metalanguage is thus the ‘high’variant in a diglossia. A word of a natural language that is at oneremove from primary reality, such as ancient Greek, or classicalChinese, or Sanskrit, is appropriate for symbolizing a phenomenon thatis at one remove from primary observation.

But when it comes to metalinguistic matters, linguistics presents aspecial case. It is not just another science. It is ‘language turned backon 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, onegrammatical 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’




297

was set up in the first place to account for a morphological phenom-enon: 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 havetried 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 morethan one, or several, or many; but the trouble is we don’t actually sayI like more than one cat , or I like many cat – we say I like cats. Themeaning 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 languagesare not good things for glossing with; in that connection, Reddyremarked, ‘As a metalanguage, English, at least, is its own worstenemy.’ We can certainly point to some deplorable habits that English

has, both in its vocabulary and in its grammar. For example, wefrequently use the same lexical item to stand both for the study of aparticular phenomenon and for the phenomenon itself, as when wetalk 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 languageand the name for the study of that stratum; and likewise with phonology

and semantics. Not even the conduit metaphor excuses a raggedpolysemy such as these.

Even worse are some of English’s grammatical pathologies. For our metalinguistic vocabulary, we usually draw on some parallel semiotic asalready illustrated, bringing in new words so as to be freed from theaccumulated associations of the old ones. (The freedom is often short-lived, since the new term may soon be borrowed into the dailylanguage, like the psychological above.) But for the grammar of our

metalanguages we are usually content to stick with the everyday formsof English; and this can lead to serious misconstructions – such as thefollowing, perpetrated by myself, when I wrote some time ago:

the Theme of an English clause is the element that is put in firstposition.

Now I meant this as Value ^ Token, with is meaning ‘is representedby’. But all such clauses in English, if they have the verb be , are




298

ambiguous; and this one was frequently misread as Token ^ Value,with is taken to mean ‘represents’. In other words, a clause that wasintended 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 mostfundamental 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 putin first position.

So there are problems in using natural language as a metalanguage,for whatever purpose: its logical and ideational systems were notdesigned for the task. Some combinations of features may be realizedin ways that are ambiguous, others may carry a baggage of unwantedcorollaries, and so on (this does sometimes lead to the creation of minor neologisms in the grammar, like the prepositional phrases thatappear 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, sincewhatever shortcomings it has are compounded by the factor of self-reference – the metalanguage being a form of the same semiotic systemthat it is also being used to describe.

The problem of self-reference is a familiar one; nevertheless it is notthe central issue. The real problem lies in the nature of language asobject, and particularly the nature of lexicogrammar. It is not so muchthat language is not good for glossing with. The problem is rather thatlanguage is not good for being glossed.

Let us take, as an example, the category of Subject. This has alwaysbeen one of the most obscure and controversial categories in westerngrammatical theory. Here is Jespersen on the subject (1909–43, Vol-

ume 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




299

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 under-

lying 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 non-western 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 functioningas Subject in any instance has meaning as actor, or has meaning as

topic; but as Subject it has none – the category of Subject has nomeaning in itself. In this view, Subject is a grammatical function whoseonly function is to be a grammatical function.

Such a view is enshrined in the terminology, in the term ‘grammat-ical Subject’ (used for example in Sweet 1891); this is in contradistinc-

tion to ‘logical Subject’ (i.e. Actor) and ‘psychological Subject’ (i.e.Theme). Compare the later Prague school interpretation, with ‘syntac-tic structure’ (Subject–Predicate) contrasted with ‘semantic structure’




300

(Actor–Action) and with ‘functional sentence perspective’ (Theme– Rheme) (Danes 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 anymore obscure than other categories. It is just as impossible to arrive atan adequate gloss for functions such as Actor, Agent, Goal (‘logicaldirect 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 thegrammarian’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 naturallanguage. But there are sound and respectable reasons for wanting todo so. The original impetus for semantic glosses on the grammar comesfrom the desire to explain the observed formal patterns: why is thisparticular 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? Toanswer questions like these one has to postulate more functional,semantically oriented generalizations; these are then used to predictfurther instances that have not yet been observed.

Then subsequently the technique is extended to the interpretationof ‘unknown’ languages, either for descriptive or for pedagogicalpurposes. 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 anevent which happens to the ‘actor’ without his volition. Thus, there isno 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 discussionof tenses (or, in his later term, assertions) in Hopi: factual or present-

past (later reportive), future (later expective), generalized or usitative(later nomic) – which initially takes up half a page, but whose semanticand ideological interpretation is the subject of an entire article (‘AnAmerican Indian model of the universe’). Such discussions typicallyinvolve implicit, or sometimes explicit, contrast with the language inwhich 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.




301

One project in which they play an essential part is the Penman projectat 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 networkof some hundreds of options. At each choice point a Chooser isactivated; the Chooser consults the environment (the Knowledge Base)

for instructions on which way to go. The Chooser’s questions arereferred to an Inquiry Operator; and they take forms such as thefollowing:

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 whichcauses 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 doesnot expect the hearer to identify uniquely by reason of the attentionwhich it currently holds, its inherent uniqueness in culture, or itsassociation 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, itmay be assumed that all grammatical generalizations have some signific-ance at a higher stratum; or alternatively, that some are simply house-keeping devices and have no semantic function. Secondly, thosegrammatical categories that are regarded as semantically significant maybe thought of as universal, or as particular to the given language, or asparticular to a given register, a functional variety of a language. (Thesewould represent fairly well the respective views of Jakobson, Hjelmslevand Firth.) Thirdly, it may be held that every such category has onemeaning 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 meaningvaries 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 gram-matical categories can be related in a systematic and in some sense‘natural’ way. Positions taken on these issues may complicate the taskof semantic interpretation: for example, if categories are assumed to beuniversal, and yet are established at an insufficiently abstract level.

But whatever beliefs are held about them, grammatical categories




302

will remain ineffable. Some of the more recalcitrant ones are categoriesthat Whorf originally called ‘covert’: ‘having no mark other thandistinctive reactances with overtly marked categories’. But by no means

all are of this kind – there is nothing covert about definiteness inEnglish, for example. The phenomenon we are concerned with hasmore to do with Whorf’s follow-up notion of a ‘cryptotype’.

Whorf remarks of these that ‘they easily escape notice and may behard to define, and yet may have profound influence on linguisticbehavior’ (1956: 92). Among cryptotypes in English Whorf citesgender, transitivity (of the verb), inherence (of the adjective), and

various more delicate categories, such as that of verbs that may bephrasalized with up. There is, of course, a connection between the twosenses in which Whorf is using ‘crypto-’: a category may be hard todefine precisely because it is hidden from view. But hidden from whoseview? It is not because they are hidden from the linguist that grammat-ical categories are hard to define; once the linguist has found them, thefact that they had escaped his notice ceases to matter. The significanceof 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 systemsof a culture; and although his observations have often been quoted, itseems 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 pointedout, in the many years since I began the study of informal speech, thatit is only in the most spontaneous, un-self-monitored kinds of discoursethat a speaker stretches his semantic resources to the utmost (cf.Halliday 1966). This does not happen in formal speech; and it certainlydoes not happen in writing. It is in unconscious spoken language thatwe typically find the truly complex sentences, with their labyrinths of hypotaxis and all their projections and expansions, from which, whilewe blunder through such sequences often losing ourselves completely

when we are engaged in the planned self-monitoring discourse of anacademic lecture, we emerge in good order and with every nodeunravelled provided we are completely unaware of what we are sayingand 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 uselanguage depends critically on our not being conscious of doing so – which is the truth that every language learner has to discover, and thecontradiction from which every language teacher has to escape.




303

Why is there this apparent contradiction, such that only when wecease to attend to the process of meaning can we ever master the abilityto mean? The immediate cause is no doubt the dynamic nature of the

medium – as with walking, or riding a bicycle: once you think aboutit 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 fromall conscious discourse, spoken and written. While the complexity of conscious language is dense and crystalline, formed by a closely-packedconstruction of words and word clusters, the complexity of unconscious

language is fluid and choreographic. Conscious language achieves itscreative force mainly by lexical means; and lexical items are semantic-ally close to experience. Unconscious language depends much morefor its creative force on grammar – and grammatical categories are far removed from experience. To quote Whorf again, grammatical cat-egories ‘represent experience . . . but experience seen in terms of adefinite linguistic scheme’ (1956: 92).

The meaning of a typical grammatical category thus has no counter-

part in our conscious representation of things. There can be no exactparaphrase of Subject or Actor or Theme – because there is no

language-independent clustering of phenomena in our experience towhich they correspond. If there was, we should not need the linguisticcategory to create one. If language was a purely passive partner,‘expressing’ a ‘reality’ that was already there, its categories would beeminently glossable. But it is not. Language is an active participant inthe semogenic process. Language creates reality – and therefore itscategories of content cannot be defined, since we could define themonly by relating them to some pre-existing model of experience, andthere is no model of experience until the linguistic categories are thereto model it. The only meaning of Subject is the meaning that hasevolved 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 alongwith the meanings; there is just one process taking place here, nottwo, though we have to interpret it as if it was two. We cannotobserve this process as it took place in the history of the community,since that would be coextensive with the evolution of the humanspecies. But we can observe it happening in the history of a human




304

child. Semogenesis begins well before the mother tongue, as the infantcreates his own protolanguage or ‘child tongue’: he constructs thissymbolic system, in interaction with those who share in his meanings,

for the twofold purpose of doing and thinking that characterizes allsuch 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 takesover the mother tongue with its ready-made grammatical categories.The symbols of the mother tongue, which have been around him fromthe start, now become his reality, at once a part of, and a key to, the

complex phenomena of his experience. Language and culture areconstrued 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 isnot intended to suggest that there is any contrast here with adults. Theproblem 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 heuses one a hundred times a day. We only have to listen to a five-year-

old 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 perman-ent duty, is only a limited set of instances. We have to infer thesystem that lies behind them; for language (if I may be allowed toinvert Chomsky’s famous dictum) is an infinite system that generatesonly a finite body of text. But what we can observe is already veryconvincing.

If I assert that a five-year-old knows what a Subject is, it is becauseI have listened to children for many years, and heard them talking inclauses which have Subjects. In my own detailed record of oneparticular child, there are about 2,500 of them; but since child language

studies became fashionable there has been an abundance of suchmaterial available, if one does not feel one can rely just on one’slistening. Now, any one of these clauses could have had the appropriateSubject by chance. Moreover, since no linguistic category is chosen inisolation – in choosing the Subject one is always making other choicesbesides (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




305

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 clauseswhich differ primarily in respect of which element is selected as theSubject. 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 thesame thematic structure, with why as Theme; and the same informationstructure, with focus of New on now . The only significant difference

lies in the choice of the Subject: in the first example that , in the secondexample mummy.

We can display the Subject prominently by turning the clause intothe 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 thatelement 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 wantsto appear in tag or response – or rather, the entity that he wants tocarry the meaning that is realized by its potential for appearing in tagor response, whether or not any tag or response is actualized. (Comparethe reason for putting something in clause-final position: so that itcarries the meaning that is realized by its potential for bearing the

unmarked focus of information, whether or not that is where the focusactually 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 couldproduce instances of either of these structures if the occasion arose. Hewould not, of course, produce them one after the other, since we donot talk in paradigms; whichever occurred would be in its appropriate

syntagmatic environment. On the other hand, I doubt whether he




306

could produce either of them under test conditions, or discriminatebetween them if presented to him; but then neither could an adultspeaker. The fact that we cannot bring to consciousness the difference

in meaning between two related forms which we nevertheless keepsystematically 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 beenactively listening for some 31 ⁄ 2 out of his five years, and assuming thathe has been exposed to a tally of about 200 Subject-featuring clauses aday (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 putit together with Whorf’s ‘sense of the cumulative value of innumerablesmall momenta’, we will not be surprised that a grammar can be learntin this way. I have been insisting for many years that a child’s semiotic

experience is extraordinarily rich – not at all the farrago of featurelessfragments that we are taught spoken language consists of! Much of itcomprises repetitions, like Come and have your lunch; but repetition isitself a semogenic factor, since it allows the child to model the languageas a probabilistic system. Every instance, whether repetitive or unique,is a configuration of meanings of different kinds, available to the childboth for storage as coded text and as evidence for construing the systemthat lies behind. There is no mystique in a child’s ability to construct alanguage 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 asingle definition, or even in an article or a book, the whole of theshared 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 unlessthe principles can be made explicit. But a language is an evolvedsystem; and evolved systems rest on principles that are ineffable – because they do not correspond to any consciously accessible categor-ization of our experience. Only the relatively trivial meanings of anatural language are likely to be reducible to (meta-)words. Funda-mental semantic concepts, like those underlying Subject, or Theme,




307

Actor, New, definite, present, finite, mass, habitual, locative, are, in anentirely 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 formsthrough which it is expressed and then asking what these forms mean,we will simply get a relabelling of each formal category. The descrip-

tion will be a gloss on its name (since the name was an attempt tocapture its meaning in the first place), together perhaps with a gloss onits relationship to other categories that were themselves formallyestablished. To pursue our earlier example from grammar, a noun will

be defined as ‘that which names a person, living being or inanimateobject (gloss on the name “noun”); which can be a participant in anaction 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 – withcategories 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 semanticsas 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 soreadily glossed in this way: they relate directly to the semantic systemthat is ‘above’ the grammar, that which interprets the ideologies of theculture (Lemke’s ‘activity structures’ and ‘thematic systems’), and codesthem 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 reversesthe perspective derived from the history of linguistics, in which alanguage 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 arethe choices in meaning in whose realization the Subject plays somepart?’ We look for a semantic paradigm which is realized, inter alia, bysystematic variation involving the Subject: in this case, that of speechfunction, in the interpersonal area of meaning. This recalls Firth’s




308

notion of ‘meaning as function in a context’; the relevant context isthat of the higher stratum – in other words, the context for understand-ing 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. Thespeaker 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 establishclause by clause what has changed and what has remained constant.

Subject Actor Theme

1 I I I Subject = Actor = Theme2 I ball I Subject = Theme; Actor 3 I I ball Subject = Actor; Theme

4 I ball ball Actor = Theme; Subject

The speaker keeps the listener’s attention by varying the Theme andthe Actor:

Theme Actor

1 about me (‘I . . .’) what I did (‘did you?’)

2 about me (‘I . . .’) what happened to me (‘were you?’)3 about the ball (‘the 3rd’) what I did (‘did you?’)

4 about the ball (‘by the 4th’) what happened to me (‘were you?’)

Clause 1 has the speaker in all three functions of prominence: interper-sonal (I as Subject), ideational (I as Actor) and textual (I as Theme).Clause 2 is marked by ideational modesty (this is what happened tome, 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 isneither 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 isno sign here of interpersonal modesty; the assertion is made to rest onI every time, and the listener’s response, correspondingly, must alwayshave a you in it – Did you?, Were you? In other words, every step in




309

the narrative has to be validated by reference to ‘me’; and that is themeaning of Subject. ‘Subject’ is not the same as ‘Actor’; nor is it thesame as ‘Theme’. But it is far from being devoid of meaning.

It is quite possible to have a clause in which all three of thesefunctions are dissociated from one another:

that teapot my aunt was given by the dukeTheme Subject Actor

This helps to give a sense of the different meaning of each. Thisparticular one is made up; but the structure of which it is an instance is

entirely familiar. (Except perhaps to philosophers of language, whotend to be disturbed by departures from the ideal of John hit Paul . Oncewhen I was giving a seminar, a participant refused to accept that such aclause was possible in English, with a non-Subject nominal in frontedposition. 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 tosay, 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 pointedthis out to him, he seemed to think I had cheated – perhaps by bringingreal language into the discussion.) Obviously, there would be no sensein dissociating the Subject from both Actor and Theme if it did notembody 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. Theproblem of ineffability is common to all grammatical categories; thereare various reasons why some may seem less problematic than others,but it is an illusion to think that any can be exhaustively defined. Andthis, as I remarked above, is not because of the shortcomings of natural

language for serving as a metalanguage, real though such shortcomingsare. Rather the converse: it is the very richness of natural language, itspower of distilling the entire collective experience of the culture into asingle manageable, and learnable, code that puts its categories beyondthe reach of our conscious attempts at exegesis.

This leads us back to the question of the Grundbedeutung . Thecategories we have been considering have been categories of thegrammar: grammatical systems and structures, and their component




310

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 samething’ 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 metafunc-tional 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 eachregister. No doubt we can also conceive of such a thing as ‘thesemantics of English’; but we also feel that (at least at the present stateof our knowledge) it is not counter-productive to envisage a morerestricted domain for semantic generalizations, and to operate withsemantic 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 thescale, 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 thepurpose was to explore what meaning potential the child must have inorder 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 exper-imental situation types. More general again is Ruqaiya Hasan’s (1983)‘message function’ network, which describes spontaneous interactionbetween children and parents, for the purpose of investigating thedevelopment of children’s learning patterns. At the other end of thescale, J. R. Martin’s (1983a, 1992) conjunction networks are likegrammatical networks in that they are set up for the language as awhole.

When we describe semantic systems, we are saying what it is that

‘preselects’ the grammatical categories: what choices in meaning call onwhat features in the grammar for their realization. It is by this processthat the grammatical categories are defined; when this is done, there isno need to gloss them further. Once the semantic system is madeexplicit, it can only be misleading to attach separate semantic descrip-tions as glosses to the categories of the grammar.

At the same time, if the semantic system is set up only for a restricteddomain, some particular register variety, then the meanings of any




311

grammatical categories that figure distinctively in that variety willappear thereby less ineffable. For example, we have no general defini-tion 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 makesonly limited demands on it, for realizing the meanings that are

engendered by that particular context. The category of ‘future in theregister of weather forecasting’ is much less resistant to being glossedthan 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 semantictraditions of Boas, Sapir and Whorf, of Malinowski and Firth, and of Mathesius and the Prague school. These were three groups of scholarswith very different orientations, but their work was complementary insignificant ways. While each had a well-rounded view of language,they emphasized, respectively, the ideational, the interpersonal and thetextual 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 onother 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. Itprovided a model of reality; but when the two did not match, sinceexperience was interpreted within the limitations of this model it couldbe disastrous in action – witness the exploding petrol drums. Mathesiusshowed how language varied to suit the context. Each sentence of thetext was organized by the speaker so as to convey the message hewanted at that juncture, and the total effect was what we recognize asdiscourse. Their work provides the foundation for a systematic func-tional semantics which enables us to bridge the gap between thecontext of culture and the language, and between the context of situation and the text. This is how we can become aware of themeaning of grammatical categories.

As a final step, let us summarize some of the alternative principlesthat can be adduced for talking about the ineffable:

A. Metonymic: the use of some semiotic system as a descriptivemetalanguage (the ‘parasemiotic’ principle)1. Parallel semiotics within the same language

(a) everyday language as folk linguistics (possible where thereis a shift of metafunction but probably not otherwise)




312

(b) open-ended definitions from theories about language (nottheories of language), e.g. rhetoric

(c) self-contained, technical systems of definition within lin-

guistics itself, e.g. Hjelmslev’s2. 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 bea 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 ingeneral greater than that of a lexical item; the sign is suscept-ible 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 differentGoals’, ‘glossed’ by two picture series each of three

events, one person kicking another – the one beingkicked either falls lower and lower (same person kickedrepeatedly) or stays in the same posture (different personeach time)

(b) Non-representational: e.g. music as metalanguage, e.g.Chandola (1970: 145) suggests that many aspects of ragascan be compared with language, and gives an example of a musical pun – but music might lend itself more toglossing dynamic aspects of the grammar such as informa-tion structure

B. Metaphoric: the use of theoretical models1. 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 ascomplementary modes of representing experience2. A general theory of register and genre, from which semantic

categories can be derived in a principled way (Martin, inpress)

3. General theories of meaning in language, e.g. Lemke’s (1985)concept of ‘making meaning’ in terms of activity structuresand thematic systems, and of ‘metaredundancy’




313

4. Perhaps as a combination of the metaphoric and the meto-nymic 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 claimedfor it, of being able to interpret all other semiotic systems (and I see noreason to assume that this is so), there are certainly limitations on theability of language to interpret itself. We may have to move outsidelanguage, to some parallel or higher order semiotic which, since it isnot itself language, can be represented in language and then refracted

to become a metalanguage for representing language. All such inter-pretation is ultimately circular; but in linguistics, we have tended tooperate within circles that are pathologically small. Until we can createa 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 why the North Star stays still

Sayer Process: Receiver Cause Carrier Process: Range

Medium Verbal Beneficiary Medium relational

intensive

Actor Process: material




314

that ’s where the magnet is they move around

Identified

Token

Process:

relational

Identified

Value

Actor Process:

material

Place

circum- Medium

stantial Attribute Carrier Process:

Medium Range Medium relational

circum-

stantial

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

Goal Process: Actor Goal Process: material Actor

Medium material Agent Medium Agent

Process type Process Medium Other elements

2 do material: middle stay still North Star Cause: Why?

4 be at relational: identifying

circumstantial: locative

is (represents) North Star Location: where

magnet is

4

[ [ ] ]

be at relational: attributive

circumstantial: locative is (has attribute) magnet 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 material: middle move other stars

1.2 Logical

Lexicogrammar : clause complex – interdependency, logical-semanticrelation

‘why?’ . . . because and but so




315

Network: experiential

clause { PROCESS

TYPE

VOICE

material 31relational 32

verbal 33

effective 41

middle 42

{

attributive a

identifying b

intensive x

circ.: place y

possessive z

[Nigel] 1–2 33.42 31.42

[Father] 3 33.42[Nigel] 4 –7 32by.41 [[3ay.42]] 31.41 31.41 31.42

Network: logical

clause

complex { INTERDEPENDENCY

(‘TAXIS’)

LOGICAL-SEMANTIC

RELATION

paratactic

hypotactic

expansion

projection

1 2 3 . . .

a b g . . .

alaborating

extending

enhancing

locution

idea

=

+

“

‘

Semantics: experiential and logical: informal gloss

24567

North Star (

not move)

cause?cause:add (effect):contrast:effect:

magnetearthearth

be-atattractnot attractmove

North Star North Star other starsother stars

‘a does not do x, because a has property p, but not-a have a notproperty 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.




316

2 Ideational semantics

2.1 Experiential

PHENOMENA

objects

processes

reactants (‘magnet’)

bodies

locating (‘be at’)

Moving (‘do’)

stars

earth

motion

force

North Star

other stars

move

not-move

(‘stay still’)

alternatively:

moving(‘do’)

{

positive

negative

uncaused (motion)

caused (force)

2.2 Logical

LOGICAL

SEMANTIC

RELATIONS

additive

causal

contrastive (‘but’)

positive (‘and’)

effect (‘so’)

cause

unknown (‘why’)

known (‘because’)




317

Interpersonal semantics

ROLECOMMODITY (proposal)

goods-&-services(proposition)information

give

demand

offer

command

statement

question

{ ROLE IN

EXCHANGE

COMMODITY

EXCHANGED

give

demand

goods-&-service

(proposal)

information

(proposition)

VEIN

tentative (‘shall I? / will you?’)

neutral (‘let . . . / let me!’)

definitive (I’ll . . . / you . . .’)

2 Interpersonal

Lexicogrammar : clause – mood

shall I tell you

Finite Subject

Mood Residue

do

Finite

mood

why the North Star stays still

WH/ Subject FiniteAdjunct Mood

Resi- due

that ’s




318

itthe other starsthey

gets attracteddon’t

move around

Subject Finite

Mood Residue

Network: interpersonal

{ Mood

indicative

imperative

declarative

interrog

1st person

1st/2nd persons

2nd person

01

yes/no 02

WH- 03

04

05

06

MOOD

PROJECTION

ELLIPSIS

direct

indirect

full

elliptical

11

12

21

22

m a j o r

c l a u s e

[Nigel][Father][Nigel]

132

02.11.2106.11.2203.12.21

yes / no interrogative2nd person imperativeWH- interrog, indirect

‘offer, tentative’‘command, neutral’‘question, projected’

4–7 01.11.21/22 declarative, full /elliptical

‘statement’




319

Semantics: informal gloss

3 Textual

3.1 Structural (THEME and INFORMATION)

Lexcogrammar : clause – theme

Shall I tell you why the North Star stays still

Inter-personal

topicalRheme

Inter-personal

topicalRheme

Theme Theme

becauseand

but

so

thatit

the other stars

they

’s where the magnet isgets attracted by the earth

don’t

move around

structural topical Rheme

Theme

1

2

4–7

discourse theme [first clause]: offer (interpersonal)

clause Themes: (interpersonal) question

clause Themes: (structural) because,

and, but, so

(topical)

(topical)

(topical)

[speaker]

the North Star

the North Star,

the other stars




320

Lexicogrammar : information unit – information

Shall I tell you why the North Star stays still because That’s Where The magnet is

Focus Given Focus Focus

New New New

and it g ets attacted by the earth but the other stars don’t so they

Focus Focus Focus Focus Focus

Given New New New Given New New

move around

Focus

New

Network: theme systems

TOPICAL

THEME

unmarked

marked

TONICITY

unmarked focus

marked focus




321

INTER-

PERSONALTHEME

{

mood finite

WH- —

modality —

vocative

—

TONALITY

STATUS

unmarked (clause = info.unit)

thematic

(other)

marked

fresh

contrastive

TEXTUAL

THEME {

continuative

—

structural

—

conjunctive

—

3.2 Cohesive

stays still

S A

move around

1

1

the North Star |R

that (=there)|R

it

S H

the other stars|R

they

2

1

1

the magnet|C

gets attracted

E

don’t Ø

3

3

the earth

E

Ø

Key:SA: synonymy – antonymsSH: synonymy – cohyponymsC: collocationE: ellipsisR: reference

1: Process + Medium2: Process + Location3: Process + Agentcolumns: lexical chainboxes: referential chain




322

4. Field

4.1 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 informationderived 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. Tenor

5.1 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).



323

Chapter Twelve

SPOKEN AND WRITTEN MODES

OF M EANING (1987)

1 Spoken language and education

It seems to me that one of the most productive areas of discussionbetween linguists and educators in the past quarter century has been

that of speech and the spoken language. Twenty-five years ago, whenI 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 tothe need for children to be articulate as well as literate. Dell Hymeshad not yet introduced “communicative competence”; the words oracyand 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, inschool, as in the community at large, meant written language.

The word language itself was hardly used in educational contexts. Inthe primary school, there was reading and writing; in the secondaryschool there was English, which meant literature and composition. Notthat a classroom was a temple of silence; but the kind of spokenlanguage that had a place, once a pupil had got beyond the infant

school, was prepared speech: reading aloud, drama, debating – languagethat 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 mediumof learning.

First published in Comprehending Oral and Written Language , 1987. Orlando, FL: AcademicPress, pp. 55–82.




324

Among linguists, by contrast, the spoken language had pride of place. One learnt in the first year of a linguistics course that speech waslogically and historically prior to writing. The somewhat aggressive

tone with which linguists often proclaimed this commitment did notendear 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 outof harmony with the facts of the English language – whereas for themit was the pronunciation that was out of step, being a distortedreflection of the reality that lay in writing.

The linguists’ professional commitment to the primacy of speech didnot, however, arise from or carry with it an awareness of the propertiesof spoken discourse. It arose from the two sources of diachronicphonology (the study of sound change) and articulatory phonetics (thestudy of speech production), which came together in twentieth centuryphonological theory. This was an interpretation of the system of speechsounds and of the phonological properties of the stream of speech; itdid not involve any attempt to study the grammar and semantics of

spoken as distinct from written language. As early as 1911, in hisdiscussion of functional variation in language, Mathesius (1964) was

referring to “how the styles of speech are manifested in the pronuncia-tion of language, in the stock of words, and in syntax” (p. 23), and to“the influence of functional styles on the lexical and semantic aspectsof speech” (p. 24); and it is clear that “speech” for him ( parole ) didencompass both spoken and written varieties. But it was not until the1950s, with the appearance of tape recorders, that natural speech couldbecome the object of systematic study. The notion of “spoken text” isstill not easily accepted, as can be seen from the confusion that prevailswhen spontaneous speech is reduced to writing in order to be analysed.

Spoken language came to figure in educational discussions in thecontext of language in the classroom: the language used by teachers tostructure, 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 notconcerned with the particular place of spoken language in the learningprocess. It was assumed, of course, that students learnt by listening; butthe expository aspects of the teacher’s language were given little atten-tion, while the notion that a student might be using his own talk as ameans of learning was nowhere part of the picture. Probably it wouldhave been felt that the principal means of learning through the spokenlanguage was by asking questions; but studies of the early seventies (for



s p ok en a nd w ritte n m ode s of m ea ning

325

example the Toronto research reported in Five to Nine) revealed thatstudents seldom do ask questions – not, that is, while they are occupyingtheir 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 somewhatremoved from the patterns of natural dialogue.

2 Complexity of natural speech

Already half a century earlier Franz Boas (1911) had stressed theunconscious character of language, unique (as he saw it) among the

phenomena of human culture. Boas’ observation was to be understoodin 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; butthe unconscious was in the air, so to speak, and playing a critical rolein the conception of systems as regularities underlying human behav-iour. But Boas may also have had in mind the unconsciousness of thebehaviour itself: the act of speaking (acte de parole) as an unconscious

act. The lack of conscious awareness of the underlying system, andthe difficulty that people have in bringing it to consciousness, are thingswhich language shares with other semiotic systems – for example, socialsystems like that of kinship; what is unusual about language is theextent to which even the manifestation of the system, the actualprocess of meaning, remains hidden from observation, by performer and receiver alike. In that respect talking is more like dancing, or evenrunning, than it is like playing chess. Speaker and listeners are of courseaware 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 butalso the speaker will ordinarily offer a paraphrase, something that istrue to the meaning but not by any means true to the wording. Tofocus attention on the wording of language is something that has to belearnt – for example if you are studying linguistics; it can be a difficultand somewhat threatening task.

About 30 years ago, as a result of being asked to teach Englishintonation to foreign students, I began observing natural spontaneousdiscourse in English; and from the start I was struck by a curious fact.Not only were people unconscious of what they themselves weresaying; they would often deny, not just that they had said something Ihad 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




326

going to have been being tested makes five serial tense choices, present inpast in future in past in future, and is also passive. This passed quiteunnoticed by both the speaker and the person it was addressed to; yet

at the time it was being seriously questioned whether a simple verbform 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, onthe basis of observed frequencies of two- and three-term tense formstogether with the constraints of the tense system; but they are providedfor 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 relativesreaching into dependent clauses, such as that’s the noise which when yousay it to a horse the horse goes faster . These are all systematic features that

people are unaware that they incorporate in their speech, and oftendeny 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 “wellit 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 observationswas the complexity of some of the sentence structures. Here are twoexamples 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, byreflecting 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 mayagain 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’sdone 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




327

usually to kick for touch – nearly always for touch because he’s milesbehind 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 handsof 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 wasthat 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 theywere remarkably well formed: although the speaker seemed to berunning through a maze, he did not get lost, but emerged at the endwith all brackets closed and all structural promises fulfilled. And thisdrew attention to a third property which I found interesting: that whilethe listeners had absorbed these passages quite unconsciously andwithout effort, they were difficult to follow in writing.

3 Lexical density

These two examples have been around for a long time; so let me turnto some recent specimens taken from recordings made by Guenter Plum to whom I am indebted for drawing them to my attention. Inthese 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-tanbitch puppy, as they told me I should have bought a bitch puppy tostart 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.






329

Table 1

Lexical Density of Texts 1A, 1B, and 1C

(1)Lexical

items

(2)Running

words

(1:2) (3)Clauses

(1:3)

1A 23 83 1:3.6 13 1.8:1

1B 26 68 1:2.6 8 3.3:1

1C 25 55 1:2.2 4 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 grammat-ical 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 thenumber of running words, and (2) as a proportion of the number of clauses. Only non-embedded clauses have been counted (if embeddedclauses are also counted, then each lexical item occurring in them iscounted 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 functionof its place on a register scale which she characterized as running frommost active to most reflective: the nearer to the “language-in-action”end of the scale, the lower the lexical density. Since written languageis characteristically reflective rather than active, in a written text thelexical density tends to be higher; and it increases as the text becomesfurther 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 withreference to the number of clauses the discrepancy stands out moresharply. Thus in the example given above, while the number of lexicalitems remained fairly constant and the number of running words fell

off slightly, the number of clauses fell steeply: from 13, to 8, to 4. Inother words, the lexical density increases not because the number of lexical items goes up but because the number of non-lexical items –




330

grammatical words – goes down; and the number of clauses goes downeven 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 meanswhereby 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 inanother 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, theyhave been bringing numerous private civil actions, which have beenmainly 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 thisindependent 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, theycan help to internalize costs, and they can help to protect the environ-ment. 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 mainlyof two kinds.




331

Table 2 Lexical Density of Texts 2A, 2B, and 2C

(1)

Lexical

items

(2)

Running

words

(1:2) (3)

Clauses

(1:3)

2A 48 87 1:1.8 5 9.6:1

2B 48 101 1:2.1 12 4.0:1

2C 51 132 1:2.6 17 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 totalnumber of words – which means, therefore, in the number of gram-matical 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 thatwritten language has a higher lexical density than spoken language; thisexpresses it as a positive feature of written discourse and suggests thatwriting is more complex, since presumably lexical density is a form of complexity. Could we then turn the formulation around, and expressthe 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 tolook at how the words and clauses are organized.

Let us consider a shorter example of a pair of texts related in thesame way, one “more written” (Text 3A), the other “more spoken”(Text 3B). I have constructed these so that they resemble the originalsof Texts 1 and 2; but they are based on a natural example occurring intwo texts in which a person had described the same experience twiceover, 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.




332

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 onesentence, since they are related by hypotaxis – only one has independ-ent 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 textconsists of four clauses where Text 3A consists of only one. It is thatText 3B consists of a clause complex consisting of four clauses. Theclauses 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 complexthan the written one. The spoken text has a lower degree of lexicaldensity, but a higher degree of grammatical intricacy.




333

Figure 2

Let us return to Text 1, in its original spoken form (Text 1A). Thisconsisted of 13 clauses. However, these 13 clauses were not strung outend to end; they were constructed into a small number of clausecomplexes of mixed paratactic and hypotactic construction: arguably

just one clause complex throughout. Here is its interpretation as oneclause complex:

Figure 3

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




334

Table 4

Other examples from the same source but from different speakersshow similar patterns; there are, obviously, individual differences(including perhaps in the preference for one or other type of interde-pendency), but the same free-flowing intricacy is noticeable all thetime, 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, whichwas 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 veterinarysurgeon, 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 awedding present,’ so immediately I bustled around looking for someone tomarry so I could get a Shepherd given to me for a wedding present, yousee, 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 lookto 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




335

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 dobecause 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 runcounter to received attitudes towards spoken language. One is thatspeech 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 thepatterns we have been illustrating, which are the patterns of the

organization of the clause complex, referred to above as grammaticalintricacy, 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 achieveconsiderable intricacy in the structure of the clause complex; it can belearnt and consciously developed as a style. Some forms of spokendiscourse, on the other hand, militate against it: rapid-fire dialoguepresents no scope for lengthy interdependencies – complex semanticpatterns can be construed between interactants, but usually withoutbeing 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 textwhich are seen as characteristic of the medium. So it is important toindicate 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 issyntactically more complex than spoken, and to suggest that, as far asone 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 acharacteristic of the most unconscious spontaneous uses of language.The more natural, un-self-monitored the discourse, the more intricate




336

the grammatical patterns that can be woven. Usually, this kind of discourse will be spoken, because writing is in essence a more consciousprocess than speaking. But there are self-conscious modes of speech,

whose output resembles what we think of as written language, andthere are relatively spontaneous kinds of writing; spoken and writtendiscourse are the outward forms that are typically associated with the

critical variable, which is that of consciousness. We can use the termsspoken and written language, to refer to the idealized types defined bythat variable.

Spoken and written language, then, tend to display different kinds

of complexity; each of them is more complex in its own way. Writtenlanguage tends to be lexically dense, but grammatically simple; spokenlanguage tends to be grammatically intricate, but lexically sparse. Butthese buts should really be ands, because the paired properties arecomplementary, not counterexpectative. It is hard to find a form of expression which will show them to be such; I have usually hadrecourse 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 complexityof spoken language is in its flow, the dynamic mobility whereby each

figure provides a context for the next one, not only defining its pointof departure but also setting the conventions by reference to which itis to be interpreted.

With the sentence of written language, there is solidarity among itsparts such that each equally prehends and is prehended by all theothers. It is a structure, and is not essentially violated by beingrepresented synoptically, as a structural unit. With the clause complex,of spoken language, there is no such solidarity, no mutual prehensionamong all its parts. Its mode of being is as process, not as product. Butsince the study of grammar grew out of writing – it is when languagecomes to be written down that it becomes an object of study, notbefore – our grammars are grammars of the written language. We havenot yet learnt to write choreographic grammars; so we look at spoken

language through the lens of a grammar designed for writing. Spokendiscourse thus appears as a distorted variant of written discourse, andnot unnaturally it is found wanting.

For example, Chafe (1982) identifies a number of regular differencesbetween speech and writing: writing is marked by more nominaliza-tion, more genitive subjects and objects, more participles, more attrib-utive adjectives, more conjoined, serial and sequenced phrases, morecomplement clauses, and more relative clauses; all of which he sum-




337

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. ButChafe has described both speech and writing using a grammar of

writing; so it is inevitable that writing comes out with positive checksall round. Not that he has no pluses on the spoken side: speech is saidto 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 languageas interpersonal action. Chafe summarizes them as features of “involve-ment” as opposed to “detachment”; but they are items of low general-ity, and negative rather than positive in their social value.

This leads me to the second point that, as I remarked above, runscounter to our received attitudes towards speech. It is not only thatspeech allows for such a considerable degree of intricacy; when speakersexploit this potential, they seem very rarely to flounder or get lost in

it. In the great majority of instances, expectations are met, dependenciesresolved, 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 disorganizedarray of featureless fragments? Here it is not just the lack of aninterpretative 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 so-called performance errors; these are regularly, more or less ritually,cited as its main distinguishing feature. There is no disputing the factthat 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 academicseminars, where I suspect much of the observation and recording hastaken place. If you are consciously planning your speech as it goesalong and listening to check the outcome, then you naturally tend tolose your way: to hesitate, back up, cross out, and stumble over thewords. But these things are not a particular feature of natural spon-taneous discourse, which tends to be fluent, highly organized and




338

grammatically well formed. If you are interacting spontaneously andwithout self-consciousness, then the clause complexes tend to flowsmoothly without you falling down or changing direction in the

middle, and neither speaker nor listener is at all aware of what ishappening. I recorded this kind of casual discourse many years agowhen 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 thephonological level: spontaneous discourse is typically more regular inits patterns of rhythm.

However, while the myth of the scrappiness of speech may havearisen at the start from the kind of discourse that was first recorded, ithas been perpetuated in a different way – by the conventions withwhich 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) youknow 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 milli-

seconds 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 structureand feeling.

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




339

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 isimportant to show the planning, trial and error, and revision work thathas gone into the production of a piece of discourse: it can have botheducational and clinical applications. This is as true of writing as it is of speech: written material of this kind has been used in neuropsychiatry




340

for most of a century. But for many purposes the discarded firstattempts are merely trivial; they clutter up the text, making it hard toread, and impart to it a spurious air of quaintness. What is much more

serious, however, is that transcribing spoken discourse in this way givesa 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 nobodybut themselves. But unfortunately this is not the way. Just when weare seeing real collaboration between linguists and educators, and theconception of “language in education” is at last gaining ground as a

field of training and research, it seems we are determined to put theclock back to a time when spoken language was not to be takenseriously 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 thecontinuum 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 modalpoints “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 impliedonce 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 toarise when attention is being paid to the process of text production.Since highly monitored discourse is typically written, these features areactually more characteristic of writing than of speech; but because mostwritten text becomes public only in its final, edited form, the hesitationsand discards are lost and the reader is shielded from seeing the processat work. Where they are likely to remain in is precisely where theyoccur least, in the more spontaneous kinds of writing such as personalletters. (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 conver-sation, where the well-recognized characteristics are those of turn-taking, such as interruptions and overlaps. But the strictly linguistic“deviations” of casual conversation are mainly systematic features thatwould not seem deviant if we had a grammar that took into accountthe specifically “spoken” resources of the linguistic system.)

Spoken and written language do differ, however, in their preferred




341

patterns of lexicogrammatical organization. Neither is more organizedthan the other, but they are organized in different ways. We havealready identified the principal variable. Spoken language tends to

accommodate more clauses in the syntagm (to favour greater “gram-matical intricacy”), with fewer lexical items in the clause. Writtenlanguage 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 clausesper clause complex will be greater in spoken language, because theremay also be a tendency towards very short ones, especially in dialogue.

It would be better to say that the greater the intricacy of a clausecomplex the more likely it is to be a product of spontaneous speech.)We must now return to this distinction in order to look through andbeyond 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 understand-ing 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 indoctri-nation 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 non-dogmatic 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 deepunderstanding of [[what Christianity is about ]] should not be bracketed




342

simply with the evangelizing aim [[to which I referred earlier ]]. ||| It is not absolutely incompatible with that aim, however, for the followingreason.||| [[What counts as indoctrination and the like ]] depends upon

a number of criteria,|| to do with the degree [[to which a teacher fails tomention 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 openone.||| The non-dogmatic teacher may be tepid;|| the open one may

be fervent.||| Fervour and indifference are not functions of closednessand openness.|||

To see how this lexical density is achieved, we can look at the firstclause. 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 nominalgroups, mainly one inside the other, and ending with an embeddedclause:

| the sympathetic induction [ of [ people ] ]] [ into [ a proper and deepunderstanding [ of [[ [ what ] < Christianity | is > about ]] ] ] ] |group or phrase boundary |embedded group or phrase [ ]enclosed elements < >(the prepositional phrase what . . . about is discontinuous, the itemsChristianity 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 lexicalitems the only ones that do not occur in nominal groups are bracketed,

simply, depends, do, and differ. It is a characteristic of written discoursethat most of the lexical information is encoded in nominal form: thatis, in nominal groups, with their structure potential of Head (typically

a noun or adjective), Premodifier (typically adjectives and nouns), andPostmodifier (typically embedded phrases and clauses, which then havefurther 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 [ Christianityor Humanism ] ]] ] ] ] ] |




343

which like the first one involves considerable embedding, to simplenominal groups such as tepid, the open one, fervour and indifference . But itis 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 nominalgroup. Since the lexical items have to go somewhere, lexical density isaccompanied 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 longnominal 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 isa nominalizing device, we have hypotaxis, which is a form of interde-pendency between clauses; and this points up the difference betweenthe two variants.

This difference is obscured, on the other hand, if the grammar failsto 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, adverbialclause). In other words, being a grammar of written language, itrecognized 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 toembedding in the strict sense, and so distinguish it from the inter-dependency relation of hypotaxis, where one element is dependent onanother but is not a constituent of it. Hypotaxis is more like parataxisthan it is like embedding; and both are characteristic of spoken rather than written language. So in order to do justice to the particular modeof organization of both spoken and written discourse, the grammar




344

needs to distinguish between the constituency relation of embedding,or rankshift, where one element is a structural part of another, and thedependency relation of ‘taxis’, where one element is bound or linked

to another but is not a part of it. Either of these relations can bereduced to a form of the other one, but only at the cost of distortingthe 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 asymptom, 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 thathave 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 hypotaxisand embedding is one that can be expressed in terms of the familiar

opposition of process and product. Written language representsphenomena as if they were products. Spoken language representsphenomena as if they were processes (see the discussion in Martin1984b).

In other words: speaking and writing – each one makes the worldlook like itself. A written text is an object; so what is represented inwriting 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 thatit happened or was done. So, in Text 3 above, the written variant tellsthe story in nouns: visit, sense, risk, attempt, action; whereas the spokenversion 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 synoptic-ally: it exists, spread out on the page. So the reader is predisposed totake a synoptic view of what it means; behind it is a tableau – like thepictures from which writing originally evolved. But when one islistening, the text reaches one dynamically: it happens, by travellingthrough the air. So the listener is predisposed to take a dynamic viewof what it means; behind it is a film, not a picture.




345

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 threeinterrelated subsystems: a semantic system (meanings), coded into alexicogrammatical system (wordings), recoded into a phonologicalsystem (sounds). A language that has a writing system has an alternativeform of expression: visual symbols as well as sounds. In such a language,a written text could, in principle, be a spoken text that has beenwritten down (a transcription); here the written version is a transcod-

ing of something that has already been coded in sound. Most writing isnot like this. Secondly, a written text could be an alternativeexpression of a given wording: in this case meanings are coded aswords and structures (“wordings”), which are then expressed either insound or in writing. If this was the norm, there would be no systematicdifference between spoken and written texts; the medium would notbe a significant register variable. But there are such differences; so, tosome 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 appropri-ate to the dynamic nature of the text process, the latter appropriate tothe synoptic nature of the text product. This is the sort of interpretationwe have been offering.

But is it the whole story? There is still a fourth possibility – thatspeech and writing can diverge already at the semantic level, so thatspoken and written discourse embody different meanings. Is there anysign that this can happen? It would of course be only a very partialeffect; no one has suggested that the two derive from different semanticsystems (or even two different lexicogrammatical systems, for thatmatter). But we should consider the possibility that there is someflowback 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 beenbringing quite a few private civil actions, which have been mainly of twokinds.




346

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 froma 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 arisebecause nominal constructions fail to make explicit many of thesemantic relations that are made explicit in clause structure. Writtendiscourse conceals many local ambiguities of this kind, which arerevealed when one attempts a more “spoken” paraphrase.

But the final sentence of Text 2 illustrates another significant featureof written language, which can be seen in the wording popular concernover the environment has stimulated private civil actions. We reworded thisas 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 asingle happening, and what happened was that A brought about B. ButA and B are themselves nominalized processes. The meaning of stimulate here is as in pruning stimulates growth. The spoken version represents thethesis 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 tolook like a process of a different kind – or, in this instance, two




347

processes, one mental and one material, have been dressed up as onewhich is neither. This coding of a semantic relation between twoprocesses as if it was the single process is very common in writing; the

sentence immediately preceding Text 2A contained another exampleof the same thing, here with the verb leads to:

A successful tort action leads to a judgment of damages or an injunctionagainst the defendant company.

But this is just one type of a more general phenomenon, somethingthat I call grammatical metaphor (Halliday 1985, Chapter 10). Written

language tends to display a high degree of grammatical metaphor, andthis is perhaps its single most distinctive characteristic.Here are three further examples of grammatical metaphor taken

from various written sources, together with suggested rewordingswhich 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 whencredit 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 wouldnot be healthy, particularly when the economy is depressed as it isnow.”

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 heused to be big.”

In all these examples nominalization plays a significant part, as itdoes in many types of grammatical metaphor; so it is perhaps worthstressing that nominalization is well motivated in English. It is not

simply a ritual feature that has evolved to make written language moreambiguous or obscure; like the passive, which is another feature whosefunctions are widely misunderstood, nominalization is an importantresource for organizing information. Take the example youth protest mounted , which is not a headline but a complete sentence from a featurearticle. We might reword this as more and more young people protested , or young people protested more and more ; but the only way to get thecombination of youth and protest as the Theme of the clause is by means




348

of a nominalization (not necessarily such a laconic one; it might havebeen the protests of the young people , but this is still a nominalizing of theprocess). So while there is a price to be paid, in that the information

being conveyed may become mildly (and sometimes severely) ambigu-ous, there is also a payoff: more choice of status in the discourse. Interms of systemic theory, there is a loss of ideational information, but a

gain in textual information. This of course favours the specialist: youneed to know the register. If you do not know the register you maymisinterpret the thesis, so the fact that it is highly coded as a message isnot very helpful to you; but if you do know it you will select the right

interpretation automatically, and the additional “functional sentenceperspective” 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 areabstractions that can enter into the structure of a clause – civil actionscan be brought, an injunction can be issued – but cannot themselvesbe 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 becomepart of our ideology, our way of knowing about the world we live in.

Patterns of this kind invade the spoken language and then act asinfiltrators, providing cover for other metaphorical nominalizations – which are still functional in speech, but considerably less so, becausespoken language has other resources for structuring the message, suchas intonation and rhythm.

Grammatical metaphor is not confined to written language: quiteapart from its tendency to be borrowed from speech into writing, thereare specific instances of it which seem clearly to have originated inspeech – most notably the pattern of lexically empty verb with theprocess expressed as “cognate object” (Range) as in make a mistake ‘err’,have a bath ‘bathe’, give a smile ‘smile’. But in its principal manifestationsit is typically a feature of writing. Writing – that is, using the writtenmedium – puts distance between the act of meaning and its counterpart

in the real world; so writing – that is, the written language – achievesthis distance symbolically by the use of grammatical metaphor. It isoften 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 meta-phorical quality. If I say technology has improved , this is presented as amessage; it is part of what I am telling you. If I say improvements intechnology, I present it as something I expect you to take for granted.




349

By objectifying it, treating it as if it was a thing, I have backgroundedit; 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; itbackgrounds, using discourse to create the context for itself. This iswhy in the world of writing it often happens that all the ideational

content is objectified, as background, and the only traces of process arethe relations that are set up between these taken-for-granted objects. Irecall a sentence from the O.S.T.I. Programme in the LinguisticProperties of Scientific English (Huddleston, Hudson, Winter and

Henrici 1968) which used to typify for us the structures found inscientific 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 anassumption; in fact each mode is metaphorical from the standpoint of the other, and the fact that the spoken is developmentally prior – theindividual listens and speaks before he reads and writes – while it meansthat the language of “process” is learnt first, does not guarantee that itis in any sense “closer to reality”. It might be a hangover from anearlier stage of evolution, like the protolanguage that precedes the

mother tongue. But personally I do not think so. I am inclined tothink the written language of the future will go back (or rather forward)to being more processlike; not only because the traditional objectlikenature of written discourse is itself changing – our reading matter istyped into a memory and fed to us in a continuous flow as the lines

follow each other up the screen – but also because our understandingof the physical world has been moving in that direction, ever sinceEinstein 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.




350

Meanwhile, grammatical analysis shows spoken and written English tobe 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 thatcombine to distinguish spoken and written discourse can be shown tobe related and encompassed within a single generalization, only when

we express this generalization in semantic terms – or at least in termsof a functional, meaning-oriented interpretation of grammar. Speechand writing will appear, then, as different ways of meaning: speech asspun out, flowing, choreographic, oriented towards events (doing,

happening, sensing, saying, being), processlike, intricate, with meaningsrelated serially; writing as dense, structured, crystalline, orientedtowards things (entities, objectified processes), productlike, tight, withmeanings related as components.

In their discussion of the comprehension and memory of discourse,Hildyard and Olson (1982: 20) suggested that meaning is preserved indifferent 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 isreceived, rather than of the linguistic features of the code that liesbehind it. The notion of different ways of meaning implies, rather, thatthere are different ways of knowing, and of learning. Spoken andwritten language serve as complementary resources for acquiring andorganizing knowledge; hence they have different places in the educa-tional process. Teachers often know, by a combination of intuition and

experience, that some things are more effectively learnt through talkand others through writing. Official policy usually equates educational

knowledge with the written mode and commonsense knowledge withthe spoken; but teachers’ actual practice goes deeper – educationalknowledge demands both, the two often relating to different aspects of the same phenomenon. For example: definitions, and structural rela-tions, are probably best presented in writing; demonstrations of howthings work may be more easily followed through speech. The twofavourite strategies for describing the layout of an apartment, reportedin the well-known study by Linde and Labov (1975), would seem to




351

exemplify spoken and written modes of symbolic exploration. We mayassume that speech and writing play different and complementary partsin 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 themother tongue; (2) the processing of new meanings into the system;

(3) the interaction between learning elements that are ready coded andlearning the principles of coding; (4) the relation between system andprocess in language; (5) the unconscious nature of linguistic categories;(6) the social construction of reality through conversation; (7) linguisticstrategies used in learning; (8) the development of functional variation,or registers; (9) the relation between everyday language and technicallanguage; and (10) the development of generalization, abstraction, andmetaphor. 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 acentral issue in any learning theory which has language as its primaryfocus.



352

Chapter Thirteen

HOW DO YOU M EAN? (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 aremeanings 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 tointervene in the processes we are trying to understand. And someprocesses 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 theinstance; it is an intrastratal relationship – that is, it does not involve amove between strata. The wording fine words butter no parsnips is aninstance, or an instantiation, of a clause. Realization, on the other hand,is prototypically an interstratal relationship; meanings are realized aswordings, wordings realized as sound (or soundings). We often use theterm to refer to any move which constitutes a link in the realizationalchain, even one that does not by itself cross a stratal boundary (for example, features realized as structures); but the phenomenon of realiza-tion only exists as a property of a stratified system. To anticipate the

discussion a little, I shall assume that realization may be formalized asmetaredundancy, as this is defined by Jay Lemke (1985). Instantiation Ishall define by making reference to the observer; it is variation in theobserver’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 byMartin Davies and Louise Ravelli. London: Pinter, pp. 20–35.



h ow d o y ou m ea n?

353

form of consciousness

action reflectiondomainof experience

1st/2nd person regulatory interactional

3rd person instrumental personal

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, obviouslyincluding the so-called “higher” mammals. In humans, meaning devel-ops, in the individual, before the stage of language proper; it beginswith what I have called “protolanguage”. So where does this mammalianexperience come from? It probably evolved out of the contradiction

between the two primary modes of experience, the material and theconscious. Material processes are experienced as ‘out there’; consciousprocesses are experienced as ‘in here’. We can see in observing thegrowth of an individual child how he or she construes this contradictionin the form of meaning. The child constructs a sign, whereby the onemode 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-pitchedsqueak), Nigel’s first sign that he produced at five months old; I glossedit as “what’s that? – that’s interesting”. In other words, Nigel wasbeginning to construe conceptual order out of perceptual chaos: ‘I amcurious (conscious) about what’s going on (material)’. This impact of the material and the conscious is being transformed into meaning by aprocess of projection, in which the conscious is the projecting and thematerial 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’, andthe other that of action, ‘I want’ – one the way things are, and theother the way they ought to be. There are also, as it happens, twodomains 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 aroundeight months, with Nigel), what is construed into meaning is not asingle sign but a two-dimensional semiotic space constituting a signsystem (Figure 1). We can justifiably refer to such a sign system as a




354

“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 throughthe projecting of the material on to the conscious, in a single two-dimensional construction. And this becomes possible because theconscious mode of experience is the social mode. We have oftenpointed out that it takes two to mean; but we still tend to refer toconsciousness as if it was an individual phenomenon, with the social asan 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, throughthe projection of the one by the other, become fused and transformedinto something that is new and different from either. We can think of this as creating a “plane of content” in the Hjelmslevian sense. If welook at this process dynamically, it is meaning-creating, or semogenic.If we look at it synoptically, as a relation construed by this process, it issemantic; 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’. Butthis is misleading. Our experience is at once both material and conscious;and it is the contradiction between the material and the conscious thatgives these phenomena their semogenic potential. The other ‘face’ isthe meaning – the signified, if you prefer the terminology of the sign.Many years ago I did my best to gloss the child’s protolinguistic meaningsusing “adult” language as metalanguage, and found myself forced intousing glosses like ‘nice to see you, 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 microfunctionalcategories just referred to. But those categories themselves were notinterpreted further. I think they can now be explained at this somewhatdeeper 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 secondinterface, a transformation back into the material, or (again, rather) intothe phenomenal – this time in its manifestation in the meaning subject’sown body: as physiological processes of articulation or gesture. This isthe phonetic / kinetic interface; the “expression plane”, in Hjelmslev’sterms. Since there can be no meaning without expression (meaning is

intersubjective activity, not subjective), the act is “doubly articulated”,




355

in Martinet’s terminology: it is the transduction of the phenomenalback into the phenomenal via these two interfaces of content andexpression. (Transduction not transformation, because as Lamb (1964)

pointed out many years ago in transformation the original is lost, ceasesto exist. And again I am suggesting that we should conceive of it asphenomenal rather than material, since both the ‘outer’ faces, that of

the content substance on the one hand and that of the expressionsubstance on the other, embody both the material and the consciousmodes 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 thatthe ‘out there’ dimension is that of person / object, provided weremember that “object” includes those persons ‘treated as’ object, i.e.third persons. Instantially, this means any person other than whoever isthe 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 bycalling it mammalian; but I am using this as a way of saying that it is apotential we hold in common with other creatures, which I think israther important. It is a rich semogenic potential; but it is also constrainedin certain critical respects. In our own specifically human history, inboth phylogenetic and ontogenetic time, it comes to be deconstructed – or rather deconstrued – and reconstrued as something else, this time inthe form of a potential for meaning that is effectively infinite, or at leastunbounded (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 theinterfaces. If we put it in terms of even more concrete metaphors, whathappens is that an entirely non-material (again, better: non-phenomenal)

system is slotted in between the two material / non-material (phenom-enal / 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 / actionand person / object (or intersubjective / objective). By ‘grammatical-




356

izing’ the process of meaning – reconstruing it so that the symbolicorganization is freed from direct dependence on the phenomenal, andcan develop a structure of its own – the collective human consciousness

created a semiotic space which is truly elastic, in that it can expand intoany number of dimensions. (We will model this more explicitly in amoment.) The immediate effect is to re-form the reflection / actionopposition into a simultaneity, such that all acts of meaning embodyboth – i.e. both reflection and action – not just as components, but assets of options, each constituting a distinct dimension of choice. Inother words they now evolve into the metafunctional categories of

ideational and interpersonal. The ideational is the dimension that isprimarily reflective (the construction of experience), the interpersonalthat which is primarily active (the enactment of social processes5). (Butnote that each engenders the other mode as a secondary motif: the waywe construe experience (by verbal reflection) disposes us to act incertain ways, e.g. as teachers structuring the role relationships in thelearning process, while the way we construct our social relations (byverbal 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 theprotolinguistic sign. The result is a semiotic of a new kind: a stratified,tristratal system in which meaning is ‘twice cooked’, thus incorporatinga 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 turnrealized by sound’. But it is also rather seriously misleading. If we followLemke’s lead, interpreting language as a dynamic open system, we canarrive at a theoretically more accurate and more powerful account. Herethe key concept is Lemke’s principle of metaredundancy.6

Consider a minimal semiotic system, such as a protolanguage – asystem 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” respec-tively 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. Thisrelationship 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 tri-rather than bi-stratal. The expressions a, b, c now realize wordingsl, m, n while the wordings l, m, n realize meanings p, q, r. In termsof redundancy, however, these are not two separate dyadic relationships.




357

Rather, there is a metaredundancy such that p, q, r redounds not withl, 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, wewould 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 realizationwas a causal relation, then it would chain: l is caused by a and p iscaused by l – it would make no sense to say “p is caused by the causingof l by a”. But realization is not a causal relation; it is a redundancyrelation, so that p redounds with the redundancy of l with a. To put itin 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 bythe realization of wording in sound.

We can of course reverse the direction, and say that sounding realizesthe 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 appropriateperspective. But for the purposes of construing the ‘higher’ levels, withlanguage as connotative semiotic realizing other semiotic systems of theculture, 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 withthe 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 aniterative one and so opens up the possibilities for construing, not onlythe context of situation, but also higher levels such as Hasan’s symbolicarticulation 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 evo-lutionary change from protolanguage to language – whereby instead of




358

Figure 2 Metaredundancy

a simple plane with two interfaces to the material (the phenomenal),we have constructed a semiotic space, a three-dimensional (potentiallyn-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 inthe transduction from one purely symbolic mode to another, and hencenot constrained by the need to interface directly with the phenomenal,

we could not have a metafunctional organization in the grammar, andwe 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 themsynoptically: by treating realization as a relation. Now, a system of thiskind could still remain fully closed: it could be a circular, self-regulatingsystem without any form of exchange with its environment. But a

language, as Lemke pointed out, is a dynamic open system; such systemsare not autostable, but metastable – they persist only through constantlychanging by interpenetration with their environment. And in order toexplain a system of this kind we have to complement our synoptic




359

interpretation with a dynamic one. This leads us into the other criticalconcept, 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 thesepersist, and they differ in systematic ways. Yet we are all very concernedabout changes in the climate, and the consequences of global warming.

What does it mean to say the climate is changing? Climate is instantiatedin the form of weather: today’s temperature, humidity, direction andspeed of wind, etc., in central Scotland are instances of climaticphenomena. As such they may be more, or less, typical: today’s

maximum is so many degrees higher, or lower, than average – meaningthe 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 centchance, 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 nomore, and no less, than the pattern set up by the instances (the weather),and each instance, no matter how minutely, perturbs these probabilitiesand 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 atdifferent distances – different time depths. To the climate observer, theweather looks like random unpredictable ripples; to the weather observer, the climate is a vague and unreal outline. So it is also withlanguage;8 language as system, and language as instance. They are nottwo different phenomena; they are the same phenomenon as seen bydifferent observers. The system is the pattern formed by the instances;and each instance represents an exchange with the environment – anincursion into the system in which every level of language is involved.

The system is permeable because each instance redounds with thecontext of situation, and so perturbs the system in interaction withthe environment. Thus both realization and instantiation are involvedin the evolution of language as a dynamic open system.

Now the relation of system to instance is in fact a cline, a continuouszoom; 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 whichend we started from. To the system observer, history takes the form of evolution; the system changes by evolving, with selection (in the senseof ‘natural selection’) by the material conditions of the environment.This is seen most clearly, perhaps, in the evolution of particular sub-




360

Figure 3 A model of semogenesis

systems, or registers, where features that are functionally well adaptedare positively selected for; but it appears also in the history of the systemas a whole once we look beyond the superficial clutter of randomfluctuations into the grammar’s cryptotypic core. To the instanceobserver, on the other hand, history is individuation: each text has itsown history, and its unique meaning unfolds progressively from thebeginning. (Note that the probability of any instance is conditionedboth 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 grammaticalmetaphor – we may be able to track it through both these histories, the

phylogenetic – its history as it evolves in the system; and what we mightcall the “logogenetic” – its history as it is built up in the course of thetext. There is of course a third kind of history, the ontogenetic, whichis different again – the cladistic model here is one of growth. This toois a mode of semogenesis; and we could follow through with the sameexample, asking how grammatical metaphor comes into being in thedevelopmental history of a child. These are in fact the three modes or dimensions of semohistory – the phylogenetic, the ontogenetic and the




361

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 thismeaning evolve, in the system? how did it develop, in the learner? andhow 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,




362

Figure 5 Microfunctional systems (Nigel’s protolanguage)

or is creolized; the individual dies, or becomes aphasic; the text comesto an end, or is interrupted.) Now, the mechanism of this increase of

meaning potential may be modelled in the most general terms as inFigure 3. Nesbitt and Plum (1988; see also Halliday 1991) showed howto do this in a corpus-based study of ‘direct speech and indirect thought’(the intersection of speech projection and thought projection with theinterdependency system of parataxis and hypotaxis). This and other postulated examples of evolutionary semogenesis are set out in Figure4. These involve relations between strata (semantics realized in gram-mar); 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 metare-dundancy relation: this is the only way it can be nudged along. Thuswhere a closed system is self-regulating (autostable) and circular, anopen system is other-regulated (metastable) and helical. And it is through

the combination of these two relations or processes, instantiation on theone hand and realization on the other, that the system exchanges withits environment, creating order in the course of this exchange and soincreasing its potential for meaning.

Thus the possibility of meaning – of acting semiotically – arises atthe intersection of the material (or phenomenal) with the conscious, asthe members of a species learn to construct themselves (“society”) inaction and to construe their experience in reflection. These two dimen-




363

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 two-

dimensional (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 con-

sists in making the transduction between them. The simple signs of theprotolanguage shape themselves into a sign system as they cluster in thefour quadrants of this semiotic plane; there is thus already a proto-system network, which we could set up in an idealized form as inFigure 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 variablesare 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 interac-tional: 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 ann– dimensional space, as the activity of meaning becomes dialogicallydynamic and metafunctionally complex: that is, it becomes possible tomean more than one thing at once, and to construe meanings into text.I have written elsewhere about how Nigel took the first step in thistransformation, using the semogenic strategy already described (combin-

ing two functionally distinct variables); it is summarized in Figure 6.




364

Here for the first time Nigel is selecting in two systems of meaning atonce, 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 fewprotolanguage remnants that persist into adult life like hi! and ah! andyum! 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 aswording (the ideational as contrasting names; the interpersonal as

contrasting mood) and ‘then’ (re)construed as sounding (names asarticulation, mood as intonation). At this second interface the child cannow combine the segmental and the prosodic choices, in this way bothrealizing 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 thematerial, as mutually contradictory forms of experience, then it is notsurprising that when experience is construed semantically, these twotypes of process, the material and the conscious, should come to besystematically distinguished. But there is a further twist. The semogenicprocess, as we saw, involves setting up a relationship between systemssuch that one is the realization of the other – that is, they stand to eachother in a relation of Token and Value. This Token–Value relationshipis set up at both interfaces, and it is also what makes it possible to prisethe two apart and wedge in a grammar in between. Here then we findthe third of the kinds of process construed by the grammar: the relationalprocess, 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 isultimately derived.

Let me return once again, finally, to the suggestion that meaning is amode 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.




365

When protolanguage evolves into language, with the stratal dimensionof realization, meaning becomes self-reflexive: and in two senses. Onthe one hand, it imposes order on itself: the textual metafunction, as

Christian Matthiessen (1992) has shown, construes a reality that is madeof 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 ametatheory in one sense of this term. At this very general level, we canthen 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 thehistory of ideas, with arguments about language – in this case arisingfrom 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 thetwo merely a function of the observer, so that patterns repeat if we waitfor them long enough, probabilities become certainties when we know

all that needs to be known? Or to put this in more specifically linguisticterms, will all the various contradictions in the grammar resolve them-

selves into some higher level of order? – I mean things like transitiveand ergative as complementary theories of process, or tense and aspectas complementary theories of time; as well as all the other indetermin-acies which arise in our polyelastic semiotic space? I am not of coursesetting out to answer these questions; I am merely pointing out that themeaning potential we have evolved for ourselves construes the possibilityof asking them. But I will allow myself one further thought in theclosing paragraphs of the paper.

It is a human failing that we usually try to impose order much toosoon. There are many examples of this in recent linguistics.10 Theattempt fails; and we then resort to ‘theories of chaos’, trying to makesense of things while remaining instance observers – looking for ‘unetheorie de la parole’, so to speak.11 Such constructs are ultimately self-

contradictory; but they serve as a way of reformulating the questionsand allow us to move back a bit, to shift our stance. A good exampleof the overimposition of order through language is provided by adesigned, 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




366

seeking ways to restore the balance: a discourse with which to construeexperience in terms of indeterminacy, of continuity and of flux. Nowby comparison with the languages of science, the ordinary evolved

language of everyday life has many of these properties. It is orientedtowards events rather than objects, and is in many respects fluid andindeterminate. But it is important to be aware that even our most

unconscious everyday language also imposes an order which we mayneed to re-examine and to deconstruct. To return to the weather: wecan say it’s raining or we can say it’s snowing – but we have to decidebetween 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 theinfluence of spoken language (Johnston 1990).) In other words, whilethe 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 notnecessarily the most functional for all times and all circumstances;

especially at times of rapid change like the present, we may need tohold it up to the light and see how it works.13 It is easy to remainunaware of the stories our grammar is telling us.

One thing I have been trying to do, in this paper, is to use thegrammar to think with about itself. Not just in the usual sense, of usinglanguage as its own metalanguage; of course I am doing that, becausethere is nothing else I can do. I mean this more specifically in the senseof using what I have called the grammatics – the concepts that we havedeveloped in order to interpret the grammar – as a means towardsunderstanding the nature and evolution of language as a whole. Thestrategy is that of treating language as ‘other’ – as if it was a differentkind of semiotic that the grammar was being used to explore.14 Thus Ihave found it helpful to think of meaning as the way consciousness (thatis, 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 oneend, and those of our bodily performance – gesture, articulation – atthe 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 anyonesays to me, “How do you mean?”




367

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 another; and the result is an act of meaning – such as Nigel’s very high-

pitched 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’ inLemke (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




368

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 asignificantly different meaning.

9. See Painter (1984), Oldenburg (1987). For an investigation of Chinese-

speaking children see Qiu (1985).

10. As pointed out by John Sinclair (1992).

11. 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.

12. But note Tigger’s defence in Winnie-the-Pooh: “You shouldn’t bounce somuch.” “I didn’t bounce; I coughed.” “You bounced.” “Well, I sort of

boffed.”

13. In a recent paper (Halliday 1990) I suggested that our present grammars

are in some respects environmentally unsound.

14. 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).



369

Chapter Fourteen

GRAMMAR AND DAILY L IFE:

CONCURRENCE AND COMPLEMENTARITY

(1998)

Let me first say what I mean by “grammar” in the title of the paper. Imean the lexicogrammatical stratum of a natural language as tradition-ally understood, comprising its syntax and vocabulary, together withany morphology the language may display: Lamb’s “lexical system”, in

his current (1992: Chapter 5) ‘three-level architecture’ – in common-sense terms, the resources of wording in which the meanings of alanguage are construed. And here I have in mind particularly theevolved, 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 aswhat they are: an outgrowth, supported by design, of the originalgrammar 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 thephenomenon itself. So while the term “grammar” is commonly usedin the way in which I have defined it, to mean the wording system,the central processing unit of a natural language, it is also usedindiscriminately to mean the study of that system: grammar

2 meaning

‘the study of grammar 1’. Since the study of language is called “linguis-

tics”, 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 byTeun A. van Dijk. Amsterdam: John Benjamins, pp. 221–37.




370

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 atheory of experience, because it is also something else besides. Like anyother theory, a grammar is something to think with. It is through

grammar that we make sense out of our experience, both of the worldwe live in (what we experience as taking place “out there”) and of theworld 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 andMatthiessen 1999).During the past twenty years leading neurobiologists, such as Harry

Jerison and John Allman, have been investigating the way the brainevolved; and they explain its evolution as the evolution of the organ-ism’s resource for constructing reality. Changes in the ecologicalenvironment require changes in the representation of experience (Edel-man 1992; Lemke 1993). One critical step was the evolution of the

cerebral cortex, which transformed the mammalian map of the externalenvironment. The second was the evolution of language, which added

a new dimension to reality, that of introspective consciousness; thislatter 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 generaand species, is that it has a grammar, an abstract stratum of coding inbetween the meaning and the expression. Grammar is what bringsabout the distinctively human construction of reality; and by the sametoken, grammar makes it possible for us to reflect on this construction.

As a teacher I have often said to my students that they should learnto ‘think grammatically’. By this I mean that they should use theunique power of the human brain to reflect on the way their experi-ence is construed in their grammar: use grammatics to think about

what grammar thinks about the world. I suggest they might do thiswith problems of any kind, such as relationships with family andfriends, or whether to go for the job that pays more or for the onethey would more enjoy. Let me give a small example of what I meanby thinking grammatically. You’re feeling a bit down. What’s thematter, 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,






372

I’m dazzled, you know . . . It’s being able to see your working lifewill span a period in which so-&-so is the topmost industry which Ifind 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 theone that’s suffering, so the Theme of the clause should more appropri-ately 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 athing separated from myself, something that I possess, with my headidentified as its location: I have an ache in my head . Better still, if myhead is used as a classifier, the ache and its location become a singlecomplex thing; and this now occupies the culminative position in theclause: I have a headache. The flow of information here is very differentfrom 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. Butthis pattern has evolved in English as the prototypical form for construing bodily qualities and states; rather than her hair is long, histhroat is sore , we tend to say she has long hair, he has a sore throat , puttingthe person rather than the body part into the thematic role.1 And incertain other languages where initial position is thematic we alsoregularly find the person, rather than the body part, lodged at thebeginning of the clause. The overall patterns are of course different: inparticular, there may be no strong bond between Theme and Subject,and this makes it clear that the relevant function is that of Theme. Wecan give examples from Chinese, Russian and French. In Chinese it ispossible to say wodi tou teng ‘my head aches’, where as in the Englishwodi tou ‘my head’ is a single element in the clause and so functions as

the Theme. The preferred form, however, is wo tou teng ‘me the headaches’, where the ‘head’ is detached from the personal pronoun; wo

‘me’ and tou ‘head’ are now independent elements in the clause andonly the first one, wo , is thematic. Again, this is the typical pattern for all such expressions in Chinese: ta toufa chang ‘her the hair (is) long’, ta houlong tong ‘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 thepreferred form. Russian however displays a different pattern: u menja

golova bolit ‘at me the head aches’, where again it is the ‘me’ that hasthematic status. In French instead of ma tete me fait mal ‘my head is



g r am m ar a n d d a ily li f e

373

hurting me’ one can use possession as in English: j’ai mal a la tete ‘Ihave an ache at the head’. French also has a further device, of detachingthe Theme altogether from the structure of the clause, and announcing

it as a key signature at the beginning: moi j’ai mal a la tete ‘me I’ve gotan ache at the head’. Neither Chinese wo 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, inthe 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 nolonger a fact about me, and my inner self, but an external fact aboutmy head. This won’t make the headache go; but it does put it in itsplace. It has now become a problem of my head, which is just one partof 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 asan 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 generaltheory of meaning: of language as a semiotic system, and of other semiotic systems brought into relation with language. Without such ageneral theory, the excursion into other languages is no more than apiece of tourism; it assumes significance only when we can show howthis small corner of experience is construed in relation to the meaningpotential of each language as a whole.

But this requires much more than a purely local explanation. Takinga fragment of the grammar of daily life, and exploring it cross-linguistically in this way, still leaves it as an isolated fragment, detachedfrom its environment in the overall system of the language. Yet this isthe critical environment to take into account. The grammar construesa 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 variousother grammatical features, many of them not obviously related in anyformal sense, which are associated topologically within this semanticspace (cf. Martin and Matthiessen 1992). Such features may cluster intoa recognizable syndrome, needing to be interpreted not piecemeal butas a whole: this is the principle of “frames of consistency” as formulatedby Whorf. Illustrations of this phenomenon may be found in Hasan’s




374

(1984b) ‘Ways of saying, ways of meaning’, where she shows how thegrammar of Urdu construes experience as collectively shared; and inMartin’s (1988) account of “grammatical conspiracies” in Tagalog. If

we are comparing the different “realities” of one language withanother, it is the syndrome rather than the single feature that is likelyto be significant.

Side by side with such frames of consistency, however, there are alsoframes of inconsistency: regions where the grammar construes a patternout 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. Meta-phorically the grammar is representing the fact that human experienceis too complex, and has too many parameters, to be construed fromany 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 thisas 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 humanexperience referred to above: between what we experience as takingplace in the world outside of ourselves and what we experience asprocesses of our own consciousness – seeing and hearing, liking,disliking, fearing, hoping, thinking, knowing, understanding and thelike. In English, the conscious or mental processes differ from theother, material kind in various respects: (1) they have a less exactpresent time; (2) they presume a conscious being taking part; (3) theydo not fall within the scope of ‘doing’, and (4) they can project – thatis, they can construe any meaning as taking place in someone’sconsciousness (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. Pro-

cesses 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 factcontradictory constructions of the same class of phenomena. Inner experience is complex and difficult to interpret; the grammar offerstwo complementary models, one with the Senser in the more activerole (by analogy with material processes), one with the Senser appearing




375

to be acted upon. Each of these brings out different agnate forms; thegrammar 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 theone 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 generalprocesses of consciousness the grammar came to favour the type of construction in which the Senser, the participant credited with con-sciousness, 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 thegrammar of all inner processes and physiological states tended to orientthe message towards the human, or humanlike, participant – perhapswith ‘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 theactive 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 externalworld). Thus in an arrow pierced his eye the arrow was the natural Subject,and remained Subject even if the narrative required the thing acted onto function as Theme. To use a constructed example, the pattern wasthat of:

The king fell to the ground; his eye an arrow had pierced.

with the Actor remaining as Subject even when displaced from initialposition in the clause. Subsequently, as already noted, this bond

between Subject and Actor was deconstructed and replaced by adifferent bond, that of Subject with Theme; this gave the modernpattern:

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, whichwas followed by a change in the tense system as passive tenses caughtup with the active ones; and various other changes took place besides.What this new alignment of grammatical forces amounted to was thatrelatively less prominence was being given to the structure of the

experience – which partner is the doer and which the done-to, so to




376

speak; and relatively greater prominence to the structure of themessage – which part is the theme, and which part is the newinformation to be attended to. Without trying to go into all the

components of the picture, let me refer briefly to three relateddevelopments.

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 onlyhis 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 intoa thematic portion and the rest. But the construction of the message ismore 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 Givenand 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 inconstruing 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, butwe prefer he made the whole story up; similarly you left the important part out (instead of you omitted . . .), they’ve taken the furniture away (insteadof they’ve removed . . .), and so on. This is the grammar’s way of makingthe happening the main item of news. The news tends to come at theend of the clause; but the happening is typically a verb, and if there aretwo parties to it – an Actor and a Goal, say – it is hard to get the verbat 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 tosplit 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 necessarilycome at the end – is marked out for news value by having a prepositionadded 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 theobservation is made to explain why you now need to borrow from




377

me, you say I gave a lot of money to my brother. The preposition to makesexplicit 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 tothe 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 themcharacteristic of unconscious, spontaneous, everyday linguistic encoun-

ters. These, and others that could be added, form a syndrome, aconcurrence of related developments, that has helped to shape themeaning potential of Modern English, giving the language its charac-teristic flavour – that “certain cut”, in Sapir’s terms, which makes eachlanguage unique. What all these have in common is that they tendtowards giving greater prominence to the organization of discourse asa flow of information, making more explicit how each element is tobe 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 leftimplicit once the concern with the message begins to take over. Most

of these effects are fairly recent in history; they reflect the changingsocial 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 canbe addressed to a stranger, who does not necessarily share the sameexpectations and norms of interaction. They can be written down in abook that is going to be printed in thousands of copies and read bypeople who have never met the author and do not even know who heis. In other words, they are features of a standard language: a form of discourse in which the flow of information will typically be renderedexplicit rather than being taken for granted. (Interestingly, many of these changes appear not to have taken place in the surviving Britishrural dialects.)

Effects like these are not the result of sudden catastrophic changes.They are trends and tendencies in a long process of evolution; and atany given time they are quantitative – changes in the relative frequencywith which this or that pattern is selected from within the system. Thegrammatics is thus a theory of probabilities, in which possible/imposs-ible is only a special case of more and less probable – and a rather uninteresting case, because meaning is a product of choice and whensomething becomes impossible there is no more choice. So, for




378

example, I have a headache is an instance of what is now the moreprobable of two agnate constructions; but in using that form thespeaker is still choosing – choosing, among other things, to map I

rather than my head on to the Theme. If my head aches had becomeobsolete, we could still have used the grammatics to explain why thestructure is as it is; but the grammar would have taken over, and the

significance of using I have . . . in any particular instance would havebeen lost. Hence the semantic features being construed in this waywould gradually disappear – just as the semantic feature construed byselecting you instead of thou in Elizabethan English disappeared after

thou had ceased to be an available alternative, although we can still usethis 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 stratalpattern: 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 levelof meanings that is separate from the lexico-grammatical level” (1988:

6). This embodies the evolutionary perspective that I remarked onabove: the evolution of lexicogrammar was the major innovation thattransformed 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. Ashe remarks, the question “whether or not [the conceptual system]should be considered part of language is . . . relatively uninteresting”: itis absurd to draw boundaries around phenomena under study and thenuse these boundaries to justify one’s intellectual stance. Such metalin-guistic boundaries are like the boundaries drawn by language itself,which as he says (1992: 121) “both help us and hinder us in our effortsto understand the world”. It is these arbitrary features of segmentationand of categorization, imposing syntagmatic and paradigmatic bound-

aries on our construction of experience, that lead to many of whatLamb calls the “thinking disorders” which arise both in everyday lifeand in scholarly life (both in language and in metalanguage). Such“disorders” arise at the interface between these two strata: “the semantic-ally generated infelicities of thinking arise because of differences betweenconcepts and the lexemes which express them” (1992: 162).

I myself take the alternative approach, of treating Lamb’s “concep-tual system” as part of language. This is because I do not think the




379

lexicogrammar is arbitrary in its construction of meaning. The grammar has to impose discontinuity on the flux of experience; but the humancondition – 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 withinone language at different stages in its history: some random, some

resonating with variation and change in human culture. But evenwithin one experiential domain, at any one moment in time, thegrammar has to contend with conflicting and often contradictorydemands; so this same interface accommodates complementarities – in

a sense analogous to that in which Niels Bohr used the term to extendHeisenberg’s uncertainty principle in quantum mechanics. The gram-mar is unable to reduce some aspect of experience to a single construc-tion and so introduces two distinct perspectives, two construals whichare mutually contradictory and yet depend on each other to provide atheory of daily life. An example would be tense and aspect as comple-mentary theories of time. These contradict each other: either time is alinear 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 verydifferent mixtures and proportions, but each amounting to a plausible

theory for coping with the everyday world. Some of these complemen-tarities display the further property that one of the two perspectives isconstrued configurationally, the other iteratively (as multivariate andunivariate structures), thus foregrounding respectively the synoptic andthe dynamic points of view. For example, the way the grammar constructs taxonomies of things involves both locating them in config-urations of properties and modifying them by means of iterativebracketing. The construction of time in English also exemplifies thispoint: the system of aspect is activated once at a time, while the systemof tense allows for successive reentries: present, past in present, futurein past in present and so on. The essence of semiotic complementarityis that it is both objective and subjective: some domain of experienceis 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 thesystem itself.)

One very pervasive complementarity is that in the grammar of agency, where the problem to be solved is: how are the processes inthe external world brought about? One theory, as construed in thegrammar of daily life, is that of “Actor, +/- Goal” – a “doer”, plus,optionally, something else that is “done to”. Thus:




380

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 configurationalmodel; there is no re-entry to the choice of agency. Then there’s asnap, 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 entitythrough which the process is actualized (here the thread), plus, option-

ally, something else as Causer that brings it about. This is an iterativemodel; here the agency relation is construed in such a way that it canrecur.

Thus there are two ways of looking at a process: one according towhich participant a acts, and the action may (or may not) extend toanother participant x (a is the constant, x the variable); the other according to which participant x “eventuates” (that is, permits theprocess to eventuate), and the event may (or may not) be broughtabout by another participant a (x is the constant, a the variable). Thefirst 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 embodysome tension between the two. Transitive and ergative are two pointsof view on the same phenomenon, that of the nature of materialprocesses and the relationship of the participants to the process and toeach other; but they are also two distinct phenomena – some processes

pattern ergatively and others transitively (cf. Halliday 1967–68; Davidse1992). This constitutes another strand in the pattern of changes thathave been taking place in English: type X has tended increasingly toprevail over type A.

Let us follow this up in a related corner of the grammar. When Ilast worked in the United States I was living in Orange County; Ifrequently travelled on the local bus services, and there was a notice onthe buses which read:




381

Federal law prohibits operation of this bus when any passenger isforward 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 youbecome 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 morerecent instances of this type like superannuee and oustee , all of which aremodelled on the pattern of employee ‘person employed’. This forms oneterm in the transitive opposition employer/employee ; the latter form wasderived from the French passive participle and matched up with theEnglish active termination -er , giving ‘the one who is acting’/’the onewho is acted upon’; cf. trainer/trainee . Here the -ee is functioning asparticipant x in type A.

Then there are some instances where rather indirect relationships are

involved: biographee ‘person whose biography is being compiled’, ampu-tee ‘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); andvarious banking terms like advisee, favouree, assignee and so on. Theseare modelled on words like referee ‘person to whom a dispute is referredfor 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 tooriginal country’. All these are like standee . When we examine them,we find that they pattern ergatively: the -ee corresponds to the functionof the Medium in the process, to participant x in a process construedas type X. There is no implication that these are functioning as theGoal: a standee is not someone who has been or is being stood. If thesewere following type A we would have stander, returner, retirer, attender and so on. The pattern is given in Figure 1:

Actor Goal Agent Medium

(type A) -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




382

In itself, each instance is trivial. It does not matter whether we writestander 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 construedand 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, observingfrom different time depths; and, especially where the grammar isunstable (as in the present-day English transitivity system), the cumulat-

ive effect of such instances is very noticeable. On the other hand, theword standee represents one perspective within a complementarity; tounderstand it we have to adopt (unconsciously, as always) a particular stance towards the phenomena we experience as taking place outsideourselves. In this perspective, where standing is grouped with beingtrained (standee, trainee ) rather than with training (trainer ), agency isinterpreted as ‘causing’ rather than ‘doing to’: the variable is not ‘doesthe 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 adifferent 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 enactingour social and personal relationships. These two metafunctions areinseparably interlocked in the system of every language: the grammar does not allow us to perform in one mode without at the same timeperforming in the other.2 In other words, while we are constructing

reality we are also acting on it through our semiotic interactions withother human beings. And this brings me back to the point from whichI began, in defining grammar as the spontaneous, natural grammar withwhich we lead our everyday lives. It is important not to set up adisjunction 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 asbeing something to think with, it is by the same token also somethingto act with. We recognize this as a feature of scientific theories: theyare not ideologically neutral, and this critically affects the domains of scientific praxis. The grammar of daily life is not neutral either. I havetried to suggest elsewhere (Halliday 1990) some of the features of our everyday grammar that seem to me to condition our attitudes, to eachother, to other species, and to the natural environment – certain aspects





384

Chapter Fifteen

ON GRAMMAR AND GRAMMATICS (1996)

1 The problem

Most of us are familiar with the feeling that there must be somethingodd about linguistics. We recognize this as a problem in the interper-

sonal sphere because as linguists, probably more than other profession-als, 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’ somephenomenon; but whereas the natural sciences are language aboutnature, and the social sciences are language about society, linguistics islanguage about language – “language turned back on itself ”, in Firth’soften quoted formulation. So, leaving aside the moral indignation somepeople seem to feel, as if linguistics was a form of intellectual incest,there is a real problem involved in drawing the boundary: where doeslanguage end and linguistics begin? How does one keep apart theobject language from the metalanguage – the phenomenon itself fromthe theoretical study of that phenomenon?

The discursive evidence rather suggests that we don’t, at least notvery 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 professionalbodies 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.



on grammar and grammatics

385

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: astronomersobserve stars, and an expression such as astronomical observations couldequally 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 studyof the psyche – has since slipped across to mean not only the study butalso that which is studied, and we talk about criminal psychology (whichmeans 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 specialdifficulties associated with linguistics. For example, it is a feature of linguistics departments that, in their actual practice, what they teach isoften not so much the study of language as the study of linguistics.(And one of the few fields where the terminological distinction is notconsistently maintained in Chinese is that of grammar, where yufa oftendoes duty also for yufaxue .) There do seem to be special categoryproblems 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 thatembodied in the term grammar . Here the slippage is in the oppositedirection 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 alreadyconfusion enough; it was made worse by the popular use of the termto mean rules of linguistic etiquette (for example bad grammar ). As away of getting round part of the problem I started using the term




386

grammatics – I think the first published occasion was in a discussion of ineffability (see above Chapter 12). This was based on the simpleproportion 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 relationshipof linguistics to language is unproblematic as long as we leave languageundefined; 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 wordis used in a non-technical sense, as in the bad grammar example, onecannot take this usage over to define a domain of systematic study: inso far as it has any objective correlate at all, this would refer to aninventory of certain marginal features of a language defined by the factthat they carry a certain sort of social value for its speakers. We canstudy ethnographically the patterns of this evaluation, and their place inthe social process; but that is a distinct phenomenal domain. Grammatics,

in fact, has no domain until it defines one for itself (or until one isdefined for it within general linguistics – exactly at what point the term

grammatics takes over from linguistics is immaterial). And it is this thatmakes the boundary hard to draw. Since both the grammar and thegrammatics are made of language, then if, in addition, each has to beused 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 tothink this through: it doesn’t matter. It does no harm if we just talkabout grammar without any clear distinction between the thing and thestudy of the thing. They are in any case much alike: if you turnlanguage back on itself, it is bound to mimic itself in certain respects.But this comforting dismissal of the problem was belied by my ownexperience. 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 samepolysemy occurs.) There was confusion in certain concepts, such as“universals of grammar” and “rule of grammar”, and in the status andscope of grammatical categories of various kinds. But also, I suspect, aproblem that has been so vexing in recent years – that of relating thesystem to the text (so often discourse is analysed as if there were nogeneral principles of meaning behind it) – is ultimately part of the sameoverall unclarity.




387

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, weexpect to find some portion or portions of the book – but not thewhole of the book – devoted to grammar. In my own work, I haveoperated with the concept of “lexicogrammar” (that is, grammar andvocabulary as a single unity), while usually referring to it simply as grammar for short; this is a stratal concept, with grammar as one amongan ordered series comprising (at least) semantics / lexicogrammar /

phonology. But whatever part–whole model is adopted, languageremains 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, andsome might deny it altogether; but most would probably accept it inone form or another. I would be inclined to characterize grammar inthe 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 theoreticalfoundation, the account of language given by Lemke (1993). Lemkecharacterizes human communities as eco-social systems which persist intime through ongoing exchange with their environment; and the sameholds true of each of their many sub-systems. The social practices bywhich such systems are constituted are at once both material andsemiotic, with a constant dynamic interplay between the two. Notethat by semiotic I mean ‘having to do with meaning’, not ‘having to dowith signs’; thus, practices of doing and practices of meaning. Theimportant feature of the material–semiotic interplay is that, as Lemkepoints out, the two sets of practices are strongly coupled: there is ahigh 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 variouskinds. In fact, we usually use the term “system” to cover both systemand process: both the potential and the instances that occur; thus asemiotic system is a meaning potential together with its instantiation inacts of meaning. Now, one special kind of semiotic system is one thathas a grammar in it: such a system “means” in two phases, having adistinct phase of wording serving as the base for the construction of meaning. In other words, its “content plane” contains a grammar as




388

well as a semantics. We could characterize this special kind of semioticsystem as a grammatico-semantic system. It is the presence of agrammar 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 frameworkof 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 morecomplex than those without.

physical + life

= biological + value

= social + meaning

= semiotics1

[primary]

+ grammar

= semiotic2

[higher order, i.e.grammatico-semantic]

S S S S S1 2 3 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, asign being a content/expression pair. Systems of this kind, which maybe called primary semiotics, are found among numerous species: allhigher animals, including our household pets; and such a system is also

developed by human infants in the first year of their lives – I referredto this as the “protolanguage” (Halliday 1975). Primary semiotic systemshave no grammar. The more complex type of semiotic system is thatwhich 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




389

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, astonish-ment or fear (rephonologized as “interjections”, like ouch! , oy! ,wow! . . .). On the other hand, human adults also develop numerousnon-linguistic semiotic systems: forms of ritual, art forms, and the like;these have no grammar of their own, but they are parasitic on naturallanguage – their meaning potential derives from the fact that those who

use them already have a grammar. (See O’Toole (1994) for a richinterpretation of visual semiotics in grammatico-semantic terms.) Thusall human semiotic activity, from early childhood onwards, is as it werefiltered through our grammar-based higher order consciousness.

What then is a grammar, if we look at it historically in this way, asevolving (in the species) and developing (in the individual)? A grammar is an entirely abstract semiotic construct that emerges between thecontent and the expression levels of the original, sign-based primary

semiotic system. By “entirely abstract” I mean one that does not interfacedirectly with either of the phenomenal realms that comprise the materialenvironment of language. The expression system (prototypically, thephonology) interfaces with the human body; the (semantic componentof the) content interfaces with the entire realm of human experience;whereas the grammar evolves as an interface between these two inter-faces – shoving them apart, so to speak, in such a way that there arisesan 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 individualchild. It emerges through deconstructing the original sign and recon-structing with the content plane split into two distinct strata, semantics

and lexicogrammar. Such a system (a higher-order semiotic organizedaround 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 createmeaning. 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 theycannot modify each other or change in the course of unfolding. By




390

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 oneanother 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 arebrought into being in contexts of function. The functional contexts of

language fall into two major types, and the constitutive function thatthe grammar performs differs as between the two types. On the onehand, language “constitutes” human experience; and in this context,the grammar’s function is to construe: the grammar transforms experi-ence into meaning, imposing order in the form of categories and their interrelations. On the other hand, language “constitutes” social proc-esses 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 “metafunc-tional” synthesis, of semiotic transformation with semiotic enactment

(of knowledge with action, if you like), by “constituting” in yet a thirdsense – creating a parallel universe of its own, a phenomenal realm thatis itself made out of meaning. This enables the semiotic process tounfold, through time, in cahoots with material processes, each provid-ing the environment for the other. To put this in other terms, thegrammar enables the flow of information to coincide with, and interactwith, 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 itsenvironment. In the experiential (or, to give it its more inclusive name,the “ideational”) metafunction, the grammar takes over the materialconditions of human existence and transforms them into meanings. Wetend to become aware of the grammatical energy involved in this

process only when we have to write a scientific paper; hence, thissemiotic transformation may appear to be just a feature of knowledgethat is systematic. But all knowledge is like this: to “know” somethingis to have transformed it into meaning, and what we call “understand-ing” is the process of that transformation. But experience is understoodin the course of, and by means of, being acted out interpersonally – and, in the same way, interpersonal relations are enacted in the courseof, and by means of, being construed ideationally. The grammar flows




391

these two modes of meaning together into a single current, such thateverything we say (or write, or listen to, or read) “means” in boththese functions at once. Thus every instance of semiotic practice –

every act of meaning – involves both talking about the world andacting on those who are in it. Either of these sets of phenomena mayof 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, Ihave also been ‘doing’ grammar: the properties have been beingconstrued in lexicogrammatical terms. In other words I have beenusing 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 touse a grammar as a resource for constructing theories is because agrammar is itself a theory. As I suggested in the previous section, thegrammar functions simultaneously as a mode of knowing and a modeof doing; the former mode – the construction of knowledge – is thetransformation of experience into meaning. A grammar is a theory of human experience.

Construing experience is a highly theoretical process, involvingsetting 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 inwhich the phenomena of our experience may be perceived as beingalike. In whichever of these terms we conceive the matter, the grammar

has to sort things out, assigning functional value selectively to thevarious possible dimensions of perceptual order. The grammar’s modelof experience is constantly being challenged and reinforced in daily life;thus it tends to change when there are major changes in the conditionsof human existence – not as a consequence, but as a necessary andintegral 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




392

evolve, whereas scientific theories are at least partially designed. (Thenearest to an independent, fully designed semiotic system is mathemat-ics. Mathematics is grounded in the grammar of natural language; but

it has taken off to the point where its operations can probably nolonger be construed in natural language wordings.) But it is still thegrammar 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 abouthuman experience and to enact human relationships. In addition to

their metafunctional organization, already alluded to as enabling theintegration of knowledge and action, I shall mention (a) their size andability 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 talkingabout grammatics.

7 How big is a grammar?

The semogenic operations performed by a grammar are, obviously,extremely complex. Neuroscientists explain the evolution of the mam-malian brain, including that of homo sapiens, in terms of its modellingthe increasingly complex relationships between the organism and itsenvironment. This explanation foregrounds the construal of experience(the ideational metafunction); so we need to make explicit also itsbringing about the increasingly complex interactions between oneorganism and another (the interpersonal metafunction). To this mustbe added the further complexity, in a grammar-based higher-order semiotic, of creating a parallel reality in the form of a continuous flowof meaning (the textual metafunction). It could be argued that, sincelanguage has to encompass all other phenomena, language itself mustbe 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 thesimple question: how big is a language? (It seems strange how seldomthis 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 demandsmade on them by the explosion of knowledge that has taken place thiscentury. In major languages of technology and science, such as English,Russian or Chinese, there must be well over a million words in use, if




393

we put together the full range of specialized dictionaries – and thedictionaries can never be absolutely exhaustive. Of course, no oneperson 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 totalmeaning 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, itcan 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 specificmodel from the grammatics, in which a grammar is represented para-digmatically as a network of given alternatives (a “system network”).Given any system network it should in principle be possible to countthe number of alternatives shown to be available. In practice, it is quitedifficult to calculate the number of different selection expressions thatare 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 itgenerates will be greater than n (n+1 if the systems are maximally

dependent) and not greater than 2n (the figure that is obtained if allsystems are independent). But that does not help very much. Given anetwork of, say, 40 systems, which is not a very large network, all ittells us is that the size of the grammar it generates lies somewherebetween 41 and 240 (which is somewhere around 1012). We do notknow how to predict whereabouts it will fall in between these twofigures.

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 treatednon-recursively in order to simplify).

This network contains 28 systems, and generates just over seventythousand 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 relativelyunconstrained: it shows no conjunct entry conditions, and it shows anunusually high degree of independence among constituent systems –

probably more than there should be, although in this respect theEnglish verbal group is somewhat untypical of (English and other)grammars as a whole. On the other hand, it is not outstandinglydelicate: 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




394

Figure 2 The English verbal group: a simplified system network

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




395

81 systems in it, Mann was probably not far wrong when he estimatedoff the cuff that it would generate somewhere between 108 and 109

clause types.

Of course there are lots of mistakes in these complex networks, andthe only way to test them is by programming them and setting themto generate at random. It is not difficult to generate the paradigm of selection expressions from a reasonably small network (already in 1966Henrici developed a program for this purpose; cf. Halliday and Martin1981), where you can inspect the output and see where it has gonewrong. 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 overallcapacity is concerned, however, they are probably not orders of magnitude out.

It has been objected that the human brain could not possibly processa 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 abstractmodel; for another thing, the number of choice points encountered in

generating or parsing a clause is actually rather small – in the networkof the verbal group it took only 28 systems to produce some 70,000selection expressions, and in any one pass the maximum number of systems encountered would be even less – probably under half thetotal, in a representative network. In other words, in selecting one outof half a billion clause types the speaker/listener would be traversing atthe most about forty choice points. So although the system network isnot a model of neural processes, there is nothing impossible about agrammar of this complexity – that is, where the complexity is such thatit can be modelled in this way, as the product of the intersection of anot very large number of choices each of which by itself is extremelysimple.

8 How does your grammar grow?

Grammars do not remain static. They tend to grow; not at an evenrate, but with acceleration at certain “moments” in the history of aculture.

On the one hand, they grow by moving into new domains. Thishappens particularly when there is an expansion in the culture’sknowledge and control: in our present era, new domains are openedup by developments in technology and science. We are likely tobecome aware of this when we meet with a crop of unfamiliar words,




396

like those associated with the recent move into nanotechnology (engin-eering the very small); but the expansion may take place anywhere inthe 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 – thetwo 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 newmeanings alive, the same material advances were sometimes made twoor 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 intheir construction of existing domains. (This has been referred to byvarious 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 , aberry 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 beextended further, with new words or new compositions of words in agrammatical structure, like the nominal group in English and manyother languages. But greater delicacy is often achieved by intersectingsemantic features in new combinations; and this is less open to casualinspection, except in isolated instances which happen to be in someway striking (like certain “politically correct” expressions in present-day English). The massive semantic innovations brought about bycomputing, word processing, networking, multimedia, the informationsuperhighway and the like, although in part construing these activitiesas new technological domains, more typically constitute them as newconjunctions of existing meanings, as a glance at any one of thousandsof current periodicals will reveal. On a somewhat less dramatic scale,

we are all aware of the much more elaborate variations in the discourseof environmental pollution and destruction than were available ageneration ago. Even a seemingly transparent piece of wording such assmoke-free construes a new confluence of meanings; indeed the wholesemogenic potential of - free as a derivational morpheme has recentlybeen transformed. (Similar expansions have happened with -wise and-hood.)

There is a special case of this second heading – perhaps even a third




397

type of grammar growth – in the form of semantic junction broughtabout 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) alexicalized 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 aroseat the intersection of ‘kill’ with ‘thingness’ (the prototypical meaningof 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 weinteract with children who are growing up (Painter 1992; Derewianka1995). This is a good context in which to get a sense of the open-endedness of a grammar. In the last resort – and in some sense that isstill 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 limitis yet being approached.

9 Grammar as multiple perspectives

In a stratified semiotic system, where grammar is decoupled fromsemantics, the two strata may differ in the arrangement of their internalspace. Things which are shown to be topologically distant at onestratum may appear in the same systemic neighbourhood at the other.(See Martin and Matthiessen 1992, where the distinction is interpretedas between topological (semantics) and typological (lexicogrammar).) Itis this degree of freedom – the different alignment of semogenic

resources between the semantics and the grammar – that enableslanguage to extend indefinitely its meaning-making potential (a strikingexample of this is grammatical metaphor, mentioned at the end of the

previous section). It is also this characteristic which explains howsyndromes of grammatical features scattered throughout differentregions 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 donot expect things which mean the same to sound the same – although




398

there is considerable seepage, which Firth labelled “phonaesthesia”(Firth 1957). But between the semantics and the grammar, this newfrontier (typically non-arbitrary) within the content plane, we expect

to find more isomorphism: things which mean alike might reasonablybe worded alike. As a general rule, they are: grammatical proportion-alities typically construe semantic ones. But not always. On the one

hand, there are regions of considerable drift in both directions; anobvious one in English is the semantic domain of probability andsubjective 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, thereare the “syndromes” mentioned above – high-level semantic motifswhich are located all around the terrain of the lexicogrammar, such asthe complex edifice of meanings that goes to make up a “standardlanguage”. People make much use of these realignments in reasoningand inferencing with language.

This stratified vision of things enables the grammar to compromiseamong competing models of reality. As pointed out above in Section

6, a grammar sorts out and selects among the many proportionalitiesthat could arise in the construal of experience. It does this by making

adjustments among the different strata. Things may appear alike fromany of three different angles: (i) “from above” – similarity of functionin context; (ii) “from below” – similarity of formal make-up; and (iii)“from the same level” – fit with the other categories that are beingconstrued in the overall organization of the system. The grammar looksat objects and events from all three angles of orientation. It takes accountof their function: phenomena which have like value for human existenceand survival will tend to be categorized as alike. It takes account of their form: phenomena which resemble each other to human percep-tions will tend to be categorized as alike. And it takes account of howthings relate to one another: phenomena are not categorized in isolationbut in sets, syndromes and domains. In other words, the grammar adoptswhat 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 infunction; and the way they relate to each other may not reflect either kind of likeness. Other things being equal, the grammar tends to givesome precedence to functional considerations: consider any crowdedlexical domain, such as that of maps, plans, charts, figures, diagrams, tablesand graphs in English; or grammatical systems that are highly critical for survival, like that of polarity in any language. But the construal of




399

categories must make sense as a whole. And this means that it needs tobe founded on compromise. The grammar of every natural language isa massive exercise in compromise, accommodating multiple perspect-

ives 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 onlybecause of the effort that goes into tidying them up. Formal logic andeven mathematics can be seen as the result of tidying up the indeter-

minacies 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, thoughrather different case.

Clines are distinctions in meaning which take the form of continuousvariables instead of discrete terms. The prototype examples in grammar are those distinctions which are construed prosodically, typically byintonation (tone contour): for example, in English, “force”, from strongto 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 distinctionswhere, although the realizations are discrete (i.e. different wordings areinvolved), 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 onthe cover of the revised edition of my Introduction to Functional Grammar 1994). In this sense, since in the grammar’s categorization of experiencefuzziness 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 arenot. Ambiguity in the strict sense, as in lexical or structural puns, is nota form of indeterminacy as considered here, because it does not involveindeterminacy 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 inmany languages, is modality; in English, oblique modal finites like




400

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 itoccurs in another. A metaphor is the limiting case of a blend.

Complementarities are found in those regions of (typically experi-

ential) semantic space where some domain of experience is construedin two mutually contradictory ways. An obvious example in English isin the grammar of mental processes, where there is a regular comple-mentarity 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 experiencewhich can be construed in different ways: here, in a process of consciousness the conscious being is on the one hand ‘doing’, withsome 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 alwaysin the same regions of semantic space (note for example the striking

complementarity of tense and aspect in Russian). One favourite domainis causation and agency, often manifested in the complementarity of transitive and ergative construals.

Strictly speaking probability is not a “fuzzy” concept; but probabilityin grammar adds indeterminacy to the definition of a category. Con-sider the network of the English verbal group in Figure 2 above. As anexercise in grammatics this network is incomplete, in that there aredistinctions made by the grammar that the network fails to show: inthat sense, as already suggested, no network ever can be complete. Butit is incomplete also in another sense: it does not show probabilities. If

you are generating from that network, you are as likely to come upwith won’t be taken as with took; whereas in real life positive issignificantly more likely than negative, active than passive, and past

than future. Similarly a typical dictionary does not tell you that go ismore likely than walk and walk is more likely than stroll , though youmight guess it from the relative length of the entries. A grammar is aninherently probabilistic system, in which an important part of themeaning of any feature is its probability relative to other features withwhich it is mutually defining. Furthermore the critical factor in register variation is probabilistic: the extent to which local probabilities departfrom the global patterns of the language as a whole; for example a




401

register of weather forecasting (and no doubt other kinds of forecastingas well), where future becomes more probable than past; or one inwhich negative and passive suddenly come to the fore, like that of

bureaucratic regulations (Halliday 1991). Probabilities are significantboth in ideational and in interpersonal meanings, as well as in thetextual 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 naturallanguage grammars which a theory of grammar – a “grammatics” – is

designed to account for. The purpose of doing this was to provide acontext for asking the questions: how does the grammatics face up tothis kind of requirement? Given that every theory is, in some sense, alexicogrammatical metaphor for what it is theorizing, is there anythingdifferent about a theory where what it is theorizing is also alexicogrammar?

There is (as far as I can see) no way of formally testing a grammar inits role as a theory of human experience: there are no extrinsic criteriafor measuring its excellence of fit. We can of course seek to evaluatethe grammar by asking how well it works; and whatever language wechoose it clearly does – grammars have made it possible for humanityto survive and prosper. They have transmitted the wisdom of accumul-ated experience from one generation to the next, and enabled us tointeract 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 dominatedby that environment to being in control of it, and therefore alsoresponsible for it; cf. Halliday 1993). I suspect that the same holds truefor the grammatics as a theory of grammar: we can evaluate such atheory, by seeing how far it helps in solving problems where languageis 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 generaldistinction between a theory and a hypothesis.) By the same token agrammatics can also have its negative effects, if it becomes reductionistor 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




402

above – the size (and growth) of the grammar, its trinocular perspective,and its fuzz – how does the grammatics handle these various parame-ters? 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 paradig-matic axis, and by building into this orientation a variable delicacy; thisensures that the grammar will be viewed comprehensively, and thathowever 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 thegrammar by also adopting a trinocular perspective, based on the stratalorganization of the grammar itself. And it copes with the indeterminacyof 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, tobecome semiotic constructs in their own right, related only very

indirectly and obliquely to observations from experience. The proto-type of such a theory is a mathematical model; and one can theorizegrammatics in this way, construing it as a formal system. But agrammatics does not need to be self-contained in this same manner. Itis, as theory, a semiotic construct; but this does not create anydisjunction between it and what it is theorizing – it remains permeableat all points on its surface. The grammatics thus retains a mimeticcharacter: it explains the grammar by mimicking its crucial properties.One could say that it is based on grammatical logic rather than onmathematical logic. In some respects this will appear as a weakness: itwill lack the rigour of a mathematical theory. But in other respects itcan be a source of strength. It is likely to be more relevant tounderstanding other semiotic systems: not only verbal art, but also

other, non-verbal art forms, as demonstrated by O’Toole’s masterlyinterpretation of painting, architecture and sculpture in terms of sys-temic grammatics, referred to already (O’Toole 1994). And the newfield 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 betweenwording and meaning (see section 20 below).




403

In the next few sections I will make a few observations about thesematching properties of the grammatics, as they seem to me to emergein a systemic perspective.

12 Paradigmatic orientation and delicacy

When many years ago I first tried to describe grammar privileging theparadigmatic axis of representation (the “system” in Firth’s frameworkof system and structure), the immediate reasons related to the theoreti-cal and practical tasks that faced a ‘grammatics’ at the time (the middle

1960s): computational (machine translation), educational (first andsecond language teaching; language across the curriculum); sociological(language and cultural transmission, in Bernstein’s theoretical frame-work, for example Bernstein (1971)); functional-variational (develop-ment of register theory) and textual (stylistics and analysis of spokendiscourse). All these tasks had in common a strong orientation towardsmeaning, and demanded an approach which stretched the grammar inthe direction of semantics. There were perhaps five main

considerations.

i: The paradigmatic representation frees the grammar from theconstraints of structure; structure, obviously, is still to beaccounted for (a point sometimes overlooked when people drawnetworks, as Fawcett (1988) has thoughtfully pointed out), butstructural considerations no longer determine the construal of the lexicogrammatical space. The place of any feature in thegrammar can be determined “from the same level”, as a function

of its relationship to other features: its line-up in a system, andthe 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 fea-ture and relating it to other features: describing anything consistsprecisely in relating it to everything else.

iii: Thirdly, the paradigmatic mode of description models languageas a resource, not as an inventory; it defines the notion of “meaning potential” and provides an interpretation of “thesystem” in the other, Saussurean sense – but without setting upa duality between a langue and a parole.

iv: Fourthly, it motivates and makes sense of the probabilisticmodelling of grammar. Probability can only be understood asthe relative probabilities of the terms in a (closed) system.




404

v: Fifthly, representing grammar paradigmatically shapes it naturallyinto a lexicogrammar; the bricks-&-mortar model of a “lexicon”of words stuck together by grammatical cement can be aban-

doned as an outmoded relic of structuralist ways of thinking.

This last point was adumbrated many years ago under the formula-tion “lexis as delicate grammar” (see above, Chapter 2); it has subse-quently been worked out theoretically and illustrated in two importantpapers by Hasan (1985; 1987). The principle is that grammar and lexisare not two distinct orders of phenomena; there is just one stratumhere, that of “(lexico)grammar”, and one among the various resources

that the grammar has for making meaning (i.e. for “realizing” itssystemic features) is by lexicalizing – choosing words. In general, thechoice of words represents a delicate phase in the grammar, in the sensethat it is only after attaining quite some degree of delicacy that wereach systems where the options are realized by the choice of thelexical item. The lexicogrammar is thus construed by the grammaticsas a cline, from “most grammatical” to “most lexical”; but it is also acomplementarity, because we can also view lexis and grammar asdifferent perspectives on the whole. The reason people write “gram-mars” on the one hand and ‘dictionaries’ on the other is that options atthe most general (least delicate) end of the cline are best illuminated byone set of techniques while options at the most delicate (least general)end are best illuminated by a different set of techniques. One canemploy 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 choiceof a lexical item, does not mean, on the other hand, that these aresystems where there is a direct correspondence of feature to item, suchthat 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 featuresare components of lexical items. (Thus, they are like the features of a

standard componential analysis, except that they form part of theoverall system network and no distinction is made between featuresthat are “lexical” and those that are “grammatical”.) Any given lexicalitem then appears as the conjunct realization of a set of systemicfeatures; and “the same” lexical item may appear many times over, in

different locations, much as happens in a thesaurus (where however theorganization is taxonomic rather than componential).




405

13 A note on delicacy

Inherent in the paradigmatic orientation is the concept of variable

delicacy, in which again the grammatics mimics the grammar: delicacyin the construal of grammar (by the grammatics) is analogous todelicacy in the construal of experiential phenomena (by the grammar).Since for the most part the “lexicalized” mode of realization isassociated 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 stageof lexical realization one has arrived at the endpoint in delicacy.

What is the endpoint, on the delicacy scale? How far can thegrammatics go in refining the categories of the grammar? In one sensethere can be no endpoint, because every instance is categoriallydifferent from every other instance, since it has a unique instantialcontext of situation. We tend to become aware of this when an

instance is codified in the work of a major writer and hence becomesimmortalized as a “quotation”. It seems trivial; but it may not be trivialin the context of intelligent computing, where the program might needto recognize that, say, turn left! , as instruction to the car, has a differentmeaning – 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 inour commonsense engagements with language, in daily life, we behaveas if there is an endpoint in delicacy: namely, that which is defined bythe orthography. We assume, in other words, that if two instances lookdifferent (i.e. are represented as different forms in writing) they shouldbe described as different types; whereas if two instances are writtenalike 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 collectivewisdom of generations of speakers/listeners. And we do allow excep-tions. (a) We recognize homonymy and, more significantly, polysemy,where the delicacy of categorization does not stop at the barrier createdby the writing system. (b) We accept that there are systematic distinc-tions which orthography simply ignores: for example, in English, allthose realized by intonation and rhythm. (c) And, as already noted, it




406

never was assumed, except perhaps among a very few linguists, that a“function word” like of has only one location in the terrain describedby 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 whatis systemic: the point where proportionalities no longer continue tohold. As long as we can predict that a : a :: b : b :: . . . , we are stilldealing with types, construed as distinct categories for purposes of

grammatical description.In practice, of course, we are nowhere near this endpoint in writingour systemic “grammars”. (I find it disturbing when the very sketchydescription of English grammar contained in Halliday (1994) is takenas some kind of endpoint. Every paragraph in it needs to be expandedinto a book, or perhaps some more appropriate form of hypertext; thenwe will be starting to see inside the grammar – and be able to rewritethe 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 littledata to support it. The trouble was that until there were first of all taperecorders and then computers, it was impossible to assemble the data agrammarian needs. Since grammars are very big, and very complex, aneffective grammatics depends on having accessible a very large corpusof diverse texts, with a solid foundation in spontaneous spokenlanguage; together with the sophisticated software that turns it into aneffective source of information.

14 A note on the corpus

A corpus is not simply a repository of useful examples. It is a treasuryof 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. Descriptivecategories do not emerge out of the data. Description is a theoreticalactivity; 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 descriptionbased on) a grammatics that is so designed as to make optimum use of the corpus data available, maximizing its value as an information source




407

for the description. (‘Corpus-based grammar’ might be a less misleadingterm.) 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 intothe 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 thegrammar from the lexical end; such a decision will of course affect theinitial design and implementation of the corpus itself, but there isnothing inherent in the nature of a corpus that requires one to take

that decision. A corpus is equally well suited to lexis-driven or togrammar-driven description. It is worth recalling that the first major corpus of English, the Survey of English Usage set up by Quirk atUniversity College London, was explicitly designed as a resource for writing a grammar in the traditional sense – that is, one that would becomplementary to a dictionary.

The most obvious characteristic of the corpus as a data base is itsauthenticity: what is presented is real language rather than sentences

concocted in the philosopher’s den. Typically in trawling through acorpus one comes across instances of usage one had never previously

thought of. But, more significantly, any kind of principled sampling islikely to bring out proportionalities that have remained entirely beneathone’s conscious awareness. I would contend that it is precisely the mostunconscious patterns in the grammar – the cryptogrammatic ones – that are the most powerful in their constitutive effect, in construingexperience and in enacting the social process, and hence in theconstruction of our ideological makeup. Secondly, the corpus enablesus to establish the probability profiles of major grammatical systems.Again, I would contend that quantitative patterns revealed in thecorpus – as relative frequencies of terms in grammatical systems – arethe manifestation of fundamental grammatical properties. The grammar is an inherently probabilistic system, and the quantitative patterns inthe discourse that children hear around them are critical to the way

they learn their mother tongues. Thirdly, the corpus makes it possibleto test the realization statements, by using a general parser and, perhapsmore effectively, by devising pattern-matching programs for specificgrammatical systems; one can match the results against one’s ownanalysis of samples taken from the data. Some form of dedicated parsingor pattern matching is in any case needed for quantitative investigations,since the numbers to be counted are far above what one could hope toprocess manually (cf. Halliday and James 1993). Fourthly, since modern




408

corpuses are organized according to register, it becomes possible toinvestigate register variation in grammatical terms: more particularly, inquantitative 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. Incategorizing the grammar, the grammarian works “from above”, “fromroundabout” and “from below”; and these three perspectives are

defined in terms of strata. Since the stratum under attention is thelexicogrammar, “from roundabout” means ‘from the standpoint of thelexicogrammar 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 morphol-ogy and phonology, how the given category relates to the expression(what it “is realized by”)’. What are being taken into account are theregularities (proportionalities) at each of the three strata.

Since the patterns seen from these three angles tend to conflict, theresulting description of the grammar, like the grammar’s own descrip-tion of experience, must be founded on compromise. This is easy tosay; it is not so easy to achieve. Often one finds oneself ‘hooked’ onone oculation – obsessed, say, with giving the most elegant account of how some pattern is realized, and so according excessive priority to theview from below; then, on looking down on it from above, one findsone has committed oneself to a “system” that is semantically vacuous.If the view from below is consistently given priority, the resultingdescription will be a collapsed grammar, so “flat” that only an impov-

erished semantics can be raised up on it. On the other hand, if one isbiased towards the view from above, the grammar will be so inflatedthat it is impossible to generate any output. And if one looks from bothvertical 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 thiscase the grammar is not so much inflated or collapsed; it is simplycurdled.

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. Butthere is more than a simple analogy embodied here. I referred above tothe notion of semiotic transformation: that the grammar transforms




409

experience into meaning. The trinocular perspective is simply that: it isthe 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, word-ings and expressions (you only have to do this, of course, when formand function cease to match; this is why the task is inescapably one of

achieving compromise). The entire stratal organization of language issimply the manifestation of this trinocular principle. Making thisprinciple explicit in the grammatics is perhaps giving substance to thenotion of ‘language turned back upon itself’.

16 Indeterminacy in grammatics

That the grammatics should accommodate indeterminacy does notneed explaining: indeterminacy is an inherent and necessary feature of a grammar, and hence something to be accounted for and indeedcelebrated in the grammatics, not idealized out of the picture – just asthe 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 youhave to intervene and actively manage it, so in semiotic practice, when

you transform something into meaning (i.e. perturb it semiotically) youalso have to manage the complexity. We can note how the grammar manages the complexity of human experience. In the first instance, itimposes artificial determinacy, in the form of discontinuities: thus, agrowing plant has to be construed either as tree or as bush or as shrub(or . . .); the line of arbitrariness precludes us from creating intermediatecategories 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 asemantic domain, with expressions like half in and half on, in between abush 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, fore-grounding something which is a property of the grammar as a whole.

Now consider the grammatics from this same point of view. The




410

categories used for construing the grammar – things like noun andsubject and aspect and hypotaxis and phrase – are also like everydayterms: 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 sameresources exist, if in a somewhat fancier form, for making the indeter-

minacy explicit: verbal noun, pseudo-passive, underlying subject , and soon.

What then about the specific construction of indeterminacy in theoverall 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 discontinuityon 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 casein point: qualitative relationships both within and between systems maybe ironed out, so that (i) the system is construed simply as a or b

(or . . .), without probabilities, and (ii) one system is either dependenton 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 alreadysuggested that the concept of lexicogrammmar (itself a cline from“most grammatical” to “most lexical”) embodies a complementarity inwhich lexis and grammar compete as theoretical models of the whole.There are many blends of different types of structure, for example theEnglish nominal group construed both as multivariate (configurational)and as univariate (iterative) but without ambiguity between them. Andthe two most fundamental relationships in the grammatics, realizationand instantiation, are both examples of indeterminacy.

I have said that a grammar is a theory of human experience. But thatdoes not mean, on the other hand, that it is not also part of thatexperience; it is. We will not be surprised, therefore, if we find that its

own complexity comes to be managed in ways that are analogous tothe ways in which it itself manages the complexity of the rest. In thelast 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 cometo construe a higher order semiotic. Realization is the name given to




411

the relationship between the strata; the verb realize faces “upwards”,such that the “lower” stratum realizes the “higher” one. (Realization isalso extended to refer to the intrastratal relation between a systemic

feature and its structural (or other) manifestation.) Instantiation is therelationship between the system and the instance; the instance is said toinstantiate the system.

It can be said that, in the elements of a primary semiotic (signs), thesignifier “realizes” the signified; but this relationship is unproblematic:although the sign may undergo complex transformations of one kindor another, there is no intermediate structure between the two (no

distinct stratum of grammar). With a higher order semiotic, where agrammar intervenes, this opens up the possibility of many differenttypes of realization. It is not necessary to spell these out here; they areenumerated and discussed in many places (for example Berry 1977;Fawcett 1980; Martin 1984; Hasan 1987; Matthiessen 1988; Eggins1994).

But there is another opening-up effect which is relevant to thepresent 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 diver-sity, among others. But this does not force us to locate the boundary atany particular place. One can, in fact, map it on to the boundarybetween system and structure, as Fawcett does (system as semantics,structure as lexicogrammar); whereas I have found it more valuable toset up two distinct strata of paradigmatic (systemic) organization. Butthe point is that the boundary is indeterminate – it can be shifted; andthis indeterminacy enables us to extend the stratal model outsidelanguage proper so as to model the relationship of a language to itscultural 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 thedialectic between observation and theory. This has always been aproblem area for linguistics: whereas the concept of a physical principlebecame clear once the experimental method had been established – a“law of nature” was a theoretical abstraction constructed mathemati-cally by the experimenter – the concept of a linguistic principle hasproved much more difficult to elucidate.




412

Saussure problematized the nature of the linguistic fact; but heconfused 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 (thelinguistic system) differs from parole (the linguistic instance) only in theposition taken up by the observer. Langue is parole seen from a

distance, and hence on the way to being theorized about. I tried tomake this explicit by using the term “meaning potential” to character-ize 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 producedand understood.Instantiation is a cline, with (like lexicogrammar) a complementarity

of perspective. I have often drawn an analogy with the climate and theweather: when people ask, as they do, about global warming, is this ablip in the climate, or is it a long-term weather pattern?, what they areasking 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 semioticclimate)? The observer can focus at different points along the cline;and, whatever is under focus, the observation can be from either of thetwo points of vantage.

18 Realization and instantiation: some specific analogiesIt is safe to say that neither of these concepts has yet been thoroughlyexplored. Problems arise with instantiation, for example, in using thecorpus as data for describing a grammar (why a special category of “corpus grammar”?); in relating features of discourse to systemicpatterns in grammar (why a separate discipline of “pragmatics”?); andin construing intermediate categories (such as Bernstein’s “code”,

which remains elusive (like global warming!) from whichever end it isobserved – which is what makes it so powerful as an agency of culturalreproduction). (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’sideas).

As far as realization is concerned, Lemke has theorized this power-




413

fully as ‘metaredundancy’ (Lemke 1984) (and cf. Chapter 14 above);but this still leaves problems in understanding how metafunctionaldiversity is achieved, and especially the non-referential, interpersonal

aspects of meaning; and in explaining the realization principles at workat 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), onissues involving the higher strata of register and genre).

I am not pursuing these issues further here. But as a final step I willshift to another angle of vision and look at realization and instantiation

from inside the grammar – turning the tables by using the grammar asa way of thinking about the grammatics. One of the most complexareas in the grammar of English is that of relational processes: processesof being, in the broadest sense. I have analysed these as falling into twomajor types: (i) attributive, and (ii) identifying. The former are thosesuch as Paula is a poet, this case is very heavy, where some entity isassigned to a class by virtue of some particular attribute. The latter arethose 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 identifiedby 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 isthe outward form, that by which the entity is recognized, while theother is the function the entity serves. This relationship of course takesa 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 isexactly one of realization: the Token realizes the Value, the Value isrealized by the Token. It is thus analogous to the relationship definedin the grammatics as that holding between different strata. The gram-mar is modelling one of the prototypical processes of experience asconstructing a semiotic relationship – precisely the one that is funda-

mental 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 placesit as one among a larger set. It was pointed out by Davidse (1992) thatthis Carrier / Attribute relationship in the grammar is actually one of instantiation: the Carrier is an instance of, or “instantiates”, the Attrib-ute. It is thus analogous to the relationship defined in the grammaticsas that holding between an instance and the (categories of the) system.




414

(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 perceptionof a phenomenon as an instance of a general class – in terms of aproperty 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. Thegrammatics evolves (or rather one should say the grammatics “isevolved”, to suggest that it is a partially designed system) as a meta-phoric 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 metaphoricproperty of being a theory about itself.

19 Centricity

Since the grammatics is a theory about a “logo” system, it is “logo-centric”, or rather perhaps “semocentric”: its task is to put semioticsystems in the centre of attention. In the same way, biological sciencesare “bio-centric”: biased towards living things; and so on. I think it isalso a valid goal to explore the relevance of grammatics to semioticsystems other than language, and even to systems of other types. Thegrammatics 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 semioticweather 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 theproperties of human language as such, and what are features specific to

a given language. The problem is compounded by the fact that there ismore than one way of incorporating the distinction (wherever it isdrawn) into one’s descriptive practice. Firth articulated the differencebetween 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 em-phasis). I have preferred to avoid talking about “universals” because itseems to me that this term usually refers to descriptive categories being




415

treated as if they were theoretical ones. As I see it, the theory modelswhat is being treated as “universal” to human language; the descriptionmodels 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, thatof Chinese sinocentric, that of French gallocentric and so on. (Notethat the theory is not anglocentric; the description of English is.) Thisis 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 thathave dominated much of linguistics during the second half of thecentury.

In practice, of course, English does exist, and it has been extensivelydescribed; so inevitably people tend to think in terms of categories setup for English – or for other relatively well-described languages. I have

suggested elsewhere some considerations which seem to me relevant todescriptive 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 onEnglish people told me I was making English look like Chinese! (Itseems ironic that, now that systemic theory is being widely applied toChinese studies, the work of mine most often cited as point of referenceis 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 eachlanguage is described in its own tongue. The protocol version of thegrammar of English is that written in English; the protocol version of the grammar of Chinese is that written in Chinese; and so on. Theprinciple 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 andscientific discourses; and this semantic baggage has some metalinguisticvalue. This applies particularly, perhaps, to the use of theoretical termsas metacategories in the description; words such as (the equivalents of)option, selection, rank, delicacy are likely to have quite significant (butvariable) loadings.

But the principle also helps to guard against transferring categoriesinappropriately. Even if descriptive terms have been translated from




416

English (or Russian, or other source) in the first place, once they aretranslated they get relocated in the semantic terrain of the newlanguage, and it becomes easier to avoid carrying over the connotations

that went with the original. So if, say, the term subject or theme appearsin a description of Chinese written in English, its status is as a translationequivalent of the definitive term in Chinese. Perhaps one should point

out, in this connection, that there can be no general answer to thequestion how much alike two things have to be for them to be calledby the same name!

20 A final note on grammatics

As I said at the beginning, when I first used the term “grammatics” Iwas 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 proportionsuch that grammatics is to grammar as linguistics is to language. Butover the years since then I have found it useful to have “grammatics”

available as a term for a specific view of grammatical theory, wherebyit is not just a theory about grammar but also a way of using grammar to think with. In other words, in grammatics, we are certainlymodelling natural language; but we are trying to do so in such a wayas to throw light on other things besides. It is using grammar as a kindof logic. There is mathematical logic and there is grammatical logic,

and both are semiotic systems; but they are complementary, and insome 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 thecontext of its application to different tasks. As Matthiessen (1991b) haspointed out, this, in general, is the way that systemic theory has movedforward. Recently, a new sphere of application has been suggested. Asmentioned 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 toadvance to the point where they really become intelligent they have tofunction the way human beings do – namely, through natural (human)

language. This view (and it is more than a gleam in the eye: Sugenohas taken significant steps towards putting it into practice) derivesultimately from Zadeh’s “fuzzy logic”; it depends on reasoning and




417

inferencing with fuzzy sets and fuzzy matching processes. But to usenatural language requires a grammatics: that is, a way of modellingnatural language that makes sense in this particular context. Systemic

theory has been used extensively in computational linguistics; and thePenman nigel grammar, and Fawcett’s communal grammar, areamong the most comprehensive grammars yet to appear in computa-

tional form (Matthiessen 1991a; Matthiessen and Bateman 1992; Faw-cett and Tucker 1990; Fawcett, Tucker and Lin 1993). But, moreimportantly perhaps, systemic grammatics is not uncomfortable withfuzziness. 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 haveif it is to have any value for intelligent computing. This is an excitingnew field of application; if it prospers, then any grammarian privilegedto interact with Sugeno’s enterprise will learn a lot about humanlanguage, as we always do from applications to real-life challengingtasks.

Note

1. This is not to question the semiotic achievements of the bonobo chimpan-

zees (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.






419

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, Codesand 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.



bibliography

420

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.

Buhler, 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 Dis-

course 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 Lin-guarum 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.



bibliography

421

Colby, B. N. and Colby, L. M. (1980) The Daykeeper: The Life and Discourse

of an Ixil Diviner . Cambridge, MA.: Harvard University Press.

Danes, 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



bibliography

422

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 Univer-sity 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.



bibliography

423

Fries, P. H. (1981) ‘On the status of theme in English: arguments from

discourse’, Forum Linguisticum 6.1. Reprinted in J. S. Petofi and E. Soozer

(eds), Micro and Macro Connexity of Texts. Hamburg: Buske.

Fries, P. H. (1995) ‘Patterns of information in initial position in English’, inP. 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’elaboration du Franc ais elementaire. 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’,


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 Univer-

sity 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.



bibliography

424

Halliday, M. A. K. (1977) ‘Text as semantic choice in social contexts’, in A.

van Dijk and J. Petofi (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 Priest-

ley’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. Putz (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 Inter-

national 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 confer-

ence 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



bibliography

425

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. Petofi (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).



bibliography

426

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. Ham-

burg: Helmut (Buske Beitrage 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 (Mem-

oirs of the International Journal of Anthropological Linguistics 2, Indiana

University Publications of Anthropology and Linguistics).



bibliography

427

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 Cognitive-

stratificational 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 strati-

fication’, in C. I. J. M. Stuart (ed.), Report of the Fifteenth Annual (First

International) Round Table Meeting on Linguistics and Language Studies. Wash-

ington, D.C.: Georgetown University Press (Monograph Series onLanguages 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



bibliography

428

(Toronto Semiotic Circle Monographs, Working Papers and Prepublica-

tions 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 Dynam-

ics 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. Petofi

and E. Sozer (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.



bibliography

429

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 congres 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 gram-mar’, 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: adiscussion 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 ).



bibliography

430

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 “know-ledge” 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, Prin-

ciples of Firthian Linguistics. London: Longman (Longman’s Linguistics

Library 19).

Mitchell, T. F. (1958) ‘Syntagmatic relations in liguistic analysis’, Transactions

of the Philological Society.

Mukarovsky, 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 technicaltexts’, 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 Reci-

procity 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 Univer-

sity Press.

Painter, C. (1984) Into the Mother Tongue: A Case Study in Early Language

Development. London: Pinter (Open Linguistics Series).



bibliography

431

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’,


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.



bibliography

432

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 theoffice 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-semioticsof 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.



433

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, 381allophone 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 122–5Buhler, 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,



index

434

category (cont.)

335, 343, 351, 354–6, 385, 397,

399–400, 408, 410, 412

causalconsequential 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, 414English 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, 342cohesive 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, 201–2conversation 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



index

435

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, 413declarative 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, 270diminuendo-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



index

436

ergativity 312

Eskimo folk tales 231

ethnographic 173, 226, 231, 236, 386

ethnographer 230ethological 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, 238extralinguistic 31, 236, 295

extratextual 39–40

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, 415Fries, 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

linguistic 111

rhetorical 226–34

speech 233–9, 268, 273, 307

syntactic 107textual 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 236semantic(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

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

givenand 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, 415lexis 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



index

437

grammatics 11, 365– 6, 369–78, 384– 417

trinocular 402–9

graphology 39

Greek 294– 6ancient 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, 301hearer-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, 221Huddleston, 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, 333see 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,

242features 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, 236infant 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, 312theory 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



index

438

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, 406written 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 414logo-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, 238Martin, 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, 71meaning-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



index

439

microclass 101, 159

Mitchell, J. F. 205, 223

modal 216, 242, 245– 6, 300, 340, 399

subject 190–1, 194modality 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 397reflective 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, 364morpheme 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, 362–4nominal 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 120O’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, 407participant 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



index

440

person 70, 230, 236–7, 268, 337, 353,

378

phatic 226

phoneme 65–6, 68, 71, 96, 219–20, 222supra-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 292diachronic 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, 376phrase 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, 371predicate 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, 407type 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, 389protolinguistic 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, 387reflection 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



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, 377Saussure, 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,

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

Documents

M. a. K. Halliday, Jonathan J. Webster] on Grammar