On Grammar (Collected Works of Mak Halliday Series) (PDFDrive) PDF

On Grammer
M. A. K. Halliday
Continuum
On Grammar
Collected Works of M. A. K. Halliday
Volume 1: On Grammar
Volume 2: Linguistic Studies of Text and Discourse
Volume 3: On Language and Linguistics
Volume 4: The Language of Early Childhood
Volume 5: The Language of Science
Volume 6: Computational and Quantitative Studies
Volume 7: Studies in English Language
Volume 8: Studies in Chinese Language
Volume 9: Language and Education
Volume 10: Language and Society
Volume 1 in the Collected Works of M. A. K. Halliday
On Grammar
M. A. K. Halliday
Edited by Jonathan Webster
Continuum
The Tower Building, 11 York Road, London, SE1 7NX
370 Lexington Avenue, New York, NY 10017-6503
First published 2002 by Continuum
Reprinted 2003, 2005
쑔 M. A. K. Halliday 2002
All rights reserved. No part of this publication may be

reproduced or transmitted in any form or by any means,
electronic or mechanical, including photocopying,
recording or any information storage or retrieval system,
without permission in writing from the publishers.
British Library Cataloguing-in-Publication Data

A catalogue record for this book is available from the
British Library.
ISBN 0-8264-4944-1 (hardback)
Typeset by SetSystems Ltd, Saffron Walden, Essex

Printed and bound in Great Britain by CPI Bath
CONTENTS
Preface vii
Acknowledgements ix
Introduction: a personal perspective 1
Professor M. A. K. Halliday
SECTION ONE: EARLY PAPERS ON BASIC CONCEPTS

Editor’s Introduction 17
1 Some aspects of systematic description and
comparison in grammatical analysis 21
2 Categories of the theory of grammar 37
3 Class in relation to the axes of chain and
choice in language 95
4 Some notes on ‘deep’ grammar 106
5 The concept of rank: a reply 118
Appendix to Section One 127
SECTION TWO: WORD–CLAUSE–TEXT

6 Lexis as a linguistic level 158
7 Language structure and language function 173
contents
8 Modes of meaning and modes of expression:

types of grammatical structure and their
determination by different semantic functions 196
9 Text semantics and clause grammar: how is a
text like a clause? 219
10 Dimensions of discourse analysis: grammar 261
SECTION THREE: CONSTRUING AND ENACTING

11 On the ineffability of grammatical categories 291
12 Spoken and written modes of meaning 323
13 How do you mean? 352
14 Grammar and daily life: concurrence and
complementarity 369
15 On grammar and grammatics 384
Bibliography 419
Index 433
vi
PREFACE
For nearly half a century, Professor M. A. K. Halliday has been

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

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

articles and chapters in this volume. Original publication details are
provided below, and also at the beginning of each chapter.
‘Some aspects of systemic description and comparison in grammatical

analysis’ from Studies in Linguistic Analysis, published by Blackwell
Publishers, 1957, pages 54–67. Reprinted by permission of Blackwell
Publishers.
‘Categories of the theory of grammar’ from Word, 17(3), 1961, pub-
lished by the Linguistic Circle of New York (now the International
Linguistic Association), pages 241–92. Reprinted by permission of the
author and the International Linguistic Association.
‘Class in relation to the axes of chain and choice in language’ from
Linguistics, vol. 2, 1963, published by Mouton (now Mouton de 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 by
permission of Cambridge University Press.
‘The concept of rank: a reply’ from Journal of Linguistics 2(1), 1966,
published by Cambridge University Press, pages 110–18. Reprinted by
permission of Cambridge University Press.
‘Lexis as a linguistic level’ from In Memory of J. R. Firth, published by
Longman, 1966, pages 148–62. Edited by C. E. Bazell, J. C. Catford,
M. A. K. Halliday and R. H. Robins.
‘Language structure and language function’ from New Horizons in
Linguistics, published by Penguin Ltd, 1970, pages 140–65. Edited by
John Lyons. 쑕 M. A. K. Halliday, 1971, collection 쑕 John Lyons,
1970. Reprinted by permission of the Penguin Group (UK).
ix
acknowledgements
‘Modes of meaning and modes of expression: types of grammatical

structure and their determination by different semantic functions’ from
Function and Context in Linguistic Analysis: Essays Offered to William
Haas, edited by D. J. Allerton, Edward Carney and David Holdcroft,
published by Cambridge University Press, 1979, pages 57–79.
Reprinted by permission of Cambridge University Press.
‘Text semantics and clause grammar: some patterns of realization’ from
The Seventh LACUS Forum, edited by James E. Copeland and Philip
W. Davies, published by LACUS, 1981, pages 31–59. Reprinted by
permission of LACUS.
‘How is a text like a clause?’ from Text Processing: Text Analysis and
Generation, Text Typology and Attrition (Proceedings of Nobel Symposium
51), edited by Sture Allen, published by Almqvist and Wiksell Inter-
national, 1982, pages 209–47. Reprinted by permission of Almqvist
and Wiksell International.
‘Dimensions of discourse analysis: grammar’ from The Handbook of
Discourse Analysis, Vol. 2: Dimensions of Discourse, published by 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 Theory
and Practice, edited by Martin Davies and Louise Ravelli, published by
Pinter, 1992, pages 20–35.
‘Grammar and daily life: concurrence and complementarity’ from
Functional Approaches to Language, Culture and Cognition, edited by Teun
A. van Dijk, published by John Benjamins Publishing Co., 2000, pages
221–37. Reprinted by permission of John Benjamins Publishing Co.
‘On grammar and grammatics’ from Functional Descriptions: Theory in
Practice, edited by Ruqaiya Hasan, Carmel Cloran and David G. Butt,
published by John Benjamins Publishing Co., 1996, pages 1–38.
Reprinted by permission of John Benjamins Publishing Co.
x
INTRODUCTION:
A PERSONAL PERSPECTIVE
The volumes in this series will contain a selection of my writings on

language, extending over the half century beginning in 1951. A few,
including one or two items written specially for the series, have not
been published before, and many of those that were published appeared
in rather inaccessible places.
The papers are arranged according to topic, beginning with the
present volume, which is oriented towards grammatical theory. But the
topical arrangement will tend to be fairly loose, partly because my
writing has always been inclined to drift, and partly because both the
editor and I prefer it to be that way – weak boundaries have always
been characteristic of my approach.
I have never really thriven in a discipline-based structure of know-
ledge. It was a feature of my century – the late and rather unlamented
twentieth, perhaps mercifully short in Eric Hobsbawm’s conception of
it – that it began by erecting walls between the disciplines, and it is
proving difficult to demolish these walls now that they have come to be
constraining rather than enabling. They had been enabling to start with,
at least for the newly founded social (and ever newer semiotic) sciences;
sociologists, psychologists and linguists had to be able to lock each other
out while sorting out and investigating their own chartered domains.
So in the mid-century many linguists sturdily proclaimed the inde-
pendence and autonomy of the discipline of linguistics, and one could
sympathize with their anxiety, because language was everybody’s
business and there would always be outsiders looking over their
shoulders and telling them how to do their job – or, more usually,
telling them they were simply wasting their time. I think I myself once
made some reference to this; if so, it will turn up in Volume 3. But
1
on grammar
what I did stress was how much linguistics had in common with other
scholarly pursuits and, when it came to asking questions about
language, I always found myself lining up with the outsiders. It seemed
fine for us, as linguists, to determine the content or domain of our
own discipline – sociologists studied society, psychologists studied . . .
whatever psychologists do study, linguists studied language.1 That’s
what we were there for. But it did not seem fine, to me at least, for us
to determine what questions should be asked about that domain. I was
interested in what other people wanted to know about language,
whether scholars in other fields or those with practical problems to be
faced and solved – including that undervalued and under-rewarded
group who have to be both scholars and practical problem-solvers,
namely teachers.
There was nothing surprising about this last perspective: not only
had my parents both been teachers, but I myself had taught languages
for thirteen years before transferring myself into a teacher of linguistics.
But even before that, while still at school, I had been trying to find out
about language – because I was keen on literature and wanted to
understand why its language was so effective, what was special about
it. There is no separating one’s personal history from the academic
paths one pursues, nor any way of detaching cause from effect in
explaining one’s chosen approach to a field of study. One way or
another, I have always found myself asking the kinds of questions about
language that arose, as it were, from outside language itself.
Of course linguists always have been located, and located themselves,
within some broader context; there is nothing unusual about that. But
at any given “moment” in space-time, there are likely to be only a few
predominant motifs by which the context of linguistic scholarship is
defined. This may even be legislated from on high, as when Stalin
writing in 1950 (or Chikobava, writing on his behalf) instructed Soviet
linguists to get on with the job of demonstrating the linguistic unity of
the Slav nations – reasonably enough, since the Soviet Union had just
taken them all over. Usually it is determined by less overtly political
factors: by particular social movements and demands, or notable
advances in knowledge in some other field.
The present era provides a noteworthy example of the latter. Since
about 1985 there has been spectacular progress in the field of neuro-
science; the combination of new technology – positron-emission
tomography, magnetic resonance imaging and its derivatives – with
new insights in evolutionary theory and its contributing disciplines has
transformed the way we understand the human brain, how it has
2
a personal perspective
evolved in the species and how it develops in the individual from birth
(and before) to maturity. And this new understanding has radically
redefined the place of language. It is now clear that language and the
brain evolve together, and that these develop together in infancy and
childhood. The development of the brain is the development of the
ability to mean; as in every aspect of human history, so in the ontogeny
of the individual human being the material and the semiotic 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, a
similar type of higher-order semiotic. On the contrary. The work of
Duane Rumbaugh and Susan Savage-Rumbaugh has brought out the
point that the bonobo chimpanzees can operate with sets of arbitrary
symbols in a way that is analogous to our own system of wording
(lexicogrammar). They lack an analogous vocal apparatus, but that is
beside the point. It is tempting to assume that they have been following
the same evolutionary path and are simply less far advanced along the
way – this is the assumption that prompts questions like ‘what age have
they reached, in terms of a human child?’ But this assumption is
probably wrong, or at least misleading, if it is used to describe an adult
chimpanzee in terms of an immature human; the adult bonobo’s brain
is fully wired up in terms of the construing of experience, and enacting
of social relations, that constitute bonobo culture. The question can
fairly be asked about bonobos brought up from birth in a human-like
semiotic environment, like Kanzi, and it is too early to say yet whether
Kanzi and the other youngsters’ development of the power of meaning
tracked that of human children and stopped at a certain level or
whether it was proceeding along a somewhat different route.
It might be argued that such new knowledge about how the brain
functions, and how it evolves and develops, has no significance for the
way linguists describe and explain language, especially at the ‘inner’
strata of lexicogrammar and phonology (wording and sounding). Pos-
sibly; although even here it seems to me to set certain constraints and
more importantly perhaps to favour certain explanations over others. It
suggests “systems thinking” rather than compositional thinking (Mat-
thiessen), grammatical logic rather than formal logic (Sugeno 1995),
fuzzy and probabilistic categories rather than clearly bounded and
deterministic ones. Since the brain is more like a jungle than like a
computer (Edelman 1992), it disfavours representations of grammar and
phonology that are influenced, however indirectly or subconsciously,
3
on grammar
by the way that computers happen to be being designed and operated

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

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

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

friend and colleague Basil Bernstein. But by nature, and also by
experience, I was (and have always remained) a generalist. So while
consciously I was trying to model language as a social phenomenon, in
fact I was acting against my own advice and trying to look at language
from every possible vantage point in turn.4
Most linguists, it seemed to me, looked at language only from the
inside, claiming the right to formulate their own questions about it –
which was why linguistics seldom interested practitioners in other
fields. This was also, of course, a perfectly valid perspective. But it did
bring with it certain problems.
When I was being trained as a dialect fieldworker, by my other great
teacher Wang Li (then Professor of Linguistics at Lingnan University,
Canton), there were still no tape recorders. We had to transcribe
responses directly into IPA script, which was excellent training for my
later investigation of child language. Professor Wang was able to acquire
a primitive version of the same thing – a wire recorder, but it was not
much use, because the wire was always breaking and would end up as
a ball of wire wool fit only for scouring a wok. That was in 1949–50.
Not that linguistics had had no base in technology. There was
already high quality instrumentation for acoustic analysis (spectrograph,
oscillograph, mingograph), as well as various techniques for investigat-
ing the articulatory mechanism of speech. Gramophone records were
widely used in language teaching: when I was taught Chinese for the
armed services at the University of London in 1942–43, the Depart-
ment had its own recording equipment on which students could
register their own performance and compare it with the recorded
model. There were archives of spoken language on disk and even on
cylinder, including dialect survey material in a number of different
languages. But there was no technology for capturing authentic speech,
natural conversation in the interactive situations of daily life, nor for
managing an extensive body of text.
As a consequence, linguistics had hardly any data. In that respect it
was about where physics had been at the end of the fifteenth century,
before technology had evolved to enable physicists to observe and
to conduct experiments. Linguists either relied on the kind of mani-
cured discourse that is produced in writing and in prepared and self-
monitored speech, or else invented data for themselves from their own
intuition of the language, and they had no way of processing large
quantities even of that.
For linguistics, the two most important advances in the latter half of
7
on grammar
the century were technological ones: the invention of the tape recorder
and the evolution of the computer. The tape recorder made it possible
to record natural speech. The computer made it possible to process
large quantities of data. The two together have given us the modern
computerized corpus, with natural speech as a significant component,
on which we can undertake quantitative analyses on a statistically
significant scale. As a bonus, the computer enables us to test our
descriptive generalizations, through text generation and analysis (“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 are
being revealed that we have known must be there, because there was a
gap where the approaches from the lexical and the grammatical poles
of the lexicogrammar converged, but which we could not see: the
nature of grammatical logic is beginning to be understood; the semo-
genic (meaning-creating) power of discourse is coming into view, both
in monologic and in dialogic mode; quantitative mechanisms of lin-
guistic change are beginning to appear on the agenda. From all this it
should be possible in the next decade or two to crack the semiotic
code, in the sense of coming fully to understand the relationship
between observed instances of language behaviour and the underlying
system of language – something that has eluded us up till now, so that
we have even turned the two into different disciplines, calling only one
of them “linguistics” and labelling the other “pragmatics”.
Some people will feel threatened by this new understanding. We
know this because there are those who already do. To bring to light
the systems and processes of society is already threatening enough, as
witness the panic reactions to Bernstein thirty years ago when he
demonstrated how social class structures are transmitted, but semiotic
systems and processes are even nearer the bone. As long as linguists
confined their attention to dead languages, codified texts or sanitized
examples like John kissed Mary and It’s cold in here, no one would feel
really at risk. But when grammar extends to the study of the meaning-
creating power of everyday real life talk, it starts to become dangerous.
Some people feel worried that the grammarian is someone who knows
what they are going to say next and even if they can be reassured that
that is not what theory is about, it is scarcely less threatening (appar-
ently) to be told what proportion of positive to negative clauses they
are going to use in their speech. And for others, just to be faced with a
record of real life conversation can be unnerving; they feel embarrassed
8
and ill-at-ease at what seems to invade the interactants’ privacy and

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

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

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

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

but still ultimately of the same kind. It should be possible to extend the
power of semantic representation so that such ways of reasoning can be
integrated into our model of language, rather than being treated as if
they were separate operations in the brain. And it will be necessary to
do this, I think, in order to achieve the kind of “intelligent computing”
that is envisaged by Michio Sugeno in his work at the Brain Sciences
Institute in Tokyo.
What the following chapters do is to illustrate some of the steps in
my own thinking that have led me in this general direction. The steps
that have seemed to me perhaps most critical in this endeavour might
be summarized as: the unity of lexicogrammar; the priority of the view
‘from above’, from meaning and function; the move into systemics
(system networks), freeing the grammar from the restrictions imposed
by structure; the metafunctional foundation, disentangling the strands
of meaning that are woven together in the syntax; the construction of
language by children, from protolanguage to mother tongue; the
decoupling and recoupling of lexicogrammar and semantics – the
phenomenon of grammatical metaphor; the conceptualizing of the
relation between system and text (instantiation) and the probabilistic
nature of linguistic systems. Some of these will be treated separately in
later volumes: in particular Volume 4 on child language and Volume 5
on grammatical metaphor and the language of science. Others will
appear in various contexts and under a variety of headings.
I am often asked about my views on “linguistic universals”. The
answer is that I follow Hjelmslev and Firth in distinguishing theoretical
from descriptive categories. The theoretical categories, and their inter-
relations, construe an abstract model of language (and other semiotic
systems); they are interlocking and mutually defining.8 The theory that
is constituted in this way is continually evolving as it is brought to bear
on solving problems of a research or practical nature. (No very clear
line is drawn between “(theoretical) linguistics” and “applied linguis-
tics” – except institutionally where, for example, an education authority
will give teachers release time and professional credit for a degree
course called “applied linguistics” but not for one called “linguistics”.)
Descriptive categories are categories set up in the description of
particular languages. When people ask about “universals”, they usually
mean descriptive categories that are assumed to be found in all
languages. The problem is that there is no mechanism for deciding
how much alike descriptive categories from different languages have to
be before they are said to be “the same thing”. There is a method,
12
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, is
proclaimed by the multifaceted architecture of the theory.
A volume of typological studies organized around the theoretical
category of metafunction will serve to illustrate this standpoint (Caf-
farel, Martin and Matthiessen 2002). My own interpretation of the
grammar of modern English will be found in An Introduction to
Functional Grammar (Halliday 1985 and later editions). Other descriptive
papers on English and on Chinese will be presented in Volumes 7 and
8. A theory-based account of the ideational semantics of English is in
Construing Experience through Meaning (Halliday and Matthiessen 1999).
I doubt whether any of the present volumes would have appeared
without the enthusiasm, energy and efficiency of “my” editor, Dr
Jonathan Webster, of the City University of Hong Kong. He brought
the whole project to life, convincing me that it was worthwhile and
convincing the publishers that it could actually come to fruition. It has
been a pleasure being driven along by his momentum.
My thanks also to the publishers, especially to Janet Joyce, who
despite years of my ineffectual attempts to get started never lost patience
with me or faith in the enterprise, and to Robin Fawcett, who set the
whole thing going and provided many rounds of valuable suggestions
and advice.
Notes
1. Psychologists, in fact, study psychology – the domain is defined by the
discipline, rather than the other way round. Hence the rather odd locutions
like “criminal psychology”, meaning the mind-set, or psyche, of criminals,
rather than psychological theories that criminals have devised. I was once
put down rather scathingly by a psychologist for suggesting that their
domain of study might be the human psyche.
2. See Volume 4 in this series.
3. Language is a system of meaning (a “semiotic” system); and semiotic
systems are of the fourth order of complexity, being also physical and
biological and social. This means that one and the same linguistic 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
13
on grammar
wander throughout the highways and byways of language. If there has

been any consistent motif, it has been ‘now how would this (phenomenon
and its explanation) seem to someone who is interested in language for
some other reason, different from the one that prompted me to explore
it?’
5. Strictly speaking, of course, it is semiotics; but semiotics has not yet
evolved into a general theory of meaning and it seems likely that, for the
time being at least, the way forward is by extending linguistics into other
semiotic systems. I use “meaning” rather than the term “information” (the
term imported from those who work on matter) because information is
only a sub-class of meaning; it is the part that can be measured, whereas,
unlike matter, meaning in general is not open to measurement (though
systemic linguistics offers one way in; see Volume 5).
6. We are of course accustomed to linguistics being dismissed in this offhand
way: “linguists always . . . (or . . . never . . .), so you needn’t bother 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).
14
SECTION ONE
EARLY PAPERS ON BASIC CONCEPTS
This page intentionally left blank
EDITOR’S INTRODUCTION
In this first section we look at five papers written and published by

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

exponents the substantial (phonic or graphic) segments of the text’.
Clearly influenced by Firth’s teaching and his scholarship, Halliday
draws on Firth’s approach to formulate a General Linguistic theory,
which is concerned with how language works at the level of grammar.
While some have referred to Halliday’s approach as neo-Firthian, such
a characterization does neither scholar justice. Realizing that his the-
oretical approach, as outlined in ‘Categories of the theory of grammar’
(Chapter 2), diverged from that of his mentor, Halliday sought to have
the opportunity to discuss the paper with Firth. Firth’s sudden passing,
however, prevented this from happening. In this paper, Halliday sets
out the following fundamental categories for the theory of grammar:
unit, structure, class and system, which relate to one another and to the
data along three distinct scales of abstraction, including rank, exponence
and delicacy. Halliday prefaces his discussion by stating what he regards
as ‘given’, among which he includes the following:
a. Texts, or observed language events, are the data to be accounted
for, whether spoken or codified in writing.
b. Description consists in relating the text to the categories of the
theory. Description is not theory; rather it is a body of method
derived from and answerable to the theory.
c. Linguistic events should be accounted for at a number of different
levels. While the primary levels are form, substance and context,
a complete framework of levels requires certain further subdivisions
and additions, including substance, whether phonic or graphic,
form on two related levels of grammar and lexis, and context,
which is an interlevel relating form to extratextual features.
d. The study of phonic substance belongs to a distinct but related
body of theory, namely General Phonetics. Phonology, on the
other hand, relates form and phonic substance, i.e. where 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 the
linguist expounds the description.
Elaborating on each of the fundamental categories for the theory of
grammar, Halliday describes units as pattern carriers. The scale on
which units are ranged in the theory is called rank. Structures are ‘the
18
editor’s introduction
ordered repetition of like events that make up the patterns’. There are
both primary and secondary structures, distinguished in terms of delicacy, or
depth of detail. Whereas class involves ‘the grouping of like events by
their occurrence in patterns’, system deals with ‘the occurrence of one
rather than another among a number of like events’. To help the reader
better understand the application of the categories of grammar, Halliday
presents a framework of categories for the description of another very
familiar kind of patterned activity, namely, eating a meal. Looking back
on this chapter after forty years, Professor Halliday provides some
background from his personal history to help readers better understand
his very careful concern for assigning things to categories:
Struggling with the grammar of Chinese, and then of English, in the

conceptual-categorial frameworks which were then available (traditional
grammar, linguist’s descriptions of languages, Jespersen and Wang Li,
Firth’s system–structure theory, Pike, Fries, Hill, Hockett, etc.), I was
constantly finding that the categories were unclear: you would find a
label attached to some patch or other, but with no indication of what
kind of category it was supposed to be and the whole battery of
technical statements never added up to a coherent picture of the whole.
I felt I needed to know where I was at any moment and where any
descriptive statement that I made fitted in to the overall account.
In a paper appearing in Linguistics in 1963, ‘Class in relation to the

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

ranks’. On one point both Halliday and Matthews agree, namely, that
rank grammar is a hypothesis about the nature of language. As Halliday
argues, it is a hypothesis worth making both for its descriptive 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 appeared
in Kress (1976).
20
Chapter One
SOME ASPECTS OF SYSTEMATIC

DESCRIPTION AND COMPARISON IN
GRAMMATICAL ANALYSIS (1957)
1 Descriptive and historical, particular and comparative

The description of a language employs, at the grammatical level as at all
other levels, systems of related categories. Such categories as are 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 points
of reference is clearly of a different order from the other two: there can
be no universal formal-linguistic categories (there might theoretically be
categories formally identified as common to all languages studied here-
tofore, but such identification is not yet a practical possibility), while
non-formal-linguistic categories, if they are to figure in the description
at all, must be implicitly regarded as universal.1
Unless it is supposed that the sole domain of linguistic science is the
study of the evolution of linguistic forms, the improvement of the
methods of linguistic description remains one of the tasks of the linguist.
In recent decades, striking advances have been made on the basis of
General Linguistic theory in the development of descriptive techniques,
especially of the first type: the description of a language in terms of
categories established within the language itself (Firth 1951). As in the
interrelated branches of any discipline, there is a constant mutual con-
tribution between descriptive and historical studies in linguistics; and it is
not surprising that many recent developments in the former have been
founded on work done in languages where it has not yet been possible to
establish a series of phonological-lexical and phonological-morphological
First published in Studies in Linguistic Analysis (Special Volume of the Philological Society),
1957, Oxford: Blackwell, pp. 54–67.
21
correspondences as a basis for genetic groupings; this especially in Britain,

where we have a long tradition of the description of the languages of
Asia, Africa and the Pacific. In such languages, for historical as well as for
descriptive purposes, improved methods have been demanded; but the
need for a general theory of description, as opposed to a universal scheme
of descriptive categories, has long been apparent, if often unformulated,
in the description of all languages. A distinctive contribution of the
twentieth century has been the progress towards its achievement.
The sort of descriptive statement which has been the fruit of these
achievements is too often characterized negatively by opposition to a
historical (the synchronic–diachronic dichotomy) or to a comparative
statement. This is probably due in part to the very fact that the techniques
have been applied to languages which have no ‘history’ (that is, no
written document of the past) and even no script, and which have not
been satisfactorily organized into families by the comparative historical
method. It is not unnatural that what is new in descriptive techniques
should have been emphasized by its being contrasted with the historical
methods which, in the modern period at least, developed earlier; but,
while the comparative historical, like all other scientific methods, will
benefit by question and scrutiny, it is no essential part of modern
descriptive linguistics that it should reject the achievements of the past,
still less that it should deny linguistic history as a field of scientific study.
If we consider general linguistics to be the body of theory which
guides and controls the procedures of the various branches of linguistic
science, then any linguistic study, historical or descriptive, particular or
comparative, draws on and contributes to the principles of general
linguistics. A simple scheme for ordering the branches of linguistic
science controlled by general linguistics might recognize two 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 texts
greater than one). Thus:
Descriptive Historical
—
Particular — —
—
—
— —
Comparative — — — —
— —
— —
22
systematic description and comparison
Any of these types of study may be undertaken with the use of

formal linguistic methods: that is, by the methods of what is sometimes
called ‘structural linguistics’. (If such a term is to be used, it should
perhaps be taken to refer neither to a branch nor to a particular school
of linguistics but to that body of general linguistic theory which
controls the application of formal linguistic techniques.) This inclusion
of historical studies in the field of application of formal linguistic
methods rests on the acceptance of the possibility of arranging language
texts according to a time–construct. On a completely a-historical view,
there will be only one vertical axis, as any number of texts treated in a
single statement could only be material for comparative study: in the
type of comparatism envisaged by Allen, in which ‘time has no
direction and there is no becoming’ (Allen 1953: 106), diachronic has
of course no meaning. In this scheme it is envisaged that there might
be a difference in the treatment of material consisting of more than one
language text according to whether or not the texts are arranged on a
time-scale and treated as exponents of the same language at different
periods (for example, by the modification of the description in such a
way as to present a continuum in which the systematic ordering of the
texts corresponds to their ordering in time).
A historical study is then formed out of material provided by a series
of descriptive studies. The distinction between this method, in which
any linguistic form is placed in its descriptive context (systematized)
and the systems in which it operates are treated historically, and the
type of historical study which ‘structural’ linguistics has excluded – in
which is traced the evolution of particular forms without descriptive
systematization – might be reflected in the addition to the diagram of a
third vertical dimension (in fact a breakaway from the second), perhaps
‘evolutionary’. If one wishes to seek an opposition between ‘structural’
linguistics and comparative philology, it must surely depend not on the
acceptance or non-acceptance of history but on the type of historical
approach. So may the social anthropologist study either the evolution
of kingship in a particular tribe or group of tribes, or the place of the
institution of kingship in the structure of a given society at different
periods of its history. The “structural” linguist does not handle > or <;
but a clear formulation of how these concepts are defined in terms of
active participants in linguistic situations would render them less
inaccessible to him and so clarify the relation between the two types of
historical study.
A third horizontal axis in terms of the scope of the material might
be the “universal”: the question is whether this is at present, to use
23
Firth’s expression, ‘on the agenda’, since what is formal when particular
or comparative tends to become imaginative when universal. One
might then summarize diagrammatically:
Descriptive Historical Evolutionary
—
Particular — —
—
—
— —
Comparative — — — —
— —
— —
——
Universal —— ——
——
——
The “structural” linguist, while attempting to develop descriptive
methods that are general, in the sense that they are scientific methods
universally valid in linguistic description and forming part of general
linguistic theory, will be unwilling to claim universality for any formally
established category; since, while, for example, it may be convenient
in the description of all languages so far studied to give the name ‘verb’
to one class of one unit, this is not a universal statement: the ‘verb’ is
redefined in the description of each language. Even if any one category
can be identified (presumably contextually) across all languages studied
so far – and this is still a long way off: possibly, if a unit ‘word’ can be
satisfactorily universalized, the word-class ‘personal pronoun’, as rela-
tively easy to identify contextually, may be the first – such a category
is universal only in the limited range of what is and has been, not what
might be. At present any universal system of categories must rest on
other than formal linguistic criteria: if such can be provided, for
example, by mathematics, so much the better – the “structural” linguist
will not “reject” it, but he cannot be expected to provide it within his
own terms of reference.
What “structural” linguistics has done has been to concentrate, partly
in redress of the balance in linguistics, on the first of the methods of
description outlined in the first paragraph above. It is in fact by the
study of systems and structures within the framework of particular
description that this body of theory, and the methods derived from it,
24
have earned the name of “structural” linguistics. The term is perhaps

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

A complete analysis at the grammatical level (Firth 1935; 1951: 121),
in a particular description in which all forms of the language are related
to systems set up within the language itself, requires the establishment
of grammatical categories, ordered as terms in interrelated systems and
having as exponents the substantial (phonic or graphic) segments of the
text. Such categories are of two types, which we may call units and
classes. The units are defined by interrelation in terms of extent: unlike
a system of classes, whose terms are both collectively exhaustive and
mutually exclusive, the single system of units forms a hierarchy in
descending progression, such that each term is defined as n times the
succeeding term, that is, as consisting of one or more members of the
succeeding term (exponentially, every exponent of a given unit is
statable either as (coextensively) a single exponent or as a sequence of
exponents of the unit next in succession).
The units are established in the grammar, by formal grammatical
criteria, though in the delimitation of exponents of each unit within
25
the text other criteria, phonetic or graphic, may contribute and may be
taken, where to do so is compatible with the general aims of simplicity
and comprehensiveness, as the primary or even sole criteria (for
example, punctuation or spacing in a written text, features of intonation
of pause in a spoken text, all of which then enter into the grammatical
description). It is probable that in the description of any language at
least two units will be required: these would be such as could be named
the sentence and the word.3
A descending order of procedure seems preferable not only for the
presentation (where indeed it may be varied for a particular purpose),
but also for the analysis of the grammar, where in such a hierarchic
progression the classification made at the level of each unit will itself
determine the classes that are to be set up for the lower units. One may
begin by establishing, and delimiting the exponents of, that unit (which
we may call the “sentence”) which, while within the scope of gram-
matical statement – not so extensive as to be incapable of systematic
analysis – is yet enabled to operate as the linguistic action of participants
in a situation: which is, in fact, “living language” and constitutes the
unit of analysis at the contextual level. In the subsequent establishment
and classification of the lower units, the statements made about each
unit will be related to values set up in the structure of the higher unit.4
For each unit there will then be set up systems of classes, formally
established in the grammar and exhaustive, such that statements may be
made which are valid for all exponents of a given unit. These classes
are set up independently of structure: that is to say, a unit (for example
clause) having been established, it is then classified by reference to
various sets of formal criteria (for example presence of, or ordering of,
certain formally defined elements); each set of criteria permits the
establishment of one system of classes (“clause–classes”). Such mutually
independent systems of classes of any unit are referred to as ‘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 placed
within it. Thus it might be that all words are either red or blue or
yellow and all words are either square or round or neutral in shape.
Two dimensions of classes are implicit in the taxonomic hierarchy of
the unit system: as characteristic of each unit except the lowest there
appears the dimension compound / simple, while for each unit except
26
the highest there may be set up the complementary system of free /

bound; the free member is exponentially identified with the simple
member of the unit next above. How far these dimensions of classes
enter into the description will depend on the extent of their 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 the
structure.5 The terms in the system of any one dimension may operate
as values in the structure of the unit next above: for example, one
could state the structure of a sentence in terms of the “aspect”
dimension in the clause (if the elements in the sentence structure are
stated in terms of the relevant values P and I, then it might be that the
structure IP is found to occur but not PI). At the same time the
sentence might also be stated in terms of the values F and B the
exponents of which are segments of the text as systematized in a free /
bound dimension of clause–classes. Similarly if a unit “word” is
established there will be dimensions of word–classes the terms in which
operate as values in clause structures: given a verb / noun / adverb
system of word–classes, it might be that the structure ANV and NAV
were admitted in the clause but NVA excluded.
The formal criteria by which the classes are established and their
exponents identified may be interior to a class (as when a clause is
identified as “bound” through the presence within it of a member of a
certain word–class) or exterior to it (as when a clause is identified as
non-sentential in virtue of its being internal to another clause (in the
sentence structure), or a word as a “verb” in virtue of certain categories
and combinations to which the class “verb” is uniquely susceptible).
The type of grammatical description outlined here may be briefly
illustrated from the description of New Chinese (Modern Pekingese).
It should be emphasized that what follows is a summary and necessarily
incomplete: it is, however, abstracted from a descriptive statement of
Modern Pekingese grammar, one among many possible such 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:
27
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 bǎ, rarely jiāng.
The passive is marked by the basic structure (N)vNVa, where a =
bound adverb, the exponent of v being the prepositive verb s, that of a
the particle di. All clauses with structure other than these are active, or
neutral in voice.
As an example of a bound word–class, the postpositive verb is placed
in the system of word–classes as follows: the verbal word–class is one of
a three-term system verbal / nominal / adverbial, and is itself made up
of the two terms free / bound, the free forming a system of the two
terms free verb / pro-verb, the bound admitting the three terms 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 vN). The
postpositive verb admits further classification into three types, two of
which operate in the verb group and are distinguished from one another
by the admission by one of certain combinatory possibilities excluded
by the other, and the third in the postpositive complex group.
Before leaving consideration of the particular description of a
language in terms of categories set up within the language itself, I might
28
mention one further instance of the application here of formal methods

of analysis. This concerns the determination of the occurrence of a
member of a particular grammatical class by formal linguistic but non-
grammatical features. The presence of a certain form in a given unit in
the syntagm may render probable the occurrence in a subsequent unit
of a member of a particular class; this is in fact a form of contextual
determination, but it may be stated, partially at least, in terms of only
the linguistic (verbal) action in a context of situation – the source of
the determination may be found to be in what might be called ‘context
of mention’. This requires the two-term system “given” / “new”, the
given being that which has been mentioned in the preceding linguistic
context.
Modern Pekingese shows some correlation between this dimension
and word order in the clause, the position of the given being regularly
precedent to that of the new. This may not only determine the relative
position of words where the basic clause structure is unaffected (for
example the relative position of preverbal free noun or pronoun and
preverbal adverb) but also permit the prediction of the occurrence of a
particular class of clause. A correlation is observable between the
occurrence of marked (not active) voice in a clause and the presence of
a marked opposition between given and new in the preceding context.
Within the categories of marked voice, it frequently appears that in an
ergative clause, both nouns are given and the verb is new; while in a
passive clause, the verb is given and the (directly) preverbal noun is
new.
The correlation is more clearly observable if context of mention is
taken to include not only repetition of the term mentioned but also
reference (for example, pronominal or synonymic). With such “context
of reference” we are well on the way towards context of situation;7 in
a spoken text, categories of given and new established in the context
of situation, though further removed from the level of grammatical
description as usually envisaged, may be expected to display such
correlation with grammatical categories to at least as great an extent as
in a written text, where contextual reference is restricted to the
linguistic context (since there is no independent context of situation
for any unit less than the whole text). Indeed the contextual categories
of given and new may aid in the identification of grammatical categor-
ies: a certain category might be identified as the form taken by the
given or that taken by the new.
29
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 as
the basis of a particular description, reference to the forms of another
language, including the language of description, may be made without
infringing the requirements of formal analysis. In any other than a
monolingual description of a language text there arises in any case the
specific problem of the relation of the forms of the language under
description to the language of description. Since language is used to
describe language, if in a formal descriptive grammar it is desired
to exclude from consideration as far as possible all forms that do
not belong to the l.u.d., the nearest approach is that outlined above,
where the aim is achieved through the creation of a metalanguage
whose terms, whatever the context of their previous usage, are to be
taken as defined only with reference to the text under description. All
identification of categories either comparative or universal is thereby
excluded.
It may, however, be desirable in a given instance or for a given
purpose to relate the forms of the l.u.d. to particular forms of the l.o.d.
The two languages may then be seen to impinge on one another at
various points. At one extreme, it is possible to make a descriptive
grammar of the l.o.d. using the same procedures as are applied to the
l.u.d. and subjecting the two descriptions to a systematic comparison
(as envisaged by Allen 1953: 88 ff.). A comparison of the systems of
categories would first establish what categories were comparable; the
latter would then be compared so as to permit the identification of
terms within the system of each category.
At the other extreme, in dealing with a limited language text it may
be possible to make a complete translation of the text under description,
a translation of the type that may be characterized as contextual – one
such as to reproduce, as nearly as possible, the creative effect in the
given situation of the original. This may be useful where the l.o.d. is
not itself described and the linguist does not consider it to be within
his terms of reference to make a description such as would permit a
systematic comparison. It may then be found that certain categories of
the l.u.d. show a regular translation equivalence to certain 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-
30
times be found preferable to reference to undefined categories of the

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

1 wǒ I
33 tāmen they
2
22,2(2)3(3)
nı̌
nı̌men
} you
3 tā { he
she
12(2), 13(3), 12(2)3(3)
12(2), 12(2)3(3)
wǒmen
zámen
} we
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 been
permitted between terms operating in systems with different numbers
of terms: in any calculation of degree of relationship this point would
of course be crucial. Prima facie, such identification would seem justifiable
where the criteria for the identification were contextual. The identifi-
cation of terms by grammatical criteria in a comparative study poses
separate problems which are touched upon briefly in the next section.
31
As a conclusion to the foregoing it may be remarked that reference

to the forms of the l.o.d. is invaluable for pedagogical purposes, where
its further systematization might prove helpful, as well as in those
instances where a single l.o.d. is common to linguists working in widely
divergent fields.8
4 Descriptive comparison
In a descriptive comparison there is no implication of genetic relation-
ship; but beyond the application to particular description permitted by
comparison of this type (since any l.u.d. may be compared with any
l.o.d.) systematic comparison itself connotes a wider purpose, seen
initially in the establishment of degrees of relationship. Here the gram-
matical identification of terms across languages is an essential basis of
comparison; and while much can be done on contextual criteria (though
not as much as in the establishment of lexical systems for comparative
purposes) some formal procedure must be found in order that statement
of degrees of relationship may assume a general significance.
Possibly, however, we should not impose a strict demarcation
between identification of grammatical terms on grammatical and on
non-grammatical (e.g. contextual) criteria. Allen, rightly rejecting both
terminological identification and identification by translation, considers
that identification by formal grammatical criteria (as opposed to the
situational–contextual criteria which are available for the semantic
identification of lexical items) seems not to be possible (Allen 1953: 99,
100). Some attempt at identification may be made with the Chinese
dialects, where speakers of more than one dialect constantly make such
identifications in practice, with or without phonological resemblances:
for example, from the personal pronouns, a Cantonese speaker equates
ne˛i with Pekingese nı̌ and, as readily, ke˛udei with Pekingese tāmen. This
identification in practice demonstrates the¯ contextual basis of the iden-
tification; its validity on grammatical criteria may be tested by reference
to the place of the class “pronoun” in the system of word–classes.
When we find that it is possible to describe both Modern Pekingese
and Modern Cantonese in terms of the same units of sentence, clause,
word and character; that at the word level we can set up for both a
three-term system of classes, verbal, nominal and adverbial, and that
one term in the system of nominal word–classes, the auxiliary noun,
enters in both into the noun group with identical position and value in
the structure, we can regard the class of auxiliary noun in the two
dialects as comparable, and are then justified in seeking to identify
32
certain terms in the two systems of auxiliary nouns.9 In this comparison,

both grammatical and contextual criteria must be invoked: the Pekingese
system has an unmarked term (ge) which is absent from the Cantonese
system, which would exclude the grammatical identification of any
terms whatsoever, since each marked term in the Pekingese system must
be grammatically identified as excluding not only the other marked
terms but also the unmarked term, to which latter the Cantonese has
no parallel. On contextual criteria, however (including linguistic–
contextual, i.e. collocational), the marked terms of the Pekingese system
may be regarded as forming a distinct sub-system comparable with the
Cantonese system, so that a term can be identified in the use of the
auxiliary noun Pekingese zhı̄, Cantonese (graphically and historically
identical) zēk, we find the following instances (with specimen English
equivalents):
Pekingese Cantonese
yı́ ge gŏu yāt zēk gáu ‘a dog’ (in general context or ‘given’)
yı̀ zhı̄ gŏu yāt zēk gáu ‘one dog’ (in specific context or ‘new’)
(nà ge) gŏu zēk gáu ‘the dog’
The non-identification of the first and third instances will not prevent
the contextual identification of the second if the total spread within that
sub-system can be shown to be the same.10
Some such procedure, for the identification of terms in grammatical
systems, once established, the field of application of descriptive com-
parison is seen to be very wide. A comparison of systems in two
languages such as Chinese and English, which at first sight seems to be
of little purpose beyond its application to the particular description of
one of the two, provides material for a formal systemic typology when
compared with other such comparisons – for, if with Allen ‘relationship
is not of languages but of systems’, the significance of such relationship
is that typology is not of languages but of systems. But without going
so far from the scope of comparative studies as usually envisaged, one
may seek a field for the application of systematic comparison both in
the comparison of languages considered to be genetically related and in
the study of languages in geographical proximity where there is no
material for genetic groupings.
With regard to the former, any discussion of the nature and impli-
cations of genetic relationship may be avoided by the choice of a simple
instance of systematic comparison in the phonology of the dialect areas
of Chinese. In the Modern Chinese dialects, in the system of nasal finals
in the phonological unit the syllable may have two or three terms, or
33
there may be only one nasal final. If we compare these systems in certain
dialects of the Mandarin, Wu and Yüeh dialect groups, we find that it
is prosodically linked, and identical in number of terms, with the system
of plosive finals or, if the latter is absent, with that of vocalic finals.
Cantonese shows the final plosive system p/t/k with prosodically
identical nasals m/n/ŋ; in Shanghai, where there is a single final plosive,
the glottal closure, there is one nasal final, with varying point of contact;
while in Pekingese, where there are no final plosives, the two terms of
the final nasal system are prosodically identical with the oral finals, so
that n:ŋ::i:u.11 The calculation of the degree of relationship will depend
on the number of final systems set up for each dialect and the number
of these in which all terms can be identified; but the comparison
immediately suggests the historical interconnection between the disap-
pearance of the final plosive system and the elimination of one term
from the final nasal system in the dialects where these have occurred.
With regard to the study of languages in geographical proximity in
an area where no, or only partial, genetic groupings have been estab-
lished, systematic comparison can determine whether the question of
grammatical affinity is to be posed at all, and if so in what form. The
superficially apparent affinity among the languages of certain areas has
long been the basis of traditional typology. An instance of how it may
be demonstrated or disproved, initially as a function of systems, might
be found in the grammar of East Asian languages. It seems possible to
set up in, for example, Pekingese, Cantonese, Vietnamese, Siamese and
Malay systems of nominal word–classes which would permit the iden-
tification as comparable of certain terms, including the category discussed
with reference to Pekingese and Cantonese above and which may be
called the auxiliary noun. If the place of the auxiliary noun in the system
of word–classes, and the terms in the various systems of the auxiliary
noun, can be compared as between these languages on the basis of
adequate criteria, we may determine whether or not a formal description
of these languages would reveal for these and other systems anything
that could be regarded as grammatical affinity – such as, that is, to
exclude from such affinity any systems set up for other languages in
other areas. Only if such affinity could be established among a significant
number of systems in the grammar of these languages would it be
possible to raise further questions of general grammatical affinity and –
as yet only an interesting speculation – of geographical gradation.
Such a systemic comparison may help to resolve the difficulty that,
on the one hand, there appears an obvious but unformulated grammat-
ical similarity among the East Asian languages as contrasted, for example,
34
with the Indo-European languages, while on the other hand there exists
the quite justifiable scepticism among linguists either as to how this is
to be explained, in comparative historical terms, where there are no
correspondences, or as to whether there is anything to be explained (or,
in systematic terms, to be stated comparatively) at all.
Notes
1. The study of the “form of the content” by plerematics, as envisaged by
Hjelmslev, would set up particular or comparative (i.e. non-universal)
categories; but the criteria for such categories, while as yet inadequately
defined, would also be such as to be regarded as formal-linguistic. A
method of classification of words on the basis of universal categories of
relation and description is to be found in the work of Brøndal (1948).
2. Seventh International Congress of Linguists, Preliminary Reports, London,
1952, p. 53: ‘Can a purely formal grammatical analysis be carried out on
languages such as Chinese, in which all or nearly all the words are
invariable, and if so, on what principle?’
3. Provided the principle of particular formal description is adhered to, the
choice of current terms seems preferable to the creation of new ones.
There is, however, no reason why, especially in the initial stage of the
process of analysis, completely non-committal terms should not be
employed. The practice of some linguists of talking (at least to themselves)
about e.g. red words and blue words can equally well be extended, for
example, to “strings” and “bits”.
4. Compare the descending order within the levels of linguistic analysis in
which meaning may be stated, as suggested by Firth (1951: 121). For an
example of the employment of this method of descending analysis in
grammatical statement, cf. Robins (1953) and Halliday (1956).
5. As pointed out by Robins (1953: 109), Firth has indicated how system
and structure require to be kept separate in General Linguistic theory. I
have attempted to follow Firth’s view of the two as distinct but related
concepts.
6. The linguistic unit of which the written character is the graphic symbol
corresponding to the syllable at the phonological level. The use of the
term ‘character’ follows the Chinese practice of calling both the linguistic
and the graphic units by the single name tzǔ (zı̀).
7. The concept of ‘context of situation’ here intended is as developed by
Firth in linguistic analysis, where it ‘is best used as a suitable schematic
construct to apply to language events’ and should be regarded as ‘a group
of related categories at a different level from grammatical analysis but
rather of the same abstract nature’ (Firth 1952: esp. 7).
8. Professor Allen has kindly drawn my attention to the paper entitled
35
‘Transfer grammar’ (Chavarria-Aguilar 1954), which is an interesting

discussion of structural comparison in language teaching. See also the
important article by Harris (1954): ‘Transfer grammar’.
9. In the same way, a comparison of case–systems as envisaged by Allen must
depend on the prior identification of the noun in the two languages
compared, whether or not on the criterion of its place in the system of
word–classes.
10. It may be argued that the easy identification on graphical and phonological
grounds of the two words ‘dog’, on the collocation with which the
identification of zhı̄ / zēk depends, is in fact (or would be, when applied
outside the dialects of a single language) a falling back on vague translation
criteria. I should, however, maintain that this is where the contextual
criteria enter into the picture: if the terms to be compared cannot
themselves be definitely identified by precise contextual reference, then
they may be formally related (for example by collocation) to other forms
which can. A more difficult instance would be the terms in the aspect
systems (of clause–classes) in Pekingese and Cantonese; here again, once
the systems are shown to be comparable, one would proceed by seeking
to identify particular terms through their collocation with contextually
identifiable forms.
11. In Pekingese the prosodic identity is in terms of “y and w”: in the
determination of the quality of the vowel, final i = n, u = ŋ . Cf. J. R.
Firth (1948: esp. 136, reference to Firth’s ‘The Chinese monosyllable
in a Hunanese dialect (Changsha)’) with B. B. Rogers: “The prosodic
diacritica (of the Hunanese syllable) included . . . yotization and labio-
velarization, symbolized by y and w”.
36
Chapter Two
CATEGORIES OF THE THEORY

OF GRAMMAR (1961)
There have been in the main two approaches to description in modern

linguistics: the “textual” and the non-textual or, for want of a better
word, “exemplificatory”. More recently a third has been added, 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 that
transformative generation should replace other types of description1 as
a linguistic method of making statements about language.2 Others,
myself included, feel that all three approaches have a fundamental place
in linguistics; that they do different things, and that the third is a
valuable supplement to the first two.3
Description is, however, not theory. All description, whether gen-
erative or not, is related to General Linguistic theory; specifically, to
that part of General Linguistic theory which accounts for how language
works. The different types of description are bodies of method which
derive from, and are answerable to, that theory. Each has its place in
linguistics, and it is a pity to deny the value of textual description
(which is appropriate, for example, in “stylistics”, the linguistic study
of literature) just because certain of the methods used in description are
found to be inadequate.
My purpose in writing this paper is to suggest what seem to me to
be the fundamental categories of that part of General Linguistic theory
which is concerned with how language works at the level of grammar,
with brief reference to the relations between grammar and lexis and
between grammar and phonology. The theory sketched out here
First published in Word, 1961, 17(3), pp. 241–92.
37
derives most of all from the work of J. R. Firth.4 At the same time I
do not of course imply that I think Professor Firth would necessarily
have found himself in accord with all the views expressed, which in
some places depart from his own; nor do I underestimate the debt to
my present colleagues and the many others whose work I have,
obviously, drawn on.5
No excuse is needed, I think, for a discussion of General Linguistic
theory. While what has made linguistics fashionable has been, as with
other subjects, the discovery that it has applications, these applications
rest on many years of work by people who were simply seekers after
knowledge. It would not help the subject if the success of these
applications led us into thinking that the theoretical problems were
solved and the basic issues closed.
1 Starting-point
It will perhaps be helpful if the point of departure is first made clear.
The following is a summary of what is taken as “given” for the
purposes of this paper.
1.1 One part of General Linguistic theory is a theory of how language

works. It is from this that the methods of Descriptive Linguistics are
derived.
1.2 The relevant theory consists of a scheme of interrelated categories

which are set up to account for the data, and a set of scales of
abstraction which relate the categories to the data and to each other.
The data to be accounted for are observed language events, observed
as spoken or as codified in writing, any corpus of which, when used as
material for linguistic description, is a “text”.6
1.3 Description consists in relating the text to the categories of the

theory. The methods by which this is done involve a number of
processes of abstraction, varying in kind and variable in degree. It is the
theory that determines the relation of these processes of abstraction to
each other and to the theory.7
1.4 The theory requires that linguistic events should be accounted

for at a number of different levels: this is found to be necessary because
of the difference in kind of the processes of abstraction involved.8
38
categories of the theory of grammar
1.5 The primary levels are form, substance and context. The substance
is the material of language: phonic (audible noises) or graphic (visible
marks). The form is the organization of the substance into meaningful
events: meaning is a concept, and a technical term, of the theory (see
below, 1.8). The context is the relation of the form to non-linguistic
features of the situations in which language operates, and to linguistic
features other than those of the item under attention: these being
together “extratextual” features.
1.6 The complete framework of levels requires certain further 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 related

body of theory, that of General Phonetics. Since phonology relates
form and phonic substance, it is the place where linguistics and
phonetics interpenetrate. Linguistics and phonetics together make up
“the linguistic sciences”.10
Figure 1 Levels of language
39
1.8 Language has formal meaning and contextual meaning. Formal mean-
ing is the “information” of information theory, though (i) it can be
stated without being quantified and was in fact formulated in linguistics
independently of the development of information theory as a means of
quantifying it,11 and (ii) formal meaning in lexis cannot be quantified
until a method is found for measuring the information of non-finite
(“open”) sets (see below, 2.1 and 8.2). The formal meaning of an item
is its operation in the network of formal relations.
1.9 Contextual meaning, which is an extension of the popular – and

traditional linguistic – notion of meaning, is quite distinct from formal
meaning and has nothing whatever to do with ‘information’.12 The
contextual meaning of an item is its relation to extratextual features;
but this is not a direct relation of the item as such, but of the item in
its place in linguistic form: contextual meaning is therefore logically
dependent on formal meaning.13
1.10 It follows from 1.8 and 1.9 that, in description, formal criteria
are crucial, taking precedence over contextual criteria; and that the
statement of formal meaning logically precedes the statement of contex-
tual meaning.14
1.11 Finally, it is necessary to distinguish not only between theory

and 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 specific
purpose intended. Description depends on the theory; theoretical
validity is demanded, and relative merit is judged by reference to
comprehensiveness and delicacy.15
2 Grammar
2.1 Grammar is that level of linguistic form at which operate closed
systems.16,17 Since a system is by definition closed, the use of the term
“closed” here is a mnemonic device; but since “system” alone will be
used as the name of one of the four fundamental grammatical categories
(see below, 6) it is useful to retain “closed system” when referring to
the system as the crucial criterion for distinguishing grammar from
lexis.
A closed system is a set of terms with these characteristics:
40
(a) The number of terms is finite: they can be listed as A B C D,

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

used the terms “hierarchy”, “taxonomy” and ‘cline’ as general scale–
types. A hierarchy is taken to mean a system of terms related along a
single dimension which must be one involving some form of logical
precedence (such as inclusion).20 A taxonomy is taken to mean a special
type of hierarchy, one with two additional characteristics: (i) there is a
constant relation of each term to the term immediately following it,
and a constant reciprocal relation of each to that immediately preceding
it; and (ii) degree is significant, so that the place in order of each one
of the terms, statable as the distance in number of steps from either
end, is a defining characteristic of that term.21 A cline resembles a
hierarchy in that it involves relation along a single dimension; but
instead of being made up of a number of discrete terms a cline is a
continuum carrying potentially infinite gradation.
2.3 In this view of linguistics description is, as already emphasized, a

body of method derived from theory, and not a set of procedures. This
has one important consequence. If description is procedural, the only
way of evaluating a given description is by reference to the procedures
themselves: a good description is one that has carried out the right
procedures in the right order, but for any more delicate evaluation
external criteria have to be invoked. Moreover, every language has to
be treated as if it was unknown, otherwise procedural rules will be
violated; so the linguist has to throw away half his evidence and a good
few of his tools.
A theory on the other hand provides a means for evaluating
descriptions without reference to the order in which the facts are
accounted for. The linguist makes use of all he knows and there is no
priority of dependence among the various parts of the description. The
best description is then that which, comprehensiveness presupposed, is
maximally grammatical: that is, makes maximum use of the theory to
account for a maximum amount of the data. Simplicity has then to be
invoked only when it is necessary to decide between fewer systems
with more terms and more systems with fewer terms; and since both
information theory and linguistic intuition favour the latter even this
preference might be built in to the theory.22
3 Unit
3.1 Language is patterned activity. At the formal level, the patterns
are patterns of meaningful organization: certain regularities are
42
exhibited over certain stretches of language activity. An essential feature

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

of course in description.25 The possibilities are sequence, inclusion and
conflation. Thus if in a given instance a unit of one rank consists of
two units of rank next below, these may appear in form as one
following, interrupting, or overlaying the other.
Three further points about the rank relation need to be clarified.
First, the theory allows for downward rankshift: the transfer of a
(formal realization of a) given unit to a lower rank. Second, it does not
allow for upward rankshift. Third, only whole units can enter into
higher units. Taken together these three mean that a unit can include,
in what it consists of, a unit of rank higher than or equal to itself but
not a unit of rank more than one degree lower than itself; and not, in
any case, a part of any unit.26
3.3 The number of units in the hierarchy is a feature of the descrip-

tion. It varies from language to language, but is fixed by the description
for each language, or rather for each describendum or “état de langue”.
The possibility of there being only one is excluded by the theory, since
a hierarchy cannot be composed of one member. It is, however,
theoretically possible to conceive of a language having only two, and
an artificial language could be constructed on these lines (whereas it
would not be possible to construct an artificial language having only
one unit). English grammar, as far as it has been studied to date, seems
to require five, though further, statistical, work on grammar might
yield at least one more.
No special status, other than that27 presupposed by rank, is assigned
by grammatical theory to any one unit. Since in any case only two, as
a minimum, are required, only two would be available for special
status. As it happens we can assign special status to two grammatical
units by reference to other levels on a “more / less” basis. There will
always be one unit which, more than any other, offers itself as an item
for contextual statement because it does the language work in 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 and
thus offers itself as an item for lexical statement;28 this we may as well
always call the word. So, in grammatical theory, all languages have at
least two units; in description, all languages have sentences and all
languages have words – but the “sentenceness” of the sentence and the
“wordness” of the word do not derive from the theory of grammar.
Various names are available for those above, below or in between
44
when they turn up. For English, for the two units between sentence
and word the terms clause and phrase are generally used. It is at the
rank of the phrase that there is most confusion – because there are here
the greatest difficulties – in the description of English; one reason is
that in English this unit carries a fundamental class division (see below,
5), so fundamental that it is useful to have two names for this unit in
order to be able to talk about it: I propose to call it the group, but to
make a class distinction within it between group and phrase. Below the
word, English has one unit, called by the general name for the unit of
lowest rank, the morpheme.29
So in the description of English the sentence30 consists of one or
more complete clauses, the clause of one or more complete groups, the
group of one or more complete words and the word of one or more
complete morphemes. The descriptive meaning of “consists of”, and
the possibilities of rankshift (including recursive rankshift), are stated as
and where applicable. One distinction that is often useful is between a
member of a unit that consists of only one member of the unit next
below and one that consists of more than one; the former may be
called simple and the latter compound, but if this is done the terms must
be kept rigorously to this, and no other, use.31
3.4 The theory requires that each unit should be fully identifiable in
description. This means that, if the description is textual, every item of
the text is accounted for at all ranks, through the various links of the
exponence chain which involve, of course, the remaining theoretical
categories. If the description is exemplificatory, exactly the same is
implied, except that the description proceeds from category to expo-
nent instead of from exponent to category.
It will be clear from the discussion in the next sections that there
can be no question of independent identification of the exponents of
the different units, since criteria of any given unit always involve
reference to others, and therefore indirectly to all the others. A clause
can only be identified as a clause if a sentence can be identified as a
sentence and a group as a group, and so on up and down the line. For
this reason description is not and can never be unidirectional: it is
essential to “shunt”, and “shunting” is a descriptive method that is
imposed on description by theory.
45
4 Structure
4.1 The unit being the category of pattern-carrier, what is the nature
of the patterns it carries? In terms once again of language as activity,
and therefore in linear progression, the patterns take the form of the
repetition of like events. Likeness, at whatever degree of abstraction, is
of course a cline, ranging from “having everything in common” to
“having nothing in common”. The commonplace that no two events
are ever identical, that the same thing can never happen twice, is of no
relevance whatever to linguistics; as soon as description starts, however
little the generalization involved, absolute identity is a necessary
hypothesis, which is then built into the theory, as one endpoint of the
likeness cline. Likeness, including absolute identity, is of course re-
defined for each level and each category.
In grammar the category set up to account for likeness between
events in successivity is the structure.32 If the relation between events
in successivity is syntagmatic, the structure is the highest abstraction of
patterns of syntagmatic relations. The scale used for talking about it,
and for its graphic display, will most naturally be the orthographic
scale: to those of us brought up on the roman alphabet this happens to
run horizontally from left to right, which is enough reason for adopting
this version of the scale. But, as in the case of the unit, it must be
stressed that linear progression itself is a feature of substance. A structure
is made up of elements which are graphically represented as being in
linear progression; but the theoretical relation among them is one of
order. Order may, but does not necessarily, have as its realization
sequence, the formal relation carried by linear progression; sequence is
at a lower degree of abstraction than order and is one possible formal
exponent of it.33
4.2 A structure is thus an arrangement of elements ordered in places.

Places are distinguished by order alone: a structure XXX consists of
three places. Different elements, on the other hand, are distinguished
by some relation other than that of order: a structure XYZ consists of
three elements which are (and must be, to form a structure) 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 a
structure of a given unit.
Each unit may display a range of possible structures, and the only
theoretical restriction is that each unit must carry at least one structure
that consists of more than one place.35 Each place and each element in
46
the structure of a given unit is defined with reference to the unit next
below. Each place is the place of operation of one member of the unit
next below, considered as one occurrence. Each element represents the
potentiality of operation of a member of one grouping of members of
the unit next below, considered as one item–grouping.36 It follows
from this that the lowest unit has no structure;37 if it carried structure,
there would be another unit below it.
4.3 In description, structures are stated as linear arrangements of

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

structure, and this can be shown by an arrow placed over the symbols
for the elements concerned. For example, in English clause structure it
씮
is a crucial criterion of the element S that it precedes P in sequence:
씮 씮 씮
structures can be stated as SPCA, SAPA, ASP, etc.41 This displays the
contrast between this situation, where S is crucially defined by position
relative to P, and realized sequences of elements which are not,
however, defined by sequence, which may be indicated by simple
linearity of the symbols.42
4.4 In the consideration of the places and elements of structure of

each unit, which of course vary from language to language and from
unit to unit within a language, a new scale enters, that of delicacy (see
below, 7.4). This is depth of detail, and is a cline running from a fixed
point at one end (least delicate, or primary) to that undefined but
theoretically crucial point (probably statistically definable) where 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 the
minimum number of elements necessary to account comprehensively
for the operation in the structure of the given unit of members of the
unit next below: necessary, that is, for the identification of every item
at all ranks. (M)H(Q), and the various possible combinations of S, P,
C, A, are primary structures: one cannot account for all words in group
structure, or all groups in clause structure, with fewer than these
elements or places.
Subsequent more delicate differentiations are then stated as secondary
structures. These are still structures of the same unit, not of the unit
next below; they take account of finer distinctions recognizable at the
same rank.43 Rank and delicacy are different scales of abstraction:
primary group structures differ in rank from primary clause structures,
but are at the same degree of delicacy; while primary and secondary
clause structures differ in delicacy but not in rank.
As the description increases in delicacy the network of grammatical
relations becomes more complex. The interaction of criteria makes the
relation between categories, and between category and exponent,
increasingly one of “more / less” rather than “either / or”. It becomes
necessary to weight criteria and to make statements in terms of
probabilities. With more delicate secondary structures, different com-
binations of elements, and their relation to groupings of the unit next
below, have to be stated as more and less probable.44 The concept of
48
“most delicate grammar”, and its relation to lexis, is discussed below

(see 6.3 and 7.4); but the “more / less” relation itself, far from being
an unexpected complication in grammar, is in fact a basic feature of
language and is treated as such by the theory. It is not simply that all
grammar can be stated in probability terms, based on frequency counts
in texts: this is due to the nature of a text as a sample. But the very fact
that we can recognize primary and secondary structures – that there is
a scale of delicacy at all – shows that the nature of language is not to
operate with relations of “always this and never that”. Grammatical
theory takes this into account by introducing a special scale, that of
delicacy, to handle the improbability of certainty; this frees the rest of
the theory from what would otherwise be the weakening effect of this
feature of language. The category of structure, for example, is the more
powerful because it can be used to state the patterns of a given unit
comprehensively at the primary degree without the assumption that it
has accounted for all the facts.
5 Class
5.1 The structure is set up to account for likeness between events of
the same rank, and it does so by referring them to the rank next below.
To one place in structure corresponds one occurrence of the unit next
below, and at each element operates one grouping of members of the
unit next below. This means that there will be certain groupings of
members of each unit identified by restriction on their operation in
structure. The fact that it is not true that anything can go anywhere in
the structure of the unit above itself is another aspect of linguistic
patterning, and the category set up to account for it is the class.
The class is that grouping of members of a given unit which is
defined by operation in the structure of the unit next above. It accounts
for a paradigmatic relation, being a grouping of items “at risk” under
certain conditions. It is related primarily to elements of structure: the
first degree of classification yields classes which stand in one / one
relation to elements of primary structures, and these we may call
primary classes.
5.2 Class, like structure, is variable in delicacy. Clearly, in the first

place, more delicate classes are the product of more delicate structures:
in fact, “secondary” classes are derived from structure in two ways.
Firstly the same element at different places in structure may yield
distinct secondary classes. If a given unit has primary structures XY,
49
XYZ, YZ and XYZY, the primary classes of the unit next below are
“class operating at X”, “class operating at Y” and “class operating at
Z”. If, however, there is a further restriction such that in XYZY,
which will now be (secondarily) rewritten XYaZYb, only a section of
the members of the class at Y can operate at Ya and only another
section (not necessarily mutually exclusive) at Yb, this yields as second-
ary classes “class operating at Ya” and “class operating at Yb”. Secondly,
with increased delicacy the elements of primary structure will be
differentiated into secondary elements. A primary structure generalized
as X . . . nYZ, of which XXXXYZ is an instance, shows a generalized
relation of X to (say) Y; but there may be internal relations within X
. . . n such that XXXX is rewritten pqrs. These will yield secondary
classes “class operating at p”, “class operating at q”, etc.
In the second place, more delicate classes appear whenever a
restriction is found which differentiates among the members of a
primary class. There may be a relation of mutual determination, or
“concord”, between two classes; each divides into two sections such
that a member of one section of one class is always accompanied by a
member of one section of the other class. Thus if the primary class at
X is 1 and that at Y is 2, a structure XY must have as its exponent
either 1.1+2.1 or 1.2+2.2. Secondary classes arrived at in this way in
description may be referred to distinctively as sub-classes, to indicate
that they are derived by differentiation from primary classes without
reference to secondary structures; but it is important to state that there
is no theoretical difference here. The relation between structure and
class is a two-way relation, and there is no question of “discovering”
one “before” the other. In any given instance there may be descriptive
reasons for stating the one without the other; but all structures
presuppose classes and all classes presuppose structures.
5.3 What is theoretically determined is the relation between struc-

ture and class on the one hand and unit on the other. Class, like
structure, is linked to unit: a class is always a class of (members of) a
given unit: and the class–structure relation is constant – a class is always
defined with reference to the structure of the unit next above, and
structure with reference to classes of the unit next below. A class is not
a grouping of members of a given unit which are alike in their own
structure. In other words, by reference to the rank scale, classes are
derived “from above” (or “downwards”) and not “from below” (or
“upwards”).
The distinction between downward and upward movement on the
50
rank scale is important in grammar, but it is a mistake to raise it to the

status of a choice between different theories, which it is not. The
“formal / functional” dichotomy is one of those which linguistics is
better rid of;45 it is misleading to say even that classes are functionally
determined, since they are set up with reference to the form of the
unit next above – the whole description is both formal and functional
at the same time, and “function” is merely an aspect of form. The
distinction does, however, need a name, and this seems the best use for
the terms “syntax” and “morphology”. Traditionally these terms have
usually referred to “grammar above the word” (syntax) and “grammar
below the word” (morphology); but this distinction has no theoretical
status.46 It has a place in the description of certain languages, “inflex-
ional” languages47 which tend to display one kind of grammatical
relation above the word (“free” items predominating) and another
below the word (“bound” items predominating). But it seems worth-
while making use of “syntax” and “morphology” in the theory, to
refer to direction on the rank scale. “Syntax” is then the downward
relation, “morphology” the upward one; and both go all the way.
We can then say, simply, classes are syntactical and not morphological.48
(A terminological alternative is to talk of “syntactic classes” and
“morphological classes”, but this has the disadvantage that it destroys
the consistency in the use of the term “class”, no longer restricting it
to a category of unique theoretical status. When a name is needed for
morphological groupings, groupings of items on the basis of likeness in
their own structure, the term “paradigm” is available.49)
5.4 In description the term “class” covers primary and secondary

classes. It is often unnecessary to specify; but it is useful to state primary
classes first, since these form the link between elements of structure
and more delicate classes.50 For the different classes of each unit, many
names are available, especially at word rank – though it is precisely the
long history of terms like “verb” and “noun” which reinforces their
need for rigorous redefinition in the theory. In other cases names can
often be found, especially where structural restrictions permit class
identification between units: if certain word–classes only (or predomi-
nantly) operate in one class of group and others only in another, one
can talk of verbal group and verb (= “verb word”), nominal group and
noun (word) and so on. At the same time it is safer not to allow
complete terminological identity between units: “verb” alone should
not serve as the name both for a class of group and for a class of word.
In the use of symbols for classes, figures have the advantage of
51
avoiding confusion with elements of structure. This is not only the-

oretically desirable, because of their different status; it has descriptive
value in that the theoretical one / one relation between elements and
classes allows for instances where two different elements of structure,
standing in different relation to each other or to a third, yield primary
classes the membership of which is coextensive: these then form a
single primary class derivable simultaneously from two elements of
structure.51 If letters are used for elements of structure, and figures for
classes, the relation between the two can be demonstrated by the use
of a colour code.
6 System
6.1 Up to this point the theory has accounted for three aspects of
formal patterning: the varying stretches that carry patterns, the ordered
repetition of like events that makes up the patterns and the grouping
of like events by their occurrence in patterns. What remains to be
accounted for is the occurrence of one rather than another from among
a number of like events.
The category set up for this purpose is the system.52 This falls under
the general definition of system given above (2.1). But this does not
yet state its place in grammatical theory, its relation to the other
fundamental categories.
The class is a grouping of items identified by operation in structure:
that is, what enters into grammatical relations of structure is not the
item itself considered as a formal realization53 but the class, which is
not a list of formal items but an abstraction from them. By increase
in delicacy, the primary class is broken down into secondary classes of
the same rank. This set of secondary classes now stands in the relation
of exponent to an element of primary structure of the unit next above.
This gives a system of classes. If class 1 is the primary class (say of
the group) operating at X in (clause) structure, and this has secondary
classes 1.1, 1.2 and 1.3, then 1.1, 1.2 and 1.3 form a system of classes
operating at X. X is now shown to presuppose a choice – a choice that
is implied by the nature of the class (as a grouping of items) but that is
displayed first still in abstraction, by reference to the category of class
itself.54
6.2 Systems of secondary classes thus allow the description to remain

at a high degree of abstraction while displaying at each step, each
increase in delicacy, a more finely differentiated range of choice. This
52
is the value of the concept of “sub-class” (above, 5.2), since wherever

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

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

tal categories. The exponent of the element S in primary clause
structure is the primary class nominal of the unit group. The primary
54
structure of the nominal group is (M . . .n)H(Q . . .n). The primary

classes of the unit word operating at H and M are respectively the noun
and the pre-noun. The element M can be broken down into secondary
structures composed of any combination of the elements D, O, E in
that order, whose exponents are the secondary classes deictic, 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, at
Da) from “a / the / some, etc.” (class 2, at Db) and various sub-classes
involving concord. Adjectives likewise include a number of secondary
classes, but in most of these the exponents form open sets, and for
further treatment of these grammar hands over to lexis.62
7 Rank, exponence and delicacy

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

show what is in fact the relation between the grammar and the
phonology of the language.64
A meaningful account of how grammar and phonology are related
in English must be based on a prior separate statement of the two.65 As
said above (1.4), it is misleading to think of the levels as forming a
hierarchy. They represent different aspects of the “patternedness” of
linguistic activity. In this respect phonology is an “interlevel”, since
there is an added degree of dependence here: the patterns of substance,
and those of form, are each fully definable in their own terms, by
internal reference, so to speak; whereas the patterns of phonology,
being the organization of substance in form, although requiring to be
stated separately because of the distinct nature of the abstractions
involved, and the fact that (not surprisingly) the patterns are carried by
different stretches of exponents, are only significant with reference to
the two levels of form and substance. Context is an interlevel in a
different sense, since it relates language to something that is not
language; it is an interlevel because it is not with the non-language
activity itself that linguistics is concerned but with the relation of this
to language form. Though there is no precedence or priority, there is
of course order among the levels, as determined by their specific
interrelations; and in the study of language as a whole form is pivotal,
since it is through grammar and lexis that language activity is – and is
shown to be – meaningful.
7.2 The scale of rank has been discussed (above, 3) with reference to
the unit, the basic category which operates on this scale. The syntactic
(‘downward’) determination of classes is a feature of the theory, so that
with respect to the category of class the rank scale appears as one
of logical precedence running from highest to lowest unit. But this
precedence applies to class criteria only, so that even in the theory it is
not a one-way relation: the theory itself embodies “shunting” (moving
up and down the rank scale) as crucial to the interrelation of the
categories. In description, all statements presuppose shunting; the
description of the sentence cannot be complete until the description of
the morpheme is complete, and vice versa. In presentation, of course,
procedure varies according to purpose and scope: downward presenta-
tion seems easier to make clear, but this may well be overridden – for
example if grammar is an adjunct to lexical description, as sometimes
in the statistical study of lexis.
Rank is distinct both from exponence and from delicacy. A shift in
56
one never by itself entails a shift in either of the others. The reason
why rank is often confused with other scales is that there are cases
where a shift in rank does accompany a shift in something else; but this
is always by virtue of the logical relations among the categories
involved. The fact that by moving from structure to class, which is (or
can be) a move on the exponence scale, one also moves one step down
the rank scale, is due to the specific relation between the categories of
class and structure, and not to any inherent interdetermination between
exponence and rank. The descriptive relevance of keeping the scales
distinct is that it is important to be able to display what happens if one
shifts on one scale, keeping the other two constant.66
7.3 Exponence is the scale which relates the categories of the theory,
which are categories of the highest degree of abstraction, to the data.67
Since categories stand in different relations to the data, it might seem
necessary to recognize four different scales of exponence, one leading
from each category. In fact, however, exponence can be regarded as a
single scale.
In the first place, each category can be linked directly by exponence
to the formal item: it is in fact a requirement of the theory that any
descriptive category should be able to be so linked. This may be stated
by way of exemplification, as when we say “the old man is (an example
of ) an exponent of S in clause structure”. This is, however, not a
description of the element S, since by relating it to its exponent at a
stage when it was not necessary to do so we should have lost generality
(cf. above, 6.3). So instead of throwing up the grammatical sponge and
moving out to lexis while this is still avoidable, the description takes
successive steps down the exponence scale, changing rank where
necessary, until (at the degree of delicacy chosen) it is brought
unavoidably face to face with the formal item.
In the second place, therefore, the step by step move from any one
category to the data can proceed via any or all of the other categories.
This is then a move down the exponence scale, and at each step, given
that delicacy is constant, one category is replaced by another (either
with or without change of rank, according to which category is
replaced by which). While therefore the categories are distinct, they
are interrelated in such a way that the relation of exponence has the
status of a single scale.
The ultimate exponent in form is the formal item. This has then to
be related, in turn, to the substance. But this relationship, though it
may also be called “exponence”, entails a new scale, in which the
57
nature of the abstraction is different and the formal item is now at the
other end – it is itself the abstraction, and is not in any way delimited
or categorized by grammar.68
When grammar reaches the formal item, either it has said all there is
formally to be said about it or it hands over to lexis. The formal item
is the boundary of grammar on the exponence scale. It is not of course
the boundary on the rank scale: whenever the formal item is anything
other than a single morpheme (whether in closed system like “seeing
(that)”, or in open set like “pickup”) the grammar can be taken further
down in rank, since it can state the structures in terms of elements
whose exponents are words and morphemes. But seeing (that) enters
into a system at group rank, while pickup emerges from the grammar as
a word, though being a lexical item it would not necessarily be an
exponent of any whole grammatical unit. Once it has been taken over
by lexis, the grammatical categories, and the grammatical exponence
scale, no longer impinge on it (see below, 8.2).
7.4 Delicacy is the scale of differentiation, or depth in detail. It is a

cline, whose limit at one end is the primary degree in the categories of
structure and class. In the theory, the other limit is the point beyond
which no further grammatical relations obtain: where there are no
criteria for further secondary structures, or systems of secondary classes
or formal items. In description, delicacy is a variable: one may choose
to describe a language without going beyond the primary degree, still
being comprehensive in rank and exponence and making use of all the
categories of the theory. Each subsequent increase in delicacy delays
the move to the exponents (cf. above, 7.3) and thus increases the
grammaticalness of the description. The limit of delicacy is set by the
means at one’s disposal.
In well-described languages, such as English, any extension in
delicacy beyond what is already known requires either or both of large-
scale textual studies with frequency counts and complex secondary
classification based on multiple criteria, criteria which often cut across
each other and may have to be variably weighted. And, as suggested
above (6.3), a point will perhaps be reached where probabilities are so
even as to cease to be significant69 and classes so delicately differentiated
that the description will have to decide on crucial criteria and ignore
the others,70 thus setting its own limits.
Delicacy is distinct from rank and the limit of delicacy applies at the
rank of all units, for example differentiation of clause structures and of
classes of the group. At one stage, therefore, it becomes a limit on the
58
grammatical differentiation of items which then remain to be lexically

differentiated: it sets an endpoint to grammar where lexis takes over.
Here the scales of delicacy and exponence meet. The endpoint set to
grammar on the exponence scale is where abstraction ceases: one has
to move from abstract category to exponential item. That set on the
delicacy scale is where differentiation ceases: the set of exponents of
each class, and of each element of structure, permits no further, more
delicate groupings. If the formal items are still not ranged in systems,
the implication in either case is that further relations among them are
lexical.
Whether or not, as discussed above (6.3), grammatical delicacy can
reach a point where there is a one / one category–exponent relation
(where each element of structure, and each class, has only one formal
item as exponent), when all formal relations, including those among
what are now treated as lexical items, can be accounted for by the
grammatical categories and stated grammatically – in other words,
whether or not, ultimately, all linguistic form is grammar, we do not
know. At present, lexical items must be treated separately, and lexical
relations established in their own right. These lexical relations do not
depend on grammatical categories71 (so they are not yet “most delicate
grammar”) and they have their own dimensions of abstraction (so not
yet “most exponential grammar”). So there must be a theory of lexis,
to account for that part of linguistic form which grammar cannot
handle.
8 Lexis
8.1 This section is intended merely to bring lexis into relation with
grammar, not to discuss the theory of lexis as such. As has been pointed
out (above, 3.3. and 6.3), there is no one / one correspondence in
exponence between the item which enters into lexical relations and
any one of the grammatical units. It is for this reason that the term
lexical item is used in preference to word, “word” being reserved as the
name for a grammatical unit, that unit whose exponents, more than
those of any other unit, are lexical items.
Not only may the lexical item be coextensive with more than one
different grammatical unit; it may not be coextensive with any 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 same
formal item) may be a number of different grammatical items, so it is
59
not true that one lexical item always has the same relation to the rank
hierarchy. So that, in English, (i) a lexical item may be a morpheme,
word or group (at least); (ii) a lexical item may be assigned to no rank,
being for example more than a word but less than a whole group, or
even both more and less than a word – part of one word plus the
whole of another, sometimes discontinuously; and (iii) one and the
same lexical item may in different occurrences cover any range of the
possibilities under (i) and (ii).72
This does not mean that lexical items cannot be identified in
grammar; it means that they are not identified by rank. They are
identified, as has been suggested (above, 6.3), by their being 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 next
stage is to relate them; and it is here that the complex relation between
lexical item and grammatical unit must be accounted for. This is exactly
parallel to what was said above (7.1) about grammar and phonology;
and, of course, it applies equally to phonology and lexis, where, after
separate description, is displayed the relationship between the lexical
item and the categories of phonology.
8.2 The task of lexis can be summed up, by illustration, as that it has
to account for the likelihood of wingless green insects and for the, by
contrast, unlikelihood of colourless green ideas.73 As in grammar, we shall
expect language to work by contrasting “more likely” with “less likely”
rather than “possible” with “impossible”; but, as has often been pointed
out,74 this particular type of likelihood is not accounted for by
grammar, at least not by grammar of the delicacy it has yet attained. It
is, however, too often assumed that what cannot be stated 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 view
that the only formal linguistics is grammar might be described as a
colourless green idea that sleeps furiously between the sheets of
linguistic theory, preventing the bed from being made. What are
needed are theoretical categories for the formal description of lexis.
It seems that two fundamental categories are needed, which we may
call collocation and set.76 The first basic distinction between these and
the categories of grammar is that in lexis there are no scales of rank and
exponence. There is no hierarchy of units; therefore no rank scale.
There is only one degree of abstraction – a set is a set of formal items
60
and a collocation is a collocation of formal items; therefore no

exponence scale (exponence there is, of course, but it is a simple
polarity). Only the scale of delicacy remains; sets and collocations can
be more and less delicate.
There is an analogy with the categories of grammar, an analogy due
to the nature of language as activity. Collocation, like structure,
accounts for a syntagmatic relation; set, like class and system, for a
paradigmatic one. There the resemblance ends.
Collocation is the syntagmatic association of lexical items, 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 with
which it is collocated being ranged from more to less probable; and
delicacy is increased by the raising of the value of n and by the taking
account of the collocation of an item not only with one other but
with two, three or more other items. Items can then be grouped
together by range of collocation, according to their overlap of, so to
speak, collocational spread. The paradigmatic grouping which is
thereby arrived at is the set. The set does not form a closed system,
but is an open grouping varying in delicacy from “having some
(arbitrary minimum) collocation in common” to subsets progressively
differentiated as the degree of collocational likeness set as defining
criterion increases.77
In lexis, as in grammar, it is essential to distinguish between formal
and contextual meaning.78 Once the formal description has identified
the categories and the items, these can and must be treated context-
ually. The formal item of lexis, the lexical item, is unrestricted gram-
matically; grammatical categories do not apply to it,79 and the
abstraction of the item itself from a number of occurrences (including,
for example, the answer to the question whether one is to recognize
one lexical item or more than one) depends on the formal, lexical
relations into which it enters. The nature of these relations is such
that formal statements in lexis require textual studies involving large-
scale frequency counts: not of course of the frequency of single items,
but of items in collocation. Since these are no longer difficult to
undertake, it should not be long before we find out much more about
how language works at this level.
61
9 An analogy
9.1 Eating, like talking, is patterned activity, and the daily menu may
be made to yield an analogy with linguistic form. Being an analogy, it
is limited in relevance; its purpose is to throw light on, and suggest
problems of, the categories of grammar by relating these to an activity
which is familiar and for much of which a terminology is ready to
hand.
The presentation of a framework of categories for the description of
eating might proceed as follows:
Units:
Daily menu
Meal
Course
Helping
Mouthful
Unit: Daily menu

Elements of primary structure E, M, L, S (“early”, “main”,
“light”, “snack”)
Primary structures EML EMLS (conflated as
EML(S))
Exponents of these elements E: 1 (breakfast)
(primary classes of unit M: 2 (dinner)
‘meal’) L: 3 (no names available; see
secondary classes)
S: 4
Secondary structures ELaSaM ELaM EMLbSb
EMSaLc
Exponents of secondary La: 3.1 (lunch)
elements (systems of Lb: 3.2 (high tea)
secondary classes of unit Lc: 3.3 (supper)
‘meal’) Sa: 4.1 (afternoon tea)
Sb: 4.2 (nightcap)
System of sub-classes of unit E: 1.1 (English breakfast)
‘meal’ 1.2 (continental breakfast)
Passing to the rank of the “meal”, we will follow through the class
“dinner”:
62
Unit: Meal, Class: dinner

씮
Elements of primary structure F, S, M, W, Z (“first”,
“second”, “main”,
“sweet”, “savoury”)
Primary structures MW MWZ MZW FMW
FMWZ FMZW FSMW
FSMWZ FSMZW
(conflated as (F(S)MW(Z))
Exponents of these elements F: 1 (antipasta)
(primary classes of unit S: 2 (fish)
“course”) M: 3 (entrée)
W: 4 (dessert)
Z: 5 (cheese*)
Secondary structures (various, involving secondary
elements
Fa–d, Ma,b, Wa–c)
Exponents of secondary Fa: 1.1 (soup)
elements (systems of Fb: 1.2 (hors d’oeuvres)
secondary classes of unit Fc: 1.3 (fruit)
“course”) 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 Fa: 1.11 (clear soup*)
“course” 1.12 (thick soup*)
S: 2.01 (grilled fish*)

2.02 (fried fish*)
2.03 (poached fish*)
Wb: 4.21 (steamed pudding*)

4.22 (milk pudding*)
Exponential systems operating Fc: grapefruit / melon

in meal structure Fd: grapefruit juice / pineapple
juice / tomato juice
Ma: beef / mutton / pork
Mb: chicken / turkey / duck /
goose
63
At the rank of the “course”, the primary class “entrée” has secondary
classes “meat dish” and “poultry dish”. Each of these two secondary
classes carries a grammatical system whose terms are formal items. But
this system accounts only for simple structures of the class “entrée”,
those made up of only one member of the unit “helping”. The class
“entrée” also displays compound structures, whose additional elements
have as exponents the (various secondary classes of the) classes “cereal”
and “vegetable”. We will glance briefly at these:
Unit: Course, Class: entrée

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 J: 1 (flesh)
(primary classes of unit T: 2 (cereal)
“helping”) A: 3 (vegetable)
Secondary structures (various, involving – among
others – secondary
elements Ja,b, Ta,b, Aa,b)
Exponents of secondary Ja: 1.1 (meat)
elements (systems of Jb: 1.2 (poultry)
secondary classes of unit
‘helping’) Ta: 2.1 (potato)
Tb: 2.2 (rice)
Aa: 3.1 (green vegetable*)

Ab: 3.2 (root vegetable*)
And so on, until everything is accounted for either in grammatical
systems or in classes made up of lexical items (marked *). The
presentation has proceeded down the rank scale, but shunting is
presupposed throughout; there is mutual determination among all units,
down to the gastronomic morpheme, the “mouthful”.
Like the morpheme, the mouthful is the smallest unit, and all eating
activity can be broken down into mouthfuls. Like the morpheme, it is
neither more nor less fundamental than any other unit. But how little
it reveals, if description proceeds in one direction from it, of the
complexity of the whole activity!
Our predecessors thought of language as an organism, and drew
64
their analogies from evolution. We reject this as misleading; but no less

misleading is its familiar substitute, according to which language is an
edifice and the morphemes are the bricks. Perhaps if language had been
thought of as activity we should never have heard of “morphemics”. It
is unfair on one who was among the greatest figures in linguistic
science to call the “bricks and edifice” view of language “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 with
many individual varieties, which has had wide currency for the last
quarter of a century and owes a very great deal to Bloomfield’s work,
I would like to consider what seem to me to be certain questionable
features of “Bloomfieldian” linguistics. What is being called in question
is really the theory (and perhaps the analogy) of how language works.
10 The seven sins

10.1 From the point of view of the present theory, there are seven
features of what is here labelled “Bloomfieldian” method which would
perhaps justify critical comment.81 These are:
(1) 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) is
conflated with the relation between two different levels, grammar and
phonology. Ranged on a single scale (“from phoneme to utterance”)
are (i) the move between phonology and grammar (which, moreover,
always goes from the “interlevel” to the level; cf. below, 10.3) and (ii)
the move between units (which, moreover, always goes from the
lowest to the highest; cf. below, 10.5) in (a) phonology and (b)
grammar. In description, this leads to unwanted complexity and thus
weakens the power of the grammar: one must be free to recognize
grammatical units whose exponents in substance both overlap with and
completely fail to coincide with the units carrying phonological con-
trasts.82 In the theory, there is a confusion of abstractions: the abstrac-
65
tion involved in relating one unit to another at the same level is quite
different from that involved in relating one level to another. The sense
in which a sentence “consists of ” morphemes is stated in description
with reference to its definition in the theory, and is totally different
from the sense in which a morpheme “consists of ” phonemes – indeed
it is doubtful if there is any meaningful sense in which a morpheme
consists of phonemes. (The contrastive use of “morph” and “mor-
pheme” is designed to build in an extra stage to account for the two
kinds of abstraction. But it does not get over the first (descriptive)
difficulty; nor in fact does it solve the theoretical one, since morph and
phoneme differ in the extent and kind of formal determination under-
lying their phonological abstraction.83)
10.2 The relation among the units also tends to be confused with
the relation between a category and its exponent(s). It is assumed that
in moving up the rank scale, from morpheme to sentence, one is also
moving up the exponence scale. It is true (as said above, 6.2 and 7.3)
that in comprehensive description, in order to display the full gram-
maticalness of language, one takes the final step on the exponence
scale at the lowest rank possible (though this, as already shown, is
by no means always the morpheme),84 and this is probably the reason
for the confusion of rank with exponence. But the scales of rank
and exponence are again different dimensions of abstraction, and
one can link any unit directly to its formal exponent (and through
this to its exponent in substance): the relation of an exponent of the
unit “sentence” to the category of sentence is exactly the same as
that of an exponent of the unit “morpheme” to the category of
morpheme.
10.3 The conflation of levels referred to is the conflation of grammar

and phonology, which follows, though is distinct, from the confusion
of level with rank (above, 10.1). The theoretical basis of this criticism
is complex but crucial. Any distinction in substance may (i) be a free
variant, in which case it is formally meaningless, or it may (ii) carry a
distinction in form, with meaning either at the grammatical or at the
lexical level.85 All formal distinctions presuppose some distinction in
substance,86 and once a distinction in substance is shown to carry a
formal distinction it must be accounted for in phonology.87 But no
relation whatever is presupposed between the categories required to
state the distinction in form (grammar or lexis) and the categories
required to state phonologically the distinction in substance which
66
carries it.88 For example: the units may not be coextensive, and a
variable relation of phonological unit to grammatical unit may be the
very thing that carries a formal system (cf. above, 10.1 n. 82); or a
single system in grammar may be carried by different phonological
distinctions, say two of its terms by tone and a third by addition of a
segment; or a phonological system, such as tone associated with a given
unit (recognized as phonological because it carries some formal distinc-
tion), may carry different formal distinctions, part grammatical and part
lexical, or terms in different grammatical systems.89
The categories required by the grammar, and the criteria for these,
should come from within grammar. They are set up to provide a
description that is comprehensive, consistent and maximally powerful.
In the definition of, for example, the unit “clause”, the requirement is
that it should yield classes and structures which make possible the
description of the sentence, the group and so on in terms of their
structures and classes: hence the mutual definition of all units and of all
grammatical categories (and, procedurally, until a description is “com-
prehensive” (primary delicacy at all ranks) all parts of it remain subject
to revision). A grammatical category is not required to be identifiable
by reference to a particular feature of substance stated phonologically:
it merely carries the potentiality of being stated in phonological terms
through a long chain of exponence.90
(The starting point here of course is the theory of grammar, so that
what is being considered, and objected to, is the identification of
grammatical categories on phonological criteria. For linguistic
theory as a whole, the question must also be formulated the other way
round: do we derive phonological categories from formal ones? For
example, in the last case mentioned in the first paragraph of this section,
would we state one phonological system of tone or more than one?
The nature of phonology as an “interlevel” suggests that its categories
should be derived from those of form; and this is usually done in
prosodic phonology, as developed by Firth and others, though it is not
an inherent requirement of prosodic method.91 Inter-level dependence
in this direction is theoretically justifiable, since the role of phonology
is to account for the formally meaningful organization of phonic
substance. But it has descriptive dangers: first, that a system carried
only by variable relation between grammatical and phonological unit
may be missed, and second, that unless a comprehensive formal
description has first been made, formal features may be distorted into a
phonological mould – the phonologist may take the “word” class
“verb” as a phonological unit, but he (or someone) must have described
67
the grammar first, or his “word” and “verb” may not turn out to be
the grammarian’s “word” and “verb”.)
To be precise, then, what is being criticized here has again both a
theoretical and a descriptive aspect. In the theory, it is the use of
phonologically stated features as crucial criteria for grammatical
categories, as when supra-segmental phonemes are used as criteria for
the category of “phrase”.92 A phrase is a phrase because it operates in
the structure of the unit above it and has its own structures in terms of
the unit below it. It has then to be related to the phonological
categories which are arrived at by a different process of abstraction on
the basis of the minimal requirement that some formal distinction is
always involved. If at any point this yields a one / one relation of
categories, so much the better: if the phrase turns out to be exponen-
tially coextensive with the tone group, the latter can be used as a
recognition signal for the former. But it remains a signal, not a criterion.
In description, the trouble arises when phonological features are used
as grammatical criteria even when they clash. If, for example, the
segment which carries tone contrasts, or is bounded by juncture
features, will not work as a grammatical unit, then tone and juncture
are no use even as recognition signals; so to “define”, say, the clause
by reference to tone or juncture one has to set up a phonological
system in which any feature in substance can be an allophone of any
term in the system. There will be clash; and if it is recognized in the
first place that there is nothing at all surprising when, say, units carrying
formal patterns do not coincide with those carrying patterns of the
organization of substance, then the search for one / one phonological
identification signals of grammatical categories, such as a phonological
statement of clause boundary, can be abandoned as being without
profit.93
10.4 In both grammar and phonology the smallest units, morpheme

and phoneme, are often assigned a special status distinct from that of
any of the other units. Since the description usually proceeds unidirec-
tionally upwards (consistently in phonology, and in grammar up to
about group rank; cf. below, 10.5) these units are treated first: they
are then rated (uniquely) fundamental, and in phonology “supra-
segmental” features are given phoneme status. The morpheme
becomes 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”.
68
The result is sometimes referred to as excessive segmentation, though

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

accounts for only 90 per cent of occurrences; and this becomes more
true with every subsequent gain in delicacy.98 If this is built in to the
theory as a basic property of language, there is no need to try to turn
all constrasts and relations into “yes / no” ones, an attempt which may
slant the choice of criteria. Criteria are sometimes chosen for this
purpose which make it necessary to account for a large number of
occurrences by changing them into something else, because they do
not display the contrast chosen as crucial. (This device is more readily
tolerated by those who give relatively little weight to textual as opposed
to exemplificatory description.)
For example, the exponent of the element “subject” in English
clause structure is sometimes identified as being that nominal group
which is in person / number concord with the verbal group. But since
a large number of English clauses do not display this concord contras-
tively99 (quite apart from those which violate it, as does the clause I
have just written), the textual application of this criterion has to rely
on the substitution of all other verbal groups by one in is, was, has or
simple present form and, if the two nominal groups are alike in
number, on the dissimilation of one of them. If instead one says that
the exponent of “subject” is that nominal group which precedes the
verbal group with no other nominal group in between, this can be
stated as a high probability relation in spoken English and is applicable
to a textual study (in which it can be quantified) without substitution.
The next step is then to make more delicate statements to account for
the instances so far not accounted for, some of which turn out to be
grammaticizable and others, for the moment, not. And it ceases to be
puzzling why, if a foreigner (or native) says my friends is inviting the
neighbour, every English speaker knows that it is the friends who are
doing the inviting and not the neighbour.
10.7 The problem of “redundancy” is complex, and needs treating

separately and at length; but the term has become a commonplace in
description and a brief reference may be made here. The name is
assigned to a number of varied phenomena, none of which is related in
any clear way to the quantifiable redundancy of information theory.100
These include relations of formal categories to exponents in form, of
formal and phonological features to exponents in substance, and of
formal features to context. Moreover, in some instances the so-called
“redundancy” is simply put in by the method of description and has
no relevance to the language at all.
70
“Redundancy” is assigned to what is displayed as multiple expon-

ence: either in form, where more than one formal item is said to be
the exponent of one grammatical category, or in substance, where a
distinction is carried by what is said to be more than one phonetic
feature. But neither of these is at all clearcut. Formal redundancy
“occurs” where there is concord, but no criteria are available for
identifying the two prerequisites of concord: that there is “more than
one” exponent as opposed to “one”, and that these are exponents of
“the same” category.101 Discontinuous morphemes, for example, may
sometimes be clearly recognizable, though at others it is impossible to
say “how many” exponents are present;102 but the question is irrelevant,
since where the description does recognize concord this concord is
itself the exponent of a distinct category of relation that is different
from the category of which the form is exponent, and that has its own
formal meaning.103 Redundancy in substance appears when formal or
phonological distinctions are related to contrastive features.104 Here
precisely the same problem arises, since it is not possible to give
rigorous criteria for deciding what is “one” phonetic feature and what
is “more than one”.105 Each time a new parameter or a further degree
of differentiation is introduced into the phonetic statement, all its
precursors are thereby made “redundant”.
In extreme cases this “redundancy” becomes completely artificial,
since it is simply inserted by the description. This happens when a
contrast (or system) is assigned to a unit lower than that to which it is
appropriate, and may result therefore from overemphasis on the lowest
unit. This tends to happen more in phonology, when the phoneme is
made to carry contrasts appropriate to a higher unit; one of the merits
of prosodic phonology is that it avoids this error.106 But it is not
unknown in grammar, where it may also arise from the use of
morphological instead of syntactic criteria for classes.107 In such cases
“redundancy” can only refer to the loss of power in the grammar
brought about by such a description: it already follows from the theory
that the “appropriate” unit for the assignment of any feature is the
highest unit that can carry it without requiring the statement to be
made twice. The best description therefore can be thought of as that
which minimalizes artificial “redundancy”. But at the same time those
instances where what is called “redundancy” is an artificial product of
the description are not essentially different from, but are merely
extreme cases of, the “multiple exponence” in form and substance to
which the same name is applied.108
What is of doubtful validity here is the implication that there are
71
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 by
something more than what “would be enough” – to the extent even
of suggesting that one may judge which of a number of exponents is
“the” non-redundant one.110 On the other hand, even if it is possible
to devise some theoretically valid criteria for “redundancy” of this
kind, its relation to the redundancy of information theory is extremely
complex and it would be better not to call it by the same name. The
redundancy of information theory is of considerable interest to linguis-
tics in the study of the information carried by grammatical systems; but
this, as far as I know, has not yet been seriously attempted. The
quantification of systems, rather than the appraisal of features as
contrastive or idle, which rests on a very partial interpretation of the
redundancy of information theory, seems the more useful role for the
concept of redundancy in linguistics.
Most, if not all, of the points made in this section can be brought
together under Chomsky’s observation that “a linguistic theory should
not be identified with a manual of useful procedures, nor should it be
expected to provide mechanical procedures for the discovery of 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 a
practical evaluation procedure for grammars”?113 This it must do. But
it can be asked to do more: to provide a framework of logically
interrelated categories (so that it can be evaluated as a theory, and
compared with other theories) from which can be derived methods
of description, whether textual, exemplificatory or transformative–
generative, which show us something of how language works.
Notes
1. It is in no way to deny the fundamental importance of Chomsky’s work
(1957) and elsewhere, if we suggest that the readiness of linguists who
had previously worked in the “Bloomfieldian” tradition to abandon
these methods in favour of Chomsky’s is in part due to their lack of
72
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 be
theoretical. But a grammar of one short text may be based on theory;
and any theory-based grammar, transformational or not, can be stated in
generative terms.
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
73
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), as
what he (perhaps unfortunately, in view of the formal use of “system”)
calls “systemic linguistics”. The difficulty of this method lies in deciding
on what Bull calls “those features of objective reality which are pertinent
to the problem” (p. 3), since this can only be known by reference to
linguistic forms: cf. e.g. “it may be assumed that normal people auto-
matically divide, on the preverbal level, all events into three categories:
those anterior to PP (point present) . . . , those simultaneous with PP
74
. . . , 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 the
steps by which the theory was arrived at in the first place – in fact less,
since the latter can at least be formulated, while the former can only be
summed up in the words of the song: “I did what I could”.
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 contextual
distinctions; but this is not a property of a system. The “addition” of a
new term is not of course considered as a process (though historical
change is one type of instance of it): it may be displayed in any
comparison of two related systems. For example, two possible systems
of first and second person pronouns used by different speakers of Italian
(quoted in oblique disjunct form; I = “interior to social group”, E =
“exterior . . .”).
75
1 me me
1+ noi noi
2I te te
2I+ voi
2E
2E+
lei
loro
} voi
(The distinctions made in written Italian are ignored, since they would
not affect the point.) The difference in format meaning is a function of
the different number of terms: in system one me excludes five others, in
system two only three. In contextual meaning only terms of the second
group are affected.
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:
76
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 can
then 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
77
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
앖 sentence
rank
앗
(
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 the
latter: 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
78
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 construction
of other partially like clauses.) But though the difference in structure has
its manifestation in substance (there can of course be ambiguity in
substance, as in Hockett’s old men and women (1957: 390n.), in form the
difference is not realized in sequence. In sequence, from the Gas Board
occupies the same place in both instances; in order, S and A stand in
different relations to P, and from the Gas Board is exponent of (part of) S
in the one case and of (the whole of) A in the other.
Sequence is presumably always manifested in phonic substance as
linear progression; the distinction is then one of exponence, “sequence”
being the name for that formal relation between formal items of which
linear progression is the manifestation in phonic substance.
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
79
noted that, since they differ in kind, “size” cannot be abstracted as

common to the two dimensions of abstraction for them to differ in.
That is, a grammatical unit can only be exponentially coextensive (or
not) with a phonological one: when it is, this is a descriptive accident
for which the linguist can be thankful (cf. reference to Allen (1956) in
7.1 n. 64 below), but the grammar cannot be made to define the units
for phonological statement (cf. the example in 4.1 n. 33 above, where
two exponents of the same grammatical unit “clause” may be (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 primitive
terms 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 formula
would 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, 씮 the arrow can be
added (where it really belongs) in the inventory: S, P, C, A. It is then
no longer required in the statement of structures, since it is presupposed.
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 distinctions씮beween puer
puellam amat, puellam puer amat, etc. In English, where SP sequence is
crucial to the definition of S (though various arrangements of C and A
are possible), more delicate grammatical distinctions, such as those
carried by intonation, must be shown secondarily.
80
43. For example the following two exponents of the (class) nominal (of the
unit) group: all the ten houses on the riverside and the finest old houses on the
riverside have the same primary structure M . . . HQ (or MMMHQ).
But a more delicate statement of M, still at group rank, shows distinct
secondary structures, the first example having D2DbO, the second
DbOE.
44. When Hockett writes (1955: 17) “In general, then, if we find 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 cannot
be 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 signifié, expression and content. These
dichotomies are a quite unnecessary nuisance, and in my opinion should
be dropped” (p. 227). Cf. Firth (1957c: 2, 3 – though here, I must
admit, Firth also rejects “form and substance”, which I find crucial (as
levels) to an understanding of how language works.
46. Cf. Firth (1957a): “It follows that the distinction between morphology
and syntax is perhaps no longer useful or convenient in descriptive
linguistics” (p. 14).
47. i.e. languages in which inflexional systems are a regular feature of word
structure. Free and bound are generalized class categories, linked to the
generalized structure categories of simple and compound (above, 2.3):
“free” is “able to stand as exponent of one-element structure of the unit
next above”, “bound” is “unable to stand, etc”. A member of a “free”
class can thus be exponent of a “simple” structure, while a member of a
bound class can operate only in “compound” structures.
48. Other terms are of course available, like Haas’ (1954: 68ff.) “synthetic
classification” and “analytic classification”. The terminological objection
to the use of “class” in both (as in “form class” and “function class”) is,
however, theoretically founded: if we say, with Haas (p. 68), that “we
distinguish two ways of classifying” linguistic units, we imply two things:
(i) a choice of (ii) procedural direction. But this is not a procedural
matter, and there is no choice. All forms are to be accounted for, and
this means stating both their class (linking them to the unit next above)
and their own structure (linking them to the unit next below). Whether
81
the “syntactic” groupings, of items operating alike in the structure of

the units next above, and the “morphological” groupings, of items alike
in their own structure, coincide or not is a descriptive variable; other
things being equal, a form of description will be chosen in which they
do, since the more they coincide the more grammatical the statement.
But they must be terminologically permitted not to coincide without
prejudice to their both being stated. Cf. Robins (1959): “When there is
a conflict of classification by morphological paradigm and syntactic
function, the latter is given preference in assigning words to word–
classes.” (p. 109) – I would add “groups to group classes, etc.”
49. For example in the structure of the English verbal group, the words
work, play operate at the same element: they are members of the same
word–class. The words works, working, worked do not operate at the
same element: they are not members of the same word–class. They
have themselves, however, the same (primary) structure: they are 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
䉳䉴
Highest
abstraction 䉱 Primary Secondary
structure 䉴 structure
Scale of
Exponence 䉲䉲
Primary Secondary
䉲 class 䉴 class
Data
51. Or nearly coextensive: the criteria for the setting up of one primary class
or two are descriptive. For example, in English clause structure S and C
are different elements standing in different relation to P. There is a high
degree of overlap between their exponents: one primary class (class
“nominal” of unit “group”) can be set up as exponent of both S and C.
The lack of exact coextensiveness will be stated by secondary elements
and classes, to account for (for example) the occurrence of the old hall,
the old town, the old town hall, this hall / town / town hall is old, this is a hall /
town / town hall, and the non-occurrence of this old is a hall, this is an old
or this hall is town.
82
52. Cf. Firth (1957b): “Various systems are to be found in speech activity
and when stated must adequately account for such activity. Science
should not impose systems on languages, it should look for systems in
speech activity, and, having found them, state the facts in a suitable
language” (p. 144). Cf. also references given in 4.1 n. 32, above.
53. Again, abstraction on the exponence scale. The formal item the old
man is exponent of (is a member of) a class (“nominal”, of the unit
“group”). The class “nominal group” is exponent of (operates at the
place of) an element of structure (S or C, of the unit “clause”). The
formal item itself, of course, has its own (and ultimate) exponents in
phonic or graphic substance.
54. Diagrammatically (axes as in 5.4 n. 50, above):
Primary structure Secondary structures

XYZY X Ya Z Yc and
X Yb Z Yc
SYSTEM
1 (at X) 䉳
2.1 (at Ya)
2 (at Y) 2.2 (at Yb)
3 (at Z) 2.3 (at Yc)
Primary classes Secondary classes

55. As already stressed (above, 2.2), the order of presentation here is for
convenience of exposition; the relations among the theoretical categories
do not involve logical precedence.
56. Since the categories are set up precisely to account for the data that are
stated as their exponents, this is not surprising. Cf. Firth (1957a): “A
theory derives its usefulness and validity from the aggregate of experience
to which it must continually refer in renewal of connexion” (p. 1). The
relation of category to exponent can be generalized as one of abstraction;
one endpoint of this relation may be any one of four categories, but
there is no scale of abstraction among the categories – their relation to
each other is such that the move from any category to its exponent may
be made either directly or via any or all of the other categories. As said
below (6.2), the route may involve rankshift; but this does not mean
that rank is to be equated with exponence or that there is any distinction
between different units as regards the kind or degree of their relation to
their exponents. (That is to say, even if one chooses to move from
“clause” to exponent via “group”, this does not mean that the group is
in any sense “nearer” to the data than the clause; indeed, the move from
83
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 . . .
exponents
⬙ structures
group (system of) classes . . .
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 scale
into the theory. Cf. above, 3.2 n. 26.
64. Cf. Allen (1956), from which the following is taken: “It frequently
occurs that an appropriate ‘bit’ of the corresponding phonological
statement (or of the orthography) is used as a label for the grammatical
unit in question . . . The price of using such labels is constant vigilance
. . . Where the phonological analysis permits of alternatives, that alterna-
tive is to be chosen which is most congruent with the grammatical
84
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 which
simplifies 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 compare
classes of one unit with classes of the unit next below: shift in rank only.
Or one may want to state and exemplify the classes of a given unit: shift
in exponence only.
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’s
major contributions to linguistic theory. References will be found
throughout the writings of Allen and of Carnochan, Eugénie 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
85
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, and
cline, 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 verbal
of 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 finite
of 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
86
item is run which is an exponent of the unit “morpheme”. In (ii) the

lexical item is ran up which is exponent of the unit ‘group’ but above
the rank of the word. (Orthography does not of course provide criteria
for grammatical units; like phonology, it must be stated separately and
then related to grammar, though if a one / one correspondence between
orthographic and grammatical categories is found to work at any point,
so much the better. The space in fact will not do in every case as
exponent of the boundary of the grammatical unit “word” in English.
Here, however, to treat (things like) run up as a single word, the only
purpose of which would be to maximalize the in any case very partial
correspondence between “word” and lexical item, considerably compli-
cates the description of the verbal group.) If the analysis is taken to the
morpheme, the lexical item is run up, which contains one morpheme
(but not the other) from the structure of the word ran plus another word
from the structure of the group – it is both more and less than a
grammatical word. Even if this type of morphemic analysis is rejected,
the grammatically discontinuous lexical item will appear in John was
running up a big bill, where the analysis of running as a two morpheme
word structure presumably would be accepted; and even where all
words have simple structure, as in John may run up a big bill, the lexical
item, though not discontinuous, is an exponent of no grammatical unit,
since it is more than a word but less than a group.
73. Cf. Chomsky (1957: 15 ff.).
74. For example by Chomsky (1957). Chomsky, however, does not 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 context
relates form to extratextual features, and is (like phonology) an interlevel,
the categories of context, like those of phonology, are not determined
by (still less, of course, do they determine) the categories of form; but
contextual statement is required to account for all (instances of the)
reflexion in form of extratextual features. For the statement of context,
as distinct from either “content” or “concept”, see especially Firth
(1957a: 5–13; 1957b: 227; 1957c: 2, 3; 1957d).
87
76. Cf. Firth (1957a: 11–13, 26–7, 31); Mitchell (1958: 108 ff.); Halliday
(1959: 156–75).
77. Analogous to the morphological grouping (the “paradigm”) of grammar
is the lexical “ordered series of words”: cf. Firth (1957b: 228) – though
Firth’s “ordered series of words” includes what I should consider a
“lexical set”, namely his “lexical groups by association”, these being (by
analogy) “syntactical”.
78. The traditional vehicle of lexical statement, the dictionary, states formal
meaning by citation and contextual meaning by definition. The 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 called
the “item and arrangement” model. It is not of course suggested that all
the points made in this section are applicable to all such studies, nor all
of them to any one study, within this type of linguistics. Since I have
been concerned to apply the present theory in the description of English,
many of the points made here were in fact first formulated with reference
to Hill (1958), which incorporates what is probably the best comprehen-
sive account of English grammar yet published and is an example of the
88
method here referred to. Hill’s book has recently been the subject of a
review article by Haas (1960); my present aim differs from Haas’ in that
I want to consider certain features of an approach in descriptive
linguistics, exemplified by Hill’s work but also by many other studies, in
the light of the theory here outlined.
82. Especially since a language may make systemic use of this variable (cf.
4.2 n. 37, above, where it is suggested that it is desirable to recognize in
English a grammatical system the exponent of which is precisely the
contrast between coextensiveness of the grammatical unit with one and
with (a sequence of) two exponents of a phonological unit).
83. Cf. above, 4.2 n. 37. Hill’s proportion, however, is
morph : morpheme : allomorph ::
sound : phoneme : allophone.
“Every morpheme must contain one phoneme and may contain several”
(p. 89) and, for English, “the occurrence of any juncture always marks
the boundary of an entity larger than a phoneme. The entity thus
bounded may be word, phrase or sentence, but must always be at least a
morph” (pp. 93–4). The phoneme thus enters into the statement both
of the unit “morpheme” and of its structure.
84. That is, the final step of the formal statement, the move to the formal
item. Strictly, since formal statement includes the placing of all forms at
all units, it would be more accurate to say “though it is by no means
always at morpheme rank that systems of formal items are to be found”.
For example in English the items when, because, if, in case, provided that,
etc., though they can of course be analysed into words and morphemes,
operate as items at the rank of the group and, as such, are members of a
particular class of the group.
85. So if we find in three languages items in substance statable phonetically
as [pata] and [pate], these may yield:
Substance Form
L1 [a] / [c] - (free variant)
L2 ’ singular / plural
(grammar)
L3 ’ ‘cat’ / ‘dog’ (lexis)
(Phonologically restricted variants may be of either type: the language
might have [pata] and [pete], never [pate] or [peta], without prejudice
to whether the contrast between [pata] and [pete] is formally meaningful
(L2,3) or not (L1).)
86. It does not matter, of course, what type of distinction is made in
substance: the pair above could equally be replaced by for example [pàt]
and [pát], or [’pata] and [pa’ta], etc.
87. Again it does not matter where: the following grammatical contrasts in
89
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 foot: shift of unit boundary (so
working re-distribution of strong
and weak syllables –
change of tonicity)
//1 he was / working // : // 4 tone replacement of one term
he was / working // group: in intonation system by
another
The phonological statement cannot necessarily be expected to cover the
systems concerned: it would be absurd, for example, to state the
affirmative: interrogative system in phonological terms – though, as said
below, it must carry the potentiality of being so stated, since it 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
앖
with the dotted line representing the (logically) final stage in which two
separate statements of abstraction are related. Cf. Robins (1959: 103):
“(grammatical distinctions are) not deducible from the phonetic shapes,
as such, of the words concerned nor from phonological rules based on
these shapes and the phonetic categories involved in them . . . While
both phonological and grammatical categories are abstraction from the
phonic material of utterance, their relation to the phonic material is
entirely different”.
89. For example the phonological system of intonation in English, operating
at the rank of the phonological unit “tone group”, carries a number of
different grammatical systems operating at different units in the
grammar.
90. Thus, the clause in English is defined by its operation in sentence
90
structure, and by its own classes and their structures. The difference
between two instances, such as an affirmative clause he saw them and an
interrogative clause did he see them?, is of course ultimately statable in
terms of substance; but grammar is not grammar if it tries to define the
class system in this way – or even to state the difference phonologically
at all. No linguist would ever try to state the grammar of clause–classes
by reference to phonology; yet the attempt to define the unit “clause”
by reference to phonological features such as juncture is no less 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) it
incorporates a rank scale, with a hierarchy of units to which contrasts
are assigned, and (ii) it is polysystemic, so that, for example, the /t/ in
10.3 n. 87, above is not “the phoneme /t/” of “the English language”
(no such entity will be postulated) but a phoneme identified by reference
to its place in the structure of the unit concerned; this would still be
true whether the phonological units concerned are (partly) taken over
from grammar or are set up independently in the phonology. Firth
stresses the very different nature of the “phoneme” in a description of
this type, and prefers to use the distinct term “phonematic unit”
(cf. 1957b, Chapter 9, passim).
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 be
recognized as the minimal element of grammatical structure; but this
does not imply that it is the most suitable element to bear the assignment
of all the grammatical functions fulfilled by the word into whose
composition it enters” (pp. 127–8). I would not follow Robins, 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
91
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 rank
transformation).
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 these
are likely in turn to account for only 80 per cent of instances (these
being now more delicately differentiated); these now demand a further
step in delicacy – and so on.
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 a
noun alone may carry four case / number distinctions, one with
adjective and noun six and one with deictic, adjective, and noun seven,
how can any two case / number forms be considered exponents of “the
same” category when they occur in different structures?
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
92
in Bantu “where there are repeated suffixes of agreement but in which

the members of the same sentence element are continuous”. Quite apart
from the arbitrary choice of assignment of redundancy, this is simply a
shift of criteria: Bantu concord is still the exponent of a relation (since
not all contiguous items are members of the same sentence element).
But even if concord and contiguity were completely mutually 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 mentioned
here is contextual redundancy. This is used, for example, by Bull (1960:
16): “Unless a language is needlessly redundant, there is little or no
likelihood that any tense system uses the point tensor formulas”; p. 27:
“English is extremely redundant. It almost always defines the axes while
Mandarin is extremely parsimonious. It defines the axes only to avoid
confusion”. In other words, the form is said to have reflected more of
the context than it need have done. This has the merit of having nothing
93
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).
94
Chapter Three
CLASS IN RELATION TO THE AXES OF

CHAIN AND CHOICE IN LANGUAGE (1963)
This paper is concerned with the nature of the class as a theoretical

term in Descriptive Linguistics: that is, with the meaning of “class” in
statements which are made in order to explain how a given language
works. Such statements are familiar in various forms, such as “there are
eight word–classes in this language: noun, verb, adverb . . .” and
“clauses can be classified into independent and dependent”.
I have assumed, for the purpose of the main points made in the
paper, that this category of “class” is to be defined syntactically. By this
I mean that the concept is introduced into the description of a language
in order to bring together those sets of items that have the same
potentiality of occurrence; in other words, sets of items which are alike
in the way they pattern in the structure of items of higher rank. Thus,
to take a typical instance from grammar, we may have morpheme
classes defined by word structure, each such class being one set of
morphemes having a given value in the structure of words: as, for
example, the morphemes of inflexion in Latin nouns. Likewise we
might have word–classes defined by group structure, or clause–classes
by sentence structure.
This use of the term “class”, to name a category defined in some
way by its relationship to a higher structure, is by no means universal
in linguistics; but it would probably be granted that some such category
is necessary to linguistic description whatever name we choose to adopt
for it. Syntactic classification (sometimes referred to as “functional
classification”, in what is perhaps a rather misleading opposition of
First published in Linguistics, 1963, 2, pp. 5–15.
95
“form” and “function”) is a central feature of linguistic method, and

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

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

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

theory. But when we say that sequence “may or may not be”
significant for structure, what we mean is that it may be significant at
varying degrees of delicacy, down to a point where a distinction
becomes so delicate that we do not know what to say about it; in such
cases we may have to be prepared to treat the particular feature of
sequence as being non-significant.
This has important implications for the category of “class”. For each
grammatical unit (i.e. sentence, clause and so on) in each language we
can recognize “primary” (least delicate) elements of structure; for the
clause in English, for example, the elements Subject, Predicator,
Complement and Adjunct. From these we derive our primary classes:
these are the sets of items of lower rank that enter into the primary
structures with the value of the elements concerned. Thus the class
corresponding to predicator is “verbal group”, that to both subject and
complement is “nominal group”, and that corresponding to adjunct is
“adverbial group”. Where the sets of items operating as two or more
elements of structure show more than a certain degree of overlap, as in
the case of subject and complement – most items that can be subject
can also be complement, and vice versa – these are conflated into a
single primary class: thus the “nominal group” is the primary class
expounding both Subject and Complement in English clause structure.
Primary classes are always chain classes. That is to say, the first step
(on the scale of delicacy) is to state the classes derived from the primary
elements of structure. When we take the class further in delicacy,
however, and recognize “secondary classes”, some of these more
delicate classes are chain classes and others are choice classes. It is clear
that primary classes cannot be choice classes, since we cannot account
for a choice until we have established that place in structure where the
choice is made: for example, the choice classes “singular nominal
group” and “plural nominal group” are meaningful only in relation to
the primary (chain) class “nominal group” which defines the context
of the choice.
For examples of secondary classes of both types we may take an
element of structure of the English nominal group, the element
“Determiner”. The class of word defined as operating with this value
is the class sometimes known as “deictic”. [Note: Halliday later reversed
these terms, using Deictic as the function (element of structure) and
determiner as the class. (Ed.)] This includes some forty items that are
fully grammatical (i.e. reducible to one-member classes by successive
steps in delicacy: the this that which whose my its a any some either each
99
other same certain, etc.), together with other items forming an open set
(i.e. that cannot be so reduced: John’s, etc., including compound ones
as in the railway company’s property).
This class of deictic may be variously subdivided along both axes. On
the one hand, there are certain sets whose members can occur in
combination, as in all my other friends; there are in fact three such
secondary groupings, the members occurring respectively in first, second
and third place in a maximum sequence. This gives three secondary
chain classes which may be called “predeictic” (for example all), “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 principal
dimensions. The deictic, for example, may be “specific” / “non-
specific” (my / every); “selective” / “non-selective” (my / the); and, as a
further subdivision of the class formed by the intersection of “specific”
and “selective”, it may be “possessive” / “demonstrative” (my / this).
These and various other systems eventually yield, by their subdivisions
and intersections, one-member classes: thus my can be uniquely classified
as “deictic: specific, selective, possessive: personal: first person”.
Secondary classes regularly cut across each other. The systems of
“specification” and “selection”, for example, form a matrix as follows:
Specific Non-specific
this / these which both all
that / those what every each
no neither
Selective my your whose
our
their
his / her
its
John’s (etc.)
the a
Non-selective some any
either another
It is not uncommon to find a large number of such intersecting

classes, which may be very difficult to sort out; the above is only one
of many possible ways of approaching the classification of the English
100
deictics. But the patterns they display are typical in their complexity: a
given class breaks down by simple subdivision into a system of more
delicate classes, but the same original class will also subdivide in a
number of different ways, so that many dimensions of classification
intersect with one another. Any given item, to be fully identified, may
require to be simultaneously classified on all such dimensions. In this
way it can be assigned to a “microclass”, this representing its value in
respect of all the properties which have been found relevant to the way
it patterns in the language. There will be, of course, a very large
number of such microclasses: for example, in a computational study of
English “phrasal verbs” (items like take up, put on) which is being
carried out at the moment, 557 such items were found to yield 125
microclasses.
Up to this point I have been concerned only with “place-ordered”
structures. These are sometimes thought of as being the normal type of
linguistic structure. By a place-ordered structure I mean one composed
of a limited number of different elements occurring nonrecursively.
Such a structure may be fully class-defining, in the sense that to each
element corresponds a distinct class of lower rank: for example the
clause structure “Subject + Predicator”, with classes respectively “nom-
inal group” and “verbal group”, as in “my friends have arrived”; or it
may be only partially class-defining, where two or more elements are
expounded by the same class but differentiated in sequence. In this
type of structure, there is no constant relation between successive (or
otherwise paired) elements: for example, in the structure ‘Subject +
Predicator + Complement’ (for example John saw Mary, my friends have
invited me) it is not true that Subject is to Predicator as Predicator is to
Complement.
This is not the only type of structure found in language, and there
seems no particular reason for assuming it to be the norm, especially in
its pure form. Language also exhibits a different kind of structure, the
“recursive” or “depth-ordered” structure. Here, as the name implies,
an element of structure, or a combination of elements, is repeated “in
depth”, a series of such elements (or combinations) thus forming a
progression. It is doubtful whether one should set a theoretical limit to
the degree of depth in recursion; rather there appears to be some
logarithmic scale of diminishing frequency, so that the number of
observations one would expect to have to make before recording a
depth of, say, ten would be extremely high. Spoken English seems to
tolerate greater depth in recursion, or at least to tolerate it more readily,
than written English; and this may be true of language generally. The
101
following are some examples of recursive structures in English, with

depth indicated by the letters of the Greek alphabet.
1) Sentence structure:
he might have come if you had told him when you rang him up
a b g
while he was packing before he went away
d e
2) Clause structure (P = Predicator; C = Complement):
she made them 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 doing
present 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 may
be distinguished by class (as in example 1, above, where the alpha-
element is expounded by the class “independent clause”), each element
of structure represented by a term in the recursive series has as exponent
one and the same class. In example 1, the class “dependent clause”
operates at beta, gamma, delta and epsilon: each item could occur at
every place. (The items themselves are of course not necessarily
mutually substitutable in every instance, since there may be more
102
delicate class restrictions such as are shown by the different forms of

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

which cuts across the scale of rank. That is to say: in non-rankshifted
structures, whether recursive or not, classes of each rank enter into a
structure of the rank immediately above: in English, morpheme classes
in word structure, word–classes in group structure, group classes in
clause structure and clause–classes in sentence structure.6 In rankshift,
this relation is broken and the classes enter into a structure of their own
rank or even of lower rank than themselves, as in examples 5 and 6:
the first shows classes of the group in group structure and the second
shows a class of the clause, in this case the relative clause, also in group
structure. Sometimes, as in the case of the relative clause, there may be
a distinct class which occurs only with rankshifted status; but this is not
necessarily so. In example 5, the adverbial group in the garden is
rankshifted to the status of Qualifier in the nominal group the peartree
103
in the garden; but this class can also operate as Adjunct in the clause,
being thus not rankshifted, as in he sat in the garden. Where a class may
occur either rankshifted or not, often ambiguity may arise, as in take
that chair in the garden: is the chair already in the garden (= ‘take that
chair which is in the garden’; rankshift) or not (= ‘take that chair and
put it in the garden’; no rankshift)?
Some of the most complex problems in the description of a
language arise where the same structure combines place-ordered with
recursive elements. Perhaps the most striking example of this in Eng-
lish is the nominal group, which is a troublesome mixture of the two
types; the earlier elements are largely place-ordered, recursive elements
being increasingly tolerated as one approaches the Head (the head
being, e.g., houses in the two old stone houses by the river) and continuing,
by rankshift, thereafter. The element immediately following the
Determiner, which may be called Ordinative and defines the classes
of cardinal and ordinal numeral, together with superlatives, is already
marginally recursive (e.g. the first three second best hotels); with later
modifying elements preceding the Head this potentiality is greatly
increased, giving items such as example 4 above and familiar also in
the language of headlines: holiday coach death crash inquiry verdict. More-
over, the linear succession of the items does not act as a constant in
showing the depth relation: compare 5-millimetre perspex boxes (gamma
beta alpha) with 6-inch perspex boxes (beta beta alpha). The description
of the word–classes entering into the structure of the English nominal
group is extremely complicated if one treats it as a simple place-
ordered structure, with classes defined for each possible position, as
the various attempts to do so have already shown. On the other hand
it is unsatisfactory to treat the whole thing as recursive in structure
and to recognize no secondary classes beyond the primary word–class
“noun”. The facts of the language here lie in between the extremes
of these two types of structure, and the best description seems to be
one which takes this into account.
These seem to me to be some of the problems that arise in intra-
linguistic classification at the level of grammar. Such problems are
probably most acute at this level, but similar ones arise also at other
levels, notably phonology, which has other additional classification
problems of its own. Lexical classification is rather a different matter,
and there are reasons for preferring not to use the term “class” in
talking about lexis; but this subject would require a separate paper. The
field of classification as a whole is one where linguists can learn much
from disciplines faced with similar or related problems. At the same
104
time perhaps the experience gained within linguistics may not be

wholly irrelevant to other sciences.
Notes
1. I am assuming here the more “abstract” view of grammatical categories
such as morpheme; cf. Palmer (1964b: 232–7).
2. For interdisciplinary purposes I have used here the terms “chain axis” and
“choice axis” in place of their less self-explanatory technical equivalents
“syntagmatic axis” and “paradigmatic axis”.
3. Strictly speaking the ultimate exponent is a token (“occurrence”) of the
type (“formal item”) fetch. If, however, we confine ourselves to the level
of grammar we can regard the formal item as the ultimate exponent.
4. Another way of drawing the same distinction between grammar and lexis
is to say that grammar is “deterministic” by contrast with lexis which is
“probabilistic”; in the sense that in grammar one can distinguish what is
possible from what is impossible (before assigning probabilities, if one
wishes, to what is possible), whereas in lexis one can only distinguish
between what is more and what is less probable.
5. In fact the verse does contain rankshifted nominal groups, as “prepositional
complement” in adverbial groups (for example that lay in the house), but
these have been ignored in the illustration for the sake of simplicity.
6. This restriction to “the rank immediately above” implies a particular
model of grammar more specifically than do most of the other points made
here. A more generally valid formulation would be “enter into a structure
of higher rank”.
105
Chapter Four
SOME NOTES ON ‘DEEP’ GRAMMAR (1966)
In the representation of syntagmatic relations in language, we may

distinguish between a linear sequence of classes, such as “adjective
followed by noun”, and a non-linear configuration of functions, such
as “modifier-head relation” or simply “modification”.1 Both of these
have been referred to as “structure”, although this term has also been
extended to cover paradigmatic as well as syntagmatic relations. For
Hjelmslev, for whom “structure” was not a technical term (see e.g.
1961: 74), “the structural approach to language . . . [is] conceived as a
purely relational approach to the language pattern” (1948: quoted in
Firth 1951: 220); among others who have emphasized the relational
aspect of such studies are Firth2 (1951: 227–8; 1957: 17ff.; cf. Robins
1953; Palmer 1964a), Tesnière (cf. Robins 1961: 81ff.) and Pike (cf.
Longacre 1964: 16). Chomsky’s (1964: 32) distinction, using Hockett’s
terms, between “surface structure” and “deep structure”, “structure”
here going beyond syntagmatic relations, is extremely valuable and
widely accepted: the surface structure of a sentence is defined as “a
proper bracketing of the linear, temporally given sequence of elements,
with the paired brackets labelled by category names”, while the deep
structure, which is “in general not identical with its surface structure”,
is “a much more abstract representation of grammatical relations and
syntactic organization”.
A representation involving the concepts of class and sequence may
thus be said to be a representation of surface structure. Here the
ordering, if each pair of brackets is said to enclose an “ordered set” of
First presented at a meeting of the Linguistics Association, Newcastle, March 1965. First
published in Journal of Linguistics, 1966, 2(1), pp. 57–67.
106
some notes on ‘deep’ grammar
classes, is interpreted in the usual sense of the word, as linear successivity,

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

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

form of “realization” relation,4 is relevant to the understanding of
syntagmatic patterns, and the distinction can be made and discussed
solely in terms of relations on the syntagmatic axis. Clearly, however,
it is relevant also to relations on the paradigmatic axis. It may be helpful
to relate this point to the distinctions made by Hjelmslev and by Firth.
In Hjelmslev’s terms (1961: 38–9), linguistic function embraces both
relation and correlation; relation is syntagmatic, within the semiotic
process, or the text, while correlation is paradigmatic, within the
semiotic system, or the language. While his view of the relation
between the two axes was somewhat different from that of Hjelmslev,
Firth likewise makes a terminological distinction, referring (1957: 17)
to syntagmatic relations as relations of structure and to paradigmatic
relations as relations of system.
Provided there is at least some syntactic element in the definition of
the classes concerned, even a syntagmic, class-sequence representation
already gives some information about the paradigmatic relations into
which an item enters, but to a very limited extent. This limitation is
again inherent in the nature of the class–sequence concept: paradig-
matic “relatedness” depends on a functionally defined environment.
Two entities can only be said to contrast if they have a functional
environment in common,5 and this environment is generally specified
in terms of syntagmatic function; it presupposes therefore a representa-
tion of structure – i.e., of “deep” syntagmatic relations. The structural
representation thus specifies the environment both for sets of paradig-
matic relations and for further networks of syntagmatic relations, those
within the lower-order constituents: for example the function of
“subject”, itself specified syntagmatically in clause structure, defines an
environment both for the syntagmatic relation of modifier–head and
for the paradigmatic relation of singular / plural. For paradigmatic
relations in the highest unit there is no functionally defined environ-
ment in this sense, so that if the sentence is said to be either declarative
or interrogative this either / or relation rests on other grounds: the
sentence as a primitive term, or some postulated higher unit not yet
structurally described; the appeal to contextually-defined sentence
functions such as “statement” and “question” is not one of these, this
being rather a way of saying that declarative and interrogative have no
environment in common.
The paradigmatic contrasts associated with a given, defined environ-
ment may be thought of as being accounted for either in a single
representation of “deep” grammar, in which are incorporated both
109
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 glossed
informally as a “deep paradigm”, a paradigm dependent on functional
environment; in a sense, and mutatis mutandis, the relation of system to
paradigm is analogous to that of structure to syntagm as these terms
were used above. One could think of “paradigmeme” as a possible
tagmemic term. In Hjelmslev the “system”, likewise a paradigmatic
concept, is defined as a “correlational hierarchy”, the underlying notion
being that of commutation (Hjelmslev 1961: 73–4). A system is thus a
representation of relations on the paradigmatic axis, a set of features
contrastive in a given environment. Function in the system is defined
by the total configuration: for example “past” by reference to “present”
and “future” in a three-term tense system, as structural function is
defined by reference to the total structural configuration for example
“modifier” by reference to “head”.
If paradigmatic relations are represented separately in this way, this
implies that the full grammatical description of a linguistic item should
contain both a structural and a systemic component. It may be useful
therefore to consider the notion of a systemic description as one form
of representation of a linguistic item, the assumption being that it
complements but does not replace its structural description. The
systemic description would be a representation of the item in terms of
a set of features, each feature being in contrast with a stated set of one
or more other features: being, in Firth’s terms, a “term in a system”.
This is exactly the sort of characterization that has been familiar for a
long time in the form of “this clause is interrogative, finite, present
tense, . . .”, given that we are told somewhere in the grammar not
merely what other tenses, moods, etc. are found in the language but
also which of them could have occurred in this particular clause all
other features being kept constant.
There is, however, one modification of a traditional “systemic
description” of this kind which may need to be considered. This
concerns the ordering of the features listed. In the traditional version
they are unordered; but if the grammar specifies not only relevant
systems but also their interrelations with one another, in particular their
hierarchization on what I have called elsewhere (Halliday 1961: 272;
1964: 18) the “scale of delicacy”, then partial ordering is introduced.
Any pair of systems, such that a feature in one may co-occur with a
feature in the other in a systemic description, may be hierarchical or
110
simultaneous; if two systems are hierarchically ordered, features assigned

to these systems are ordered likewise. So for example the system whose
terms are declarative / interrogative would be hierarchically ordered
with respect to the system indicative / imperative, in that selection of
either of the features declarative and interrogative implies selection of
indicative. If this is taken together with another system, unpredicated
theme / predicated theme, likewise dependent on indicative, then the
item John has seen the play may be represented in respect of these
features as:
(indicative : (declarative / unpredicated theme))
where : indicates hierarchy and / simultaneity. Then it’s John who has
seen the play contrasts with it in respect of one feature:
(indicative : (declarative / predicated theme))
and is it John who has seen the play in respect of two features:
(indicative : (interrogative / predicated theme))
The systemic description would represent a selection from among the
possibilities recognized by the grammar. As far as these examples are
concerned, the grammar would show that in a given environment
selection is made between indicative and imperative; and if indicative
is selected, there is also simultaneous selection between declarative and
interrogative and between unpredicated theme and predicated theme,
the two latter selections being independent of one another. Any item
thus contrasts with others in respect of such features and combinations
of features as the ordering of the system permits.
For any set of systems associated with a given environment it is
possible to construct a system network in which each system, other
than those simultaneous at the point of origin, is hierarchically ordered
with respect to at least one other system. The point of origin is
specified syntagmatically, so that all features are associated with a
syntagmatic environment; at the same time the system network 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, the
one concerned with paradigmatic and the other with syntagmatic
relations. On the other hand it might be useful to consider some
possible consequences of regarding systemic description as the under-
111
lying form of representation, if it turned out that the structural

description could be shown to be derivable from it. In that case
structure would be fully predictable, and the form of a structural
representation could be considered in the light of this. It goes without
saying that the concept of an explicit grammar implied by this formu-
lation derives primarily from the work of Chomsky, and that steps
taken in this direction on the basis of any grammatical notions are
made possible by his fundamental contribution. My own more specific
debt here is to Lamb, whose formalization of stratification theory is
based on general notions closely akin to those which I had adopted (cf.
Hockett 1965: 198). The present paper, however, attempts no more
than an informal discussion of the question of a grammatical description
in terms of features, here based on the notion of a feature as one of a
set of contrastive “terms in system”.
Presenting the systemic description of a linguistic item as the
underlying grammatical representation of that item would seem to
imply that its paradigmatic relation to other items of the language was
in some way its more fundamental property, from which its internal
(syntagmatic) structure is considered to be derived. This would seem
to be Hjelmslev’s view, in his discussion of system and process (1961:
39–40). But the priority which is implied is not one between paradig-
matic and syntagmatic relations as such, but rather between the external
relations, both paradigmatic and syntagmatic, into which an item enters
(the point of origin of a system network being defined in syntagmatic
function) and its internal relations of structure. If one talks of simplicity,
this means the simplicity of the whole description; underlying grammar
is semantically significant grammar, whether the semantics is regarded,
with Lamb, as input or, with Chomsky, as interpretation. What is
being considered therefore is that that part of the grammar which is as
it were “closest to” the semantics may be represented in terms of
systemic features. This would provide a paradigmatic environment for
the “relatedness” of linguistic items, a contrast being seen as operating
in the environment of other contrasts. Structure would then appear as
the realization of complexes of systemic features, involving in places
both neutralization and diversification as defined in Lamb’s terms
(Lamb 1964a: 64).
If the structural representation is not required to account for paradig-
matic relations, the question of how “deep” it needs to be is determin-
able by reference to other considerations: it should give an adequate
account of syntagmatic relations, and permit the explicit realization of
the systemic description in terms, ultimately, of a sequence of classes.
112
This may be illustrated from the example it’s John who has seen the play.
Leaving aside variation that is immaterial to the discussion, there would
seem to be three possible representations of its structure:
(1) it Subject, is Predicator, John who has seen the play Complement
(2) it . . . who has seen the play Subject, is Predicator, John
Complement
(3) it’s John who Subject, has seen Predicator, the play Complement
(1) would presumably be an attempt merely to state the simplest
sequence of classes in the syntagm, although it could be shown to be
unsatisfactory even on class-distributional grounds. (2) is distributionally
acceptable and would account adequately for the syntagmatic relations;
but it fails to account for the paradigmatic relations in that it does not
show the “relatedness” of this clause to John has seen the play, etc. If the
structural description is required to show the paradigmatic as well as
the syntagmatic relations of the grammar we need some representation
such as (3) in which John is the Subject. This leads to complexity in
the realization, since a nominalization of the form it’s John who seems
to add no new insight elsewhere in the grammar. A more serious
difficulty arises in relation to the element “Subject” in English, which
is a complex element within which it is possible to distinguish three
components, or features; each of these may contrast independently of
the other two, although there is a general, and generalizable, tendency
to co-variation among them.
The three contrasts can be seen independently in (i) John has seen the
play, with tonic on play, versus, respectively, (ii) the play has been seen by
John (Subject as actor versus Subject as goal); (iii) the play John has seen (=
“the play, John has seen, but . . .”, Subject as theme versus Subject
nonthematic); (iv) John has seen the play (with John tonic; Subject as
“given” versus Subject as “new”). Each of these three is related paradig-
matically to the original item, and each of them contrasts with it in
respect of one feature only. By a further contrast, that of “unpredicated
theme” versus “predicated theme”, (iv) is related to (v) it’s John who has
seen the play with John tonic.6 Thus, despite the difference in constituent
structure, (v) differs from (iv) in respect of only one feature. Such
patterns, where different complexes of (paradigmatic) features may be
combined in what is syntagmatically one and the same element of
structure, here the Subject, involve some complexity for a structural
description; if they were handled in systemic terms, the structure need
represent only their realization in syntagmatic relations. We could then
adopt a form of structural representation such as (2) above.
113
The examples cited might be regarded as irrelevant on the grounds

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

constituent structure is adopted for the statement of paradigmatic
relations, and is then made to determine the constituent structure, then
provided the structural description adequately handles the syntagmatic
relations there is no need for everything to be accounted for at a
constituent stage of representation. This is most obviously relevant to
phonological features of the prosodic type, but could be extended also
to items identified as being markers of, rather than elements in,
syntactic relations.
The crucial factor in the designation of any feature as present in the
grammar would thus be its assignment to a place in the systemic
network. A putative feature which could not be shown to contrast
independently with one or more others at some point would not be a
distinct feature; each feature that is recognized is thus a term in a
system, which system is located in hierarchical and simultaneous
relation to other systems. The location is “polysystemic”: the recog-
nition of a system, and the assignment of a feature to it, depends on
the potentiality of contrast in the stated environment. For example,
there might seem to be a proportionality in English such that he can go
is to can he go as he is wondering if he can go is to he is wondering can he go;
but the related features are different in the two cases: that is, the two
syntagmatic environments determine different sets of paradigmatic
relations. The ordering of the systems in delicacy would thus be
important in the identification of the systemic features.
It would be necessary also to specify the syntagmatic environment,
in order to define the point of origin of a system network. This can be
done in terms of the notion of rank, where the initial identification
and labelling of certain stages in a constituent hierarchy in such general
terms would provide a starting point for the delimitation of different
more specific environments. The designation of rank, in other words,
is a possible first step in the specification of what Haas (1966: 125) calls
“functional relations”, relevant here in that it makes possible the
assignment of a system to a place determined solely by constituent
status (for example all clauses) and allows further specification of the
environment to be in terms of features: a feature x may be associated
with constituents having the features y and z rather than with constitu-
ents having a given syntagmatic function. The possibility of contrast
between active and passive in the clause depends on other features of
the clause, not on its function in the sentence.
In stratificational terms, rank defines an inner series of strata, or sub-
strata, within the outer grammatical stratum, with each rank character-
115
ized by a different network of systems. While many, though by no

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

with its deeper patterns of organization.
Notes
1. This paper was first presented at a meeting of the Linguistics Association,
Newcastle, March 1965. I am grateful to R. M. W. Dixon and R. D.
Huddleston for their subsequent valuable comments and suggestions.
2. Hass (1966: 131) points out that Firth’s position here has been 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” precisely
for this non-linear relation among the components (“elements”) of a
structure, contrasting it with “sequence”.
4. I use Lamb’s term “realization” instead of the earlier “exponence”. Lamb’s
term is more widely known; it also corresponds closely to my own use,
whereas as Palmer (1964b) pointed out my use of “exponence” differed
materially from that of Firth.
5. This may be interpreted either as “if there is at least one set of conditions
under which both could occur” or as “if both could occur under the given
set of conditions”. It is the latter interpretation which I take to be the basis
of (one aspect of) Firth’s ‘polysystemic’ approach. Firth himself was
inconsistent in referring to a ‘system’ of word–classes noun, verb, etc.
6. Since the subject normally has the feature “given”, that of “new” being
realized in other elements, the realization of the feature “new” in the
subject is often accompanied by its predication as in it’s John who has seen
the play. This explains why it’s John who has seen the play usually, though
not obligatorily, has John and not play tonic, while the opposite is true of
John has seen the play.
117
Chapter Five
THE CONCEPT OF RANK: A REPLY (1966)
I am most grateful to Peter Matthews (who I am pleased to note seems

to deplore, as I do, “missionary fervour” in linguistics1) for allowing
me the opportunity of replying in the same issue to his criticisms of the
concept of “rank” and to the various other more or less related
criticisms which he has incorporated into his discussion (1966).
Having embarked on a detailed reply, I soon discovered that it
would take up far too much space: for example, Matthews’ strictures
on my use of terminology demanded, besides a lengthy documentation
of my own care in this respect and of the reasons for particular choices,
a demonstration, with citations, that my own practice in regard to
terminology is in no way different from that of other linguists. Yet
Matthews is well aware that I could cite numerous examples of the
redefinition of traditional terms, especially polysemic ones like “word”,
many of them much more at variance with earlier usage than mine –
and presumably therefore much more likely to “mislead” the “public”.
Such tu quoque arguments seem to me trivial; and yet how else can one
answer a charge such as ‘If Halliday does not mean by “word”, in
particular, what the ordinary linguist on the Clapham omnibus means
by “word”, why has he never said so?’ other than by pointing out,
what is surely obvious, that linguists do use terms in different ways and
that apparently I have said so, since that is Matthews’ starting point?
Such strictures seem designed merely to defy comment.
A discussion of rank does, however, demand detailed consideration.
The main issue, that of total accountability, has been a subject of
discussion over many years; if Matthews is here covering old ground,
First published in the Journal of Linguistics, 1966, 2(1), pp. 110–18.
118
the concept of rank: a reply
that is my fault rather than his. I plead guilty unreservedly to the charge
of inadequate documentation of my own work, although the relevant
bibliography of others’ writings is rather longer than Matthews seems
to suggest.
Some of the points raised in my paper, ‘Some notes on “deep”
grammar’ (see above, Chapter 4), are I think relevant to Matthews’
comments; more important in this connection is Huddleston (1965),
written without knowledge of Matthews’ article. Here I shall try to
take up with reasonable brevity some of the issues as Matthews himself
sees them.
Two assertions seem crucial to Matthews’ argument. (1) A rank-free
constituency grammar is to be preferred to a rank constituency gram-
mar. (2) All formulations of rank grammars are either theoretically
insignificant or empirically unsound: unsound if they differ materially
from rank-free grammars and insignificant if they do not. (1) follows
from (2), except that, if they do not differ materially, this by itself gives
us no reason for preferring either. Let us examine each of these in turn.
By a rank grammar I mean one which specifies and labels a fixed
number of layers in the hierarchy of constituents, such that any
constituent, and any constitute, can be assigned to one or other of the
specified layers, or ranks. The European linguistic tradition, by its use
of terms such as “sentence” and “clause”, has always implied the
possibility of such a grammar, although, as is well known, the absence
of criteria regulating the necessary modifications of the simple con-
stituency relation led to various difficulties. I return below to the
concept of rank as suggested in such formulations as “clause used as a
word” and the like.
A rank grammar is, as Matthews observes, a hypothesis about the
nature of language. This leads us to ask, first, whether it can be falsified,
and second, whether it is worth making in the first place. Like many
other hypotheses, both in linguistics and in other disciplines, its
empirical falsification, given that it cannot be logically disproved, is
unlikely to take the form of the discovery of a clear counter-example
– in this case, of a language which it is impossible to describe in rank
constituent terms. It would be likely to take the form rather of
demonstrating that the hypothesis prompts no interesting new questions
and leads to unnecessarily complex or otherwise unsatisfactory accounts
of the facts. Neither this limitation on the conditions of its falsification
nor the fact that disagreements may arise (as in all subjects) over
whether a given hypothesis has in fact been shown to be inadequate of
themselves deprive it of interest. It seems to me that this is a hypothesis
119
which is capable of being falsified in these terms – that there are, in

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

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

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

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

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

intuitive adequacy, but at the cost of some surface complexity; this is
something to be avoided if possible, and Huddleston (1965) has
introduced an important modification into the definition of rank which
goes a long way in this direction.
Given an adequate representation of the underlying grammar, there
is no need to insist that every element should be assigned constituent
status at all; it is quite usual not to recognize intonation features as
constituents, and the same considerations could apply, as Matthews
points out, provided limitations were stated, to markers such as and and
or. I do not know how to specify in a general formation the conditions
under which accountability in constituent terms would not be required.
But this problem is no more difficult for a rank grammar, which has at
least the concept of total accountability to refer to as a point of
departure, than for a rank-free grammar, which has not. Matthews’
proposal (p. 104), that “Any generalizations we want to make can be
made, once and for all, by specifying a single element of structure at
which all the relevant classes or systems may operate: it does not matter
if some of these classes are classes of phrases, words or morphemes
rather than clauses”, is not a solution, since in many cases it does
matter; otherwise no constituency grammar would have introduced
singulary branching. Matthews has shown no convincing reasons for
abandoning the generalizations which the notion of rank facilitates; to
the extent that his objections are purely terminological they could of
course be accommodated if one knew what they were.
Even if all singulary branching was excluded, while this would be
failing to exploit the concept of rank it would not render it insignifi-
cant. It would remain a form of labelling of greater generality than
structure labelling; and it would embody the hypothesis that there are
a finite number of points of origin of distinct systemic choices in
language and a finite number of layers of constituent structure at which
distinct types of syntagmatic relation are contracted.
There remains the question of the king of England’s hat, where
Matthews has invented a problem on my behalf by insisting that either
the king of England or ’s must be a word. In fact no such requirement is
implied. I am not sure whether Matthews wishes to suggest, as at one
point he appears to do, that there is no problem here except the one he
ascribes to me; to many linguists it has seemed that there is. It is beyond
the scope of a discussion of this nature to attempt to do it justice, and I
do not want to argue here in favour of any one approach; but a brief
comment on Matthews’ own observations seems to be called for.
125
It is perfectly true that the fact that something is “in some sense” a
word does not make it an immediate constituent of anything. But it is
surprising to be told that ‘there is no argument for treating the king of
England’s . . . as (a word) . . . which does not simply take as a premise
the requirement that “each unit should be fully identifiable in 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 rested
on such grounds; in any case we may reasonably ask for more than the
mere reiteration that it “is” a phrase. Many questions need to be raised:
what, for example, are the implications for constituent structure of the
many cases where ’s and of are not parallel, for example that hat is the
king of England’s but not this hat is of the king of England? How does the
grammar take account of the fact that the placement of the ’s is
paralleled by the placement of the tonic (in the king of England) which
likewise depends on location in the syntagm and not on the constituent
structure? and so on. If Matthews is suggesting that the problem is
created, or even that it is complicated, by the introduction of rank into
the grammar, it seems to me that he has given no evidence of this.
Here also the issue seems to be partly one of labelling. This does not
make it any the less appropriate for discussion; but whatever the
usefulness, or otherwise, of the concept of rank in this and other
contexts, it is surely something that can be discussed in its own
linguistic terms.
Notes
1. In this connection I find the concept of ‘a “neo-Firthian” ’, especially one
“committed” to certain “statements”, rather extraordinary. Must we all be
labelled in this way? There is an important principle at stake here: that a
scholar is responsible for what he says and writes, not for what others say
and write. If I express agreement with something another linguist has put
forward this neither makes that linguist responsible for my views nor
commits me to acceptance of the whole of his. I may be wrong, but I
feel that there are undesirable limitations on this principle inherent in
Matthews’ first two paragraphs.
2. I am sorry if my own formulation was unclear here. But in that case the
whole of this part of Matthews’ argument rests on one purely 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.
126
APPENDIX TO SECTION ONE
TABLE 1
Tonality: Distribution of utterance into tone groups (location of tone
group boundaries)
Tonicity: Distribution of tone group into tonic and pretonic (location
of tonic foot)
Tone (primary; pitch movement on tonic)
Tone (secondary)
This paper was written between May and August 1964 and formed the substance of a
course on the description of English at the University of Indiana. First published in System
and Function in Language, 1976, edited by G. R. Kress, London: Oxford University Press,
pp. 101–35.
127
Symbols
// tone group boundary (always — tonic syllable
also foot boundary) . silent ictus
/ foot boundary . . . pause
Tone, primary and secondary, is shown by Arabic figures, alone or
with diacritics, placed immediately after the tone group boundary
marker.
128
appendix
TABLE 2
Tonicity: location of ‘information focus’
tonic = final lexical item (neutral)
tonic = pre-final item or final grammatical item (contrastive)
Tone (assuming tonality neutral):
Place of clause in sentence structure: final main 1, non-final
coordinate 3, non-final subordinate 4
Declarative clauses
reservation: 1 unreserved, 4 reserved
involvement: 1 neutral, 3 uninvolved, 5 involved
agreement: 1 neutral, 3 confirmatory, 2 contradictory
information: 1 one information point, 13 two information points
‘key’: 1 neutral, 1 + strong, 1 – mild
Interrogative clauses, WH-type
‘key’: 1 neutral, 2 (with final tonic) mild
relation to previous utterance: 1 unrelated, 2 (with WH-tonic) echo
Interrogative clauses, yes/no type
‘key’: 2 neutral, 1 strong
involvement: 2 neutral, 3 uninvolved, 5 involved
place in alternative question: 2 first alternative, 1 second alternative
specification of point of query: 2 unspecified, 2 specified
Imperative clauses
‘key’ (positive): 1 neutral, 3 moderate, 13 mild
‘key’ (negative): 3 neutral, 1 strong, 13 mild
force: 1 neutral, 4 compromising, 5 insistent
function: 1 etc. command, 2 question
‘Moodless’ clauses (also as declarative)
function: 1 answer etc., 2 question, 3 warning, 5 exclamation
In the section under ‘tone’, the headings (e.g. ‘reservation’) indicate
the nature of the choice, the entries under each heading representing
the terms in the choice with their appropriate tone. Thus ‘1 unreserved,
4 reserved’ means ‘in this choice tone 4 indicates reservation, by
contrast with one 1 which indicates no reservation’. Secondary tones
are indicated where relevant.
129
1 Systems of tone
appendix
2 Systems of the clause

2.1 Dependent clause
131
2.2 Declarative clause
132
appendix
2.3 Interrogative 1
133
2.4 Interrogative 2
134
appendix
2.5 Imperative
135
2.6 Transitivity
extensive: (effective/operative: (goal-intransitive))

John threw; Mary washed (sc. the clothes)
extensive: (effective/operative: (goal-transitive: non-benefactive))
John threw the ball; Mary washed the clothes
extensive: (effective/operative: (goal-transitive: benefactive))
John gave the dog a bone; Mary washed the boys their clothes
extensive: (effective/middle)
Mary washed (sc. herself)
extensive: (effective/receptive: (agent-oriented: non-benefactive))
the ball was thrown; the clothes were washed
extensive: (effective/receptive: (agent-oriented: benefactive: (goal-receptive)))
the bone was given the dog
extensive: (effective/receptive: (agent-oriented: benefactive: (beneficiary-receptive)))
the dog was given the bone
extensive: (effective/receptive: (process-oriented))
the books sold; the clothes washed
extensive: (effective/operative)
the sergeant marched the prisoners
extensive: (descriptive/middle: (+ range))
Peter jumped the wall
extensive: (descriptive/middle: (ⳮ range))
Peter jumped; the prisoners marched
extensive: (descriptive/receptive)
the prisoners were marched
intensive: non-benefactive
Mary seemed happy; Mary made a good wife
intensive: benefactive
Mary made John a good wife
136
appendix
2.7 Theme
Examples of tone-expounded systems in declarative clause (from 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/normously / long //
// 1 – . I’m / not / sure that it’s / worth it //
// 1 . and there was a / photograph of a / rabbit // I quivering his / ears a //
1 – lovely / white / rabbit and //1 – he was a / little man / sitting behind
the / rabbit //
//4 . oh / mine / aren’t parasites //
//4 no worse than / anyone / else //
//3 six / foot //3 I don’t know //
//3 . they / just find a / comfortable / place in your / gut and they //3 stick
their / hooks in and //1 stay there //
/5 he was a / very / famous / man //
137
//5 . it’s / very / interesting //

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

Key to examples
Examples
1 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 -
138
appendix
//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
Examples
1 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
139
3 Imperative systems: jussive

Key to examples
Examples
1 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
140
appendix
4 Theme systems: examples

Key to examples
Examples
1 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 spontaneous
conversation)
in the first month one was too ill to move
adjudicator I thought they were called
the sound that went floating out on the air I didn’t know I had it in me
aged legal gentlemen all like pipes
the metal container somehow it turns your coffee rather sour
141
Britain it’s all roads

it’s inhaling that’s harmful
it’s the side that has possession is at an advantage
it was that part I didn’t enjoy
it’s rather good coffee this
it was quite fascinating to see her
it does interest me how memory works
imagining some suffering is worse than experiencing it oneself
//this of course de/pends on the / country where they / live //
// . I / thought / cats always / ate them //
// how / long do these / changes / take //
// . that’s why it’s so / awful to / have to get / rid of it //
// . it / looked rather / odd having those / needles //
// . no / I saw the / first one //
// . but / in A/merica they //layer / things //
// all the / dialect forms are //marked / wrong //
Paradigm examples corresponding to those on preceding page
yesterday I saw John
John I saw yesterday
John I saw him yesterday
I saw John yesterday
John yesterday he saw me
John he saw me yesterday
it was John that saw me yesterday
it was John saw me yesterday
it was John I saw yesterday
he saw me yesterday John
it was strange to see John
it was strange how I saw John
seeing John was strange
//. I / saw / John / yesterday //
// I saw / John / yesterday //
//. I / saw / John //
//. it was / strange to / see / John //
//. it was / strange / seeing / John //
// I saw / John / yesterday //
// yesterday I // saw / John //
// I saw / John // yesterday //
142
appendix
5 Systems of the verbal group 1
6 Systems of the verbal group 2
143
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 6
present in past 7
present 8
future 9
future in past 10
present 11
future 12
past in future in past 13
144
appendix
present 14
future 15
present in past in past 16
present 17
future 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 25
present 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 33
present 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 taking
9 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
145
16 had been taking

17 has been taking
18 will have been taking
19 was going to be taking
20 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 taken
28 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 taking
36 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.
146
appendix
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 taking
9 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
147
8 Nominal group 1
8.1 Principal systems
148
appendix
8.2 Determiners
149
8.3 Determiners
150
appendix
8.4 Quantifiers
151
SECTION TWO
WORD–CLAUSE–TEXT
In the second section, which includes works spanning two decades

from the mid-1960s to the mid-1980s, we observe how Halliday’s
developing theoretical framework is applied in the analysis and descrip-
tion of patterns at various linguistic levels ranging from lexical item to
clause to text. The clause as lexico grammatical construct is different in
kind from both lexis and text. While each is defined differently –
lexical item defined by reference to collocation, clause by reference to
structure, and text by reference to context of situation – they are
nevertheless analogous in nature and systemic in orientation.
‘Lexis as a linguistic level’ (Chapter 6), published in 1966, first
appeared in a volume dedicated to Halliday’s mentor, Professor J. R.
Firth. Noting the importance Firth gave to lexical studies in descriptive
linguistics, Halliday explores some of the issues involved in handling
lexical patterns in language, which he regards as being different in kind,
not just in degree of delicacy, from grammatical patterns. What
Halliday is proposing is a lexical model “with distinct, though analo-
gous, categories and forms of statement”. In lexical analysis, the
occurrence of an item appears to have more to do with collocational
restrictions than a “known and stated set of terms in choice relation”.
This being the case, Halliday concludes, “even such a thing as a table
of most frequent collocates of specific items, with information about
their probabilities, unconditioned and lexically and grammatically 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’
155
word–clause–text
(Chapter 7) introduces the three grammatically relevant ‘language

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

analogy with the text as model, and can thus represent the meanings of
a text in a rich variety of different ways’. In ‘How is a text like a
clause?’, Halliday elaborates on how the textual properties of structure,
coherence, function, development and character have their analogous
counterparts in the organization of the clause.
The final paper in this section, ‘Dimensions of discourse analysis:
grammar’ (Chapter 10), published in 1985, illustrates the application of
systemic-functional grammar to the analysis of a sample of spoken
language, i.e. a discussion between an adult and three nine-year-old
schoolgirls. The analysis is presented in ten steps, ranging from 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 rhythm
2. analysis into clauses and clause complexes, showing interdepen-
dencies and logical semantic relations
3. analysis of clauses and clause complexes, for thematic (Theme–
Rheme) structure
4. comparison of clauses and information units, and analysis of the
latter for information (Given–New) structure
5. analysis of finite clauses for mood, showing Subject and Finite
6. analysis of all clauses for transitivity, showing process type and
participant and circumstantial functions
7. analysis of groups and phrases (verbal group, nominal group,
adverbial group, prepositional phrase)
8. analysis of grammatical and lexical cohesion
9. identification, rewording and re-analysis of grammatical
metaphors
10. description of context of situation and correlation with features
of the text
The goal of the analysis is to show how the text being studied derives
from the linguistic system and how it comes to mean what it does.
157
Chapter Six
LEXIS AS A LINGUISTIC LEVEL (1966)
At a time when few linguists, other than lexicographers themselves,

devoted much attention to the study of lexis, and outlines of linguistics
often contained little reference to dictionaries or to other methods in
lexicology, J. R. Firth repeatedly stressed the importance of lexical
studies in descriptive linguistics.1 He did not accept the equation of
“lexical” with “semantic”,2 and he showed that it was both possible
and useful to make formal statements about lexical items and their
relations. For this purpose Firth regarded the statement of collocation
as the most fruitful approach, and he sometimes referred, within the
framework of his general views on the levels of linguistic analysis, to
the “collocational level”.3
The aim of this paper is to consider briefly the nature of lexical
patterns in language, and to suggest that it may be helpful to devise
methods appropriate to the description of these patterns in the light of
a lexical theory that will be complementary to, but not part of,
grammatical theory. In other words, the suggestion is that lexis may be
usefully thought of (a) as within linguistic form, and thus standing in
the same relation to (lexical) semantics as does grammar to (grammat-
ical) semantics, and (b) as not within grammar, lexical patterns thus
being treated as different in kind, and not merely in delicacy, from
grammatical patterns. This view is perhaps implicit in Firth’s 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.
158
lexis as a linguistic level
of delicacy greater than has hitherto been attained, and it is rightly

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

is less easily compatible with the second view than is a model which
does not make this distinction and which allows statistical statements a
place in linguistic description; nevertheless it is not wholly incompatible
with it.6 Lexical statements, or “rules”, need not be statistical, or even
corpus-based, provided that their range of validity is defined in some
other way, as by the introduction of a category of “lexicalness” to
parallel that of grammaticalness.
One may validly ask whether there are general grounds, independent
of any given model, for supplementing the grammar by formal state-
ments of lexical relations, at least (given that the aim of linguistics is to
account for as much of language as possible) until these are shown to
be unnecessary. It may be a long time before it can be decided whether
they are necessary or not, in the sense of finding out whether all that is
explained lexically could also have been incorporated in the grammar;
there still remains the question whether or not it could have been
explained more simply in the grammar. The question is not whether
formal lexical statements can be made; they are already made in
dictionaries, although at a low level of generality, in the form of
citations. The question of interest to linguists is how the patterns
represented by such citations are to be stated with a sufficient degree
of abstraction, and whether this can best be achieved within or outside
the framework of the grammar.
Let us consider an example. The sentence he put forward a strong
argument for it is acceptable in English; strong is a member of that set of
items which can be juxtaposed with argument, a set which also includes
powerful. Strong does not always stand in this same relation to powerful:
he drives a strong car is, at least relatively, unacceptable, as is this tea’s too
powerful. To put it another way, a strong car and powerful tea will either
be rejected as ungrammatical (or unlexical) or shown to be in some
sort of marked contrast with a powerful car and strong tea; in either case
the paradigmatic relation of strong to powerful is not a constant but
depends on the syntagmatic relation into which each enters, here with
argument, car or tea.
Grammatically, unless these are regarded as different structures,
which seems unlikely, they will be accounted for in a way which,
whatever the particular form of statement the model employs, will
amount to saying that, first, strong and powerful are members of a class
that enters into a certain structural relation with a class of which
argument is a member; second, powerful (but not strong) is a member of a
class entering into this relation with a class of which car is a member;
160
and third, strong (but not powerful) is a member of a class entering into
this relation with a class of which tea is a member. It would be hoped
that such classes would reappear elsewhere in the grammar defined on
other criteria. Argument, car and tea will, for example, already have been
distinguished on other grounds on the lines of “abstract”, “concrete
inanimate” and “mass”; but these groupings are not applicable here,
since we can have a strong table and powerful whisky, while a strong device
is at least questionable.
The same patterns do reappear: he argued strongly, I don’t deny the
strength of his argument, his argument was strengthened by other factors.
Strongly and strength are paralleled by powerfully and power, strengthened
by made more powerful. The same restrictions have to be stated, to
account for the power (but not the strength) of his car and the strength (but
not the power) of her tea. But these involve different structures; elsewhere
in the grammar strong, strongly, strength and strengthened have been
recognized as different items and assigned to different classes, so that
the strong of his argument has been excluded on equal terms with the
strong of his car. Strong and powerful, on the other hand, have been
assigned to the same class, so that we should expect to find a powerful
car paralleled by a strong car. The classes set up to account for the
patterns under discussion either will cut across the primary dimension
of grammatical classification or will need to be restated for each primary
class.
But the added complexity involved in either of these solutions does
not seem to be matched by a gain in descriptive power, since for the
patterns in question the differences of (primary) class and of structure
are irrelevant. Strong, strongly, strength and strengthened can all be
regarded for this present purpose as the same item; and a strong argument,
he argued strongly, the strength of his argument and his argument was
strengthened all as instances of one and the same syntagmatic relation.
What is abstracted is an item strong, having the scatter strong, strongly,
strength, strengthened, which collocates with items argue (argument) and
tea; and an item power (powerful, powerfully) which collocates with argue
and car. It can be predicted that, if a high-powered car is acceptable, this
will be matched by a high-powered argument but not by high-powered tea.
It might also be predicted, though with less assurance, that a weak
argument and weak tea are acceptable, but that a weak car is not.
As far as the collocational relation of strong and argue is concerned, it
is not merely the particular grammatical relation into which these items
enter that is irrelevant; it may also be irrelevant whether they enter
into any grammatical relation with each other or not. They may be in
161
different sentences, for example: I wasn’t altogether convinced by his

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

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

up1 and make up2 it must recognize a new and independent dimension
of class membership on the basis of relations to which the previous
dimension is irrelevant. Any one example can of course be handled by
ad hoc grammatical devices: here for instance the potentialities of make
up2 in intransitive structures are more restricted. But such clearly
grammatical distinctions, even when present, are so restricted in their
range of validity that the generalizing power of a grammatical model is
of little value as compared with the cost, in increased complexity, of
the cross-classifications involved.
It may be worth citing a further example of a similar kind. We can
distinguish grammatically, but not lexically, between they want the pilot
to take off (= “so that they can take off” and = “they desire him to do
so”): these are not necessarily distinguished by intonation, although the
unmarked tone selections are different. On the other hand it is easier
to distinguish lexically than grammatically between he took two days off
(= “he did not work”) and (= “he reduced the time available”). In the
takeoff of the president (= “his becoming airborne”) and (= “the imitation
of the president”) the distinction can usefully be made both in grammar
and in lexis.
On the syntagmatic axis, it may be useful to recognize a lexical item
which has no defined status in the grammar and is not identified as
morpheme, word or group. For example, in he let me in the other day for
a lot of extra work, one could handle let in for as a single discontinuous
item in the grammar; but this complexity is avoided if one is prepared
to recognize a lexical item let in for without demanding that it should
carry any grammatical status. Similarly the ambiguity in he came out with
a beautiful model may be explained, instead of by giving two different
grammatical descriptions, by identifying two distinct lexical items, come
and come out with (and of course two different lexical items model).
It is not suggested of course that such non-coextensiveness between
the items of grammar and those of lexis is the norm, but merely that
for certain purposes it is useful to have a descriptive model of language
that allows for it. At the same time the above considerations suggest
that the lexical component requires not, as it were, a second ‘run-
through’ of the model designed for the grammar but rather a specifically
lexical model with distinct, though analogous, categories and forms of
statement.
Nor is it suggested that the set of patterns recognized as language
form is neatly divided into two types, the grammatical and the lexical.
A model for the description of language form may recognize only one
164
kind of pattern and attempt to subsume all formal relations within it:
some grammatical models, as has been noted, envisage that it is the
grammar’s task to distinguish strong from powerful as well as to dis-
tinguish a from the and “past” from “present”; while a lexicographical
model in which a and the, as well as strong and powerful, are entered in
the dictionary and described by means of citations could be regarded as
in a similar way attempting to subsume grammar under lexis. Even
where the model recognizes two distinct kinds of pattern, these still
represent different properties of the total phenomenon of language, not
properties of different parts of the phenomenon; all formal items enter
into patterns of both kinds. They are grammatical items when described
grammatically, as entering (via classes) into closed systems and ordered
structures, and lexical items when described lexically, as entering into
open sets and linear collocations. So in a strong cup of tea the grammar
recognizes (leaving aside its higher rank status, for example as a single
formal item expounding the unit “group”) five items of rank “word”
assignable to classes, which in turn expound elements in structures and
terms in systems; and the lexis recognizes potentially five lexical items
assignable to sets.
But, to take a further step, the formal items themselves vary in
respect of which of the two kinds of pattern, the grammatical or the
lexical, is more significant for the explanation of restrictions on their
occurrence qua items. The items a and of are structurally restricted, and
are uniquely specified by the grammar in a very few steps in delicacy;
collocationally on the other hand they are largely unrestricted. For the
item strong, however, the grammar can specify uniquely a class (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 from
other members of the class (and if the members of its ‘scatter’ strong,
strength, etc., turn out to operate collocationally as a single item then
this conflated item is not even specifiable qua class member); colloca-
tionally, however, it is restricted, and it is this which allows its
specification as a unique item. There might then appear to be a scale
on which items could be ranged from “most grammatical” to “most
lexical”, the position of an item on the scale correlating with its overall
frequency ranking. But these are three distinct variables, and there is
no reason to assume a correlation of “most grammatical” with either
“least lexical” or “most frequent”. The “most grammatical” item is one
which is optimally specifiable grammatically: this can be thought of as
“reducible to a one-member class by the minimum number of steps in
delicacy”. Such an item may or may not be “least lexical” in the sense
165
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 a
stated environment relative to its total frequency of occurrence. While
a and of are unlikely to occur in any collocationally generalizable
environment with a probability significantly different from their overall
unconditioned probabilities, there will be environments such that strong
occurs with a probability greater than chance. This can be regarded, in
turn, as the ability of strong to “predict” its own environment. As
extreme cases, fro and spick may never occur except in environments
including respectively to and span (the fact that to and fro accounts for
only a tiny proportion of the occurrences of to, while spick and span
may account for all occurrences of this item span, is immaterial to the
specification of fro and spick); here it is likely that, for this very reason,
to and fro and spick and span are to be regarded as single lexical items.
It is the similarity of their collocational restriction which enables us
to consider grouping lexical items into lexical sets. The criterion for the
definition of the lexical set is thus the syntactic (downward) criterion of
potentiality of occurrence. Just as the grammatical system (of classes,
including one-item classes) is defined by reference to structure, so the
lexical set (of items) can be defined by reference to collocation. Since
all items can be described lexically, the relation of collocation could be
regarded as being, like that of structure, chain-exhausting, and a lexical
analysis programme might well begin by treating it in this way; but this
is not a necessary condition of collocation, and if closed-system items
turn out, as may be predicted, to be collocationally neutral these items
could at some stage be eliminated by a “deletion-list” provided either
by cross-reference to the grammar or, better, as a result of the lexical
analysis itself. Once such “fully grammatical” items are deleted, 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 is
a chain-exhausting unit (text items being then fully accounted for in
sentence structure and again in clause structure and so on), it does not
seem useful to postulate such an ordered hierarchy for lexis. Lexical
items may indeed enter into a sort of rank relation: it is likely, for
example, that on collocational criteria we would want to regard stone,
grindstone and nose to the grindstone each as a separate lexical item, and
though triads of this kind may be rare it looks as though we need the
166
categories of simple and compound, and perhaps also phrasal, lexical

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

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

cosy chat and a friendly discussion, is it the simple co-occurrence of the
two items that is unlikely, or their occurrence in this particular
structure? All that has been said above has implied an approach in
which grammatical relations are not taken into account, and reasons
have been given for the suggestion that certain aspects of linguistic
patterning will only emerge from a study of this kind. But it is essential
also to examine collocational patterns in their grammatical 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 meaningful
one.9
An investigation on the lines suggested requires the study of very
large samples of text. The occurrence of an item in a collocational
environment can only be discussed in terms of probability; and,
although cut-off points will need to be determined for the purpose of
presenting the results, the interest lies in the degree of “lexicalness” of
different collocations (of items and of sets), all of which are clearly
regarded as “lexical”. Moreover the native speaker’s knowledge of his
language will not take the form of his accepting or rejecting a given
collocation: he will react to something as more acceptable or less
acceptable on a scale of acceptability. Likely collocations could be
elicited by an inquiry in which the subject was asked to list the twenty
lexical items which he would most expect to find in collocation with a
given node;10 but the number of such studies that would be required
to cover even the most frequent lexical items in the language is very
large indeed. Textually, some twenty million running words, or
1500–2000 hours of conversation, would perhaps provide enough
occurrences to yield interesting results. The difficulty is that, since
lexical patterns are of low generality, they appear only as properties of
very large samples; and small-scale studies, though useful for testing
methods, give little indication of the nature of the final results.
It is hard to see, however, how the results could fail to be of interest
and significance for linguistic studies. Their contribution to our know-
ledge of language in general, and of one language in particular, may
perhaps be discussed in relation to the use of the term “semantics”. If
lexis is equated with semantics, the implication is that lexical patterns
can only be described either externally (that is, as relations between
language and non-language, whether approached denotatively or con-
textually) or lexicogrammatically (that is, in dependence on grammati-
169
cal patterns). This restriction leaves two gaps in our understanding of

language: the internal relations of lexis, and the external relations of
grammar – that is, lexis (lexical form) and grammatical semantics. But
linguistics is concerned with relations of both types, both internal
(formal, within language) and external (contextual or “semantic”,
between language and non-language); and all linguistic items and
categories, whether operating in closed contrasts, like the and a, or
“past” and “present”, or in open ones, like strong and powerful, enter
into both. Moreover, as Firth stressed, both these types of relation are
“meaningful”: it is part of the meaning of “past” that it contrasts with
“present”, and it is part of the meaning of strong that it collocates with
tea. The fact that the labels for grammatical categories are chosen on
semantic grounds should not be taken to imply that they represent an
adequate substitute for grammatical semantics; but equally the existence
of traditional methods in lexical semantics does not mean that lexical
items display no internal, formal patterns of their own.
A thesaurus of English based on formal criteria, giving collocationally
defined lexical sets with citations to indicate the defining environments,
would be a valuable complement to Roget’s brilliant work of intuitive
semantic classification in which lexical items are arranged “according
to the ideas which they express”.11 But even such a thing as a table of
the most frequent collocates of specific items, with information about
their probabilities, unconditioned and lexically and grammatically 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. These
include studies of register and of literary style, of children’s language,
the language of aphasics and many others. In literary studies in particular
such concepts as the ability of a lexical item to “predict” its own
environment, and the cohesive power of lexical relations, are of great
potential interest.12 Lexical information is also relevant to foreign
language teaching; many errors are best explained collocationally, and
items can be first introduced in their habitual environments.13 A further
possible field of application is information retrieval: one research group
in this field is at present undertaking a collocational analysis of the
language of scientific abstracts. 14
Only a detailed study of the facts, such as that now being undertaken
by Sinclair,15 can show in what ways and to what extent the introduc-
tion of formal criteria into the study of lexis, as implied by the
recognition of a “lexical level”, are of value to any particular applica-
tions of linguistics. But there seem to be adequate reasons for expecting
170
the results to be interesting; and if they are, this is yet another indication
of the great insight into the nature of language that is so characteristic
of J. R. Firth’s contribution to linguistic studies.
Notes
1. See especially Firth (1935) (reprinted in Firth 1957b).
2. Firth (1957b: 195–6): ‘It must be pointed out that meaning by collocation
is not at all the same thing as contextual meaning, which is the functional
relation of the sentence to the processes of a context of situation in the
context of culture . . . Meaning by collocation is an abstraction at the
syntagmatic level and is not directly concerned with the conceptual or
idea approach to the meaning of words.’ Compare also Firth (1935) (in
“lexical items”): “This (sc. the lexical) function should not be misnamed
semantic”.
3. See Firth (1957a: 12). In the present paper “lexical level” has been used
in preference to “collocational level” in order to suggest greater generality
and parallelism with the grammatical level.
4. It is also stated explicitly by Firth (1957a: 12): “Collocations of a given
word are statements of the habitual or customary places of that word in
collocational order but not in any other contextual order and emphatically
not in any grammatical order”. Note that here “order” refers to the
“mutual expectancy” of syntagmatically related categories, such as 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 these
in turn define further sub-classes on various dimensions within previously
defined classes.
6. The place of collocational restrictions in a transformational grammar is
considered by Matthews (1961).
7. For a discussion of the relation between grammatical and lexical patterns
see McIntosh (1961).
8. Sinclair (1966).
9. The implication is, in effect, that “wellformedness” is best regarded as
“lexicogrammaticalness”, and that a departure from wellformedness may
be ungrammatical, unlexical or unlexicogrammatical. That the last two
are distinct is suggested by such examples as sandy hair, sandy gold and
sandy desk: sandy desk is unlexical, in that this collocation is unlikely to
occur in any grammatical environment, whereas sandy gold is merely
unlexicogrammatical: there is nothing improbable about golden sand. An
analogous distinction is observable in clichés: in shabby treatment the
mutual expectancy is purely lexical, and is paralleled in they treated him
171
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 disponibilité of lexical items in the
development of “Français 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, fashionable
airliner, modern cosy flights, economical experience, delightfully stressed,
serious stupid people, shining values, a wobbly burden, light possibility,
luxurious man, whose skin was bleeding, driving a bicycle, old and
disturbed bits of brick wall, a comprehensive traffic jam, her throat
became sad, my head is puzzled, people touched with assurance, thoughts
are under a strain, a sheer new super car.
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).
172
Chapter Seven
LANGUAGE STRUCTURE AND

LANGUAGE FUNCTION (1970)
1 The functions of language

Why is language as it is? The nature of language is closely related to
the demands that we make on it, the functions it has to serve. In the
most concrete terms, these functions are specific to a culture: the use
of language to organize fishing expeditions in the Trobriand Islands,
described half a century ago by Malinowski, has no parallel in our own
society. But underlying such specific instances of language use are more
general functions which are common to all cultures. We do not all go
on fishing expeditions; however, we all use language as a means of
organizing other people, and directing their behaviour.
A purely extrinsic account of linguistic functions, one which is not
based on an analysis of linguistic structure, will not answer the question;
we cannot explain language by simply listing its uses, and such a list
could in any case be prolonged indefinitely. Malinowski’s ethnographic
account of the functions of language, based on the distinction between
“pragmatic” and “magical”, or Bühler’s well-known tripartite division
into the “representational”, “expressive” and “conative” functions,
show that it is possible to generalize; but these generalizations are
directed towards sociological or psychological inquiries, and are not
intended primarily to throw light on the nature of linguistic structure.
At the same time, an account of linguistic structure that pays no
attention to the demands that we make of language is lacking in
First published in New Horizons in Linguistics, 1970, edited by John Lyons. Harmondsworth:
Penguin, pp. 140–65.
173
perspicacity, since it offers no principles for explaining why the

structure of language is organized one way rather than in another.
Here, therefore, we shall consider language in terms of its use.
Structural preoccupations have been dominant in linguistics for some
time; but the usefulness of a synthesis of structural and functional
approaches has long been apparent from the work of the Prague
linguists (Vachek 1966) who developed Buhler’s ideas, especially in the
study of grammar. The particular form taken by the grammatical system
of language is closely related to the social and personal needs that
language is required to serve. But in order to bring this out it is
necessary to look at both the system of language and its functions at
the same time; otherwise we will lack any theoretical basis for 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 add
something new, whether or not to include our own judgement, and so
on. It would be better, in fact, to say that we “opt”, since we are
concerned not with deliberate acts of choice but with symbolic
behaviour, in which the options may express our meanings only very
indirectly: in the same sense we may be said to “opt” between a long
vowel and a short one, or between a straight arm and a bent one
(where the meaning is likewise mediated through the symbolic signifi-
cance of the distinction between a handshake and a salute). The system
of available options is the “grammar” of the language, and the speaker,
or writer, selects within this system: not in vacuo, but in the context of
speech situations. Speech acts thus involve the creative and repetitive
exercise of options in social and personal situations and settings (Ellis
1966; Pike 1967; Firth 1968).
It is fairly obvious that language is used to serve a variety of different
needs, but until we examine its grammar there is no clear reason for
classifying its uses in any particular way. However, when we examine
the meaning potential of language itself, we find that the vast numbers
of options embodied in it combine into a very few relatively indepen-
dent “networks”; and these networks of options correspond to certain
basic functions of language. This enables us to give an account of the
different functions of language that is relevant to the general 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
174
language structure and language function
speaker’s experience of the real world, including the inner world of his
own consciousness. We may call this the ideational function, though it
may be understood as easily in behavioural as in conceptual terms
(Firth 1968: 91). In serving this function, language also gives structure
to experience, and helps to determine our way of looking at things, so
that it requires some intellectual effort to see them in any other way
than that which our language suggests to us.
2. Language serves to establish and maintain social relations: for the
expression of social roles, which include the communication roles
created by language itself – for example the roles of questioner or
respondent, which we take on by asking or answering a question; and
also for getting things done, by means of the interaction between one
person and another. Through this function, which we may refer to as
interpersonal, social groups are delimited, and the individual is identi-
fied and reinforced, since by enabling him to interact with others
language also serves in the expression and development of his own
personality.
These two basic functions, to each of which corresponds one broad
division in the grammar of a natural language, are also reflected in
Bernstein’s studies of educational failure (e.g. Bernstein 1970). Bern-
stein’s work suggests that in order to succeed in the educational system
a child must know how to use language as a means of learning, and
how to use it in personal interaction; these can be seen as specific
requirements on his control of the ideational and interpersonal func-
tions of language.
3. Finally, language has to provide for making links with itself and
with features of the situation in which it is used. We may call this the
textual function, since this is what enables the speaker or writer to
construct “texts”, or connected passages of discourse that is situationally
relevant; and enables the listener or reader to distinguish a text from a
random set of sentences. One aspect of the textual function is the
establishment of cohesive relations from one sentence to another in a
discourse (Hasan 1968).
All these functions are reflected in the structure of the clause. Here
we attempt to show, by reference to English, what a clause is: how it
serves for the realization of a number of very general meanings, or
semantic options, relating to the interpersonal, ideational and textual
functions of language; and how these are expressed through various
configurations of structural “roles” – functional elements such as
“process” and “actor” that derive from these basic functions. For a
more detailed exemplification we shall consider an aspect of ideational
175
meaning, the system of transitivity; the remaining areas, which have

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

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

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

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

The roles which appear in the expression of processes are of different
kinds. First there is the process itself, usually represented by a verb, for
example built in (2). Then there are the participant functions, the
specific roles that are taken on by persons and objects, for example
Wren and gazebo; and finally there are what we may call the circumstan-
tial functions, the associated conditions and constraints such as those of
time, place and manner (Fillmore 1968, where the two together are
referred to as “cases”; Halliday 1967–68).
It has been customary to recognize three participant functions in
English, namely “actor”, “goal” (or “patient”), and “beneficiary”.
Various subdivisions and modifications have been proposed, such as the
distinction between goal and ‘object of result’ (Lyons 1968: 439; cf.
“factitive” in Fillmore 1968: 25) as in:
(3i) the Borough Council restored this gazebo
(3ii) Sir Christopher Wren built this gazebo
where this gazebo is goal in (3i) but object of result in (3ii); in (3ii) the
gazebo comes into existence only as a result of the process of building.
Similarly, the beneficiary may be the recipient of an object, as Oliver in
(4i), or the recipient of a service, as Frederick in (4ii):
(4i) I’ve given Oliver a tie
(4ii) I’ve made Frederick a jacket
These subclassifications are not made arbitrarily; they account for
systematic distinctions in the grammar, e.g. the related prepositional
form is to Oliver but for Frederick in (4), and in (3) restore but not build
can be substituted by do to (what they did to this gazebo was restore it). But
there may be many, often contradictory, criteria to choose from (see
below, Section 4); moreover, the more categories one sets up, the
178
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 grammatical
subject only in (5i). Conversely in the book sells well, the book is gram-
matical subject but “logical direct object”. The concepts of actor, goal
and beneficiary are represented in Sweet’s account as “logical subject”,
“logical direct object” and “logical indirect object” respectively.
The linguistic expression of processes, and of the participants (and,
by extension, the circumstances) associated with them, is known by the
general term transitivity. Transitivity comes under what we have called
the ‘ideational’ function of language. Actor, Goal and Beneficiary are
structural functions, or roles, in transitivity; and just as the same
transitivity function may be realized in more than one way, as in (5),
so also the same constructional form may express different transitivity
functions. Thus by the fire is Actor in (6i), Place in (6ii):
(6i) it was singed by the fire

(6ii) it was stored by the fire
This also illustrates the conflict of criteria. In (6i), by the fire might be
considered instrument rather than actor, on the grounds that it is
inanimate. Fillmore (1968) distinguishes actor and instrument as,
respectively, the “typically animate perceived instigator of the action”
(his “agentive”; cf. below, Section 8) and the “inanimate force or
object causally involved in the action”; the latter may also be 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).
179
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 by
nominal groups and circumstances by adverbial groups – the last often
in the form of prepositional phrases. There are also incongruent forms
of expression, with functions of one type expressed by classes primarily
associated with another type, as in (8):
(8) dinner of roast beef was followed by a swim
Here the processes of eating and swimming are expressed by nouns;
the temporal relation between them by the verb follow; and of the two
participants, one is omitted and the other (roast beef ) is made to qualify
dinner (contrast in the evening they ate roast beef and then swam).
The circumstantial functions seem less central to the process than do
the participant functions; this is related to their inability to take on the
role of subject. But this peripheral status is not a feature of all
circumstantial elements, which can be subdivided into an “inner” and
“outer” type. Within the function “place”, in:
(9i) he was throwing stones at the bridge
(9ii) he was throwing stones on the bridge
at the bridge (the “inner” type) seems more central to the process than
on the bridge: we can say what was he throwing stones at? and not (in this
sense) what was he doing at the bridge? (On the other hand, we can say
180
what was he doing on the bridge? and not what was he throwing stones on?)
However, the sense of “inner” and “outer” is contributed to by various
factors not all of which coincide. For example, in (10) the place
element is obligatory in (i) but optional in (ii):
(10i) he put all his jewels in the wash
(10ii) he lost all his jewels in the wash
In (11), there is a difference of clause type; (i) is a relational clause
(see below, Section 7) whereas (ii) is an action clause (Fillmore, from
whom (11) is taken, gives this as an instance of dependency between
functions: the place element is “outer” if an actor is present and “inner”
otherwise):
(11i) John keeps his car in the garage
(11ii) John washes his car in the garage
5 Inherent functions
The distinction between obligatory and optional roles helps us to relate
transitivity functions to a system of clause types. As, however, this
involves recognizing that an “obligatory” element may in fact be
absent, we shall use the term “inherent” rather than “obligatory”. An
inherent function is one that is always associated with a given clause
type even if it is not necessarily expressed in the structure of all clauses
of that type. (We are not here talking about ellipsis, which is a matter
of textual structure.)
Consider a pair of clauses such as (12):
(12i) Roderick pelted the crocodile with stones
(12ii) the crocodile got pelted
The verb pelt, as it happens, is always associated with three participant
roles: a pelter, a pelted and something to pelt with; and this holds for
(ii) as well as for (i) (cf. Svartvik (1966), on “agentless agentives”).
Similarly there are inherently benefactive clauses without a beneficiary,
such as we’re giving a silver coffee-pot. So:
(12iii) Roderick pelted the crocodile
is “(inherently) instrumental”, and although no instrument is men-
tioned the receiver interprets the process as having an instrumental role
associated with it.
The same verb may occur in clauses of more than one type. But
181
within one type there may be different sets, and different alignments,
of participants; this is the function of the system of voice – of the choice
between active and passive, though the actual patterns are more
elaborate than this. The options in the voice system (simplifying
somewhat) are (a) middle / non-middle (see next paragraph); if non-
middle, then (b) “active” / “passive” (not exactly equivalent to active
and passive in the verb; see Halliday, 1967–68: § I 39ff., where they
are referred to as “operative” and “receptive”); if “active”, then (c)
plus / minus goal; if ‘passive’, then (d) plus / minus actor. The reason
for choosing one rather than another of these options lies in the textual
function of language (see Sections 11 and 12 below); but which options
are available to choose from depends on transitivity.
Voice is concerned with the roles of Actor and Goal (but see below,
Section 8), both as inherent and as actualized roles. A “middle” clause
is one which has only one inherent participant, which for the moment
we will continue to refer to as the “Actor”; examples are Hector sneezed,
the cat washed. A “non-middle” clause is one which has two, an Actor
and a Goal, but one or the other may not be actualized: if “active”,
there may be no Goal, for example Mary is washing (“the clothes”), and
if ‘passive’, no clear Actor, e.g. the clothes have been washed (“by Mary”).
All actions are classified into those involving one participant role and
those involving two; there are then different ways of presenting the
situation in those cases where there are two.
The point was made earlier that the notion of “participant” derives
from the more fundamental concept of syntactic function, or “role”.
The basic elements of transitivity structure are the various roles associ-
ated with processes; and two or more such roles may be combined
in one participant, as in a reflexive clause such as John is washing
(“himself ”) where John is both Actor and Goal at the same time. The
elements that operate as Actor, Goal, etc. also play a part, simul-
taneously, in other structures of the clause, expressing aspects of the
interpersonal and textual functions of language. The principle of
combining a number of roles in a single complex element of structure
is fundamental to the total organization of language, since it is this that
makes it possible for the various functions of language to be integrated
in one expression. We return to this below, Section 9.
6 Transitivity clause types: action clause

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

equative forms as in example (32) below, having do or happen in them,
such as:
(13i) what Lionel did was (to) jump off the roof
(13ii) what happened to Lionel was that he fell off the roof
The following table shows the full range of possibilities of voices in
action clauses, together with the roles associated with each of them:
voice roles voice example

(clause) (verb)
middle Actor active the gazebo has collapsed
n
o active Actor, Goal active the Council are selling the gazebo
{
n active
m passive
i
d passive
d
l passive
e
Actor, (Goal) active the Council won’t sell
Goal
Goal, Actor
active the gazebo won’t sell
passive the gazebo has been sold by the Council
Goal, (Actor) passive the gazebo has been sold
The role in parentheses are inherent but not expressed.

Not all clauses are of the “action” type. English appears to recognize
three main types of process: action, mental process and relation. Mental
process clauses, and clauses of relation, are associated with what are at
first sight rather different sets of participant roles.
7 Transitivity clause types: mental process clauses, relation

clauses
In mental process clauses, such as:
(14) I liked your hairstyle
we cannot really talk of an Actor and a Goal; it is not possible to say,
for example, what I did was like your hairstyle, or what I did to your
hairstyle was like it. The inherent roles are those of a human, or at any
rate animate, being whose consciousness is impinged upon, and some
phenomenon which impinges upon it. Let us refer to these as the
“Processer” and the “Phenomenon”. The voice potentialities are now
somewhat different; among the non-middle (two participant) clauses
there are two types: those having the Phenomenon as Subject in active
183
voice (15i), and those having the Processer (15ii). In the first types, the
passive form is much more frequent than the passive in action clauses;
in the second type it is much less so:
(15i) the gift pleased her / she was pleased by (with) the gift
(15ii) she liked the gift / the gift was liked by her
This is because the passive is a means of bringing the element governed
by by into prominence as the focus of information (see below, Section
12); in (15ii) the by element, i.e. her, is the Processer, and in English
this tends to be the “given” element in the situation (she must have
been referred to already in the text), and thus does not appropriately
carry such prominence.
Mental process clauses express (a) perception, e.g. see, look; (b)
reaction, e.g. like, please; (c) cognition, e.g. believe, convince; (d) verbali-
zation, e.g. say, speak. They are distinct in that the Phenomenon – that
which is perceived, reacted to, etc. – is not limited, as are the
participants in action clauses, to the class of “thing”, namely persons,
objects, abstractions and the rest of the phenomena on the plane of
experience.
What is perceived or felt or thought of may be a simple 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 were
been filtered through the medium of language. Here words as well as
things may participate in the process.
For example, in (16) all the “processed” entities are simple 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 clauses
accept only reports, and “reported speech” is the meaning of clauses of
184
this type. The difference between fact and report is that a “fact” is a
representation at the semantic level, where the truth lies in the meaning
– (she regretted) that he had gone away; whereas a “report” is a represen-
tation at the lexicogrammatical, or syntactic, level, where the truth lies
in the wording – (she said) that he had gone away.
In relational clauses, the “process” is simply a form of relation
between two roles. One type is the attributive, such as:
(19i) Marguerite is a poet
(19ii) Marguerite looks desperate
where the relation is one of class membership: “Marguerite belongs to
the class of poets,” “. . . the class of people who look desperate”. This
is a relation between entities of the same order of abstraction but
differing in generality.
The other type, exemplified by (20):
(20i) Templecombe is the treasurer
(20ii) the treasurer is Templecombe
has two functions, resembling the two terms of an equation, where the
one serves to identify the other, as in x = 2. Here the two entities are
alike in generality but differ in abstraction: the identifying element may
be of a higher order of abstraction, as in (21i), where the treasurer
expresses Templecombe’s function, or of a lower order, as in (21ii)
where the fat one expresses Templecombe’s form, how he is to be
recognized:
(21i) (which is Templecombe?) Templecombe is the treasurer
(21ii) (which is Templecombe?) Templecombe is the fat one
(21i) could be interpreted in the sense of (21ii) if the committee were
in view on the platform; there is in fact partial ambiguity between
these two sub-types.
These two major types of relational clause, the attributive and the
equative, differ in various respects. The attributive are non-reversible
(e.g. we can say that man is a poet but not a poet is that man), have the
role Attribute which may be an adjective and is usually indefinite,
express class inclusion, are usually questioned by what? or how? and are
expressed by the verbs be, get, turn, keep, remain, seem, sound, look, etc.
The equative are reversible (i.e. have a “voice” system), have the role
Identifier which must be as noun and is usually definite, express class
identity, are usually questioned by who? or which? and are expressed by
the verbs be, equal, represent, resemble, stand for, etc.
185
It is interesting to note that, in relational clauses, quite unlike clauses

of action or mental process, the verb is regularly unstressed. This is a
symptom of its much weaker function in the clause. Contrast the
pronunciation of equals in (22i) and (22ii):
(22i) England Equals (Australia’s Total of) 512 [action]
(22ii) 29 = 512 [relation]
8 The ergative
As far as the ideational component of grammar is concerned, the
English clause shows the three principal types – action, mental process
and relation – and associates with each a set of different inherent roles,
or structural functions. The system of clause types is a general frame-
work for the representation of processes in the grammar; possibly all
languages distinguish three such categories. We need to ask, at this
point, whether the structural functions can be generalized across clause
types; whether, for example, an Actor in an action clause can be shown
to be equivalent to a Processer (one who does the thinking, etc.) in a
mental process clause. This may be approached through a reconsidera-
tion of the functions in action clauses, a reconsideration which such
clauses demand anyway.
If we look at examples like (23i and ii):
(23i) the sergeant led the recruits
(23ii) the sergeant marched the recruits
they appear to be clearly distinct, (i) being transitive, with Actor and
Goal, (ii) causative, with Initiator and Actor. However, there is a
problem with (23iii):
(23iii) the sergeant trained the recruits
Is it like (23i) or like (23ii)?
Actually it is like both; (23i) and (23ii) are not really different as far
as transitivity is concerned. In English no very clear distinction is made
between doing something to someone and making someone do some-
thing, so that (23iii) can be interpreted in either way without any sense
of ambiguity. This is why so many verbs are labelled “vb trans. &
intrans.” in the dictionary.
The concepts of actor and goal are not well suited to describing this
situation, since with these we are forced to describe (23i) and (23ii)
differently. The distinction between them is by no means entirely
186
unreal, since there are verbs like lead which are normally transitive (two
inherent 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 verb
bounce, between he bounced the ball and the ball bounced. In addition there
are a number of verbs which, while themselves clearly transitive or
clearly intransitive, group into pairs differing only in transitivity, so that
Mary put out the fire is to the fire went out as Polly lit the fire is to the fire
lit.
It has been pointed out by various linguists (Halliday 1967–68: § 3;
Anderson 1968; Fillmore 1968) that action clauses in English seem to
be organized on an ergative rather than on a transitive (or “nomina-
tive”) basis. This means that, with any action clause, there is associated
one inherent role which is that of the participant affected by the process
in question. Fillmore describes this as the “semantically most neutral”
function, and labels it the “objective”; I used the term “affected”,
which I will retain here. In (23) the recruits has the role of “affected” in
every case, even through it is Goal (if an Actor–Goal analysis is used)
in (23i) and Actor in (23ii); in general, the affected is the Goal in a
transitive and the Actor in an intransitive clause.
We have now turned what was the borderline case, such as (23iii),
into the most central clause type. This is the type in which both middle
(one-participant) and non-middle (two-participant) forms are equally
normal; it may be considered the “favourite” clause type of Modern
English. The transitive and intransitive types – those with non-middle
as norm and with middle as norm respectively – are the marginal ones,
and they seem to be becoming more marginal as time goes on.
Hence all the examples in (24i) have the same structure, with a
Process and an Affected. Those in (24ii) also have a Causer (Fillmore’s
“agentive”):
(24i) they’re being led
they’re being trained / they’re training
they’re being marched / they’re marching
(24ii) he’s leading them
he’s training them
he’s marching them
These two ways of representing processes, the transitive and the
ergative, are very widely distributed; possibly all languages display one
or the other, or (perhaps always) both, in different mixtures. In English,
187
the two occur side by side. The transitive system asks “does the action
extend beyond the active participant or not?”, the ergative, “is the
action caused by the affected participant or not?” The ergative com-
ponent is more prominent now than it was in Middle English, and this
appears in various ways, for example, the change from impersonal to
personal forms in mental process clauses (formerly methinks, it likes me).
In the modern form I like, I cannot be explained as an Actor (among
other things we cannot say what he does to jam is like it); but it can be
shown on various grounds to have the function Affected.
As this suggests, the ergative pattern, whereby a Process is accom-
panied by an obligatory Affected participant and an optional Causer, is
more readily generalizable than that of Actor and Goal. It extends
beyond action clauses to those of mental process, and perhaps even to
clauses of relation as well. We want to say that Paul has the same
function in both (25i) and (25ii):
(25i) Paul fears ghosts

(25ii) ghosts scare Paul
– not that they are identical in meaning, but that the transitivity roles
are the same. This is not possible in Actor–Goal terms. But in an
ergative system there is considerable evidence for regarding Paul as the
‘Affected’ participant in both cases. The ergative, therefore, represents
the more general model of the transitivity patterns of Modern English
– that is, of the options available to the speaker of English for talking
about processes of all kinds.
9 Other dimensions of clause structure

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

(26ii) these beads I was given by my mother
188
The solution was to recognize different kinds of subject. For Sweet,

my mother was “logical subject”, I was “grammatical subject”; these beads
came to be known as ‘psychological subject’. In (26i), all these coincide.
The notion of subject conflates three distinct roles which, although
they are typically combined into one element, are nevertheless 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 selected
unless there is good reason for selecting otherwise (cf. Jakobson 1963:
268ff.).
These three “kinds of subject” relate to the functions of language as
described above. The logical subject is the actor; this is a transitivity
role, deriving from the ideational function. The other two have
different sources, though they are no less meaningful. The grammatical
subject derives from the interpersonal component in language function:
specifically, it has to do with the roles taken on by the performer and
receiver in a communication situation. The psychological subject
belongs to the textual component; it is concerned with the organization
of the clause as a message, within a larger piece of discourse. The next
two sections will examine these in turn.
10 Mood
As we have said, one function of language is to provide for interaction
between people, by allowing the expression of statuses, social and
individual attitudes, assessments, judgements and the like; and this
includes participation in linguistic interaction. Language itself defines
the roles which people may take in situations in which they are
communicating with one another; and every language incorporates
options whereby the speaker can vary his own communication role,
making assertions, asking questions, giving orders, expressing doubts
and so on. The basic “speech functions” of statement, question,
response, command and exclamation fall within this category (though
they do not exhaust it), and these are expressed grammatically by the
system of mood (cf. Sweet 1891: 105), in which the principal options
are declarative, interrogative (yes / no and wh- types) and imperative.
The difference between he can and can he? is a difference in the
communication role adopted by the speaker in his interaction with a
listener.
The notion “grammatical subject” by itself is strange, since it implies
a structural function whose only purpose is to define a structural
189
function. Actually, just as the “logical subject” is a function defined by

transitivity, so the “grammatical subject” is a function defined by
mood. If we consider an example such as (27):
(27) Tigers can climb trees. – Can tigers climb trees? – They can
climb trees, can’t they? – No they can’t.
we find that one part, tigers can, has the function of expressing mood
throughout; it also typically carries the positive / negative option. It
consists of the finite element of the verb, plus one nominal (noun or
noun group) which is the “grammatical subject”.
The function of the “grammatical subject” is thus a meaningful
function in the clause, since it defines the communication role adopted
by the speaker. It is present in clauses of all moods, but its significance
can perhaps be seen most clearly in the imperative, where the meaning
is “request you to . . .”; here the speaker is requiring some action on
the part of the person addressed, but it is the latter who has the power
to make this meaning “come true” or otherwise, since he can either
obey or disobey. In the usual form of the imperative, this modal entity,
or “modal subject” as we may call it, is the listener; and the only
option is plus or minus the speaker himself, as in let’s go home as
opposed to (you) go home. Hence, in a passive imperative such as be
guided by your elders, although the actor is your elders, the modal subject
is “you”; it is the listener who accedes, potentially, to the request,
fulfilling the modal function defined by the speaker’s role.
11 Theme
The basic unit of language in use is not a word or a sentence but a
“text”; and the “textual” component in language is the set of options
by means of which a speaker or writer is enabled to create texts – to
use language in a way that is relevant to the context. The clause, in
this function, is organized as a message; so in addition to its structure
in transitivity and in mood, it also has structure as a message, what is
known as a “thematic” structure. (It was linguists of the Prague school
who first studied this aspect of language, cf. Mathesius 1928; Firbas
1959; 1964; Svoboda 1968 and references therein.)
The English clause consists of a Theme and a Rheme. The Theme
is another component in the complex notion of subject, namely the
“psychological subject”; it is as it were the peg on which the message
is hung, the theme being the body of the message. The Theme of a
clause is the element which, in English, is put in first position; in (28
190
i–v) the Theme is the item outside the brackets, what is inside being
the 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 unless
there is good reason for them not to be (cf. (26) above). Where they
are not, the tendency in Modern English is to associate Theme and
modal Subject; and this is the main reason for using the passive. The
passive has precisely the function of dissociating the Actor from this
complex, so that it can either be put in focal position at the end or,
more frequently, omitted, as in (29):
(29i) this gazebo was built by Sir Christopher Wren
(29ii) this gazebo is being restored
The typical theme of declarative clause is thus the modal Subject (or
“grammatical subject” – this gazebo in both cases); in interrogatives,
however, the picture is different. If we ask a question, it is usually
because we want to know the answer, so that the typical Theme of an
interrogative is a request for information. Hence we put first, in an
interrogative clause, the element that contains this request for 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 by
putting something in first position in the clause. The Theme is the
point of departure for the message; a paradigm form of it is the
headword in a definition, for example a gazebo in (31):
(31) a gazebo is a pavilion or summerhouse on an eminence, open
for the view
In addition to the selection of a particular element as the theme, the
speaker has other options in thematic structure open to him (Halliday
1967–68: § 2); for example, any clause can be split into two parts by
the use of nominalization, as in
191
(32) the one who built this gazebo was Sir Christopher Wren
where the Theme is the whole of whichever part comes first – here the
one who built this gazebo.
12 Information structure
Thematic structure is closely linked to another aspect of the textual
organization of language, which we may call information structure. This
refers to the organization of a text in terms of the functions Given and
New. These are often conflated with Theme and Rheme under the
single heading “topic and comment”; the latter, however, is (like the
traditional notion of “subject”) a complex notion, and the association of
Theme with Given, Rheme with New, is subject to the usual “good
reason” principle already referred to – there is freedom of choice, but
the Theme will be associated with the Given and the Rheme with the
New unless there is good reason for choosing some other alignment.
In English, information structure is expressed by intonation. Con-
nected speech takes the form of an unbroken succession of distinctive
pitch contours, or tone groups; each tone group represents what the
speaker decides to make into one unit of information. This is not
necessarily the same length as a clause, though it often is so. The
information unit consists of an obligatory New element – there must
be something new, otherwise there would be no information – and an
optional Given element; the main stress (tonic nucleus) marks the end
of the New element, and anything that is Given precedes it, unless
with good reason – which means, here, unless it is a response to a
specific question, either asked or implied. The function Given means
‘treated by the speaker as non-recoverable information’: information
that the listener is not being expected to derive for himself from the
text or the situation.
(33) illustrates the interaction of information structure with thematic
structure (information unit (“i.u.”) boundaries are marked by //; main
stress is indicated by bold type; 4 = falling-rising tone; 1 = falling
tone):
(33) //4 this gazebo //1 can’t have been built by Wren//
(clause: Theme . . . Rheme. . . . . . . . . . . . . . . . . . . .)
(i.u.(1): New. . . . . . .; (2): New. Given. . . . . . . . . . . . . . . .)
meaning ‘I am talking (Theme), specifically, (New) about this gazebo:
the fact is (Rheme) that your suggestion (Given) that Wren built it is
192
actually (New) quite impossible’. No such suggestion need actually

have been made for this clause to occur; one of the features of the
Given–New structure is its use for various rhetorical purposes, such as
bullying the listener. Given and New thus differ from Theme and
Rheme, though both are textual functions, in that Given means ‘here
is a point of contact with what you know’ (and thus is not tied to
elements in clause structure), whereas Theme means ‘here is the
heading to what I am saying’.
The functions of given and new link up in turn with the functions
in transitivity. It was noted earlier (see 3 above) that a number of
participant roles may be expressed in either of two ways, either directly
or through the mediation of a preposition, for example the Beneficiary
in:
(34i) I’ve offered Oliver a tie
(34ii) I’ve offered the tie to Oliver
The members of such a pair have the same ideational meaning but
differ in information. Typically, the prepositional form of the Benefici-
ary is associated with the function New, the other form with the
function Given; and if we assume here the expected intonation pattern,
then in (34ii) Oliver is New and the tie is Given, the implied question
being ‘who did you offer the tie to?’, while in (34i) a tie is New and
Oliver is Given, answering ‘what have you offered to Oliver?’ (note
that one of the meanings of definiteness – not the only one – is Given,
hence the likelihood of the tie in (34ii)).
A general principle underlies the existence of these two information-
ally distinct forms, one with a preposition and one without, for
expressing participant roles. The textual function of language requires
that, for effective communication, new information should be made
grammatically explicit. New lexical content has to be backed up, as it
were, by adequate quanta of grammar; specifically, it has to be made
clear what is the ideational function of any new material in the
discourse, and here it is the preposition that indicates the role of the
unfamiliar element. The use of a preposition to specify function in the
clause in just those cases where the element in question is typically
New (compare the use of by with the Actor in a passive construction)
illustrates how the “texture” of discourse is achieved through the
interplay of varied grammatical resources expressing different facets of
the total meaning.
193
13 Conclusion
The subject, in its traditional sense, is thus a complex of four distinct
functions, three in the structure of the clause (cf. Lyons 1968: 343–4):
1. Actor (“logical subject”): ideational
2. modal Subject (“grammatical subject”): interpersonal
3. Theme (“psychological subject1”): textual
together with a fourth function which is in the structure of the
‘information unit’
4. Given (‘psychological subject2’): textual
These coincide unless there is “good reason” for them not to do so;
thus in (35i) the Borough Council is Actor, modal Subject and Theme,
whereas in (35ii) the Borough Council is Actor, this gazebo is modal
Subject and next year is Theme:
(35i) the Borough Council will restore this gazebo next year
(35ii) next year this gazebo will be restored by the Borough Council
No mention has been made of subject and predicate as a logical
relation. We might introduce “predication” as another dimension of
clause structure, with the Borough Council in (35i) being also “subject in
predication” and the rest predicate; but the subject in this sense would
be identical with the modal subject. The subject-predicate structure is
entirely derivable from mood, and has no independent significance (cf.
Fillmore 1968: 17; and Fillmore’s reference, ibid., to Tesniēre 1959:
103–5). As a form of generalization, it may be useful in that it expresses
the fact that Actor, modal Subject and Theme are regularly associated;
but it obscures the equally important fact that they are distinct and
independent structure roles.
The multiple function of language is reflected in linguistic structure;
this is the basis for the recognition of the ideational (including logical),
interpersonal and textual functions as suggested here. It is not necessary
to argue that one function is more abstract, or “deeper”, than another;
all are semantically relevant. The investigation of these functions
enables us to relate the internal patterns of language – its underlying
options, and their realization in structure – to the demands that are
made on language in the actual situations in which it is used. As
performers and receivers, we simultaneously both communicate
through language and interact through language; and, as a necessary
condition for both of these, we create and recognize discourse (the
194
textual function is thus instrumental to the other two). A speech act is

essentially 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 broad
range of more specific patterns relating to the creative and the repetitive
aspects of language in use.
195
Chapter Eight
MODES OF MEANING AND MODES

OF EXPRESSION: TYPES OF
GRAMMATICAL STRUCTURE AND THEIR
DETERMINATION BY DIFFERENT
SEMANTIC FUNCTIONS (1979)
1 Preamble
Let us say that a code is a system of signs having a content and an
expression: for example, a traffic control code (Figure 1).
Figure 1
The relation between the content and the expression is one of
realization. Then, a semiotic, or semiotic system, is a code having two
or more realizational cycles in it, so that the expression of content1 (call
it expression1) is itself a content (content2) that in turn has its own
expression (expression2). Hence there will be at least three levels, or
strata, in such a system:
level one: content1
two: (realized as) expression1 = content2
three: (realized as) expression2 . . .
A semiotic, in other words, is a stratified, or stratal, system, in which
the output of one coding process becomes the input to another.
First published in Function and Context in Linguistic Analysis: A Festschrift for William Haas,
1979, edited by D. J. Allerton, Edward Carney and David Holdcroft. Cambridge:
Cambridge University Press, pp. 57–79.
196
modes of meaning and modes of expression
In this sense language is a semiotic. It consists (at least) of three strata

with, therefore, two realizational cycles; these are set out in Figure 2,
(i) in everyday and (ii) in technical terminology. The formulation “at
least” is intended to allow for the addition of further strata above the
semantic system, since the semantic system itself can be regarded as the
realization of some higher level semiotic. In principle this may be
associated with any of a number of different orders of meaning,
cognitive, social, aesthetic and other things besides. At any particular
time, attention is likely to be focused on one or other of these higher
orders.
It follows from this that as far as the elements of a semiotic system
are concerned, we may in principle consider the organization of any
one part of the system from three different aspects:
(1) at its own level – its relation to other elements identified at the
same level as itself
(2) from above – its relation to elements at the next (or some)
higher level
(3) from below – its relation to elements at the next (or some)
lower level
Figure 2
Now it is typical of semiotic systems that the different strata are not
isomorphic; there is no relation of biuniqueness (one–one correspon-
dence) between one level and the next. This is bound to be the case in
a system such as language, where the coding not only converts elements
of one kind into elements of another kind – meanings into wordings
into sounds – but also reduces both the size and the inventory of the
basic components. By any usual definition of linguistic units, units of
speech sound are both smaller than and fewer than units of form; and
units of form are both smaller than and fewer than units of meaning.
197
Hence when one organizational system is represented in terms of

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

Let us focus first on the semantic system, and introduce a broad
generalization along the following lines. The semantic system of a
natural language is organized into a small number of distinct compo-
nents, different kinds of meaning potential that relate to the most
general functions that language has evolved to serve. Here are the
headings we shall use:
IDEATIONAL INTERPERSONAL TEXTUAL
EXPERIENTIAL LOGICAL
The first of these is language as representation: the semantic system
as expression of experience, including both experience of what is round
about us in the outside world and experience of the world of 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, animate
and inanimate objects, institutions, abstractions) and circumstantial
features (extent, location, time and space, cause, manner and so on);
and a logical, where we represent experience “indirectly” in terms of
certain fundamental logical relations in natural language – ‘and’,
‘namely’, ‘says’, ‘is subcategorized as’, etc. – which are not those of
formal logic but rather are the ones from which the operations of
198
formal logic are ultimately derived. These two, the logical and the
experiential, 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 as
interaction: it is meaning in the active mode. Here the semantic system
expresses the speaker’s intrusion in the speech event: his attitudes,
evaluations and judgements; his expectations and demands; and the
nature of the exchange as he is setting it up – the role that he is taking
on himself in the communication process, and the role, or rather the
role choice, that he is assigning to the hearer. This component is
therefore both speaker- and hearer-oriented; it is interpersonal – what
Hymes called “socio-expressive” – and represents the speaker’s own
intrusion into the speech situation.
All discourse involves an ongoing simultaneous selection of meanings
from both these components, which are mapped into a single output
in the realization process. But there is also a third component, which
we are calling the textual, whereby the meanings of the other two
kinds take on relevance to some real context. Here the semantic system
enables the speaker to structure meaning as text, organizing each
element as a piece of information and relating it significantly to what
has gone before. If the ideational component is language as reflection
(the speaker as observer of reality), and the interpersonal component is
language as action (the speaker as intruder in reality), the textual
component is language as relevance (the speaker as relating to the
portion of reality that constitutes the speech situation, the context
within which meanings are being exchanged). The textual component
provides what in modern jargon we might refer to as the ecology of
the text. For example, from the Walrus and the Carpenter, in Alice
Through the Looking-Glass, when the Carpenter says to the Walrus
Cut us another slice!
the ideational meaning is the representation of a material process,

cutting, in which three entities participate: the one who cuts, the thing
that is cut and the one that the thing is cut for; also the place of cut in
the taxonomy of actions and of slice in the taxonomy of things. The
interpersonal meaning is a demand for goods-and-services, “I want you
to do something for me”, embodied in the selection of the imperative
mood, direct, explicit and without any special modulation. The textual
meaning is the internal organization of this as a message with the focus
on what is demanded, together with its relation to the preceding text
199
through presuppositions – a slice of something, of which I have already

had at least one.
3 Above and below the semantic system

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

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

components of the meaning potential which are activated by different
components of the social context (cf. Halliday 1977). This appears to
be the basis on which interactants make predictions about the meanings
that will typically be exchanged in any particular situation with which
they find themselves confronted.
4 The experiential mode

For the rest of this paper we shall be concerned with what is “below”
the semantic system: with the question of what kinds of structural
mechanism are typically involved in the realization of these various
components of meaning. The suggestion will be that here too the same
categories are relevant, since they tend to be expressed through
fundamentally different types of structural organization (cf. Mathesius
1964: 24–5 of Czech original).2
Let us consider the experiential function first. Here we are con-
cerned with the semantic (linguistic) encoding of experience; particu-
larly our experience of the processes of the external world and of the
internal world of our own consciousnesses. We tend to encode such
experience in terms of configurations of elements each of which has a
special and distinct significance with respect to the whole. Typically,
we recognize a process itself, and various more or less specialized
participants and circumstantial elements.
For example, suppose there is a flock of birds flying overhead. We
represent this in language as something like There are birds flying: that is,
a process of “flying” and, separated out from this, an entity that is
doing the flying, namely “birds”. This is, certainly, one valid way of
encoding it; but it is not the only one – we might have said, instead,
It’s winging. If we did say this in English we would be treating the
phenomenon as a single unanalysed process, not as a process plus a
participant; this is, after all, what we do with It’s raining (although not,
for some reason, with The wind’s blowing). No doubt it is useful to be
able to talk about birds doing other things than flying, and about flying
being done by other things than birds. Some languages feel the same
about rain: in Chinese one says, liberally translated, “There’s rain
falling”; and in one south Chinese dialect, a variety of Cantonese, there
are usually two participants in the pluvial process, which is encoded as
“The sky is dropping water”.
So there is no reason for assuming that each particular process will
always be encoded as just this or that particular configuration of
202
elements. But we can formulate the general principle that this is how
processes are represented in languages. This means that a structure
which represents experiential meanings will tend to have this form: it
will be a configuration, or constellation, of discrete elements, each of
which makes its own distinctive contribution to the whole.
We usually represent this kind of structure linguistically as a func-
tionally labelled constituent structure as shown, for example, in Figure
3.
Figure 3
There is no particular reason at this stage why the representation should
have to be linear; that is no doubt the form of the final output, after
the other structures have been mapped on to it, but experiential
structures are not in fact sequential and we could just as well represent
this as in Figure 4. If ordering is to be introduced into the representa-
tion, there is the possibility of using
Figure 4
a dependency construct having the Process at the centre, as in Figure
5. But a more appropriate ordering would have a nucleus consisting of
Process plus Goal, with the other elements clustering around it, as in
Figure 6. (We will leave aside here the question of the appropriateness
of the functional labels themselves.)
Figure 5 Figure 6
203
Each one of these elements may have a substructure of the same

general type, with different labels as, for example, in Figure 7. This
again can be reinterpreted as Figure 8. The layers
Figure 7 Figure 8
introduced into the structure in this way are the source of the linguist’s
classic mode of representation of constituent structure in the form of
trees, or labelled bracketings. But the appropriate bracketing for func-
tionally labelled structures is minimal, not maximal (“many ICs” rather
than “few ICs”, in the terms of Hudson 1967). Maximal bracketing
imposes too much structure for a functional grammar: for example, in
four young oysters it is reasonable to recognize young oysters as a
constituent provided the labelled elements are classes, since it is a
nominal group, but in no way does this correspond to any meaningful
functional constituent. The nonlinear representation implies more of a
molecular model of structure, with a taxonomy similar to cell: mol-
ecule: atom: subatomic particle.
Experiential meanings are typically realized as elemental structures of
this type. The basic structural mechanism is that of constituency, with
larger units constituted out of layered clusters or bracketed strings of
smaller units, each part having its own specific function with respect to
the whole. We could call this “segmental”, except that it is better
perhaps to reserve the term “segment” for an element in the final
output – the syntagm – that serves as input to the next realizational
cycle. So let us say that experiential meanings are realized through
some kind of constituent structure.
This expresses the particular way in which we order our experience
of reality when we want to turn it into meaning. The bounded entities
that enter into constituent structures with specific functions like Pro-
cess, Actor, Goal, Extent or Manner offer a presentation of reality in
terms of “things” – doings by, and happenings to, persons and objects,
in the environment of other persons and objects, with yet other persons
and objects, and also times and places and so on, as attendant circum-
stances; and including various “metathings” (facts and reports), which
are complex things that have already been encoded in language and so
acquired a status which enables them to participate in certain types of
204
process as objects in their own right. Such elements naturally form

constituent-like structures which allow us to isolate them and continue
to refer to them as discrete entities.
5 The interpersonal mode

Now let us consider interpersonal meanings. These are expressed
through very different structural devices.
Think of an example such as the following, uttered when somebody
is sick: I wonder if perhaps it might be measles, might it d’you think? The
experiential meaning is: Attribute “measles” plus Time “now”. With
this the speaker has combined an interpersonal meaning of “I consider
it possible”, together with an invitation to the hearer to confirm the
assessment.
This interpersonal meaning, however, is strung throughout the
clause as a continuous motif or colouring. It appears as I wonder, perhaps,
might, might and d’you think; each of these expresses the same modality,
and each one could occur by itself. When they all occur, the effect is
cumulative; with each one the speaker reaffirms his own angle on the
proposition.
The intonation contour is another mode of the realization of
interpersonal meanings. It expresses the “key”, the particular tone of
assertion, query, hesitation, doubt, reservation, forcefulness, 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 mapped
on to the clause is a whole, running through from beginning to end.
We shall refer to this mode of realization as “prosodic”, since the
meaning is distributed like a prosody throughout a continuous stretch
of discourse (cf. Mitchell 1958). It is characteristic of interpersonal
meanings that they are expressed in this prosodic fashion. Mood and
modality, tone and key, intensity and other attitudinal meanings are
typically realized through this kind of structural pattern. Swearwords
and obscenities, also, may occur at any or all points in the clause; it
does not matter what segments they are attached to – many writers
have noted that such elements readily occur even in the middle of a
word. The speaker who says
Christ they beat the hell out of those bastards
is in fact using a very regular and well-established resource for the
expression of meanings of this kind.
205
It is not difficult to see the rationale behind this mode of realization.

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

The “textual”, or text-forming, resources in the semantic system
generate structures of still another kind. Consider the example:
Why did you let the big one get away?
This clause has, in effect, two points of prominence. It is a 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 associated
with first position in the clause; in fact it is realized by first position,
since putting something first is what gives it the status of a Theme.
But there is also another point of prominence here, at the end. To
be aware of it we have to consider the clause in its spoken mode, since
this takes the form of tonic prominence: that is, the location of the
tonic accent, which is the dynamic centre of the pitch contour, the
206
place where the greatest pitch movement takes place. (This may be a
falling movement or a rising movement or some kind of complex
movement, depending on which kind of melody it is. It corresponds
to what is sometimes called “primary stress”, although it is not, in fact,
a stress feature.) Suppose we represent the intonation unit (the tone
group) as bounded by double slash, and intermediate rhythmic units
(the feet) by single slash, with tonic prominence as bold type; the likely
form of the utterance would be:
// why did you / let the / big one / get a/way //
The meaning of tonic prominence is the focus of information; it

signals the climax of what is new in the message. This kind of focal
prominence can be assigned at any point in the clause; it is not realized
by final position, in the way that thematic prominence is realized by
initial position. But it is typically located at the end, and any other
focus is “marked” and so explicitly contrastive. In the typical form of
the message, in other words, the speaker puts what is new at the end.
So there is a peak of prominence at the beginning, which is the
Theme; and another peak of prominence, usually at the end, which is
the focus of information or, simply, the New. The two are different in
meaning. The Theme is speaker-oriented; it is the speaker’s signal of
concern, what it is that he is on about – he may even make this
explicit, by starting ‘as far as . . . is concerned’. The New is hearer-
oriented (though still, of course, selected by the speaker); it is the
speaker’s presentation of information as in part already recoverable to
the hearer (the Given) and in part not recoverable (the New). These
two types of prominence are independent of each other. But both
contribute to the “texture”, to fashioning the fabric of the text.
What these text-forming systems do is to organize discourse into a
succession of message units, quanta of information such that each has
its own internal texture, provided by the two systems of prominence
just mentioned. The message unit corresponds, typically (i.e. in the
unmarked case), to a clause. Hence it is possible in such instances to
represent both thematic and focal prominence as constituent-like
structures of the clause, by recognizing the functional significance of
the non-prominent part. So Theme contrasts with Rheme, and New
contrasts with Given, as in Figure 9. In fact, the information unit is not
always coextensive with the clause; to return to the Walrus and the
Carpenter, in The moon was shining sulkily (following the earlier The sun
was shining on the sea, shining with all his might)
207
Figure 9
we should presumably have the structure shown in Figure 10. But
because a clause is typically one information (focal) unit, and the focus
in the information unit is typically at the end, final position in the
clause carries a potential for prominence which highlights it in the
same way that initial position is highlighted by prominence of a
thematic kind.
Figure 10
The structures that realize options in the textual component are
what we may call “culminative” structures. They are not 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 the
textual component does is to express the particular semantic status of
elements in the discourse by assigning them to the boundaries; this
gives special significance to “coming first” and “coming last”, and so
marks off units of the message as extending from one peak of 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, thematic
prominence, 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 the
prominence is also in part phonological – and can be heard. The
periodicity is further reinforced by the use of conjunctives to link one
sentence with another; these contribute to the texture by relating a
clause cohesively to what went before it, and they also occur at the
boundaries – usually the beginning, but sometimes, especially in casual
speech, at the end.
208
Perhaps the clearest instance of the periodicity of texture is to be

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

Figure 11 summarizes the three types of structure we have recognized
so far. It is important to stress that when we associate each of these
structural types with one of the functional semantic components, we
are talking of a tendency not a rule. Experiential structures tend to be
more elemental in character, interpersonal structures tend to be 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 that
each should take this form – may differ very considerably from one
language to another.
Figure 11
Given structures of these very general kinds, it is clear that each can
be reduced to some form of constituency; but not all with the same
success. Experiential structures are quite constituent-like; whereas 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
209
a form of reductionism. An example of this is the attempt to reduce an

intonation contour to a sequence of pitch phonemes (which are then
attached to specific places in a string).
Figures 12 and 13 represent an English clause first in non-constituency
terms and then in constituency terms. The clause is:
On Sunday perhaps we’ll take the children to the circus, shall we?
If we consider the major traditions in linguistic thought, we find,
not at all surprisingly, perhaps, that those in the psycho-philosophical
tradition, who are firmly committed to language as an ideational
system, have usually worked with constituency models of structure:
Notes: (1) experiential : clause as representation (of process); (2) inter-

personal : clause as interaction; (3) textual : clause as message
Figure 12
210
Figure 13
American structuralist and transformationalist theories, for example. By

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

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

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

That you have wronged me doth appear in this
Another example of rankshift is a restrictive relative clause, for example
the lights went out in:
The day the lights went out
These are not true recursive structures. The recursion-like effect that
is produced is an incidental outcome of the selection, at a particular
place in structure, of an item from the same rank or from a higher rank
in the constituent hierarchy. Clearly, this effect may appear more than
once, as in a “house that Jack built” routine; but it is strictly a non-
event – there is no function involved that we could identify as a
recursive function.
True recursion arises when there is a recursive option in the
network, of the form shown in Figure 14, where A:x,y,z . . . n is any
system and B is the option ‘stop or go round again’. This I have called
“linear recursion”; it generates lineally recursive structures of the form
a1 + a2 + a3 + . . . (not necessarily sequential):
Figure 14
These are of two kinds: paratactic and hypotactic. The paratactic
involve relations like “and” and “equals”, which are logically transitive
(A ‘&’ B, and B ‘&’ C, implies A ‘&’ C; A ‘=’ B, and B ‘=’ C, implies
A ‘=’ C). The hypotactic are logically non-transitive; these include
relations such as “if” and “says”, where a “if” ß and ß “if” y does not
imply a “if” y, nor does a “says” ß and ß “says” y imply a “says” y.
In paratactic structures, because they are transitive, recursive order
is expressed by structural (linear) sequence: there is no other way in
which it can be expressed – no way, that is, in which a sequence A
and B and C could realize an order “A & C & B”. The only departure
from this strict sequence is nesting, where a sequence A and B and C
and D represents an order, say, “A & (B & C) & D”. These are
typically signalled in the phonological structure, e.g.:
213
// 13 soup // 3ˆ/ ham and / eggs // 3^/ apple/pie // 1^ and / tea

or / coffee //
i.e. A ‘and’ (B ‘and’ C) ‘and’ D ‘and’ (E ‘and’ F)
In hypotactic structures there is no such restriction; recursive order is

not always signalled by the sequence, which may reverse it ( ß before
a) or modify it in other ways ( ß inside a, for example).
In English the principal instances of ‘recursive structures’ (i.e. struc-
tures deriving from recursive options) are as set out in Table 1.
We should distinguish structures of this kind from those which
happen to be realized as lists, like counting. Counting 1 2 3 4 5 6 . . .
is not a recursive operation, linguistically; it is simply the enumeration
of a list, like Sunday Monday Tuesday Wednesday Thursday Friday
Saturday, but one which happens to be endless because there is a (non-
linguistic) recursive mechanism for generating the items in the list. In
Table 1
214
the same way, the days of the week do not form a closed system, in
the linguistic sense; they form a lexical set whose members happen to
be limited and fixed. In both instances, the infiniteness of the set of
natural numbers and the finiteness of the set of days of the week are
properties not of the language but of the social system.
Logical structures are different in kind from all the other three. In
the terms of systemic theory, where the other types of structure –
particulate (elemental), prosodic and periodic – generate simplexes
(clauses, groups, words, information units), logical structures generate
complexes (clause complexes, group complexes, etc.) (Huddleston
1965). The apparent exception is the sentence, which is generated by
logical structures; but this is merely a terminological exception – the
sentence is, in fact, simply a name for the clause complex. While the
point of origin of a non-recursive structure is a particular rank – each
one is a structure “of” the clause, or of the group, etc. – recursive
structures are in principle rank-free: coordination, apposition, subcate-
gorization are possible at all ranks. The more restricted ones, like tense
and report, are also the ones that are nearer the borderline; they are
only just logical structures. Tense is particularly interesting because it
has only come into this category within the last two to three centuries,
and English appears to be unique in treating tense in this way.
9 Postscript
As a postscript, it should be noted that what is conceptually one and
the same kind of relation may be coded in the semantic system in more
than one way, i.e. may be realized through more than one of the
functional components. The “and” relation, for example, may be coded
in a logical system, expressed as coordination; or in a textual system,
expressed as conjunction. The same is true of “yet”, “so” and “then”;
these are much more weakly represented in the logical mode, but on
the other hand they can also be coded experientially. Table 2 shows
different codings of the temporal “then” relation.
Although the distinct types of structure discussed in this paper appear
at all ranks throughout the grammar, in English at least it is at the rank
of the clause that they are most clearly in evidence.
As an experiential construct, the clause is the locus of transitivity: it
is the representation of the processes, participants and circumstances
that constitute our experience of the real world. As an interpersonal
construct it is the locus of mood and modality: the speaker’s adoption
and assignment of speech roles and his judgement of the validity of the
215
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 the
speaker’s concern and his presentation of what is “news”.
The clause, therefore, is a multiply structured concept; it is clause as
representation, clause as interaction and clause as message. And each of
these provides its characteristic contribution to the total, characteristic
in terms of the kinds of structure we have been talking about. The
clause is orchestrated as melody (the experiential component, 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 is
structurally neutral, not based on the concept of constituent structure
as the norm to which all grammatical patterns are expected to conform.
In a systemic interpretation, language is treated as a resource, a
potential; the various kinds of structure are the different means of
expression of this potential. “Structure” is then no longer the basic
organizing concept; instead, structural representations are derived from
a more abstract conceptual framework that is paradigmatic rather than
syntagmatic, according to which the semantic system of a language is
comprised of sets of options in meaning, each of which makes some
contribution to the expression in its final shape.
10 Summary
If we study the semantic system of a language we find that it consists
of three major functional components: an ideational, an interpersonal
and a textual; with the ideational further subdivided into an experiential
and a logical.
This pattern appears clearly at the semantic level itself: within each
component, the networks of systemic options are closely inter-
connected, whereas between one component and another there are
216
relatively few connections. In other words, choices made in one

component affect other choices within the same component but hardly
at all affect the choices in other components.
It also appears clearly at a higher level, in the relation between
language and situation. Broadly speaking, ideational meanings reflect
the field of social action, interpersonal meanings reflect the tenor of
social relationships and textual meanings reflect the mode of operation
of the language within the situation.
But it is at the lower level, that is in their grammatical realization,
that these functional components are made manifest in the linguistic
structure. In English, experiential options tend to generate constituent-
like structures, actually constellations of elements such as can be fairly
easily represented in constituency terms. Interpersonal options generate
prosodic structures, extending over long stretches (for example inton-
ation contours), which are much less constituent-like. Textual options
generate culminative structures, elements occurring at the boundaries
of significant units, and give a kind of periodicity to the text, which is
part of what we recognize as “texture”. Logical options generate
recursive structures, paratactic and hypotactic, which differ from all the
other three in that they generate complexes – clause complex, group
complex, word complex – and not simple units.
Systemic theory takes the system, not the structure, as the basis of
the description of a language, and so is able to show how these types
of structure function as alternative modes of the realization of systemic
options. They are then mapped on to each other to form the syntagm,
which is the “output” of the lexicogrammatical system.
Notes
1. Written language may be (i) an alternative coding of meanings (ideograms);
(ii) an alternative coding of wordings (characters, as in Chinese); or (iii) an
alternative coding of sounds (syllabaries and alphabets).
2. Mathesius’s observation relates to ‘functional styles’ (orientation towards
different functions in the use of language), not to functionally derived
components of the system; but it is pertinent nevertheless: ‘The influence
of functional styles on the lexical and semantic aspects of speech was
stressed especially by Gröber, . . . [who] distinguishes the subjective
expression . . . and the objective expression . . . The subjective expression
differs from the objective both quantitatively (inasmuch as it expresses by a
pause, by tone or by gesture what the latter expresses by words; and
further, as it repeats what could be expressed only once) and qualitatively (by
217
choosing other words than factual names of the things referred to), and,
finally, locally (by placing sentence elements into positions not pertaining to them
in objective speech). Both ways of expression are often combined in actual
speech.’ (My italics throughout.)
218
Chapter Nine
TEXT SEMANTICS AND CLAUSE GRAMMAR:

HOW IS A TEXT LIKE A CLAUSE? (1981)
1 Patterns of wording in the clause

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

‘Text semantics and clause grammar: some patterns of realization’, first published in The
Seventh LACUS Forum 1980, 1981, edited by James E. Copeland and Philip W. Davies.
Linguistic Association of Canada and the United States, pp. 31–59.
‘How is a text like a clause?’, first published in Text Processing: Text Analysis and Generation,
Text Typology and Attrition (Proceedings of Nobel Symposium 51), 1982, edited by Sture Allen.
Stockholm: Almqvist and Wiksell, pp. 209–47.
219
1961), in which he discussed the different statuses of morpheme and

phoneme and contrasted the two possible pathways between them.
A clause, that is to say, is not only bigger than a syllable; it is an entity
of a different kind, at another level of abstraction. And it is this second
relationship, that of realization, or coding, that is the critical one; the
size distinction is typically associated with it, but not obligatorily – there
can be a clause encoded as a single syllable. It took a surprisingly long
time to clarify this two-dimensional relationship, and to accept that the
relation of syllable to clause (or of phoneme to morpheme) was not
simply one of part to whole, although it should have been fairly obvious
seeing that the two are separated by the Saussurean line of arbitrariness.
It is much harder to establish that a similar shift along two dimensions
separates the clause from the text. A text is likewise – typically but not
necessarily – bigger than a clause. But it is also, and more importantly,
of a different level of abstraction. A text is a semantic entity rather than
a formal, lexicogrammatical one; and this distinction is less easy to draw,
because between the semantics and the grammar there is no such line of
arbitrariness. (I shall return to this point below.)
There is a problem in discussing text, if only because a text can be
such a large object: every example takes up a great deal of space. One
solution to this is to write rules for generating text but never actually
to generate any. Another is that once suggested by Peter Wexler when
he proposed to introduce a talk with the words “This paper is about
the language of this paper”, so making the same entity serve both as
text and as metatext. I have usually approached the problem in a
different way, by using text that is so familiar that the audience can
supply the missing parts for themselves, like Mother Goose or Alice in
Wonderland. We need to face up to it somehow or other; there is
something discouraging about a publication where the author is insist-
ing on the importance of context but cites nothing longer than a
decontextualized clause.
When text is discussed in this way, with reference made only to
isolated clauses, it is perhaps being assumed that the relation of clause
to text is simply one of constituency. If a text is the same kind of thing
as a clause only bigger, we can reasonably use clauses as instances while
making observations about text. And perhaps this approach to text in
turn reflects a presemantic view of language, in which it is assumed
that the linguistic system is no more than grammar and phonology; so
a text must be a grammatical unit, something that consists of clauses in
the same way that a clause consists of words.
Since there is no line of arbitrariness between semantics and gram-
220
text semantics and clause grammar
mar, this view is plausible. It is natural to think of text in the sense of

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

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

of structure would be “filled” by classes of the clause (perhaps with some
intermediate units) in the same way that elements of structure of the
clause are filled by classes of the group. But it is difficult to specify text
structures in a way which represents the text simply as a higher-rank
grammatical constituent; the configurations are different in kind, and
the relationship to the wording is both indirect and complex. Functional
elements of text structure are not translatable into strings of clauses.
A text is therefore not “like” a clause in the way that a clause is like
a word or a syllable like a phoneme. But by the same token, just
because clause and text differ on two dimensions, both rank (size level)
and exponence (stratal level), there can exist between them a relation
of another kind: an analogic or metaphorical similarity. A clause stands
as a kind of metaphor for a text. In this paper I shall refer to some
well-known properties of a text, and then, drawing on some recent
text-linguistic studies in a systemic-functional framework, try to show
that these are paralleled in significant ways by properties of a clause
that are in some sense (not always the same sense) analogous. The
textual properties to be considered are the following:
1. 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 structure
of a text as a configuration of functions (Hasan 1979). A generalized
structural representation is likely to include some elements that are
obligatory and others that are optional; and the sequence in which the
elements occur is likely to be partly determined and partly variable.
Most of the actual formulations of text structure that have been put
forward seem to relate to one broad genre, that of narrative. The
original source of inspiration for these was Propp’s theory of the folk
tale. The structure of traditional oral narrative has been investigated in
detail within tagmemic and stratificational frameworks, on foundations
provided by Longacre and Gleason. A well-known representation of
another kind of narrative is Labov and Waletsky’s structural formula
for narratives of personal experience:
222
Abstract ^ Orientation ^ Complication • Evaluation • Resolution ^

Coda
Outside narrative registers, Mitchell (1957) set up structures for the
“language of buying and selling” in Cyrenaican Arabic, recognizing
three subvarieties having some common and some variant features:
Market and shop transactions: Salutation ^ Enquiry as to object of
sale ^ Investigation of object of sale ^ Bidding ^ Conclusion
Auctions: Opening ^ Investigation of object of sale ^ Bidding ^
Conclusion
Mitchell refers to these as “stages” and comments that “stage is an
abstract category and the numbering of stages does not necessarily
imply sequence in time”. Hasan considers that structure is a property
of texts in all registers. For any register, specified at any appropriate
degree of delicacy, it should be possible to state a generalized structure
by reference to which any actual text can be interpreted. Her suggested
formula for a particular class of transactions, retail sale in a personal
service food store, is as follows:
((Greeting •) Sale initiation ^) ((Sale inquiry •) (Sale request ^ Sale
compliance) ^) Sale ^ Purchase ^ Purchase closure (^ Finis)
Martin (1980), who uses “functional tenor”, or rhetorical purpose, as
the superordinate concept for characterizing registers, gives the 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 to
be defective in some way is that they do not “hang together”: they
“lack coherence”. A text has coherence; it forms a unity, a whole that
is more than the sum of its parts.
Coherence is a complex property to which many factors contribute.
One way to approach it is through the category of cohesion, as defined
by Halliday and Hasan (1976). Cohesion is a necessary but not a
sufficient condition of coherence. The different types of cohesive
relation are the fundamental resources out of which coherence is built.
But the mere presence of cohesive ties is not by itself a guarantee of a
223
coherent texture. These resources have to be organized and deployed

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

text by the semantic relations of ‘and, or, nor, viz, yet, so, then’:
additive, including alternative and appositive; adversative; causative,
and termporal. These are described and illustrated, like the other
cohesive systems, with reference to cohesion between adjacent 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 in
Halliday and Hasan, eliminating the category of “adversative” and
grouping “as against” with additive and “contrary to expectation” with
consequential (causal), is expressed in the network in Appendix 2,
pp. 249–51. The category of “implicit”, also not in Halliday and
Hasan, accounts for those instances where the semantic relationship is
present but there is no conjunction or textual (discourse) adjunct
making it explicit. Martin’s analysis of a short passage of dialogue
includes instances of conjunctive relations bridging a number of inter-
mediate turns (Appendix 2).
2. Conjunctive relations may be set up between passages of any
extent. Not only a turn in a dialogue, but an episode, argument, scene
or any other functional element may be “picked up” conjunctively in
a succeeding portion of the text. In this way the presence of conjunc-
tive relations creates coherence over extended passages of discourse. An
example of this would be a section beginning Because of this, where this
refers to the whole of some preceding argument.
3. Halliday has suggested (1975) that the “textual” properties of
a text – the cohesive patterns and those of ‘functional sentence
perspective’ – tend to be determined by the “mode”, the function
ascribed to the text in the given context of situation, the purpose it is
intended to achieve. Thus the mode would determine the balance
among the different types of cohesive resource – reference, ellipsis,
conjunction, lexical cohesion; and within conjunction, the relative
weight accorded to internal and external conjunctive relations and to
the various semantic alternatives within each. In this way the kind of
conjunctive relations found in the text will be characteristic of the
register (as defined on the dimension of ‘mode’) to which the text
belongs.
An illustration of this principle is provided by Mary Ann Eiler (1979)
in her study of expository writing by ninth graders in an American
225
high school. She has shown how the conjunctive relation of specific
instance to general principle, coded in Martin’s network as:
(internal / comparative) : similarity : nonexhaustive
is the major conjunctive factor giving coherence to these texts. For
example, in one part of a text there occurs the sentence
Odysseus’ friends and anyone else who heard his story respected him.
Elsewhere in the same text we find:
Heroic men are very much respected and idolized.
(Either member of such a pair may come first.) There is lexical cohesion
between individual items – the repetition of respect. There is cohesion
between Odysseus and heroic (men), with ‘Odysseus’ being a hyponym
of ‘hero’. But between the two sentences as wholes there is a conjunct-
ive relation – itself an extension of hyponymy – such that the second
one stands as a general principle of which the first one offers a specific
instance; and this type of conjunction is a distinguishing feature of the
sort of expository discourse she is investigating.
1.1.3 A text has function

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

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

A text is a dynamic process; it has a semantic ‘flow’, a development of
ideational and interpersonal meanings. This flow or development is
carried forward by the interaction of speaker and listener; obviously so
in the case of dialogue, but so also in monologic modes where the active
participation of a listener still contributes to the construction of meaning.
Even in written language the semogenic process is essentially of the same
nature; researchers in writing theory now strongly insist on the part
played by the imagined audience in the process of written composition.
227
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 to
the concepts of theme and rheme. His argument proceeds in four steps:
(1) the pattern of theme–rheme organization in the clause is a function
of the register; (2) the pattern in the choice of theme is a function of the
method of development of the text; (3) “theme–rheme” is clearly distinct
from, but also clearly related to, “given–new”; (4) the theme–rheme
organization of the clause “fits into a larger pattern governing the informa-
tion flow in sequences of sentences in English discourse in general”.
Fries regards the theme as a “ground”: “In English discourse at least
there seems to be a strong tendency to set up certain information as a
ground first, and then to introduce later information using that ground
as a basis for evaluation and comparison”. (It is reasonable to understand
“information” rather broadly here; presumably the ground may be any
configuration of ideational and interpersonal meanings.) So there is a
movement from the ground, to something that is defined by it as “not
ground”. There is also, we may add, a complementary movement in a
text, which is a movement towards rather than away from: a movement
to what we may call the “point” (generalizing from Fries’ “main
point”), from something that is defined by it as “not point”. Rhetorical
theory has always stressed the beginning and the end: topic sentence,
introductory paragraph &c. on the one hand, and culmination, climax,
summation &c. on the other. But it is important to stress that this is
not a single movement. A text is a kind of diminuendo – crescendo,
beginning and ending with prominence; but the prominence is of two
different kinds. Rather than thinking this time of a flow, a unidirec-
tional current with a set of rapids at each end, we should perhaps
change the metaphor to that of a gift, or rather an exchange, in which
there is a shift of focus from donor to recipient in the course of the
exchange, or rather from giving to receiving. The process begins as
giving and ends as receiving; but “giving – not giving” is not the same
movement as “not receiving – receiving”. Moreover although the
process must start with the giving – until then there is no exchange –
it need not necessarily end with the receiving, which may occur quite
early and be followed up in various ways.
In the development of a text, phasing out the “ground” goes along
with phasing in the “point”. This pattern is one that can be repeated
over as many levels in the hierarchy of constituents as the text has in it,
from the entire text down to the individual clause.
228
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 a
specific character of its own distinct from that of other texts within the
same register. Some texts are highly valued as individual texts, and one
of the interests of text studies is the stylistic one, the attempt to
understand the unique qualities of a highly valued text, and what it is
that makes it highly valued.
(A) The generic character of a text is in principle predictable from
its context of situation. Taking the categories of field, tenor and mode
as a predictive framework, Halliday proposed that the ideational mean-
ings of a text tend to be determined by the field, the interpersonal
meanings by the tenor, and the textual meanings by the mode,
suggesting that this was how listeners and readers make predictions
about what is coming next – predictions that they must make if
meanings are to be successfully exchanged. For illustrations of this see
Halliday (1975, 1977), Halliday and Hasan (1980).
An example of a register variable is provided by Jean Ure’s study of
lexical density (1971). Ure shows that the lexical density of a text is a
function of its level of formality, the amount of self-monitoring done
by the speaker or writer; writing has a higher density than speech, with
what she calls “language-in-action” having the lowest density of all.
Lexical density can be defined as the number of lexical items per unit
grammar (per clause, as the most natural measure), though Ure 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 values
range from 57 per cent (formal written) to 24 per cent (casual spoken),
and all texts with a density of 35 per cent and below are dialogue.
Charles Taylor (1979) has used both these measures in his study of the
language of high school textbooks in New South Wales.
Robin Melrose (1979) has suggested another variable that defines
the generic character of a text, one that relates to the field instead of
the mode. He finds that each text will tend to be characterized by a
particular “message type”. Melrose distinguishes factual, phenomenal
and relational message, with various subcategories; deriving these from
the material, mental and relational processes of the transitivity system
(Halliday 1967–68; 1975). An instance of a text with relational :
attributive messages is given in Appendix 4, p. 255–9.
229
According to Melrose, certain other features are associated with

these message types: different patterns of theme, conjunction and lexical
cohesion, and also different kinds of “message superordinate”, the
summative expression described by Winter (1977); the table in 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 predictable
from the contextual configuration. This has sometimes been character-
ized as deviation, the text creating its own rules; but “breaking the
rules” is a minor and relatively insignificant form of uniqueness. What
a text does is to create its own norms, its own unique selection from
the resources of the system by which it is generated.
Many texts in daily life are not unique at all; the same things have
been said countless times before. Such texts are often of particular
interest to an ethnographer (and, one might add, to a linguist). Other
texts are presumed to be unique; this class includes all those texts we
think of as literature. But any text can be described and interpreted as
an event that is sui generis. If the qualities that we perceive as specific to
a text reside not merely in the particular combination of features
selected but also in a special highlighting of some aspect of these, we
usually try to relate this highlighting to our interpretation of the
underlying theme, seeking the kind of semiotic convergence that
would explain the particular impact that the text has on us.
A text is a polyphonic composition of ideational, interpersonal and
textual “voices”. The ideational voice provides the content: the things,
facts and reports; processes, participants and circumstances; the logical
relations of different kinds. The interpersonal voice provides the
interaction: mood, modality, person, polarity, attitude, comment, key.
The textual voice provides the organization: thematic and informational
prominence; grammatical and lexical cohesion among the parts. The
“character” of the text is its pattern of selections in these various voices,
and the way they are combined into a single whole.
The accompanying extract from J. B. Priestley’s An Inspector Calls
(Appendix 5, pp. 257–9) is a piece of dramatic dialogue which is dis-
tinctively characterized by the foregrounding of modality. The first
part is dominated by modalized assertions, which move from probabil-
ity to obligation, the second part by assertions about obligations; the
final speech shifts into narrative, returning at the end to the assertive
mode but this time without modalities. Herein lies the movement of
the play, which is concerned with social responsibility (obligation)
acted out through chance (objective probability); the interplay of these
230
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 lexicogrammatical
object, a structure of wording; whereas a text is a semantic object, a
structure of meaning. The resemblance is like that of clause to syllable.
A syllable is a phonological object, and therefore not part of a clause;
but it has structure in the same sense.
With one difference, however: between clause and syllable runs the
line of arbitrariness in language. In the realization of wording in sound,
natural symbols are the exception. But there is no such line of arbitrar-
iness separating the clause from the text. The realization of meaning in
wording is largely “natural”, non-arbitrary. This leads us to speculate
whether the text may display the same kind of multiple structuring that
is found in the clause, ideational, interpersonal and textual. The repres-
entations of text structure proposed by Hasan and others suggest that
we might want to interpret a text as having, potentially at least, an
ideational structure relating to its field and an interpersonal structure
relating to its tenor, rather than (or as well as) a single structure deriving
from the mode (functional tenor) as proposed by Martin.
Benjamin N. Colby and Lore M. Colby (1980) analyse traditional
and other oral narratives in terms of “eidons”, which are ideational
elements of text structure set up to allow for the interpretation of the
text as an ethnographic document, as a window on the culture. The
theory and method are set out in Colby’s study (1973) of Eskimo folk
tales. The notion of the eidon Colby ascribes to Gregory Bateson’s
interpretation (1936) of Iatmul culture and the Naven ceremonies.
Bateson talks of the “eidos” and the “ethos” of a culture, and of
“eidological” and “ethological relationships” – the realization of eidos
and ethos in cognitive and affective aspects of cultural behavior. Colby’s
interpretation of text structure is “eidological”, corresponding to the
ideational component in the structure of the clause; he suggests the
possibility of a parallel “ethological” interpretation corresponding to
the interpersonal component in the structure of the clause.
231
We should not press the analogy too far. But if it seemed useful to
set up simultaneous structures in a text along these lines we might ask
whether there is the same kind of structural variation as we find in the
clause, with the eidological structure being “particulate” (represented
by definable segments of the text) and the ethological being “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 its
structure, whereas coherence in a text largely depends on cohesion.
Cohesion is the resource that takes over, as it were, when grammatical
structure no longer holds (i.e. above the clause complex). We could
point out that cohesion also obtains within clauses; we find reference,
substitution, ellipsis, conjunction and lexical cohesion all operating
between elements in the same clause, for example:
M’s evening speech caused more fuss than his morning one had.
C R L S E
But this is a superficial similarity. A more significant analogy can be

found with the notions of cohesive harmony and conjunctive relations
discussed in 1.1.2 above.
Ruqaiya Hasan’s work showed how lexico-referential motifs enter
into a text not as isolated motifs but as interlocking chains having some
kind of regular functional relationship with each other. But these
functional relationships are relationships within the clause; and this
reflects the fact that the elements in these chains themselves cannot
occur as isolated entities. Names have no place in language except in
function with other names; and the functions are defined within the
structure of the clause.
The conjunctive relations discussed by J. R. Martin (1992) are also
derived from relations with the clause. Consider a series such as the
following:
She didn’t know the rules. She died.

She didn’t know the rules. So she died.
She died, because she didn’t know the rules.
She died because of not knowing the rules.
That she died was because she didn’t know the rules.
232
That she died was caused by her not knowing the rules.
Her ignorance of the rules caused her death.
The same conjunctive relation, the “external causal”, can be coded in

very many ways. It can appear as a relationship within the clause,
realized lexicogrammatically; but it can also serve to link segments of
the text, at any distance and of any extent. The kind of coherence that
is achieved by the presence in the text of semantic relationships of the
conjunctive kind is essentially a clause-type coherence, one that is
based on relations defined systematically within the transitivity system
of the clause.
The notion that a text has function is again closely related to an
analogous feature of the clause, also one that is coded in the lexico-
grammatical system: that a clause has a speech function, realized by the
mood system. The speech function of the clause – in simplest terms, as
statement, question, command, offer, or a minor speech function – is
represented by the grammatical categories of declarative, interrogative
and so on; this is the rhetorical function of the clause, and the whole
range of rhetorical functions that we assign to text are simply the
“mood” of the text. (Cf. Martin (1980) for speech-functional analysis
of dialogue.)
The development of the text again has its analogy in the clause. This
has already been made clear from the example cited in 1.1.4, since
Peter Fries used the theme–rheme structure of the clause as the source
from which to derive the method of development of the text. We can
generalize this still further by bringing in the notion of information
structure, the given–new movement within the clause.
In its “textual” aspect, a clause has a wave-like periodic structure
created by the tension between theme–rheme (where theme is the
prominent element) and given–new (where new is the prominent
element); the result is a pattern of diminuendo–crescendo, with a peak
of prominence at each end. There is a balance of development (i) away
from the theme, and (ii) towards the new. But these are separate
movements. They are in phase in the unmarked, “default” case, where
the theme is selected from what is given, and the news is put into the
rheme. But they can also be out of phase, and this gives an alternative
pattern of texture to the clause. Putting the two out of phase means
locating the new (the focus of information) somewhere other than at
the end of the rheme; this as it were changes the wave shape but does
not disturb the essential periodicity.
This pattern is the “method of development” of the clause. It is
233
closely analogous to what takes place in a text; not only over the whole
text but also in structurally defined intermediate units within the text.
The classic movement of a paragraph, beginning with a topic sentence
(from theme to elaboration) and culmination – having a high point,
unmarkedly but not obligatorily final – in a climax (from prelude to
main point), is one of the clearest manifestations of the analogy between
clause and text. It is in the clause that this movement is displayed in
the most systematic and clearly motivated form.
Finally a clause can be said to have “character” in both the generic
and the specific sense. If a text is typified by virtue of its being
organized around the expression of processes of a particular type, the
clause is the unit in which these processes are realized and categorized.
The clause is the locus of the transitivity system: the system of material,
mental and relational processes, together with their numerous subcat-
egories. Thus analogous to the major types or genres of text are the
major types or classes of the clause, each being characterized by the
selection of a dominant process type.
But each clause is also a unique combination, or potentially unique
combination, of features deriving from the different semantic functions,
ideational, interpersonal and textual. Moreover any one or other of
these may be foregrounded: the clause may display an orientation
towards any one, or any combination, of the various systems and their
subsystems. The extract from J. B. Priestley’s An Inspector Calls
(Appendix 5, pp. 257–9) gave an instance of a clause that is oriented
towards a certain type of modality: interpersonal meanings are high-
lighted, with the speaker’s skeptical doubting as the predominant
rhetorical function. The passage cited is a unique interplay between the
exploration of probabilities and the assertion of obligations, and so is
the entire text. No one clause can recapitulate the whole; but all
contribute, and some achieve a remarkable likeness – a likeness that is
possible because the systems of the clause not only embody all the
semantic components from which the text is built but do so in a way
that allows an infinitely varied, almost text-like balance among them.
Thus the properties that we recognize in a text are also, in a
transformed way, properties that we recognize in a clause. A clause is a
kind of metaphor for a text – and a text for a clause. That this is
possible is due to two things: one, that a text is not only (typically)
larger than but also more abstract than a clause; two, that on the other
hand there is no line of arbitrariness between clause and text as there is
between clause and syllable. Hence it is not only in a formal sense that
a text is like a clause. It is no accident that it is possible to illustrate so
234
many of the relations in a text by reference to relations in a clause. The

illustrations 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 the
same time showing that a clause has all these things too, though in an
interesting variety of different ways. Presumably this is how clauses
evolved – as the most efficient means of encoding text.
SUMMARY
TEXT CLAUSE
Structure: configuration of configuration of
contextually motivated semantically grammatical
semantic functions functions
Coherence: by cohesion by structure
(i) cohesive harmony: (i) names (things)
chains interrelate by interrelate by function in
function in (semantic) (grammatical) transitivity
transitivity (ii) conjunctive relation:
(ii) conjunctive relation: between parts of clause, as
between messages or major or minor process
larger parts of text
Function: has “functional tenor” has “speech function”
(rhetorical function as (rhetorical function as
text) speech act)
Development: has “information flow”: has “information
ground - - - - > structure”:
{
- - - - > point } theme - - - - >
{ - - - - > focus / new }
Character: generic: selects generic: selects process
“favourite” process type type
as message type specific: foregrounds
specific: foregrounds one selections from one or
or more systems more systems
1.2 A functional interpretation

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

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

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

accustomed to thinking of this in structural terms: that each semantic
component generates its own particular tree, a configuration of parts
each having a distinct function with respect to the organic whole. The
ideational component generates “actor–action–goal”-type structures:
configurations of Process, Medium, Agent, Beneficiary, Range, Extent,
Location, Manner, Cause and so on. The interpersonal component
generates so-called “modal-propositional” structures: configurations of
Subject, Finite, Modality, Predicator, Complement and Adjunct. And
the textual component generates thematic and informational structures,
configurations of Theme and Rheme, and Given and New; as well as
cohesive elements of a non-configurational kind.
We can represent all these in structural terms, using the linguist’s
traditional notion of structure: the simplest of all possible forms of
organization, that of parts into wholes. Because this notion of constitu-
ent structure is so simple, it is natural that a linguist should want to do
as much as possible with it. And it can be made to do quite a lot. But
there comes a point where it ceases to be appropriate; where moulding
the facts so that they fit the notion of constituency will distort them
rather than just simplifying them. With a multifunctional interpretation
of the clause we reach this point.
As outlined in Chapter 8, the contributions that are made by the
three functional-semantic components to the form of the clause are of
three rather different kinds. As far as the ideational systems are con-
cerned, these do tend to generate part–whole structures; they are realized
by organic configurations which themselves, and whose constituents,
are reasonably clearly bounded, such that it can be specified where one
clause element leaves off and the next one starts. But this is not nearly
so true of interpersonal systems. Interpersonal systems tend to generate
prosodic patterns that run all the way through the clause: not only
intonation contours, though these provide a clear instance, but also
reiterations of various kinds like those that are typical of modality in
English, e.g. surely . . . can’t . . . possibly . . . can . . . d’you think in:
Surely they can’t possibly be serious about it can they d’you think ?
Textual systems generate patterns that differ from both of these,
culminative patterns formed by peaks of prominence; and since these
peaks typically appear at the beginning or the end of the clause, where
there is a sequence of clauses they result in a kind of periodicity, a
movement from a clause-initial peak via an off-peak medial state to a
clause-final peak which is then sustained to form the initial peak of the
238
succeeding clause. Thus a clause is at one and the same time particle,
field and wave, as Pike suggested more than twenty years ago (Pike
1959), although the details of this interpretation are not quite the same
as those worked out by Pike.
Now, the significance of this step in our interpretation lies not only
in establishing that these three distinct patterns of realization go to
make up the English clause, but also in the fact that they appear to be
non-arbitrary; this is clearly important when we come to ask whether
such tendencies are found in every language. The grammar of languages
is a natural grammar; as I expressed it earlier, there is no line of
arbitrariness between semantics and grammar as there is between
grammar and phonology. If the clause is at once particle, field and
wave this is because the meanings it has to express have different
semiotic contours, to which these three realizational forms correspond
in a natural, non-arbitrary way.
The particular nature of ideational structures reflects the relative
discreteness of the phenomena of our experience. Consider cows eat
grass: we know where the cow begins and ends, what eating is and is
not, which part of reality consists of grass and which part consists of
other things. Many of our perceptions are schematized into entities that
are bounded in this way, and the constituent-like form of the wording
reflects this fact: the word cow has an outline because the object cow
has an outline. Of course not all experience is like this; indeed I have
always tended to emphasize the unboundedness of many phenomena,
the indeterminacy and the flux; and I share Whorf’s view that language
itself, once it has been constituted in this way, strongly influences the
forms that our perceptions take. Nevertheless there is a basic fit between
the discreteness of words and the discreteness of things; otherwise we
should not be able to talk about the things at all, or explain contrastively
those instances where the fit of words to things is less than perfect.
By contrast, the interpersonal kind of meaning is a motif that runs
throughout the clause; and this is represented by lexicogrammatical or
phonological motifs that are likewise strung unboundedly throughout.
The speaker’s attitudes and assessments; his judgements of validity and
probability; his choice of speech function, the mode of exchange in
dialogue – such things are not discrete elements that belong at some
particular juncture, but semantic features that inform continuous
stretches of discourse. It is natural that they should be realized not
segmentally but prosodically, by structures (if that term is still appropri-
ate) that are not particulate but field-like. The linguist’s tree is an
inappropriate construct for representing structures of this kind.
239
Thirdly, the undulatory movement by which textual meanings are

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

“I shall sit here,” the Footman remarked, “till to-morrow – ”
240
The Footman, just like the rest of us, favours himself as an unmarked
Theme.
So the patterns of wording in the clause, which is the basic unit of
lexicogrammatical organization, display a variation that derives from
the different kinds of meaning they express; and the structural shape is
in each case a natural product of the semantic functions. A functional
grammar is an interpretation of the primary semiotic purposes that
language has evolved to serve, and of the different ways in which
meanings relating to these different purposes tend to be encoded (and
the patterns just described are only tendencies). When we go on to
observe the developmental processes whereby young children construct
their language, we gain a further insight into the steps by which
grammar may have evolved on the way towards its present form.
2 From clause to text

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

recognize functional constituents of a text, always allowing (as in the
clause) for some variation in sequence and a certain amount of
overlapping. These structural elements have been identified most
clearly, perhaps, in narrative; and the researches of Pike, Longacre and
their co-workers on the one hand, and of Gleason and his colleagues
on the other, have provided a rich body of empirical findings about
the structure of narrative in languages and cultures from every 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 each
with its own configuration of functional elements. Ruqaiya Hasan
(1979) considers that this structural organization is a general feature of
texts of all genres; in her studies of transactional discourse she recognizes
optional and obligatory elements, variations in sequence, recursive
options and the like, all of which make the text structure look rather
similar to the ideational structure that is characteristic of the clause. We
can sum this up by saying that, in at least some genres, and perhaps in
all, a text is a configuration of functional elements, collectively 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 be
coded in a great variety of different ways. They appear in many forms
within the clause and even within clause constituents; most typically,
perhaps, they link clauses in a hypotactic or paratactic clause complex.
But they also function as semantic links over longer passages of
discourse. Martin (1992) has interpreted these relations in a generalized
system network and suggested how they may be accounted for as an
aspect of the ideational structure of a text.
2. Interpersonal. A clause has an interpersonal pattern of organiza-
tion, including a modal structure (mood, modality and key) which
expresses its character as a speech event. In the same way a text has a
unified character as a rhetorical event. In a recent study, Melrose (1979)
makes the suggestion that a text or portion of a text derives its character
from the type of process, in the transitivity system, that is predominant
in it: material including action, event, behaviour; mental, including
perception, reaction, cognition; verbal; or relational, including attribu-
tive, equative, existential. This “type of process” is of course an
ideational category; but a clustering of processes of the same kind
expresses the rhetorical design of the text rather in the same way that a
242
particular complex of ideational features such as “I am certain” or “I

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

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

form a text not as isolated chains but as role-chains in an ongoing
configurational movement (Appendix 1, pp. 247–50).
2. Interpersonal. How does one recognize the unique rhetorical
flavour of a text? Partly at least from the overall pattern of interpersonal
features of the individual clauses. A text has its own character as an
intersubjective event, and this tenor of discourse is manifested primarily
through the cumulative force of the options taken up in the interper-
sonal systems of meaning. In Priestley’s play An Inspector Calls, the
underlying theme, or rather one of the underlying themes, is that of
social responsibility: we are all members of one body. This confers
obligations on all of us, each one towards others. In the course of the
play these obligations are acted out – or rather the consequences of
their not being fulfilled are acted out – through the step-by-step
uncovering of a chain of irresponsibility, compounded by sheer chance
and observed through a confusion of prejudices and doubts. Now, the
three conceptual fields of probability, opinion and obligation together
comprise the semantic raw material of the complex system of modality
in the grammar of English. It is not surprising therefore that the
underlying semiotic of the play is worked out metaphorically, at a
critical point in the action, through the highlighting of modal selections
within the clause, backed up by lexical choices from the same semantic
fields. The clauses in this key passage, each with its own small
momentum, combine to produce a powerful semantic movement, a
motif first of chance and then of duty, both hedged around by opinion,
and culminating, after a narrative monologue serving as commentary,
in a burst of direct assertion in which the modalities are finally swept
away. As audience we respond to this movement even though the
events which call it forth are in themselves trivial, no more than an
argument over the identification of a photograph. Here the interper-
sonal features of the clause stand as a metaphor for the social semiotic
of the text, as an exploration of the complex symbolic structures
binding men to their fellow men (Appendix 5, pp. 257–60).
3. Textual. The last two examples suggest that we can ‘read off’
significant aspects of the semiotic quality of a text from looking at the
transitivity and modal features that predominate in the individual
clauses. When we come to consider the rhetorical organization of a
text, this too can be discovered from a reading of the clause patterns,
in this case those having to do with functional sentence perspective:
what are the elements that function predominantly as theme, and what
are the elements that function predominantly as news. In his study of
245
the thematic organization of discourse, Fries (1981) has shown how

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

clause does this effectively because it has itself evolved by analogy with
the text as a model, and can thus represent the meanings of a text in a
rich variety of different ways.
Appendices
Appendix 1
from Hasan (1980).
Text A
1. once upon a time there was a little girl
2. and she went out for a walk
3. and she saw a lovely little teddybear
4. and so she took it home
5. and when she got home she washed it
6. and when she took it to bed with her she cuddled it
7. and she fell straight to sleep
8. and when she got up and ( ) combed it with a little wirebrush
the teddybear opened his eyes
9. and ( ) started to speak to her
10. and she had the teddybear for many many weeks and years
11. and so when the teddybear got dirty she used to wash it
12. and every time she brushed it it used to say some new words
from a different country
13. 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 ship
2. and he’s coming home with a dog
3. and the dog wants the boy and the girl
4. and -----
they don’t know the bear’s in the chair
5. and the bear’s coming to go to sleep in it
6. and -----
they find the bear in the chair
7. they wake him up
-----
8. and (-----) chuck him out the room
9. and (-----) take it to the zoo
10. the sailor takes his hat off
247
11. and the dog’s chased the bear out the room
12. and the boy will sit down in their
----- chair what the bear was
sleeping in
Underlined items are those which enter into lexico-referential chains.
Broken underlining indicates that one item incorporates more than one
token, for example -----
they referring to the girl, the boy, the sailor and
the dog. Empty underlining within parenthesis ( ), (-----) indicates a
token or tokens presupposed by ellipsis.
The number of lexico-referential tokens in the two texts is not very
different: 66 in Text A, 56 in Text B. But whereas 43 of those in Text
A (65 per cent) occur in chain interaction, the comparable figure for
Text B is only 20 (36 per cent). Text A thus displays considerably
greater cohesive harmony.
When subjects were asked to judge the coherence of the two texts,
Text A was consistently rated “more coherent” than Text B.
248
Chain Interaction – Text A
girl 왗 왘
went
home 왗 왘
girl got
home girl
girl 왗 왘
took 왗 왘
teddybear
girl had teddybear
girl took-to-bed
girl 왗 왘
fell-to-sleep
girl got-up
girl washed teddybear

girl 왗 왘
combed teddybear
girl washed 왗 왘
teddybear
girl brushed teddybear
lovely 왗 왘
teddybear
dirty teddybear
speak
왗 왘
teddybear say
teddybear words
English
speak 왗 왘
Scottish
All-the-rest
Chain Interaction – Text B
bear chair
bear chair
왗 왘
go-to-sleep 왗 왘
bear chair
sleep bear chair
chuck 왗 왘
bear 왗 왘
room
chase bear room
go 왗 왘
sailor
come sailor
Each rectangle corresponds to one chain; subdivisions in the chain are indicated by boxing. Each box contains those
items which are in a constant functional relation (shown by a double-headed arrow) to items in some box in a
different chain; for example in Text A, between girl (4) and the box containing washed . . . brushed there is an actor–
action relation; between the latter and teddybear (4), a relation of action–goal.
Appendix 2
from 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 well
11. uh to use a good toner
12. A: Hmm.
13. B: and moisturiser is a must
14. 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).
251
Figure 1 Conjunctive relations in Talking Shop (scene 21)
252
Figure 2
253
Appendix 3
from Peter Fries (1981).
1. A. The English Constitution – that indescribable entity – is a
living thing, growing with the growth of men, and assuming
ever-varying forms in accordance with the subtle and 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 English
gave it one of its essential peculiarities – the system of a
cabinet independent of the Crown and subordinate to
the Prime Minister.
5. C The wisdom of Lord Grey saved it from petrification and
set it upon the path of democracy.
6. C. Then chance intervened once more.
7. D. A female sovereign happened to marry an able and
pertinacious man,
8. D. and it seemed likely that an element which had been
quiescent within it for years – the element of irre-
sponsible administrative power – was about to
become its predominant characteristic and change
completely the direction of its growth.
9. C. But what chance gave chance took away.
10. D. The Consort perished in his prime,
11. D. and the English Constitution, dropping the dead limb
with hardly a tremor, continued its mysteric life as if
he had never been.
Queen Victoria, Lytton Strachey (p. 192)
[The above] is a well constructed paragraph which contains within it

three lexical systems; the first concerns living, growing, changing, the
second system concerns wisdom versus chance and the third system
concerns concepts having to do with government. From reading the
paragraph it is clear that the main point of the paragraph is that the
English constitution is living, growing and changing, that the paragraph
is developed via the opposition between wisdom and chance and that
the lexical system having to do with government plays no particular
role within the structure of the paragraph. On examining the paragraph
one finds that the terms having to do with living, growing and
254
changing typically occur within the rhemes of the component sentences

of the paragraph. The terms having to do with wisdom and chance,
with certain exceptions which can be explained, occur within the
themes of the component sentences. The terms having to do with the
form of government occur more or less equally within the themes and
rhemes of the component sentences of the paragraph. Thus the
consistent placement of the terms of a lexical system inside or outside
the themes of the component sentences of the paragraph affect the
perceived role of that lexical system within the paragraph as a whole.
[Hence] a) the lexical material placed initially within each sentence
of a paragraph (i.e. the themes of each sentence of a paragraph) indicates
the point of departure of the message expressed by that sentence, and
b) the information contained within the themes of all the sentences of
a paragraph creates the method of development of that paragraph. Thus
if the themes of most of the sentences of a paragraph refer to one
semantic field (say location, parts of some object, wisdom vs chance,
etc.) then that semantic field will be perceived as the method of
development of the paragraph. If no common semantic element runs
through the themes of the sentences of a paragraph, then no simple
method of development will be perceived.
Appendix 4
From Robin Melrose (1979).
The remaining eleven sections deal with every aspect of life, regulating
it at every stage and aspect, ordering everything, forcing everything into
a symmetrical pattern: the cities are uniform, married life is strictly
controlled, education is minutely prescribed. Philosophy is confined
within rigid limits, the fine arts somewhat less so (. . .) This planned
paradise is enforced by drastic penal laws. Machinery of government is
paternalistic and pyramidal. It is based on division into families, tribes,
cities and provinces, and, in the case of the different crafts and 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. Subordinate
senates of cities are controlled by the Supreme Senate. At the head of
the state is the General.
Totalitarianism, Leonard Schapiro (pp. 87–8)
255
Theme:
The remaining eleven sections ; the cities ; married life ; education ;
Philosophy ; This planned paradise ; Machinery of government ; It ;
To each unit of work ; Each paterfamilias over fifty ; each family ;
Subordinate senates of cities ; At the head of state
Lexical Cohesion:
Group A :
regulating ; ordering ; forcing ; uniform ; controlled ; prescribed ;
confined ; planned ; enforced ; penal laws ; paternalistic and
pyramidal
This is a particularly clear example of an attributive message. It
begins with a summation, “every aspect of life”, with the general noun
“aspect” acting as Head of a nominal group. It is this summation which
determines the Theme of the clauses that follow: thematic prominence
is assigned precisely to aspects of life, so that there is a relationship of
superordinate to hyponym between summation and Theme in the
message, reinforced by Theme in the last five clauses, which is in a
relationship of hyponym to “machinery of government”, itself an aspect
of life.
There is no explicit conjunction of interest : the chief conjunctive
relationship is an implicit one, of the internal additive type. More
worthy of attention is the lexical string of Group A. Just as Theme was
determined by the summation, so the lexical string of Group A is
determined by the non-finite clauses dependent on the clause of which
the summation is an element. Together with the three verbs in these
non-finite clauses, nine lexical items of the message proper constitute a
string of synonyms, near-synonyms, and collocates – and of these nine,
six function as Complement, and so form a kind of pattern in the
Rheme. Thus it may be seen that in this attributive message the
summation and the clause complex of which it is an element are both
closely related to the message proper that follows: the summation is
hyperonymously linked to the Theme, and the clause complex (or,
more precisely, the non-finite clause) is synonymously connected –
with one exception – to the Rheme. Or, to put it another way, it is
most often the case that “aspects of life” are encoded in the Subject,
while “regulation” is realised in the Complement.
256
Characteristics of message types

Table adapted from Melrose’s Table 4, p. 50.
Factual Phenomenal Relational
Type of material (doing, mental (seeing, relational (being –
Process happening) feeling, thinking); attribute, identity)
verbal (saying)
Characteristic main participant cognizant / sayer synonym or
Theme or phenomenon / hyponym of
discourse summative element
Typical external temporal external or internal internal additive or
Conjunction additive or adversative
temporal
Summative general noun, of general noun, of general noun +
Element which message is which message is expansion, of
meronym hyponym which message is
meronym
Appendix 5
from M. A. K. Halliday (1982).
Mrs. Birling: I think we’ve just about come to an end of this
wretched business.
Gerald: I don’t think so. Excuse me.
(He goes out. They watch him go in silence. We hear
the front door slam.)
Sheila: (to Inspector) You know, you never showed him that
photograph of her.
Inspector: No. It wasn’t necessary. And I thought it better not
to.
Mrs. Birling: You have a photograph of this girl?
Inspector: Yes. I think you’d better look at it.
Mrs. Birling: I don’t see any particular reason why I should –
Inspector: Probably not. But you’d better look at it.
Mrs. Birling: Very well.
(He produces the photograph and she looks hard at
it.)
Inspector: (taking back the photograph) You recognize her?
257
Mrs. Birling: No. Why should I?

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

opinions relating to duties, and, as a minor motif, duties relating to
opinions. The two themes are closely interwoven. We have already
seen that this is a projection into the text of a relation that exists
between them in the system. The scales of “possible–certain” and
“allowed–required” both typically combine with a common semantic
258
feature, that of “subjective”, in the sense of representing the speaker’s

judgment; and this is symbolized by the use of modal verbs as one form
of the realization of both.
The significance of the lexicogrammatical selections in the text can
only be fully revealed by a consideration of their value in the semantic
system. Textually, the passage under discussion centres around the
scrutiny and recognition of a photograph. The words and structures
which, in their automatic function as the “output” of semantic choices,
carry forward the movement of the text, also become de-automatized
and so take on a life of their own as engenderers of meaning.
Example of modalized clause complex

Inspector: Of course she might have changed lately, but I can’t
believe she could have changed so much.
Clause 1 polarity positive
modality low / (indicative : probability) / (subjective :
congruent)
Clause 2 polarity negative : transferred
modality high /
(a) can (imperative : inclination) / (subjective :
congruent)
(b) I . . . not believe (indicative / probability) /
(subjective : explicit)
(c) could (indicative / probability) /
(subjective : congruent)
259
indicative imperative
(MODALITY) (MODULATION)
‘it is’ ‘do!’
probability temporality obligation inclination
modal ‘it either is ‘it both is ‘you do!’ ‘me do!’
or isn’t’ and isn’t’
categorical positive is do
high certain must be always required must do desperate
median probable will be usually supposed will do keen
low possible can be sometimes allowed can do able
categorical 䉲 negative isn’t don’t
high: must ought to need has/had to subjective, congruent: must &c. (modal auxiliary verbs)
median: will would shall should subjective, explicit: I think &c. (mental process clauses)
low: can could may might objective: certain(ly) &c. (modal adverbs, etc.)
Chapter Ten
DIMENSIONS OF DISCOURSE ANALYSIS:

GRAMMAR (1985)
This chapter presents a brief sketch of a lexicogrammatical text 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 is
the text in standard orthography and punctuation (Hasan 1965: 65):
Adult: Do you – when you have a small baby in the house, do you
call it ‘it’, or do you call it ‘she’ or ‘he’?
Elsie: Well if it’s just – if you don’t know what it is I think you
ought to call it ‘it’, because you don’t know whether you’re
calling it a boy or a girl, and if it gets on and if you start
calling it ‘she’ then you find out that it’s a boy you can’t
stop yourself cause you’ve got so used to calling it ‘she’.
Lacey: Em – Mrs. Siddons says that if – if some neighbour has a
new baby next door and you don’t know whether it’s a he
or a she, if you refer to it as ‘it’ well then the neighbour will
be very offended.
Tilly: Well if it’s in your family I think you should call it either
‘he’ or ‘she’ or else the poor thing when it grows up won’t
know what it is.
Adult: Well what did Mrs. Siddons suggest you should do if . . .
your neighbour has a baby and you don’t know whether it’s
a boy or a girl?
First published in The Handbook of Discourse Analysis, Vol. 2: Dimensions of Discourse, 1985.
London: Academic Press, pp. 29–56.
261
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 the
paradigmatic notion of choice. It is built on the work of Saussure,
Malinowski and Firth, Hjelmslev, the Prague school, and the American
anthropological linguists Boas, Sapir, and Whorf; the main inspiration
being J. R. Firth. It is a tristratal construct of semantics (meaning),
lexicogrammar (wording), and phonology (sound). The organizing
concept at each stratum is the paradigmatic system: A system is a set of
options with an entry condition, such that exactly one option must be
chosen if the entry condition is satisfied. Options are realized as
syntagmatic constructs or structures; a structure is a configuration of
functional elements – functions or function bundles. The functions are
motivated (nonarbitrary) with respect to the options they realize; the
grammar as a whole is motivated with respect to the semantics. The
only line of (relative) arbitrariness is that between content and
expression (between the lexicogrammar and the phonology).
A text in systemic-functional grammar is an instantiation of the
system (in the Hjelmslevian sense of “the linguistic system”). (Note
that “instantiation” is not the same thing as “realization”; the two
concepts seem to be confused in Saussure.) Text may be studied as
process or as product; in either case, interpreting a text means showing
how it derives from the system and therefore why it means what it
does. It is not possible here to present the networks of systems from
which the text is derived, since that would involve representing large
portions of the grammar. Instead we employ structural notations, with
brief discussion of some of the options from which the structural
functions are derived.
The analysis is given in 10 steps, with a short commentary on each.
The 10 steps are as follows:
1. transcription and analysis of intonation and rhythm
2. analysis into clauses and clause complexes, showing 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 the
latter for information (Given–New) structure
5. analysis of finite clauses for mood, showing Subject and Finite
262
dimensions of discourse analysis
6. analysis of all clauses for transitivity, showing process type and

participant and circumstantial functions
7. analysis of groups and phrases (verbal group, nominal group,
adverbial group, prepositional phrase)
8. analysis of grammatical and lexical cohesion
9. identification, rewording and reanalysis of grammatical
metaphors
10. description of context of situation, and correlation with features
of the text
1 Transcription and analysis of intonation and rhythm

The text is transcribed orthographically with notation for intonation
and rhythm:
Figure 1
263
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
264
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 具具 ⳯b ) 典典 ^ 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 具具典典 won’t know ||
4.4 2 ⳯b 具具 when it grows up 典典
4.5 2 a ‘b || what it is |||
Clause complex 5 ‘‘b a 具具 ) a ( 典典 ^ ⳯b (1 ^ +2 ( a ^ 1b ) ) )
5.1 ‘‘b a ||| well what 具具典典 you should do ||
5.2 a 具具 did Mrs. Siddons suggest 典典
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
265
The major portion of the text consists of five turns, each made up
grammatically of one clause complex. These contain, respectively, 3,
11, 6, 5, and 5 clauses. They show a preference for hypotactic (17)
over paratactic (7) interdependencies; the predominant logical-semantic
relation is that of enhancement, typically hypothetical (11 instances) –
the discussion centers around what to do if and when a certain situation
arises. Of the 15 other instances, 5 are extension, “and / or”; the
remainder (8) are projection, of which 2 are saying (the teacher as
Sayer) and 6 are knowing, mainly negative and with a generalized
Senser “(if) you don’t know” – this being an aspect of the problem
under discussion. (The two instances of I think [new addition] are
metaphorical modalities, not projections, as can be shown by adding a
tag: the tagged form of I think you should call it he or she is shouldn’t you?
not don’t I?)
All these clause complexes are not only complex but also impeccably
well formed, as is typical of casual spontaneous speech (including that
of children!).
So much for the “reasoning” component of the discussion. The
remainder consists of “suggesting,” partly humorous and partly serious,
and here the turns are short, one or two clauses each. The clause
complexes are even shorter, since each consists of just one clause.
There is no parataxis or hypotaxis. (6.1 and 7.1, She didn’t. I don’t
suppose she knew [new addition] could be considered to form a paratactic
elaboration, given the tone concord; but the latter, though necessary,
is probably not a sufficient condition.)
As far as the organization of the information is concerned, a
comparison of the two transcriptions shows that in the majority of
instances one clause is one information unit, this being the unmarked
(default) situation in English. This holds throughout, with the following
systematic exceptions: (1) seven out of the eight projections have both
projecting and projected clause on one tone group, for example, // if
you / don’t know / what it / is // – this is the locally unmarked form;
and (2) one clause consists of two information units, one for the Theme
and one for the Rheme: // if some / neighbour has a // new / baby next
/ door // – this is the predominant pattern when two information units
are mapped on to one clause. For the analysis of theme and information
structure see the next two sections.
266
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
267
The Theme is the (speaker’s) point of departure for the clause. It is

realized in English by position in the sequence: thematic elements are
put first. Hence the thematic structure of the clause is Theme ^ Rheme.
Each of the three metafunctional components of the content plane
– ideational, interpersonal, textual – may contribute thematic material.
The textual Theme is some combination of continuative (for example
oh, well), conjunctive (for example then, if) or relative (for example that,
which). The interpersonal Theme is modality (for example perhaps),
interrogative mood marker (WH-element or Finite verbal auxiliary),
or Vocative. The topical Theme is any element functioning in the
transitivity structure of the clause. The typical sequence is textual ^
interpersonal ^ topical, and the Theme in any case ends with the
topical element: in other words, the Theme of a clause extends up to
the first element that has a function in transitivity.
The unmarked Theme for any clause is determined by the choice of
mood: Subject in declarative, WH- or Finite element in interrogative,
[zero] in imperative and minor clauses. Semantically, the unmarked
Theme is the natural starting point for the particular speech function:
in a question, “this is what I want to know” – the information-seeking
(WH-) or polarity-carrying (Finite) element; in a statement, “this is the
entity on which the argument rests” (Subject).
The ongoing choice of clause Themes reveals the method of
development of the text. In the example, every clause has an unmarked
topical Theme. At first, the impersonal you predominates, followed
later by specific third person participants: the teacher (Mrs. Siddons),
the neighbour, and the baby. Many are preceded by textual Themes,
continuative and / or conjunctive. Thus the text develops as a
discussion of a general topic with particular personalities brought in to
carry it forward, the whole being linked together both dialogically and
logically. At the higher rank of the clause complex, on the other hand,
the logical structure of the argument becomes the dominant motif:
here there are a number of marked Themes, in the form of hypotactic
(dependent) clauses introduced by if. The picture is that of a shared
discourse being developed as a logical generalization with hypothetical
cases, without much concern for attitudinal rhetoric (the only interper-
sonal Themes are questions and uncertainties), and with some concret-
ization towards the end.
268
4 Information structure
Each information unit is analysed for information structure (numbers
refer to clause complexes, letters to information units):
Figure 5
269
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 to
carry the information focus. Anything after the focal element is thereby
marked as Given, while anything preceding it may be Given or may
be New (there are rather subtle intonational and rhythmic variations
serving as signals).
An information unit is not necessarily coextensive with a clause, but
that is its unmarked status: each ranking clause (i.e. independent or
dependent, but not embedded) is typically one information unit. The
principal systematic variants are (1) two clauses / one information unit:
failing dependencies, that is, a ^ b sequences; (2) one clause / two
information units: thematic focus, that is, // Theme // Rheme //
information patterns.
Analysis of the New elements will reveal the “main point” of the
text. In the example it is to do with babies, what sex they are, and
how they are to be referred to in cases of doubt. Subsidiary points of
270
attention are the baby’s growing up, the children’s understanding and
obligations, and the adult world’s possible displeasure.
The Theme in a clause is what is prominent for the speaker; it is
“what I am on about”. The New in an information unit is what is
made prominent (by the speaker) to the listener; it is “what you are
being invited to attend to”. When clause and information unit are
mapped on to each other, the result is a wavelike movement from
speaker to listener, the diminuendo of the speaker’s part being as it
were picked up by the crescendo of the listener’s part.
The effect of this movement is cumulative over the text as a whole.
The present text is typical in the way that the sequence of Themes
represents the “method of development” of the dialogue, while the
sequence of News represents the “main point” of the discussion, with
each speaker contributing her part to the construction of the overall
pattern – all unconsciously, of course.
This interplay of two distinct waves of prominence is possible
because Theme–Rheme and Given–New are not (as often conceived)
one single structure, but two distinct structures that interpenetrate. As
a result, they can vary independently, allowing for all possible 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, and
marked calling it she as Given; the effect of the latter would have been
to bring out the repetitive facet of calling it she, thus reinforcing its
cohesion with 2e, but by the same token to deprive it of its status as a
main point for attention. The interaction of the thematic and 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:
271
Figure 6
The system of mood expresses the speech function of the clause.

Typical patterns of realization are as shown in Figure 7 (where
means ‘is typically realized as’).
272

Figure 7
In addition there are the minor speech functions of exclamation,

greeting, and call, realized by minor clauses, and by exclamatory
declaratives.
Offers, commands, and statements may be tagged, for example, I’ll
help you, shall I? The tag makes explicit the speaker’s request for the
listener to perform his complementary role: accept offer, carry out
command, acknowledge and confirm statement. There are no tagged
clauses in the text under consideration.
The Subject expresses the participant in respect of which the particular
speech function is validated: performer in the case of goods-&-services,
bearer of the argument (i.e. the one on whom the validity is made to
rest) in the case of information. In a declarative, the Subject is typically
also the Theme (hence “unmarked Theme”; see the discussion of
Theme above); but whereas the Theme is a discourse (textual) function,
displaying patterns over the text as a whole, the Subject is an interper-
sonal function having significance just for the particular exchange. Here
it is frequently the impersonal you, showing that these statements are to
be arguable as statements that are valid in general; in other cases, it is
the baby, Mrs. Siddons, or the hypothetical neighbour.
The Finite element expresses the deicticity of the process, by
reference to either (1) speaker-now (primary tense: past, present, or
future) or (2) speaker judgement (modality: probability, usuality, obli-
gation, inclination, or ability; high, median, or low value). Almost all
the finiteness in this text is combined with present tense; the children’s
text proceeds as a series of declaratives, some independent and some
dependent, the mood-bearing constituent of which consists of general-
ized Subject plus Finite present (and there is very little secondary tense).
273
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 just
three expressions of obligation (ought and two instances of should).
Other than in combination with finiteness, there are again only three
modalities, in this case expression of probability: (I suppose and two
instances of I think). As it happens, in two paired instances the two
kinds of modality are associated: where the speaker expresses a 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:
274
Figure 8
Transitivity is the representation, in the clause, of the experiential

component of meaning: specifically the processes, the participants in
them, and the attendant circumstances.
275
This text is a discussion of a problem: what to do if a certain

situation arises. The majority of the clauses in it relate to the situation,
the problem, its solution, and the process of problem-solving:
(i) 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 (two
material: 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 and
verbal); Type III, being (relational and existential). They are distin-
guished in the grammar in various ways; the principal distinctions are
as follows:
Figure 9
276
In the example text, the clauses in (i–iii) above are all type III,
relational; the issue is one of being, partly attribution (having a baby,
which in this text means possessing it, not bearing it; being male or
female) and partly identification (being the name of). Of these, (i) and
(ii) are middle – there is a Medium (the neighbour, the baby) but no
Agent; (iii) however is effective – there is an Agent in the identification
process, always represented as you, but moving from personal “you” in
clause complex 1 to impersonal “you” thereafter. The clauses in (iv)
and (v) are type II, mental-verbal; (iv) are cognitive, with the Medium
(Senser) being you, I, Mrs. Siddons or the baby when it grows up; (v)
are verbal, the Medium (Sayer) being Mrs. Siddons. All interactants,
real and hypothetical, are involved in thinking about the problem –
including the baby, at some future date, if a solution is not reached
now; and the teacher has put the problem into words.
A summary of process types and the relevant participant functions
follows:
Figure 10
277
7 Other group features

Groups and phrases are analysed with respect to features that are
relevant to the inquiry:
Figure 11
The English verbal group carries a recursive three-term (past /

present / future) tense system of the type “present in past in . . .”
where any tense selection may become the point of reference for
another one, subject to certain restraints that limit the total number of
278
possible combinations. In the “full” (finite) system of tenses the num-

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

tense, polarity, and so on, the nominal group further specifies the
entity represented by the Head noun. There are similarities between
the two types of word group; but the nominal group has much more
lexical material, since entities have a more developed taxonomic organi-
zation. Hence the nominal group has a functional structure Deictic–
Numerative–Epithet–Classifier–Thing–Qualifier, with left–right ordering
279
from the most instantial, situation-bound to the most permanent

characteristic, modified by a purely syntactic principle which puts
anything that is embedded at the end.
There is little elaboration of nominal groups in the text. Most of
them are simply personal pronouns, functioning cohesively; those that
have noun as Head contain just enough specification to establish the
general point being made, for example, a small baby, your neighbour, the
house. The only Qualifier is the nonfinite clause to calling it she following
the Attribute used; and this is a metaphor for a modality “have so
usually been calling it she”.
Figure 13
8 Grammatical and lexical cohesion

The text is analysed for grammatical and lexical cohesion:
280
Figure 14
The headings given in the key are the basic lexicogrammatical

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

what is presumed is some semantic representation: of a participant, for
example, as when it refers back to “a small baby”, but also of a semantic
construct of any extent. In substitution and ellipsis, on the other hand,
what is presumed is a lexicogrammatical representation, some piece of
wording that has to be retrieved, as when she didn’t requires the
restitution of suggest (anything); this is a different kind of textual
retrieval and rarely extends beyond one clause complex. Conjunction
refers to the nonstructural representation of logical-semantic relations
that may also be expressed structurally; for example, on the other hand,
semantically related to paratactic but and hypotactic although. Lexical
cohesion is created by the repetition of a lexical item (e.g., call . . . call);
the use of a synonym (e.g., call . . . refer to); the use of a high-frequency
collocate (e.g., house . . . family); or the use of a hyponym or 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 and
personal reference; the discourse unfolds around a small number of
entities that are constantly being referred to. When a new instance is
brought in, the link is achieved by collocation: in the house . . . next
door. There is very little conjunction, because the logical-semantic
relations are realized by the hypotactic organization of the clause
complex: The reasoning is systematic and explicit. There is also little
ellipsis, which comes in only when the reasoning gives way to a more
dialogic pattern with shorter exchanges.
9 Grammatical metaphor
Grammatical metaphors are identified, reworded, and reanalysed.
Most examples of adult English contain some instances of grammat-
ical metaphor: clauses in which one type of process is represented in
the grammar of another; for example, the fifth day saw them at the summit
“on the fifth day they arrived at the summit”, or guarantee limited to
refund of purchase price of goods “we guarantee only to refund the price
for which the goods were purchased”.
Children’s speech is largely free of grammatical metaphors of this
kind; this is in fact the main distinction between child and adult
language. There are no examples of it in the present text.
There are however certain grammatical metaphors that have been
built into the language, so that the metaphorical version has become
the norm; for example, she gave a nod “she nodded”, he has a long nose
282
“his nose is long”. One very common type of these is the use of a
mental process (cognitive) clause to express a modality, such as I think,
I don’t believe. It was pointed out above that the tagged form of I think
it’s broken is I think it’s broken, isn’t it?, not I think it’s broken, don’t I?,
showing that this is a metaphor for it’s probably broken. We can use one
of these as an example:
|| I don’t suppose || she came ||| a ˆ ´ b
mental: cognitive material
negative positive
reworded as:
||| she probably didn’t come ||| (single clause)
material
modalized:probability/median
negative
In cases like this it saves time if the analysis moves directly to the
nonmetaphorical version, since the rewording is quite automatic. In
other instances, however, the principle is as follows:
1. Analyse the clause as it stands.
2. Reword it, in nonmetaphorical form.
3. Analyse the reworded version.
Both analyses figure in the interpretation. Sometimes it takes several
steps in rewording to reach a nonmetaphorical version, and there may
be more than one possible route; all are potentially relevant to an
understanding of the text.
10 Context of situation
The context of situation of the text is described in terms of field, tenor,
and mode. The “field” is what is going on: the nature of the social-
semiotic activity. The “tenor” is who are taking part: the statuses and
mutual roles of the interactants. The “mode” is what part the language
is playing: the rhetorical and communicative channels. This description
is then used to interpret the lexicogrammatical features of the text.
Field. A general, imaginary problem of verbal behaviour: how to
refer to a baby whose sex is unknown, without offending
against the parents, the baby (later in life), or the language.
283
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 in
orientation, moving towards (partly humorous) resolution.
These features determine the choice of register: that is, the kinds of
meanings that are likely to be exchanged. Like the rest of the linguistic
system, the patterns are probabilistic: given these features of the 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
284
learn the language in the first place: the systematic relationship between
these categories of the situation and the metafunctions of the content
system. By and large, characteristics of the field predict experiential
meanings, those of the tenor predict interpersonal meanings, and those
of the mode predict textual meanings.
In analysing a text, we identify those features of the lexicogrammar
which in a text-generation program might reasonably be expected to
be called upon if the situation was represented in this way.
11 Conclusion
Three final points should be made about an outline of this type. One
is that it is just an outline: Obviously, the analysis under every heading
could be developed much further in delicacy, and other headings could
be added. The guiding principle is to select and develop whatever is
needed for the particular purpose in hand. There are many different
purposes for analysing a text, and the scope and direction of the analysis
will vary accordingly. Often we may want to scrutinize only one or
two features, but to follow them through to a considerable depth.
Secondly, a text analysis is a work of interpretation. There are
relatively few absolute and clearcut categories in language; there are
many tendencies, continuities, and overlaps. Many actual instances can
be analysed in two or more different ways, none of which can be ruled
out as impossible; some may be less sensible than others, and so can be
discarded, but we may still be left with valid alternatives. Especially in
a literary text it is to be expected that we will find clauses with multiple
grammars; but this is not a distortion of the system – it is a richer use
of its natural resources. All analyses may need to figure in the
interpretation.
Thirdly, the lexicogrammatical analysis is only a part of the task. It is
an essential part; all text is made of language, and the central processing
unit of the linguistic system is the lexicogrammar. But just as the
grammatical system does not itself create text – text is a semantic
creation, with the grammar functioning largely (though not entirely) as
the automatic realization of the semantic choices – so the analysis of
the grammar does not constitute the interpretation of a text. (There
has been some misunderstanding on this point, for example in the use
of cohesion as a method of text analysis. Cohesion is an essential
property of texts, but it is the way the cohesive resources are deployed
that makes the difference between text and non-text, and between one
text and another.)
285
Editor’s note: Examples of text analysis based on the grammatical

principles outlined in this chapter are presented in Volume 2, Linguistic
Studies of Text and Discourse. See Hasan (1985/1989) and Martin (1992)
for theoretical discussion and illustration of the place of grammar in the
analysis of discourse.
286
SECTION THREE
CONSTRUING AND ENACTING
Linguistics like other sciences requires a metalanguage for representing

its object of study, which in the case of linguistics is language itself.
However, as Halliday points out in ‘On the ineffability of grammatical
categories’ (Chapter 11), unlike other sciences, linguistics is “language
turned back on itself”, to use Firth’s expression. The problem lies,
Halliday argues, in the nature of language as object. Because language
is an evolved system, not a designed system, it rests on principles that
are ineffable. Its very richness, “its power of distilling the entire collective
experience of the culture into a single manageable, and learnable, code
. . . puts its categories beyond the reach of our conscious attempts at
exegesis”. This richness is most apparent in unconscious, spontaneous,
un-self-monitored language, or as Halliday notes, “our ability to use
language depends critically on our not being conscious of doing so”.
In ‘Spoken and written modes of meaning’ (Chapter 12), originally
published as a chapter in Comprehending Oral and Written Language
(1987), Halliday elaborates further on the differences between uncon-
scious and spontaneous spoken discourse, and its more conscious and
self-monitored counterpart, written language. Arguing against assump-
tions that written language is syntactically more complex or more richly
endowed than spoken, Halliday maintains that each is highly organized
and complex in its own way: “Written language tends to be lexically
dense, but grammatically simple; spoken language tends to be gram-
matically intricate, but lexically sparse”. Halliday describes written
language as “crystalline” and spoken language as “choreographic”.
Comparing the two, Halliday notes how speaking and writing impart
their own character onto the world they depict. Written language
objectifies. “A written text is an object, so what is represented in
writing tends to be given the form of an object. But when one talks,
289
construing and enacting
one is doing; so when one talks about something, one tends to say that
it happened or was done.” By means of grammatical metaphor, written
language symbolically distances “the act of meaning and its counterpart
in the real world”. Each plays complementary roles when it comes to
using language to acquire knowledge and reflect on one’s experience.
Halliday maintains that this complementarity must figure into any
attempt at developing a linguistic theory of learning.
In ‘How do you mean?’ (Chapter 13), appearing in Advances in
Systemic Linguistics: Recent Theory and Practice (1992), meaning is taken
as a mode of action occurring at the intersection of the conscious and
material modes of experience. In particular, Halliday examines the
evolution of protolanguage into language, or how the two-dimensional
semiotic that defines the mammalian experience evolves into ‘a semiotic
of a new kind: a stratified, tri-stratal system in which meaning is “twice-
cooked”, thus incorporating a stratum of “pure” content form. The
two dimensions of protolanguage, a minimal semiotic system, include
“the ‘inner’ dimension of reflective / active, ‘I think’ as against ‘I want’,
and the ‘outer’ dimension of intersubjective / objective, ‘you and me’
as against ‘he, she, it’ ”. The third dimension results from the introduc-
tion of grammar, or as Halliday describes it: ‘a purely symbolic mode
of being between these two interfaces’. The processes of instantiation
and realization make possible this dynamic open system we call language.
In the two final works in this section, ‘Grammar and daily life’,
which first appeared in Functional Approaches to Language, Culture and
Cognition (2000), a Festschrift for Sydney Lamb, and ‘Grammar and
grammatics’, published as Volume 121 of the series Current Issues in
Linguistic Theory (1996), Halliday describes how grammar enables us,
unconsciously, to construe our reality, and interpret our experience,
while grammatics makes it possible for us to reflect consciously on how
this theory of our human experience works. Halliday introduced the
term “grammatics”, in contradistinction to grammar, to distinguish
between a particular stratum of a natural language and the study of this
stratum. “Thinking grammatically”, or “using grammatics to think
about what grammar thinks about the world”, may help us better
understand this ‘grammatical energy’ or ‘grammatical logic’ that powers
language and also conditions our attitudes to each other and to the
world around us. “To be a linguist”, Halliday writes, “is inevitably to
be concerned with the human condition”, and those who ‘think
grammatically’ will be better prepared not only to address issues of
social injustice and inequality, but also to contribute to the develop-
ment of new applications of linguistics such as intelligent computing.
290
Chapter Eleven
ON THE INEFFABILITY OF GRAMMATICAL

CATEGORIES (1984)
1 The problem of the ineffable

We live in an age of growth, in which every day more and more
things come into our lives; and things, and all their parts, need names.
So more and more words come in with them – new words, or new
ways of exploiting, embellishing and combining the old ones; and in
this way the balance is maintained. There is no sign that our onomastic
resources are drying up; indeed we are likely to run out of natural
resources long before we run out of names for the things we make out
of them. But there are signs that the things are becoming more resistant
to being named. There is no natural way of referring to the various
small plastic objects that lie around the house, or the toys we give our
children, or the furniture we now have to assemble for ourselves. They
no longer fit our taxonomies. We live in modules, sit on units and
entertain ourselves with systems.
Behind these nameless objects is a technology and a science that
produce them; and there, less visible to us, is another realm of things
that have to be named. Many of them are abstract things, the categor-
ies and concepts of a theory; and some of these also prove recalcitrant
to ordinary onomastic processes – they only come to be ‘named’ by
some mathematical formula, like a function of the co-ordinates of x and y,
the integer over psi1 and psi2, and so on. But somehow they have to
be enmeshed in language; otherwise they are not brought under
control.
First published in The Tenth LACUS Forum, 1984, edited by Alan Manning, Pierre Martin
and Kim McCalla, pp. 3–18. Amsterdam: John Benjamins.
291
In this respect linguistic categories are no different from those of

other theories; they too need names. It is true that in one instance a
theory evolved without them: Chinese phonology, in its first few
centuries, had neither terms nor definitions – and still managed to give
an account of the syllable that was explicit enough to enable later
scholars to reconstruct the system. But western linguists were always
more namebound; they either created a terminology, or borrowed it
from some other field. The earliest known Arabic grammarian, Al-
Khalil ibn Ahmad, took various terms over from architecture; his great
pupil Sibawayhi, who had been trained as a lawyer and so placed a
high value on names and definitions, replaced them with a more
systematic terminology derived from legal models (see Carter 1981).
In ancient Greek linguistics, by contrast, the technical terms evolved
out of everyday language. The process was a gradual one, extending
over three or four centuries; and in the course of this time the original
terms had moved some distance from their non-technical meanings,
evolving as the theory evolved. Thus, in everyday language, onoma
meant ‘a name’, rhēma meant ‘a saying’; logos meant ‘speech, discourse’
and grammatikē meant ‘writing’. As grammatikē evolved into ‘grammar’,
onoma came to mean ‘a noun’; rhēma became first ‘rheme’, in the
Prague sense, and then ‘a verb’; while logos came to mean ‘a sentence’.
Here it was the folk linguistic terms for forms of discourse which
became the source of technical nomenclature in grammar.
Once having reified these abstract categories by naming them, the
Greek grammarians went on to ask what the names meant. What ‘is’ a
noun? they wondered. At first, this was a question of: how do I
recognize a noun when I see it? how do I know whether something is
a noun or not? But before long the questions came to be asked in the
other direction: what ‘is’ a noun, in the sense of what function does it
serve? At this second stage, instead of treating ‘noun’ as the Value and
then supplying a Token for it, the definition now treats ‘noun’ as the
Token and seeks to supply a Value for it (for the terms Token and
Value see Halliday (1985: Chapter 5); also this volume, Chapter 7,
p. 173). Instead of ‘a noun is a word that inflects for number and case’,
we have ‘a noun is the name of a person, creature or thing’. This is a
decoding definition, one which embodies a notion of ‘what the
category means’.
To define a linguistic term by encoding is relatively simple: one
hops along the realizational chain of grammatical categories until
reaching some form of output. Defining a noun in this way would
involve, altogether, three steps: (1) a move in rank – a noun inflects
292
on the ineffability of grammatical categories
for case; (2) a move in delicacy – case is nominative, genitive, dative

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

head my attention may wander and lose touch with the drift of it.
293
This is accompanied by an antinomy of count / mass, which produces

expressions such as a pane of glass, conceptualized as ‘a pane with glass
in it’ (like a cup of tea), and extending to time construed as a container
as in an hour of bliss. Reddy takes this argument one step further and
shows how the metaphor determines the way we talk about communi-
cation, with ideas, meanings and emotions being packaged inside words
and sentences and piped along a conduit. He lists about 150 such
expressions, like get your thoughts across, his feelings came through, the lines
are empty of meaning.
All this, according to Reddy, is innocent enough. But it may become
pathological, when the conduit metaphor pervades the whole termino-
logy of linguistics, because the effect of this metaphor is that all words
for kinds and quantities of discourse, like poem, message, text, are split
into two meanings: (i) the content, and (ii) the container. In our terms,
there is a stratal polysemy: these words refer both to a piece of meaning
(semantic stratum) and to a piece of wording (lexicogrammatical
stratum). Reddy’s view is that as long as one remains within the
conduit metaphor the effect is still benign, because the one sense is
then metonymic to the other: that is, the meaning is ‘contained in’ the
wording. So text2 ‘wording’ ‘contains’ text1 ‘meaning’, and no great
harm is done. It is if one tries to escape from the metaphor that this
kind of polysemy becomes pathological.
It will have been noticed that this stratal polysemy is precisely the
process by which the original Greek terms had come to be extended
so as to serve in the grammar: starting as names of semantic (or pre-
semantic) categories, they were then transferred to become names for
the lexicogrammatical categories by which the former were (typically)
expressed. So logos2 ‘sentence’ is that by which logos1 ‘discourse’ is
realized. Likewise with onoma and rhēma: onoma2 ‘noun’ is what realizes
onoma1 ‘name’ – while with rhēma there is an additional step: rhēma1
‘saying’ is realized by rhēma2 ‘Rheme’, which is in turn realized by
rhēma3 ‘verb’.
It is not hard to see why this happens. It soon becomes obvious,
once one begins to be aware of language as an object and starts to
investigate its central processing unit, that the categories of this coding
system are not arbitrary, but relate systematically to the meanings. It is
natural, therefore, to name them by reference to their semantic
function.
With terms for classes and units, like those above, the polysemy is
confined within language. Both ‘noun’ and ‘name’ are linguistic
entities; the difference is simply that one is grammatical, the other
294
semantic. The same is true of the pair ‘sentence’ and ‘discourse’. But
with the majority of grammatical terms – those for functions, systems
and terms in systems (features) – this does not hold. These too have to
be imported into grammar from outside; but in this case they come
not from folk semantics but from outside language altogether. When
categories of this type come to be named, the terms that are introduced
for the purpose interpret the categories by reference to some aspect of
extralinguistic experience.
A typical example would be a complex of categories such as ‘(system)
number: (features) singular / plural’. Consider a label such as ‘plural’.
This derives from the ideational meaning of the category: it expresses a
relation between that category and the speaker’s experience of the
world. The term ‘plural’ is the name of this relationship. But the same
term is also used as the name of the grammatical category which realizes
this relationship; a noun will be said to be ‘plural (in number)’. And
this can cause problems.
Typically in the history of western linguistics the reasoning has
proceeded as follows. In the morphology we are presented with a
certain category, let us say a two-term system, of the form ‘a : x/y’;
thus, ‘all a are either x or y’. By inspecting, or perhaps introspecting,
typical contexts in which these forms are used we recognize a redun-
dancy, such that x redounds with one of a set of countable things and
y redounds with two or more. The abstract labels ‘a : x/y’ are then
replaced by the substantive labels ‘number : singular / plural’. This is
of course an idealized model of the process; such abstract labels have
never been used, as far as I know, at least until modern times. But it
helps to bring the issue into focus.
What happens next? The categories that have been labelled in this
way then come to be reified and the question is asked what they mean.
The answer given is: ‘singular means one of a thing, plural means more
than one’. In fact, these are definitions of the words singular and plural;
but they are made to serve as definitions of the metawords, the terms
of the metalanguage. Next, the terms are evaluated for their predictive
power: will they correctly predict text from situation, or situation from
text? Given a plural form, will it refer to more than one of a thing?
Given more than one of a thing, will it be referred to by a plural form?
The answer this time is: yes, with a certain degree of probability –
high enough for many purposes, but inadequate for some, and disturb-
ing for those who like their categories pure. This then gives rise to a
theory of ‘core meanings’: a term of the metalanguage is said to
represent the ‘core meaning’ of the category. With this defence, in
295
spite of fruit and furniture, scissors and oats (or their counterparts in the
language in question), ‘singular’ and ‘plural’ can continue to be used.
This problem – that of interpreting a symptom and then labelling its
interpretation – is common to all sciences. It arises in any realm of
discourse involving explanation and abstraction. Somehow, a 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 swelling
somewhere on my body. He looks at it, and pronounces ‘You’ve got
an oedema.’ What is oedema? – Greek for ‘swelling’. But what he is
diagnosing is a more abstract swelling – it is that which is manifested
by the swelling on my body. An oedema is the Value of which a
swelling is the Token. The use of a different term in this way allows
for stratal diversification: not all swellings realize oedemata, and not all
oedemata are realized as swellings. The relationship is a probabilistic
one – and hence invites further, more delicate investigation.
Where do such terms come from? Ideally, they come from a parallel
but distinct semiotic. It should be a natural language, in order to
maintain the non-arbitrariness of the relation between the symptom
and the underlying condition – given swelling and oedema we would
predict that, in default of any special circumstances, they will refer to
the same thing. But it should be a different language, or at least a
different sub-language, in order to allow for instances where they do
not. And it should be a higher status code, in order to symbolize a
higher order of abstraction (and also the social value of abstract
knowledge). The ideal source of a metalanguage is thus the ‘high’
variant in a diglossia. A word of a natural language that is at one
remove from primary reality, such as ancient Greek, or classical
Chinese, or Sanskrit, is appropriate for symbolizing a phenomenon that
is at one remove from primary observation.
But when it comes to metalinguistic matters, linguistics presents a
special case. It is not just another science. It is ‘language turned back
on itself’, to use Firth’s (very British) expression; or, in Weinreich’s
(very American) formulation, ‘language as its own metalanguage’. As a
consequence, where other sciences need two terms, we need three:
one for the phenomenon, and two for the metaphenomenon, one
grammatical and the other semantic. To return to the example of
number: we need to be able to say that the grammatical category of
‘plural’ typically expresses (realizes) the semantic category of, say,
‘manifold’, which typically expresses (redounds with) more than one of
a thing.
But notice what has happened. The grammatical category of ‘plural’
296
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 have
tried to escape from the circle by finding a gloss for –s / –z / –iz –
that is, an exact synonym for it, in natural language wording; and that
is an extremely difficult thing to do. We might try glossing it as more
than one, or several, or many; but the trouble is we don’t actually say
I like more than one cat, or I like many cat – we say I like cats. The
meaning of the –s on cats is impossible to gloss in natural language,
except by means of itself. The category is, quite simply, ineffable.
2 Difficulties with the subject

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

with is taken to mean ‘represents’. In other words, a clause that was
intended to say how the Theme in English is to be recognized was
taken as a statement of how it is to be defined – one of the most
fundamental confusions in linguistics. It would all have been avoided if
the verb be had had a passive; I should, therefore, have created the
appropriate metagrammar and written:
The Theme of an English clause is been by the element that is put
in first position.
So there are problems in using natural language as a metalanguage,
for whatever purpose: its logical and ideational systems were not
designed for the task. Some combinations of features may be realized
in ways that are ambiguous, others may carry a baggage of unwanted
corollaries, and so on (this does sometimes lead to the creation of
minor neologisms in the grammar, like the prepositional phrases that
appear in the language of mathematics (I mean mathematical English)
such as the inequations over O, symmetry about a certain point for various
angles of rotation). And using natural language as a metalanguage for
natural language itself is likely to inflate the problems still further, since
whatever shortcomings it has are compounded by the factor of self-
reference – the metalanguage being a form of the same semiotic system
that it is also being used to describe.
The problem of self-reference is a familiar one; nevertheless it is not
the central issue. The real problem lies in the nature of language as
object, and particularly the nature of lexicogrammar. It is not so much
that language is not good for glossing with. The problem is rather that
language is not good for being glossed.
Let us take, as an example, the category of Subject. This has always
been one of the most obscure and controversial categories in western
grammatical theory. Here is Jespersen on the subject (1909–43, 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
298
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 functioning
as Subject in any instance has meaning as actor, or has meaning as
topic; but as Subject it has none – the category of Subject has no
meaning in itself. In this view, Subject is a grammatical function whose
only function is to be a grammatical function.
Such a view is enshrined in the terminology, in the term ‘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’
299
(Actor–Action) and with ‘functional sentence perspective’ (Theme–

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

of tenses (or, in his later term, assertions) in Hopi: factual or present-
past (later reportive), future (later expective), generalized or usitative
(later nomic) – which initially takes up half a page, but whose semantic
and ideological interpretation is the subject of an entire article (‘An
American Indian model of the universe’). Such discussions typically
involve implicit, or sometimes explicit, contrast with the language in
which they are written; in this case, English.
More recently, there has evolved a third context for semantic glosses:
research in text generation, in the framework of artificial intelligence.
300
One project in which they play an essential part is the Penman project
at the Information Sciences Institute (Mann and Matthiessen 1983;
Mann 1983). In Penman the motive power is provided by the
grammar, Nigel, which is a systemic grammar consisting of a network
of some hundreds of options. At each choice point a Chooser is
activated; the Chooser consults the environment (the Knowledge Base)
for instructions on which way to go. The Chooser’s questions are
referred to an Inquiry Operator; and they take forms such as the
following:
Is this concept inherently multiple, i.e. a set or collection of things,
or is it unitary?
Is the process one which conceptually has some sort of entity which
causes the process to occur?
Does this represent a concept which the speaker expects the hearer
to find novel, not previously mentioned or evoked, and thus does
not expect the hearer to identify uniquely by reason of the attention
which it currently holds, its inherent uniqueness in culture, or its
association with an identifiable entity?
All such glosses are attempts to get at the grammar beneath the skin;
and they may be supported by a variety of different beliefs. First, it
may be assumed that all grammatical generalizations have some signific-
ance at a higher stratum; or alternatively, that some are simply house-
keeping devices and have no semantic function. Secondly, those
grammatical categories that are regarded as semantically significant may
be thought of as universal, or as particular to the given language, or as
particular to a given register, a functional variety of a language. (These
would represent fairly well the respective views of Jakobson, Hjelmslev
and Firth.) Thirdly, it may be held that every such category has one
meaning that is common to all its manifestations, and the problem is to
find the right semantic generalization to cover all cases; or alternatively,
that some categories at least are polysemous, so that their meaning
varies in ways that are not predictable from the context (cf. Ikegami
1980: 59). Fourthly, there is a range of beliefs about the place of
grammar, and the need to postulate some higher level semiotic system
(‘semantics’, ‘semology’, ‘the conceptual level’, etc.) to which gram-
matical categories can be related in a systematic and in some sense
‘natural’ way. Positions taken on these issues may complicate the task
of semantic interpretation: for example, if categories are assumed to be
universal, and yet are established at an insufficiently abstract level.
But whatever beliefs are held about them, grammatical categories
301
will remain ineffable. Some of the more recalcitrant ones are categories
that Whorf originally called ‘covert’: ‘having no mark other than
distinctive reactances with overtly marked categories’. But by no means
all are of this kind – there is nothing covert about definiteness in
English, for example. The phenomenon we are concerned with has
more to do with Whorf’s follow-up notion of a ‘cryptotype’.
Whorf remarks of these that ‘they easily escape notice and may be
hard to define, and yet may have profound influence on linguistic
behavior’ (1956: 92). Among cryptotypes in English Whorf cites
gender, transitivity (of the verb), inherence (of the adjective), and
various more delicate categories, such as that of verbs that may be
phrasalized with up. There is, of course, a connection between the two
senses in which Whorf is using ‘crypto-’: a category may be hard to
define precisely because it is hidden from view. But hidden from whose
view? It is not because they are hidden from the linguist that grammat-
ical categories are hard to define; once the linguist has found them, the
fact that they had escaped his notice ceases to matter. The significance
of this concept of a cryptotype is that it is something that escapes the
notice of the speakers of the language.
Franz Boas long ago drew attention to the unconscious nature of
language, contrasting it in this respect with the other meaning systems
of a culture; and although his observations have often been quoted, it
seems to me that their significance is seldom fully taken into account.
There is a fundamental relationship between the unconsciousness of
language and the nature of its semantic categories. I have often pointed
out, in the many years since I began the study of informal speech, that
it is only in the most spontaneous, un-self-monitored kinds of discourse
that a speaker stretches his semantic resources to the utmost (cf.
Halliday 1966). This does not happen in formal speech; and it certainly
does not happen in writing. It is in unconscious spoken language that
we typically find the truly complex sentences, with their labyrinths of
hypotaxis and all their projections and expansions, from which, while
we blunder through such sequences often losing ourselves completely
when we are engaged in the planned self-monitoring discourse of an
academic lecture, we emerge in good order and with every node
unravelled provided we are completely unaware of what we are saying
and attending only to whatever it is we are involved in at the time.
(That sentence is best taken orally, at high speed.) Our ability to use
language depends critically on our not being conscious of doing so –
which is the truth that every language learner has to discover, and the
contradiction from which every language teacher has to escape.
302
Why is there this apparent contradiction, such that only when we

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

Meaning is formed in action; people create meaning, by exchanging
symbols in shared contexts of situation. The symbols evolve along
with the meanings; there is just one process taking place here, not
two, though we have to interpret it as if it was two. We cannot
observe this process as it took place in the history of the community,
since that would be coextensive with the evolution of the human
species. But we can observe it happening in the history of a human
303
child. Semogenesis begins well before the mother tongue, as the infant
creates his own protolanguage or ‘child tongue’: he constructs this
symbolic system, in interaction with those who share in his meanings,
for the twofold purpose of doing and thinking that characterizes all
such systems; and in the same process he also constructs the objects of
his action and his reflection. But the child tongue has its limitations,
for both these purposes; so he moves into the adult mode, and takes
over the mother tongue with its ready-made grammatical categories.
The symbols of the mother tongue, which have been around him from
the start, now become his reality, at once a part of, and a key to, the
complex phenomena of his experience. Language and culture are
construed as one.
Does a child, then, know what a Subject is? We cannot ask him;
nor can we set up a test situation to find out – if only because children,
given an unnatural task, will respond with unnatural behavior. (It is
not intended to suggest that there is any contrast here with adults. The
problem lies again in the unconscious nature of linguistic processes,
which adults cannot reproduce experimentally either.) Nevertheless it
is clear, surely, that a child does know what a Subject is, because he
uses one a hundred times a day. We only have to listen to a five-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 the
system that lies behind them; for language (if I may be allowed to
invert Chomsky’s famous dictum) is an infinite system that generates
only a finite body of text. But what we can observe is already very
convincing.
If I assert that a five-year-old knows what a Subject is, it is because
I have listened to children for many years, and heard them talking in
clauses which have Subjects. In my own detailed record of one
particular child, there are about 2,500 of them; but since child language
studies became fashionable there has been an abundance of such
material available, if one does not feel one can rely just on one’s
listening. Now, any one of these clauses could have had the appropriate
Subject by chance. Moreover, since no linguistic category is chosen in
isolation – in choosing the Subject one is always making other choices
besides (and this will apply whatever category is used as illustration) –
in any one instance we could always claim that the appropriateness of
the Subject was a consequence of some other choice. But if countless
304
children produce countless clauses each with an appropriate Subject,

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

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

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

that 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. The
speaker is telling a story of a sporting experience, and he says:
I caught the first ball, I was beaten by the second; the third I stopped
– and by the fourth I was knocked out.
Let us identify the different grammatical variables, and establish
clause by clause what has changed and what has remained constant.
Subject Actor Theme
1 I I I Subject = Actor = Theme
2 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 and
the 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 to
me, not what I did), and clause 3 is marked by textual modesty (now
I’ll tell you about the ball, not about myself). In clause 4, the speaker
gains further merit by ending on a doubly modest note, in which he is
neither the Actor nor the Theme.
But in regard to the speech function, the picture is quite different.
The speaker retains himself in the role of Subject throughout. There is
no sign here of interpersonal modesty; the assertion is made to rest on
I every time, and the listener’s response, correspondingly, must always
have a you in it – Did you?, Were you? In other words, every step in
308
the narrative has to be validated by reference to ‘me’; and that is the

meaning of Subject. ‘Subject’ is not the same as ‘Actor’; nor is it the
same as ‘Theme’. But it is far from being devoid of meaning.
It is quite possible to have a clause in which all three of these
functions are dissociated from one another:
that teapot my aunt was given by the duke
Theme Subject Actor
This helps to give a sense of the different meaning of each. This
particular one is made up; but the structure of which it is an instance is
entirely familiar. (Except perhaps to philosophers of language, who
tend to be disturbed by departures from the ideal of John hit Paul. Once
when I was giving a seminar, a participant refused to accept that such a
clause was possible in English, with a non-Subject nominal in fronted
position. So I asked him a question – I forget what the question was,
but any question would have served; and when he answered, I said
‘Yes – that answer I’ve been given by other people too.’ Needless to
say, he raised no eyebrow; he did not notice that I had used a clause
with the structure he had just rejected as impossible. When I pointed
this out to him, he seemed to think I had cheated – perhaps by bringing
real language into the discussion.) Obviously, there would be no sense
in dissociating the Subject from both Actor and Theme if it did not
embody a meaning of its own, distinct from either of the other two.
4 Talking about the ineffable

What I have been trying to show with this illustration is that while,
with a category like Subject, it is impossible to answer the question
‘what does it mean?’, this does not signify that it has no meaning. The
problem of ineffability is common to all grammatical categories; there
are various reasons why some may seem less problematic than others,
but it is an illusion to think that any can be exhaustively defined. And
this, as I remarked above, is not because of the shortcomings of natural
language for serving as a metalanguage, real though such shortcomings
are. Rather the converse: it is the very richness of natural language, its
power of distilling the entire collective experience of the culture into a
single manageable, and learnable, code that puts its categories beyond
the reach of our conscious attempts at exegesis.
This leads us back to the question of the Grundbedeutung. The
categories we have been considering have been categories of the
grammar: grammatical systems and structures, and their component
309
features and functions; and since grammar is the central processing unit,
where meanings of different kinds are brought together as wordings,
we expect its categories to be valid for the language as a whole. All
uses of English involve ‘Mood : Subject + Finite’, or ‘tense : past /
present / future’; and these are assumed to be in some sense ‘the same
thing’ in all contexts, since otherwise we would not be looking for
definitions of them.
No such constraint figures in our conception of semantics. The
grammar is the grammar: it has internal organization, of a 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 each
register. No doubt we can also conceive of such a thing as ‘the
semantics of English’; but we also feel that (at least at the present state
of our knowledge) it is not counter-productive to envisage a more
restricted domain for semantic generalizations, and to operate with
semantic sub-systems each relevant to a specific universe of discourse.
In principle, the domain of a semantic description may be anything
from ‘the whole language’ down to a single text. At one end of the
scale, I have found it useful to set up a semantic system relating to just
one dialogue of 35 words long; this was a child–adult dialogue, and the
purpose was to explore what meaning potential the child must have in
order to be able to construe such a discourse (see Appendix, p. 313).
Geoffrey Turner’s (1973) semantic networks define a rather broader
range of texts, such as mother–child control patterns in specific exper-
imental situation types. More general again is Ruqaiya Hasan’s (1983)
‘message function’ network, which describes spontaneous interaction
between children and parents, for the purpose of investigating the
development of children’s learning patterns. At the other end of the
scale, J. R. Martin’s (1983a, 1992) conjunction networks are like
grammatical networks in that they are set up for the language as a
whole.
When we describe semantic systems, we are saying what it is that
‘preselects’ the grammatical categories: what choices in meaning call on
what features in the grammar for their realization. It is by this process
that the grammatical categories are defined; when this is done, there is
no need to gloss them further. Once the semantic system is made
explicit, it can only be misleading to attach separate semantic descrip-
tions as glosses to the categories of the grammar.
At the same time, if the semantic system is set up only for a restricted
domain, some particular register variety, then the meanings of any
310
grammatical categories that figure distinctively in that variety will

appear 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 makes
only limited demands on it, for realizing the meanings that are
engendered by that particular context. The category of ‘future in the
register of weather forecasting’ is much less resistant to being glossed
than the general category of ‘future in English’.
This interpretation of semantic systems is a kind of functional
semantics, and it derives from the twentieth century functional semantic
traditions of Boas, Sapir and Whorf, of Malinowski and Firth, and of
Mathesius and the Prague school. These were three groups of scholars
with very different orientations, but their work was complementary in
significant ways. While each had a well-rounded view of language,
they emphasized, respectively, the ideational, the interpersonal and the
textual aspects of meaning.
For Malinowski, language was a means of action; and since symbols
cannot act on things, this meant a means of interaction – acting on
other people. Language need not (and often did not) match reality; but
since it derived its meaning potential from use, it typically worked. For
Whorf, on the other hand, language was a means of thought. It
provided a model of reality; but when the two did not match, since
experience was interpreted within the limitations of this model it could
be disastrous in action – witness the exploding petrol drums. Mathesius
showed how language varied to suit the context. Each sentence of the
text was organized by the speaker so as to convey the message he
wanted at that juncture, and the total effect was what we recognize as
discourse. Their work provides the foundation for a systematic func-
tional semantics which enables us to bridge the gap between the
context of culture and the language, and between the context of
situation and the text. This is how we can become aware of the
meaning of grammatical categories.
As a final step, let us summarize some of the alternative principles
that can be adduced for talking about the ineffable:
A. Metonymic: the use of some semiotic system as a descriptive

metalanguage (the ‘parasemiotic’ principle)
1. Parallel semiotics within the same language
(a) everyday language as folk linguistics (possible where there
is a shift of metafunction but probably not otherwise)
311
(b) open-ended definitions from theories about language (not

theories of language), e.g. rhetoric
(c) self-contained, technical systems of definition within lin-
guistics itself, e.g. Hjelmslev’s
2. Translation and commentary in another language, such as
Whorf on Hopi: ‘A first approximation to the meaning of
the . . . projective is “does with a forward movement” ’
(1956: 103) – a (presumably non-existent) example would be
a grammar of English in (non-westernized) Hopi
3. Deaf sign as metalanguage: the visual modality of sign gives it
a different semiotic potential (the semantic field of a sign is in
general greater than that of a lexical item; the sign is 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 different
Goals’, ‘glossed’ by two picture series each of three
events, one person kicking another – the one being
kicked either falls lower and lower (same person kicked
repeatedly) or stays in the same posture (different person
each time)
(b) Non-representational: e.g. music as metalanguage, e.g.
Chandola (1970: 145) suggests that many aspects of ragas
can be compared with language, and gives an example of
a musical pun – but music might lend itself more to
glossing dynamic aspects of the grammar such as informa-
tion structure
B. Metaphoric: the use of theoretical models
1. Interpretative metaphors for particular features of language;
e.g. Whorf’s interpretation of assertion in Hopi as ‘two grand
cosmic forms’ (1956: 59n.); transitivity and ergativity as
complementary modes of representing experience
2. A general theory of register and genre, from which semantic
categories can be derived in a principled way (Martin, in
press)
3. General theories of meaning in language, e.g. Lemke’s (1985)
concept of ‘making meaning’ in terms of activity structures
and thematic systems, and of ‘metaredundancy’
312
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 claimed
for it, of being able to interpret all other semiotic systems (and I see no
reason to assume that this is so), there are certainly limitations on the
ability of language to interpret itself. We may have to move outside
language, to some parallel or higher order semiotic which, since it is
not itself language, can be represented in language and then refracted
to become a metalanguage for representing language. All such inter-
pretation is ultimately circular; but in linguistics, we have tended to
operate within circles that are pathologically small. Until we can create
a greater distance between the semiotic object and the metasemiotic,
grammatical categories are bound to remain ineffable.
Appendix
Nigel at 5;4 (from Halliday (1984))
Nigel: Shall I tell you why the North Star stays still?
Father: Yes, do.
Nigel: Because that’s where the magnet is, and it gets attracted by the
earth; but the other stars don’t so they move around.
//2 shall I / tell you / why the / North / Star / stays / still//
//1 yes // .1 do//
//4 . because / that’s //1 where the / magnet / is and it //1 gets at/
tracted //1 by the / earth //4 . but the / other / stars //4 don’t so /
/4 they //1 move a/round //
1. Ideational
1.1 Experiential
Lexicogrammar: clause – transitivity [ Nigel ]
shall I tell you why the North Star stays still
Sayer Process: Receiver Cause Carrier Process: Range
Medium Verbal Beneficiary Medium relational
intensive
Actor Process: material
313
that ’s where the magnet is they move around

Identified Process: Identified Actor Process: Place
Token relational Value material
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 is (represents) North Star Location: where
circumstantial: locative magnet is
4
[[ ]] 冀
be at relational: attributive
circumstantial: locative 冁 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-semantic
relation
‘why?’ . . . because and but so
314
Network: experiential
attributive a
{
왘
material 31 identifying b
{
PROCESS relational 32
왘
clause TYPE verbal 33 intensive x
왘 circ.: place y
VOICE effective 41 possessive z
왘
middle 42
[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
INTERDEPENDENCY paratactic 123...
{
왘
(‘TAXIS’) hypotactic abg...
clause alaborating =
complex expansion 왘 extending +
LOGICAL-SEMANTIC enhancing ⳯
왘
RELATION
projection locution “
왘
idea ‘
Semantics: experiential and logical: informal gloss

2 North Star not move cause?
4 ( ) cause: magnet be-at North Star
5 add (effect): earth attract North Star
6 contrast: earth not attract other stars
7 effect: move other stars
‘a does not do x, because a has property p, but not-a have a not
property p, so not-a do x’
Nigel at 5;4 (from Halliday (1984))
Nigel: Shall I tell you why the North Star stays still?
Father: Yes, do.
Nigel: Because that’s where the magnet is, and it gets attracted by the
earth. But the other stars don’t so they move around.
315
2 Ideational semantics
2.1 Experiential
reactants (‘magnet’)
North Star
objects 왘 stars 왘
bodies 왘 other stars
earth
PHENOMENA locating (‘be at’)
processes 왘 move
motion 왘
Moving (‘do’) 왘 not-move
(‘stay still’)
force
alternatively:
positive
moving
(‘do’) { 왘
왘
negative
uncaused (motion)
caused (force)
2.2 Logical
contrastive (‘but’)
additive 왘
LOGICAL positive (‘and’)
SEMANTIC
왘
RELATIONS effect (‘so’)
causal 왘
unknown (‘why’)
cause 왘
known (‘because’)
316
Interpersonal semantics
COMMODITY (proposal) (proposition)
ROLE goods-&-services information
give offer statement
demand command question
give
{
ROLE IN
왘
EXCHANGE
demand
tentative (‘shall I? / will you?’)
VEIN
goods-&-service 왘 neutral (‘let . . . / let me!’)
COMMODITY (proposal)
왘
EXCHANGED definitive (I’ll . . . / you . . .’)
information
(proposition)
2 Interpersonal
Lexicogrammar: clause – mood
shall I tell you
Finite Subject
Mood Residue
do
Finite
mood
why the North Star stays still

WH/ Subject Finite
Adjunct Mood
Resi- due
that ’s
317
it gets attracted
the other stars don’t
they move around
Subject Finite
Mood Residue
Network: interpersonal
declarative 01
indicative 왘 yes/no 02
interrog 왘
WH- 03
{
Mood
왘
1st person 04
clause
imperative 왘 1st/2nd persons 05
2nd person 06
direct 11
MOOD
왘
PROJECTION
indirect 12
major
full 21
ELLIPSIS
왘
elliptical 22
[Nigel] 1 02.11.21 yes / no interrogative ‘offer, tentative’

[Father] 3 06.11.22 2nd person imperative ‘command, neutral’
[Nigel] 2 03.12.21 WH- interrog, indirect ‘question, projected’
4–7 01.11.21/22 declarative, full / ‘statement’

elliptical
318
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- topical Inter- topical
personal Rheme personal Rheme
Theme Theme
because that ’s where the magnet is

and it gets attracted by the earth
but the other stars don’t
so they move around
structural topical Rheme
Theme
1 discourse theme [first clause]: offer (topical) [speaker]

(interpersonal)
2 clause Themes: (interpersonal) question (topical) the North Star
4–7 clause Themes: (structural) because, (topical) the North Star,
and, but, so the other stars
319
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
왗 New New 왗 New
and it gets attacted by the earth but the other stars don’t so they
Focus Focus Focus Focus Focus

Given 왗 New 왗 New New Given New New
move around
Focus
왗 New
Network: theme systems

unmarked unmarked focus
TOPICAL TONICITY
왘 왘
THEME marked focus
marked
320
{
mood finite TONALITY unmarked (clause = info.unit)
왘 왘
— WH- marked thematic
INTER- (other)
PERSONAL modality
왘
THEME — fresh
STATUS
왘
vocative contrastive
왘
—
{
continuative
왘
—
TEXTUAL structural
왘
THEME —
conjunctive
왘
—
3.2 Cohesive
stays still 1
왗 왘 the North Star
|R
that (=there) 왗 2 왘 the magnet
|R |C
it 1
왗 왘 gets attracted 3
왗 왘 the earth
S A
S H E E
the other stars 왗 왘

1 don’t Ø 왗 왘
3 Ø
|R
1
move around 왗 왘 they
Key:
SA: synonymy – antonyms 1: Process + Medium
SH: synonymy – cohyponyms 2: Process + Location
C: collocation 3: Process + Agent
E: ellipsis columns: lexical chain
R: reference boxes: referential chain
321
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 information
derived from teacher (probably supplemented by charts / pictures;
possibly with referent in direct experience of night sky):
observational stars move, North Star doesn’t move;
problem: why is North Star exceptional?;
explanation: held by magnet.
5. 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).
322
Chapter Twelve
SPOKEN AND WRITTEN MODES

OF MEANING (1987)
1 Spoken language and education

It seems to me that one of the most productive areas of discussion
between linguists and educators in the past quarter century has been
that of speech and the spoken language. Twenty-five years ago, when
I launched the “Linguistics and English Teaching” project in London,
which produced Breakthrough to Literacy and Language in Use, it was still
rare to find references to the place of spoken language in school, or to
the need for children to be articulate as well as literate. Dell Hymes
had not yet introduced “communicative competence”; the words oracy
and orality had not yet entered the field (Andrew Wilkinson’s Some
Aspects of Oracy appeared in 1967); David Abercrombie (1963) had only
just published his ‘Conversation and spoken prose.’ Language, in
school, as in the community at large, meant written language.
The word language itself was hardly used in educational contexts. In
the primary school, there was reading and writing; in the secondary
school there was English, which meant literature and composition. Not
that a classroom was a temple of silence; but the kind of spoken
language that had a place, once a pupil had got beyond the infant
school, was prepared speech: reading aloud, drama, debating – language
that was written in order to be spoken, or at least was closely monitored
in the course of its production. Spoken language in its natural form,
spontaneous and unselfconscious, was not taken seriously as a medium
of learning.
First published in Comprehending Oral and Written Language, 1987. Orlando, FL: Academic
Press, pp. 55–82.
323
Among linguists, by contrast, the spoken language had pride of

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

students seldom do ask questions – not, that is, while they are occupying
their student role (i.e. in class). It is the teachers that ask the questions;
and when they do so, both question and answer may be somewhat
removed from the patterns of natural dialogue.
2 Complexity of natural speech

Already half a century earlier Franz Boas (1911) had stressed the
unconscious character of language, unique (as he saw it) among the
phenomena of human culture. Boas’ observation was to be understood
in its contemporary context as a characterization of the language system
(langue); not that, writing in 1911, he could have read Saussure’s Course
in General Linguistics, any more than Mathesius could have done; but
the unconscious was in the air, so to speak, and playing a critical role
in the conception of systems as regularities underlying human behav-
iour. But Boas may also have had in mind the unconsciousness of the
behaviour itself: the act of speaking (acte de parole) as an unconscious
act. The lack of conscious awareness of the underlying system, and
the difficulty that people have in bringing it to consciousness, are things
which language shares with other semiotic systems – for example, social
systems like that of kinship; what is unusual about language is the
extent to which even the manifestation of the system, the actual
process of meaning, remains hidden from observation, by performer
and receiver alike. In that respect talking is more like dancing, or even
running, than it is like playing chess. Speaker and listeners are of course
aware that the speaker is speaking; but they are typically not aware of
what he is saying, and if asked to recall it, not only the listeners but
also the speaker will ordinarily offer a paraphrase, something that is
true to the meaning but not by any means true to the wording. To
focus attention on the wording of language is something that has to be
learnt – for example if you are studying linguistics; it can be a difficult
and somewhat threatening task.
About 30 years ago, as a result of being asked to teach English
intonation to foreign students, I began observing natural spontaneous
discourse in English; and from the start I was struck by a curious fact.
Not only were people unconscious of what they themselves were
saying; they would often deny, not just that they had said something I
had observed them to say, but also that they ever could say it. For
example, I noticed the utterance it’ll’ve been going to’ve been being tested
every day for the past fortnight soon, where the verbal group will have been
325
going to have been being tested makes five serial tense choices, present in
past in future in past in future, and is also passive. This passed quite
unnoticed by both the speaker and the person it was addressed to; yet
at the time it was being seriously questioned whether a simple verb
form like has been being tested, which one can hear about once a week,
could ever occur in English. Five-term tense forms are, predictably,
very rare – one can in fact make a reasonable guess as to how rare, on
the basis of observed frequencies of two- and three-term tense forms
together with the constraints of the tense system; but they are provided
for within the resources of the spoken language. Another instance I
observed was they said they’d been going to’ve been paying me all this time,
only the funds just kept on not coming through.
Other things I noted regularly included present in present participial
non-finites like being cooking in I never heard you come in – it must have
been with being cooking; marked thematic elements with reprise pronoun,
as in that poor child I couldn’t get him out of my mind; and relatives
reaching into dependent clauses, such as that’s the noise which when you
say it to a horse the horse goes faster. These are all systematic features that
people are unaware that they incorporate in their speech, and often
deny having said even when they are pointed out; or at least reject as
unsystematic – after “I didn’t say it”, the next line of defence is “well
it was a mistake”. But of course it was not a mistake; it was a regular
product of the system of spoken English.
But perhaps the most unexpected feature of those early observations
was the complexity of some of the sentence structures. Here are two
examples from recordings made at the time:
(i) It’s very interesting, because it fairly soon is established when you’re
meeting with somebody what kind of conversation you’re having: for
example, you may know and tune in pretty quickly to the fact that
you’re there as the support, perhaps, in the listening capacity – that
you’re there, in fact, to help the other person sort their ideas; and
therefore your remarks, in that particular type of conversation, are aimed
at drawing out the other person, or in some way assisting them, by
reflecting them, to draw their ideas out, and you may tune in to this, or
you may be given this role and refuse it, refuse to accept it, which may
again alter the nature of your conversation.
(ii) The other man who kicks is the full-back, who usually receives the ball
way behind the rest of his team, either near his line or when somebody’s
done what the stand-off in the first example was doing, kicked over the
defenders; the full-back should be able then to pick it up, and his job is
326
usually to kick for touch – nearly always for touch because he’s miles
behind the rest of his side, and before he can do anything else with the
ball he’s got to run up into them, before he can pass it, because he can’t
pass the ball forward, and if he kicks it forward to another of his side
the other man’s automatically off-side.
And you get a penalty for that, do you, the other side?
Depending on whether it’s kicking or passing forward. Passing forward
– no, it’s a scrum. If you kick it forward and somebody else picks it up
that will be a penalty.
And if not, if the other side picks –
If the other side picks it up that’s all right; but the trouble is this is in
fact tactics again, because you don’t want to put the ball into the hands
of the other side if you can avoid it because it’s the side that has
possession, as in most games of course, is at an advantage.
Examples such as these were noteworthy in two respects. One was

that they embodied patterns of parataxis (combining with equal status)
and hypotaxis (combining with unequal status) between clauses which
could run to considerable length and depth. The other was that they
were remarkably well formed: although the speaker seemed to be
running through a maze, he did not get lost, but emerged at the end
with all brackets closed and all structural promises fulfilled. And this
drew attention to a third property which I found interesting: that while
the listeners had absorbed these passages quite unconsciously and
without effort, they were difficult to follow in writing.
3 Lexical density
These two examples have been around for a long time; so let me turn
to some recent specimens taken from recordings made by Guenter
Plum to whom I am indebted for drawing them to my attention. In
these spontaneous narratives Plum regularly finds sequences such as the
following:
1A I had to wait, I had to wait till it was born and till it got to about eight
or ten weeks of age, then I bought my first dachshund, a black-and-tan
bitch puppy, as they told me I should have bought a bitch puppy to
start off with, because if she wasn’t a hundred percent good I could
choose a top champion dog to mate her to, and then produce something
that was good, which would be in my own kennel prefix.
327
This displays the same kind of mobility that the earlier observations
had suggested was typically associated with natural, unselfconscious
speech – which is what it was. I asked myself how I would have
expressed this in writing, and came up with two rewordings; the first
(1B) was fairly informal, as I might have told it in a letter to a friend:
1B I had to wait till it was born and had got to about eight or ten weeks of
age; that was when I bought my first dachshund, a black-and-tan bitch
puppy. By all accounts I should have bought a bitch puppy at the start,
because if she wasn’t a hundred percent good I could mate her with a
top champion dog and produce a good offspring – which would carry my
own kennel prefix.
My second rewording (1C) was a more formal written variant:
1C Some eight or ten weeks after the birth saw my first acquisition of a
dachshund, a black-and-tan bitch puppy. It seems that a bitch puppy
would have been the appropriate initial purchase, because of the
possibility of mating an imperfect specimen with a top champion dog,
the improved offspring then carrying my own kennel prefix.
The aim was to produce a set of related passages of text differing
along one dimension, which could be recognized as going from “most
likely to be spoken” to “most likely to be written”. How such variation
actually correlates with difference in the medium is of course problem-
atic; the relationship is a complicated one, both because written /
spoken is not a simple dichotomy – there are many mixed and
intermediate types – and because the whole space taken up by such
variation is by now highly coded: in any given instance the wording
used is as much the product of stylistic conventions in the language as
of choices made by individual speakers and writers. Here I am simply
moving along a continuum which anyone familiar with English usage
can readily interpret in terms of “spoken” and “written” poles.
The kind of difference that we find among these three variants is
one that is often referred to as a difference of ‘texture’, and this familiar
rhetorical metaphor is a very appropriate one: it is as if they were the
product of a different weave, with fibres of a different yarn. But when
we look behind these traditional metaphors, at the forms of language
they are describing, we find that much of the difference can be
accounted for as the effect of two related lexicosyntactic variables. The
written version has a much higher lexical density; at the same time, it
has a much simpler sentential structure. Let us examine these concepts
in turn.
328
Table 1
Lexical Density of Texts 1A, 1B, and 1C
(1) (2) (1:2) (3) (1:3)
Lexical Running Clauses
items words
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 the
number of running words, and (2) as a proportion of the number of
clauses. Only non-embedded clauses have been counted (if embedded
clauses are also counted, then each lexical item occurring in them is
counted twice, since it figures in both the embedded and the matrix
clause – i.e., both in the part, and in the whole of which it is a part).
The figures are given to the nearest decimal.
As Jean Ure showed (1971), the lexical density of a text is a function
of its place on a register scale which she characterized as running from
most active to most reflective: the nearer to the “language-in-action”
end of the scale, the lower the lexical density. Since written language
is characteristically reflective rather than active, in a written text the
lexical density tends to be higher; and it increases as the text becomes
further away from spontaneous speech.
Jean Ure measured lexical density as a proportion of running words;
but as is suggested by the figures given above, if it is calculated with
reference to the number of clauses the discrepancy stands out more
sharply. Thus in the example given above, while the number of lexical
items remained fairly constant and the number of running words fell
off slightly, the number of clauses fell steeply: from 13, to 8, to 4. In
other words, the lexical density increases not because the number of
lexical items goes up but because the number of non-lexical items –
329
grammatical words – goes down; and the number of clauses goes down
even more.
Let us attempt a similar rewording the other way round, this time
beginning with a passage of formal written English taken from Scientific
American:
2A Private civil actions at law have a special significance in that they provide
an outlet for efforts by independent citizens. Such actions offer a means
whereby the multiple initiatives of the private citizens, individually or
in groups, can be brought to bear on technology assessment, the
internalization of costs and environmental protection. They constitute a
channel through which the diverse interests, outlooks and moods of the
general public can be given expression.
The current popular concern over the environment has stimulated
private civil actions of two main types.
2B is my attempt at a somewhat less “written” version; while 2C is in
another step nearer to speech:
2B Private civil actions at law are especially significant because they can be
brought by independent citizens, so enabling them to find an outlet for
their efforts. By bringing these actions, either as individuals or in groups,
private citizens can regularly take the initiative in assessing technology,
internalizing costs and protecting the environment. Through the use of
these actions as a channel, the general public are able to express all their
various interests, their outlooks, and their moods.
Because people are currently concerned about the environment, they
have been bringing numerous private civil actions, which have been
mainly of two types.
2C One thing is especially significant, and that is that people should be
able to bring private civil actions at law, because by doing this
independent citizens can become involved. By bringing these actions,
whether they are acting as individuals or in groups, private citizens can
keep on taking the initiative; they can help to assess technology, they
can help to internalize costs, and they can help to protect the 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 mainly
of two kinds.
330
Table 2 Lexical Density of Texts 2A, 2B, and 2C

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

“simple sentence” in traditional grammar. The second version (3B)
consists of four clauses (assuming that ended up feeling and tried to do are
each single predicators); but these too have to be transcribed as one
sentence, since they are related by hypotaxis – only one has 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 text
consists of four clauses where Text 3A consists of only one. It is that
Text 3B consists of a clause complex consisting of four clauses. The
clauses are not strung together as one simple sentence after another;
they are syntactically related. Looked at from the point of view of the
sentence structure, it is the spoken text that appears more complex
than the written one. The spoken text has a lower degree of lexical
density, but a higher degree of grammatical intricacy.
332
Figure 2
Let us return to Text 1, in its original spoken form (Text 1A). This
consisted of 13 clauses. However, these 13 clauses were not strung out
end to end; they were constructed into a small number of clause
complexes of mixed paratactic and hypotactic construction: arguably
just one clause complex throughout. Here is its interpretation as one
clause complex:
Figure 3
Sequences of this kind extend to a considerable length and depth in

parataxis and hypotaxis. A typical pattern is one in which both these
kinds of “taxis”, or interdependency, occur, with frequent alternation
both between the two and also among their various subcategories, as
in the example here. The relationships between successive pairs of
clauses in Text 1A are set out in Table 4.
333
Table 4
Other examples from the same source but from different speakers
show similar patterns; there are, obviously, individual differences
(including perhaps in the preference for one or other type of interde-
pendency), but the same free-flowing intricacy is noticeable all the
time, as in Texts 4–6:
4 Roy was always interested in dogs and unfortunately he’d never had the
opportunity to have a dog of his own, just because of circumstances –
where he lived and what not, and so I bought him a Shepherd pup, which
was supposedly, you know, pure-bred Shepherd, but unfortunately people
sold it because it didn’t have papers with it, so it was a ‘pup’.
5 Now how I got a German Shepherd was that I worked with a veterinary
surgeon, as I’ve told you before, and there used to be a lady that brought
her Shepherds along to the clinic and I used to admire them greatly, and
she said, ‘Well,’ she said, ‘if you get married I’ll give you one as a
wedding present,’ so immediately I bustled around looking for someone to
marry so I could get a Shepherd given to me for a wedding present, you
see, so that’s how that worked out well, not quite! However I got my
Shepherd and he was my first dog, mainly because when I was a youngster
I always wanted a dog but I lived with grandparents who wouldn’t have
dogs or cats and I was a very frustrated animal lover at that stage of the
game, so as soon as I got out on my own I sort of went completely berserk!
6 So we rang up the breeder, and she sort of tried to describe the dog to us,
which was very hard to do over the phone, so we went over to have a look
to see what they were like, and we bought Sheba, because at that stage
Bob was away a lot on semitrailers with the army and it used to get quite
334
bad with the exercises – you’d have prowlers and perverts through the
married quarters, so if we, you know, got a dog, which we could do
because it didn’t matter what sort of dog anyone had, it’d bark and they
wouldn’t bother us.
5 Types of complexity
Two distinct points need to be made here, and both of them run
counter to received attitudes towards spoken language. One is that
speech is not, in any general sense, ‘simpler’ than writing; if anything,
it is more complex. There are, of course, many different kinds of
complexity, and we have already noted one measure – lexical density
– whereby speech will appear as the simpler of the two. But the
patterns we have been illustrating, which are the patterns of the
organization of the clause complex, referred to above as grammatical
intricacy, would seem to be at least as central to any conception of
complexity; and in this respect, speech appears as the more complex.
The “syntactic complexity expected in writing”, with which Deborah
Tannen (1982) introduces her discussion of oral and literate strategies,
does not turn out to be a characteristic of written discourse.
Of course, there are many other variables. Some writers achieve
considerable intricacy in the structure of the clause complex; it can be
learnt and consciously developed as a style. Some forms of spoken
discourse, on the other hand, militate against it: rapid-fire dialogue
presents no scope for lengthy interdependencies – complex semantic
patterns can be construed between interactants, but usually without
being realized in syntactic terms. And the categories of “written” and
“spoken” are themselves highly indeterminate – they may refer to the
medium in which a text was originally produced, or the medium for
which it was intended, or in which it is performed in a particular
instance; or not to the medium at all, but to other properties of a text
which are seen as characteristic of the medium. So it is important to
indicate specifically which variable of discourse is being referred to,
when one variety is being said to display some distinctive characteristic.
My point here is to question the assumption that written language is
syntactically more complex than spoken, and to suggest that, as far as
one particular kind of syntactic complexity is concerned – the intricacy
(I do not want to call it “structure” because that assumes a particular
interpretation) of the sentence or “clause complex” – this is more a
characteristic of the most unconscious spontaneous uses of language.
The more natural, un-self-monitored the discourse, the more intricate
335
the grammatical patterns that can be woven. Usually, this kind of

discourse will be spoken, because writing is in essence a more conscious
process than speaking. But there are self-conscious modes of speech,
whose output resembles what we think of as written language, and
there are relatively spontaneous kinds of writing; spoken and written
discourse are the outward forms that are typically associated with the
critical variable, which is that of consciousness. We can use the terms
spoken and written language, to refer to the idealized types defined by
that variable.
Spoken and written language, then, tend to display different kinds
of complexity; each of them is more complex in its own way. Written
language tends to be lexically dense, but grammatically simple; spoken
language tends to be grammatically intricate, but lexically sparse. But
these buts should really be ands, because the paired properties are
complementary, not counterexpectative. It is hard to find a form of
expression which will show them to be such; I have usually had
recourse to metaphors of structure versus movement, saying for
example that the complexity of written language is crystalline, whereas
the complexity of spoken language is choreographic. The complexity
of spoken language is in its flow, the dynamic mobility whereby each
figure provides a context for the next one, not only defining its point
of departure but also setting the conventions by reference to which it
is to be interpreted.
With the sentence of written language, there is solidarity among its
parts such that each equally prehends and is prehended by all the
others. It is a structure, and is not essentially violated by being
represented synoptically, as a structural unit. With the clause complex,
of spoken language, there is no such solidarity, no mutual prehension
among all its parts. Its mode of being is as process, not as product. But
since the study of grammar grew out of writing – it is when language
comes to be written down that it becomes an object of study, not
before – our grammars are grammars of the written language. We have
not yet learnt to write choreographic grammars; so we look at spoken
language through the lens of a grammar designed for writing. Spoken
discourse thus appears as a distorted variant of written discourse, and
not unnaturally it is found wanting.
For example, Chafe (1982) identifies a number of regular differences
between speech and writing: writing is marked by more nominaliza-
tion, more genitive subjects and objects, more participles, more attrib-
utive adjectives, more conjoined, serial and sequenced phrases, more
complement clauses, and more relative clauses; all of which he sum-
336
marizes by saying, ‘Written language tends to have an ‘integrated’

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

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

without self-consciousness, then the clause complexes tend to flow
smoothly without you falling down or changing direction in the
middle, and neither speaker nor listener is at all aware of what is
happening. I recorded this kind of casual discourse many years ago
when studying the language spoken to and in the presence of a small
child, and was struck by its fluency, well-formedness, and richness of
grammatical pattern. Interestingly, the same feature is apparent at the
phonological level: spontaneous discourse is typically more regular in
its patterns of rhythm.
However, while the myth of the scrappiness of speech may have
arisen at the start from the kind of discourse that was first recorded, it
has been perpetuated in a different way – by the conventions with
which it is presented and discussed. Consider, for example, Beattie
(1983: 33):
Spontaneous speech is unlike written text. It contains many mistakes,
sentences are usually brief and indeed the whole fabric of verbal
expression is riddled with hesitations and silences. To take a very simple
example: in a seminar which I recorded, an articulate (and well-known)
linguist was attempting to say the following:
No, I’m coming back to the judgements question. Indeterminacy
appears to be rife. I don’t think it is, if one sorts out which are
counterexamples to judgement.
But what he actually said was:
No I’m saying I’m coming back to the judgements question (267) you
know there appear to (200) ah indeterminacy (1467) appears to be rife.
I don’t think it is (200) if one (267) if one sorts out which are
counterexamples (267) to judgement, I mean observing.
Here, the brief silences (unfilled pauses) have been measured in 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 structure
and feeling.
In other words: when you speak, you cannot destroy your earlier
drafts. If we were to represent written language in a way that is
comparable to such representations of spoken language, we should be
including in the text every preliminary scrap of manuscript or type-
script, with all the crossings out, misspellings, redraftings and periods of
338
silent thought; this would then tell us what the writer actually wrote.
Figure 4 is a specimen.
Figure 4 Written discourse
Now, there are undoubtedly research purposes for which it is

important to show the planning, trial and error, and revision work that
has gone into the production of a piece of discourse: it can have both
educational and clinical applications. This is as true of writing as it is of
speech: written material of this kind has been used in neuropsychiatry
339
for most of a century. But for many purposes the discarded first
attempts are merely trivial; they clutter up the text, making it hard to
read, and impart to it a spurious air of quaintness. What is much more
serious, however, is that transcribing spoken discourse in this way gives
a false account of what it is really like. It may seem a harmless piece of
self-gratification for a few academics to present spoken language as a
pathological phenomenon; one might argue that they deceive nobody
but themselves. But unfortunately this is not the way. Just when we
are seeing real collaboration between linguists and educators, and the
conception of “language in education” is at last gaining ground as a
field of training and research, it seems we are determined to put the
clock back to a time when spoken language was not to be taken
seriously and could have no place in the theory and practice of
education.
Let us recapitulate the argument. Speech and writing as forms of
discourse are typically associated with the two modal points on the
continuum from most spontaneous to most self-monitored language:
spontaneous discourse is usually spoken, self-monitored discourse is
usually written. We can therefore conveniently label these two modal
points “spoken” and “written” language. Spoken and written language
do not differ in their systematicity: each is equally highly organized,
regular, and productive of coherent discourse. (This is clearly implied
once we recognize them both as “language”.)
Discourse in either medium can be characterized by hesitation,
revision, change of direction, and other similar features; these tend to
arise when attention is being paid to the process of text production.
Since highly monitored discourse is typically written, these features are
actually more characteristic of writing than of speech; but because most
written text becomes public only in its final, edited form, the hesitations
and discards are lost and the reader is shielded from seeing the process
at work. Where they are likely to remain in is precisely where they
occur least, in the more spontaneous kinds of writing such as personal
letters. (Not all discourse features that are regarded as pathological, or
assigned negative value, are of this self-monitoring kind. One form of
discourse that has received a lot of critical attention is casual 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 that
would not seem deviant if we had a grammar that took into account
the specifically “spoken” resources of the linguistic system.)
Spoken and written language do differ, however, in their preferred
340
patterns of lexicogrammatical organization. Neither is more organized

than the other, but they are organized in different ways. We have
already identified the principal variable. Spoken language tends to
accommodate more clauses in the syntagm (to favour greater “gram-
matical intricacy”), with fewer lexical items in the clause. Written
language tends to accommodate more lexical items in the clause (to
favour greater “lexical density”), with fewer clauses in the syntagm.
(This does not imply, of course, that the average number of clauses
per clause complex will be greater in spoken language, because there
may also be a tendency towards very short ones, especially in dialogue.
It would be better to say that the greater the intricacy of a clause
complex the more likely it is to be a product of spontaneous speech.)
We must now return to this distinction in order to look through and
beyond it.
7 A closer look at the difference

Let us illustrate with another passage of written discourse (Text 7):
Thus the sympathetic induction of people into a proper and deep 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 deep
understanding of [[what Christianity is about ]] should not be bracketed
341
simply with the evangelizing aim [[to which I referred earlier]]. ||| It
is not absolutely incompatible with that aim, however, for the following
reason.||| [[What counts as indoctrination and the like]] depends upon
a number of criteria,|| to do with the degree [[to which a teacher fails to
mention alternative beliefs]] , the tone of voice [[used]], the lack of
sympathy for the criticism [[levelled at Christianity or Humanism]]
and so on. |||A dogmatic teacher or lecturer differs from an open
one.||| The non-dogmatic teacher may be tepid;|| the open one may
be fervent.||| Fervour and indifference are not functions of closedness
and openness.|||
To see how this lexical density is achieved, we can look at the first
clause. After the cohesive thus, it begins with a nominal group the
sympathetic induction of people into a proper and deep understanding of what
Christianity is about. The Head is induction; the Postmodifier consists of
a series of alternating embedded prepositional phrases and nominal
groups, mainly one inside the other, and ending with an embedded
clause:
| the sympathetic induction [ of [ people ] ]] [ into [ a proper and deep

understanding [ of [[ [ what ] < Christianity | is > about ]] ] ] ] |
group or phrase boundary |
embedded group or phrase [ ]
enclosed elements <>
(the prepositional phrase what . . . about is discontinuous, the items
Christianity and is being enclosed within it)
This nominal group contains a large amount of lexical information;

and if we take this passage as a whole we find that out of the 52 lexical
items the only ones that do not occur in nominal groups are bracketed,
simply, depends, do, and differ. It is a characteristic of written discourse
that most of the lexical information is encoded in nominal form: that
is, in nominal groups, with their structure potential of Head (typically
a noun or adjective), Premodifier (typically adjectives and nouns), and
Postmodifier (typically embedded phrases and clauses, which then have
further nominal groups inside them).
Not every instance of a nominal group has a complex structure, of
course; the remaining ones in this passage range from:
| the lack [ of [sympathy [ for [ the criticisms [[ levelled | at [ Christianity

or Humanism ] ]] ] ] ] ] |
342
which like the first one involves considerable embedding, to simple

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

or rankshift, where one element is a structural part of another, and the
dependency relation of ‘taxis’, where one element is bound or linked
to another but is not a part of it. Either of these relations can be
reduced to a form of the other one, but only at the cost of distorting
the nature of discourse.
The distinction between embedding and hypotaxis – between, for
example, the conviction [[that he failed]] / [of failure] and was convinced ||
that he had failed; or between the effect [of such a decision] would be [[ that
no further launchings could take place]] and if they decide that way || no
further . . . – is an important one; but it is really an instance, and a
symptom, of a more general and fundamental divergence. As always,
when we talk about these phenomena, and when we illustrate them,
they will appear as dichotomies: either this way or that. As always,
however, at least in the present context (but also in most issues that
have to do with language), they must be seen as tendencies – more or
less continuous variation along a line, but with most actual instances
(most texts, in this case) tending towards one pole or the other. The
divergent tendency that is manifested in the distinction of hypotaxis
and embedding is one that can be expressed in terms of the familiar
opposition of process and product. Written language represents
phenomena as if they were products. Spoken language represents
phenomena as if they were processes (see the discussion in Martin
1984b).
In other words: speaking and writing – each one makes the world
look like itself. A written text is an object; so what is represented in
writing tends to be given the form of an object. But when one talks,
one is doing; so when one talks about something, one tends to say that
it happened or was done. So, in Text 3 above, the written variant tells
the story in nouns: visit, sense, risk, attempt, action; whereas the spoken
version tells it in verbs: visited, ended up feeling, might get hurt, tried to do.
This is to look at it from the point of view of the writer or speaker.
For reader or listener, there is a corresponding difference in the way
the discourse is received. To the reader, the text is presented synoptic-
ally: it exists, spread out on the page. So the reader is predisposed to
take a synoptic view of what it means; behind it is a tableau – like the
pictures from which writing originally evolved. But when one is
listening, the text reaches one dynamically: it happens, by travelling
through the air. So the listener is predisposed to take a dynamic view
of what it means; behind it is a film, not a picture.
344
8 Grammatical metaphor
Where then in the linguistic system do spoken and written discourse
diverge? A language, if it is not written down, consists of three
interrelated subsystems: a semantic system (meanings), coded into a
lexicogrammatical system (wordings), recoded into a phonological
system (sounds). A language that has a writing system has an alternative
form of expression: visual symbols as well as sounds. In such a language,
a written text could, in principle, be a spoken text that has been
written down (a transcription); here the written version is a transcod-
ing of something that has already been coded in sound. Most writing is
not like this. Secondly, a written text could be an alternative
expression of a given wording: in this case meanings are coded as
words and structures (“wordings”), which are then expressed either in
sound or in writing. If this was the norm, there would be no systematic
difference between spoken and written texts; the medium would not
be a significant register variable. But there are such differences; so, to
some extent at least, spoken and written discourse must represent
alternative wordings. In this third case, meanings are coded either as
“speakable wordings” or as “writeable wordings”, the former appropri-
ate to the dynamic nature of the text process, the latter appropriate to
the synoptic nature of the text product. This is the sort of interpretation
we have been offering.
But is it the whole story? There is still a fourth possibility – that
speech and writing can diverge already at the semantic level, so that
spoken and written discourse embody different meanings. Is there any
sign that this can happen? It would of course be only a very partial
effect; no one has suggested that the two derive from different semantic
systems (or even two different lexicogrammatical systems, for that
matter). But we should consider the possibility that there is some
flowback into the meaning.
Consider the last sentence of Text 2, in its original written form
(2A):
The current popular concern over the environment has stimulated private civil
actions of two main types.
We “translated” it into something more speech-like as:
At present, people are concerned about the environment; so they have been
bringing quite a few private civil actions, which have been mainly of two
kinds.
345
But this could be wrong; it may have meant:

At present, people are concerned about the environment; so there have been
mainly two kinds of action being brought by private citizens.
There is no way of deciding: by reference to the spoken version, the
written version is simply ambiguous. Compare the following, also from
a written text:
A further complication was the 650-ton creeper cranes poised above the end
of each 825-foot arm.
Does this mean:
Above the end of each 825-foot arm there were poised 650-ton creeper
cranes, and they made the work more complicated.
or does it mean:
. . . and this made the work more complicated.
(i.e., not the cranes, but the fact that they were poised where they
were)? Another example is:
Slavish imitation of models is nowhere implied.
This could be reworded either as it is nowhere implied that models have
been slavishly imitated, or as . . . that models should be slavishly imitated.
Examples of this kind could be added to indefinitely; they arise
because nominal constructions fail to make explicit many of the
semantic relations that are made explicit in clause structure. Written
discourse conceals many local ambiguities of this kind, which are
revealed when one attempts a more “spoken” paraphrase.
But the final sentence of Text 2 illustrates another significant feature
of written language, which can be seen in the wording popular concern
over the environment has stimulated private civil actions. We reworded this
as people are concerned about the environment, so they have been bringing
private civil actions. The original is one clause with the verb stimulate
representing the Process; in other words, the thesis is encoded as a
single happening, and what happened was that A brought about B. But
A and B are themselves nominalized processes. The meaning of stimulate
here is as in pruning stimulates growth. The spoken version represents the
thesis as two distinct processes, linked by a relation of cause; cf. if the
tree is pruned, it will grow.
Here one kind of process has been dressed up by the grammar to
look like a process of a different kind – or, in this instance, two
346
processes, one mental and one material, have been dressed up as one
which is neither. This coding of a semantic relation between two
processes as if it was the single process is very common in writing; the
sentence immediately preceding Text 2A contained another example
of the same thing, here with the verb leads to:
A successful tort action leads to a judgment of damages or an injunction
against the defendant company.
But this is just one type of a more general phenomenon, something
that I call grammatical metaphor (Halliday 1985, Chapter 10). Written
language tends to display a high degree of grammatical metaphor, and
this is perhaps its single most distinctive characteristic.
Here are three further examples of grammatical metaphor taken
from various written sources, together with suggested rewordings
which are less metaphorical:
Issue of the specially-coded credit cards will be subject to normal credit
checking procedures.
“Credit cards have been specially coded and will be issued only when
credit has been checked in the normal way.”
Strong Christmas sales were vital to the health of the retail industry,
particularly in the present depressed climate.
“Unless many goods were sold at Christmas the retail industry would
not be healthy, particularly when the economy is depressed as it is
now.”
He also credits his former big size with much of his career success.
“He also believes that he was successful in his career mainly because he
used to be big.”
In all these examples nominalization plays a significant part, as it
does in many types of grammatical metaphor; so it is perhaps worth
stressing that nominalization is well motivated in English. It is not
simply a ritual feature that has evolved to make written language more
ambiguous or obscure; like the passive, which is another feature whose
functions are widely misunderstood, nominalization is an important
resource for organizing information. Take the example youth protest
mounted, which is not a headline but a complete sentence from a feature
article. We might reword this as more and more young people protested, or
young people protested more and more; but the only way to get the
combination of youth and protest as the Theme of the clause is by means
347
of a nominalization (not necessarily such a laconic one; it might have

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

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

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

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

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

assume that speech and writing play different and complementary parts
in the construction of ideologies (Hasan 1986), since each offers a
different way of knowing and of reflecting on experience.
Considerations of this kind are an essential element in any linguistic
theory of learning. The development of such a theory is perhaps the
most urgent task of educational linguistics; and certain components of
it can already be recognized: (1) the child’s construction of language,
from presymbolic communication through protolanguage to the
mother tongue; (2) the processing of new meanings into the system;
(3) the interaction between learning elements that are ready coded and
learning the principles of coding; (4) the relation between system and
process in language; (5) the unconscious nature of linguistic categories;
(6) the social construction of reality through conversation; (7) linguistic
strategies used in learning; (8) the development of functional variation,
or registers; (9) the relation between everyday language and technical
language; and (10) the development of generalization, abstraction, and
metaphor. The absence of any general theory of learning based on
language has been a significant gap in educational thinking and practice.
This provides an important context for our current concern, since the
complementarity of spoken and written language will certainly be a
central issue in any learning theory which has language as its primary
focus.
351
Chapter Thirteen
HOW DO YOU MEAN? (1992)
I realize that the title might well prompt someone to ask, ‘How do
you mean, “How do you mean?”?’ I could have written, ‘How are
meanings made?’ – although I prefer the more personalized version.
The question is meant theoretically; but, like so many theoretical
questions, it becomes relevant in practice the moment we want to
intervene in the processes we are trying to understand. And some
processes of meaning are involved in more or less everything we do.
I shall need to talk about two fundamental relationships, those of
realization and instantiation; so let me begin by distinguishing these
two. Instantiation I take to be the move between the system and the
instance; it is an intrastratal relationship – that is, it does not involve a
move between strata. The wording fine words butter no parsnips is an
instance, or an instantiation, of a clause. Realization, on the other hand,
is prototypically an interstratal relationship; meanings are realized as
wordings, wordings realized as sound (or soundings). We often use the
term to refer to any move which constitutes a link in the realizational
chain, even one that does not by itself cross a stratal boundary (for
example, features realized as structures); but the phenomenon of 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 as
metaredundancy, as this is defined by Jay Lemke (1985). Instantiation I
shall define by making reference to the observer; it is variation in the
observer’s time depth. Firth’s concept of exponence is the product of
these two relations: his “exponent” is both instantiation and realization.1
First published in Advances in Systemic Linguistics: Recent Theory and Practice, 1992, edited by
Martin Davies and Louise Ravelli. London: Pinter, pp. 20–35.
352
how do you mean?
form of
consciousness
action reflection
domain
of 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, obviously
including the so-called “higher” mammals. In humans, meaning devel-
ops, in the individual, before the stage of language proper; it begins
with what I have called “protolanguage”. So where does this mammalian
experience come from? It probably evolved out of the contradiction
between the two primary modes of experience, the material and the
conscious. Material processes are experienced as ‘out there’; conscious
processes are experienced as ‘in here’. We can see in observing the
growth of an individual child how he or she construes this contradiction
in the form of meaning. The child constructs a sign, whereby the one
mode of experience is projected on to the other. In my own observations
this took the form of what I coded as “v.h.p.s.” (very high-pitched
squeak), Nigel’s first sign that he produced at five months old; I glossed
it as “what’s that? – that’s interesting”. In other words, Nigel was
beginning to construe conceptual order out of perceptual chaos: ‘I am
curious (conscious) about what’s going on (material)’. This impact of
the material and the conscious is being transformed into meaning by a
process of projection, in which the conscious is the projecting and the
material the projected.2
But there are two possible modes of such projection – two forms
that the consciousness may take: one, that of reflection, ‘I think’, and
the other that of action, ‘I want’ – one the way things are, and the
other the way they ought to be. There are also, as it happens, two
domains of the material experience: one, that of ‘you and me’, and the
other that of ‘the rest (it, them)’. So, once the process begins (at around
eight months, with Nigel), what is construed into meaning is not a
single sign but a two-dimensional semiotic space constituting a sign
system (Figure 1). We can justifiably refer to such a sign system as a
353
“language”; but since it lacks the essential properties of an adult language,

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

back into the phenomenal via these two interfaces of content and
expression. (Transduction not transformation, because as Lamb (1964)
pointed out many years ago in transformation the original is lost, ceases
to exist. And again I am suggesting that we should conceive of it as
phenomenal rather than material, since both the ‘outer’ faces, that of
the content substance on the one hand and that of the expression
substance on the other, embody both the material and the conscious
modes of being.)
What is construed in this way, by this total semogenic process, is an
elastic space defined by the two dimensions given above: the ‘inner’
dimension of reflective / active, ‘I think’ as against ‘I want’, and the
‘outer’ dimension of intersubjective / objective, ‘you and me’ as against
‘he, she, it’. (Again, there is a naming problem here; we could say that
the ‘out there’ dimension is that of person / object, provided we
remember that “object” includes those persons ‘treated as’ object, i.e.
third persons. Instantially, this means any person other than whoever is
the interlocutor at the time; systemically it means any person not forming
part of the subject’s (the child’s) meaning group.)
This two-dimensional ‘elastic space’ defines what I have called the
mammalian experience. Obviously I am begging lots of questions by
calling it mammalian; but I am using this as a way of saying that it is a
potential we hold in common with other creatures, which I think is
rather important. It is a rich semogenic potential; but it is also constrained
in certain critical respects. In our own specifically human history, in
both phylogenetic and ontogenetic time, it comes to be deconstructed –
or rather deconstrued – and reconstrued as something else, this time in
the form of a potential for meaning that is effectively infinite, or at least
unbounded (to use an analogue rather than a digital mode of expression).
This reconstrual is the explosion into grammar. If we keep to the
‘interface’ conception, it is the evolution of an interface between the
interfaces. If we put it in terms of even more concrete metaphors, what
happens is that an entirely non-material (again, better: non-phenomenal)
system is slotted in between the two material / non-material (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 / action
and person / object (or intersubjective / objective). By ‘grammatical-
355
izing’ the process of meaning – reconstruing it so that the symbolic

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

l, m, n but with the redundancy of l, m, n with a, b, c; thus:
l, m, n a, b, c p, q, r (l, m, n a, b, c)
Why has it to be like this? Because there is not, in fact, a chain of
dyadic relationships running through the system. (If there was, we
would not need the extra stratum.) It is not the case, in other words,
that p l and l a, p, q, r is realized by l, m, n; but the system at l,
m, n is sorted out again for realization by a, b, c, so that what p, q, r
is actually realized by is the realization of l, m, n by a, b, c. This is the
fundamental distinction between redundancy and causality. If realization
was a causal relation, then it would chain: l is caused by a and p is
caused by l – it would make no sense to say “p is caused by the causing
of l by a”. But realization is not a causal relation; it is a redundancy
relation, so that p redounds with the redundancy of l with a. To put it
in more familiar terms, it is not that (i) meaning is realized by wording
and wording is realized by sound, but that (ii) meaning is realized by
the realization of wording in sound.
We can of course reverse the direction, and say that sounding realizes
the realization of meaning in wording:
p, q, r l, m, n (p, q, r l, m, n) a, b, c
For the purpose of phonological theory this is in fact the appropriate
perspective. But for the purposes of construing the ‘higher’ levels, with
language as connotative semiotic realizing other semiotic systems of the
culture, we need the first perspective. Thus when we extend ‘upwards’
to the context of situation, we can say that the context of situation s, t,
u redounds with the redundancy of the discourse semantics p, q, r with
the redundancy of the lexicogrammar l, m, n with the phonology a,
b, c. Thus:
s, t, u (p, q, r (l, m, n a, b, c))
(cf. Figure 2). Once the original protolinguistic redundancy has been
transformed into metaredundancy in this way, the relation becomes an
iterative one and so opens up the possibilities for construing, not only
the context of situation, but also higher levels such as Hasan’s symbolic
articulation and theme in verbal art, or Martin’s strata of genre and
ideology.
The metaredundancy notion thus formalizes the stratal principle in
semogenesis. What makes meaning indefinitely extendable is the evo-
lutionary change from protolanguage to language – whereby instead of
357
Figure 2 Metaredundancy
a simple plane with two interfaces to the material (the phenomenal),

we have constructed a semiotic space, a three-dimensional (potentially
n-dimensional) system in which there is a purely symbolic mode of
being between these two interfaces. It is this that we call grammar, or
more explicitly lexicogrammar. Without this semiotic space, situated in
the transduction from one purely symbolic mode to another, and hence
not constrained by the need to interface directly with the phenomenal,
we could not have a metafunctional organization in the grammar, and
we could not have the phenomenon of grammatical metaphor.7
The metaredundancy theory explains the ‘stratal’ organization of
language, and the semiotic principle of realization. It explains them
synoptically: by treating realization as a relation. Now, a system of this
kind could still remain fully closed: it could be a circular, self-regulating
system without any form of exchange with its environment. But a
language, as Lemke pointed out, is a dynamic open system; such systems
are not autostable, but metastable – they persist only through constantly
changing by interpenetration with their environment. And in order to
explain a system of this kind we have to complement our synoptic
358
how do you mean?
interpretation with a dynamic one. This leads us into the other critical
concept, that of instantiation.
Consider the notion of climate. A climate is a reasonably stable
system; there are kinds of climate, such as tropical and polar, and these
persist, and they differ in systematic ways. Yet we are all very concerned
about changes in the climate, and the consequences of global warming.
What does it mean to say the climate is changing? Climate is instantiated
in the form of weather: today’s temperature, humidity, direction and
speed of wind, etc., in central Scotland are instances of climatic
phenomena. As such they may be more, or less, typical: today’s
maximum is so many degrees higher, or lower, than average – meaning
the average at this place, at this time of year and at this time of day.
The average is a statement of the probabilities: there is a 70 per cent
chance, let us say, that the temperature will fall within such a range.
The probability is a feature of the system (the climate); but it is no
more, and no less, than the pattern set up by the instances (the weather),
and each instance, no matter how minutely, perturbs these probabilities
and so changes the system (or else keeps it as it is, which is just the
limiting case of changing it).
The climate and the weather are not two different phenomena. They
are the same phenomenon seen by two different observers, standing at
different distances – different time depths. To the climate observer, the
weather looks like random unpredictable ripples; to the weather
observer, the climate is a vague and unreal outline. So it is also with
language;8 language as system, and language as instance. They are not
two different phenomena; they are the same phenomenon as seen by
different observers. The system is the pattern formed by the instances;
and each instance represents an exchange with the environment – an
incursion into the system in which every level of language is involved.
The system is permeable because each instance redounds with the
context of situation, and so perturbs the system in interaction with
the environment. Thus both realization and instantiation are involved
in the evolution of language as a dynamic open system.
Now the relation of system to instance is in fact a cline, a continuous

zoom; and wherever we focus the zoom we can take a look into history.
But to know what kind of history, we have to keep a record of which
end we started from. To the system observer, history takes the form of
evolution; the system changes by evolving, with selection (in the sense
of ‘natural selection’) by the material conditions of the environment.
This is seen most clearly, perhaps, in the evolution of particular sub-
359
Figure 3 A model of semogenesis
systems, or registers, where features that are functionally well adapted

are positively selected for; but it appears also in the history of the system
as a whole once we look beyond the superficial clutter of random
fluctuations into the grammar’s cryptotypic core. To the instance
observer, on the other hand, history is individuation: each text has its
own history, and its unique meaning unfolds progressively from the
beginning. (Note that the probability of any instance is conditioned
both systemically (a register is a resetting of the overall probabilities of
the system) and instantially, by the transitional probabilities of the text
as a Markoff chain.) Given any particular feature – say grammatical
metaphor – we may be able to track it through both these histories, the
phylogenetic – its history as it evolves in the system; and what we might
call the “logogenetic” – its history as it is built up in the course of the
text. There is of course a third kind of history, the ontogenetic, which
is different again – the cladistic model here is one of growth. This too
is a mode of semogenesis; and we could follow through with the same
example, asking how grammatical metaphor comes into being in the
developmental history of a child. These are in fact the three modes or
dimensions of semohistory – the phylogenetic, the ontogenetic and the
360
how do you mean?
Figure 4 Postulated examples of semogenic evolution in relation to some

systems of Modern English. (Note. Those on the right are labelled merely for
identification, not in terms of their systemic features in the grammar.)
logogenetic; in the dynamic perspective, we can ask: how did this

meaning evolve, in the system? how did it develop, in the learner? and
how did it unfold, in the text?
In all these histories, the meaning potential typically tends to increase.
(Where it decreases, this is generally catastrophic: the language dies out,
361
Figure 5 Microfunctional systems (Nigel’s protolanguage)
or is creolized; the individual dies, or becomes aphasic; the text comes

to an end, or is interrupted.) Now, the mechanism of this increase of
meaning potential may be modelled in the most general terms as in
Figure 3. Nesbitt and Plum (1988; see also Halliday 1991) showed how
to do this in a corpus-based study of ‘direct speech and indirect thought’
(the intersection of speech projection and thought projection with the
interdependency system of parataxis and hypotaxis). This and other
postulated examples of evolutionary semogenesis are set out in Figure
4. These involve relations between strata (semantics realized in 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. Thus
where a closed system is self-regulating (autostable) and circular, an
open system is other-regulated (metastable) and helical. And it is through
the combination of these two relations or processes, instantiation on the
one hand and realization on the other, that the system exchanges with
its environment, creating order in the course of this exchange and so
increasing its potential for meaning.
Thus the possibility of meaning – of acting semiotically – arises at
the intersection of the material (or phenomenal) with the conscious, as
the members of a species learn to construct themselves (“society”) in
action and to construe their experience in reflection. These two dimen-
362
how do you mean?
Figure 6 Nigel’s first stratifications (age one year two months)
sions – action / reflection, and you + me / other – define a semiotic field.

At first this is a plane, a rubber sheet so to speak, elastic but 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 the
protolanguage shape themselves into a sign system as they cluster in the
four quadrants of this semiotic plane; there is thus already a proto-
system network, which we could set up in an idealized form as in
Figure 5. (I use the variables that emerged from my own Nigel data;
this should be compared with studies by Clare Painter (1984) and by
Jane Oldenburg (1987), where other systemic variables may appear
more prominent.9) We must leave open the question of what variables
are the ones in respect of which the protolinguistic system is typically
construed, but I think it will be a fairly small set. Nigel’s seemed to be
(1) in instrumental: polarity, (2) in regulatory: intensity, (3) in 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 an
n– dimensional space, as the activity of meaning becomes dialogically
dynamic and metafunctionally complex: that is, it becomes possible to
mean more than one thing at once, and to construe meanings into text.
I have written elsewhere about how Nigel took the first step in this
transformation, using the semogenic strategy already described (combin-
ing two functionally distinct variables); it is summarized in Figure 6.
363
Here for the first time Nigel is selecting in two systems of meaning at
once, and by this token the initial move into grammar has been made.
Through the second year of life this new stratified system will gradually
replace the protolinguistic one, and all meanings (except for a few
protolanguage remnants that persist into adult life like hi! and ah! and
yum! and ouch!) will come to be stratally and metafunctionally complex.
So in Nigel’s first exemplar, just cited, we have (i) proto-metafunctions
(proto-ideational – different persons; proto-interpersonal – seeking /
finding), and (ii) proto-strata, with the meaning ‘first’ construed as
wording (the ideational as contrasting names; the interpersonal as
contrasting mood) and ‘then’ (re)construed as sounding (names as
articulation, mood as intonation). At this second interface the child can
now combine the segmental and the prosodic choices, in this way both
realizing and also iconically symbolizing the two different modes of
meaning that are combined at the first interface. The resources for
making meaning are now in place.
It is probably not a coincidence that, as the ideational grammar
evolved, so in the system of transitivity the field of processes was
construed into different process types along precisely the lines that (if
my understanding is right) went into the making of meaning in the first
place. If meaning arises out of the impact of the conscious and the
material, as mutually contradictory forms of experience, then it is not
surprising that when experience is construed semantically, these two
types of process, the material and the conscious, should come to be
systematically distinguished. But there is a further twist. The semogenic
process, as we saw, involves setting up a relationship between systems
such that one is the realization of the other – that is, they stand to each
other in a relation of Token and Value. This Token–Value relationship
is set up at both interfaces, and it is also what makes it possible to prise
the two apart and wedge in a grammar in between. Here then we find
the third of the kinds of process construed by the grammar: the relational
process, based on identifying a Token with a Value. The grammar of
natural language, in its ideational metafunction, is a theory of human
experience; thus it may reasonably be expected to take as its point of
departure the very set of contrasts from which its own potential is
ultimately derived.
Let me return once again, finally, to the suggestion that meaning is a
mode of action engendered at the intersection of the material (or
phenomenal) and the conscious, as complementary modes of experience.
Now, the effect of this impact is to construe order. By the act of
meaning, consciousness imposes order on the phenomena of experience.
364
how do you mean?
When protolanguage evolves into language, with the stratal dimension

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

experience in terms of indeterminacy, of continuity and of flux. Now
by comparison with the languages of science, the ordinary evolved
language of everyday life has many of these properties. It is oriented
towards events rather than objects, and is in many respects fluid and
indeterminate. But it is important to be aware that even our most
unconscious everyday language also imposes an order which we may
need to re-examine and to deconstruct. To return to the weather: we
can say it’s raining or we can say it’s snowing – but we have to decide
between them. We may accommodate an intermediate form, it’s sleeting,
this cuts up the continuum more finely, but still into discrete parts –
‘rain or sleet or snow’.12 (Contrast in this respect the semantics of sign,
such as Auslan, which often allows a more continuous interpretation of
experience – though of course it is constantly being modified under the
influence of spoken language (Johnston 1990).) In other words, while
the order – that is, the particular mix of order and chaos – that our
grammar construes has served us well, and continues to do so, it is not
necessarily the most functional for all times and all circumstances;
especially at times of rapid change like the present, we may need to
hold it up to the light and see how it works.13 It is easy to remain
unaware of the stories our grammar is telling us.
One thing I have been trying to do, in this paper, is to use the
grammar to think with about itself. Not just in the usual sense, of using
language as its own metalanguage; of course I am doing that, because
there is nothing else I can do. I mean this more specifically in the sense
of using what I have called the grammatics – the concepts that we have
developed in order to interpret the grammar – as a means towards
understanding the nature and evolution of language as a whole. The
strategy is that of treating language as ‘other’ – as if it was a different
kind of semiotic that the grammar was being used to explore.14 Thus I
have found it helpful to think of meaning as the way consciousness (that
is, mental processes), by a type of projection, construes a relationship
(that is, a Token = Value identity, or a nested series of such identities)
between two sets of material processes (those of our experience, at one
end, and those of our bodily performance – gesture, articulation – at
the other). I do not know how useful anyone else will find this strategy.
But at least it is something I can answer with, the next time anyone
says to me, “How do you mean?”
366
how do you mean?
Notes
1. For Firth’s concept of ‘exponence’ see especially his ‘Synopsis of Linguistic
Theory’ in Firth (1957a).
2. I am speaking here of phylogenesis; but the process is recapitulated in the
growth of the individual, where it can be observed in the form of
behaviour. A child experiences certain phenomena as ‘out there’ – as lying
beyond the boundary between ‘me’ and ‘non-me’: some perturbation seen
or heard, like a flock of birds taking off, or a bus going past, or a coloured
light flashing. At the same time, he also experiences a phenomenon that
is ‘in here’: his own consciousness of being curious, or pleased, or
frightened. At first these two experiences remain detached; but then
(perhaps as a result of his success in grasping an object that is in his line of
sight – in Trevarthen’s terms, when “pre-reaching” becomes reaching,
typically at about four months) a spark flies between them by which the
material is projected on to the conscious as ‘I’m curious about that’, ‘I like
that’ and so on. Now, more or less from birth the child has been able to
address others and to recognize that he is being addressed (Catherine
Bateson’s “proto-conversation”). The projection of the material on to the
conscious mode of experience maps readily on to this ability to address an
other; and the result is an act of meaning – such as Nigel’s very 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’ in
Lemke (1984).
7. It is impossible to have metaphor in a protolanguage at all, unless one
chooses to call “metaphor” (or perhaps “proto-metaphor”) what is taking
place when, for example, Nigel transfers a particular sign [gωg gωg gωg]
from ‘I’m sleepy’ to ‘let’s pretend I’m going to sleep’. See Halliday (1975:
Chapter 2).
8. The analogy should not, of course, be pressed too far. Specifically, while
367
the relation of instantiation holds both for language / speech (langue /

parole) and for climate / weather, that of realization does not. It could be
said that climate is in fact modelled as a stratified system (in the semiotic,
not the atmospheric sense!); but this would be using ‘stratified’ with a
significantly different meaning.
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 so
much.” “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).
368
Chapter Fourteen
GRAMMAR AND DAILY LIFE:

CONCURRENCE AND COMPLEMENTARITY
(1998)
Let me first say what I mean by “grammar” in the title of the paper. I
mean the lexicogrammatical stratum of a natural language as tradition-
ally understood, comprising its syntax and vocabulary, together with
any morphology the language may display: Lamb’s “lexical system”, in
his current (1992: Chapter 5) ‘three-level architecture’ – in common-
sense terms, the resources of wording in which the meanings of a
language are construed. And here I have in mind particularly the
evolved, spontaneous grammar that construes the discourse of daily life.
This is not to exclude from the picture the elaborated grammars of
scientific and other metalanguages; but these can only be understood as
what they are: an outgrowth, supported by design, of the original
grammar that is learnt at mother’s knee and on father’s shoulders.
Now English is not very efficient at creating technical nomenclature,
since it tends to confuse the study of a phenomenon with the
phenomenon itself. So while the term “grammar” is commonly used
in the way in which I have defined it, to mean the wording system,
the central processing unit of a natural language, it is also used
indiscriminately to mean the study of that system: grammar2 meaning
‘the study of grammar1’. Since the study of language is called “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 by
Teun A. van Dijk. Amsterdam: John Benjamins, pp. 221–37.
369
a resource for meaning, the critical functioning semiotic by means of

which we pursue our everyday life. It therefore embodies a theory of
everyday life; otherwise it could not function in this way. A grammar
is a theory of human experience: or rather, let us say, it includes a
theory of experience, because it is also something else besides. Like any
other theory, a grammar is something to think with. It is through
grammar that we make sense out of our experience, both of the world
we live in (what we experience as taking place “out there”) and of the
world that lives in us (what we experience as taking place “in here”,
inside our own consciousness), construing a “reality” such that the one
can be reconciled against the other (Matthiessen 1991; Halliday and
Matthiessen 1999).
During the past twenty years leading neurobiologists, such as Harry
Jerison and John Allman, have been investigating the way the brain
evolved; and they explain its evolution as the evolution of the organ-
ism’s resource for constructing reality. Changes in the ecological
environment 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 external
environment. The second was the evolution of language, which added
a new dimension to reality, that of introspective consciousness; this
latter step is associated with the development of the prefrontal zone of
the cortex, allowing a major reorganization of neural circuitry (Dunbar
1992). Linguists can show that the corresponding unique feature of
human language, distinguishing it from semiotic systems of other genera
and species, is that it has a grammar, an abstract stratum of coding in
between the meaning and the expression. Grammar is what brings
about the distinctively human construction of reality; and by the same
token, grammar makes it possible for us to reflect on this construction.
As a teacher I have often said to my students that they should learn
to ‘think grammatically’. By this I mean that they should use the
unique power of the human brain to reflect on the way their experi-
ence is construed in their grammar: use grammatics to think about
what grammar thinks about the world. I suggest they might do this
with problems of any kind, such as relationships with family and
friends, or whether to go for the job that pays more or for the one
they would more enjoy. Let me give a small example of what I mean
by thinking grammatically. You’re feeling a bit down. What’s the
matter, someone asks. ‘I have a headache.’ So how does the grammar
construe your unfortunate condition? Of course, you construed it,
using your grammatical potential; but you did so quite unconsciously,
370
grammar and daily life
in the way that it has been done countless other times by countless
other people, so it is reasonable to talk about the condition being
construed ‘by the grammar’.
In I have a headache the grammar construes a kind of thing, called an
ache; it then uses a part of the body to classify this thing, setting up a
taxonomy of aches including stomachache, backache and various others.
(Not all the parts of the body are allowed to ache, however; you cannot
have a footache or a thighache.) The grammar then sets up a configuration
of possession between the ache and some conscious being, in this case
the speaker I. The speaker becomes the owner of one specimen of that
complex class of things. It is not a prototypical form of possession; the
possessor does not want the thing possessed but cannot get rid of it –
cannot give it away, or put it back where it came from. Why then
does the grammar not favour my head aches; or my head’s aching? – in
which the aching is a process, a state of being, rather than a thing, and
the entity involved in that state of being is my head rather than me.
The grammar has no trouble in constructing the clause my head aches;
yet it is not the most usual way in which the experience is worded.
Why is I have a headache preferred instead?
In English, as in many other languages (though not all), there is a
particular meaning associated with being the first element in the clause.
What is put first is being instated by the speaker as the theme of the
coming message; it is the setting for the information that follows (Fries
1995). This pattern of the clause, a structure of “Theme + Rheme”,
was apparently identified by the earliest rhetorical grammarians of
ancient Greece, the sophists, who seem to have recognized in the
thematic organization of the clause a potent resource for constructing
legal and political discourse. In modern times it was first investigated in
detail by Mathesius, the founder of the Prague school; it is a particularly
prominent feature of English, appearing not only in the clause but also
as a “fractal” pattern in both smaller and larger structures – inside word
groups, both nominal and verbal, on the one hand and extending over
a nexus of clauses on the other. The following example, taken from
natural conversation, shows thematic predication of a whole clause
complex (from Svartvik and Quirk 1980: 304):
. . . in my last year at college I said to myself: “You want to do

applied chemistry, right? What industries are now just being born
which will blossom in the next quarter of a century, which is going
to be my working lifetime?” And I said “Plastics, sure as the nose
on your face. I’m going to get into this.” . . .
371
I’m dazzled, you know . . . It’s being able to see your working life
will span a period in which so-&-so is the topmost industry which I
find so dazzling.
Now if I say my head aches, the first element in that clause is my head:
I have constructed a message in which my head is enunciated as Theme.
My head is instated as what I want to elaborate on. But it isn’t; I’m the
one that’s suffering, so the Theme of the clause should more 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 a
thing separated from myself, something that I possess, with my head
identified as its location: I have an ache in my head. Better still, if my
head is used as a classifier, the ache and its location become a single
complex thing; and this now occupies the culminative position in the
clause: I have a headache. The flow of information here is very different
from that of my head’s aching.
If this was just a feature of the grammar of localized aches and pains,
it might remain a curiosity, a special effect rather than a principle. But
this pattern has evolved in English as the prototypical form for
construing bodily qualities and states; rather than her hair is long, his
throat is sore, we tend to say she has long hair, he has a sore throat, putting
the person rather than the body part into the thematic role.1 And in
certain other languages where initial position is thematic we also
regularly find the person, rather than the body part, lodged at the
beginning of the clause. The overall patterns are of course different: in
particular, there may be no strong bond between Theme and Subject,
and this makes it clear that the relevant function is that of Theme. We
can give examples from Chinese, Russian and French. In Chinese it is
possible to say wǒdi tóu tèng ‘my head aches’, where as in the English
wǒdi tóu ‘my head’ is a single element in the clause and so functions as
the Theme. The preferred form, however, is wǒ tóu tèng ‘me the head
aches’, where the ‘head’ is detached from the personal pronoun; wǒ
‘me’ and tóu ‘head’ are now independent elements in the clause and
only the first one, wǒ, is thematic. Again, this is the typical pattern for
all such expressions in Chinese: tā tóufǎ cháng ‘her the hair (is) long’, tā
hóulóng tòng ‘him the throat (is) sore’ and so on. In Russian, likewise,
one can say moja golova bolit ‘my head aches’; but this also is not the
preferred form. Russian however displays a different pattern: u menja
golova bolit ‘at me the head aches’, where again it is the ‘me’ that has
thematic status. In French instead of ma tête me fait mal ‘my head is
372
hurting me’ one can use possession as in English: j’ai mal à la tête ‘I
have an ache at the head’. French also has a further device, of detaching
the Theme altogether from the structure of the clause, and announcing
it as a key signature at the beginning: moi j’ai mal à la tête ‘me I’ve got
an ache at the head’. Neither Chinese wǒ nor Russian u menja nor
French moi is Subject; what they have in common is the status of
Theme.
At this point we might think once more of the sufferer and say to
him or her: pity you’ve got a headache. But try de-construing this, in
the grammar, and then re-construing it – rewording it – as my head
aches; or better still my head’s aching, which makes it an external rather
than an internal phenomenon. This is rather less self-centered: it is no
longer a fact about me, and my inner self, but an external fact about
my head. This won’t make the headache go; but it does put it in its
place. It has now become a problem of my head, which is just one part
of my physical make-up. One might offer this as a form of logotherapy,
a kind of grammatical acupuncture. But here I just want it to serve as
an instance of “thinking grammatically”.
Thus the grammar enables us, unconsciously, to interpret experience;
and the metagrammar, or grammatics, enables us to reflect consciously
on how it does so. The grammatics, of course, is part of a more general
theory of meaning: of language as a semiotic system, and of other
semiotic systems brought into relation with language. Without such a
general theory, the excursion into other languages is no more than a
piece of tourism; it assumes significance only when we can show how
this small corner of experience is construed in relation to the meaning
potential of each language as a whole.
But this requires much more than a purely local explanation. Taking
a fragment of the grammar of daily life, and exploring it cross-
linguistically in this way, still leaves it as an isolated fragment, detached
from its environment in the overall system of the language. Yet this is
the critical environment to take into account. The grammar construes
a unitary semantic space, elastic and many-dimensioned; and whatever
aspect of the grammar we are considering (such as the selection of
person as Theme, in the examples above), there will usually be various
other grammatical features, many of them not obviously related in any
formal sense, which are associated topologically within this semantic
space (cf. Martin and Matthiessen 1992). Such features may cluster into
a recognizable syndrome, needing to be interpreted not piecemeal but
as a whole: this is the principle of “frames of consistency” as formulated
by Whorf. Illustrations of this phenomenon may be found in Hasan’s
373
(1984b) ‘Ways of saying, ways of meaning’, where she shows how the
grammar of Urdu construes experience as collectively shared; and in
Martin’s (1988) account of “grammatical conspiracies” in Tagalog. If
we are comparing the different “realities” of one language with
another, it is the syndrome rather than the single feature that is likely
to be significant.
Side by side with such frames of consistency, however, there are also
frames of inconsistency: regions where the grammar construes a pattern
out of tensions and contradictions – where the different “voices” of
experience conflict. To put this another way, the grammar’s theory of
experience embodies complementarity as well as concurrence. Meta-
phorically the grammar is representing the fact that human experience
is too complex, and has too many parameters, to be construed from
any one angle alone. It is the combination of these two perspectives –
concurrence and complementarity – that is the salient characteristic of
the grammar of daily life.
Let me first try to illustrate the complementarity, and then use this
as a point of departure for exploring concurrence, looking at a more
general syndrome of features within which the earlier, more particular
example might be located. Many grammars (perhaps all) make a rather
clear distinction between the two fundamental modes of human
experience referred to above: between what we experience as taking
place in the world outside of ourselves and what we experience as
processes of our own consciousness – seeing and hearing, liking,
disliking, fearing, hoping, thinking, knowing, understanding and the
like. In English, the conscious or mental processes differ from the
other, material kind in various respects: (1) they have a less exact
present time; (2) they presume a conscious being taking part; (3) they
do not fall within the scope of ‘doing’, and (4) they can project – that
is, they can construe any meaning as taking place in someone’s
consciousness (as “direct or indirect thought”). In addition, these inner
processes display another feature not found with the grammar of
processes of the external, material type: they are bi-directional. 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 fact
contradictory constructions of the same class of phenomena. Inner
experience is complex and difficult to interpret; the grammar offers
two complementary models, one with the Senser in the more active
role (by analogy with material processes), one with the Senser appearing
374
to be acted upon. Each of these brings out different agnate forms; the
grammar of daily life, in English, accommodates both.
In late Middle to early Modern English the common verbs of
consciousness such as like and think changed their allegiance from the
one pattern to the other: from ‘it likes/thinks (to) me’ to I like it, I
think so. This happened at about the same time as the emergence of the
pattern discussed earlier: I have a headache, etc. For very general
processes of consciousness the grammar came to favour the type of
construction in which the Senser, the participant credited with 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 the
grammar of all inner processes and physiological states tended to orient
the message towards the human, or humanlike, participant – perhaps
with ‘I’, the individual self, as the prototypical member of this class.
This in turn leads us to another feature. At the same period of
history another shift took place affecting processes of the external kind,
those experienced as happening ‘out there’. In earlier English the
grammatical Subject in such processes had been overwhelmingly the
active participant, whether human or not (in fact the distinction
between human and non-human, or conscious and non-conscious,
plays no part in the construction of these processes of the external
world). Thus in an arrow pierced his eye the arrow was the natural Subject,
and remained Subject even if the narrative required the thing acted on
to function as Theme. To use a constructed example, the pattern was
that of:
The king fell to the ground; his eye an arrow had pierced.
with the Actor remaining as Subject even when displaced from initial
position in the clause. Subsequently, as already noted, this bond
between Subject and Actor was deconstructed and replaced by a
different bond, that of Subject with Theme; this gave the modern
pattern:
The king fell to the ground; his eye had been pierced by an arrow.
This change led to an increase in the frequency of passive verbs, which
was followed by a change in the tense system as passive tenses caught
up with the active ones; and various other changes took place besides.
What this new alignment of grammatical forces amounted to was that
relatively less prominence was being given to the structure of the
experience – which partner is the doer and which the done-to, so to
375
speak; and relatively greater prominence to the structure of the

message – which part is the theme, and which part is the new
information to be attended to. Without trying to go into all the
components of the picture, let me refer briefly to three related
developments.
First, the grammar developed a battery of resources such that any
representation of a process can be construed in all possible patterns of
information flow; given ‘an arrow pierced his eye’ we have not only
his eye was pierced by an arrow but also what pierced his eye was an arrow,
what the arrow pierced was his eye, what the arrow did to his eye was pierce it,
and so on. These evolved as different ways of dividing the clause into
a thematic portion and the rest. But the construction of the message is
more fluid and more complex than that simple formulation suggests.
The flow of information is made up of two distinct currents: a linear
movement from Theme to Rheme, and an oscillation between Given
and New which is not encoded in the sequence but in which the
“New” – the part presented by the speaker as ‘to be attended to’ –
tends to build up at the end. And just as various features in the grammar
conspire to construe the Theme, so various others come together in
construing the resources for the New; and this leads in to the second
of the three developments being mentioned here.
Secondly, then, another feature of Modern English grammar is the
motif of the “phrasal verb”; we can say he invented the whole story, but
we prefer he made the whole story up; similarly you left the important part
out (instead of you omitted . . .), they’ve taken the furniture away (instead
of they’ve removed . . .), and so on. This is the grammar’s way of making
the happening the main item of news. The news tends to come at the
end of the clause; but the happening is typically a verb, and if there are
two parties to it – an Actor and a Goal, say – it is hard to get the verb
at the end: we cannot say he the whole story invented (we can say the
whole story he invented; but that changes the thematic balance by
marking the Theme). What the “phrasal verb” construction does is to
split the verb into two parts, so that the second part of it can come at
the end: he made the whole story up is the grammar’s suppletion for *he
vented the whole story in.
Thirdly, there is an analogous pattern whereby one of the other
elements in the clause – one which could but would not necessarily
come at the end – is marked out for news value by having a preposition
added to it. If you want to tell me that you supported your brother
financially you could say I gave my brother a lot of money; but if the
observation is made to explain why you now need to borrow from
376
me, you say I gave a lot of money to my brother. The preposition to makes
explicit your brother’s role as a participant in the process, and is added
just in those positions which are prominent in the information flow
(likewise if the brother appears as a marked Theme: to my brother I
gave a lot of money). It is precisely this same principle which adds by to
the Actor when the clause is passive: his eye had been pierced by an arrow/
by an arrow his eye had been pierced.
All the features I have sketched in here are features of the grammar
of daily life: some more global, some more local, but all of them
characteristic of unconscious, spontaneous, everyday linguistic encoun-
ters. These, and others that could be added, form a syndrome, a
concurrence of related developments, that has helped to shape the
meaning potential of Modern English, giving the language its charac-
teristic flavour – that “certain cut”, in Sapir’s terms, which makes each
language unique. What all these have in common is that they tend
towards giving greater prominence to the organization of discourse as
a flow of information, making more explicit how each element is to
be construed as part of a message. As a corollary to this, less prominence
is given to the experiential patterning, much of which is in fact left
implicit once the concern with the message begins to take over. Most
of these effects are fairly recent in history; they reflect the changing
social conditions of the language over the past five hundred years. Or
rather: they do not reflect them – they help to bring them about.
These features in the grammar construe the kind of discourse that can
be addressed to a stranger, who does not necessarily share the same
expectations and norms of interaction. They can be written down in a
book that is going to be printed in thousands of copies and read by
people who have never met the author and do not even know who he
is. In other words, they are features of a standard language: a form of
discourse in which the flow of information will typically be rendered
explicit rather than being taken for granted. (Interestingly, many of
these changes appear not to have taken place in the surviving British
rural dialects.)
Effects like these are not the result of sudden catastrophic changes.
They are trends and tendencies in a long process of evolution; and at
any given time they are quantitative – changes in the relative frequency
with which this or that pattern is selected from within the system. The
grammatics is thus a theory of probabilities, in which possible/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 when
something becomes impossible there is no more choice. So, for
377
example, I have a headache is an instance of what is now the more

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

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

What is she sewing?
She’s sewing her old jacket.
Mum is the doer and the jacket is the done-to. This is a configurational
model; there is no re-entry to the choice of agency. Then there’s a
snap, and the grammar takes up the story again:
Bother! the thread snapped.

What snapped it?
The machine snapped it.
Yes? Who made the machine snap it?
I did, of course.
What made you make the machine snap it?
My own impatience, I suppose.
and so on. This second theory says that there is a Medium, an entity
through which the process is actualized (here the thread), plus, option-
ally, something else as Causer that brings it about. This is an iterative
model; here the agency relation is construed in such a way that it can
recur.
Thus there are two ways of looking at a process: one according to
which participant a acts, and the action may (or may not) extend to
another participant x (a is the constant, x the variable); the other
according to which participant x “eventuates” (that is, permits the
process to eventuate), and the event may (or may not) be brought
about by another participant a (x is the constant, a the variable). The
first of these (let us call it type A) is the transitive theory of processes,
the second (type X) is the ergative; and probably all languages embody
some tension between the two. Transitive and ergative are two points
of view on the same phenomenon, that of the nature of material
processes and the relationship of the participants to the process and to
each other; but they are also two distinct phenomena – some processes
pattern ergatively and others transitively (cf. Halliday 1967–68; Davidse
1992). This constitutes another strand in the pattern of changes that
have been taking place in English: type X has tended increasingly to
prevail over type A.
Let us follow this up in a related corner of the grammar. When I
last worked in the United States I was living in Orange County; I
frequently travelled on the local bus services, and there was a notice on
the buses which read:
380
Federal law prohibits operation of this bus when any passenger is

forward of the standee line.
If you are standing, on the bus, you are a standee. Why not a stander?
You are a passenger, not a passengee (and if you cannot get on the bus
you may be a bystander); but once you are a standing passenger you
become a standee, and you have your standee line, and must keep
behind it. What kind of participant is construed in the grammar as an -
ee?
There are familiar ones like nominee, trainee, appointee, and more
recent instances of this type like superannuee and oustee, all of which are
modelled on the pattern of employee ‘person employed’. This forms one
term in the transitive opposition employer/employee; the latter form was
derived from the French passive participle and matched up with the
English active termination -er, giving ‘the one who is acting’/’the one
who is acted upon’; cf. trainer/trainee. Here the -ee is functioning as
participant x in type A.
Then there are some instances where rather indirect relationships are
involved: biographee ‘person whose biography is being compiled’, 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); and
various banking terms like advisee, favouree, assignee and so on. These
are modelled on words like referee ‘person to whom a dispute is referred
for decision’, refugee ‘person to whom a place of refuge is offered’.
Then, with escapee ‘person who escapes’ as an early model, we now
have conferee and attendee ‘person attending a conference or lecture’,
retiree ‘person having retired’, and returnee ‘person trying to get back to
original country’. All these are like standee. When we examine them,
we find that they pattern ergatively: the -ee corresponds to the function
of the Medium in the process, to participant x in a process construed
as type X. There is no implication that these are functioning as the
Goal: a standee is not someone who has been or is being stood. If these
were following type A we would have stander, returner, retirer, attender
and so on. The pattern is given in Figure 1:
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
381
In itself, each instance is trivial. It does not matter whether we write

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

however, those who “think grammatically” are enabled thereby to act
grammatically, whether in developing forms of praxis for educational
and other professional tasks, or in combating sexism, racism and other
prevailing inequalities. To be a linguist is inevitably to be concerned
with the human condition; it takes a linguist of the stature of Sydney
Lamb to explain how so much of what constitutes the human condition
is construed, transmitted, maintained – and potentially transformed –
by means of language.
Notes
1. Notice on the other hand that in the interrogative this pressure is much
less strong: we readily say does your head ache? is your throat sore? as well as
have you got a headache/a sore throat? This is because in the interrogative the
grammar preempts the thematic slot to signal that the clause is, in fact, a
question, by putting at the beginning the part of the verb that selects for
‘yes or no’, the Finite operator, does/is: does your head ache? signals ‘my
message is concerned with whether it does or not’. As a result there is
relatively little thematic weight left over; the difference in information
flow between is your throat sore? and have you got a sore throat? is very much
less noticeable than that between the agnate declarative pair my throat’s sore
and I’ve got a sore throat, where the full thematic weight is felt on either my
throat or I.
2. Thus the grammar signals metaphorically that meaning is a social process.
We might put this together with the recent neurobiological finding by
Robin Dunbar (1992), that species living in large social groups have
proportionally larger cortices. “Dunbar’s explanation is that large group
sizes demand greater social cohesion and hence more advanced skills for
communicating . . .”.
383
Chapter Fifteen
ON GRAMMAR AND GRAMMATICS (1996)
1 The problem
Most of us are familiar with the feeling that there must be something
odd about linguistics. We recognize this as a problem in the 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’ some
phenomenon; but whereas the natural sciences are language about
nature, and the social sciences are language about society, linguistics is
language about language – “language turned back on itself ”, in Firth’s
often quoted formulation. So, leaving aside the moral indignation some
people seem to feel, as if linguistics was a form of intellectual incest,
there is a real problem involved in drawing the boundary: where does
language end and linguistics begin? How does one keep apart the
object language from the metalanguage – the phenomenon itself from
the theoretical study of that phenomenon?
The discursive evidence rather suggests that we don’t, at least not
very consistently. For example, the adjective linguistic means both ‘of
language’, as in linguistic variation, and ‘of linguistics’ as in linguistic
association (we never know, in fact, whether to call our professional
bodies linguistic associations or linguistics associations). But a situation
First published in Functional Descriptions: Theory in Practice, 1996, edited by Ruqaiya Hasan,
Carmel Cloran and David G. Butt. Amsterdam: John Benjamins, pp. 1–38.
384
on grammar and grammatics
analogous to this occurs in many disciplines: objects in nature have

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

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

ineffability (see above Chapter 12). This was based on the simple
proportion grammatics : grammar :: linguistics : language. I assumed it
was unproblematic: the study of language is called linguistics; grammar
is part of language; so, within that general domain, the study of
grammar may be called grammatics.
But this proportion is not quite as simple as it seems. The relationship
of linguistics to language is unproblematic as long as we leave language
undefined; and we can do this – as linguists, we can take language for
granted, as sociologists take society for granted, treating it as a primitive
term. Grammar, on the other hand, needs defining. Although the word
is used in a non-technical sense, as in the bad grammar example, one
cannot take this usage over to define a domain of systematic study: in
so far as it has any objective correlate at all, this would refer to an
inventory of certain marginal features of a language defined by the fact
that they carry a certain sort of social value for its speakers. We can
study ethnographically the patterns of this evaluation, and their place in
the social process; but that is a distinct phenomenal domain. Grammatics,
in fact, has no domain until it defines one for itself (or until one is
defined for it within general linguistics – exactly at what point the term
grammatics takes over from linguistics is immaterial). And it is this that
makes the boundary hard to draw. Since both the grammar and the
grammatics are made of language, then if, in addition, each has to be
used to define the other, it is not surprising if they get confused.
Now you may say, as indeed I said to myself when first trying to
think this through: it doesn’t matter. It does no harm if we just talk
about grammar without any clear distinction between the thing and the
study of the thing. They are in any case much alike: if you turn
language back on itself, it is bound to mimic itself in certain respects.
But this comforting dismissal of the problem was belied by my own
experience. If I had become aware of the polysemy in the word
grammar it was because it got in the way of clear thinking – my own,
and that of the students I was trying to teach. (It does not help,
incidentally, to take refuge in the term syntax, where precisely the same
polysemy occurs.) There was confusion in certain concepts, such as
“universals of grammar” and “rule of grammar”, and in the status and
scope of grammatical categories of various kinds. But also, I suspect, a
problem that has been so vexing in recent years – that of relating the
system to the text (so often discourse is analysed as if there were no
general principles of meaning behind it) – is ultimately part of the same
overall unclarity.
386
3 Defining grammar
In the simplest definition grammar is part of language. If we pick up a
book purporting to describe a language, or to help us to learn it, we
expect to find some portion or portions of the book – but not the
whole of the book – devoted to grammar. In my own work, I have
operated with the concept of “lexicogrammar” (that is, grammar and
vocabulary as a single unity), while usually referring to it simply as
grammar for short; this is a stratal concept, with grammar as one among
an ordered series comprising (at least) semantics / lexicogrammar /
phonology. But whatever part–whole model is adopted, language
remains the more inclusive term.
But there is a further step, by which grammar is not just one among
various parts of language; it is a privileged part. The exact nature of
this privilege will be interpreted differently by different linguists, and
some might deny it altogether; but most would probably accept it in
one form or another. I would be inclined to characterize grammar in
the first instance as the part of language where the work is done.
Language is powered by grammatical energy, so to speak.
Let me approach the definition of grammar, however, from a
somewhat different angle. I shall assume here, as a general theoretical
foundation, the account of language given by Lemke (1993). Lemke
characterizes human communities as eco-social systems which persist in
time through ongoing exchange with their environment; and the same
holds true of each of their many sub-systems. The social practices by
which such systems are constituted are at once both material and
semiotic, with a constant dynamic interplay between the two. Note
that by semiotic I mean ‘having to do with meaning’, not ‘having to do
with signs’; thus, practices of doing and practices of meaning. The
important feature of the material–semiotic interplay is that, as Lemke
points out, the two sets of practices are strongly coupled: there is a
high degree of redundancy between them. We may recall here Firth’s
concept of ‘mutual expectancy’ between text and situation.
Underlying the semiotic practices are semiotic systems of various
kinds. In fact, we usually use the term “system” to cover both system
and process: both the potential and the instances that occur; thus a
semiotic system is a meaning potential together with its instantiation in
acts of meaning. Now, one special kind of semiotic system is one that
has a grammar in it: such a system “means” in two phases, having a
distinct phase of wording serving as the base for the construction of
meaning. In other words, its “content plane” contains a grammar as
387
well as a semantics. We could characterize this special kind of semiotic

system as a grammatico-semantic system. It is the presence of a
grammar that gives such a system its unique potential for creating (as
distinct from merely reflecting) meaning.
4 The emergence of grammar through time

We could locate grammatico-semantic systems within the framework
of an evolutionary typology of systems, as in Figure 1. In this frame,
semiotic systems appear as systems of a fourth order of complexity, in
that they are at once physical and biological and social and semiotic.
Within semiotic systems, those with a grammar in them are more
complex than those without.
physical + life
= biological + value
= social + meaning + grammar
= semiotics1 = semiotic2
[primary] [higher order, i.e.
grammatico-semantic]
S S S S S
1 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, a
sign being a content/expression pair. Systems of this kind, which may
be called primary semiotics, are found among numerous species: all
higher animals, including our household pets; and such a system is also
developed by human infants in the first year of their lives – I referred
to this as the “protolanguage” (Halliday 1975). Primary semiotic systems
have no grammar. The more complex type of semiotic system is that
which evolves in the form of Edelman’s “higher order consciousness”.
This higher order semiotic is what we call language. It has a grammar;
and it appears to be unique to mature (i.e. post-infancy) human beings.
In other words, it evolved as the “sapiens” in homo sapiens. (I say this
without prejudice; I would be happy – indeed very excited – to learn
388
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 numerous
non-linguistic semiotic systems: forms of ritual, art forms, and the like;
these have no grammar of their own, but they are parasitic on natural
language – their meaning potential derives from the fact that those who
use them already have a grammar. (See O’Toole (1994) for a rich
interpretation of visual semiotics in grammatico-semantic terms.) Thus
all human semiotic activity, from early childhood onwards, is as it were
filtered through our grammar-based higher order consciousness.
What then is a grammar, if we look at it historically in this way, as
evolving (in the species) and developing (in the individual)? A grammar
is an entirely abstract semiotic construct that emerges between the
content and the expression levels of the original, sign-based primary
semiotic system. By “entirely abstract” I mean one that does not interface
directly with either of the phenomenal realms that comprise the material
environment of language. The expression system (prototypically, the
phonology) interfaces with the human body; the (semantic component
of the) content interfaces with the entire realm of human experience;
whereas the grammar evolves as an interface between these two inter-
faces – shoving them apart, so to speak, in such a way that there arises
an indefinite amount of “play” between the two.
5 Grammar in semiotic function

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

or correspond to pre-existing states of affairs. The stratal pattern of
organization, with an entirely substance-free stratum of grammar at its
core, makes it possible to construct complex open-ended networks of
semantic potential in which meanings are defined relative to one
another and hence can modify each other and also can change in
interaction with changes in the ongoing (semiotic and material)
environment.
The grammar does not, of course, evolve in isolation; meanings are
brought into being in contexts of function. The functional contexts of
language fall into two major types, and the constitutive function that
the grammar performs differs as between the two types. On the one
hand, language “constitutes” human experience; and in this context,
the grammar’s function is to construe: the grammar transforms 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 third
sense – creating a parallel universe of its own, a phenomenal realm that
is itself made out of meaning. This enables the semiotic process to
unfold, through time, in cahoots with material processes, each provid-
ing the environment for the other. To put this in other terms, the
grammar enables the flow of information to coincide with, and interact
with, the flow of events (Matthiessen 1992; 1995).
This metafunctional interdependence is central to the evolution of
language, and to its persistence through constant interaction with its
environment. In the experiential (or, to give it its more inclusive name,
the “ideational”) metafunction, the grammar takes over the material
conditions of human existence and transforms them into meanings. We
tend to become aware of the grammatical energy involved in this
process only when we have to write a scientific paper; hence, this
semiotic transformation may appear to be just a feature of knowledge
that is systematic. But all knowledge is like this: to “know” something
is to have transformed it into meaning, and what we call “understand-
ing” is the process of that transformation. But experience is understood
in the course of, and by means of, being acted out interpersonally –
and, in the same way, interpersonal relations are enacted in the course
of, and by means of, being construed ideationally. The grammar flows
390
these two modes of meaning together into a single current, such that
everything we say (or write, or listen to, or read) “means” in both
these functions at once. Thus every instance of semiotic practice –
every act of meaning – involves both talking about the world and
acting on those who are in it. Either of these sets of phenomena may
of course be purely imaginary; that in itself is a demonstration of the
constitutive power of a grammar.
6 Grammar as theory
So far I have been talking about various properties of grammar. But in
“talking about” grammar, I have been “doing” grammatics – it is my
discourse that has been construing grammar in this way. Naturally, I
have also been ‘doing’ grammar: the properties have been being
construed in lexicogrammatical terms. In other words I have been
using grammar to construct a theory about itself.
Every scientific theory – in fact every theory of any kind, whether
‘scientific’ or otherwise – is constructed in similar fashion, by means of
the resources of grammar. A theory is a semiotic construct (see Lemke
(1990) for a powerful presentation of this point). That we are able to
use a grammar as a resource for constructing theories is because a
grammar is itself a theory. As I suggested in the previous section, the
grammar functions simultaneously as a mode of knowing and a mode
of doing; the former mode – the construction of knowledge – is the
transformation of experience into meaning. A grammar is a theory of
human experience.
Construing experience is a highly theoretical process, involving
setting up categories and relating each category to the rest. As Ellis
(1993) points out, there are no natural classes: the categories of
experience have to be created by the grammar itself. Or, we might say,
there are indefinitely many natural classes: indefinitely many ways in
which the phenomena of our experience may be perceived as being
alike. In whichever of these terms we conceive the matter, the grammar
has to sort things out, assigning functional value selectively to the
various possible dimensions of perceptual order. The grammar’s model
of experience is constantly being challenged and reinforced in daily life;
thus it tends to change when there are major changes in the conditions
of human existence – not as a consequence, but as a necessary and
integral element, of these changes.
The difference between a grammar, as a “commonsense” theory of
experience, and a scientific theory (such as grammatics) is that grammars
391
evolve, whereas scientific theories are at least partially designed. (The

nearest 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 no
longer be construed in natural language wordings.) But it is still the
grammar of natural language that is deployed in the designing of
scientific theories (cf. Halliday and Martin 1993).
In the next few sections I shall discuss some of the properties of
grammars that enable them to function as they do: to theorize about
human experience and to enact human relationships. In addition to
their metafunctional organization, already alluded to as enabling the
integration of knowledge and action, I shall mention (a) their size and
ability to expand, (b) their multiplicity of perspective, and (c) their
indeterminacy. In talking about these features, of course, I shall still be
“doing” grammatics. Then, in the final sections, I shall turn to talking
about grammatics.
7 How big is a grammar?

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

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

than one secondary tense. (And, it should be pointed out, the options
shown are all simply the variant forms of one single verb.) So when I
prepared a network of the English clause as the first grammar for
William Mann’s “Penman” text generation project in 1980, which had
394
81 systems in it, Mann was probably not far wrong when he estimated
off the cuff that it would generate somewhere between 108 and 109
clause types.
Of course there are lots of mistakes in these complex networks, and
the only way to test them is by programming them and setting them
to generate at random. It is not difficult to generate the paradigm of
selection expressions from a reasonably small network (already in 1966
Henrici developed a program for this purpose; cf. Halliday and Martin
1981), where you can inspect the output and see where it has gone
wrong. But even if the program could list half a billion expressions it
would take a little while to check them over. As far as their overall
capacity is concerned, however, they are probably not orders of
magnitude out.
It has been objected that the human brain could not possibly process
a grammar that size, or run through all the alternative options whenever
its owner said or listened to a clause. I am not sure this is so impossible.
But in any case it is irrelevant. For one thing, this is a purely abstract
model; for another thing, the number of choice points encountered in
generating or parsing a clause is actually rather small – in the network
of the verbal group it took only 28 systems to produce some 70,000
selection expressions, and in any one pass the maximum number of
systems encountered would be even less – probably under half the
total, in a representative network. In other words, in selecting one out
of half a billion clause types the speaker/listener would be traversing at
the most about forty choice points. So although the system network is
not a model of neural processes, there is nothing impossible about a
grammar of this complexity – that is, where the complexity is such that
it can be modelled in this way, as the product of the intersection of a
not very large number of choices each of which by itself is extremely
simple.
8 How does your grammar grow?

Grammars do not remain static. They tend to grow; not at an even
rate, but with acceleration at certain “moments” in the history of a
culture.
On the one hand, they grow by moving into new domains. This
happens particularly when there is an expansion in the culture’s
knowledge and control: in our present era, new domains are opened
up by developments in technology and science. We are likely to
become aware of this when we meet with a crop of unfamiliar words,
395
like those associated with the recent move into nanotechnology (engin-
eering the very small); but the expansion may take place anywhere in
the lexicogrammar, as new wording, in any form. The grammar is not
simply tagging along behind; technological developments, like other
historical processes, are simultaneously both material and semiotic – the
two modes are interdependent. Early on in his researches into science
and technology in China, Needham noted how in the medieval period,
when there was no adequate institutional mechanism for keeping new
meanings alive, the same material advances were sometimes made two
or three times over, without anyone realizing that the same technology
had been developed before (Needham 1958).
On the other hand, grammars grow by increasing the delicacy in
their construction of existing domains. (This has been referred to by
various metaphors: refining the grid or mesh, sharpening the focus,
increasing the granularity and so on. I shall retain the term “delicacy”,
first suggested by Angus McIntosh in 1959.) This is a complex notion;
it is not equivalent to subcategorizing, which is simply the limiting case
– although also the one that is likely to be the most easily recognized.
The grammar does construct strict taxonomies: fruit is a kind of food, a
berry is a kind of fruit, a raspberry is a kind of berry, a wild raspberry is a
kind of raspberry; these are typically hyponymic and can always be
extended further, with new words or new compositions of words in a
grammatical structure, like the nominal group in English and many
other languages. But greater delicacy is often achieved by intersecting
semantic features in new combinations; and this is less open to casual
inspection, except in isolated instances which happen to be in some
way striking (like certain “politically correct” expressions in present-
day English). The massive semantic innovations brought about by
computing, word processing, networking, multimedia, the information
superhighway and the like, although in part construing these activities
as new technological domains, more typically constitute them as new
conjunctions of existing meanings, as a glance at any one of thousands
of current periodicals will reveal. On a somewhat less dramatic scale,
we are all aware of the much more elaborate variations in the discourse
of environmental pollution and destruction than were available a
generation ago. Even a seemingly transparent piece of wording such as
smoke-free construes a new confluence of meanings; indeed the whole
semogenic potential of -free as a derivational morpheme has recently
been transformed. (Similar expansions have happened with -wise and
-hood.)
There is a special case of this second heading – perhaps even a third
396
type of grammar growth – in the form of semantic junction brought

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

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

(Firth 1957). But between the semantics and the grammar, this new
frontier (typically non-arbitrary) within the content plane, we expect
to find more isomorphism: things which mean alike might reasonably
be worded alike. As a general rule, they are: grammatical proportion-
alities typically construe semantic ones. But not always. On the one
hand, there are regions of considerable drift in both directions; an
obvious one in English is the semantic domain of probability and
subjective assessment, which is construed in many different regions of
the grammar – each of which may in turn construe other semantic
features, such as obligation or mental process. On the other hand, there
are the “syndromes” mentioned above – high-level semantic motifs
which are located all around the terrain of the lexicogrammar, such as
the complex edifice of meanings that goes to make up a “standard
language”. People make much use of these realignments in reasoning
and inferencing with language.
This stratified vision of things enables the grammar to compromise
among competing models of reality. As pointed out above in Section
6, a grammar sorts out and selects among the many proportionalities
that could arise in the construal of experience. It does this by making
adjustments among the different strata. Things may appear alike from
any of three different angles: (i) “from above” – similarity of function
in context; (ii) “from below” – similarity of formal make-up; and (iii)
“from the same level” – fit with the other categories that are being
construed in the overall organization of the system. The grammar looks
at objects and events from all three angles of orientation. It takes account
of their function: phenomena which have like value for human existence
and survival will tend to be categorized as alike. It takes account of
their form: phenomena which resemble each other to human percep-
tions will tend to be categorized as alike. And it takes account of how
things relate to one another: phenomena are not categorized in isolation
but in sets, syndromes and domains. In other words, the grammar adopts
what we may call a “trinocular” perspective.
It often happens that the various criteria conflict: things (whether
material or semiotic) that are alike in form are often not alike in
function; and the way they relate to each other may not reflect either
kind of likeness. Other things being equal, the grammar tends to give
some precedence to functional considerations: consider any crowded
lexical domain, such as that of maps, plans, charts, figures, diagrams, tables
and graphs in English; or grammatical systems that are highly critical for
survival, like that of polarity in any language. But the construal of
398
categories must make sense as a whole. And this means that it needs to
be founded on compromise. The grammar of every natural language is
a massive exercise in compromise, accommodating multiple 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 only
because of the effort that goes into tidying them up. Formal logic and
even mathematics can be seen as the result of tidying up the 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, though
rather different case.
Clines are distinctions in meaning which take the form of continuous
variables instead of discrete terms. The prototype examples in grammar
are those distinctions which are construed prosodically, typically by
intonation (tone contour): for example, in English, “force”, from strong
to mild, realized as a continuum from wide to narrow pitch movement
– if the tone is falling, then from wide fall (high to low) to narrow fall
(midlow to low). But one can include in this category those distinctions
where, although the realizations are discrete (i.e. different wordings are
involved), the categories themselves are shaded, like a colour spectrum:
for example, colours themselves; types of motorized vehicles (car, bus,
van, lorry, truck, limousine . . . etc.); types of process (as illustrated on
the cover of the revised edition of my Introduction to Functional Grammar
1994). In this sense, since in the grammar’s categorization of experience
fuzziness is the norm, almost any scalar set will form a cline: cf. humps,
mounds, hillocks, hills and mountains; or must, ought, should, will, would,
can, could, may, might.
Blends are forms of wording which ought to be ambiguous but are
not. Ambiguity in the strict sense, as in lexical or structural puns, is not
a form of indeterminacy as considered here, because it does not involve
indeterminacy of categorization. Blends also construe two (or more)
different meanings; but the meanings are fused – it is not a matter of
selecting one or the other. A favourite area for blends, apparently in
many languages, is modality; in English, oblique modal finites like
399
should provide typical examples, for example the brake should be on,
meaning both ‘ought to be’ and ‘probably is’. There is then the further
indeterminacy between an ambiguity and a blend, because a wording
which is clearly ambiguous in one context may be blended when it
occurs in another. A metaphor is the limiting case of a blend.
Complementarities are found in those regions of (typically experi-
ential) semantic space where some domain of experience is construed
in two mutually contradictory ways. An obvious example in English is
in the grammar of mental processes, where there is a regular 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 experience
which can be construed in different ways: here, in a process of
consciousness the conscious being is on the one hand ‘doing’, with
some phenomenon defining the scope of the deed, and on the other
hand ‘being done to’ with the phenomenon functioning as the doer.
All languages (presumably) embody complementarities; but not always
in the same regions of semantic space (note for example the striking
complementarity of tense and aspect in Russian). One favourite domain
is causation and agency, often manifested in the complementarity of
transitive and ergative construals.
Strictly speaking probability is not a “fuzzy” concept; but probability
in grammar adds indeterminacy to the definition of a category. Con-
sider the network of the English verbal group in Figure 2 above. As an
exercise in grammatics this network is incomplete, in that there are
distinctions made by the grammar that the network fails to show: in
that sense, as already suggested, no network ever can be complete. But
it is incomplete also in another sense: it does not show probabilities. If
you are generating from that network, you are as likely to come up
with won’t be taken as with took; whereas in real life positive is
significantly more likely than negative, active than passive, and past
than future. Similarly a typical dictionary does not tell you that go is
more likely than walk and walk is more likely than stroll, though you
might guess it from the relative length of the entries. A grammar is an
inherently probabilistic system, in which an important part of the
meaning of any feature is its probability relative to other features with
which it is mutually defining. Furthermore the critical factor in register
variation is probabilistic: the extent to which local probabilities depart
from the global patterns of the language as a whole; for example a
400
register of weather forecasting (and no doubt other kinds of forecasting

as well), where future becomes more probable than past; or one in
which negative and passive suddenly come to the fore, like that of
bureaucratic regulations (Halliday 1991). Probabilities are significant
both in ideational and in interpersonal meanings, as well as in the
textual component; they provide a fundamental resource for the
constitutive potential of the grammar.
11 Some matching features

In the last few sections I have picked out certain features of natural
language grammars which a theory of grammar – a “grammatics” – is
designed to account for. The purpose of doing this was to provide a
context for asking the questions: how does the grammatics face up to
this kind of requirement? Given that every theory is, in some sense, a
lexicogrammatical metaphor for what it is theorizing, is there anything
different about a theory where what it is theorizing is also a
lexicogrammar?
There is (as far as I can see) no way of formally testing a grammar in
its role as a theory of human experience: there are no extrinsic criteria
for measuring its excellence of fit. We can of course seek to evaluate
the grammar by asking how well it works; and whatever language we
choose it clearly does – grammars have made it possible for humanity
to survive and prosper. They have transmitted the wisdom of accumul-
ated experience from one generation to the next, and enabled us to
interact in highly complex ways with our environment. (At the same
time, it seems to me, grammars can have quite pernicious side-effects,
now that we have suddenly crossed the barrier from being dominated
by that environment to being in control of it, and therefore also
responsible for it; cf. Halliday 1993). I suspect that the same holds true
for the grammatics as a theory of grammar: we can evaluate such a
theory, by seeing how far it helps in solving problems where language
is centrally involved (problems in education, in health, in information
management and so on); but we cannot test it for being right or wrong.
(This point was made by Hjelmslev many years ago, as the general
distinction between a theory and a hypothesis.) By the same token a
grammatics can also have its negative effects, if it becomes reductionist
or pathologically one-sided.
The special quality of a theory of grammar, I think, is the nature of
the metaphoric relationship that it sets up with its object of enquiry. If
we consider just those features of language brought into the discussion
401
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; this
ensures that the grammar will be viewed comprehensively, and that
however closely we focus on any one typological or topological domain
this will always be contextualized in terms of the meaning potential of
the grammar as a whole. It copes with the trinocular vision of the
grammar by also adopting a trinocular perspective, based on the stratal
organization of the grammar itself. And it copes with the indeterminacy
of the grammar by also being indeterminate, so that the categories of
the theory of grammar are like the categories that the grammar itself
construes.
Theories in other fields, concerned with non-semiotic systems, begin
by generalizing and abstracting; but they then take off, as it were, to
become semiotic constructs in their own right, related only very
indirectly and obliquely to observations from experience. The proto-
type of such a theory is a mathematical model; and one can theorize
grammatics in this way, construing it as a formal system. But a
grammatics does not need to be self-contained in this same manner. It
is, as theory, a semiotic construct; but this does not create any
disjunction between it and what it is theorizing – it remains permeable
at all points on its surface. The grammatics thus retains a mimetic
character: it explains the grammar by mimicking its crucial properties.
One could say that it is based on grammatical logic rather than on
mathematical logic. In some respects this will appear as a weakness: it
will lack the rigour of a mathematical theory. But in other respects it
can be a source of strength. It is likely to be more relevant to
understanding other semiotic systems: not only verbal art, but also
other, non-verbal art forms, as demonstrated by O’Toole’s masterly
interpretation of painting, architecture and sculpture in terms of sys-
temic grammatics, referred to already (O’Toole 1994). And the new
field of “intelligent computing”, associated with the work of Sugeno,
and explicitly defined by him as “computing with (natural) language”,
requires a theory that celebrates indeterminacy (it is a development of
fuzzy computing) and that allows full play to the interface between
wording and meaning (see section 20 below).
402
In the next few sections I will make a few observations about these
matching properties of the grammatics, as they seem to me to emerge
in a systemic perspective.
12 Paradigmatic orientation and delicacy

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

into a lexicogrammar; the bricks-&-mortar model of a “lexicon”
of words stuck together by grammatical cement can be 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 important
papers by Hasan (1985; 1987). The principle is that grammar and lexis
are not two distinct orders of phenomena; there is just one stratum
here, that of “(lexico)grammar”, and one among the various resources
that the grammar has for making meaning (i.e. for “realizing” its
systemic features) is by lexicalizing – choosing words. In general, the
choice of words represents a delicate phase in the grammar, in the sense
that it is only after attaining quite some degree of delicacy that we
reach systems where the options are realized by the choice of the
lexical item. The lexicogrammar is thus construed by the grammatics
as a cline, from “most grammatical” to “most lexical”; but it is also a
complementarity, because we can also view lexis and grammar as
different perspectives on the whole. The reason people write “gram-
mars” on the one hand and ‘dictionaries’ on the other is that options at
the most general (least delicate) end of the cline are best illuminated by
one set of techniques while options at the most delicate (least general)
end are best illuminated by a different set of techniques. One can
employ either set of techniques all the way across; but in each case
there will be diminishing returns (increasing expenditure of energy,
with decreasing gains).
To say that, as the description moves towards the lexical end, one
eventually reaches systems where the options are realized by the choice
of a lexical item, does not mean, on the other hand, that these are
systems where there is a direct correspondence of feature to item, such
that feature 1 is realized by lexical item a, feature 2 by lexical item b
and so on. What it means is that one reaches systems where the features
are components of lexical items. (Thus, they are like the features of a
standard componential analysis, except that they form part of the
overall system network and no distinction is made between features
that are “lexical” and those that are “grammatical”.) Any given lexical
item then appears as the conjunct realization of a set of systemic
features; and “the same” lexical item may appear many times over, in
different locations, much as happens in a thesaurus (where however the
organization is taxonomic rather than componential).
404
13 A note on delicacy
Inherent in the paradigmatic orientation is the concept of variable
delicacy, in which again the grammatics mimics the grammar: delicacy
in the construal of grammar (by the grammatics) is analogous to
delicacy in the construal of experiential phenomena (by the grammar).
Since for the most part the “lexicalized” mode of realization is
associated with fairly delicate categories in the grammar, we can talk of
“lexis as delicate grammar” (this refers to lexical items in the sense of
“content words”; grammatical items, or “function words”, like the, of,
it, not, as, turn up in the realization of very general systemic features).
But this is not the same thing as saying that when one reaches the stage
of lexical realization one has arrived at the endpoint in delicacy.
What is the endpoint, on the delicacy scale? How far can the
grammatics go in refining the categories of the grammar? In one sense
there can be no endpoint, because every instance is categorially
different from every other instance, since it has a unique instantial
context of situation. We tend to become aware of this when an
instance is codified in the work of a major writer and hence becomes
immortalized as a “quotation”. It seems trivial; but it may not be trivial
in the context of intelligent computing, where the program might need
to recognize that, say, turn left!, as instruction to the car, has a different
meaning – and therefore a different description – at every instance of
its use. This is the sense in which a grammar can be said to be an
“infinite” (i.e. indefinitely large) system. But if we are literate, then in
our commonsense engagements with language, in daily life, we behave
as if there is an endpoint in delicacy: namely, that which is defined by
the orthography. We assume, in other words, that if two instances look
different (i.e. are represented as different forms in writing) they should
be described as different types; whereas if two instances are written
alike they should be described as tokens of the same type – however
delicate the description, it will not tease them apart. The orthography
is taken as the arbiter of paradigmatic boundaries: the way things are
written determines their identity.
There is sense in this: writing represents the unconscious collective
wisdom of generations of speakers/listeners. And we do allow excep-
tions. (a) We recognize homonymy and, more significantly, polysemy,
where the delicacy of categorization does not stop at the barrier created
by the writing system. (b) We accept that there are systematic distinc-
tions which orthography simply ignores: for example, in English, all
those realized by intonation and rhythm. (c) And, as already noted, it
405
never was assumed, except perhaps among a very few linguists, that a
“function word” like of has only one location in the terrain described
by the grammatics. These exceptional cases challenge the implicit
generalization that the orthographic form always defines a “type”
within the wording.
A more explicit principle could be formulated: that, as far as the
grammatics is concerned, the endpoint in delicacy is defined by what
is systemic: the point where proportionalities no longer continue to
hold. As long as we can predict that a : a⬘ :: b : b⬘ :: . . . , we are still
dealing with types, construed as distinct categories for purposes of
grammatical description.
In practice, of course, we are nowhere near this endpoint in writing
our systemic “grammars”. (I find it disturbing when the very sketchy
description of English grammar contained in Halliday (1994) is taken
as some kind of endpoint. Every paragraph in it needs to be expanded
into a book, or perhaps some more appropriate form of hypertext; then
we will be starting to see inside the grammar – and be able to rewrite
the introductory sketch!) We are only now beginning to get access to
a reasonable quantity of data. This has been the major problem for
linguistics: probably no other defined sphere of intellectual activity has
ever been so top-heavy, so much theory built overhead with so little
data to support it. The trouble was that until there were first of all tape
recorders and then computers, it was impossible to assemble the data a
grammarian needs. Since grammars are very big, and very complex, an
effective grammatics depends on having accessible a very large corpus
of diverse texts, with a solid foundation in spontaneous spoken
language; together with the sophisticated software that turns it into an
effective source of information.
14 A note on the corpus

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

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

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

the process of transforming anything into meaning – of “semioticizing”
it in terms of a higher order, stratified semiotic. Construing the
phenomena of experience means “parsing” them into meanings, word-
ings and expressions (you only have to do this, of course, when form
and function cease to match; this is why the task is inescapably one of
achieving compromise). The entire stratal organization of language is
simply the manifestation of this trinocular principle. Making this
principle explicit in the grammatics is perhaps giving substance to the
notion of ‘language turned back upon itself’.
16 Indeterminacy in grammatics
That the grammatics should accommodate indeterminacy does not
need explaining: indeterminacy is an inherent and necessary feature of
a grammar, and hence something to be accounted for and indeed
celebrated in the grammatics, not idealized out of the picture – just as
the grammar’s construal of experience recognizes indeterminacy as an
inherent and necessary feature of the human condition.
But construing indeterminacy is not just a matter of leaving things
as they are. Construing after all is a form of complexity management;
and just as, in a material practice such as looking after a wilderness,
once you have perturbed the complex equilibrium of its ecosystem you
have to intervene and actively manage it, so in semiotic practice, when
you transform something into meaning (i.e. perturb it semiotically) you
also have to manage the complexity. We can note how the grammar
manages the complexity of human experience. In the first instance, it
imposes artificial determinacy, in the form of discontinuities: thus, a
growing plant has to be construed either as tree or as bush or as shrub
(or . . .); the line of arbitrariness precludes us from creating intermediate
categories like shrush. Likewise, one thing must be in or on another;
you are either walking or running, and so on. At the same time, however,
each of these categories construes a fuzzy set, whose boundaries are
indeterminate: on and run and tree are all fuzzy sets in this sense.
Furthermore, the grammar explicitly construes indeterminacy as a
semantic domain, with expressions like half in and half on, in between a
bush and a tree, almost running and the like. The specific types of
indeterminacy discussed in Section 10 above, involving complex
relationships between categories, are thus only special cases, fore-
grounding something which is a property of the grammar as a whole.
Now consider the grammatics from this same point of view. The
409
categories used for construing the grammar – things like noun and
subject and aspect and hypotaxis and phrase – are also like everyday
terms: they impose discontinuity. Either something is a noun or it is a
verb (or . . .); we cannot decide to construe it as a nerb. But, in turn,
each one of these itself denotes a fuzzy set. And, thirdly, the same
resources exist, if in a somewhat fancier form, for making the indeter-
minacy explicit: verbal noun, pseudo-passive, underlying subject, and so
on.
What then about the specific construction of indeterminacy in the
overall edifice constructed by such categories? Here we see rather
clearly the grammatics as complexity management. On the one hand,
it has specific strategies for defuzzifying – for imposing discontinuity
on the relations between one category and another; for example, for
digitalizing the grammar’s clines (to return to the example of “force”,
cited in section 10, it can establish criteria for recognizing a small,
discrete set of contrasting degrees of force). A system network is a case
in point: qualitative relationships both within and between systems may
be ironed out, so that (i) the system is construed simply as a or b
(or . . .), without probabilities, and (ii) one system is either dependent
on or independent of another, with no degrees of partial association.
But, at the same time, the grammatics exploits the various types of
indeterminacy as resources for managing the complexity. I have already
suggested that the concept of lexicogrammmar (itself a cline from
“most grammatical” to “most lexical”) embodies a complementarity in
which lexis and grammar compete as theoretical models of the whole.
There are many blends of different types of structure, for example the
English nominal group construed both as multivariate (configurational)
and as univariate (iterative) but without ambiguity between them. And
the two most fundamental relationships in the grammatics, realization
and instantiation, are both examples of indeterminacy.
I have said that a grammar is a theory of human experience. But that
does not mean, on the other hand, that it is not also part of that
experience; it is. We will not be surprised, therefore, if we find that its
own complexity comes to be managed in ways that are analogous to
the ways in which it itself manages the complexity of the rest. In the
last resort, we are only seeing how the grammar construes itself.
17 A note on realization and instantiation

I referred earlier to these two concepts as being critical when we come
to construe a higher order semiotic. Realization is the name given to
410
the relationship between the strata; the verb realize faces “upwards”,
such that the “lower” stratum realizes the “higher” one. (Realization is
also extended to refer to the intrastratal relation between a systemic
feature and its structural (or other) manifestation.) Instantiation is the
relationship between the system and the instance; the instance is said to
instantiate the system.
It can be said that, in the elements of a primary semiotic (signs), the
signifier “realizes” the signified; but this relationship is unproblematic:
although the sign may undergo complex transformations of one kind
or another, there is no intermediate structure between the two (no
distinct stratum of grammar). With a higher order semiotic, where a
grammar intervenes, this opens up the possibility of many different
types of realization. It is not necessary to spell these out here; they are
enumerated and discussed in many places (for example Berry 1977;
Fawcett 1980; Martin 1984; Hasan 1987; Matthiessen 1988; Eggins
1994).
But there is another opening-up effect which is relevant to the
present topic: this concerns the nature and location of the stratal
boundary between the grammar and the semantics. This is, of course,
a construct of the grammatics; many fundamental aspects of language
can be explained if one models them in stratal terms, such as metaphor
(and indeed rhetorical resources in general), the epigenetic nature of
children’s language development, and metafunctional unity and diver-
sity, among others. But this does not force us to locate the boundary at
any particular place. One can, in fact, map it on to the boundary
between system and structure, as Fawcett does (system as semantics,
structure as lexicogrammar); whereas I have found it more valuable to
set up two distinct strata of paradigmatic (systemic) organization. But
the point is that the boundary is indeterminate – it can be shifted; and
this indeterminacy enables us to extend the stratal model outside
language proper so as to model the relationship of a language to its
cultural and situational environments.
Instantiation is the relationship which defines what is usually thought
of as a “fact” – in the sense of a physical fact, a social fact and so on.
Facts are not given; they are constructed by the theorist, out of the
dialectic between observation and theory. This has always been a
problem area for linguistics: whereas the concept of a physical principle
became clear once the experimental method had been established – a
“law of nature” was a theoretical abstraction constructed mathemati-
cally by the experimenter – the concept of a linguistic principle has
proved much more difficult to elucidate.
411
Saussure problematized the nature of the linguistic fact; but he

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

It is safe to say that neither of these concepts has yet been thoroughly
explored. Problems arise with instantiation, for example, in using the
corpus as data for describing a grammar (why a special category of
“corpus grammar”?); in relating features of discourse to systemic
patterns in grammar (why a separate discipline of “pragmatics”?); and
in construing intermediate categories (such as Bernstein’s “code”,
which remains elusive (like global warming!) from whichever end it is
observed – which is what makes it so powerful as an agency of cultural
reproduction). (See Francis 1993 for the concept of corpus grammar;
Martin 1992 for showing that there can be a system-based theory of
text; Bernstein 1990 for code; Hasan 1989; 1992b for interpretation of
coding orientation; and also Sadovnik 1995 for discussion of Bernstein’s
ideas).
As far as realization is concerned, Lemke has theorized this power-
412
fully as ‘metaredundancy’ (Lemke 1984) (and cf. Chapter 14 above);

but this still leaves problems in understanding how metafunctional
diversity is achieved, and especially the non-referential, interpersonal
aspects of meaning; and in explaining the realization principles at work
at strata outside language itself (see Thibault (1992) and Matthiessen
(1993a) on issues relating to the construal of interpersonal meanings;
Eggins and Martin in press, Hasan (1995), Matthiessen (1993b), on
issues involving the higher strata of register and genre).
I am not pursuing these issues further here. But as a final step I will
shift to another angle of vision and look at realization and instantiation
from inside the grammar – turning the tables by using the grammar as
a way of thinking about the grammatics. One of the most complex
areas in the grammar of English is that of relational processes: processes
of being, in the broadest sense. I have analysed these as falling into two
major types: (i) attributive, and (ii) identifying. The former are those
such as Paula is a poet, this case is very heavy, where some entity is
assigned to a class by virtue of some particular attribute. The latter are
those such as Fred is the treasurer/ the treasurer is Fred, the shortest day is
22nd June/ 22nd June is the shortest day, where some entity is identified
by being matched bi-uniquely with some particular other. (See Halliday
1967–8; 1994.)
The identifying relationship, as construed in the grammar of English,
involves two particular functions, mutually defining such that one is
the outward form, that by which the entity is recognized, while the
other is the function the entity serves. This relationship of course takes
a variety of more specific guises: form / function, occupant / role, sign
/ meaning, and so on. I labelled these grammatical functions “Token”
and “Value”. This Token / Value relationship in the grammar is
exactly one of realization: the Token realizes the Value, the Value is
realized by the Token. It is thus analogous to the relationship defined
in the grammatics as that holding between different strata. The gram-
mar is modelling one of the prototypical processes of experience as
constructing 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 places
it as one among a larger set. It was pointed out by Davidse (1992) that
this Carrier / Attribute relationship in the grammar is actually one of
instantiation: the Carrier is an instance of, or “instantiates”, the Attrib-
ute. It is thus analogous to the relationship defined in the grammatics
as that holding between an instance and the (categories of the) system.
413
(In that respect the original term “ascriptive”, which I had used earlier
to name this type of process, might better have been retained, rather
than being replaced by “attributive”.) Here too, then, the grammar is
construing a significant aspect of human experience – the perception
of a phenomenon as an instance of a general class – in terms of a
property of language itself, where each act of meaning is an instance of
the systemic meaning potential.
Of course, the boot is really on the other foot: the grammatics is
parasitic on the grammar, not the other way around. It is because of
the existence of clause types such as those exemplified above that we
are able to model the linguistic system in the way we do. The
grammatics evolves (or rather one should say the grammatics “is
evolved”, to suggest that it is a partially designed system) as a 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 metaphoric
property 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 semiotic
systems in the centre of attention. In the same way, biological sciences
are “bio-centric”: biased towards living things; and so on. I think it is
also a valid goal to explore the relevance of grammatics to semiotic
systems other than language, and even to systems of other types. The
grammatics is also “totalizing”, because that is the job of a theory. Of
course, it focuses on the micro as well as on the macro – the semiotic
weather as well as the semiotic climate; but that again is a feature of
any theoretical activity.
It has always been a problem for linguists to discover what are the
properties of human language as such, and what are features specific to
a given language. The problem is compounded by the fact that there is
more than one way of incorporating the distinction (wherever it is
drawn) into one’s descriptive practice. Firth articulated the difference
between two approaches: “what is being sketched here is a general
linguistic theory applicable to particular linguistic descriptions, not a theory of
universals for general linguistic description” (Firth 1957: 21; Firth’s em-
phasis). I have preferred to avoid talking about “universals” because it
seems to me that this term usually refers to descriptive categories being
414
treated as if they were theoretical ones. As I see it, the theory models
what is being treated as “universal” to human language; the description
models each language sui generis, because that is the way to avoid
misrepresenting it.
Thus while the theory as a whole is logocentric, the description of
each language is what we might call “glottocentric”: it privileges the
language concerned. The description of English is anglocentric, that
of Chinese sinocentric, that of French gallocentric and so on. (Note
that the theory is not anglocentric; the description of English is.) This
is not an easy aim to achieve, since it involves asking oneself the
question: “how would I describe this language as if English (or other
languages that might get used as a descriptive model) did not exist?”
But it is important if we are to avoid the anglocentric descriptions that
have dominated much of linguistics during the second half of the
century.
In practice, of course, English does exist, and it has been extensively
described; so inevitably people tend to think in terms of categories set
up for English – or for other relatively well-described languages. I have
suggested elsewhere some considerations which seem to me relevant to
descriptive practice (Halliday 1992). As far as my own personal history
is concerned, I worked first of all for many years on the grammar of
Chinese; I mention this here because when I started working on
English people told me I was making English look like Chinese! (It
seems ironic that, now that systemic theory is being widely applied to
Chinese studies, the work of mine most often cited as point of reference
is the descriptive grammar of English.)
In my view an important corollary of the characterological approach
(that is, each language being described in its own terms) is that each
language is described in its own tongue. The protocol version of the
grammar of English is that written in English; the protocol version of
the grammar of Chinese is that written in Chinese; and so on. The
principle of “each language its own metalanguage” is important,
because all descriptive terminology carries with it a load of semantic
baggage from its use in the daily language, or in other technical and
scientific discourses; and this semantic baggage has some metalinguistic
value. This applies particularly, perhaps, to the use of theoretical terms
as metacategories in the description; words such as (the equivalents of)
option, selection, rank, delicacy are likely to have quite significant (but
variable) loadings.
But the principle also helps to guard against transferring categories
inappropriately. Even if descriptive terms have been translated from
415
English (or Russian, or other source) in the first place, once they are
translated they get relocated in the semantic terrain of the new
language, and it becomes easier to avoid carrying over the connotations
that went with the original. So if, say, the term subject or theme appears
in a description of Chinese written in English, its status is as a translation
equivalent of the definitive term in Chinese. Perhaps one should point
out, in this connection, that there can be no general answer to the
question how much alike two things have to be for them to be called
by the same name!
20 A final note on grammatics

As I said at the beginning, when I first used the term “grammatics” I
was concerned simply to escape from the ambiguity where “grammar”
meant both the phenomenon itself – a particular stratum in language –
and the study of that phenomenon; I was simply setting up a proportion
such that grammatics is to grammar as linguistics is to language. But
over the years since then I have found it useful to have “grammatics”
available as a term for a specific view of grammatical theory, whereby
it is not just a theory about grammar but also a way of using grammar
to think with. In other words, in grammatics, we are certainly
modelling natural language; but we are trying to do so in such a way
as to throw light on other things besides. It is using grammar as a kind
of logic. There is mathematical logic and there is grammatical logic,
and both are semiotic systems; but they are complementary, and in
some contexts we may need the evolved logic of grammar rather than,
or as well as, the designed logic of mathematics.
This reflects the fact that, as I see it, grammatics develops in the
context of its application to different tasks. As Matthiessen (1991b) has
pointed out, this, in general, is the way that systemic theory has moved
forward. Recently, a new sphere of application has been suggested. As
mentioned above in Section 10, Sugeno has introduced the concept of
“intelligent (fuzzy) computing”: this is computing based on natural
language (Sugeno 1995). He has also called it “computing with words”,
although as I have commented elsewhere (Halliday 1995) this is really
“computing with meanings”. Sugeno’s idea is that for computers to
advance to the point where they really become intelligent they have to
function the way human beings do – namely, through natural (human)
language. This view (and it is more than a gleam in the eye: Sugeno
has taken significant steps towards putting it into practice) derives
ultimately from Zadeh’s “fuzzy logic”; it depends on reasoning and
416
inferencing with fuzzy sets and fuzzy matching processes. But to use
natural language requires a grammatics: that is, a way of modelling
natural language that makes sense in this particular context. Systemic
theory has been used extensively in computational linguistics; and the
Penman nigel grammar, and Fawcett’s communal grammar, are
among the most comprehensive grammars yet to appear in computa-
tional form (Matthiessen 1991a; Matthiessen and Bateman 1992; Faw-
cett and Tucker 1990; Fawcett, Tucker and Lin 1993). But, more
importantly perhaps, systemic grammatics is not uncomfortable with
fuzziness. That is, no doubt, one of the main criticisms that has been
made of it; but it is an essential property that a grammatics must have
if it is to have any value for intelligent computing. This is an exciting
new field of application; if it prospers, then any grammarian privileged
to interact with Sugeno’s enterprise will learn a lot about human
language, as we always do from applications to real-life challenging
tasks.
Note
1. This is not to question the semiotic achievements of the bonobo 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.
417
BIBLIOGRAPHY
Abercrombie, D. (1963) ‘Conversation and spoken prose’, English Language

Teaching 18.1. Reprinted in D. Abercrombie, Studies in Phonetics and
Linguistics. London: Oxford University Press (Language and Language
Learning 10).
Allen, W. S. (1953) ‘Relationship in comparative linguistics’, Transactions of
the Philological Society.
Allen, W. S. (1956) ‘Structure and system in the Abaza verbal complex’,
Transactions of the Philological Society.
Anderson, J. M. (1968) ‘Ergative and nominative in English’, Journal of
Linguistics 4. 1–32.
Bateson, G. (1936) Naven. Stanford: Stanford University Press (2nd edn
1958).
Bazell, C. E. (1953) Linguistic Form. Istanbul: Istanbul University Press.
Beattie, G. (1983) Talk: An Analysis of Speech and Non-verbal Behaviour in
Conversation. Milton Keynes: Open University Press.
Bernstein, B. (1970) ‘A sociolinguistic approach to socialization’, in J. J.
Gumperz and D. Hymes (eds), Directions in Sociolinguistics. New York:
Holt, Rinehart & Winston.
Bernstein, B. (1971) Theoretical Studies Towards a Sociology of Language. Vol. 1,
Class, Codes and Control. London: Routledge & Kegan Paul.
Bernstein, B. (1990) The Structuring of Pedagogic Discourse. Vol. 4, Class, Codes
and Control. London: Routledge & Kegan Paul.
Berry, M. (1957) Structures and Systems. Vol. 1, Introduction to Systemic
Linguistics. London: Batsford Academic.
Berry, M. (1977) Levels and Links. Vol. 2, Introduction to Systemic Linguistics.
London: Batsford Academic.
Berry, M. (1981a) ‘Systemic linguistics and discourse analysis: a multi-layered
approach to exchange structure’, in R. M. Coulthard and M. M.
419
bibliography
Montgomery (eds), Studies in Discourse Analysis. London: Routledge &

Kegan Paul.
Berry, M. (1981b) ‘Towards layers of exchange structure for directive
exchanges’, Network 2.
Bloomfield, L. (1933) Language. Chicago: University of Chicago Press.
Boas, F. (1911) Introduction to the Handbook of American Indian Languages.
Vol. 40.1, Bulletin of the Bureau of American Ethnology. Washington, D.C.:
Smithsonian Institution. Reprinted in C. I. J. M. Stuart, Report of the
Fifteenth Annual (First International) Round Table Meeting on Linguistics and
Language Study (Monograph Series in Languages and Linguistics 17).
Brøndal, V. (1948) Les Parties du discours, trans. by Naert. Copenhagen:
Munksgaard.
Bühler, 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.
420
bibliography
Colby, B. N. and Colby, L. M. (1980) The Daykeeper: The Life and Discourse
of an Ixil Diviner. Cambridge, MA.: Harvard University Press.
Daneš, F. (1966) ‘A three-level approach to syntax’, Travaux linguistiques de
Prague 1.
Davidse, K. (1992a) ‘A semiotic approach to relational clauses’, Occasional
Papers in Systemic Linguistics 6.
Davidse, K. (1992b) ‘Transitivity/ergativity: the Janus-headed grammar of
actions and events’, in M. Davies and L. Ravelli (eds), Advances in Systemic
Linguistics: Recent Theory and Practice. London: Pinter.
Derewianka, B. (1995) ‘Language development in the transition from childhood to
adolescence: the role of grammatical metaphor.’ Unpublished PhD dissertation,
Macquarie University.
Doughty, P. S., Pearce, J. J. and Thornton, G. M. (1971) Language in Use.
London: Edward Arnold (for the Schools Council).
Dunbar, R. (1992) Summary in ‘Secret life of the brain’, New Scientist 1850
(Supplement 4).
Ebeling, C. L. (1960) Linguistic Units. The Hague: Mouton ( Janua Linguarum
12).
Eco, U. (1983) The Name of the Rose, trans. by W. Weaver. London: Secker
& Warburg. (Italian original Il nome della rosa. Milano: Bompiani.)
Edelman, G. M. (1992) Bright Air, Brilliant Fire: On the Matter of the Mind.
New York: Basic Books.
Eggins, S. (1994) An Introduction to Systemic Functional Linguistics. London:
Pinter.
Eggins, S. and Martin, J. R. (1997) ‘Genres and registers of discourse’, in T.
A. van Dijk (ed.), Discourse: A Multidisciplinary Introduction. London: Sage.
Eiler, M. A. (1979) ‘Meaning and choice in writing about literature: a study of
cohesion in the expository texts of ninth graders.’ Unpublished PhD dissertation,
Illinois Institute of Technology.
Ellis, J. M. (1993) Language, Thought and Logic. Evanston Il.: Northwestern
University Press.
Ellis, J. O. (1966a) Towards a General Comparative Linguistics. The Hague:
Mouton (Janua Linguarum).
Ellis, J. O. (1966b) ‘On contextual meaning’, in C. E. Bazell et al. (eds) In
Memory of J. R. Firth. London: Longman.
Fawcett, R. P. (1980) Cognitive Linguistics and Social Interaction. Heidelberg and
University of Exeter: Julius Groos (Exeter Linguistic Studies 3).
Fawcett, R. P. (1988) ‘What makes a “good” system network good?’, in J. D.
Benson and W. S. Greaves (eds), Systemic Functional Approaches to Discourse.
Norwood, N.J.: Ablex.
Fawcett, R. P. and Tucker, G. H. (1990) ‘Demonstration of GENESYS: a
very large semantically based systemic functional grammar’, Proceedings of
421
bibliography
the Thirteenth International Conference on Computational Linguistics, Helsinki,

Vol. 1.
Fawcett, R. P., Tucker, G. H. and Lin, Y. Q. (1993) ‘How a systemic
functional grammar works: the role of realisation’, in H. Horacek and M.
Zock (eds), New Concepts in Natural Language Generation. London: Pinter.
Fillmore, C. J. (1968) ‘The case for case’, in E. Bach and R. T. Harms (eds),
Universals in Language. New York: Holt, Rinehart & Winston.
Firbas, J. (1959) ‘Thoughts on the communicative function of the verb in
English, German and Czech’, Brno Studies in English 1. 39–63.
Firbas, J. (1964) ‘On defining the theme in Functional Sentence Analysis’,
Travaux linguistique de Prague 1. 267–80.
Firth, J. R. (1951) ‘General linguistics and descriptive grammar’, Transactions
of the Philological Society. Reprinted in J. R. Firth 1957b.
Firth, J. R. (1948) ‘Sounds and prosidies’, in Transactions of the Philological
Society.
Firth, J. R. (1951) ‘Modes of meaning’, in J. R. Firth 1957b.
Firth, J. R. (1952) ‘Personality and language in society’, Sociological Review
42. 2.
Firth, J. R. (1955) ‘Structural linguistics’, Transactions of the Philological Society.
Reprinted in F. R. Palmer (ed.), Selected Papers of J. R. Firth 1952–1959.
London: Longman (Longman’s Linguistic Library).
Firth, J. R. (1957a) ‘A synopsis of linguistic theory’, in J. R. Firth et al.
(eds), Studies in Linguistic Analysis. Oxford: Blackwell (Special Volume of
the Philological Society). Reprinted in F. R. Palmer (ed.), Selected Papers
of J. R. Firth 1952–1959. London: Longman (Longman’s Linguistic
Library).
Firth, J. R. (1957b) Papers in Linguistics 1934–1951. London: Oxford 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.
422
bibliography
Fries, P. H. (1981) ‘On the status of theme in English: arguments from

discourse’, Forum Linguisticum 6.1. Reprinted in J. S. Petöfi and E. Soözer
(eds), Micro and Macro Connexity of Texts. Hamburg: Buske.
Fries, P. H. (1995) ‘Patterns of information in initial position in English’, in
P. H. Fries and M. Gregory (eds), Discourse in Society: Systemic Functional
Perspective. Norwood, N.J.: Ablex.
Garvin, P. (1957) Report of the Seventh Annual Round Table Meeting on Linguistics
and Language Study. Washington, D.C.: Georgetown University Press
(Languages and Linguistics 9).
Gougenheim, R. et al. (1956) L’élaboration du Français élémentaire. 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.
423
bibliography
Halliday, M. A. K. (1977) ‘Text as semantic choice in social contexts’, in A.

van Dijk and J. Petöfi (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. Pütz (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
424
bibliography
Language in a Social-Semiotic Perspective. Tokyo: Sophia University Graduate

School of Linguistics (Sophia Linguistica 6).
Halliday, M. A. K. and James, Z. L. (1993) ‘A quantitative study of polarity
and primary tense in the English finite clause’, in J. McH. Sinclair, M.
Hoey and G. Fox (eds), Techniques of Description: Spoken and Written
Discourse. A Festschrift for Malcolm Coulthard. London and New York:
Routledge.
Halliday, M. A. K. and Martin, J. R. (eds) (1981) Readings in Systemic
Linguistics. London: Batsford Academic.
Halliday, M. A. K. and Martin J. R. (1993) Writing Science: Literary and
Discursive Power. London: Falmer Press.
Halliday, M. A. K. and Matthiessen, C. (1999) Construing Experience through
Meaning: A Language-based Approach to Cognition. London: Continuum.
Halliday, M. A. K., McIntosh, A. and Strevens, P. (1964) The Linguistic
Sciences and Language Teaching London: Longman (Longmans’ Linguistics
Library).
Halliday, M. A. K. and Poole M. (1978) Notes on Talking Shop: Demands on
Language for Use with the Film Australia Production. Lindfield: Australian Film
Commission.
Harris, Z. S. (1951) Methods in Structural Linguistics. Chicago: University of
Chicago Press.
Harris, Z. S. (1952) ‘Discourse analysis’, Language 28.1,4.
Harris, Z. S. (1954) ‘Transfer grammar’, International Journal of Linguistics 20.
Hasan, R. (1965) The Language of Nine-year-old Children. Leeds: Nuffield
Foreign Languages Teaching Materials Project, Child Language Survey
(Transcript No. 2b).
Hasan, R. (1968) Grammatical Cohesion in Spoken and Written English Vol. 1.
London: Longman.
Hasan, R. (1979) ‘On the notion of text’, in J. Petöfi (ed.), Text Versus
Sentence: Basic Questions of Text Linguistics. Hamburg: Helmut Buske Verlag.
Hasan, R. (1980) ‘The texture of a text’, in M. A. K. Halliday and R. Hasan
(eds), 1980.
Hasan, R. (1983) ‘Coherence and cohesive harmony’, in J. Flood (ed.),
Understanding Reading Comprehension. Newark, DE.: International Reading
Association.
Hasan, R. (1984a) ‘The structure of the nursery tale: an essay in text typology’,
in L. Coveri (ed.), Linguistica Testuale: Proceedings of the 15th Internatonal
Congress of the Italian Linguistics Society. Rom: Bulzoni.
Hasan, R. (1984b) ‘Ways of saying: ways of meaning’, in R. P. Fawcett, M.
A. K. Halliday, S. M. Lamb and A. Makkai (eds), Language as Social
Semiotic. Vol. 1, The Semiotics of Culture and Language. London and Dover:
Pinter (Open Linguistics Series).
425
bibliography
Hasan, R. (1984c) ‘Coherence and cohesive harmony’, in J. Flood (ed.),

Understanding Reading Comprehension. Newark, DE.: International Reading
Association.
Hasan, R. (1985) ‘Lending and borrowing: from grammar to lexis’, in J. E.
Clark (ed.), The Cultivated Australian: Festschrift for Arthur Delbridge. Ham-
burg: Helmut (Buske Beiträge 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).
426
bibliography
Hockett, C. F. (1957) ‘Two models of grammatical description’, in M. Joos

(ed.), Readings in Linguistics. Washington: American Council of Learned
Societies.
Hockett, C. F. (1958) Course in Modern Linguistics. New York: Macmillan.
Hockett, C. F. (1961) ‘Linguistic elements and their relations’, Language 37.
Hockett, C. F. (1965) ‘Sound change’, Language 41.
Huddleston, R. D. (1965) ‘Rank and depth’, Language 41.
Huddleston, R. D., Hudson, R. A., Winter, E. O. and Henrici, A. (1968)
Sentence and Clause in Scientific English. London: University College London,
Communication Research Centre (O.S.T.I. Programme in the Linguistic
Properties of Scientific English, Final Report).
Hudson, R. A. (1967) ‘Constituency in a systematic description of the English
clause’, Lingua 18. 225–50.
Hymes, D. H. (1971) ‘Competence and performance in linguistic theory’, in
R. Huxley and E. Ingram (eds), Language Acquisition: Models and Methods.
London and New York: Academic Press.
Ikegami, Y. (1980) ‘On some categories for describing the semolexemic
structure’, in J. E. Copeland and P. W. Davis (eds), Papers in 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 on
Languages and Linguistics 17).
Lamb, S. M. (1988) Discussion in M. A. K. Halliday, S. Lamb and J. Regan
(eds), In Retrospect: Using Language and Knowing How. Claremont, CA.:
Claremont Graduate School (Issues in Communication 12).
Lamb, S. M. (1992) Outlines of a Cognitive Theory of Language: A Work in
Progress (pre-publication draft).
Lemke, J. L. (1984) Semiotics and Education. Toronto: Victoria University
427
bibliography
(Toronto Semiotic Circle Monographs, Working Papers and 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. Petöfi
and E. Sözer (eds), Micro and Macro Connexity of Texts. Hamburg: Buske.
Martin, J. R. (1983b) ‘The development of register’, in R. O. Freedle and
J. Fine (eds), Developmental Issues in Discourse. Norwood, N.J.: Ablex.
Martin, J. R. (1984a) Process and text: two aspects of human semiosis’, in
J. D. Benson and W. S. Greaves (eds), Systemic Perspectives on Discourse:
Selected Theoretical Papers from the Ninth International Systemic Workshop.
Norwood, N.J.: Ablex.
Martin, J. R. (1984b) ‘Functional components in a grammar: a review of
deployable recognition criteria’, Nottingham Linguistic Circular 13.
Martin, J. R. (1988) ‘Grammatical conspiracies in Tagalog: family, face and
fate – with regard to Benjamin Lee Whorf ’, in J. D. Benson, M. J.
428
bibliography
Cummings and W. S. Greaves (eds), Linguistics in a Systemic Perspective.

Amsterdam: J. Benjamins (Current Issues in Linguistic Theory 39).
Martin, J. R. (1992) English Text: System and Structure. Amsterdam: J.
Benjamins.
Martin, J. R. (1993) ‘The model’, in M. A. K. Halliday and J. R. Martin
(eds), 1993.
Martin, J. R. (forthcoming), ‘Lexical cohesion, field and genre: parcelling
experience and discourse goals’, in J. E. Copeland (ed.), Text Semantics and
Discourse Semantics: Proceedings of the Second Rice Symposium in Linguistics and
Semiotics. Houston: Rice University.
Martin, J. R. and Matthiessen, C. (1992) ‘Systemic typology and topology’,
in F. Christie (ed.), Literacy in Social Processes. Darwin: Northern Territory
University, Centre for Studies in Language Education.
Mathesius, V. (1928) ‘On linguistic characteriology with illustrations from
Modern English’, in Actes du premier congrès international de linguistes à 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: a
discussion of the dimensions of change’, Network 17.
Matthiessen, C. (1992) ‘Interpreting the textual metafunction’, in M. Davies
and L. Ravelli (eds), Advances in Systemic Linguistics: Recent Theory and
Practice. London: Pinter.
Matthiessen, C. (1993a) ‘The object of study in cognitive science in relation
to its construal and enactment in language’, in M. A. K. Halliday (ed.),
Cultural Dynamics 6.1–2 (Language as Cultural Dynamic).
429
bibliography
Matthiessen, C. (1993b) ‘Register in the round: diversity in a unified theory

of register analysis’, in M. Ghadessy (ed.), Register Analysis: Theory and
Practice. London: Pinter.
Matthiessen, C. (1995) ‘Theme as an enabling resource in ideational “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.
Mukařovský, J. (1977) The Word and Verbal Art: Essays, trans. by J. Burbank
and P. Steiner. New Haven, CT.: Yale University Press.
Mullie, J. (1937) The Structural Principles of the Chinese Language, Vol. 2.
Beijing, Peitang Lazarist Press.
Needham, J. (1958) ‘The translation of old Chinese scientific and technical
texts’, in A. H. Smith (ed.), Aspects of Translation. London: Secker &
Warburg (Studies in Communication 2).
Nesbitt, C. and Plum, G. (1988) ‘Probabilities in a systemic grammar: the
clause complex in English’, in R. P. Fawcett and D. J. Young (eds).
Nida, E. (1960) A Synopsis of English Syntax. Norman: University of
Oklahoma, Summer Institute of Linguistics (Linguistic Series 4).
Nystrand, M. (1987) The Structure of Written Communication: Studies in 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).
430
bibliography
Painter, C. (1996) ‘The development of language as a resource for thinking: a

linguistic view of learning’, in R. Hasan and G. Williams (eds), Literacy in
Society. London: Longman.
Palmer, F. R. (1958a) ‘Linguistic hierarcy’, Lingua 7.
Palmer, F. R. (1958b) ‘Comparative statement and Ethiopian semitic’,
Palmer, F. R. (1964a) ‘ “Sequence” and “order” ’, in C. I. J. M. Stuart (ed.),
Report of the Fifteenth Annual (First International) Round Table Meeting on
Linguistics and Language Studies. Washington, D.C.: Georgetown University
Press (Monograph Series on Languages and Linguistics 17).
Palmer, F. R. (1964b) ‘Grammatical categories and their phonetic exponents’,
in H. G. Lunt (ed.), Proceedings of the Ninth International Congress of Linguists.
The Hague: Mouton.
Pike, K. L. (1959) ‘Language as particle, wave and field’, Texas Quarterly 2.
Quirk, R. (1959) ‘Review of W. Nelson Francis: “The structure of American
English” ’, Archivum Linguisticum 11.
Reddy, M. J. (1979) ‘The conduit metaphor: a case of frame conflict in our
language about language’, in A. Ortony (ed.), Metaphor and Thought.
Cambridge: Cambridge University Press.
Robins, R. H. (1953) ‘Formal divisions in Sundanese’, Transactions of the
Philological Society.
Robins, R. H. (1957a) ‘Vowel nasality in Sundanese’, in J. R. Firth et al.
(eds), Studies in Linguistic Analysis. Oxford: Blackwell (Special Volume of
the Philological Society).
Robins, R. H. (1957b) ‘Aspects of Prosodic Analysis’. Vol. 1, Proceedings of the
University of Durham Philosophical Society 1. Durham: Durham University
Philosophical Society (series B).
Robins, R. H. (1959) ‘Status of grammar’, Lingua.
Robins, R. H. (1961) ‘Syntactic analysis’, Archivum Linguisticum 13.
Roget, P. M. (1960) Thesaurus of English Words and Phrases. London: Longman.
Russell, B. (1925) ABC of Relativity. London: Allen & Unwin (3rd revised
edn 1977).
Sadovnik, A. R. (ed.) (1995) Knowledge and Pedagogy: The Sociology of Basil
Bernstein. Norwood, N.J.: Ablex.
Schapiro, L. (1972) Totalitarianism. London: Pall Mall Press.
Sinclair, J. McH. (1966) ‘Beginning the study of lexis’, in C. E. Bazell et al. (eds),
In Memory of J. R. Firth. London: Longman (Longman’s Linguistic Library).
Sinclair, J. McH. (1991) Corpus, Concordance, Collocation. Oxford: Oxford
University Press.
Sinclair, J. McH. (1992) ‘Trust the text: the implications are daunting’, in M.
Davies and L. Ravelli (eds), Advances in Systemic Linguistics: Recent Theory
and Practice. London: Pinter.
431
bibliography
Sinclair, J. McH. and Coulthard, R. M. (1975) Towards an Analysis of Discourse:

The English Used by Teachers and Pupils. London: Oxford University Press.
Sinclair, J. McH., Jones, S. and Daley, R. (1970) English Lexical Studies.
Birmingham: University of Birmingham, School of English (Report to the
office of Scientific and Technical Information).
Smart, N. (1968) Secular Education and the Logic of Religion. London: Faber &
Faber.
Sugeno, M. (1995) ‘Intelligent fuzzy computing’. Paper presented to the
second conference of the Pacific Association of Computational Linguistics
(PACLING II), University of Queensland, Brisbane.
Svartvik, J. and Quirk, R. (eds) (1980) A Corpus of English Conversation. Lund:
C. W. K. Gleerup (Lund Studies in English 56).
Svoboda, A. (1968) ‘The hierarchy of communication units and fields as
illustrated by English attributive constructions’, Brno Studies in English 7.
49–101.
Sweet, H. (1891) A New English Grammar: Logical and Historical. Oxford:
Clarendon.
Tannen, D. (1982) ‘Oral and literate strategies in spoken and written
narrative’, Language 58.
Taylor, C. V. (1979) English for Migrant Education Curriculum. Canberra:
Education Development and Research Committee (Report 18).
Thibault, P. J. (1992) ‘Grammar, ethics and understanding: functionalist reason
and clause as exchange’, Social Semiotics 2.1.
Turner, G. J. (1973) ‘Social class and children’s language of control at age five
and age seven’, in B. Bernstein (ed.).
Ure, J. (1971) ‘Lexical density and register differentiation’, in G. E. Perren
and J. L. M. Trim (eds), Applications of Linguistics: Selected Papers of the
Second International Congress of Applied Linguistics, Cambridge 1969. London:
Cambridge University Press.
Ure, J. N. and Ellis, J. O. (1979) ‘Register in descriptive linguistics and
linguistic sociology’, in O. U. Villegas (ed.), Issues in Sociolinguistics. The
Hague: Mouton.
Vachek, J. (1966) The Linguistic School of Prague. Bloomington: Indiana
University Press.
van Leeuwen, T. (1988) ‘Music and ideology: notes towards a socio-semiotics
of mass media music’, Sydney Association for Studies in Society and Culture
Working Papers 2:1/2.
Winter, E. (1977) ‘A clause–relational approach to English texts: a study of
some predictive lexical items in written discourse’, Instructional Science 6.
Yngve, V. H. (1960) ‘A model and an hypothesis for language structure’,
Proceedings of the American Philosophical Society 104.5.
432
INDEX
abstraction 38, 41, 52–69, 185, 296, 351, arbitrary 3, 47, 114, 294, 300, 378–9,
411 397
degree of 46–7, 53, 114, 158 art 5, 357, 389, 402
higher order of 185, 296 article 98, 299
level of 220–1 articulation 354, 357, 364, 366
accusative 293 articulatory 7, 324
active 28–9, 400 attributive 185, 229, 243, 276, 336,
and passive 115, 182–3 413
Malinowski’s 227, 236 autostable 358, 362
actor 113, 176, 179–94, 204, 212, 224, auxiliary 28, 268
299 noun 32–4
additive 223 axis 23, 95, 97–8, 100, 109–10, 120,
adjective 55, 108, 187, 302, 336, 342, 163–4, 221, 402–3
384
adjectival 343 benefactive 181
adjunct 47, 56, 96, 98–9, 104, 225, Beneficiary 178–81, 193, 238
238 Bernstein, B. 7–8, 175, 403, 412
adverb 28–9, 95, 180 Bloomfield, L. 65, 219
adverbial 28, 32, 343 Bloomfieldian 4, 65
group 96, 103, 180, 263 Boas, F. 246, 262, 302, 311, 325
adversative 225 bonobo 3
Affected 187–8 branching
Agent 238, 244, 277, 298, 300, 381 singulary 122–5
allophone 68 Bühler, K. 173–4, 226–7, 234
alternative 223
ambiguity 104, 159, 163–4, 185–6, 343, Cantonese 32–4, 202
385, 399–400, 410, 416 Carrier 244, 413
ambiguous 298, 346–8, 385, 399–400 category 12–13, 21–34, 37–72, 95,
apposition 215 97–9, 106, 160–4, 167, 170,
appositive 225 177–9, 186–7, 200–2, 209, 215,
Arabic 223, 292 223–33, 242–3, 284–5, 291–313,
433
index
category (cont.) complex(es) 215, 217, 232, 242,

335, 343, 351, 354–6, 385, 397, 262–70, 281–2, 332–6, 341, 343,
399–400, 408, 410, 412 371
causal type(s) 181–7, 233, 395, 414
consequential 225 English 47, 70, 97–9, 186, 190, 239,
external 233 297–8, 394
relation 357 code 8, 196, 226, 233, 296, 309, 350,
causality 412
vs. redundancy 357 cognitive 116, 197, 231, 276–7, 282
causation 400 coherence 222–6, 232–5, 244–5
causative 186, 225 coherent 41, 216, 340
Cause 238, 305 cohesion 223–44, 263, 271, 280–5
chain 45, 67, 95, 97–100, 208, 224, cohesive 170, 175, 342
232–5, 244–6, 292, 352, 357, 360 cohesive harmony 224, 232, 235
chain-exhausting 166 collocation(al) 33, 60–1, 158–70, 201,
chain-exhaustive 120 282
chaos 353, 365–6 communication 175, 189–200, 193,
Chinese 25–33, 167, 202, 294–301, 199–201, 294, 351
372–3, 385, 415–16 communicative 236, 283
box 43 complement 47, 96–102, 113, 238, 374
choice 33, 51–3, 95–9, 100, 116–18, clauses 336
163, 174, 182, 192, 198–200, 228, conative 226–7, 236
239–40, 262, 268, 279, 283, 301, conceptual 173, 216, 245, 301, 363, 378
304–7, 310, 328, 348, 356, 364, concord 55, 70–1
377, 380, 395, 404 tone 266
chooser 301 constituent structure 113–16, 121–6,
points 395 204–6, 216, 238
Chomsky, N. 4, 6, 72, 106, 112, 304 construe 3, 12–13, 306, 310, 353, 357,
class(es) 24–34, 41, 45, 49–68, 95–104, 362–6, 369–77, 382, 390–1, 393,
106–13, 123–5, 159–68, 184–5, 396, 398–9, 402, 407–14
212, 222, 234, 294, 307, 371, de-construe 373, 355, 363
374–5, 397, 412–14 context 23–33, 39, 56, 70, 96, 99, 174,
class-defining 101 190, 198–202, 220–1, 225–9,
class-sequence 109 243–4, 293, 300–1, 306–11,
class-structure 50 324–5, 336, 344, 349–51, 400–5
class-type 108 function in 398
natural 391 of situation 29, 201, 221–9, 243, 263,
clause 26–32, 45–58, 68–70, 95–104, 283, 311, 357–9, 405
109–15, 121–4, 175–94, 205–17, social 10, 201–2
219–22, 228–47, 262–83, 297–8, conversation 7–8, 169, 326, 346, 351,
305, 308–9, 329–48, 352, 371–7, 371
394–5, 414 coordination 215
clause-classes 26 coreferential 224
clause-final 238, 252 corpus 8, 9, 38, 466–8, 412
clause-initial 238 corpus-based 159–60, 362, 407
clause-like entities 221 grammar 412
clause-to-text analogy 241 cortex 272
434
index
creolized 362 of choice 356

crescendo 228, 240, 271 of classes 26–9
criteria 24–34, 40, 42, 45, 48, 55–6, 58, of realization 365
61, 67–72, 96–7, 107, 114, 119, diminuendo 228, 240, 270
161, 166–8, 170, 178–9, 341–2, diminuendo-crescendo 233, 243
398–401, 410 discourse 7–8, 10–11, 175, 189, 193–4,
(situational-) contextual 32–3 199–209, 225–8, 239, 242–3,
cryptogrammatic 407 245–7, 261, 268, 270–3, 282, 285,
crypotype 302 292, 294, 296, 302–3, 311, 324–5,
crypotypic 360 329–31, 335–50, 365–6, 369, 371,
crystalline 303, 336, 350 377, 386–7, 391, 396–7, 407, 412,
415
Davidse, K. 380, 413 spoken 270, 324, 331, 335–7, 340,
declarative 109, 111, 189, 233, 268, 273, 403
305 spontaneous 325, 337–40
deep written 331, 335–6, 340–6, 348–50
grammar 116
structure 106, 116 ecolinguistics 9
definite 98, 185, 299–300, 307 Eggins, S. 411, 413
deictic 55, 99–101, 243 eidological 232
deicticity 272 eidon 231
postdeictic 100 eidos 231
predeictic 100 ellipsis 181, 225, 232, 237, 281–2
delicacy 40–1, 48–70, 98–9, 114–15, Ellis, J. 174, 227, 391
158–9, 165, 223, 285, 293, 396, embedding 126, 343–4
402–6, 415 embedded 227, 270, 280, 329, 341–2
degree of 48, 54, 57, 69, 99, 159, 223, empirical(ly) 119, 242
404 enact 3, 5, 356, 382, 390, 392, 407
most delicate grammar 49, 54, 59, 405 encode 202
depth 48, 58, 101–4, 107, 120, 285, 327, encoded 202, 204, 220, 240–1, 307,
333 342, 346
depth-ordered 97 encoding 202, 235, 292–3
determiner 99, 104 endocentric word groups 243
diachronic 23, 324 English 33, 44–8, 54–5, 58, 60, 70,
synchronic-diachronic 22 96–104, 113–15, 120, 160, 167,
dialect 7, 32–4, 202, 377 175–92, 202, 206, 209–10, 214–17,
British rural dialects 377 228–46, 266, 270, 276, 279, 282,
Wu and Yueh dialect groups 34 297–312, 323–30, 343, 347–50,
dialogue 225–33, 239–40, 271, 283, 308, 362, 369–82, 385, 392–400, 405–7,
325, 335, 341 410, 413–16
dictionary 54, 158, 160, 165, 186, modern 187–8, 191, 261, 375–7
392–3, 400, 407 spoken 70, 101, 326, 343, 350
diglossia 296 written 101, 261, 330, 343, 350
dimension(s) 22, 42–3, 100–1, 159, 161, equative 183, 185, 243, 300
164, 188, 194, 222, 225, 232, 241, ergative 28–9, 186–8
328, 355–62, 391 and transitive 380–1, 400
of abstraction 59, 66, 96 see also voice
435
index
ergativity 312 linguistic 111

Eskimo folk tales 231 rhetorical 226–34
ethnographic 173, 226, 231, 236, 386 speech 233–9, 268, 273, 307
ethnographer 230 syntactic 107
ethological 227, 232, 236 textual 175–6, 182, 193–5, 237, 273
ethos 4, 231 functional
evolution 65, 303, 349, 355–62, 370, categories 209, 300
377–8, 390, 392, 413 component(s) 200–1, 211, 215, 241
evolutionary semogenesis 362 element(s) 175, 225, 242, 262
exemplificatory 45, 70, 72 environment 110, 122
existential 243, 275 interpretation 200, 235
exophoric 201 labels 107, 203
experiential 198–217, 224, 276, 284, schemata 236
377, 379, 390, 400, 405 semantic(s) 209, 237, 311
exponence 41, 45, 53–72, 97, 221–2, tenor 227, 231
293 theories 226, 235–6
Firth’s concept of 352 variety 301, 307
scale 57–61, 66 variational 403
see also rank fuzzy 3, 210, 417
exponent(s) 23–7, 45–72, 97, 116, 221 computing 402, 416
expressive 227, 236 set(s) 409–10, 417
Extent 204, 212, 238
extralinguistic 31, 236, 295 gender 302
extratextual 39–40 generative 37
genetic 22, 32–4
Field genitive 293, 300, 336
field, tenor, mode 201, 217, 221, genre(s) 209, 222, 230, 234, 242, 312,
227–31, 243, 283–4, 364 357, 413
particle, field and wave 209–11, 241 gesture 354, 356, 366
Fillmore, C. 178–9, 181, 187, 194 given
Firth, J. R. 7, 12, 21, 24, 25, 38, 67, 106, and new 29, 190–1
109, 110, 158, 170–1, 174–5, 210, Given 192–4, 207, 209, 238, 270–1, 376
219, 262, 296, 301, 307, 311, 384, Gleason, H. A. 222, 244
387, 398, 403, 414 Goal 113, 176, 178–83, 186–8, 203–4,
foreground 235 212, 300, 312, 376, 379, 381
foregrounded 234 Goal/Medium 305
foregrounding 230, 349, 379, 409 goods and services 199, 273
free/bound 27–8 grammar
French 372, 373, 381, 415 descriptive 30, 163, 415
Fries, Peter 228, 233, 245–6, 371 lexis 37–67, 165, 379, 404, 410
function 12, 107–16, 122, 173, 202–4, phonology 56–68, 220, 239
212–47, 262, 281, 292–4, 299–308, protocol version 415
329–50, 376–82, 389–416 rank 121–7
experiential 202, 224 semantics 220, 239, 306, 324
ideational 175–93 systemic(-functional) 261–2, 332
imaginative 227 theory 41, 44, 67, 370, 401–2, 416
interpersonal 272, 382 see also grammatics
436
index
grammatics 11, 365–6, 369–78, 384–417 indeteminacy 179, 239, 310, 335, 338,
trinocular 402–9 365–6, 392, 399–402, 409–11
graphology 39 indicative 111
Greek 294–6 Indo-European 35, 236
ancient 292, 365 infant 3–4, 304, 323, 388–9
ancient Greece 371 inferential relations 11
alphabet 104 information
flow 184
focus 207–8, 233, 270
Hasan, R. 11, 175, 222–5, 229, 231, 242,
prominence 271
244, 261, 285, 351, 404, 406,
retrieval 170
411–13
structure 192, 216, 233, 266, 269–70,
hearer 199, 205, 207, 301
305, 312
hearer-oriented 199, 207, 240
theory 40, 42, 70, 72
hesitation 205, 337–8, 340
unit 192, 207, 215, 262, 266, 270–2
hierarchy 25–6, 42–4, 56, 59–60,
inherence 302
110–11, 115, 119–24, 166, 213,
instantial 10, 279, 405
228, 242
instantiation 12, 262, 352, 359, 362, 387,
Hjemslev, L. 4, 5, 12, 106, 109, 110,
410, 412–13
112, 219, 236, 262, 301, 312, 354,
interaction 175, 189, 199, 201, 210,
401
216, 227, 230, 310, 377, 382, 390,
Hockett, C. 106, 112, 219, 221
392
Huddleston, R. D. 120, 125, 215, 349
interactional 353, 363
Hudson, R. 204, 349
interlevel 39, 56, 67
hyponomy 12, 226
interpersonal 175, 182, 189, 199–217,
hyponym 226, 282
227–45, 268, 273, 284, 307–8, 311,
hyponymic 396
337, 356, 364, 382, 384, 390, 392,
hypotaxis 107, 266, 302, 327, 332–3,
401, 413
333, 343–4, 362, 410
interrogative 111–13, 189, 191, 200, 233,
hypotactic 213–17, 242, 266, 266,
268, 274
282, 333
intersubjective 245, 354–5
see also parataxis
intonation 55, 114, 123–5, 192–3
contour(s) 205, 217, 238, 270, 399
iconicity 312 pattern 114, 123, 193
ideational 210–11, 216–17, 227–44, 268, of pause 26
298, 308, 311, 348–9, 356, 364, and rhythm 114, 262–3, 405
382, 384, 392, 401 intransitive 28, 164, 187
component 186, 198–200, 208, 237, intricacy 331–7, 341, 343
242
features 243 Jacobson, R. 189, 226, 301
function 175, 177, 189, 193 Jesperson, O. 298–9
meaning 177, 188, 193, 199, 201, 217, juncture 55, 68
229, 295 see also tone
semantics 13
structure 231, 241–2, 244 Labov, W. 222, 350
voice 230 Lamb, S. 5–6, 108, 112, 198, 355, 378,
indefinite 185, 299 383, 389
437
index
language 1–13, 21–30, 37–72, 95–104, Location 238

106–17, 119–25, 158–71, 173–95, logic 3, 6, 8, 198–9, 212, 399, 402, 416
197–217, 219–20, 226–46, 261, logical-semantic 262, 264, 266, 281–2
282–5, 291–313, 323–51, 353–66, logo-centric 414
369–83, 384–417 logo-genetic 360
child language development 3–4, 310, Longacre, R. E. 106, 222, 242
411 Lyons, J. 114, 178, 194
“language-in-action” 229, 329
teaching 7, 170, 403 McIntosh, A. 227, 396
natural 175, 198, 237, 296–309, 364, Malinowski, B. (Kasper) 173, 226–7,
369, 389–417 236–7, 262, 311
spoken 4, 7, 10, 261, 302–6, 323–48, Mann, W. C. 301, 394–5
366, 406 Manner 204, 212, 238
written 43, 227, 323–51 Martin, J. 13, 223, 225–7, 231–3, 242,
langue 44, 236–7, 403, 412 285, 310, 312, 344, 357, 373–4,
Latin 25, 95 389, 392, 395, 397, 412–13
Lemke, J. L. 306–7, 312, 352, 356, 358, Mathesius, V. 190, 202, 311, 324–5, 371
370, 382, 387, 391, 412–13 Matthiessen, C. 3, 13, 301, 365, 370,
lexical density 229, 327–32, 335, 341–2 373, 390, 397, 407, 411, 413,
lexicogrammar 3, 6, 8–12, 163, 168–9, 416–17
185, 217–21, 231, 233, 239, 241, meaning 3, 9–13, 23, 39–41, 45, 61, 66,
243, 246, 261–2, 281–5, 294, 298, 71, 95–8, 121, 170, 174–93,
341, 345, 357–8, 369, 378–9, 387, 196–207, 221–4, 227, 231, 237–47,
389, 391, 393, 396–8, 401, 403–4, 262, 276, 294–5, 297–312, 323,
408, 411–12 325, 346, 348, 350–66, 369–74,
lexicoreferential 224, 232, 244 386–93, 397–414
lexico semantic 246 act of 201, 348, 354, 356, 391, 414
lexico syntactic 328 choices in 307, 310
lexicogrammatical entities 241 contextual 40, 61
lexicogrammatical system 345 formal 40, 71
see also wording(s) meaning-creating 8, 354
lexis 37, 39–43, 49, 54–61, 67, 98, 104, meaning-making 397
158, 162–70, 379, 404, 410 meaning-oriented 350
lexis-based 407 Medium 238, 274, 277, 380–1
lexis-driven 407 Melrose, R. 227–8, 242
linguistics 1–12, 21–30, 37–40, 42, 46, metafunction 12–13, 268, 284, 310–11,
51, 56, 60, 68, 72, 95, 98, 105, 356–8, 364–5, 382, 390–2, 411,
118–19, 123, 158–60, 170, 174–7, 413
219, 236–7, 243, 292, 294–8, 307, metalanguage 30, 293–300, 309–13, 354,
311–13, 324–5, 351, 365, 384–6, 366, 369, 378, 384, 415
406, 411, 415–17 metaphor
listener 175, 189–90, 192–3, 209, 227, clause as 222, 234, 245
237, 240, 243, 270–3, 338, 344, conduit 293–4, 297
350 grammatical 12, 280, 282, 345–51,
listener-prominence 240 358–60, 397, 400–1
literate 323, 325, 405 rhetorical 328
literature 2, 5, 37, 230, 313, 323 metaredundancy 352, 356–8, 362
438
index
microclass 101, 159 185, 190, 280, 292–3, 295, 300,

Mitchell, J. F. 205, 223 307, 342–3, 397, 410
modal 216, 242, 245–6, 300, 340, 399
subject 190–1, 194 order 43, 46, 116, 213–14, 269, 362,
modality 200, 205, 215, 230, 234, 237–8, 364–6, 390
242, 245, 268, 271–4, 278–82, 399 higher order consciousness 388–9
visual 312 higher order semiotic 313, 388,
mode 221, 227, 229, 283 410–11, 409
active 199, 237 social 356, 390
experiential 202 word 29
interpersonal 205 orthography 39, 261, 400
logical 211–15 orthographic 46, 124, 406
metaphoric 397 orthographies 120
reflective 199, 237 O’Toole, M. 389, 402
textual 206
model 159–67 Painter, C. 363, 397
bricks-&-mortar 404 Palmer, F. R. 106, 108, 116
modifier 47, 108 paradigmatic 49, 61, 106, 109–16,
modifier-head 108–9 120–3, 160–3, 216, 262, 378,
mood 189–90, 194, 199–200, 205, 215, 402–5, 411
230, 233, 237, 242, 262, 268, paragraph 211, 228, 234, 243, 246
271–3, 364 paraphrase 114, 325, 346
morpheme 45, 53–69, 71, 95–6, 103, parataxis 266, 327, 333, 343, 362
124–5, 164, 219–20, 295, 298 paratactic 213, 216–17, 242, 266, 282,
morphemics 69 333
see also hypotaxis
parole 236, 324–5, 365, 403, 412
narrative 222–4, 227, 230–1, 237, 242, see also langue
245, 309, 327, 375 parser 407
Navaho 312 parsing 246, 395, 407
Nesbitt, C. N. 362, 407 participant 23, 26, 31, 178–93, 200–2,
network 12, 40, 48, 109, 111–16, 174, 215, 224, 227, 230, 237, 244, 263,
200–1, 212–13, 225–30, 242, 301, 268, 273–4, 277, 303, 307, 374–5,
310, 363, 390, 393–410 377, 380–1
neural 5, 370, 395 participial 326
New 192–3, 207, 238, 240, 243, 270–1, participle 336, 381
300, 305, 307, 326 particle 204, 209, 211, 239
news 231, 238, 244, 269–70, 376 particulate 211, 215, 232, 239, 242
Nigel 301, 353–4, 362–4 passive 28–9, 35–7, 182, 190–3,
nominal 32–4, 54, 70, 104, 190, 309, 298–300, 400–1
342, 348, 371 imperative 190
group 51, 55, 70, 96–9, 102–4, 180, tenses 375
204, 212, 243, 263, 279, 342–3, verbs 375
396, 410 past 102, 273, 279, 326, 379, 400
nominative 189, 293, 300 pedagogical 32, 300
non-finite 40, 278, 326 Pekingese 27, 29, 31–4
noun 28–34, 51, 55, 95, 108, 167, 180, Penman 301, 394, 417
439
index
person 70, 230, 236–7, 268, 337, 353, mental 183–8, 282, 398
378 relational 229, 234, 364, 413
phatic 226 semogenic 227, 355, 364
phoneme 65–6, 68, 71, 96, 219–20, 222 social 201, 227, 386, 407
supra-segmental 68 type 177, 235, 243, 263, 274, 277,
phonetic 26, 71 282, 364, 399
phonetics 39, 324 pronominal 29, 31, 305
phonetic/kinetic 354 pronoun 24, 29, 31, 326, 372
phonic 25, 39, 67 prosodic 67, 71, 115, 205–17, 238,
phonology 3, 6, 33, 37–9, 56, 60, 65–71, 364
104, 120, 206, 219–20, 239, 262, prosodies 206, 232
297, 324, 357, 387, 389, 408 protolanguage 4, 304, 349–51, 353–65,
Chinese 292 378, 389
diachronic 324 protolinguistic 354–64
phonological 32–3, 39, 55, 66–71,
114–15, 208, 213, 231, 239, 293, question 174–6, 189, 191–3, 233, 268
324, 338, 345, 357 question-and-answer 283
phonological-lexical 22 Quirk, P. 9, 371, 407
phonological-morphological 22
prosodic 67, 71 Range 238, 300, 348
see also grammar rank 41–69, 95–103, 115–16, 118–26,
phrasal verb 101, 376 159–66, 212–15, 221–2, 242, 268,
phrase 45, 68, 121, 125–6, 180, 206, 292, 343–4, 415
263, 278, 298, 336, 342, 410 scale 51–69, 123
phylogenetic 355, 360 see also exponence
Pike, K. 106, 174, 211, 239, 242 rankshift(ed) 103–4, 121–4
Pinyin 33 upward rankshift 122–4
plosive 34 realization 44, 46, 52, 108, 112–16, 124,
Plum, G. 327, 362 175, 194, 196–212, 217, 220–1,
plural 295, 307 231, 236–46, 273, 285, 307, 310,
polarity 12, 61, 230, 279, 363, 398 352, 357–9, 362–5, 404–13
polarity carrying element 189, 266 realizational chain 292, 352
polysemy 294, 297, 386, 405 realizational cycles 196–7, 204
possession 327, 371, 373 Recipient 305
possessive 100, 275 recursion 101, 213
postmodifier 342 recursive 45, 102–4, 121, 212–17, 242,
pragmatic 10, 11, 412 279, 343
Prague School 174, 190, 262, 292, 299, Reddy, M. 293–4, 297
311, 371 redundancy 65, 70–2, 295, 356–7, 387
predicate 102, 188, 194, 298–9 reflection 197, 227, 236, 353, 355–6,
predicator 47, 99, 101–2, 113, 238 362, 382, 389
preposition 193, 376–7 reflective 197, 237, 329, 356
prepositional 178, 180, 193, 263, 298, 342 relational 106–7, 227, 241, 277–8
present 70, 102, 110, 274–5, 326, 379 clause(s) 181, 185–6
Priestley, J. B. 234, 245 see also process
process 187–8, 203–5, 212, 224, 238, representational, conative and 226–7,
244, 274, 305, 346 236
440
index
rhema 292, 294 speaker/listener 395, 405

rheme 188–91, 205, 226, 231, 236, 264, speech 6–8, 174, 176–7, 192, 195,
266, 269, 300, 371, 376 197–9, 206–10, 215, 219, 229–35,
rhetorical 191, 199, 221, 224–6, 231–3, 239–43, 266, 268, 273, 282–3, 302,
240–4, 283, 328, 371, 411 305–8, 323–31, 335–41, 345,
rhythm 114, 216, 262–3, 338, 348, 405 348–51, 362
rhythmic 207, 269 see also language, spoken
Rumbaugh and Rumbaugh 3, 4 speech-functional 233
Russell, Bertrand 349 speech-like 345
Russian 372–3, 392, 400, 416 spontaneous 302, 310, 323–41, 369, 377,
382, 406
Sanskrit 296 see also discourse
Sapir, E. 219, 246, 262, 311, 377 strata/stratal/stratum 3, 115, 196–7, 222,
Saussure, F. 4, 220, 262, 325, 403, 412 262, 294, 296–7, 301, 308, 352–65,
science 5–6, 9, 296, 396 369–70, 378–9, 387–90, 397–8,
language(s) of 12, 365–6 402–16
linguistic 21–2, 65 stratified 196, 356, 364, 389, 397–8,
technology and 291, 392, 395 409
semantic 11–12, 32, 114, 159, 170, 175, stratification(al) 112, 115, 222
185, 196–216, 220–46, 283–7, Subject 96–101, 113, 176, 183, 191, 194,
294–312, 324–50, 354, 373–8, 238, 262, 268, 271–4, 298–310,
389–416 372–5, 410, 416
component(s) 200, 234–41, 389 psychological 189–94
system 197–216, 237, 310–11, 345 Sugeno, M. 3, 12, 402, 416–17
semiotic 1–12, 109, 196–247, 296–313, Svartvik, J. 181, 371
325, 353–66, 370–82, 387–414 Sweet, H. 179, 189, 299
higher-order semiotics 3, 389, 392 syllable 33, 219–22, 231, 234, 292–3
space 353–65 syntagmatic 27, 46, 61, 106–16, 120–5,
system(s) 8, 12, 109, 196–7, 298, 160–4, 216, 262, 305, 378
301, 311, 313, 325, 356–7, syntactic 56, 71, 95–7, 106–7, 109, 115,
370–3, 387–8, 392, 397, 402, 406, 166–7, 182, 185, 280, 335
414 syntax 12, 51, 369, 386
transformation 390, 408 systemic
semogenic 8, 227, 303–6, 354–64, description 110–12
392–7 feature(s) 112–15, 362, 404–5, 411
semogenesis 357–62 grammar(s) 301, 332, 406
Sinclair, J. McH. 168, 170, 242, 407 grammatics 402, 417
Slav 2 meaning potential 414
sociolinguistics 10 network 115
Soviet 2 theory 215, 348, 415–16
speaker 31, 174–6, 188–92, 199–200, typology 33
205–7, 226–7, 229, 236, 240, 243,
270–4, 301–11, 325–6, 337–8, 344, Tagalog 374
371, 376 tagmemic 107, 110, 121, 222
speaker-now 273 taxonomic 26, 279, 404
speaker-oriented 207, 240 taxonomies 291, 379, 396
speaker-prominent 240–3 taxonomy 42–3, 199, 264, 371
441
index
tense 25, 112, 215, 273–4, 279–80, 300, universal 21–31, 95, 120, 301
310, 326, 365, 375, 379, 393–4, universals 12, 209, 414
400 unmarked 33, 164, 167, 207, 233,
text 7–12, 22–30, 38, 45, 49, 109, 166, 240–1, 264–9, 279, 305
169, 175, 184, 192, 199–201, Urdu 374
207–11, 217–47, 261–87, 294–5, Ure, J. 227, 229, 327
300–11, 324, 328–35, 338–47, 350,
360–3, 386–7, 394, 412 verb 25, 28–30, 51, 95, 167, 178,
text-forming 206–7 180–3, 186–7, 190, 236, 297–302,
text-generation 284 312, 326, 346–8, 376, 394, 410–11
text-like 234 postpositive 28
text-linguistic 222 prepositive 28
texture 207–8, 211, 224–5, 233, 271, pro-verb 28
281 verbal group 47, 51, 70, 102, 243, 263,
Theme 111, 113, 190–4, 206–9, 216, 280, 325, 393–5, 400
228, 230, 233–5, 238–46, 266–8, Vietnamese 34
270–3, 297–300, 303–9, 347, 357, voice 28–9, 182–4
371–8, 416 active 29
Theme-Rheme 209, 228, 233, 262, ergative 28–9
271
(un)predicated 111 WH- 189, 206, 268, 275
Token 292–3, 296–8, 364, 366, 413 Whorf, B. 219, 262, 293, 302–3, 306,
tone 67–8, 164, 192, 205, 270 311, 312, 373, 397
concord 266 word 26–34, 44–5, 51, 55, 58–60, 69,
contour 269, 399 95–104, 106, 116, 121–6, 164, 180,
group(s) 55, 192, 207–8, 266, 269 190, 205–6, 217, 221–2, 239, 243,
key 203 280, 296, 303, 371
tonic 113–14, 126, 192, 270 class(es) 28, 31, 34, 95–6, 104
accent 206, 208, 270 wording(s) 3, 222, 345
prominence 206–7, 270 meaning 197, 231, 293–4, 310,
segment 270 356–7, 402
topic 266 patterns of 219–20, 241
and comment 299 see also lexicogrammar
sentence 211, 228, 234, 243 writing 227–9, 302, 405
Topic 300 expository 225
transcription 262–3, 345 scientific 349
transitivity 176–93, 200, 215, 224, 229, speech vs. 323–51
233–7, 243, 246, 263, 268, 274, theory 227
276, 302, 305, 312, 364, 382 see also language, written
typological 13, 120, 397, 402
typology 9, 33–4, 167, 388, 399 Zadeh, L. A. 399, 416
442

On Grammar (Collected Works of Mak Halliday Series) (PDFDrive) PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

On Grammar (Collected Works of Mak Halliday Series) (PDFDrive) PDF

Uploaded by

Copyright:

Available Formats

On Grammer

First published 2002 by Continuum

Reprinted 2003, 2005

All rights reserved. No part of this publication may be

British Library Cataloguing-in-Publication Data

ISBN 0-8264-4944-1 (hardback)

Typeset by SetSystems Ltd, Saffron Walden, Essex

SECTION ONE: EARLY PAPERS ON BASIC CONCEPTS

SECTION TWO: WORD–CLAUSE–TEXT

8 Modes of meaning and modes of expression:

SECTION THREE: CONSTRUING AND ENACTING

For nearly half a century, Professor M. A. K. Halliday has been

8 Studies in Chinese Language

We are grateful to the original publishers for permission to reprint the

‘Some aspects of systemic description and comparison in grammatical

‘Modes of meaning and modes of expression: types of grammatical

The volumes in this series will contain a selection of my writings on

by the way that computers happen to be being designed and operated

An exceptional ﬁgure in this period was Sydney Lamb, who took

but you have to know language in order to understand them. Com-

human society, as expounded by my teacher J. R. Firth and by my

and ill-at-ease at what seems to invade the interactants’ privacy and

was to say that he or she was “data-oriented”. Data were said to be

If linguistics had not fragmented in this way, with the system of

This sort of thing was often used to challenge those of inferior

familiar in the lexicogrammar, like hyponymy and polarity and voice,

based on the (theoretical) category of system, for matching up descrip-

wander throughout the highways and byways of language. If there has

In this ﬁrst section we look at ﬁve papers written and published by

categories, ordered as terms in interrelated systems and having as

Struggling with the grammar of Chinese, and then of English, in the

In a paper appearing in Linguistics in 1963, ‘Class in relation to the

constituent can be assigned to one or other of the speciﬁed layers, or

SOME ASPECTS OF SYSTEMATIC

1 Descriptive and historical, particular and comparative

correspondences as a basis for genetic groupings; this especially in Britain,

Any of these types of study may be undertaken with the use of

have earned the name of “structural” linguistics. The term is perhaps

2 Particular description in grammar: units and classes

the highest there may be set up the complementary system of free /

clause–classes: 1. verbal / nominal.

mention one further instance of the application here of formal methods

3 Language under description, language of description

times be found preferable to reference to undeﬁned categories of the

Reference Chinese English

Contextual reference is as follows: 1, speaker; 2, addressee; 22, address-

As a conclusion to the foregoing it may be remarked that reference

certain terms in the two systems of auxiliary nouns.9 In this comparison,

‘Transfer grammar’ (Chavarria-Aguilar 1954), which is an interesting

CATEGORIES OF THE THEORY

There have been in the main two approaches to description in modern

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

1.1 One part of General Linguistic theory is a theory of how language

1.2 The relevant theory consists of a scheme of interrelated categories

1.3 Description consists in relating the text to the categories of the

1.4 The theory requires that linguistic events should be accounted

1.6 The complete framework of levels requires certain further subdi-

(a) Substance may be either phonic or graphic.

1.7 The study of phonic substance belongs to a distinct but related

Figure 1 Levels of language

1.9 Contextual meaning, which is an extension of the popular – and

1.11 Finally, it is necessary to distinguish not only between theory

(a) The number of terms is ﬁnite: they can be listed as A B C D,