Syntax: A Linguistic

Introduction to Sentence
Structure
 p
2*15
Syntax: A Linguistic Bi
Introduction to Sentence
Structure

E. K. Brown and J. E. Miller

HUTCHINSON
London Melbourne Sydney Auckland Johannesburg
Hutchinson & Co. (Publishers) Ltd
An imprint of the Hutchinson Publishing Group
24 Highbury Crescent, London N5 irx
Hutchinson Group (Australia) Pty Ltd
30-32 Cremome Street, Richmond South, Victoria 3121
PO Box 151, Broadway, New South Wales 2007
Hutchinson Group (NZ) Ltd
32-34 View Road, PO Box 40-086, Glenfield, Auckland 10
Hutchinson Group (SA) (Pty) Ltd
PO Box 337, Bergvlei 2012, South Africa
First published 1980
© E. K. Brown and J. E. Miller 1980

The paperback edition of this book is sold

subject to the condition that it shall not, by
way of trade or otherwise, be lent, re-sold,
hired out or otherwise circulated without the
publisher’s prior consent in any form of binding
or cover other than that in which it is published
and without a similar condition including this
condition being imposed on the subsequent purchaser
Set in vip Times Roman by Preface Ltd, Salisbury, Wilts.
Printed in Great Britain by The Anchor Press Ltd
and bound by Wm Brendon & Son Ltd,
both of Tiptree, Essex

British Library Cataloguing in Publication data

Brown, E. K.
A linguistic introduction to sentence structure.
1. Grammar, Comparative and general - Syntax
I. Title II. Miller, J. E.
415 P291

isbn 0 09 138620 9 cased

0 09 138621 7 paper
Contents

Preface 7
Introduction 9

Part one Constituent structure

1 Constituent structure 21
2 Form classes 32
3 Constituent structure grammar 40
4 Some verb classes in English 50
5 Formal grammars 60
6 Optional and obligatory constituents 69
7 Selection restrictions 85
8 Relations between sentences 102
9 Embedding and recursion 134
10 The sentence 149

Part two Morphology

11 Words and morphemes 161
12 Morphemes and morphs 173
13 The analysis of words 186
14 The order of morphs and the order of morphemes 205
15 Lexical morphology 222
16 Form classes and grammatical categories 231

Part three Functional relations

17 Syntagmatic and paradigmatic relations in syntax 253
18 Processes and participants 288
6 Contents
19 Subjects, objects, complements and adjuncts 330
20 Theme, rheme and end focus; topic and comment;
given and new 357

Postscript 383
Further reading 386
References 388
Index 391
Preface

We would like to explain briefly why we have decided to write

an introduction to syntax when there are many such books
already on the market.
This volume arose from the teaching of syntax to first-year
undergraduate students in the University of Edinburgh. Since
the first-year course is designed to give students a general
overview of linguistics, we thought that the syntax component
should cover those basic concepts that have been and still are
being used in the description of the syntactic structure of
languages. This decision rules out those many textbooks that
purport to be introductions to syntax or to grammar in general
but are really introductions to transformational grammar. Of
course, many basic concepts - constituent structure, embedding,
recursion, inter-sentential relations, and so on - are integral
parts of transformational grammar, but they are not exclusive to
this particular approach to description. Other equally basic
concepts - dependency (syntagmatic) relations, propositional
processes and participants, the notions of theme and rheme,
topic and comment, and so on - are not well handled in
transformational models. Of the other textbooks available, none
treats the range of concepts we want to consider in enough
detail to form the single text round which a course could be
built.
We recognize the commanding position that the
transformational approach has assumed in the field of syntactic
description over the past decade. It is still the dominant
linguistic paradigm and has contributed more to our
understanding of the nature of language than any other single
model: indeed the companion volume to this is a fully
formalized transformational grammar. But the concepts we wish
to discuss are independent of and prior to any single model of
grammar: it is precisely notions such as constituent structure,
8 Preface
embedding, recursion, dependency relations and so on that any
model seeks to account for, and linguists criticize one or another
model precisely because it deals with one or another of these
concepts inadequately. Our approach does not rule out formal
grammar. A formal approach to language description is one of
the corner stones of linguistics, and we hope to encourage
students to look at the formal structure of language. It will be
clear that much of the formalization we introduce in Parts one
and two derives from a transformational approach, and we
introduce some transformational rules. This book is not
intended, however, to be an introduction to a fully formalized
transformational grammar, and a formal approach is not
exclusive to transformational grammar. The companion volume,
however, Brown and Miller, in press, discusses the problem of
constructing a fully formalized transformational description
of constituent structure, compares two transformational models
and explores ways of formalizing other aspects of syntax.
There is another reason, external to linguistics, why we take
this approach. At least half of the students who take a first-year
course do not proceed any further with linguistics, though many
do continue language study, and have occasion to use grammars.
Most, if not all, detailed grammars of English or other languages
are not set out as a formalized transformational grammar, but
rather make informal use of such concepts as constituent
structure, subject and object, complement and adjunct, agent
and patient, government and concord, and so on. At least part
of our purpose therefore is to train students to understand such
concepts, so they can make the best use of the grammars they
come across.
This book has been written with undergraduate students in
mind, but it should be suitable for beginners at any level. We
think it appropriate to have as few references as possible in the
actual text, as our subject matter is those basic concepts that
have become the common property of linguistics. The references
at the end of the book give some idea of the sources from which
we have drawn inspiration: they are not an exhaustive list of the
writings on each topic, but an indication of books the student
should find accessible.
Introduction

It is indisputable that language is central to all communities of

human beings. One of the first tasks of the young human being
is to learn the language of the society into which he or she is
born, and it takes many years to acquire the mastery of
language expected of mature adults. Language is essential for
the regulation of every community: the instruction of its young,
the creation of laws, the development of its culture, the
identification of its members. Consequently, language, as well as
being a fascinating phenomenon in itself, is a necessary part of
any investigation into human social organization and psychology.
This book is concerned with language for its own sake. To be
more accurate, it is concerned with one part of language,
namely syntax. But before we embark on the study of syntax,
we should consider some general questions about the nature of
language study. Let us begin by examining an instance of one
person talking to another. If both speaker and hearer are
mature speakers of the same language and neither suffers from
any defects - such as deafness or laryngitis - that impede the
production and the hearing of speech, we can say that the
speaker produces a sequence of sounds, and that the listener
hears the sounds and understands the message. This
rudimentary account ignores many factors necessary in a
detailed description of communication by language, but it will
suffice for present purposes.
If both participants were literate, and perhaps not in close
proximity, the speaker could have decided to convey his
message in writing. The person receiving the message would
then have been required to read a series of marks on paper, the
marks being organized in sequences of different lengths
separated by spaces. If the message was a long one, there might
be several series of marks, each series being separated from the
other by a space on the paper. Of course, the sequences are
10 Introduction
ordinarily called words and the series are ordinarily called
sentences.
If we compare a written message with a tape-recording of a
spoken message one obvious discrepancy is that nothing in the
series of noises corresponds to the spaces between the words on
paper, not even a pause or a slight cough. Furthermore, while
there may be slight pauses corresponding to the gaps between
the sentences, there need not be. The machine records a
sequence of mere noises. If the language being spoken is one we
know, we can interpret the noises, reconstructing the words,
phrases, and sentences encoded in them. But if we know
nothing of the language being spoken, we will make nothing at
all of the transmission.
It is obvious - once it has been pointed out - that the same
message can be conveyed either in speech or writing. This
suggests that the notion of ‘language’ is rather more abstract
than one might suppose. Language, we might say, does not
intrinsically depend on either speech or writing (or indeed on
any specific means of expression, like the system of manual
signs employed by the deaf), but is a more abstract notion. We
might say that a language is the system of relationships that a
speaker has in his mind, and that the language can be expressed
in speech or writing or by another medium. The linguist usually
works with this more abstract notion of language. At the same
time much can be said about written and spoken language in
general or about a written and spoken language in particular,
witness the success with which linguists write descriptions of
other languages and assiduous students can acquire enough
knowledge from these descriptions to communicate in them.
These remarks about a speaker and a listener have an
important consequence for anyone who wishes to study
language. To talk about language in any sensible fashion we
have to draw a number of distinctions that are not normally
drawn overtly in school grammars of languages.
In the first place, languages are organized on two levels: the
level of expression - to put it crudely, the level of the noises in
which the message is encoded - and the level of content. The
word ‘content' is unfortunate, since this level encompasses both
the organization of words into sentences and the meanings that
are associated with these words and sentences, but it is the
standard term.
 Introduction 11
The linguist needs a sound knowledge of both levels of
organization. Since language may be viewed as something
abstract and messages can be encoded in sounds or writing,
in principle the study of the level of expression will extend
to both writing systems and sound systems. In practice, most
attention has focused on the expression of language in sound,
reflecting the general consensus among linguists that spoken lan¬
guage is more basic than written language. The study of the level
of expression of sounds belongs to phonetics and phonology.
Phonetics is concerned with the description of physical sounds,
the functioning of the organs that are involved in their produc¬
tion and the range of sounds the speech organs can produce,
and phonology is concerned with the ways different languages
make systematic use of sounds for the transmission of messages.
We shall not be directly concerned with either in this book.
We said in the previous paragraph that spoken language is
more basic than written language, yet the written language
enjoys greater prestige in all communities in possession of a
writing system and a literature. Written language has a greater
range of vocabulary, a more complex syntax and, except perhaps
in hastily composed letters, none of the stops, starts and
repetitions of spoken language. So great is the power of written
language that in some formal situations, such as the delivering
of judgment in court or the preaching of a sermon, it is felt
appropriate to use the syntax and vocabulary of written
language for proceedings conducted in speech.
Nevertheless, it is the spoken language which is fundamental.
Every normal human being learns to speak before he or she
learns to read and write. Even where literacy is universal most
people conduct most of their communication in speech
(including radios, television sets and telephones), and many
books and newspapers are written in a style that approaches the
style of spoken language.
The history of languages contributes two facts that put the
primacy of spoken language beyond doubt. Firstly, no case is
known of a community that had a written language before it had
a spoken one. Speech always developed a long time before the
emergence of a writing system. The second is that changes in
language over time by and large originate in the spoken
language and gradually find their way into the written language.
12 Introduction
From the level of expression let us pass to the level of content,
bearing in mind that the term ‘content’ is too narrow. Study of
the content level has traditionally been split between syntax and
semantics. Syntax is concerned with ways in which words
combine to make sentences, and semantics deals with meaning.
Traditionally a third component was also recognized -
morphology, dealing with the structure of words. It is evident
that if sentences can be split up to make words, words can be
split up into smaller units. For example, ineffectiveness consists
of four parts: in, effect, ive and ness. Morphology is the subject
of Part two of this book. It is in many ways a sort of ‘bridge’
between the content and the expression level. It is related to
syntax in a natural manner since the parts of words can be
considered as participating in the structure of sentences along
with the words themselves. Morphology is also relevant to the
study of the expression level, because the structure of words is
relevant to the functioning of systems of sounds. However, that
aspect of morphology is not of major concern in this book.
The relationship between syntax and semantics is difficult and
controversial. It has always been recognized that there is some
distinction to be drawn between the meaning and the structure
of sentences, but it is far from clear just where this boundary
should be drawn. Language is a means of conveying meanings of
various kinds, and it would be surprising indeed if
considerations of meaning and of structure were totally separate.
This book takes the view, however, that syntactic and
morphological structure should be studied in their own terms as
far as possible, the results furnishing a basis for investigations
into semantics. Syntax and morphology are gateways to
meaning, and it is important to have secure syntactic and
morphological criteria against which intuitions about meaning
can be measured. But, as we shall see, discussions of syntax
cannot, and should not, totally exclude considerations of
meaning.
In the course of our discussion we will need to redefine and
refine a number of concepts and terms commonly used in
language description. One we can usefully tackle immediately is
the ‘sentence’. In ordinary usage this term refers sometimes to
the actual sequences of sounds produced by a speaker,
sometimes to an orthographic unit, and sometimes to something
much more abstract. In this book we use the term ‘utterance’ to
 Introduction 13
refer to actual sequences of sounds, and restrict the term
‘sentence’ to the more abstract use. We consider the sentence to
be a unit described in the syntax at the abstract level of content.
In order to understand why we make this distinction, consider
the relationship between a sentence and its corresponding
utterances. More than one utterance corresponds to a given
sentence, in fact, a multitude of utterances. Take the sentence
which we may represent orthographically as What did you think
of the programme on Comecon? Even in the speech of one and
the same speaker this sentence may be expressed (or realized)
by a large number of different utterances. If speech is slow there
may be a sound in the utterance corresponding to the last letter
in what and a separate sound to each of the letters in did, to the
first letter in you and to the last two letters in you. (To talk of
‘separate sounds’ in this way is to skate over some very
formidable problems in the analysis of speech.) But if the
speaker speaks at even ordinary conversational speed there may
be only one separate sound corresponding to the last letter in
what and the first letter in you. This part of the utterance might
be represented as whadja, with dj representing one sound.
The discrepancy between the number of ‘letters’ and the
number of separate sounds in our sentence may be considerable.
Moreover, other factors increase the number of possible
utterances corresponding to our sentence - all within the speech
of our one speaker. For example, the speaker’s attention may be
distracted, causing him or her to repeat a word, or perhaps to
miss out a word entirely. The speaker may inhale a crumb of
toast and finish the utterance in gasps and coughs. The speaker
may suffer from laryngitis that has all but paralysed some of the
organs of speech, so the utterance is deformed.
Furthermore different speakers can produce a range of
utterances that are interpreted as realizations of our original
sentence. The quality of voice, for instance, is clearly different
between men and women; and no two men or women have
exactly the same quality of voice. Similarly, speakers of the
different accents of English may not aim at the ‘same’ sounds. If
all these factors are put together the utterances which can
realize our sentence vary enormously.
Yet all of these different utterances are related by the hearer
to the same, abstract, sentence. Native speakers clearly practise
such an abstraction in everyday conversation. Normally we
14 Introduction

simply do not notice if a word is repeated or a couple of sounds

are slurred (unless the interference is so great that the intended
sentence cannot be reconstructed). Indeed we are often able to
guess at the complete sentence after hearing only part of the
utterance. It is this abstract notion of the sentence that the
linguist usually works with.
Let us accept then that the sentence is an abstract unit. Lin¬
guists set up such a unit to explain how we relate a range of
utterances to the same abstract sentence, and to explain patterns
within the sentence. For the moment we can suppose that these
patterns are built up of words: it turns out that the only satisfac¬
tory way to make sense of the arrangements of words is to con¬
sider the patterns they form within sentences.
Words are dependent on each other in many different ways.
For example, they may have to occur in a certain order. In
English, words like the and a precede the nouns they go with,
but in some other languages what corresponds to the in English
is placed after the noun. Again, one word may not be able to
occur unless another word is present. In the sentence Foreign
books are expensive, foreign cannot occur unless there is a word
like books. The forms of words may also differ according to
what other words are present in the sentence. In Latin, for
instance, the adjective corresponding to good is bon-, the dash
indicating that various items can be added. If the adjective goes
with a noun like puer it ends as bonus but if it goes with a noun
like puella it ends bona. These relationships or dependencies
between words can be sensibly described only if we establish the
sentence as a unit in our description. We define the sentence
then as an abstract unit postulated in order to account for the
dependencies between units of syntactic structure.
When the abstractness of language was first mentioned, along
with the distinction between the levels of content and expres¬
sion, there was a short discussion of the relationship between
written and spoken language. Linguists, it was argued, have
good reason to assign priority to spoken language. Closely
related to the linguists’ attitude to spoken and written language
is a very important point concerning the task of describing the
syntax of a language.
To a great extent we treat the topics that used to be gathered
together under the heading of ‘grammar’, a word that for many
people has a strong association with the idea of ‘good grammar’.
 Introduction 15
The popular understanding seems to be that the grammar of a
language, whether your native language or a foreign language,
contains information that enables its user to write and speak the
language correctly. This understanding has some foundation, but
the notion of correctness is far from simple.
The complexities of the notion can be illustrated by the situa¬
tion in Britain. A child learning English in Exeter, Manchester
or Edinburgh produces utterances that satisfy the expectations
of the people he or she comes in contact with: relatives, the
children he or she plays with, the local shopkeepers. (This is the
simplest of cases. For many children the learning of the appro¬
priate English is not so straightforward, for instance the child of
Bengali-speaking parents in an area inhabited by English speak¬
ers.) The child cannot string any sort of noises together in any
order. He or she must produce approximately the noises that
the people roundabout produce, and the noises must be com¬
bined into sequences as they are by those other people.
Every community has rules of language that have to be
observed by its members. The person who does not observe
them is not understood and, more seriously, is open to ridicule.
Conversely, the child who produces appropriate utterances
speaks correctly from the point of view of the local community.
When the child goes to school, he may speak incorrectly from
the point of view of his or her new community. At some stage
in his or her education the child is encouraged to adapt to a new
standard of correctness, usually that of the literary language.
Different children adapt to different degrees. The question
whether adaptation is necessary or desirable is highly emotive
and divisive, and one we do not tackle here. From a linguistic
viewpoint, the literary language is only one dialect of English,
even though many may consider it the ‘correct’ form. All
dialects of English, literary and non-literary, have their own
rules, and these rules are capable of description. The grammar
itself makes no evaluative judgements - it is users of the lan¬
guage that make such judgements, for reasons (like, for
instance, social prestige) external to grammatical considerations.
What is considered to be ‘correct’ usage can change over
time. Let us illustrate this. Twenty years ago the linguistic
observer in Britain would have recorded utterances that could
be related to sentences like This book is different from that book
and This book is different to that book. A manual of ‘faults’ to
16 Introduction

be avoided by the public speaker would have listed the different

to construction as one to be avoided. The linguist, however,
would have observed and recorded utterances corresponding to
sentences with the different to construction. He would also have
brought them into his description. He might also have sought to
account for the rise of this construction by noting the similarity
in meaning between X being ‘different from’ Y and X being
‘opposed to’ y, or by observing the occurrence of to with adjec¬
tives like opposite and opposed. Change has not stopped there.
In the late 1970s the linguist might record utterances related to
sentences like This book is different than that book. He might
bring this too into his description, accounting for the construc¬
tion by noting the connection between saying that X and Y are
‘different’ and that X and Y are ‘to be compared’. The compara¬
tive construction makes use of than, as in X is smaller than Y,
and we note constructions like other than and rather than.
The different than construction is probably even less accept¬
able to the authors of public-speaking manuals than the different
to one. Yet many speakers of English use both constructions,
and in the course of time both may become acceptable even to
the purist. If the linguist declined to record such constructions
he would not only be falsifying the facts of the language, but he
would also be throwing away data that affords an insight into
how a language functions as a system, and how such systems
change.
We do not assert that linguists never come across ‘incor¬
rect’ data - our previous discussion of slips of the tongue, the
omission of words and so on illustrated that they do. We main¬
tain, however, that ‘correctness’ is a notion relative to particular
uses of language and to particular dialects or subdialects. The
description of a language may concentrate on the literary dialect
of the language or on some other dialect, but it is inappropriate
to assert that the former deals with ‘good’ and the latter with
‘bad’ grammar.
Finally let us say that the study of language can be different
and exciting. No complete grammar of any language, be it a
literary language or a dialect, has yet been written. Even the
enormous amount of work done recently on the syntax of
English, far from solving all the problems of English, has only
served to increase the number of complexities that require
investigation.
 Introduction 17

Anybody with a little training in syntax (and in other

branches of linguistics) will find much to occupy his attention.
Let us recall that people have to be taught to observe; whether
it be the doctor learning to observe and ‘read’ the human body;
the artist learning to observe and ‘read’ great paintings; or the
geologist and geomorphologist learning to observe and ‘read’
the landscape. At the very least, a knowledge of syntax should
enable the student to observe and ‘read’ the utterances he hears,
opening his ears to a multitude of curiosities.
'
Part one

Constituent structure
1 Constituent structure

The analysis of the structure of sentences was traditionally

known as ‘parsing’. Part of the Oxford English Dictionary's
definition of PARSE is: ‘to resolve (a sentence etc.) into its
component parts of speech and describe them grammatically’.
The two operations suggested by this definition suffice for our
present purposes:
(i) the analysis of the sentence into its relevant parts;
(ii) the grammatical description of these parts.

These two operations are clearly closely related, and in

linguistic work have commonly been referred to as constituent
structure analysis. A constituent is some proper subpart of a
sentence relevant to its analysis. Part of a constituent structure
description of a language, therefore, tells us how to break
sentences down into their constituent parts, and which strings of
words are, and which are not, considered to be constituents. Or,
seen the other way round, the description tells us how the
smallest constituents can combine to form other, larger,
constituents, and how these, in their turn, can form yet larger
constituents until we have constructed a sentence. This suggests
that there are different types of constituents, and we need to
know the degree of likeness and difference between different
types, which constituents can combine with which other
constituents to make sentences, what order constituents can, or
must, occur in, and so on: this involves the naming of different
types of constituent so that we can identify them.
Let us, as an example, consider how to describe the structure
of the sentence:
1 The dog frightened the child
We have to make three initial assumptions, all of them justified
in due course. Firstly, although we are considering only a single
22 Part one: Constituent structure
sentence, it is not the only sentence in the language, and we
can, indeed must, assume that we can use our knowledge of
other potential sentences to guide our analysis. The description
we give to this sentence must be compatible with the analysis we
would like to make of other sentences we might have chosen:
our ultimate aim is the description not of a single sentence but
of the language as a whole.
Secondly we assume that the smallest constituents with which
we are concerned are the five words represented in 1. This
assumption begs a number of questions. For one thing, it is
clearly not quite true, since a word like frightened is analysable
into the smaller constituents frighten and -ed, on the analogy of
other forms of FRIGHTEN like frighten-ing, frighten-s or indeed
frighten in sentences like:

2a Dogs frighten children

2b The dog is frightening the child
2c The dog frightens the child

and so on.1 We also know that constituents like -ed, -ing and -5
occur with other verbs like KILL: kill-ed, kill-ing, kill-s. We
consider the analysis of words in due course; for the present we
need merely to acknowledge that the word is a traditionally
recognized unit, and that a unit of this sort indeed seems a
relevant constituent in English.
Our third assumption concerns the sentence. Suppose we use
the neutral term ‘string’ to refer to any sequence of constituents:
we can then refer to the words the dog or frightened the child or
indeed the whole of 1 as strings without a commitment as to
their status as constituents or any identification of the type of
constituents we believe them to be. What entitles us to refer to
the string 1 as a sentence, when the other strings mentioned do
not seem to be sentences? Unfortunately, as with the word,
there are considerable problems about the identification of
sentences: to call a string a sentence implies that it has a certain
sort of unity, but it is far from easy to describe exactly what sort
of unity a string needs to have for it to be referred to as a

1 We use small capitals to refer to the dictionary or lexical entry of particular

items. Thus frighten is the lexical entry which relates the actually occurring
forms of frigh ten, like frightens, frightening, frightened etc. The form we use
is the one typically chosen in English dictionaries. For further discussion see
page 166.
 Constituent structure 23
sentence. We might regard the study of syntax as an attempt to
answer just this question. For the time being we merely accept
that 1 is a sentence; again our provisional justification is that it
is supposed that we would all agree that 1 is indeed a sentence.

1 The dog frightened the child

With these three assumptions in mind we can turn to the

analysis of 1. What relevant units can we discern? At an
intuitive level the dog, frightened and the child appear relevant
constituents in a way that strings like frightened the or dog
frightened do not. Furthermore, again at an intuitive level, the
strings the dog and the child seem to be constituents of the same
type. We can back these intuitions by formal criteria if we
consider the relationships between our sentence and other
sentences in the language. Basically two types of criteria are
used: one has to do with features of a constituent related to its
internal structure, the other with aspects of the external
distribution of the constituent. Both of these criteria show that a
particular constituent has the unity that justifies its being treated
as a unit.
Consider first the strings the dog and the child. In their
internal structure, we note that both strings contain the word
the, that this word does not co-occur with frightened:

3 *The dog the frightened the child2

and that it must be the initial word in its constituent:

4 *Dog the frightened child the

Furthermore, these strings can be substituted by a single word,

and the sentence as a whole remains grammatical:

5a Fido frightened the child

5b The dog frightened Johnny
5c It frightened him

Similarly, these constituents can be substituted by other, longer,

strings, which we consider to be ‘expansions’ of the strings in
question, and the sentence remains grammatical:

2 Here and throughout the book an asterisk (*) indicates an ungrammatical

sentence.
24 Part one: Constituent structure
6 The big dog frightened the child next door
The big dog that lives next door frightened the child

Similar substitutions and expansions are not possible for strings

like frightened the. If we now turn to the word frightened, we
can make comparable observations. Since it is a single word,
and we are treating words as the smallest constituents, we can
say nothing about its internal structure, but it can certainly be
substituted by various expansions:

7a The dog has frightened the child

7b The dog may have frightened the child

We may note that the expansions appropriate here (has, may

have) are impossible with constituents like the child:

8a *The may have child frightened the dog

8b *May have the child frightened the dog

just as the expansions appropriate to constituents like the child

are impossible with constituents like frightened:

9a *The dog big frightened the child

9b *The dog frightened that lives next door the child

So much for the internal structure of constituents: particular

types of constituent may have characteristic structural properties
(the child, but not *child the), complex constituents may be sub¬
stitutable by single words (the dog, Fido, it) and constituents
may be appropriately expanded (the dog, the big dog; frightened,
has frightened).
The second criterion has to do with the external distribution
of a particular constituent: in what range of environments can
the constituent be found? Does it retain its integrity as a
constituent in other structural positions within sentences?
Consider once more the strings the dog and the child. To begin
with they are mutually substitutable:

10a The dog frightened the child

10b The child frightened the dog

but neither of them is substitutable for frightened. Next note

that these strings retain their unity in sentences like those in 11,
all of which, you would probably agree, are systematically
related to our original sentence 1:
 Constituent structure 25
11a The child was frightened by the dog
lib It was the dog that frightened the child
11c What the dog did was frighten the child
lid What happened to the child was that the dog
frightened it

We may also note the relevant question and answer sequence:

12a What frightened the child? The dog

12b What did the dog frighten? The child

The constituents we are interested in have a unity difficult, or

perhaps impossible, to dissolve. We also note that the
substitutions and expansions discussed above show exactly the
same unity, since substitution of Fido or the big dog that lives
next door for the dog in any of the sentences in 11 or 12 still
yields grammatical sentences. There is no way in which strings
like frightened the show a comparable unity.
The discussion shows, then, that there is some justification for
regarding the constituents of our sentence as:

The dog - frightened - the child

The discussion also shows that constituents like the dog and the
child are constituents of the same type, whereas frightened is a
constituent of a different type. Following traditional usage we
call constituents like the dog Noun Phrases, abbreviated as NP,
and constituents like frightened a Verb, abbreviated as V.
Our first analysis, then, suggests that the sentence has three
constituents, which we can represent by the string:

13 NP + V+NP
(Note that we use string to refer both to strings of words, and to
strings of constituent names.)
We should now consider the constituents of these constituents
since the NP clearly has internal structure: it consists of the
word the followed by either child or dog. Child and dog appear
to be the same sort of constituent since they are mutually
substitutable for each other, and different from either frightened
or the since neither is substitutable for either of these words: we
will refer to them, again following the traditional terminology,
as Nouns abbreviated as N. We have already noted that the is a
different type of constituent from either of the Ns we have
identified or the V: we have noted that it does not co-occur with
26 Part one: Constituent structure
a V, and that it may co-occur with an N, but that if it does it
must precede the N in the NP: there are no strings *dog the or
* child the. Again following traditional terminology, we call the
an Article, abbreviated as Art.

Since we have agreed to stop with the word, no further analysis

is possible. Let us see how it meets the two requirements set out
at the beginning of the chapter (page 21).
We have analysed the sentence into a tripartite structure:

14 NP + V + NP

we have analysed each of the NPs into a structure

15 Art + N

and we have identified each of the words as belonging to one of

three types of constituents:

16 frightened is a V
dog and child are Ns
the is an Art

We have also given a partial grammatical description of these

items. The is an Article, an item which can only occur in an NP,
and must precede the other constituent of the NP, the N. Ns
only occur in NPs and they must follow the Art. The V
frightened occurs between two NPs: there are no sentences:

17 *Frightened the child the dog

*The dog the child frightened

There are two common ways of representing this information

about constituent structure. One is to indicate constituent
structure by bracketing:

18 (((The)(dog))(frightened)((the)(child)))

There are brackets round each word, since each word is a

constituent. Since the strings the dog and the child are also
constituents, these are also bracketed. Since the sentence as a
whole is a unit, there are brackets round the whole string.
(There are always as many right-facing brackets as left-facing.)
Bracketing shows the constituent structure but omits
information about the type of constituent. This can be included
by labelling each bracket with a suitable subscript:
 Constituent structure 27

19 S(NP(Art(the)Art N<d°8)N>NP V<fri8htened>V

NP<Art(,he)Art N<child>N> NP^S

The subscript identifying the type of constituent is to the left of

the bracket at the beginning of the constituent, and to the right
of the bracket at the end of the constituent. Alternatively we
can put subscripts only on the opening bracket:

20 S^N Art^6 Wd°g^^frightened)

NP(Art(the)N(child)))

Representations of this kind are called labelled and bracketed

strings.
Sometimes, however, such representations are not easy to
read, especially if there is a good deal of structure to represent;
an alternative method, often preferred, is the tree diagram,
illustrated in Figure 1.

the dog frightened the child

the dog frightened the child

Art N Art N

NP lNp-l
V
I
Figure 1 •s■

Whichever way up the tree is, it conveys the same information

as the labelled and bracketed string. It is more usual nowadays
to find tree diagrams as in the top half of Figure 1, with the
28 Part one: Constituent structure
constituents depending from the S node. In tree diagrams, those
places where there are labels are called nodes: thus we can refer
to the NP node, the V node, and so on. Where two or more
lines lead from a single node this is called branching: thus the S
and NP nodes branch. The other nodes do not branch - all
nodes do not have to branch.
At this point it is convenient to introduce further technical
vocabulary. First, the notion of domination: a node is said to
dominate all the constituents that are traceable back to it. Thus
in Figure 1, the first NP dominates the string the dog, but it
does not dominate, say, frightened. A more particular notion
is that of immediate domination: a node A is said to immedi¬
ately dominate another node B if B is traceable back to A
with no other node intervening. Thus S dominates everything in
the sentence - indeed it has to do so - but it only immediately
dominates NP, V and NP. S does not immediately dominate
Art + N.
A different way of talking about the same relationships uses
the term constituent: corresponding to the notion of domination
is the notion of constituents. Thus, the and dog are constituents
of the first NP. Similarly, corresponding to the notion of
immediate domination is the notion of immediate constituents:
Art and N are the immediate constituents of the first NP.
Concomitant with constituent is the notion of a construction:
two or more constituents are in construction with each other if
they are dominated by the same node. Thus Art and N are in
construction in the first NP. Finally ‘family tree’ terminology is
often used to describe the relationships between items; thus in
Figure 1, NP, V and NP are daughters of S, the leftmost NP
being the left daughter, and the rightmost NP being the right
daughter. Analogously, the leftmost NP is the left sister of V
(both are daughters of the same mother), and the rightmost NP
is the right sister of V, for the same reason.
A final observation about the analysis of Figure 1 is to note
its hierarchical structure: the sentence is analysed into its
immediate constituents, each of these constituents into its
immediate constituents, and so on. The hierarchical nature of
the analysis illustrates what appears to be a universal principle
of the organization of human languages, that they have
hierarchical syntactic structures. It is not immediately obvious
why this should be so, nor why this analysis is preferable to an
 Constituent structure 29

analysis in terms of a single string (a ‘string constituent

analysis’) as illustrated, for sentence 1, in Figure 2.

Art N V Art N

the dog frightened the child

Figure 2

Let us compare Figures 1 and 2. Firstly, Figure 2 does not

indicate that the NP is a relevant sentence constituent, yet our
previous discussion suggests it is an important constituent, in
that a number of generalizations about syntactic structure can be
more clearly formulated using such a constituent. Secondly, a
hierarchical analysis shows that the relationship between any
pair of constituents must be considered in terms of relationships
established within the tree as a whole; mere contiguity for
instance is not necessarily a particularly interesting relationship.
Thus, in Figure 1, there is a close relationship between the
contiguous items the and dog, they are constituents of NP, but
there is no immediate relationship between dog and frightened,
even though they are contiguous. Rather the V frightened has a
relationship to the NP the dog, since both are constituents of S.
This can be seen more clearly if we compare the relationships
between the strings the child and frightened in the two sentences
analysed in Figures 3 and 4. (We do not justify the analysis of
Figure 4 further here - it can be established in just the way we
have established our analysis of 1.) The string the rumour that
the dog chased the child is an NP which has the sentence the
dog chased the child embedded in it (such structures are further
discussed on pages 134ff.). In Figure 3 both the child and
frightened are daughters of S: in traditional terms the child is
called the ‘subject’ of the verb frightened; this sentence would
be used when the child in question ‘did’ the ‘frightening’. In
Figure 4 the child is a daughter of the embedded S, and in
traditional terms the ‘object’ of the verb chased (it is ‘the child’
30 Part one: Constituent structure

V NP

Art N Art N

the child frightened the dog

Figure 3

Art N Pro

Comp NP V NP
/\ /\
Art N Art N
ii ii
the rumour that the dog chased the child frightened me

Figure 4

that is ‘chased’), and frightened is a daughter of the

superordinate S. Whereas in Figure 3 the child is the subject of
frightened, in Figure 4 the subject of frightened is the NP the
rumour that the dog chased the child. The relationship between
the contiguous items the child and frightened is close in Figure 3,
but remote in Figure 4.
Finally, if we adopt a hierarchical analysis we discover that
‘the same’ constituents keep appearing at different places within
constituent structures. We have already observed this in the case
of the NP: we may now observe in Figure 4, that this is also
true of sentence structures, since one sentence of the structural
form NP + V + NP is embedded inside another sentence of
precisely the same form. This ‘nesting’ property of constituents
is an important characteristic of language, and one to which we
shall return.
 Constituent structure 31

Technical terms

bracketing immediate domination

branching in construction with
constituent labelled and bracketed string
constituent structure labelling node
construction sister of
daughter of string
domination substitution
hierarchical structure tree diagram
immediate constituent
2 Form classes

The previous chapter discussed the sentence:

1 The dog frightened the child

and assigned to it the constituent structure of Figure 5.

NP V
I
/\
Art N

the dog frightc

Figure 5

Many hundreds of sentences in English can be analysed in an

analogous fashion, for example:

2a The dog chased the child

2b The cat killed the bird
2c The child hit the dog
2d The dog caught the cat

etc. In these sentences the words fall into groups that are
mutually substitutable within the sentence structure shown in
Figure 5, as summarized in Figure 6.

Selecting one word from each column yields an acceptable

sentence with the analysis of Figure 5. Tables like these, which
may be familiar from foreign-language teaching manuals, are
called substitution tables, since they use the principle of simple
substitution. They are also very simple grammars.
 Form classes 33

dog frightened dog

the child chased the child
cat killed cat
bird hit bird
caught

Figure 6

A set of words like {dog, child, cat, bird} where the individual
words are mutually substitutable is known as a word class. Our
analysis has established three classes; the class containing the
one member {the}; the class containing {dog, child, cat, bird};
and the class containing {frightened, chased, killed, caught, hit}.
We have named them, following traditional usage, respectively,
Art(icle) N(oun) and V(erb).
Word classes are not the only classes of item needed in a
grammar, as is clear from the discussion in the preceding
chapter, where we saw that strings of words like the dog and the
child also need to be accounted for: we identified these as Noun
Phrases (NP). For this reason linguists have usually preferred
the more general term form class, since form may be applied
indifferently to words, parts of words, strings of words, etc. Our
substitution table can accommodate information about form
classes other than word classes by labelling and bracketing
within the table, as shown in Figure 7.

NP V NP

Art N Art N

dog frightened bird

the child chased the dog
cat killed cat
bird

Figure 7
34 Part one: Constituent structure
Two important and closely related points about form classes
are, first, the membership or the internal structure of the class
in question, and second, its distribution.
As far as word classes are concerned we can only list their
membership, since we are not considering here the internal
structure of words. Thus, the class of Articles consists of the
word {the} (so far the only article we have met); and so on. In
the case of other form classes it would be absurd and cumber¬
some to state their membership in terms of strings of words: the
class NP consists of the strings [the dog, the cat, the bird . . .};
etc. Instead we describe their internal structure in terms of other
form classes: the class NP consists of strings of the form
Art + N, leaving to our description of the membership of Art
and N a specification of what these classes contain.
We now turn to the question of distribution. By the
distribution of an item, we mean the set of environments in
which it occurs. In sentence 1 the distribution of frightened is:

3 The dog_the child

the sign indicating the place within the sentence, the

environment, in which this item occurs. Similarly we can say that
the distribution of the is the two environments:

4a _dog frightened the child

4b The dog frightened_child

But to state environments in this way is highly redundant; we

can make a more economical statement of distribution by using
notions of class membership, since this makes the most
economical statement of the relevant environment. In this case
frightened is a V, and the distribution of V is:

5 NP_NP

Similarly the distribution of the is the most economically stated

by stating its membership of the class Art and that the
distribution of Art is:

6 NP (_N)

i.e. articles precede nouns in NPs.

For each of the classes discussed so far a description can be
made in terms of both criteria: the membership, or the internal
 Form classes 35
structure of the class in question; and the distribution of the
class. In this way we can begin to make a grammatical
description of various classes. What is an Article? An Article is
a class of words, including the, which occurs in Noun Phrases
preceding a Noun.
Form classes may contain as members not only single words,
but also strings of words and indeed other form classes.
Consider the membership of the class NP. Let us first widen our
data base to include sentences like:

7a The boy chased Fido

7b Fido chased the child
7c Fido chased John

This data introduces a new form class, with the membership

{Fido, John}.
We call this class Proper Names (PN). The distribution of PN
is different from that of N in that members of this class may not
be preceded by an Art (*The Fido). We have already assigned
strings Art + N to a class labelled NP. We must now consider
whether the class PN is itself a member of the class NP, or
whether PN and NP are quite separate classes. If PN and NP
are considered quite separate classes, the analysis is as shown in
Figures 8 and 9.

Art N

Figure 9 Fido frightened the child

36 Part one: Constituent structure
In this analysis the two sentences concerned appear rather
different, since at no level do they have the same structure: the
immediate constituents of the two sentences are:

8 NP + V + NP (Figure 8)
PN + V + NP (Figure 9)

If, on the other hand, PN is considered a subclass of NP, the

analysis is as shown in Figures 10 and 11.

NP V

Art N

the
I I
dog frightened

Figure 10

NP V NP

PN

Fido frightened

Figure 11

Under this analysis the two sentences have the same structure at
the most ‘fundamental’ level since the constituents of the
sentence are, in both cases:

9 NP + V + NP

and the two sentences differ only in the internal structure of the
initial NP. This solution seems the more satisfactory on an
intuitive basis, and formal evidence supports it.
First of all, let us widen the data base further to include the
following sentences:
 Form classes 37
10a The boy yawned
10b The dog slept
10c Fido yawned
lOd John slept

As before, we consider strings like the dog and the boy as strings
with the structure Art + N, and members of the class NP. Fido
and John are once more members of the class PN. Yawned and
slept also form a substitution class, traditionally called
Intransitive Verbs, which we abbreviate as VI. Members of this
class cannot be put into the same class as words like frightened,
chased, etc., since these items are not mutually substitutable -
there are no sentences:

11 *The dog yawned the boy

*Fido frightened

etc. We therefore re-label the class containing frightened,

chased, etc., as the class Transitive Verb, abbreviated as VT.
This raises the question of why we refer to VT and VI as
different sub-classes of verbs, a question to which we return
later.
Using these two new verb classes, VT and VI, if we analyse
NP and PN as distinct classes, we end up with six distinct
sentence types:

12a NP + VI (sentences 10a,b)

12b PN + VI (sentences 10c,d)
12c NP + VT + NP (sentence 2)
12d NP + VT + PN (sentence 7a)
12e PN + VT + NP (sentence 7b)
12f PN + VT + PN (sentence 7c)

whereas if we take PN as a subclass of NP we need account for

only two types of sentence:

13 NP + VI
NP + VT+ NP

The distinction between Art + N and PN is now a matter of

the NP. This leads to a more economical description,
particularly if we seek to extend our data base still further, as
we will. This solution also makes possible a more economical
statement of the distribution of our two verb classes. If NP and
38 Part one: Constituent structure
PN are distinct classes, then we must describe the distribution of
our verbs in such terms as:

14 VI occurs in the environments: NP_#;PN — #'

VT occurs in the environments: NP_NP; NP_PN
PN_NP; PN_NP

If we consider PN a subclass of NP, a more economical

statement results:

15 VI occurs in the environment NP_#

VT occurs in the environment NP_NP

The evidence we have considered has been entirely based on

the substitutability of PN for strings Art + N. Also, the external
distribution of PN is the same as that of strings Art + N, as can
be determined by the appropriate substitutions in sentences like
11-12 discussed in the previous chapter. All this strongly
suggests that PN and Art + N form a single class, NP.
Let us now turn back to the statements of 15. Might not VI
and VT + NP themselves constitute a form class, sometimes
labelled Verb Phrase (VP)? An argument in favour of this
analysis is the fact that these strings are clearly substitutable for
each other following the initial NP:

16 NP + {vt + NP }

Furthermore, using the sort of argument adopted earlier (cf.

discussion of Figures 8 and 9), an analysis of this sort shows that
at the most fundamental level all our sentences have the same
structure, as shown in Figures 12 and 13.

S S

VT NP VI
Figure 12 Figure 13

1 The symbol # indicates a ‘constituent boundary’: in this case the end of the
sentence. We use this symbol to indicate that VI does not occur with a following
NP, as distinct from VT.
 Form classes 39
(Structure below the NP level is not shown since it is irrelevant
to the argument.) All the sentences are now analysed as having
the single structure:

17 NP + VP

rather than as being of two distinct structures. This analysis is

fine, but is only partially supported by other kinds of evidence.
For instance, the NP has a unity in external distribution, but
comparable evidence for the unity of the VP is not so strong.
For instance, although we do find sentences like:

18 John kissed Mary and Harry did so too

(i.e. Harry - kissed Mary)

where did so might be thought of as a substitution for the VP;

we also find sentences like:

19a John kissed Mary and Charlie Agnes

(i.e. Charlie - kissed - Agnes)
19b John kissed and Harry fondled Mary
(i.e. John kissed - Mary)

where constituents other than the VP are involved; so evidence

in favour of identifying a constituent of the VP type is less
conclusive than that leading us to recognize the NP. It can,
however, prove a useful constituent to have recognized, as will
be seen later, and it has the advantage that all our sentences are
analysed as having the same structure at the ‘deepest’ level: i.e.
NP + VP.

Technical terms
distribution substitution table
environment word class
form class
3 Constituent structure grammar

In chapter 1 we said that two of the tasks of a grammar are to

provide a systematic account of the constituent structure of
sentences and to account for the distribution of forms into
classes. In this chapter we produce a formal and explicit
grammar for the data discussed so far. To begin with it is
convenient to summarize our description.

A: Constituent structures. The constituent structures

discussed in the previous chapters can be summarized as:

1 A Sentence (S) has the constituents: NP + VP

A Verb Phrase (VP) has the constituents:

A Noun Phrase (NP) has the constituents:

B: Word class membership. There are a number of ways the

description can be arranged. We can organize our description so
that it answers the question, ‘What is the membership of the
classes VT, VI, . . . , etc.?’ or so that it answers the question,
‘To what class (VT, VI, . . . , etc.) does a particular word
belong?’ The first question would be appropriately answered by
a description which enumerates the membership of each class:

2 Art: {the}
N: {dog, child, boy, cat}
PN: {Fido, John}
VT: {frightened, chased}
VI: {yawned, slept}

The second question can be answered by providing a ‘lexicon’ or

word list in which the class membership of each individual word
is specified. For convenience, and to follow traditional
dictionaries, this list can be arranged alphabetically, though a list
 Constituent structure grammar 41
arranged by some other principle, or even an unordered list
serves our purpose equally well. A lexicon looks like this:

3 boy: N dog: N slept: VI

cat: N Fido: PN the: Art
child: N frightened: VT yawned: VI
chased: VT John: PN

The description above is itself a grammar in that it constitutes,

as required, a systematic account of the language. We shall,
however, further formalize our grammar, by using the follow-
ing sysmbols:

symbol interpreted as:

‘consists of’, ‘has the constituents’,
‘is to be expanded as’
+ ‘followed by’
{X,Y} ‘either X or Y, but not both’
X
‘either X or Y, but not both’
Y

The first symbol is used in rules like:

4 S —* NP + VP

In ordinary language we can interpret this as meaning, ‘The

sentence consists of the constituent NP followed by the
constituent VP.’ Rules of this sort are known as ‘rewrite rules’.
In rewrite rules there must be only a single symbol on the
left-hand side of the arrow: the right-hand side of the arrow
may have one or more symbols. If the left-hand side of the
arrow is not restricted to a single symbol it is unclear exactly
what is a constituent of what; for instance:

5 A + B—»X + Y + Z

This rule could mean X + Y is an expansion of A and Z an

expansion of B:

X Y Z
 42 Part one: Constituent structure

or the rule could mean X is an expansion of A, and Y + Z is an

expansion of B:

X Y Z

To specify the first analysis the rules are:'

8 A X + Y
B -» Z

To specify the second analysis the rules are:

9 A^ X
B -> Y + Z

Our rules must lead to an explicit and unambiguous account of

constituent structure: for this reason we must only ever expand
one symbol at a time, and hence only one symbol may be on the
left-hand side of the arrow.
The second symbol, +, indicates an ordering relationship
between constituents. Thus the rule:

10 A —> X + Y

means that A is to be expanded into two constituents, X

followed by Y. If the constituents may be in either order, the
rule states this explicitly:

X + Y
11 A
Y + X

This leads us to our two rule types involving ‘curly’ brackets.

Curly brackets indicate choices and mean that the symbol on the
left of the arrow may be expanded into one and only one
member of the set of items enclosed within curly brackets. The
alternatives are on different lines, when writing constituent
structure rules:

X + Y
12 A
P + Q
 Constituent structure grammar 43

The alternatives are on the same line and separated by commas,

when writing rules for the insertion of lexical items:

13 A —> {X, Y, Z}

Using this formulation we can write our grammar like this:

14 A Constituent structure rules:

IS -> NP + VP

B Lexical rules:
4 Art —> {the}
5 N —» {dog, child, boy, cat}
6 PN —» {Fido, John}
7 VT —» {frightened, chased}
8 VI —» {yawned, slept}

Rules 1-3 are called constituent structure rules since they

introduce and develop constituent structures. We call the first
symbol, S, the initial symbol, since it is this symbol that is
expanded first and the derivation of the whole sentence starts
with this symbol. In this grammar the symbols Art, N, PN, VT
and VI are terminal symbols since the constituent structure
rules cannot develop them further. VP and NP can be further
developed by the rules and are called non-terminal symbols.
This grammar uses the first of our two suggested approaches to
lexical insertion - the lexical rules enumerate the class
membership of each terminal symbol. The second approach to
lexical insertion, the one using a lexicon, involves scrapping the
lexical rules above and replacing them by a lexicon as in 3
above and a general lexical insertion rule:

15 Lexical insertion rule:

For any terminal symbol of the constituent
structure rules, select from the lexicon a
word that is described as being a member of
the class named by the terminal symbol in
question and attach this word as a daughter
of the relevant symbol.
44 Part one: Constituent structure
The output of both grammars (constituent structure rules and
lexical rules, or constituent structure rules, lexical insertion rule
and lexicon) is exactly the same. There are certain advantages in
using the approach involving the lexical insertion rule and a
lexicon, and this is the approach we shall use. However, the
lexical insertion rule is, formally speaking, a different type of
rule from the constituent structure rules - it is not a rewrite rule
as the constituent rules are:1
This proposed grammar can be used for an explicit analysis of
the sentences in our data. We start with the first rule, that
sentences have a structure NP + VP. This can be represented
by the partial tree in Figure 14:

Figure 14 NP VP

Note the convention in drawing trees from the grammar that a

rule like S^-NP + VP is interpreted as ‘write the symbol S,
and draw from it as many branches as there are constituents on
the right-hand side of the rule, and label the node at the end of
each branch by the appropriate symbol’; and similarly for the
other rules. Tree diagrams can be drawn on the basis of the
rules of the grammar. A grammar will contain information
about the possible structural configurations in the data it seeks
to account for. A tree diagram shows the structural
configuration of some one particular sentence. The second rule
in our grammar develops the VP as either VT + NP or as VI.
We chose the first alternative, Figure 15:

1 The rules presented above as ‘constituent structure rules’ are an informal

presentation of what are known in Transformational Grammars as ‘Phrase
Structure Rules’. The Lexical Insertion Rule is an informal presentation of a
type of Transformational Rule’. Transformational Rules are discussed at more
length on pages 102ff. For a fully formal account of the properties of Phrase
Structure Rules and Transformational Rules the reader is referred to any of the
books specifically about transformational grammar listed on pages 386-7.
 Constituent structure grammar 45

The third rule says that NP is to be developed as either Art + N

or as PN. There are two NPs; let us choose one of each,
Figure 16:

Art N

Figure 16 PN

We now come to the choice of lexical items. By the lexical

insertion rule we may insert from our lexicon words with the
appropriate class membership. This gives the sentence shown in
Figure 17:

Art N VT NP
I
PN
I
The boy chased Fido
Figure 17

The same result is obtained by using the lexical rules instead of

the lexical insertion rule. We have produced a labelled and
bracketed analysis of a sentence; providing we have operated
the grammar correctly - and providing that the grammar itself is
correct - we should have a grammatical sentence, as indeed we
do. The grammar also produces a definition of what is meant by
a grammatical sentence: a sentence that our grammar generates
is, by definition, a grammatical sentence. Other strings are not.
Our particular grammar can generate only a tiny number of
sentences in English, but the principle is clear: the grammar
defines grammatical sentences.
We can also use our grammar to analyse sentences. However,
the particular way we have chosen to write the grammar entails
a somewhat cumbersome procedure. Consider the sentence:
46 Part one: Constituent structure
16 The boy chased the cat
Our lexicon assigns these words to classes as shown in
Figure 18:

Art N VT Art N

Figure 18 the boy chased the cat

Inspection of the constituent structure rules shows that the

sequence Art + N constitutes an N P, but that no other sequence
of forms is a possible constituent. Incorporating this
information, we get Figure 19:

VT NP

chased cat

Inspecting the constituent structure rules again, we see that

VT + NP constitutes a VP; and finally we note that NP + VP
constitutes an S. Incorporating this information in a tree
diagram (Figure 20), we once more have a labelled and
bracketed analysis of a sentence:
 Constituent structure grammar 47

In principle a grammar tells you how to construct a sentence, or

how to analyse a sentence into its relevant parts, and in both
cases assigns a labelled and bracketed description to the
sentence. It also accounts for sentences other than those on
which it was immediately constructed - in other words it
predicts which sentences of the language are grammatical and
assigns descriptions to them. Grammars that can do this are
generative grammars. Our grammar can do all of these things,
though in the form we have here only for a very limited part
of English. Later chapters expand the description.

Technical terms
category lexical insertion
constituent structure rules lexical rule
expansion non-terminal symbol
generative grammar rewrite rule
grammatical terminal symbol
initial symbol

Exercises

1 Some verb classes in English

This is an exercise in verb subclassification.

1 The man seemed angry

2 The man laughed
3 The man terrified the children
4 The children cried
5 The woman telephoned the policeman
6 The policeman looked strong

There are three different sentence structures and hence three

different subclasses of verb. The relevant environments for the
subclassification of the verbs are shown below. Say which verbs
occur in which environments:

The verbs . . . occur in the environment — Adj

The verbs ... occur in the environment —NP
The verbs . . . occur in the environments — #
(The environment _ # means the verb occurs with no following
constituent.)
48 Part one: Constituent structure
Draw a tree diagram for each sentence type. Use the class labels
VCop (Copulative verb); VT (Transitive verb); and VI
(Intransitive verb) for the three classes of verb identified above.
Now consider the following additional data:

7 The man ran away

8 The man hid under the hedge
9 The policeman looked down
10 The policeman saw the man
11 The policeman caught the man
12 The woman was happy
13 The policeman stood near the playground

This data introduces two new sentence types and hence two new
verb subclasses (and also includes further examples of struc¬
tures we have already analysed). What are the environments for
these two new verb classes? Use the additional constituent
labels PP (Prepositional phrase) and Part(icle), and use the
labels VbPP and VbPart for the two new classes of verb. A
Prepositional phrase has the constituents Prep(osition) + NP.
The class Prep includes items like {under, near, in, on . . .}.
The NP can be any NP as described earlier.
Example 14 in this chapter presented a grammar for a re¬
stricted set of English sentences. Extend this gramipar to in¬
clude the data from this exercise. The extended grammar should
generate all the sentences in the data and others of a similar
pattern and assign to each a constituent structure tree.

2 The structure of the NP in English

There are restrictions on the possible order of constituents in

English. We find NPs like all their old cotton dresses, a lovely
pale blue silk shirt, five worn-out jumpers, but not NPs like
*their all old cotton dresses, *a pale lovely silk blue shirt.
Construct twenty NPs each of not more than six words using as
‘head noun’ one of the following: dress, shirt, jumper, sock,
petticoat, apron, blouse, and as ‘premodifiers’ of the head noun
appropriate items from the following list:

the, big, new, her, five, all, lovely, worn-out, woollen, blue,
spectacular, small, my, two, few, splendid, quite, silk, pale,
your, a, old, some, red, cotton, very
 Constituent structure grammar 49
(The notions ‘head noun’ and ‘premodifiers’ are discussed further
on pages 254-5: in NPs of the sort you are asked to consider the
head noun is always the final, and obligatory, constituent,
which may be preceded by a number of optional premodifiers.)

(a) Assign each word above to one of the form classes:

Quantifier; ail, several . . . Num(eral) five, two . . .
Art(icle) a, the . . . Adj(ective) blue, red . . .
Poss(essive) my, your . . . Int(ensifier) very, pale . . .

(b) There are ordering restrictions on the adjectives - thus old

blue dress but not *blue old dress. Assign adjectives to ‘order
classes’ labelled Adj1; Adj2, etc. (Can you make any semantic
generalizations about these order classes?)

(c) There are also co-occurrence restrictions on the

Quantifiers, thus all their socks, but not *several their socks.
Assign Quantifiers to one of the two subclasses Quant! and
Quant2.

(d) Write formulae to show possible orders of elements in the

NP (and to exclude impossible orders): e.g. all blue dresses
(Quant Adj N); all dresses (Quant N). Try and collapse the
formulae into a single general statement using brackets for
‘optional’ constituents: e.g. (Quant)(Adj) N - which will cover
the examples above: the N is an obligatory constituent and so is
unbracketed. You will need to add a note about the difference
between singular and plural head nouns - to account for the
fact that we find a with singular NPs (a red sock *a red socks)
and all with plural NPs (all red socks, *all red sock), etc.
4 Some verb classes in English

Implicit in the preceding chapter is the notion that decisions

about constituent structure have implications for the description
of class membership, and vice versa, and that these decisions
have implications for the type of grammar we write. Some of
these problems can be illustrated by considering a classification
of some verbs in English. The criterion used is the number and
type of constituents that follow a verb, a very traditional
criterion. It is what follows the verb that is criterial: what
precedes the verb is non-criterial since all verbs must be
preceded by a ‘subject’ NP. Some traditionally recognized verb
classes include the following:

‘Copular’ or ‘linking’ verbs

These occur in the environment:

2a John is a soldier
2b John is strong

These verbs, of which the ‘copula’ BE1 is the prime exemplar,

may be followed by either an adjective, as in 2b, or an NP, as in
2a; if they are followed by an NP the NP must agree in number
with the subject NP of the sentence:

3 He is a soldier
4 They are soldiers
5 *They are a soldier. *He is soldiers

1 It will be recalled that we introduced the convention of referring to the

dictionary or lexical entry of a particular item by the use of small capitals (see
note 1, page 22). Thus be is the lexical entry relating the various forms of be -
am, is, are, were, and so on.
 Some verb classes in English 51

(There are some exceptions to this generalization, for example if

the NP to the right of BE contains a ‘collective’ noun; They are
a crowd of lazy idlers', They are the best regiment in the army.)
Only copular verbs can be followed by either an NP or an
Adjective so we can treat NP and Adj here as members of a
single form class, defined by the environment We call
Cop
this form class, again following traditional terminology,
Predicate (Pred). This means we must change the statement at
the head of this section to read, These occur in the
environment:

6 _Pred

We then need a rule in our grammar to the effect:

The introduction of the labelled node, Pred, has two

advantages: it offers a unique environment for the classification
of copular verbs, and it enables us to identify which NPs are
‘predicate NPs’. We need to know this to ensure that predicate
NPs agree in number with the subject of the sentence. Other
copular verbs include seem, become, look.

‘Intransitive’ verbs

These occur in the environment:

8 _ #
9a The women wept
9b The children cried

Such verbs can occur as the final constituent of a sentence, as is

implied by the statement in 8 that they occur in the environment
of # , a symbol indicating a constituent boundary. The relevant
constituent boundary in this case is the right hand end of the
VP, for reasons which will become apparent. Intransitive verbs
include die, talk, work.
52 Part one: Constituent structure

‘Transitive’ verbs

These occur in the environment:

10 _N P
11a The dog bit the man
lib The child thrashed the dog

Verbs like BITE and THRASH were traditionally called

transitive verbs because ‘the action of the verb’ was considered
to ‘pass over’ from the subject to the object.2 Transitive verbs
cannot typically occur without a following NP:

12a *The dog bit

12b *The child thrashed

With all true transitive verbs, for any sentence of the form:

13a NP + V + NP
13b The dog bit the man

there are corresponding sentences of the form:

14a The man was bitten by the dog

14b What the dog did to the man was bite him
14c What happened to the man was that the dog bit him

Such sentences are impossible with copular verbs:

15a He became a soldier

15b *A soldier was become by him
15c *What happened to the soldier was that the man
became him

Other transitive verbs include kill, cook, find.

2 The grammar discussed in the preceding chapter, and elaborated at the end
of this chapter, can be used to give a formal definition to the terms 'subject’ and
‘object’. The subject NP is that NP which is immediately dominated by S; the
object is that NP immediately dominated by VP. In these terms the NP
dominated by Pred which occurs after the copula verb be discussed in 1-7 is not
an object NP - in traditional terms it is referred to as a ‘complement’. Notions
such as subject, object, complement, etc. are discussed in more detail on pages
330ff.i
 Some verb classes in English 53

‘Di-transitive’ verbs

These occur in the environment:

16 _NP + NP
17 John gave Mary the book

These verbs are called ‘ditransitive’ since they are typically

followed by two NPs, and most verbs in this class must be
followed by two NPs.3 Characteristically, sentences of the form
illustrated in 17 have corresponding sentences where the two
NPs are reversed in order and a preposition, typically to, is
introduced:

18a John gave Mary the book

18b John gave the book to Mary

Like sentences containing transitive verbs, sentences like those

in 18a,b correspond to sentences like:

19a Mary was given the book by John

19b The book was given to Mary by John

Either object NP can become the subject of the corresponding

‘passive’ sentence. Other di-transitive verbs include PRESENT,
TELL, TAKE.

‘Intransitive locative’ verbs

These occur in the environment:

20 _PP
21a The lamp stood on the table
21b The gun leant against the wall

Most verbs of this class require to be followed by a PP:

22a *The lamp stood

22b *The gun leant

3 In terms of our previous definition of object, such verbs may be said to

have two objects - hence the description ‘di-transitive’. The two objects are
usually distinguished as the ‘indirect object’, the first of the two NPs in the
structure shown, and the ‘direct object’, the second of the two NPs. (For further
discussion see chapter 19.) In sentences with only one object, like 11, this is
called the direct object.
54 Part one: Constituent structure

The PP typically indicates a location, hence the name for this

class of verbs. Other intransitive locative verbs are HANG, SIT,
SLUMP.

‘Transitive locative’ verbs

These occur in the environment:

23 _NP + PP
24 John stood the lamp on the table
25 Mary leant the gun against the wall

Frequently, as in the examples given, these verbs correspond to

the verbs found as intransitive locative verbs, except for the
object NP immediately following the verb. These verbs
resemble transitive verbs in that there are corresponding
sentences like:

26 The gun was leant against the wall by Mary

27 What happened to the gun was that Mary leant it
against the wall

and so on, in a manner analogous to transitive verbs (as 17).

They differ from simple transitive verbs in that, typically, they
require to be followed by a PP as well as an NP. We find no
sentences:

28 *John stood the lamp

etc. These verbs differ from di-transitive verbs in that they

cannot be followed by two NPs:

29 John stood the lamp on the table

30 *John stood the table the lamp.

Other transitive locative verbs are PUT, lay, hang.

We have looked at six different verb classes. Each can be
distinguished in terms of the environments in which the
members of the class can occur, and in terms of the sentences
that can be related to a sentence containing a verb from a
particular class. Some of these have been illustrated; others the
reader is left to discover for himself.
How might we account for these different verb classes in a
formal grammar? One way is to label each of the verb classes,
producing a constituent structure grammar of the following sort:
 Some verb classes in English 55
31 S —» NP + VP
V + Pred
cop
V.
(V = verb copulative;
V +NP ' cop r
VP-^ 1 V. = verb intransitive, and
V, +NP + NP l
dt
so on)
V., + PP
ll
V + NP + PP

together with a set of lexical rules of the form:

32 V {seem, turn, appear, ...},

cop 1 ’ J’
V. -> {die, talk, work, smile, ...}.

and so on.
An approach of this sort poses difficulties. It leads to the
establishment of a large number of verb classes: we already
have six, and with little difficulty can establish more. This makes
the second rule in our grammar cumbersome. Furthermore, we
find that a considerable amount of ‘cross-classification’ is
involved: i.e. many verbs occur in more than one class -
examples we have seen are the verbs STAND and LEAN. An
alternative approach is to have a lexicon listing the class
membership of each verb, and a lexical insertion rule as
discussed in the preceding chapter. This approach leads to the
listing in the lexicon of the different class memberships of each
verb, in the form:

33 STAND V.„ V ,
ll tl

and so on, but does not solve the clumsiness of the constituent
structure rules.
A different approach lists in the lexicon the environments in
which a particular verb can occur, without actually seeking to
name each individual verb class. Thus we list STAND as:

34 STAND V; _ (NP) PP

In this representation V indicates that STAND is a verb. We call

56 Part one: Constituent structure
such characterizations (and others like N, Adj, Prep, etc.)
‘inherent subcategorization’ since they describe the form class
to which a particular item belongs, which seems to be an
inherent property of the item itself. The representation_(NP)
PP indicates that stand must occur with a following PP, and
may, optionally, occur with an immediately following NP. This
is equivalent to saying that STAND is either an intransitive
locative, or a transitive locative verb, since these two
environments define these verb classes. We call such
characterizations ‘strict subcategorization’: this always refers
only to the syntactic environment relevant to the sub¬
categorization of the item in question which means other
constituents of the VP. This approach simplifies our grammar:
instead of listing individually all the different expansions of VP
we can present the rules as follows:

35 S —» NP + VP

Recall that, by our conventions, items curly-bracketed together

are alternatives; items in ordinary parentheses are optional. The
VP rule here allows us to produce the strings: V + Pred; V;
V + NP, V + PP, V + NP + NP; V + NP + PP - this is just
the set of environments described above. The strict sub¬
categorization frames for a given verb include just those
categories that are introduced as sisters of V in the NP. Note
that only the sisters of V are necessary for strict
subcategorization.
We shall also have to amend the lexical insertion rule given
on page 43. Our rule will now be:

36 Lexical insertion rule

For any terminal symbol of the constituent
structure rules, select from the lexicon a
word that is described as being a member of
the class named by the terminal symbol in
 Some verb classes in English 57

question and attach this word as a daughter

of the relevant symbol. If the item selected
is characterized by a strict subcategorization
frame, then the environment into which the
item is inserted must not conflict with the
strict subcategorization frame.

Technical terms

copula linking verb

complement object
copular verb predicate
di-transitive verb strict subcategorization
inherent subcategorization strict subcategorization frame
intransitive verb subject
indirect object transitive verb
intransitive locative verb transitive locative verb
locative verb

Exercises

1
The text of this chapter distinguished six verb classes. To which
of these classes do the following verbs belong? Note that some
belong to more than one class. For each verb draw up a lexical
entry like that shown in the text for STAND:

SEEM LIE LIVE ANTAGONIZE HANG

SPREAD SING LOCK FIND CRY
APPEAR SHOOT EAT GROW DIE
SMELL GROWL WRITE GIVE PASS
PUZZLE REMAIN ARRIVE BECOME SEND

We noted in the text that sentences like Moggy is a cat and Kate
bought a cat are, at one level of structure, apparently similar -
NP V NP - but that we need to distinguish between copular
verbs (like be) and transitive verbs (like buy). Summarize the
reasons for making this distinction.
58 Part one: Constituent structure

Now consider the following sentences:

1 The girls made John happy (NP V NP Adj)

2 The committee elected John Chairman (NP V NP NP)
3 The committee gave John a cheque (NP V NP NP)

Using the same sorts of criteria that you used to distinguish

between copular and transitive verbs, distinguish between the
three apparently similar sentence types (in particular think of
syntactically related sentences - The committee elected John as
Chairman, etc.). To which of the verb classes you establish do
the following belong:

GIVE PRESENT VOTE CONSIDER OFFER

THINK MAKE SHOW CHOOSE IMAGINE

Many of the verbs that we have looked at so far can take NP

objects, where the NP is expanded to dominate a noun (like
The hunters shot THE LIONS). Many verbs can also have
sentences as objects:

I believe that the world is flat

I think that Edinburgh is a lovely city

Suppose that such sentences, which we will examine in more

detail later, have a structure like this:

I believe that the world is flat

Note that in the analysis proposed the sentential object of the

verb is treated as an NP which has been expanded as that + S.
 Some verb classes in English 59

(This analysis will be defended in more detail later on.) For our
present purposes we describe the relevant environment as:

_ S

and we can add this environment to the other environments

discussed above. We note one further similar environment,
illustrated by:

I told Mary that I loved her (NP V NP S)

Given these two new environments, write lexical entries for the
following verbs. For example:

TELL V,_NPNP (I told Mary a bedtime story)

V,_NP S (I told Mary that I loved her)

ASSUME PERSUADE ARGUE FORETELL PREDICT

SAY TELL EXPECT SUPPOSE ASSURE
DENY HEAR AGREE PROMISE SUPPOSE
5 Formal grammars

Grammars of the sort discussed so far are referred to as ‘formal’

grammars. Formal here can be taken in two senses.
In one sense the grammar is formal in that it is concerned
with the distribution of forms, but not with the sense of
individual forms, nor with the meaning of whole sentences nor
whether any appropriate context of use can be found for the
sentence. It is concerned only with the distribution of forms.
Form classes are established in terms of their distribution -
items with the same distribution are placed in the same form
class. In this sense ‘formal’ may be opposed to ‘notional’. A
notional description would include in the same class items held
to have a common element of meaning, but not necessarily with
the same distribution.
The distinction between a formal and a notional approach can
be illustrated by considering the following data from the African
language Akan, from central and southern Ghana:

la Kofi ware Kofi is tall

lb Kofi reware Kofi is getting tall
lc Kofi aware Kofi has grown tall
2a Kookoo no bere the cocoa is ripe/red
2b Kookoo no rebere the cocoa is ripening
2c Kookoo no abere the cocoa has got ripe
3a Kofi kasa Kofi speaks
3b Kofi rekasa Kofi is speaking
3c Kofi akasa Kofi has spoken
4a Kofi ye kesee Kofi is big
4b Kofi reye kesee Kofi is getting big
4c Kofi aye kesee Kofi has become big

Sentence sets 1 to 3 contain the verbs ware ‘to be or become

tali’; bere ‘to be or become ripe or red’; kasa ‘to speak’. Set
 Formal grammars 61
4 has a verb Ye ‘to be or become’ followed by an adjective
KeSEe ‘big’. We can identify the verbs in formal terms as those
forms that can be inflicted by various verbal affixes: the a
sentences contain the simple verb stem; the b sentences the
prefix re-\ the c sentences the prefix a-} By contrast adjectives,
like KeSEe, are invariant, so there are no sentences like:
5 *Kofi kesee
*Kofi rekesee
*Kofi akesee

Furthermore adjectives must, as in English, be preceded by a

copular verb, in this case the verb Ye ‘to be or become’. Verbs
do not co-occur with the copula, so we do not find:
6 *Kofi ye ware
*Kookoo no ye bere
In formal terms items like WARE ‘to be tall’ or bere ‘to be
red’ are clearly verbs, just as kasa ‘to speak’ is a verb.
Grouping WARE ‘to be tall’ and KeSEe ‘big’ into the same class
falsifies the facts of the language. Both are ‘descriptive’ words,
but their grammatical behaviour is different, and the grammar
should be based on grammatical considerations (like the ability
to bear verbal affixes) rather than notional considerations (like
the fact that they are descriptive words).
We can make the same point by considering the following
English sentences:
7 John talks; John is talking
8 The fruit ripens; The fruit is ripening
9 The fruit is ripe; The fruit is getting ripe

TALK and RIPEN in 7 and 8 are clearly verbs (they occur in the
verbal forms ripens and IS ripen ING, etc.); equally clearly RIPE
in 9 is an adjective (it occurs with the copula, BE, and in
constructions with the verb GET). We cannot use the verb
RIPEN with the syntax of an adjective, nor the adjective RIPE
with the syntax of a verb, as can be seen by comparing 8 and 9
with the ungrammatical:
10 *The fruit ripes; *The fruit is riping
11 *The fruit is ripen; *The fruit is getting ripen

1 This is not the whole story since tonal distinctions are also involved, and
these are not marked. However this account is adequate for our purposes.
62 Part one: Constituent structure
The fact that RIPE and RIPEN are lexically related, though
true, is irrelevant: RIPE is an adjective and RIPEN a verb. In
attempting to describe the grammar of a language, formal
criteria are to be preferred to notional criteria.
Members of a formally established form class may have a
common notional element; indeed this is often the case and is
no surprise if we hold that language is a system for communi¬
cating meanings. On the other hand the principles behind the
establishment of formal and notional classes are quite different,
and formal criteria are in general preferable.
The second sense of formal implies that the grammar should be
formalized, that is, presented in terms of a set of formal rules,
like those of the previous chapter, explicitly and economically. It
must be clear just what data is accounted for and how it is being
accounted for, just what assumptions are made about the nature
of the language, and just what claims about the structure of the
language are presented. In this sense formal is opposed to informal.
Formal in the first sense does not necessarily exclude
considerations of meaning: this would be impossible and un¬
desirable. Ostentatious attempts to expel considerations of
meaning have usually allowed meaning to creep surreptitiously
back under some other guise. We can see how meaning is
smuggled in if we consider the sort of problem sometimes
presented for analysis to beginning linguistics students. Such
problems often present data from a little-known language which
can sometimes be analysed without knowing the meaning of the
sentences, rather like a puzzle from a Sunday newspaper. Here
is an example from Scots Gaelic (we have omitted glosses to
demonstrate a formal analysis):
12 Bha an cu dubh
13 Bha an cat ban
14 Bha Calum mor
15 Bha an cu sgith
16 Bha Calum sgith
17 Bha Mairi beag
18 Bha an gille mor
19 Bha an cu beag
20 Bha Mairi ban
21 Bha an gille beag
22 Bha an cat mor
23 Bha Mairi beag
 Formal grammars 63
We are provided with the additional information that these
strings are all sentences, and that no other word order is
possible for these sentences — the following, for example, are
impossible:

24 *An cat dubh bha

25 *Bha cat an dubh
26 *Bha dubh an cat

etc. We are also told that strings like:

27 *Bha an Calum sgith

28 *Bha cat dubh

are not found.

It is possible to write a grammar for this data without
knowing what the sentences mean, using only the techniques of
expansion and substitution. Thus, we allocate dubh, sgith, and
beag to the same form class since they are all mutually
substitutable following bha an cu (12, 15, 19). mdr and beag are
mutually substitutable in 18 and 21, and beag and ban are
substitutable in 17 and 20. Since these classes have overlapping
membership we assume that all these items are members of the
same form class, which we give the arbitrary label A. We could
test this supposition, if a Gaelic speaker were handy, by
enquiring whether a string such as:

29 Bha an cu mor

which we do not find in our data, but which this analysis

predicts, is well formed. (It is.)
Next we note that our data includes a class of words {cii, gille,
cat) that only occur after an, and we see from the
ungrammaticality of 28 that at least one member of this class
cannot appear without a preceding an: we assume this is typical
of the class, which we give the arbitrary label B.
There is a class of words in our data never preceded by an:
Calum and Mairi, and 27 shows that Calum at least may not be
preceded by an. We give this class the arbitrary label C. This
leaves only bha and an unaccounted for; the evidence suggests
these must be considered separate one-member classes, D and
E. So much for simple substitution. We can summarize our
findings in substitution tables, one for each string type (Figures
21 and 22):
64 Part one: Constituent structure

D E B A

bha an cu dubh
cat ban
gille sgith
beag
mor

Figure 21

D C A

Calum dubh
bha
Mairi ban
sgith
beag
mor

Figure 22

At this point we note that strings of the form E + B appear

substitutable for members of the class C. This suggests a new
form class, F, which can be expanded as either E + B or C. This
too we can represent in the substitution table (Figure 23):

D F A

E B

cu dubh
bha an cat ban
gille mor
beag
C sgith
Calum
Mairi

Figure 23
 Formal grammars 65
Or we can write a little grammar for our data:

30 S —» D + F + A
E + B
F
C

Lexical rules:
D —> {bha}
E -> {an}
etc.

At this point we must give up our pretence. If we had glossed

the sentences - glosses for a few of the sentences are shown
below, and reference to the lexicon below will yield glosses for
the others - the reader would quickly have identified the class
arbitrarily labelled A as a class of Adjectives, B as a class of
Nouns, C as a class of Proper Nouns, D as a copular verb and E
as an article. Using these more familiar class names we could
write the grammar:

31 Constituent structure rules

S -» VCop + NP + Adj Copular verb/
Noun Phrase/
Adjective

Article/Noun
Proper Noun

Lexicon:
an Art ‘the’
bha VCop ‘was’
cat N ‘cat’
ban Adj ‘white; fair haired (of people)’
etc., etc.

Exercises like this with or without glosses are useful. They

encourage a formal (rather than a notional) approach to
description; they illustrate the principles of distributional
analysis; and they can be used to practise writing formal
grammars. They also draw the student’s attention to the ways in
which languages can differ with respect to such features as
66 Part one: Constituent structure

constituent order within the sentence or class membership: note

the difference in constituent order between English (subject +
verb + complement) and Gaelic (verb + subject +
complement). In the Akan data at the beginning of the chapter
the class Verb in Akan includes items that translate as
Adjectives in English. The exercises also illustrate, albeit in an
oversimplified way, an approach to language analysis: sentences
with similar structure are gathered together so that they may be
compared directly and the likenesses and differences between
them clearly observed.
Such exercises are, however, artificial. The data has been
carefully selected and arranged: and it is here that meaning has
crept surreptitiously in. A random sample of data from a native
speaker of a language speaking naturally would not look
anything like the data we have examined; it would be impossible
to make a formal grammar on the basis of a dozen or so
utterances collected at random.

Lexicon for the Gaelic data:

an Art ‘the’ cu N ‘dog’
ban Adj ‘white’ dubh Adj ‘black’
beag Adj ‘small’ gille N ‘boy’
bha Vcop ‘was’ Main PN ‘Maire’
Calum PN ‘Calum’ mor Adj ‘big’
cat N ‘cat’ sgith Adj ‘tired’

The sentences 12-23 translate as

The dog was black (12)

The cat was white (13)
Calum was big (14)
etc.

Technical terms

formal (of forms) vs. notional

formal (formalized) vs. informal

Exercises

The data in this and the following exercise are from Scottish
Gaelic.
 Formal grammars 67
1

(a) Group the words into classes: use the class names
Art(icle); N(oun); PN (Proper Noun); VT (Transitive verb);
VI (Intransitive verb); Pro(noun).

(b) Find an appropriate translation meaning for each word.

(c) Write a grammar for these data and incorporate it into the
grammar given in the text.

1 Chunnaic mi an cu I saw the dog

2 Bhuail e an gille He struck the boy
3 Ghlac Calum breac Calum caught a trout
4 Reic Calum an cu Calum sold the dog
5 Sgriobh mi litir I wrote a letter
6 Ghlac Calum an cu Calum caught the dog
7 Bhuail Tearlach an gille Charlie struck the boy
8 Chunnaic an cii Calum The dog saw Calum
9 Bhasaich Tearlach Charlie died
10 Fhuair Tearlach cu Charlie got a dog
11 Sheinn mi I sang
12 Bhasaich an cu The dog died
13 Chunnaic Calum cu Calum saw a dog

In the sentences below, there is an additional constituent not

present in the corresponding sentences in the previous exercise.
Let us refer to this as a Prepositional Phrase (PP). What are the
constituents of the PP? Use the class name Prep(osition) in
addition to the class names already available to you.
Enlarge the grammar you wrote for the previous exercise to
accommodate this new data.

1 Bhuail Tearlach an gille le Charlie struck the boy with

bata a stick
2 Ghlac Calum breac anns Calum caught a trout in
an linne seo this pool
3 Chunnaic mi an cii aig an I saw a dog at the door
dorus
4 Sgriobh mi litir gu Calum I wrote a letter to Calum
5 Bhasaich Tearlach anns an Charlie died in Italy
Eadailt
68 Part one: Constituent structure

3
The following data are from Kwahu, an Akan language from
Ghana:

1 okyerekyerefoo no rekasa The teacher is speaking

2 abofora no regoro The child is playing
3 Amma regoro Amma is playing
4 Kofi rehwe okyerekyerefoo Kofi is looking at the
no teacher
5 okyerekyerefoo no reboa The teacher is helping
Amma Amma
6 okyerekyerefoo no reboro The teacher is beating
abofora no the child
7 abofora no resu The child is crying
8 Amma reboa abofora no Amma is helping the child

(a) Group the words into classes: use the class names - Article
(Art), Noun (N), Proper Noun (PN), Verb Transitive (VT),
Verb Intransitive (VI).
(b) Find an appropriate translation meaning for each word
based on the English gloss. Since we are assuming that words
are the smallest constituents of interest to us, you will have to
identify a word like regoro as a member of the class VI and
give it a gloss like ‘is playing’.
(c) There are two types of Noun Phrase (NP) shown in the
data. What are the constituents of these NPs?
(d) Write a grammar for this data.
(e) Using your grammar, draw tree diagrams to represent the
grammatical structure of sentences 1 and 6.
6 Optional and obligatory
constituents

We have noted that the subject NP is non-distinctive for

purposes of verb sub-classification (page 50). Since all sentences
require a subject NP, only what follows the verb can be distinc¬
tive. Let us now consider some other constituents that are non-
distinctive for the purpose of verb subclassification. Consider
these sentences:

1 The women wept in the bathroom

2 The dog bit the man in the bathroom
3 John gave Mary the book in the bathroom
4 John stood on the table in the bathroom
5 John stood the gun against the wall in the bathroom

We can analyse each of these sentences as having a constituent

in the bathroom in construction with one of our example
sentences from Chapter 4.1 Traditionally these constituents are

1 Some of these sentences are ambiguous. For example, 2 could be analysed as:
i The dog - bit - the man - in the bathroom
or as: ii The dog - bit - the man in the bathroom

In sentence i the constituent in the bathroom indicates where the man was
bitten, and can be related to sentences like:
iii It was in the bathroom that the dog bit the man
Where the dog bit the man was in the bathroom
Where did the dog bite the man? In the bathroom

In sentence ii the constituent the man in the bathroom is an NP. This sentence
is related to other sentences like:
iv It was the man in the bathroom that the dog bit
Who did the dog bite? The man in the bathroom
We are immediately concerned with the first of these analyses, i. (We return to
the other analysis later.) We also observe that constituents like in the bathroom
do not typically occur in sentences with copular verbs:

v *John is tall in the bathroom

This is why such sentences are not included in 1-5.
70 Part one: Constituent structure

called ‘adverbs’, and specifically ‘adverbs of place’, since they

indicate where the action took place. There are a number of
things we should observe about these constituents in deciding
the constituent structure of 1-5. Firstly, in each case in the
bathroom is an optional constituent: without it, an acceptable
sentence remains. By contrast we cannot leave out the
constituent immediately following the verb. We can have:

6 The dog bit the man

but not

7 *The dog bit in the bathroom

Secondly the constituent in the bathroom is not criterial for

verb subclassification: it can occur with all the verb classes
illustrated, so cannot be criterial. Finally, constituents like in the
bathroom can occur in a distinct sentence, for example:

8 The women wept. This happened in the bathroom

9 John stood the gun against the wall. This happened in
the bathroom

8, 9 roughly paraphrase the sentences 1, 5. Note that we cannot

find sentences resembling 8, 9 which correspond to any of the
other constituents in 1-5:

10 *The dog bit. This happened to the man

This last characteristic of constituents like in the bathroom is

particularly valuable when we contrast them with superficially
similar constituents like against the wall and on the table in
sentences 4 and 5. The following are acceptable:

11 John stood on the table. This happened in the

bathroom
12 John stood the gun against the wall. This happened in
the bathroom

but the following are nonsense:

13 *John stood. This happened on the table

14 *John stood the gun. This happened against the wall

In Chapter 4 we observed that the constituents criterial for strict

subclassification are those constituents that are sisters of V, and
daughters of VP with V. This suggests that we should consider
 Optional and obligatory constituents 71
our adverbial constituents as constituents of some node other
than VP. We will treat them as sisters of VP, and daughters,
along with VP, of S, as shown in Figure 24.

John stood the gun against the wall in the bathroom

Figure 24

This analysis is captured in the following grammar:

15 S —> NP + VP (+ Adv)
VP^ V(NP) (jPPjj
Adv —> PP •
PP-^ Prep + NP
Art + N I
NP
PN J

This grammar clearly distinguishes those PP that are within the

VP - and are hence constituents criterial for strict
subclassification, as with stand - from those PP not in the
VP. These latter are adverbs and not criterial for
subclassification.
Traditionally items like adverbs are called ‘modifiers’;
traditional grammars indeed often refer to ‘modifying adverbs’.
(We discuss modification in more detail on pages 254-9.) In
informal notional terms, a modifier is a constituent that restricts
the possible range of reference of some other constituent. In the
sentence the woman wept in the bathroom, the adverbial
constituent in the bathroom modifies the ‘nuclear’ sentence the
woman wept by specifying the particular place in which the
72 Part one: Constituent structure

action described by the nucleus took place. We use the term

‘scope’ to refer to the constituents which the modifier modifies:
thus the scope of the modifier in the bathroom is the nuclear
sentence the woman wept. Syntactically, modifiers are optional
constituents, and this is shown in the first rule in 15 by enclosing
the adverb constituent in brackets. The scope of the modifier
can, to some degree, be shown by the rule that introduces it. In
the case of 15 the adverb is introduced as an optional
constituent of the sentence, so its scope is the other constituents
of the sentence, i.e. the nuclear sentence. Adverbial expressions
like in the bathroom are thus ‘sentence adverbs’ or ‘sentence
modifiers’.
The previous paragraph referred to the constituent the woman
wept as ‘nuclear’. We can draw a useful distinction between
‘nuclear’ and ‘non-nuclear’ constituents. Nuclear constituents of
the sentence are NP + VP and all that is immediately
dominated by VP. Other sentence constituents, like Adv, are
non-nuclear. Nuclear constituents are either obligatory for the
sentence to be accepted as grammatical, or criterial for verb
subclassification. Non-nuclear constituents are optional, and,
typically, modifiers. Frequently, non-nuclear constituents can be
placed in a separate sentence, as illustrated by 8 and 9. In these
terms PPs deriving from an expansion of VP are nuclear, PPs
deriving from Adv are non-nuclear. Nuclear sentences are the
most ‘basic’, simplest, sentence types in the language. Defining
nuclear sentences is not, however, entirely straightforward.
The distinction between nuclear and non-nuclear constituents
is not a simple matter of optionality. We saw, for instance, in
the previous chapter that the NP that immediately follows a
verb like STAND is not obligatory: both He stood the lamp on
the table and The lamp stood on the table are grammatical. It is,
however, nuclear, as it is part of the necessary strict
subcategorization frame. Other verb classes with optional but
nuclear constituents include ‘verbs of motion’ like RUN, WALK:

16 John is running
17 John is running to school

where the to school constituent is nuclear but not obligatory: 17

cannot be paraphrased by

18 *John is running. This happened to school.

 Optional and obligatory constituents 73
Another similar case involves verbs like read and write
which take an optional object NP:

19 John is reading
20 John is reading a book

The constituent a book can hardly occur in a distinct sentence

on the model of 8 or 9. Furthermore it makes no sense to say
that a book ‘modifies’ John is reading in a comparable fashion
to the way in the bathroom modifies the woman wept in the
sentence The woman wept in the bathroom.
Let us now look further at the expansion of the Adv node
shown in the third rule of the grammar in 15. The constituents
we have been concerned with are traditionally called ‘adverbs of
place’: are there other types of constituent that may be called
adverbs of place? Are there other types of adverb in addition
to adverbs of place? The answer to both questions is yes.
First, consider another type of adverb of place:

21 The dog bit the man upstairs

(21, like 2, is ambiguous as between the bracketing The dog -

bit - the man - upstairs and The dog - bit - the man upstairs.
We are concerned with the former analysis.) In 21 upstairs has
the same distribution as the PP in the bathroom. This suggests
they belong to the same form class, which we call Adv
place
This leads to the constituent structure rule:

22

where PP expands into a prepositional phrase, and Place adverb

dominates such items as upstairs, downstairs, here, there, etc.
Sentences with Adv constituents correspond to
place
interrogative sentences of the form:
23a Where did the dog bite the man?
23b Upstairs/In the bathroom, etc.

Consider two other types of adverb, adverbs of time and

adverbs of manner. Adverbs of time, as the name suggests,
locate events, etc. in time:
24a The dog bit the man yesterday
24b John lived in London for twenty years
74 Part one: Constituent structure
Once again there are prepositional phrases (for twenty years)
and simple time adverbs (yesterday). Sentences with adverbs of
time correspond to interrogative sentences with when:

25 When did the dog bite the man?

In independent sentences, adverbs of time most naturally follow

rather than precede adverbs of place (though both orders may
occur):

26a The dog bit the man upstairs yesterday

26b The dog bit the man yesterday upstairs

Whereas adverbs of time occur freely in sentence-initial

position, adverbs of place there are often clumsier:

27a Yesterday the dog bit the man

27b Upstairs the dog bit the man

Adverbs of manner, as their name suggests, indicate the

manner in which the event described by the verb is carried out:

28 The dog bit the man viciously

Once again there are manner adverbs (viciously) and prep¬

ositional phrases:

29 The man beat the dog with apparent enjoyment

Many manner adverbs are formed by the suffixation of Ay to the

corresponding adjective form (happy :happily; vicious -.viciously),
and there is then a typical paraphrase in an [Adj] manner:

30a The man beat the dog viciously

30b The man beat the dog in a vicious manner

Manner adverbs typically follow the VP immediately and

precede other adverbs:

31a The man beat the dog viciously in the garden

yesterday
31b ?The man beat the dog yesterday in the garden
viciously

Manner adverbs, unlike other adverbs, may also occur within

the verb constituent, but occur less readily in sentence initial
position.
 Optional and obligatory constituents 75

32a The man viciously beat the dog

32b The man is viciously beating the dog
32c *The man in the garden beats the dog
32d Viciously the man beat the dog

Finally manner adverbs correspond to how in interrogative

sentences:

33a How did the man beat the dog?

33b Viciously/With apparent enjoyment, etc.

Given this new data, we amend the first rule of 15 to:

34 S —» NP + VP (+Adv
,, . , , v manner (+Advplace>
(+Adv . )
v time7

together with appropriate rules, on the analogy of 22, to

accommodate the expansion of the adverb nodes. This leaves us
with the problem of describing the ‘alternative’ placements of
adverbs of manner and time (either within the verbal constituent
or the sentence initially), which we return to in a later chapter.

Let us now turn to some optional constituents of the NP. The

footnote on page 69 observed that the constituent in the bath¬
room could be understood as an adverb of place, which we
have discussed, or as a constituent of the NP. If we understand
it as an NP constituent, the analysis is that of Figure 25.

NP

Art N PP

the dog bit the man in the bathroom

Figure 25
76 Part one: Constituent structure
Under this analysis we consider the constituent the man in the
bathroom to be an NP since it has the same distribution as an
NP:

35a The man in the bathroom was bitten by the dog

35b Who did the dog bite? The man in the bathroom

etc. It corresponds to sentences of the form:

36 The dog bit the man who was in the bathroom

where the constituent the man who was in the bathroom is also
an NP. We defer consideration of such NPs, involving what are
traditionally called ‘relative clauses’, to pages 137-44.
We can account for the structure of NPs like the man in the
bathroom by proposing the constituent structure rules:

37a NP-h> Art + N (+PP)

37b PP^Prep + NP

Note that in 37a PP is an optional constituent. We can, in the

terms of our previous discussion, refer to it as modifying the
constituents Art + N, and we see that the scope of the
modification is defined by the rule. Such PPs are ‘NP
modifiers’. Just as adverb modifiers could be ‘separated out’ into
a separate sentence, so too NP modifiers can be related to
structures that have a relation to separate sentences -
in this case the relative clause.
In using the rules 37a, 37b to derive a sentence, we need to
‘cycle’ through them: we apply 37a to introduce a PP
constituent, and then, having used rule 37b to expand the PP,
return to rule 37a to expand the NP introduced by 37b. This
second application of 37a may produce another PP, which in
turn can produce another NP, etc. This property of ‘cycling’
through rules over and over again is called recursion (it is
discussed in more detail on pages 134ff.). Recursive rules are
necessary for structures like that in Figure 26.
Now consider the NPs in sentences like:

38 The foolish boy stood on the burning deck

which we can analyse as in Figure 27.

Adjectives can be introduced into NPs by a rule like:

39 NP —> Art (+Adj) + N

 Optional and obligatory constituents 77

NP

Art N PP

the house by the bridge over the river beyond the town

Figure 26

PP

stood on the burning deck

Figure 27
78 Part one: Constituent structure
The rule correctly indicates the optional nature of adjectives,
captures the fact that adjectives ‘modify’ nouns (reflected in a
traditional characterization as ‘modifying adjectives’) and shows
that the scope of the modification is the NP. But as it stands it
does not allow us to account for NPs containing more than one
adjective:

40a A big red car

40b A large old comfortable settee
40c A large red ripe juicy plum

and so on. We therefore modify the rule by introducing a new

notation:

41 NP -> Art (+Adj)* + N

where the symbol * after a constituent means we can choose as

many of this constituent as we wish - one in 38, two in 40a, and
so on. The rule will lead to structures like that in Figure 28.

NP

Art Adj Adj Adj N

1 1 1
1 1 1
Figure 28 a large old comfortable settee

The implication of such structures is that all the adjectives

modify the noun equally, and we see that an NP like 40b can be
paraphrased by

42 A settee which is large and [which is] old and [which

is] comfortable

This rule is satisfactory for many similar adjective constructions,

but obviously not for all. For example,

43 A dark red car

can hardly be paraphrased as

44 *A car which is dark and [which is] red (compare 42)

but rather by

45 A car which is dark red

 Optional and obligatory constituents 79
Similarly

46 A red sports car

can be paraphrased by neither of

47a *A car which is red and [which is] sports {compare 42)
47b *A car which is sports red (compare 46)

In 43 dark modifies red, and dark red as a whole modifies car;

in 46 sports modifies car, and red modifies sports car. This
suggests the analyses in Figures 29 and 30 for 43 and 46
respectively.

NP NP

Art Adj N

N N

a dark red car a red sports car

Figure 29 Figure 30

In Figure 29 dark has been categorized as an intensifier (Int).

This class includes other items like light, bright, brilliant, deep,
very, and so on, all of which may modify colour adjectives. In
Figure 29 we have categorized sports as a noun, and the string
sports car also as a noun. We do this because sports used as a
separate item is a noun (What sports do you play?, The sports
have been rained off, etc.) and because sports car is a
‘compound noun’ like detective story, water mill, billiard ball,
etc., all of which typically have separate dictionary entries. (We
discuss compound nouns further on pages 226-7.)
To accommodate structures like those in Figures 29 and 30
we need rules like:

48a NP Art (+ Adj)* + N

48b (OP) Adj Int + Adj
48c (OP) N ^N + N

Another notational convention is introduced: the symbol (OP)

before a rule indicates that it is optional. In Figure 29 we use
80 Part one: Constituent structure
the rule 48b, but not 48c; in Figure 30 we use 48c but not 48b.
Rules like these generate large numbers of NPs, and go some
way towards capturing the necessary relationships, of
optionality, of modification and of the scope of modification,
that we would want our grammar to account for. Without much
ingenuity the reader will be able to think of other patterns of
modification within NPs that are not shown by the rules, and he
is invited to try and write rules to accommodate them.
Let us now summarize the rules of grammar we have so far
considered: they are set out in Figure 31:

NP + VP( +Advmanner)(+Advplace)(+Advlime)

Pred

(NP)

Pred

Manner adverb
Adv„
PP

Place adverb
Adv place
PP

Time adverb
Adv.;
PP

PP —* Prep + NP

NP Art(+Adj)* + N(+PP)

(OP)Adj —> Int + Adj

Figure 31 (OP)N —» N + N
 Optional and obligatory constituents 81
We conclude the chapter with a brief discussion of syntactic
ambiguity, since this matter has arisen several times in our
discussion, and seems a natural consequence of the
organizational principles of language. We have already noted
that a sentence like:

49 The dog bit the man in the bathroom

is ambiguous as between two interpretations (Figures 32

and 33):

Figure 32

Figure 33

(Internal constituent structure is omitted where irrelevant to the

argument.) Ambiguities of this sort may be called ambiguities of
bracketing, since the ambiguity rests on whether in the bathroom
82 Part one: Constituent structure
is a constituent of NP (Figure 33) or is an adverb of place
(Figure 32). Another type of ambiguity involves ambiguities
arising from a particular item belonging to more than one form
class, often called ambiguities of labelling: an example is shown
in Figures 34 and 35.

NP VP Adv
manner

Manner Adverb

V Pred

Adj

I
Figure 35 she looked hard

In Figure 34, hard is a member of the class of manner adverbs

(cf. she looked intently, she looked carefully, etc.) whereas in
Figure 35, it is a member of the class of predicate adjectives
(cf. she looked pretty, she looked careful, etc.). The dual class
membership of hard leads to the ambiguity.
Ambiguity has always exerted a special fascination: it can vary
from the simple sort that children at a certain age revel in
(Wanted: a grand piano by an old lady with carved legs) to the
complex ambiguity that poets have found particularly fruitful.
The grammarian’s interest in ambiguity lies in the descriptive
challenge it offers: if one particular string has two clear and
distinct meanings, the grammar should ascribe to it two different
analyses, compatible with its two meanings. The grammar we
have been dealing with is capable of doing this. This is not to
 Optional and oligatory constituents 83

claim that our grammar can deal with all kinds of ambiguity - it
can only cope with those that arise from labelling or bracketing
or both. Ambiguities such as those in:

50 The shooting of the hunters was disgraceful

I find it hard to sympathize with revolting students
Call me a taxi. OK, you’re a taxi

fall outside our present scope. We shall try to deal with such
ambiguities by machinery introduced in later chapters.

Technical terms
adverb (of place, time, manner) non-nuclear constituent
ambiguity (of labelling and nuclear constituent
of bracketing) obligatory constituent
basic sentence optional constituent
cycle recursion
modifier scope

Exercises

1 Ambiguity

The ambiguity in the phrase young men and women can be

shown by bracketing: young {men and women) - they are all
young - as against {young men) and women - only the men are
young. The difference can also be shown in tree diagrams:

NP NP

Adj

Adj N N

young young men and women

The following sentences are all ambiguous in one way or

another: show their ambiguity by drawing appropriate tree
diagrams.
84 Part one: Constituent structure

1 My brother teaches history in a school for young boys

and girls of wealthy parents
2 The bride and groom left early last night
3 He greeted the girl with a smile
4 We would like to attract more intelligent students
5 I bought an old French dictionary
6 They decided on the boat

Can you think of paraphrases that resolve the ambiguities one

way or the other? How commonly do you think that strings like
those illustrated are in fact ambiguous in the everyday use of
language?

2 Optional and obligatory constituents

The text suggests that obligatory constituents are always nuclear,

but that nuclear constituents are not always obligatory. It is further
suggested that nuclear constituents are important for the strict
subcategorization of verbs. In the sentences below the relevant
constituents have been separated by strokes. Which of them
would you consider to be obligatory and which non-obligatory,
and which nuclear and which non-nuclear? You should give
reasons to support your decisions: these may be based on the
unacceptability of some sentence if a constituent is omitted,
considerations that arise from strict subcategorization,
considerations relating to possible transformations of the
sentences in question and so on.

1 I/agree/with John
2 I/went/to the cinema/with John/yesterday
3 The fifteenth Music Festival/will be held/in the
Assembly Rooms/in May 1979
4 John/went/to Bristol/in 1972
5 Harriet/always/plays/with a rubber duck/in the bath
6 Mary/will write/a letter/to mother/for you
7 Harriet/is/in the bath
8 They/elected/Charlie/president
9 We/sent/a parcel/to Mary/for Christmas
10 We/pierced/the membrane/with a needle
11 They/thought/Charlie/foolish
7 Selection restrictions

In our normal use of language we do not expect all linguistic

forms freely to co-occur with all other linguistic forms. For
example, intransitive verbs do not co-occur with a following NP;
transitive verbs must co-occur with a following NP. The
specification of such co-occurrence restrictions we have called
‘strict subcategorization’ and have proposed that this
information should be included in the lexicon (pages 55-6).
We have also seen that items can be characterized as Nouns,
Verbs, Articles and so on, and have seen that this
characterization too can be used to restrict co-occurrence.
Articles co-occur with Nouns within the NP {the man, etc.), but
not with Verbs {*the frightens). We have proposed that this
information should be included in the lexicon by marking
individual items as N, V, Art and so on. We have called these
descriptions ‘inherent subcategorization’ (pages 55-7).
This chapter examines another type of co-occurence
restriction, ‘selection restrictions’, and considers whether these
too should be included in our description. The notion of
selection can be illustrated in the following:

la The man admired the picture

lb *The picture admired the man
2a The picture frightened the man
2b *The man frightened the picture

Typically ADMIRE requires an animate subject expression, but

will tolerate any object expression: in notional terms only
animates can ‘admire’ things, but they can admire other
animates, inanimates, abstract objects, and so on. This accounts
for the acceptability of la, and the oddness of lb. Conversely,
FRIGHTEN takes any subject expression, but must have an
animate object expression: in notional terms, one cannot
‘frighten’ things! We say that admire ‘selects’ an animate
86 Part one: Constituent structure

subject expression, and that FRIGHTEN ‘selects’ an animate

object expression. Statements such as these are statements of
selection restrictions.
Such statements rely on an intuition of what is the ‘normal’ or
‘literal’ use of language, and they beg an important question:
are sentences like:

3 Sincerity admires the young poet

‘unacceptable’, since ADMIRE here has an abstract subject

expression? Or are such sentences quite acceptable, but involve
a ‘figurative’ or ‘metaphorical’ use of language? We exclude the
possibility that Sincerity is a girl’s name (in which case the
sentence is perfectly well formed in terms of the selection
restriction stated above, and is non-figurative). We might find
such a sentence in some allegorical tale: in this case we
would speak of ‘personification’. Traditional grammars of
Indo-European and other languages sometimes recognize this
problem by giving a passing mention to various kinds of
figurative language. In doing so they usually assume that the
reader knows the difference between the literal and figurative
use of language and makes allowances accordingly: implying that
we can distinguish ‘normal’ states of affairs and a literal use of
language from ‘abnormal’ states of affairs and figurative
language. It is not at all clear just how this might be done, nor
even, if it can be done, whether it is to be considered a
specifically linguistic ability (like knowing that in English
adjectives precede the noun they modify) or whether it rests on
our knowledge of the world, or the way in which we choose to
conceptualize the world. A thorough account of such matters is
beyond our scope: the essential thing for the reader is to
recognize that deep questions lie behind any attempt to include
any information about selection restrictions in a grammar.
How might we take some account of such information in the
grammar? Suppose we can somehow identify a ‘literal’ use of
language and that we will restrict our description to this sort of
language. In the case of frighten and admire the
necessary information is that the subject of ADMIRE and the
object of frighten should be animate. We can capture this
generalization by subcategorizing nouns into classes of ‘animate’
and ‘inanimate’ nouns. We express this in terms of ‘features’
attached to the nouns. Thus MAN (and other similar nouns like
 Selection restrictions 87

BOY, GIRL, CAT, DOG, etc.) has the feature [+animate] and
PICTURE (and other nouns like table, chair,
TYPEWRITER, etc.) has the feature [-animate]. We can now
use this categorization to specify a selection restriction for
FRIGHTEN and ADMIRE in much the same way as we
specified their strict subcategorization. ADMIRE can be inserted
into a structure of the form:

4 ^ p(N[+animate]) NP

We call such a structure a ‘selection frame’, and take it to mean

that ADMIRE requires an animate subject expression, but can
have any object expression. 4, as stated, is obviously schematic
since it takes no account of other constituents that may be
relevant, such as other constituents of the subject NP, but it will
serve our purposes. Similarly a selection frame for FRIGHTEN
is:

5 NP p(N[+animate])

We add this specification to other specifications in the lexicon.

The lexical entries for admire and frighten are now:

6 ADMIRE V;_NP; Np(N[+animate]) —NP

FRIGHTEN V;_NP; NP~Np(N[+animate])

We have met the first two types of categorization: V indicates

that the items are verbs and _N P indicates that they are
transitive verbs. The third specification is the selection
restriction. Whereas strict subcategorization only includes other
constituents from the VP, selection restrictions must also
include information about the subject, and indeed may have to
include still other constituents of the sentence as a whole not
relevant to strict subcategorization. Specifying verbs as in 6
means we must also specify nouns with their appropriate
features. So our lexicon will include entries for nouns like:

7 man N; [+animate]
PICTURE N; [-animate].

The specifications in 6 and 7 can now be used to account for the

acceptability of la and 2a and the unacceptability of lb and 2b.
We will see exactly how this might be done in due course.
88 Part one: Constituent structure
Before we do so, let us consider some other noun features
that will be useful for a specification of selection restrictions.
We will examine the features [±human], [±male], [±concrete],
[±count] and [±common].
[±human] distinguishes between ‘human’ and ‘non-human’
animates. PRAY and MARRY, for example, co-occur with
human nouns, since animals do not ‘pray’ or ‘marry’ (in a literal
use of language). [±male] distinguishes between ‘male’ and
‘female’ animates: PREGNANT, for example, co-occurs with
female animates and not with males. [±concrete] distinguishes
between ‘concrete’ and ‘abstract’ nouns: fall, for example,
requires a concrete subject noun, since abstracts cannot ‘fall’.
[±count] distinguishes between ‘count’ and ‘mass’ nouns: count
nouns are typically those nouns that are used to refer to discrete
objects that can, in a literal senge, be counted (one chair, two
chairs, etc.); mass nouns are typically those nouns that are used
to refer to non-discrete objects, like OIL, BUTTER, CHEESE,
that cannot be counted (*one butter, *two butters, etc.).1 A verb
like FLOW typically co-occurs with a mass rather than a count
subject noun: we find The oil flowed over the floor but not *The
chair flowed over the floor, [icommon] distinguishes between
‘proper’ nouns, like JOHN, FIDO and ‘common’ nouns like
MAN, DOG.
Some of these features ‘cross classify’. Thus, JOHN and FIDO
are both [—common], and MAN and DOG are both
[Fcommon]; JOHN and man are both [+human] and FIDO
and DOG are both [—human]. Similarly BEAUTY is
[-concrete],[-count], truth is [-concrete],[+count], crowd
is [Tconcrete],[-count] and man is [+concrete],[+count].
Others of them are ‘hierarchical’ - the distinction [±male] is
contingent on the prior choice of [+animate], since sex is only
distinguished in animates.
We can use the features to specify various nouns:2

1 butter, and other mass nouns, can sometimes occur in sentences like Our
grocer stocks two butters, French and Danish. But note that in such sentences
butters has the sense ‘kind(s) of butter’. We discuss cases like this in more detail
on pages 242-4. For the present we assume that butter is a mass noun.

2 For any lexeme we have given only one specification. In some cases the
items in question have more than one possible use. Thus man in the sense
specified indicates ‘individual human male’; there is another sense of man as
‘mankind’: this use is [-count] and is not accounted for here. In the entry for
 Selection restrictions 89
8 MAN N; [+concrete],[+count],[+animate],
[+human],[+male],[+common]
WOMAN N; [+concrete],[+count],[+animate],
[+human],[—male],[+common]
JOHN N; [+concrete],[+count],[+animate],
[+human],[+male],[—common]
BEAUTY N; [ - concrete],[ - count],[+common],
[—animate]

OIL N; [+concrete],[-count],[+common],
[—animate]

DOG N; [+concrete],[+count],[+animate],
[-human],[± male]

We can now use these features to specify selection restrictions

for some of the verbs etc. discussed in the text:3

9 FLOW V; _ PP; Np(N[-count])_

p p(Prep N p(N[+concrete]))

PREGNANT Adj; Cop _ ; ^p(N[+female]) _

FALL V; _ #; Np(N[+concrete]) _

PRAY V; _ #; Np(N[+human]) _

MARRY V; — NP; Np(N[+human]) _

Np(N[+human])

dog we have specified [±male] to indicate the sense where dog is used to
refer to a canine irrespective of sex. Another sense of dog would need to be
specified as [+male] in contrast to bitch [—male].

3 As with the nouns, the specifications are illustrative rather than exhaustive. A
full specification of the entries would require to account for all possible uses.
Even the examples as they stand need modification in some cases. For example
the entry for marry should indicate that if the subject N is [+male] the object
noun should be [-male] and vice versa. We could use the specification
[a feature], meaning ‘either + or — ’ and the specification [—a feature] to
indicate the opposite polarity. So if [a male] has the value [ + male], then
[—a male] has the value [—male].
90 Part one: Constituent structure
Together these descriptions account for the acceptability or
otherwise of the following:
10a The man/oil/dog fell
10b * Beauty fell {non-concrete subject)
10c The woman/dog is pregnant
lOd *The man/John is pregnant {non-female subject)
lOe *The dog prayed {non-human subject)
lOf *The woman flowed over the beauty {count subject,
non-concrete N in PP)
The examples show the sort of judgements that formally stated
selection restrictions force on us, and bring us back to a
question mentioned at the beginning of the chapter - is
‘figurative language’ to be considered in some way ‘deviant’?
Perhaps what we recognize as ‘metaphorical language’ is
language that ‘breaks’ a selection restriction. But since so much
of our ‘ordinary’ use of language is ‘metaphorical’ we might
question the validity of such an account - it is far from clear
where, or how, to draw the boundary.
The discussion so far has assigned features to nouns on a
semantic basis, in terms of some characteristic of the object, etc.
that the noun in question is typically used to refer to. So, for
example, we have assigned to woman the features
[+animate],[+human],[—male], etc. since this lexeme is indeed
used to refer to ‘animate human females’. We may now observe
that all the features we have discussed have relevance for
grammatical processes of one sort or another. For example, the
pronoun typically used for [—human] nouns is it; the pronoun
for [+human] nouns is he or she depending on whether the
noun in question is [+male] or [—male]. Similarly we note that
[—common] nouns do not usually co-occur with an article: Fido
but not *the Fido, *a Fido, etc. There is a complex web of
co-occurrence restrictions between articles, and nouns
characterized as [±count]. Some of these are illustrated in 11 for
NPs with no article, and with the articles a, the and some, when
pronounced /sm/.

11 [+count] [—count]
singular the chair the butter
a chair *a butter
*sm chair sm butter
* chair butter
 Selection restrictions 91

[+count] [-count]
plural the chairs *the butters
*a chairs *a butters
sm chairs *sm butters
chairs *butters

The reader is invited to insert the NPs into a sentence frame

like:

12 I am going out to buy_

Grammatical sentences are only possible with the unstarred

forms: the starred forms produce ungrammatical sentences, as I
am going out to buy butter but not *1 am going out to buy chair,
etc. (we agreed, note 1, page 88, to disregard the reading ‘kind
of butter’ ‘packets of butter’ for the examples involving butter).
The other features can be shown to control similar grammatical
restrictions. For these reasons, features like those we have
discussed are known as ‘syntactic features’.
We can use syntactic features for two rather different
purposes: to specify certain grammatical restrictions (like those
illustrated in 11) and to control selection (as in 8). We defer for
the moment the question of whether it is sensible to use the
same set of features for two rather different purposes (which
may on occasion conflict).
Having demonstrated the relevance of syntactic features to
what seem to be grammatical operations, we might account for
them in the grammar itself rather than, as we have supposed
hitherto, in the lexicon. We now consider the consequences of
trying to incorporate subclassification into the grammar. At first
glance this seems quite straightforward. We could use rules like:

13 ♦
N animate
•

^inanimate
N
human

non-human

etc.; but this will not do since nouns are typically not just
animate or human, but both animate and human etc. Perhaps
rules like 14 might be appropriate:
92 Part one: Constituent structure

14 N
anim
N
N.
man

N anim,hum
• K
N .
anim
N • u
anim,non-hum

f^anim, hum, common 1

N
anim,hum
• u

g anim, hum, non-common J

etc. Or perhaps rules like:

15
hum
N
non-hum

N
hum,anim
N,
hum N
hum,non-anim

N
hum,anim,common
N,
hum,anim N
hum,anim,non-common

etc. Or indeed any other ordering of these categories. 14 and 15

are clearly absurd: there is no obvious reason to prefer one type
of subcategorization to another, and the number of rules must
be immense to accommodate all possible combinations. The
difficulty is that characterizations like animate : non-animate;
common : non-common cross-classify (as we have seen). Rewrite
rules cannot handle cross-classification in any satisfactory way.
Furthermore rules like these lead to trees like that shown in
Figure 36, which is ridiculous. From a syntactic point of view an
NP contains only one N, not a series of constituents N, Nr
anim
N . , etc.
anim,hum
A further problem arises if we use rules like this for purposes
of selection as well: the ‘direction’ in which such restrictions
apply. For example, in discussing frighten and ADMIRE we
said (page 85) that admire selects an animate subject, and
FRIGHTEN selects an animate object. The direction of the
selection is from verb to noun. This seems more natural than
 Selection restrictions 93

N,

N
1 anim, hum

Figure 36 N„anim,hum,common

to say that an animate object selects FRIGHTEN as its verb.

Unfortunately, if the subcategorization of nouns is to be
handled in the grammar, then selection is most easily handled in
terms of nouns selecting Verbs - the direction we would rather
avoid. To see that this is so consider the grammar and lexicon
in 16:

16 NP + VP
V+ NP
Art + N
fN anim
. V
N.
man

MAN N
anim
PICTURE N.
man
frighten V;_NP; N_N .
’ anim
ADMIRE V;_NP; N . _N
’ anim

This grammar produces trees like that shown in Figure 37. The
subject noun is categorized as N. . We cannot therefore

lexicalize this tree with admire, since admire requires an

animate subject. In other words, if the subcategorization of
nouns is handled in the grammar, the direction of selection is
from noun to verb; this seems inappropriate.
For reasons such as these it seems wrong to handle
subcategorization in the grammar. We therefore return to our
94 Part one: Constituent structure

original proposal: the subcategorization of nouns should be a

matter for lexical specification. Given this decision, let us see
how lexical insertion is to be handled. We suppose that our
grammar has rules like:

17 S —» NP + VP
VP^ V(+ NP)
NP —> (Art+)N

and our lexicon has entries like those illustrated in 8 and 9. We

need a lexical insertion rule along the following lines:

18 Lexical insertion rule:

For any terminal symbol of the constituent
structure rules:
(i) Select from the lexicon an item
characterized as a member of the class
named by the terminal symbol in
question.
(ii) Attach this item as a daughter of the
relevant symbol.
(iii) The subcategorization frame for the
relevant item must not conflict with
the environment in which the item is
to be inserted.
(iv) The selection frame for the relevant
item must not conflict with the
environment in which the item is to be
inserted.
(v) The V node is to be lexicalized first.
 Selection restrictions 95

Let us see how the rules will work. The constituent structure
rules of 17 can generate the tree of Figure 38:

Lexical insertion must now take place. By condition (v) the V

node is to be lexicalized first. Suppose we chose FRIGHTEN. It
is a V, thus satisfying condition (i). Its strict subcategorization
frame is _NP, which does not conflict with the environment
into which we wish to insert it, thus satisfying condition (iii).
Condition (iv) cannot apply since no selection environment is
yet available. We may therefore attach this item to the relevant
node, yielding Figure 39:

NP

Art N

Figure 39 FRIGHTEN

We now come to the lexicalization of the N nodes. Condition

(v) is satisfied - all the V nodes are lexicalized. Condition (iii) is
vacuous since we have not specified subcategorization frames for
the nouns in our lexicon. Condition (iv) must be satisfied since
FRIGHTEN carries a selection frame. In terms of the frame we
must select an animate object noun, but can select any subject
noun. Thus we might lexicalize to produce Figure 40:
96 Part one: Constituent structure

S
NP VP

Art N V NP

Art N

PICTURE FRIGHTEN MAN

Figure 40

which will yield a sentence like:

19 The picture frightened the man
given the appropriate choice of articles, tense, etc.
We can extend the principle illustrated above to account for
the co-occurrence restrictions we noted in 9-11 that exist
between articles and mass nouns. Suppose we amplify the rules
for the NP as follows:4
20 NP -» (Art) N + Num (Num = Number)
(sing = singular)
(pi = plural)
Art —> {a, the, sm}
Suppose also that the following general convention attaches to
the lexicalization of the N node within an NP:
21 Lexicalization of NP
N[—count] may be inserted to N within the NP
providing
(i) Art is not a
(ii) Num is not pi
N[+count] may be inserted to N within NP providing
(i) If Num is sing; Art is present, but not sm
(ii) If Num is pi; Art is not a

4 The first rule in 20 introduces a feature that we discuss in more detail on

pages 187ff. We suppose that a plural NP like the boys is derived from a string
Det + N + pi (the + boy H—s). Analogously, a singular NP like the boy derives
from a string Det + N + sing (the + boy): in this case the item sing has no overt
realization.
 Selection restrictions 97

The rules of 20 generate trees like those of Figures 41(a) and

41(h):

NP NP

Art N Num N Num

1
1
sm sing Pi
Figure 41(a) Figure 41(b)

In terms of our rules for the lexicalization of NP, a [+count]

noun, like BOY, can be inserted into the tree in Figure 41(h),
since it does not offend the restrictions for [+count] nouns (yield¬
ing boys. (A [-count] noun could not be inserted into this tree
since it offends rule (ii) for [—count] nouns.) By contrast, a
[-count] noun can be inserted into Figure 41(a), since it does not
offend the relevant rules; a [+count] noun could not. The lexi¬
calization rules will allow for all of the possible NPs shown in
11, and none of the impossible ones.
An approach of the sort illustrated blurs the boundary
between what we might call ‘grammatical’ categorization, and
what we think of as ‘lexical’ categorization. This is clearly true.
However, the decision as to what is counted as ‘lexical’ and
what ‘grammatical’ is not in any sense ‘given’: it is something
the analyst needs to work out for the particular language that he
is describing. We have in any case already seen advantages in
treating such notions as ‘transitive’ and ‘intransitive’ for verbs as
matters of strict subcategorization in the lexicon - in principle
the case illustrated above is no different. There are other
reasons why this approach is preferred. One is the sheer
complication of writing constituent structure rules to account for
the sort of distribution noted in 11; this illustrates only a part of
the whole complex web of co-occurrence restrictions between
articles and nouns. A second reason has to do with the relation
between categories like [± count] and various types of semantic
characterization - a subject we will explore (pages 242-4).
We now return to the question raised on page 91: is it
appropriate to use syntactic features both to control grammatical
occurrence, illustrated in 11, and selectional co-occurrence,
illustrated in 10? It does not seem appropriate for two main
98 Part one: Constituent structure
reasons, both to do with the fact that selection restrictions
appear more a matter of semantic compatibility between lexical
items than of the grammaticality of sentences as such.
A semantic and a syntactic characterization of some item need
not be congruent. In setting up a set of syntactic features, we
would be wisest to pay attention to syntactic facts. This can be
illustrated with the category [±count] in English. Our initial
characterization of the distinction was semantic - as between
‘discrete’ and ‘non-discrete’ objects. This characterization holds
rather generally but not universally. Some nouns, like GRAPE,
PLUM, PEAR are normally [+count]; others, like
GRAPEFRUIT, APPLE, may be either [+count] or [-count], as
can be verified by checking the nouns in NPs on the lines of 11.
I could enquire of a guest

22 Would you like some (sm) grapefruit?

but hardly

23 *Would you like some (sm) plum?

even though what was being offered was similar - in the one
case a plate of segmented grapefruit and in the other a plate of
stewed plums. The categorization of nouns as [±count] is not
only to some degree idiosyncratic within a language; it is also
idiosyncratic between languages.
This point, the mismatch between a syntactic and a semantic
characterization, can be made even more clearly by reference to
a language that has a grammatical category of ‘gender’. In French,
for example, all nouns are categorized as being in one of two
classes (with a few exceptions which may be in both classes)
which are usually given the names ‘masculine’ and ‘feminine’. The
grammar needs to know the gender of a noun in order that
appropriate grammatical rules can be applied: for example the
rules of ‘concord’ that ensure that a masculine noun co-occurs
with the masculine form of the article, le, un, etc., and a
feminine noun co-occurs with a feminine form of the article, la,
une, etc. Gender distinctions in French are not typically a
semantic matter, since all nouns are assigned to one gender class
or the other (c.f. le stylo ‘the pen’, but la table ‘the table’) even
when the notion of sex is irrelevant. We could handle this by a
syntactic feature [±masculine], but that feature would hardly be
appropriate for stating selection restrictions. For the correct
 Selection restrictions 99

selection restriction of an adjective like ENCEINTE ‘pregnant’,

for example, we need to know the sex of the referent, not the
grammatical gender of the noun. This is particularly clear in the
case of those nouns that are masculine in gender but may be
used to refer to individuals of either sex, like le docteur ‘doctor’.
This question also has marginal relevance in English: some
kinds of object, like a ‘ship’ and a ‘car’, are often
pronominalized with she, rather than it (one may perhaps
speculate that this is because they are used to refer to objects
men find attractive!), and conversely a ‘baby’ is often referred
to as it.
The second reason is that a complete specification of the
selection restrictions of some item clearly cannot be handled by
syntactic features alone. Consider the verb eat. In its literal
sense this means ‘consume as food’. The object NP might be
specified as [+concrete] since we normally eat ‘things’. But
clearly not all ‘things’ are ‘edible’. We might then suppose that
we should add a feature [+edible]. If we do, this is clearly not a
syntactic feature in the sense discussed, since it has no relevance
at all for any grammatical process, syntactic co-occurrence
restriction, etc. And even a feature like [+edible] does not solve
all our semantic problems: what is to be classed as [+edible]
depends on what is ‘doing the eating’. Humans do not normally
eat wood, coal, carrion or clothes - but termites, fires, crows
and clothes-moth caterpillars do! Even a man might on occasion
be said to eat his hat - though this might be considered
‘figurative’ language.
Our conclusion then is this: nouns are marked in the lexicon
with a set of syntactic features (inherent subcategorization)
which may be used to account for various grammatical features
(like co-occurrence with articles). In a similar way verbs are
marked with a strict subcategorization frame.5 While it is
possible to include a specification for selection restrictions using
the syntactic features which characterize nouns, it is not
desirable to do so: selection restrictions are primarily a matter

5 We have not considered any cases of nouns with strict subcategorization

frames, but this is appropriate in some cases (see page 136). Similarly we have
not considered inherent subcategorization for verbs, other than the marking V,
though this too will be appropriate in some cases (see discussion on page
238).
100 Part one: Constituent structure

for a semantic description. This suggests that we should not

include within the grammar or the lexicon (insofar as it impinges
on the grammar) the selection frames so laboriously established
at the beginning of this chapter. This is exactly what we propose
to do, but not because questions of selection are unimportant,
or not a matter of interest for the linguist. They clearly are of
considerable interest. Our proposal simply implies that we
should not consider the specification of selection restrictions as a
part of the grammatical description of a language. So the
sentences marked in 10 as deviant are considered only as
semantically and not as grammatically deviant. This seems a
more suitable conclusion: so, for instance, with respect to 10b,
while the sentence might well be considered as ‘unusual’, it is
not uninterpretable - but the onus of interpretation falls on the
semantic component of a complete description, not on the
grammatical component. This frees a semantic description of
such matters from an unnecessarily intimate link with a
particular grammatical description, which may in some cases
unduly restrict its freedom. It also frees the grammatical
description from a semantic description where the two are not
congruent. Both consequences seem desirable.
The decision does not however mean that we wish to discard
the lexical subcategorization of nouns in terms of syntactic
features. We have seen that these can be useful to specify
grammatical co-occurrence, etc., and we wish to retain them.
Nor does the decision imply that we wish to sever the bonds
between the syntactic and the semantic description of a
language. The discussion has made clear that there is some
degree of congruence between semantic features, of the sort we
might use for selection, and syntactic features, of the sort we
might use to control grammatical co-occurrence, etc. This is why
we were able to use syntactic features at the beginning of the
chapter to make some quite useful remarks about selection. This
should not surprise us - language is after all, a means of
conveying meanings. The congruence between a semantic and a
syntactic characterization should cause no surprise - it would be
more surprising if there were a complete mismatch! Syntactic
features, however, can only be used to account for the grossest
facts about selection, and only insofar as they do not conflict
with semantic characterizations. Our conclusion therefore is that
selection restrictions are not part of a grammatical description.
 Selection restrictions 101

Technical terms

cross-classification selection restriction

feature selection frame
inherent subcategorization strict subcategorization
lexicalization

syntactic features used for subcategorization of nouns:

[± animate]; [±human]; [±male]; [±concrete]
(concrete : abstract); [±count] (count: mass); [±common]
(common : proper)

Exercise
You are probably familiar with playground jingles like:

One fine day in the middle of the night

Two dead men got up to fight,
One blind man to see fair play,
Two dumb men to shout, ‘Hurray.’
Back to back they faced each other,
Drew their swords and shot each other.

Would you wish to consider such jingles as ‘ungrammatical’? If

not, how could you handle a linguistic description of the obvi¬
ously anomalous nature of such jingles?
8 Relations between sentences

Pairs of sentences like:

1 The cat has eaten the mouse

2 The mouse has been eaten by the cat

are related to each other both semantically and syntactically.

The semantic relationship lies primarily in the fact that both
sentences share the same ‘agent’ (the cat), responsible for the
action described by the verb, and the same ‘patient’ (the mouse),
affected by the action described by the verb. (Notions like
‘agent’ and ‘patient’ are discussed more fully on pages 288ff.)
One could say that 1 describes the situation from the cat’s point
of view, and 2 describes the same situation from the point of
view of the mouse (we shall discuss these questions on pages
357ff.); for our present purposes the relevant semantic rela¬
tionship is that in both sentences the cat is the eater and the
mouse is the eaten. This is reflected in the traditional
description of sentences like 1 as ‘active’ (agent subject) and 2
as ‘passive’ (patient subject).
The structural relationship between these sentences can be
diagrammed as in Figure 42.

Figure 42

There are three syntactic differences between 1 and 2:

 Relations between sentences 103
i The NP (the mouse) that is the object of the verb in 1
‘becomes’ the subject of the verb in 2.
ii Correspondingly, the NP (the cat) that is the subject in 1 has
the preposition by adjoined to it to form the Prepositional
Phrase (PP) by the cat, and this now follows the verb.
iii In 1 the verbal constituent has eaten consists of the auxiliary
verb has followed by the main verb eaten. In the passive
sentence the verb constituent is changed by the addition of
been, a form of the verb be, which is added immediately
after (and as a right sister to) the auxiliary verb. We call this
form of the verb be the ‘passive auxiliary’. We discuss
auxiliary verbs and the rules for the formation of verb forms
later (pages 105ff.).

The active-passive relationship can be generalized to any

sentence containing a transitive verb:

3a The dog will bite the cat

3b The cat will be bitten by the dog
4a The man is beating the boy
4b The boy is being beaten by the man

etc. As before, the objects of the a sentences ‘become’ the

subjects of the corresponding b sentences; the subjects of the a
sentences ‘become’ PPs with by and follow the verb, and a
passive auxiliary, a form of be, is introduced after the auxiliary
(will BE bitten, is BEING beaten).
This suggests that we can generalize the syntactic statement of
this relationship by abstracting away from particular lexical
items in particular sentences, and state the relationship in terms
of general syntactic structure, as in Figure 43.

Figure 43
104 Part one: Constituent structure
We can alternatively state the relationship in formulaic terms as:

5 NP — Aux — V— NP2

O NP2 - Aux + Pass — V — by + NP^

The formula is interpreted as follows. For any sentence that has

a constituent structure analysis as shown on the left-hand side of
the formula (NP^ - Aux - V - NP2), there is a corresponding
sentence which has the analysis shown on the right-hand side of
the formula (NP2 - Aux + Pass - V — by + NPj). The symbol
’ separates those constituents that are of relevance to the
statement of the rule: we are not for example concerned with
the internal structure of the NPs in question, since this is not
relevant to the formulation of the relationship: both of the
following are good active-passive pairs of sentences:
6a The little old lady who lives upstairs has fed our cat
6b Our cat has been fed by the little old lady who lives
upstairs
7a John painted that remarkable picture hanging on the
wall over there
7b That remarkable picture hanging on the wall over there
was painted by John

We have used the symbol ‘ + ’ to indicate that some particular

pair of constituents form a construction: we want to indicate
that the string by + NP in the passive sentence is a constituent.
The symbol ‘Pass’ indicates an appropriate form of the passive
auxiliary be. We have used the symbol ‘o’ to indicate that the
two structures are related.
Before considering the active : passive relationship further, we
must turn our attention to auxiliary verbs. Consider the
sentences:
8a The dog bit the cat
8b The dog has bitten the cat
8c The dog is biting the cat
8d The dog has been biting the cat
8e The dog will bite the cat
8f The dog may have bitten the cat

All the sentences have the same subject NP (the dog) and the
same object NP (the cat). They differ in the internal structure
 Relations between sentences 105

of what we call the Verb Group. Each Verb Group consists

minimally of a form of the verb BITE, which we call the main
or lexical verb. It is always the last constituent. The main verb
may be preceded by one or more other verbs - in our examples
we have forms of the verbs BE, HAVE, WILL and MAY. We
call these auxiliary verbs.
We will consider first those sentences like 8b-8f which contain
auxiliaries, and return later to consider sentences like 8a, which
has only a main verb. An appropriate constituent structure
analysis for sentences 8b-8f is, schematically, Figure 44:

Figure 44 Aux V NP

The node Aux dominates one or more auxiliary verbs, and V

dominates the main verb.
The principle function of the main verb is to introduce the
appropriate lexical content - all the sentences have to do with
‘biting’. The principal function of the auxiliaries on the other
hand is to relate the sentence to ‘temporal’ and ‘aspectual’
distinctions:1

9a The dog is biting the cat (now)

9b The dog was biting the cat (yesterday afternoon)
9c The dog will bite the cat (tomorrow, I expect)

1 Temporal distinctions are often related to a grammatical category of ‘Tense’

- temporal distinctions (though not always tense distinctions, see pages 246-8)
refer to the location of an event, etc. in time - before, after or concurrent with
the moment of utterance.
‘Aspectual’ distinctions are often related to a grammatical category of
‘Aspect’. Aspectual distinctions commonly refer to the distribution of an event
over time. The main aspectual distinctions usually identified in English are
‘progressive’ (indicating that an action is, or was, etc. in progress, the cat is/was
biting the dog), and ‘perfective’, indicating that an action has been completed
(the dog has bitten the cat). Other aspectual distinctions often referred to in
English include ‘habitual’ (I drive to work) and ‘stative’ (7 am clever). The rela¬
tionship between distinctions of tense and aspect and particular verb forms is
complex and is discussed on pages 247-8.
106 Part one: Constituent structure
or to estimations of the probability, likelihood, etc. of the event
described in the sentence having taken place, being likely to
take place etc.:

10a The dog may have bitten the cat (but that’s unlikely)
10b The dog could have bitten the cat (but I hope it didn’t)

One obvious difficulty about the analysis of auxiliary verbs in

English is that each auxiliary requires that the verb, auxiliary or
main, that immediately follows it, should take a particular form.
So, if we have one of the auxiliaries MAY, WILL or CAN the
following verb will be in its ‘base’ form:

11a The dog will bite the cat

lib The dog may be biting the cat
11c The dog can’t have bitten the cat

Such auxiliaries, and others like SHALL and MUST, are usually
known as ‘modal’ auxiliaries (because one of their principle
semantic functions is to mark the ‘modal’ distinctions of
probability, etc. noted above). They constitute a form class
insofar as they are mutually substitutable in a particular verb
group, but no verb group can contain more than one of them:

12a The dog may/will/must/could have bitten the cat

12b *The dog may will have bitten the cat

If the auxiliary is a form of HAVE (usually called the

‘perfective’ auxiliary) then the following verb, main or auxiliary,
is in the ‘past participle’ form. Exactly what shape the past
participle takes depends on the particular verb involved, but it is
frequently a form which ends in the suffix -ed or -en:

13 The dog has frightened the cat

The dogs have bitten the cat
The dog has been biting the cat

though not always:

14 The man has hit the dog

John has swum the channel

To remind us of the formation rules for the perfective auxiliary

we represent it as have + pp, where HAVE reminds us that
the auxiliary involved is a form of have and pp indicates that
the following verb is in the past participle form.
 Relations between sentences 107
If the auxiliary is a form of the non-passive be (usually called
the ‘progressive’ auxiliary, since one of its main functions is to
indicate that the action, etc. described by the verb is in
progress) the following verb is in the -ing form:

15 The dog is biting the cat

The dog may be bitmg the cat

As before, to remind us of the formation rules for the

progressive auxiliary, we describe this as be + -ing: be
indicating a form of the verb BE (which one depends on what
other verb this follows) and -ing to indicate that the following
form is an -ing form.
If the auxiliary is the passive be (which we call the ‘passive’
auxiliary since it is involved in the formation of passive verb
groups), then the following verb once again is in the ‘past
participle’ form:

16 The cat has been frightened by the dog

The cat has been bitten by the dogs
The cat is be ing bitten by the dog

We describe this as BE + pp, on the same principles as before.

The last of the examples in 16 has both the progressive BE
(BE + ing: is beiNG) and the passive BE (BE + pp: being
bitlen). Both involve forms of BE, but are distinguished by the
form of the following verb.
The above account shows a number of peculiar features about
the structure of verb groups in English: in particular, the choice
of a particular auxiliary has consequences for the verb that
immediately follows it. This fact makes it extremely
cumbersome to describe the English verb group using
constituent structure rules, so we resort to the more ‘abstract’
type of representations of the auxiliaries as have + pp,
BE + -ing and BE + pp, noting that the representations pp and
-ing relate to what happens to the following item. We can now
say that the verb group consists of, minimally, a main verb, and
the main verb may, optionally, be preceded by one or more
auxiliaries. These must be in the order shown schematically in
Figure 45.
In the example sentence the first auxiliary MAY is followed
by the base form of the perfective auxiliary have, have. This is
108 Part one: Constituent structure

VP

e.g. (the cat) may have been being bitten (by the dog)

Figure 45

followed by the past participle form of BE, been. This BE is the

progressive BE and is in turn followed by the -ing form, being.
This BE is the passive BE, which is itself followed by the past
participle form of the main verb, bitten.

In the example sentence in Figure 45 all the auxiliaries have

been chosen. Obviously this is not obligatory: any combination
of auxiliaries is possible, providing the order shown in Figure 45
is maintained. Note that if verb groups are formed with the
auxiliaries in another order they are ungrammatical:

17 * Modal Prog Perf

*The dog will be having bitten the cat
18 * Modal Pass Prog
*The cat will be been biting by the dog
19 *Prog Modal
*The dog is canning bite the cat

We must now look at verb groups with no auxiliary:

20a The dog bites the cat (every day)

20b The dog bit the cat (yesterday)
 Relations between sentences 109
The distinction shown between the forms bites and bit is
generally described as one of ‘tense’, the form in 20b being the
‘past tense’ form, and that in 20a the ‘non-past’ form. The
typical function of the ‘past’ form is to refer to events located in
‘past’ time, i.e. times previous to the moment of utterance of the
sentence involved. The other form is usually contrasted as the
‘non-past’ form, not the ‘present’ form, since it is not typically
used to refer to events taking place at the moment of utterance
of the relevant sentence. The form often used in this sense is the
present progressive (the dog is biting the cat (at this moment)),
and this verb group involves the progressive auxiliary. A typical
use of the non-past form is to refer to habits (as in 20a), general
truths (dogs bite cats), etc. In the orthographic form of the
language the past form of verbs is generally formed by the
addition of the suffix -ed to the stern of the main verb (kill-ed;
frighten-ed\ want-ed, etc.), but this is not the case for all verbs,
as, for example, bit in 8a, rather than *bited (and similarly
run: ran, think : thought, etc.). For now we follow the
traditional account and simply refer to all such forms as ‘the
past form of kill’, ‘the past form of bite’, etc. (for a fuller
discussion see pages 168-9).
The past: non-past alternation shown in 20a,b is also
characteristic of verb groups that contain auxiliaries:

21a The dog has bitten the cat

21b The dog had bitten the cat
22a The dog is biting the cat
22b The dog was biting the cat
23a The dog can bite the cat
23b The dog could bite the cat
24a The dog bites the cat
24b The cat is bitten by the dog
25a The dog bit the cat
25b The cat was bitten by the dog

In all cases the first verb in the verb group shows the alternation
in tense, however many items there are in the group. Only this
first position in the verb group is open to this alternation, since
the form of verbs other than the first is determined by what
precedes them. This is most clearly shown in the last two pairs
of examples. In 24a and 25a the first verb in the verb group is
110 Part one: Constituent structure
the main verb itself, and this consequently shows the tense
alternation. In the corresponding passive sentences 24b and 25b
the passive auxiliary is introduced before the main verb: the
passive auxiliary is now the first verb in the verb group, and
consequently shows the tense alternation: the main verb itself
must be in the past participle form, since this form, as we have
seen, is the form that follows the passive auxiliary.
How are we to treat these facts? The description we adopt
(discussed further on pages 207-10) treats tense as a sort of
auxiliary, and, furthermore, as the first auxiliary in the verb
group. This entails saying that this tense auxiliary is not realized
itself as a discrete item, but that its presence is felt as a marker
on the first item to follow it within the verb group. This gives it
a status much like -ing and pp in our description of the
progressive and perfect auxiliaries. It also captures the
generalization that tense is shown on the first member of the
verb group. We must now modify our schematic constituent
structure representation for the verb group from that of Figure
45, to that shown in Figure 46:
VP

Tense Modal Perf Prog Pass

VP -> Aux + V

Aux —> Tense (Modal) (Perf) (Prog) (Pass)

Tense
Figure 46

The constituent structure rules shown in Figure 46 state in the

expansion rule for Aux that only Tense is an obligatory
constituent: the other auxiliaries are optional. This explains the
generalization that all verb groups are ‘tensed’, but not all
groups contain an auxiliary. We can now show diagrammatically
how we can deal with the verb groups in 24 and 25 in Figures
47-50. The reader can himself draw comparable structures for
the verb groups in 21-3.
 Relations between sentences 111

VP

Aux V

Tense

non-past BITE

Figure 47 bites (24a) Figure 48

The structures in Figure 47 and Figure 49 are identical except

for the introduction of the passive auxiliary - and similarly
Figures 48 and 50.
One final remark on this analysis of auxiliaries: it introduces a
degree of abstraction into the description absent from the simple
constituent structure grammars we have considered hitherto. This
abstraction arises because we no longer have a simple
one-to-one relationship between the order of words, and parts
of words, as they actually occur in sentences, and as we describe
them in our grammar. Each of the elements we have labelled
-ing, pp and Tense is not realized in the actually occurring
sentences in the place we have described it in the grammar, but
on the following constituent. We will see in chapter 14 how we
can relate these abstract descriptions to the actually occurring word
forms - at that time we will also consider a few residual features
of the analysis of auxiliaries that we have ignored in this
account, in particular the question of ‘agreement’ in ‘number’

VP

Aux V

Tense Pass Tense Pass

I
non-past BE
A pp BITE past be pp BITE

is
\/
bitten
A/
was
\/
bitten

Figure 49 (24b) Figure 50 (25b)

112 Part one: Constituent structure
between subject expressions and auxiliaries (The man is going,
The men ARE going).

After this lengthy digression on auxiliaries, let us return to our

description of the passive. We formulated the active : passive
relationship as:

26 NP - Aux - V - NP2

oNP^- Aux + Pass — V — by + NP

This formula gives the correct analysis of passive verb groups.

We noted before that the precise internal structure of the NPs is
irrelevant (c.f. discussion of 6-7); we can now observe that the
precise internal structure of Aux is also irrelevant. It must
consist minimally of Tense, but it does not matter which - if any
- of the other auxiliaries are also present. In all cases the Pass
auxiliary in the corresponding passive sentence appears as a
right sister of the other auxiliaries, immediately preceding the
main verb. This is illustrated in Figures 49 and 50, and the
reader is invited to experiment with other active : passive pairs.
Let us now turn our attention to pairs of sentences like:

27 The cat has eaten the mouse

28 Has the cat eaten the mouse?

Once again there is both a syntactic and a semantic relationship.

The most usual interpretation works from a semantic difference
between 27 and 28: 27, the syntactic form called ‘declarative’, is
most usually used to make a statement, whereas 28, with the
syntactic form usually called ‘interrogative’, is usually used to
ask a question. However, the relationship of ‘agent’ (the cat) and
‘patient’ (the mouse) remains constant, as it did in the
active : passive sentences discussed earlier: to this extent the
pair of sentences has semantic material in common. As before
we can diagram the syntactic relationship between the
sentences:

Figure 51
 Relations between sentences 113
Once again we can generalize to other syntactically similar
sentences:

29a The cat will eat the mouse

29b Will the cat eat the mouse?
30a The cat has been eating the mouse
30b Has the cat been eating the mouse?
31a John has left
31b Has John left?
32a Mary can sleep in the spare bedroom
32b Can Mary sleep in the spare bedroom?
33a The baby has always been fed at midnight
33b Has the baby always been fed at midnight?

In all the interrogative sentences, the b sentences, an auxiliary

appears before the subject NP (Will - the cat . . . (29b), Has -
the cat . . . (30b), etc.); it is always the same as the first auxiliary
of the verb group of the corresponding declarative sentence.
The rest of the sentence is the same. As a first approach to a
formalization of the relationship between these pairs we can
diagram them as in Figure 52:

Figure 52

(By Auxj we mean ‘the first auxiliary’ and by X we mean ‘any¬

thing that follows the first auxiliary’.) In our examples ‘anything’
includes the main verb (29), another auxiliary (30) and an
adverb (33).
In terms of the description of the verb group given earlier this
analysis cannot be quite right: the first auxiliary in our previous
analysis was Tense, and clearly in the cases illustrated above one
of the auxiliary verbs (will, have, can) also appears before
the subject NP. So let us examine the problem more closely: in
Figure 53 each example is followed by an analysis in the terms
already established.
114 Part one: Constituent structure

In these examples, the tense auxiliary does indeed move round

the subject, whether or not the tense auxiliary is itself further
developed as past or non-past. The other element that follows
Tense to a position before the MV is the item HAVE, BE or a
Modal. The pp or -ing element is evidently ‘left behind’, as can
be seen by inspection of the examples. These last two observa¬
tions suggest we cannot simply talk in terms of Tense + Perf or
Tense + Prog, etc, as being involved, since only part of Perf or
Prog is involved.2 Finally, observe that only Tense and have or
be or Modal is involved. We can now reformulate Figure 52 as
Figure 54. (X still stands for ‘anything that follows the items
specified in the second box’, including the elements pp and -ing
left behind when HAVE or BE follow Tense to a position in
front of NPj.)
There are other kinds of interrogative sentence where there is
no overt auxiliary. Thus corresponding to the declarative
sentences:

37a John leaves (every morning at six o’clock)

37b John left (yesterday)

there are the interrogative sentences:

38a Does John leave . . . ?

38b Did John leave . . . ?

In English the main verb itself cannot ‘hop’ the subject NP:
there are no sentences like * Left John?, *Leaves John?

2 A minor advantage of formulating the rule as involving be and have

rather than the progressive and perfective auxiliary, is that it will also catch BE
and have when they are ‘main verbs’, be as a main verb always forms inter¬
rogative sentences in the pattern:

She is pretty Is she pretty?

He is at home Is he at home?

have is somewhat more problematical; its syntax varies to some degree from
speech community to speech community. For some
She has a car Has she a car?

are acceptable. Other communities treat have more like a ‘main verb’, which
we discuss next:

She has a car Does she have a car?

 Relations between sentences 115

34a John has left NP non-past have PP V

X
34b Has John left? non-past have NP PP V

35a John was coming NP past BE -ing V

35b Was John coming? past BE NP -ing V

36a John will be coming NP non-past Modal B E -ing V

t
36b Will John be coming? non-past Modal NP BE -ing V

Figure 53

Figure 54

The tense distinction shown in 37a and 37b, where it is carried

by the main verb (leaves : left), is still maintained in 38a and
38b, now carried by forms of the verb DO (does : did), and the
116 Part one: Constituent structure
main verb appears in its base form. The verb DO appears in the
same structural position in 38 as do the auxiliaries has, was, etc.
in 34-6. Perhaps what has ‘hopped’ the subject NP in these
cases is simply the tense auxiliary. We can now return to an
earlier observation: the tense auxiliary cannot appear as a
separate element all by itself; it needs a verb, as it were, to
carry it. Since, under the analysis proposed, there is no verb
available, do is introduced as a ‘dummy’ verb, a ‘carrier’ of the
tense. We can now add a further piece of description to
Figure 54: this is shown in Figure 55:

Figure 55

This relationship only holds when Tense immediately precedes

the main verb (MV), since forms like:

39 *John do be leaving
*Do John be leaving?

are not acceptable in standard English (though they are part of

some regional varieties of English).
Let us consider next negative sentences:

40 The cat has eaten the mouse

41 The cat hasn’t eaten the mouse

Once more there is a syntactic and a semantic relationship.

Once again the semantic similarity lies in the fact that the agent
is the same in both sentences, as is the patient: obviously the
sentences differ in that 40 may be used to assert and 41 to deny
that the cat ate the mouse. The syntactic relationship lies in the
fact that both sentences have the same structure, except that
there is a negative marker, -n’t in the example, affixed to the
first auxiliary. This, as the reader can easily check, is regular
 Relations between sentences 117
when there is an actual auxiliary verb present. Once again
descriptive problems arise where there is no overt auxiliary:

42a John leaves (every morning at six o’clock)

42b John doesn’t leave (every morning at six o’clock)
43a John left (yesterday)
43b John didn’t leave (yesterday)

Again the DO verb appears, in this instance seemingly because

the negative marker, (-n’t) cannot be affixed to a main verb:
there are no forms *John leaven’t, Vohn leftn’t, etc. The
negative marker can only be affixed to an auxiliary verb; there
is no auxiliary verb, so DO is supplied. But Tense is also carried
by the DO verb: there are no forms Vohn doesn’t left, etc.,
where Tense appears on the main verb and the negative is
carried by the DO verb. Both Tense and Negative must be
marked on the same verb - as can be seen by considering
negative questions:

44 Hasn’t the cat eaten the mouse?

Won’t John come back tonight?

and most strikingly:

45 Doesn’t John like strawberries?

Didn’t Mary come to your party?

We are now able to summarize in tabular form the

affirmative : negative relationship:

Figure 56
118 Part one: Constituent structure

Figure 57

These tables again use the symbols X and Y to indicate ‘any

arbitrary constituents’. It doesn’t matter what other constituents
an affirmative sentence has: providing it has the constituents
specifically named, there is a corresponding negative sentence
with the analysis shown. In other words, for every affirmative
sentence, there is a corresponding negative sentence.

Let us pause at this point and summarize. We have discussed

three types of sentence relatedness illustrated in:

46a The dog has bitten the cat active

46b The cat has been bitten by the dog passive
47a The dog has bitten the cat declarative
47b Has the dog bitten the cat? interrogative
48a The dog has bitten the cat affirmative
48b The dog hasn’t bitten the cat negative

We can summarize the relevant structural relationships in terms

of formulae rather than diagrams. In the formulae which follow
we use the following conventions: the relevant constituents are
separated by dashes (—); internal structure within a constituent
is symbolized by a plus sign (+); the fact that the structures in
question are systematically related is shown by using a
double-headed arrow (<s>). Note also the use of square brackets:
an item in square brackets on one side of the rule is to be
matched with the corresponding item in the square brackets on
the other side.
 Relations between sentences 119
49 Active <=> Passive (cf. Figure 43)
NPj - Aux- V- NP2

^ NP^ — Aux + Pass — V — by + NP^

50 Affirmative o Negative (cf. Figures 56 and 57)

HAVE
X — Tense BE - Y
.Modal
HAVE
X — Tense + Neg BE - Y
.Modal.
X - Tense - MV - Y
o X — Tense + Neg + DO — MV — Y

51 Declarative Interrogative (cf. Figures 54 and 55)

'HAVE
NP — Tense + BE - X
.Modal
HAVE
o Tense + BE - NP-X
. Modal
NP — Tense — MV — X
oTense + DO - NP - MV - X

All the a sentences in 46-8 are identical; it has long been

traditional linguistic wisdom that there is something ‘basic’
about ‘active, declarative, affirmative’ sentences, like 52a below,
in comparison with, say, ‘passive, interrogative, negative’
sentences like 52b:

52a The dog has bitten the cat

52b Hasn’t the cat been bitten by the dog?

The description we have offered can perhaps go some way

towards explaining why sentences like 52a seem more basic than
those like 52b. There are two reasons.
In the first place, if the sentences in each pair in 46-8 are
related, and if 46a, 47a and 48a are identical as they are, then
the corresponding b sentences must also be related. In each case
the relationship is, as it were, mediated through the common a
sentence. A distinct structural relationship between 46b and 48b
could be established in the manner shown in Figure 58:
120 Part one: Constituent structure

Figure 58

Prima facie there is no reason why this can’t be done, but it

seems more revealing to interpret the relationship as shown in
Figure 59:

passive

t
1
active
affirmative

negative

Figure 59

where one parameter of change is accommodated at a time: the

intermediary is our basic sentence.
A second reason - which is implicit in the preceding account -
is that the various relationships we have been discussing can be
compounded. Thus, related to the basic sentence:

53 The dog has bitten the cat

is the corresponding passive:

54 The cat has been bitten by the dog

and related to this is the corresponding passive negative:

55 The cat hasn’t been bitten by the dog

 Relations between sentences 121
and related to this is the corresponding passive interrogative
negative:
56 Hasn’t the cat been bitten by the dog?

The affirmative : negative relationship holds whether the

sentence is active or passive; the declarative : interrogative
relationship holds whether the sentence is active or passive,
affirmative or negative. Breaking down these statements of
sentence relatedness into ‘elements’, active: passive,
affirmative : negative and so on, allows us to state the most
basic relationships, and to combine these basic relationships to
form more complex relationships. In doing so, the most ‘basic’
of all the sentences on which we can operate is the declarative,
active passive.
Relationships of this kind are known as transformational
relationships. They are relationships of one structure (with the
constituent structure analysis shown on one side of the rule),
to another structure (that has the constituent structure shown
on the other side of the rule). This can be illustrated by the two
analyses in Figures 60 and 61 on page 122. The rule that relates
these two structures is, as we have seen (cf. 51):

57 NP - Tense - MV - X<>Tense + DO - NP - MV + X
The following question now arises: in statements like 57, or
the other statements shown in 49-51, are we to suppose that the
structures are simply related, or is the one structure to be
derived from the other? In formulating these rules, we have
supposed that it is simply a relation between two structures: this
has been symbolized by the use of the double-headed arrow
which may be taken to mean ‘any structure of the form found
on one side of the rule can be matched by a structure found on
the other side of the rule’. If you find a structure as shown on
the left, you also find one as shown on the right; and vice versa,
for any structure as shown on the right, there is a corresponding
structure as shown on the left. This is a statement of relatedness.
We can, however, reinterpret such rules with minor modification
as rules of derivation, and symbolize the fact that the rules are
derivational by using a single-headed arrow =>. This we take to
mean that, given the structure on the left-hand side of the rule,
a structure can be formed as shown on the right-hand side of
the rule by performing the changes noted in the rule. Thus in
122 Part one: Constituent structure

Figure 60 (declarative)

Figure 61 (interrogative)

the derivational sense we would say that Figure 61 could be

formed from Figure 60 by (i) moving tense to the left-hand side
of NP and (ii) inserting DO after Tense.
Most grammars that make use of transformational
relationships formalized in the way that we have begun to
formalize them are ‘derivational’ rather than ‘relational’: the
rules are formulated using the symbol => (can be changed into
the structure) rather than the symbol o (can be related to the
structure). There are a number of reason§ why. One is that it
 Relations between sentences 123

simplifies the associated constituent structure grammar. If we

wish to use a relational type of rule, then the constituent
structure rules must be able to generate all types of sentences
directly, including passive sentences, interrogative sentences,
negative interrogative sentences, and so on. To get a correct
account of the constituent structure for all these diverse
sentence types, together with all necessary co-occurrence
restrictions, means a very complex set of constituent structure
rules. For example, we need to allow for a negative element to
be generated before the subject (for negative interrogative
sentences) and for another negative element to be generated
after the subject (for negative declarative sentences): but we
need to specify that both negative elements cannot be present in
one sentence to avoid nonsense like *Hasn’t the man been’t
fishing? Furthermore, the relational transformational rules will
be a sort of ‘addendum’ to the grammar, rather than a fully
integrated part of it.
If, however, we use a derivational type of rule, as in
transformational grammars, then the constituent structure rules
need to account only for the structure of the ‘basic’ sentences
we discussed earlier. Other types of sentence are derived from
these basic sentences by ‘changing them into’ some derived
structure. Under this account the problem of the multiple
negatives mentioned in the previous paragraph (and similar
problems) simply does not arise. Furthermore, the
transformational rules are fully integrated into the grammar,
which then formally contains not only a statement of basic
constituent structure patterns, but also a statement of regular
relationships between sentences. Such a grammar consists of two
components - a constituent structure component (or ‘phrase
structure component’) and a transformational component. This
book does not study transformational grammar in any more
detail than is considered in this and the next chapter. In
particular we do not discuss further the formalisms used in such
grammars - the interested reader is referred to the books listed
on pages 386-7. It is, however, useful to have a flavour of the
way in which such grammars are presented, so we devote the
remainder of this chapter to reformulating the rules already
presented as a derivational transformational grammar.
First we need constituent structure (or phrase structure) rules
to generate basic structures. The following will serve:
124 Part one: Constituent structure

s NP + VP
VP Aux + V (+NP)
NP —» Art + N
Aux —> Tense (Modal) (Perf) (Prog)

Tense —> {past

[ non-past,1J
Perf HAVE + pp
Prog -> B E + -ing

This grammar is clearly only illustrative - it could be

elaborated with some of the other rules we have discussed in
previous chapters. Associated with these rules we also need a
lexicon and a lexical insertion rule, as discussed in chapter 7.
These two - the constituent structure, or phrase structure, rules
and the lexicon - comprise what is referred to as the ‘base
component’. From the base component we can derive lexicalized
trees which we call underlying structures.
The transformational component contains transformational
rules that operate on these underlying structures to produce a
derived structure. These rules may rearrange elements of the
underlying structure, add elements to it, or delete elements from
it. After the operation of all the relevant transformational rules
we have a syntactic surface structure.
In some cases the underlying structure and the syntactic
surface structure are almost identical. This is the case with
active declarative sentences (the ‘basic’ sentences discussed on
page 119), where we only need to use rules that account for
the actual ordering of elements like past, -ing and pp which, as
you will see, the grammar generates preceding the item to which
they will eventually be affixed: we discuss machinery to
accomplish this on pages 207-9. In other kinds of sentences a
more radical restructuring occurs. An example is the passive.
The rules in 58 derive elements in the order of active sentences
only (the expansion of Aux contains, for instance, no provision
for a passive auxiliary). The transformational rule re-orders the
subject and object NPs and introduces a passive auxiliary. We
discuss this immediately, but the reader may now like to consult
Figure 62, which illustrates graphically the changes the passive
transformation effects. The structure at the top of this figure
is the underlying structure: the annotations show how this struc¬
ture is ‘changed’ into the derived structure shown at the bottom
 Relations between sentences 125

of the figure. This derived structure still needs to go through

the rules to re-order the elements past, -ing and pp mentioned
above to produce the syntactic surface structure.
Since the transformational rules are to be derivational rather
than relational, we need to amend our previous rules in various
small ways. The way we formulate transformational rules in this
and the following chapter is illustrated by considering the pas¬
sive transformation:

59 Active => Passive (optional)

SA: NPj — Aux — V — NP^

SC: NP2-Aux(Aux + Pass(BE + PP»

- V-ppfby + NPj)

i NP^ is moved to the front of the S, and adjoined

as a daughter of S
ii A new constituent, Pass, expanded as be + pp,
is adjoined as a daughter of Aux, and the right
sister of all other Aux constituents
iii by is adjoined as left sister of NP^ to form a new

constituent PP, and this is adjoined as right sister

of the VP.

SA is an abbreviation for ‘structure analysis’: it indicates that

the rule only applies to trees that can be analysed into the con¬
stituents shown in the analysis. The relevant constituents are
separated by - as before. The abbreviation SC indicates ‘struc¬
ture change’: it shows how the structure is to be changed. The
notes below present, in ordinary language, an account of the
relevant changes. We have also noted that the rule is optional -
the passive does not have to apply whenever the structural
analysis is met (otherwise we would find no active transitive sen¬
tence!). The additional detail in this formulation is an attempt to
specify with some precision what the derived structure of the
passive sentence will look like. This needs to be specified care¬
fully since the passive structure may itself become the ‘input’ to
a subsequent transformational rule (like that for forming a pas¬
sive negative for example).
The effect of the rule is shown diagrammatically in Figure 62,
which is annotated to show the relevant changes:
126 Part one: Constituent structure

Figure 62

The other rules are amended as follows:

50 Affirmative => Negative (optional)

( HAVE
S A: X — Tense — BE - Y

. Modal
HAVE
SC: X - jense(Tense + Neg) — BE - Y
. Modal.
SA: X - Tense - MV - Y
 Relations between sentences 127

SO. X - Tense(Tense + Neg) - MV - Y

i X and Y are any arbitrary constituents.

ii Neg is adjoined as a daughter of Tense, and a
right sister of whatever is dominated by Tense.

51 Declarative => Interrogative (optional)

'HAVE'
SA: NP — Tense + BE - X
Modal,
HAVE'
SC: Tense + be ; NP-X
. Modal.
SA: NP - Tense - MV - X
SC: Tense - NP - MV - X

i X is any arbitrary constituent.

ii Tense and have, be or any Modal verb are
moved to the front of the sentence. If Tense is
immediately followed by the main verb, Tense
alone will move to the front of the sentence.

52 DO support (obligatory)
SA: X - Tense - Y
SC: X — Tense + DO — Y
i X and Y are any arbitrary constituents or none
(in the case of X), on condition that the left-most
constituent of Y is not a verb, main or auxiliary.
ii do is introduced as a right sister of Tense when
Tense is thus ‘stranded’ with no verb to its
immediate right.
There are a number of small changes in these ‘derivational’
rules as compared with the corresponding ‘relational’ rules in
49-51. Most of them are necessary precisely because these new
rules are derivational, and we must specify the derived con¬
stituent structure in detail. Note that we have also formulated
‘DO support’ as a separate rule, rather than including it in the
‘Negative’ and ‘Interrogative’ rules as previously. This enables
us to generalize about the circumstances that require ‘DO sup¬
port’: whenever Tense is ‘stranded’ - i.e. separated from an
immediately following verb, main or auxiliary - do is supplied
to ‘carry’ the tense marker. Making ‘DO support’ a separate rule
128 Part one: Constituent structure
means we can then use this rule (with some small further mod¬
ification) for other circumstances in which DO occurs. For
example, we find DO in negative imperative sentences:

53a Go!
53b Don’t go!

in some types of emphatic sentence (italics indicate stress):

54a John is going (John’s going)

54b John is going
55a John went
55b John did go

and in ‘Tag’ questions:

56a John is going, isn’t he?

56b John will go, won’t he?
56c John went, didn't he?

The description is clearly not yet complete (even in its outline

form!). We need rules that relate the various elements we have
been juggling with to actually occurring word forms or segments
of word forms. This subject is tackled in Part Two, and we leave
further discussion until then.

The discussion shows that transformational rules potentially

form an important and powerful addition to our descriptive
armoury. They enable us to relate together in a systematic fash¬
ion those sentences we perceive to be structurally related, and
the relationship is formalized within the grammar itself.
We close this chapter by observing that we can call upon
transformational rules to solve a descriptive problem noted on
page 75, which we were not then able to account for - the
alternative positions open to sentence adverbials. We saw that in
sentences like:

57a The dog bit the man yesterday

57b Yesterday the dog bit the man

the time adverbial could occur either at the end of the sentence
(its ‘normal’ position) or at the beginning. A constituent struc¬
ture rule like:

58 S —> (Adv ) NP + VP (+Adv . )

time7 v time'
 Relations between sentences 129
allows us to generate an adverb either at the beginning or the
end of the sentence. It also allows two adverbs, which we do not
want:

59 *Yesterday the dog bit the man last week

To make a rule like 58 work we need additional notation to the

effect that Advtjme was to be chosen once only. A more
illuminating solution is a constituent structure rule:

60 S -* NP + VP (+Adv . )
v time

and then a transformational rule optionally fronting the adverb:

61 Time adverb fronting (optional):

SA: NP-VP-Adv.
time
SC: Adv. -NP-VP
time

Similar transformational rules will account for the distributional

possibilities of the other adverbs discussed on pages 73-5.

Technical terms

active negative
affirmative passive
aspect passive auxiliary
auxiliary verb perfective auxiliary
base component phrase structure
basic sentence progressive auxiliary
declarative structure analysis
derivation structure change
derived structure syntactic surface structure
DO support tense
interrogative transformation
main verb underlying structure
modal auxiliary verb verb group

Exercises

In these exercises, you are asked to proceed by stages.

The first stage is to examine some data that can be described
in terms of a transformational rule insofar as there are formally
130 Part one: Constituent structure

describable structural relationships between sentences, and to

attempt to formulate these relationships in ordinary language.
This involves assigning the appropriate constituent structure to
the data in question so that the statement can deal with
structures of the maximum generality.
The second stage attempts to formulate the rules as
transformational rules of the sort discussed in the preceding
chapter. This usually involves having to decide on one structure
as a ‘basic’ structure from which the other structures are to be
derived: you need to find reasons why one rather than another
structure is to be considered ‘basic’ and be prepared to justify
them. Then you can write a constituent structure grammar
which generates the basic structures.
Finally you need to formulate transformations that operate on
the structures generated by the constituent structure grammar to
produce the ‘derived’ structures.

1 Dative movement

Consider pairs of sentences like:

la Mary baked John a cake

lb Mary baked a cake for John
2a Jane sent Simon a Valentine
2b Jane sent a Valentine to Simon

Describe the structural relationship between these sentences.

Which sentence would you choose as ‘basic’? (Hint: think of the
problem of specifying the right preposition - to or for.)
Now consider the following extra data:

lc *Mary baked John it

Id Mary baked it for John
2c *Jane sent Simon it
2d Jane sent it to Simon

Describe the restrictions on ‘dative movement’ introduced by

the pronominalization of a cake in 1, etc. Does your original
hypothesis about which sentence is ‘basic’ still hold?
Now write a grammar which generates the basic sentences
(and others which operate syntactically like them) and an
appropriate transformational rule. Write two grammars. In one
 Relations between sentences 131
suppose that pronouns (it, him, etc.) are introduced into
structures by a constituent structure rule, perhaps of the
following sort:
Det + Nl
NP
Pro )

and consider what restrictions you must place on your

transformation ‘dative movement’. In the other grammar
suppose that pronouns are introduced ‘transformationally’, so
the constituent structure rules generates NPs by a rule like:
NP^ Det + N
(i.e. they do not introduce pronouns). There will be a
transformational rule which we may informally describe as:
‘replace some NP by an appropriate pronoun: he, if the
replaced NP refers to a single male human; she if it refers to a
single female human’, etc. What restrictions do you now need to
introduce on your rule of dative movement?

2 Phrasal verbs

Consider sentences like:

la The woman rang up the policeman

2a The policeman stood near the hedge

Note that la is related to sentences like:

lb The woman rang the policeman up

lc The woman rang him up
Id *The woman rang up him

and 2a is related to sentences like:

2b *The policeman stood the hedge near

2c *The policeman stood it near
2d The policeman stood near it

Describe the structural relationships that hold between the

different sets of sentences. Decide which sentence in each set
you consider ‘basic’, and why. Describe the restrictions on
‘particle movement’ (as the transformation involved is usually
called).
132 Part one: Constituent structure

Write a grammar which generates these sets of sentences —

and others like them (collect a set of verbs that operate in each
structure). As with the previous exercise, you can write a
grammar assuming that pronouns are generated either in the
constituent structure rules, or are introduced by transformation.

3 Tag questions
Tags are constituents like those which follow the comma in the
sentences below; we call the constituent that precedes the
comma the main sentence:

1 You are coming, aren’t you?

2 John has been stupid, hasn’t he?
3 Mary will be coming, won’t she?
4 You can’t understand, can you?
5 Mary won’t be coming, will she?
6 The shops haven’t closed already, have they?

Assume that the main sentence is the basic form, and that a tag
is derived by rule from the basic sentence. Formulate a rule for
the derivation of tags. At this stage ignore sentences that do not
contain an auxiliary verb (i.e. sentences like Mary came), or that
contain an auxiliary verb other than HAVE, BE, WILL,
WOULD, CAN, COULD, SHALL, SHOULD.
Now consider the formation of tags in sentences like:

7 He likes hamburgers,_?
8 Your mother died in 1969,_?
9 You don’t like punk rock,_?
10 John didn’t break the window,_?

Extend your first description to include such sentences.

In addition to the auxiliaries listed above, the following are
often considered to be auxiliaries in English: may, might,
OUGHT TO, USED TO, NEED TO, HAVE TO. How do these
function in your dialect? (Dialects of English differ as to whether
they categorize all or only some of these items as auxiliaries or
as main verbs for the purposes of tag formation.) Incorporate
your findings into your statement about tag-formation.
Now consider the formation of tags to sentences which
contain have as a possessive verb, or BE:
 Relations between sentences 133
11 John has a big house,_?
12 Mary is pretty,_?

and incorporate this into your description.

Next consider some other uses of have, e.g.:

13 You had a good time at the party last night,

_?
14 Mary had a bath this morning,_?

Now write a grammar including a transformational rule (or

rules) to account for the tags you can form in your dialect of
English (you will probably find it useful to use the analysis of
auxiliaries developed in the preceding chapter).
Your rules will not account for all possible types of tag. For
instance, they will only handle ‘reversed polarity’ tags like those
illustrated. (The system of negation, with the terms ‘positive’
and ‘negative’ is often referred to as one of ‘polarity’: the tags
you have been looking at are all reversed polarity tags because
the polarity of the main verb is reversed in the tag.) They will not
handle non-reversed polarity tags like I’ll open the window, shall
I? Nor will they handle tags like I think it’s Monday, isn’t it (note
that *1 think it’s Monday, don’t I? is deviant!).
9 Embedding and recursion

Sentences like:

1 Charlie married Mary

2 The rumour that Charlie married Mary is untrue
3 Fred believes that Charlie married Mary

can be analysed as in Figures 63-5.

NP VP

PN

Figure 63 Charlie married Mary

Figures 64 and 65 contain, as a constituent, a structure identical

to the structure shown in Figure 63, with the same analysis as
the independent sentence 1 but preceded by the
‘complementizer’ that. In Figures 64 and 65, the structure
dominated by the topmost S is called the ‘superordinate’ or
‘matrix’ sentence, and the structure dominated by the lower S
the ‘subordinate’ or ‘embedded’ sentence. Embedded sentences,
then, are sentences which are themselves constituents of other
sentences.
In the structures under discussion, the embedded sentences
are analysed as constituents of an NP in the matrix sentence.
This analysis can be supported by noting their distribution in
transformationally related sentences, for example:
 Embedding and recursion 135

VP

Adj

is untrue
Figure 64

Fred believes that Charlie married Mary

Figure 65

4a What is untrue is the rumour that Charlie married

Mary
4b That Charlie married Mary is believed by Fred

In both 2 and 3 the embedded sentence is preceded by that.

136 Part one: Constituent structure

We call such items ‘complementizers’ (Comp), as in the analysis:

and the particular type of embedding illustrated is called
‘complementation’. One of the functions of complementizers is
to indicate that the sentence structure immediately following
is a subordinate or embedded sentence. Two types of
complementation are illustrated: Noun complementation, as in
2, where the embedded sentence is the complement of the noun
RUMOUR; and Verb complementation, as in 3, where the
embedded sentence is the complement of the verb believe.
Not all nouns can co-occur with complement sentences:

5 *The house that Charlie married Mary is unsafe

and we can indicate which nouns can take complements by

including an appropriate subcategorization frame (page 55) in
the lexical specification of these nouns, like rumour, story,
fact, insinuation, etc, as follows:

6 rumour: N;_S

Similarly, not all verbs can take complement sentences; we do

not find

7 *John killed that Charlie is married

For those verbs that do take complementsentences, like

believe, expect, think, SUPPOSE, etc., we can, once
again, indicate this fact in the lexical specification:

8 BELIEVE: V; _ NP[S]

where the strict subcategorization frame_NP[S] indicates that

the object of believe can be a complement sentence.
To make the constitu6nt~structure rules account for sentences
like those in Figures 64 and 65 we need rewrite rules of the
form:
NP+ VP
V (+NP)
f Art + N (+S)) (for structures like Figure 64)
(S J (for structures like Figure 65)

In these rules the rule expanding NP reintroduces the initial

symbol S. To expand this S we need to return to the beginning
of the grammar and go through the rules a second time. This
‘cycling’ through the rules is known as recursion. Recursion in
 Embedding and recursion 137
the grammar permits embedding. It will be noted that the rules
in 9 do not account for the introduction of the complementizer
that. For our immediate purposes we can suppose that this will
be introduced by a transformational rule.
The subject of verb complementation in English is complex
and we do not consider it in this book. Those interested in
pursuing the subject are referred to any of the books on
transformational grammar mentioned in ‘Further reading’.
Instead we now turn to consider a different embedded
construction within the NP usually referred to as ‘relativization’.
This describes the formation of ‘relative clauses’. These can be
illustrated by
10 The woman who you saw me with last night is my wife
of which the italicized portion is the relative clause. The whole
string the woman who you saw me with last night is an NP, as
can be verified by checking it in appropriate transformations;
and the relative clause is clearly a part of this. The relative
clause cannot stand on its own as a sentence:
11 *who you saw me with last night
yet it clearly relates to the sentence:
12 You saw me with the woman last night
which has all the properties of an independent sentence. Let us
suppose that the structure underlying 10 can be represented as
in Figure 66. The structure to be related to the relative clause is
represented as a sentence embedded within the NP. It has been
generated by a constituent structure rule, which can be added to
the rules in 9:
13 NP —» NP + S
All relative clauses are derived from this rule. Note that this
rule is distinct from the rule in 9 that introduces noun
complement sentences. Since the syntax of noun complement
sentences and relative clauses is different, we need a different
rule for each type of structure. Rule 13 develops a structure as
shown in Figure 66, the underlying structure (see page 124) for
sentence 10. We call the N P the woman in Figure 66 the ‘head’
of the construction and the S from which the relative clause
derives the embedded or relative sentence.
138 Part one: Constituent structure

T tie woman You saw me with the woman last night is my wife

Figure 66

We can derive the sentence 10 by the rule shown in 14. The

rule is formulated as a derivational transformational rule.
14 Relative clause rule: (obligatory)
SA: np(nps(x-np-y»

scWNPs([p™] "X"Y))
i Identify the NP in the embedded S that is
identical to the head.
ii Move this NP to the front of the embedded S.
iii Replace it by the appropriate relative pronoun.
rel
The symbol indicates ‘the appropriate relative pronoun’:
pr°_
who if the NP to be replaced contains a ‘human’ noun; which
for all other nouns. X and Y, as before, indicate any arbitrary
constituents, or none. In the case of ‘subject relatives’, X is null
(i.e. the NP in the relative clause is the first constituent in its
clause so X must be null); in the case of ‘object relatives’, X is not
 Embedding and recursion 139
null (it includes the subject NP and the verb at least), but Y
may be null, since the object NP may be the last constituent in
the clause (the object need not, of course, be the final
constituent, as it isn’t in 10: in that case neither X nor Y is null).
Subject and object relative clauses are illustrated in 16. The
reader should try to derive the items in 16 using the rule 14.
Applying rule 14 to the relevant string in Figure 66 gives:
15 Figure 66: the woman (you saw me with the woman
last night) is my wife
Rule 14 i: the woman (you saw me with the woman
last night) is my wife
Rule 14 ii: the woman (the woman you saw me with
last night) is my wife
Rule 14 iii: the woman (iwho you saw me with last
night) is my wife

This produces the desired result. The rule 14 predicts the form
of infinitely many relative clauses:
16a The apples (the apples are ripe)
The apples which are ripe
16b The man (the man lives next door)
The man who lives next door
16c The girl (I love the girl)
The girl who I love
16d The knife (she stabbed her lover with the knife)
The knife which she stabbed her lover with
16e The girl (John gave the book to the girl)
The girl who John gave the book to

etc. The rule applies whether the NP which is identical to the

head is the subject (16a,b), the object (16c), the indirect object
(16e), an NP in an ‘instrumental phrase’ (16d), etc.
We can now go a stage further. In 16d,e the NPs in the
embedded S that are identical with the head are constituents of
a PP (with the knife; to the girl). In the examples shown the
NPs have been lifted out of their PPs and moved to the front of
the embedded S. It is, however, possible to take the preposition
as well:
17 The knife with which she stabbed her lover
The girl to whom John gave the book
140 Part one: Constituent structure
This suggests that we should amend 14 as follows:
14' Relative clause rule: (obligatory)
SA: (NPS(X- NP-Y))
NP

SC:
NP (NPs<P - X“Y»

SA: ,(NPS(X — pp(PreP - NP) - Y))

NF

SC: (a) Np(NPs(Prep-M -X -Y))

(b) Np(NP rd 1 - X - Prep -Y))

L pr° J

i Identify the NP in the embedded S that is

identical to the head.
ii Move this N P to the front of the embedded S
(and, optionally, if the NP is in construction with
a preposition in a PP, also move the preposition).1
iii Replace the NP by the appropriate relative pronoun.

1 We need to include the restriction that this rule applies only to cases where
the preposition is part of a PP in order to avoid producing nonsense when we
deal with ‘phrasal verbs’. Phrasal verbs are items like run up in the sense
‘amass, accumulate’ or put down in the sense ‘kill’. We describe these as
having a structure verb stem + particle. We can then distinguish between items
like up and down when they are used as particles (in phrasal verbs) and when
the ‘same’ items are used in prepositional phrases like up the hill, down the
drain, etc. For example, given the structure:

the cat (we put down the cat last week) was very old
we want to avoid producing:
*the cat down which we put last week was very old
while still allowing for the acceptable:
the cat which we put down last week was very old
The distinction between a ‘phrasal verb’ and a verb + PP string can be
observed in the following-

the bill (John ran up the bill) was enormous run up = ‘amass’
the bill which John ran up was enormous
*the bill up which John ran was enormous
the hill (John ran up the hill) was enormous run + PP
the hill which John ran up was enormous
the hill up which John ran was enormous
 Embedding and recursion 141
We can go further still. In sentences 16c,d,e the relative
pronoun can be omitted without affecting the acceptability of
the sentences in question:
18 the girl (who) I love
the knife (which) she stabbed her lover with
the girl (who) John gave the book to
The relative pronoun cannot be omitted in 16a,b:
19 the apples which are ripe (need to be picked)
*the apples are ripe (need to be picked)
the man who lives next door (is the Lord Mayor)
*the man lives next door (is the Lord Mayor)
nor can it be omitted in the NP in 17:
20 The knife with which she stabbed her lover
*The knife with she stabbed her lover.
The general principle which governs the omission of the rela¬
tive pronoun may be formulated as:

21 Relative pronoun deletion rule: (optional)

SA:np<npsC] -Np-X»

SC: Np(NPs(NP-X))

The relative pronoun, who or which, may be deleted,

optionally, if
i it directly follows the head noun
and
ii it immediately precedes the subject NP of the em¬
bedded S.
Both of these conditions are satisfied in the sentences in 18, so
the relative pronoun can be deleted. In the sentences in 19 the
first condition is met, but the second is not, since the relative
pronoun has replaced the subject noun. In sentences 20 the sec¬
ond condition is met, but the first condition is not.
There are a number of reasons why it is attractive to derive
relative clauses in the way suggested by rules 14' etc. It accounts
for the fact that relative clauses, although not independent
142 Part one: Constituent structure

sentences, seem to have many of the characteristics of full

sentences: we can show this by supposing that they are derived
from full sentences, but that the rules alter them in various
ways. It explains why the pronoun who in 16b is a ‘subject rela¬
tive pronoun’ and the ‘same’ pronoun in 16c an ‘object relative
pronoun’ - they derive from the subject and object respectively
of the embedded sentence. It also accounts for the fact that
even though a verb like LOVE is a ‘transitive verb’ (i.e. one that
requires an object NP) no object NP appears following the verb
in the actual structure of a sentence like 16c. Without this way
of accommodating that information, a constituent structure
grammar must specify, for example, that LOVE is a transitive
verb, and must therefore co-occur with an object NP, except
when it is found in a relative clause in which the relative pro¬
noun is understood to be the object relative pronoun. Our deri¬
vation is more straightforward, and in addition accounts for a
variety of facts about relative clauses.
Let us now continue some lines of investigation that the
account so far suggests.
Consider the sentences:

22a The man (the man is scratching his head)

22b The man who is scratching his head
22c The man scratching his head

These sentences seem systematically related in a way that can be

matched by numberless other sentences of a similar kind:

23a The music (the music is on the piano)

23b The music which is on the piano
23c The music on the piano
24a The man (the man was run over by a bus)
24b The man who was run over by a bus
24c The man run over by a bus

In each case the c sentences omit the relative pronoun and a

form of the verb BE. We make a rule to cover this case:

25 Relative clause reduction rule (optional):

SA: np(NP S^[pro] ~ Tense ~ BE ~

SC:Np(NPs(Y))
 Embedding and recursion 143

i Optionally, the relative pronoun, tense and an

immediately following form of BE may be deleted.

It does not appear to matter whether the BE in question is the

BE found in a ‘locative’ sentence like 23; the be which is the
‘progressive’ auxiliary as in sentences like 22; or the ‘passive’
auxiliary BE as in sentences like 24. Note also that the rule only
deletes a form of BE - in 22c for example the -ing in scratching,
which derives from the progressive BE +-ing (as we noted
earlier, pages 107-8) is not affected by the rule: there is of
course no NP like *the man scratch his head where both the BE
and the -ing have been deleted.
As formulated, the rule also applies to the copula BE in NPs
like:

26 The apples (which are ripe)

For many adjectives, there is a relation between an adjective in

a relative clause, and the same adjective in a simple NP:

27 The ripe apples

The apples which are ripe

We can maintain that rule 25 does apply to these, and that an

extra rule is needed which would reverse the order of noun and
adjective. The derivation is as follows:

28a The apples (the apples are ripe)

28b The apples (which are ripe) (by rule 14)
28c The apples ripe (by rule 25)
28d The ripe apples

The step between 28c and 28d is accounted for by postulating

the rule:

29 Adjective shift rule: (obligatory)

SA: Np( NP(Art ~ N) S(Adj) )

SC: Np( Art-Adj-N)

i Obligatorily, in a structure of the form N + Adj reverse

order of these two constituents.

Handling the derivation of adjectives in this way has descrip¬

tive advantages over the constituent structure account of adjec¬
tives we discussed on pages 76-9, but it is not without
144 Part one: Constituent structure
problems. The rules do make interesting generalizations about
systematic structural relationships between sets of sentences. They
also go some way towards explaining their paraphrase relation -
they derive from a common underlying structure, later obscured
by transformational rearrangements. A further advantage is that
the rules simplify the account, however it is handled, of co¬
occurrence restrictions between adjectives and nouns. Instead of
having to state that, say, ripe can modify a noun like apple
both when it precedes it (the ripe apple) and when it follows it
in a relative clause (the apple which is ripe), a single statement
suffices. If the apple is ripe is semantically well formed, then so
too is the ripe apple. By contrast, if the apple is careless is ill
formed, then so too is the careless apple. Yet another advantage
is to simplify further the constituent structure rules. We can
replace rule 30a by 30b, since both adjectives and prepositional
phrases within the NP are derived from ‘reduced’ relative
clauses.

30a NP —» Art (+Adj)* + N (+ PP)

(see rule illustrated in Figure 31, page 80)

I Art + N
30b NP
(np + s
We do, however, have to ‘pay for’ this simplification. To
begin with there is the additional machinery of transformational
rules. We also note that not all adjectives go through the deriva¬
tion proposed. Some do not appear in relative clauses:

31a The major difficulty

31b *The difficulty which is major
31c The same problem
31d *The problem which is same

and conversely some can appear in relative clauses but not pre¬
ceding nouns:

32a *The alike girls

32b The girls who are alike
32c *The ablaze building
32d The building which is ablaze

We could perhaps handle such cases by marking items like

MAJOR in the lexicon as obligatorily having to undergo rules 25
 Embedding and recursion 145
and 29, and items like ablaze as being unable to go through
these rules, but this seems a somewhat ad hoc solution.
A wide range of structures in English involve embedding. We
do not have space to examine more in detail, but some are
worth mentioning. We find embedded sentences in temporal
adverbials:

33 When you go to London you should visit the National

Gallery

in place adverbials:

34 You will find your book where you put it down last
night

in manner adverbials:

35 He walks in a manner that reminds me of his father

in adverbial clauses of reason and purpose:

36 You cannot have another glass of wine because the

bottle is empty
He emigrated to Australia in order to escape from his
mother-in-law

and in conditional clauses:

37 You should go and live in France if you want to learn

to speak French well

We can also use the approach outlined in this chapter to

explain certain sorts of ambiguity. We noted on page 83 that the
ambiguity in sentences like:

38 The shooting of the hunters was disgraceful

cannot easily be accounted for in terms of simple labelling and

bracketing. The ambiguity arises from the fact that the shooting
of the hunters can have a ‘subjective’ interpretation (the hunters
shot ‘something’) or an ‘objective’ interpretation (‘somebody’
shot the hunters). If we derive ‘nominalizations’ like the shooting
of the hunters from sentence-like structures of the sort indicated,
then we have a systematic account of such ambiguities.
Embedding is one of the most remarkable features of lan¬
guage. It enables a speaker to construct not only simple sen¬
tences, but also sentences of considerable syntactic complexity.
146 Part one: Constituent structure
Our grammar, however it is formulated, needs to be able to cap¬
ture this potential. Transformational grammar claims that it
enables the grammarian to capture this ‘creative’ power in a
language. The constituent structure rules cannot only generate
simple sentences; they can also, by the use of recursive rules
(which enable us to cycle again and again through the same
small set of rules) generate sentences containing complex
embedding. A small set of transformational rules, operating on
these embedded sentences as many times as is necessary will
produce the complex sentences we wish to account for.

Technical terms
‘adjective shift’ recursion
complementation relative clause
complementizer relative clause reduction
constituent sentence relative clause rule
creativity relative pronoun deletion rule
embedding subordinate sentence
head matrix sentence superordinate sentence
noun complementation verb complementation

Exercise

Relative clauses in English

We noted in the text that relative clauses can be regarded as
being derived by a set of rules from an underlying structure
which contains a ‘full’ sentence. We looked at four such rules.
To save space we can represent the effect of these rules in terms
of bracketed strings which ‘chart’ the derivation of an NP. Thus
the derivation of 16a can be shown as:

Underlying structure: The apples {the apples are ripe)

Relative clause rule: The apples (which are ripe)
18 can be charted as:
Underlying structure: The girl (I love the girl)
Relative clause rule: The girl (who I love)
Rel. Pro. deletion rule: The girl (Hove)
 Embedding and recursion 147
and 28 can be represented by:

Underlying structure: The apples (the apples are ripe)

Relative clause rule: The apples (which are ripe)
Rel. Cl. reduction rule: The apples (ripe)
Adjective shift: The ripe apples

Check that these derivations are correct in terms of the

description in the text.
Note that some of the other transformations looked at on
pages 102-33 can be applied to the embedded sentence before
it undergoes the relativization rules. Thus, for example, consider
the following derivation:

Underlying structure: The man (a bus ran over the man)

Passive to embedded S: The man (the man was run over
by a bus)
Relative clause rule: The man (who was run over
by a bus)
Rel. Cl. reduction rule: The man (run over by a bus)

Write derivations like the above for the following NPs:

i The music on the piano

ii The girl who won the tennis match
iii The book I bought yesterday
iv The criminal imprisoned by the authorities
v The guilty man
vi The knife father cut the turkey with
vii The woman to whom he is married

The text discussed relative clauses formed with the relative

pronouns who and which. In addition we encounter relative
clauses with the ‘genitive’ relative pronoun whose:

the man whose leg was broken

the man whose dog I rescued from the river

If we derive these also from underlying structures containing a

full sentence, then they have underlying structures like this:
The man (the man’s leg was broken)
The man (I rescued the man’s dog from the river)
Write a rule like those in the text which will derive such
sentences, and then write derivations for the following NPs:
148 Part one: Constituent structure

viii the woman whose dog died

ix the boy whose mother I gave it to
x the people whose cars have been stolen

In most forms of English, in addition to relative clauses

formed with a relative pronoun (who, which etc.), you find
relative clauses formed with that'.

The girl that I love

A man that I would like to meet

The rules for wh- and that relative clauses are slightly different,
and to indicate this we distinguish between a 'Wh-rel rule’ and a
‘That-re\ rule’

Underlying structure: The girl (I love the girl)

That-rel rule: The girl (that I love)

See if you can formulate the differences between the two types
of relative clause forming rules by considering the different rela¬
tive clauses you can form from the following underlying struc¬
tures:

xi The girl (I gave the book to the girl)

xii The man (I envy the man’s cellar)
10 The sentence

Lyons defines the sentence as the maximum unit of grammatical

analysis: ‘A grammatical unit between the constituent parts of
which distributional limitations and dependencies can be
established, but which can itself be put into no distribution class’
(1968, 173), and we adopt this definition. In other words,
formal statements can be made about the distribution of
sentence constituents, but not about sentences as wholes. Thus,
we can talk about the distribution of NP (in environments like
_VP; V_; Prep _ etc.) but no comparable distributional
statements can be made with respect to the sentence itself,
except, perhaps, that a text consists of a series of sentences S &
S & S . . . 7
Lyons’s definition implies that the sentence has a certain sort
of unity: it is grammatically complete; it can stand on its own
independent of context; and it has a degree of semantic
independence. This is largely true of the units so far identified
as sentences and used as examples. What, however, of the
relationship between such units and, say, the orthographic
sentences out of which this chapter is composed? Lyons suggests
that no distributional statements can be made about such units -
a text it is suggested consists of a series of sentences S & S &
S . . . . Consider, for example, the three orthographic sentences
that form the preceding paragraph. The second and third begin
with In other words and Thus, expressions which tie the relevant
sentences to their predecessors. In other words indicates that the
1 This is a well-established descriptive viewpoint with a long history. Consider
the following quotation from Hermes, by James Harris (1751): ‘The Extensions
of Speech are quite indefinite, as may be seen if we compare the Eneid to an
Epigram of Martial. But the longest Extension with which Grammar has to do is
the Extension here considered, that is to say a sentence. The greater Exten¬
sions (such as Syllogisms, Paragraphs, Sections and complete Works) belong not
to Grammar but to Arts of a higher order; not to mention that all of them are
but Sentences repeated.’
150 Part one: Constituent structure
sentence is a rephrasing of the point made in the sentence
before: Thus introduces an example of the point just made. We
call such expressions ‘binding expressions’ since they bind
sentences together into text. Without the binding expressions
the three sentences could be rearranged in some other order -
3—2—1; 1-3-2; 2-1-3, etc. - with no substantial difference to the
message conveyed, and with few other consequential
amendments to the text (other binding expressions, or none,
would be needed). Furthermore, if we disregard the binders,
each sentence can stand on its own as a grammatically complete
unit, and forms, in some sense, a complete message.
Orthographic sentences in texts are not always capable of
such ready permutation. Consider, for instance, the following
(read it as a text - the numbers are for reference):

1 When John and Mary got back home, they found the
front door had been left open. 2 John accused Mary. 3
She denied it. 4 John asked her to shut it. 5 She
wouldn’t.

The first sentence is grammatically and semantically capable of

standing as an isolate, but none of the others is. If the second
sentence were found as an isolate, one might reasonably
enquire: ‘What did John accuse Mary of?’ In an isolate sentence
ACCUSE requires a sentence frame NP_NPPP, where the
subject NP is the ‘accuser’, the object NP is the ‘accused’ and
the PP contains the ‘accusation’. In connected text, however,
ACCUSE permits the ellipsis or omission of the ‘accusation’,
providing the ‘accusation’ is recoverable from the context, as in
the example. Similarly if 3 were found as an isolate, we might
enquire: ‘Who denied what?’. The proforms she and it are
understood in context as, respectively, ‘Mary’ and ‘that she left
the door open’. 4 further illustrates the use of proforms. 5 is
another example of ellipsis - here understood as ‘Mary wouldn’t
shut the door’.
The sentences 2-5 cannot stand alone. They are to a greater
or lesser degree either grammatically and semantically
incomplete or both: grammatically incomplete insofar as
constituents have been elided; and semantically incomplete
inasmuch as the sentences can only be understood with
reference to their context. In both cases, these ‘sentence
fragments’ are only fully understood by reference to some
 The sentence 151
‘understood’ fuller form, where the items that have been elided,
reduced to proforms, etc., are restored:

6a John accused Mary of leaving the door open

6b Mary denied that she had left the door open
6c John asked Mary to shut the door
6d Mary wouldn’t shut the door

The amount of grammatical and contextual interpretation

necessary can be appreciated if we consider, for example, 5.
Only in this particular context is it understood as 6d. In another
context it might be understood as ‘Mary wouldn’t feed the cat’;
‘Jane wouldn’t give up smoking’; etc.
Strings like 2-5 are entirely appropriate and natural in a text.
By contrast the series of sentences l,6a-d do not form a text
under any normal interpretation of text. On the other hand each
of l,6a-d can stand alone as a well-formed grammatical
sentence, and each is semantically self-contained. These
sentences can be re-ordered to form a different text, perhaps:

7 John accused Mary of leaving the door open when they

got back home and found it like that. 8 She denied it,
but wouldn’t shut it when John asked her to.

This time the ‘accusation’ is restored to ACCUSE: it is hardly

possible in this context to elide this constituent since there is
now no context from which it would be recoverable. Wouldn’t
has, this time, lost its subject, but gained its main verb, shut, the
object of which now turns up as a proform; and asked has lost
its complement which must now be recovered from the context.
Different arrangements of the ‘understood’ sentences have
different consequences in terms of elision, pronominalization,
etc., and these processes produce sentence fragments, incapable
of standing alone as sentences and relying for their
interpretation on context and on the ability of the reader to
reconstruct the ‘understood’ sentences to which they relate.
Such processes can create sentence fragments that relate to
almost any part of a sentence. Consider, as well as the examples
in 2-5, 7-8, the relevant strings from the examples below, which
are to be understood as exchanges between two speakers:

9a Nobody wants to go to the pictures

9b Except John
152 Part one: Constituent structure

10a Anybody remember to feed the cat?

10b Yes, Kate did
11a What has John done now?
lib Broken another plate
12a Gill adores pastrami
12b Not adores, just likes

All of these might be appropriate exchanges, and the b

fragments comprehensible and appropriate in context; all of
them rely for their interpretation on the hearer’s reconstructing
the understood forms to which they relate.
The question arises: What is the status of sentence fragments?
If our grammar generates these directly as expansions of the
initial S symbol, that leads to an incoherent grammar, and
destroys the very foundations of the grammar. We can postulate
two sets of rules: firstly grammatical rules which generate full
sentences like 1, 6a-d and a set of ‘text formation rules’ which
produce, inter alia, sentence fragments. The rules allow for
the ellipsis of contextually recoverable constituents, for inter-
sentential proform formation, for the insertion of binding expres¬
sions, and so on. Text formation rules depend on linguistic and
non-linguistic context. The rules for the formation of sentences,
by contrast, are in principle independent of such constraints.
This is an attractive proposal, but it is far from clear whether
it can be accomplished in any straightforward fashion. Let us
look at the problems. We may begin by drawing a distinction
between simple and complex sentences. Suppose simple
sentences are those sentences that contain a single main verb,
sentences with the sort of structure examined in earlier chapters:

13 The dog frightened the cat

The man put the lamp on the table

Complex sentences are those sentences that can be analysed as

consisting of a number of simple sentences. We have already
considered (page 134) a sentence like:

14 John believes that Bill has married Mary

as having a constituent sentence (Bill has married Mary) embed¬

ded in a matrix sentence (John believes NP). The constitu¬
ent sentence is traditionally called a subordinate sentence,
and the general structure of complex sentences involving
 The sentence 153
S

Figure 67

subordination can be shown schematically as in Figure 67, where

the constituents of the matrix sentence are represented as X, Y, Z
and the subordinate sentence is shown as a constituent of one of
these constituents.
Many subordinate sentences bear overt markers of
subordination. Thus, complementizers like that in 14 are
markers of subordination, as are items like when and after in:

15a When Mary came into the room, John stood up

15b After entering the room, John took off his coat
15c John wanted Bill to go
15d I don’t like your smoking so much

In 15 the strings deriving from embedded subordinate sentences

are italicized. There are not only overt markers of subordination
in complementizers (that) and subordinating conjunctions (when,
etc.), but also grammatical constructions only open to subordinate
strings: the ‘infinitive construction’ (Bill to go), and the ‘gerund
construction’ (your smoking so much) (the derivation of these
constructions is not considered further in this volume.
The distribution of these subordinate constructions can be
accounted for in terms of the distribution of the sentence
constituent into which they are embedded. Thus NPs with
sentences embedded in them operate much like other NPs (see
page 135; similarly temporal subordinate sentences (e.g. 15a,b)
form a class with adverbs of time and are distributed like them
(cf. page 74).
Subordinate sentences may be important for the strict
subcategorization of various lexemes - we have already noted
that nouns like rumour and verbs like believe are
subclassified precisely in terms of the fact that they can occur
with complement sentences.
154 Part one: Constituent structure

Finally, complex subordinate sentences cannot typically be

resolved into a string of separate sentences without doing some
violence to the dependency relations between the various
constituents involved in a single sentence. This is obviously the
case with complex sentences like:

16 John wanted Bill to go

which do not resolve into anything like:

17 *John wanted. Bill was to go

or even
18 John wanted something. Bill was to go

since this breaks the relation of dependence: in these forms it is

not necessarily the case that ‘what John wanted’ was for ‘Bill to
go’. The same argument can be applied, though with perhaps
less force, to examples like:

19a Mary left the room because John started talking

politics
19b Mary left the room. John started talking politics
19c John started talking politics. Mary left the room

In 19a the causal relation is explicit. In 19c the relationship may

be causal, but it may equally well not be. In 19b, even if the
relationship is considered causal, and it need not be, the
causality is ‘reversed’.
The discussion leads to a number of conclusions. Our
grammar needs to account for the formation of such sentences
as those in 15 - it is not satisfactory to derive them by the use
of text formation rules of any sort. The grammar must contain
rules that account for strings that often resemble sentence
fragments: a point we will return to.
First, however, consider another type of complex sentence,
illustrated by:

20a Mary came into the room and John stood up

20b Jane doesn’t buy books, but Mary buys a book
whenever she passes a bookshop
Such sentences are ‘co-ordinate complex sentences’, and forms
like and and but are ‘co-ordinating conjunctions’. The structure
of co-ordinate sentences can be shown schematically as in Figure
68. Comparison with Figure 67 shows That in subordinate
 The sentence 155
S

S and S

Figure 68 Z_A Z_A

constructions the subordinate sentence is embedded in the

matrix sentence, but in co-ordinate constructions, the
co-ordinate sentences are sisters: one sentence is not sub¬
ordinate to the other. In principle, then, each of the conjunct
sentences should be grammatically independent. Disregarding
the conjunctions, this is true of the sentence in 20, which can be
resolved into two separate sentences:

21 Mary came into the room. John stood up

Now this is not the case for all co-ordinate sentences. Consider,
for example, the following:

22a John wanted Harry to come and Bill to go

22b John uses a fountain pen when writing to his mother
and when writing letters to the press
22c John asked Mary to shut the door but she wouldn’t

In each of these the second conjunct is a string identical to the

sort of string seen earlier in the chapter as sentence fragments.
We now find ourselves on the horns of a dilemma. One
uncomfortable horn prods us into allowing that sentences like 22
should be generated by our grammar, so the grammar will have
rules for producing sentence fragments. The other equally
uncomfortable horn goads us into proposing that co-ordinate
sentences are formed by text formation rules and not by rules of
the grammar at all!
The former position entails an initial rule for the grammar of
the sort:

23 S -*S*

(where S* is to be understood as allowing expansions S, S & S,

S & S & S, etc.: a similar rule for adjectives was proposed
on page 78). This rule enables sentences to be generated of
infinite length, identical to texts, except that where texts are
typically subdivided into orthographic units the units are simply
conjoined by conjunction. Children often produce texts of this
156 Part one: Constituent structure

sort (We went to the seaside last Sunday and we played on the
sands and Daddy bought us an ice cream and . . .). This
approach also allows us to treat binding expressions (of the sort
discussed at the beginning of this chapter) as a form of
conjunction, which in many cases they are! But the approach
does have problems. We cannot make distributional statements
about constituent sentences produced by such a rule since the
rule produces no distinct environments we can use. We also
need to postulate some set of rules to accommodate sentential
sequence, and these rules must take account of matters that go
beyond the simple sentence as we have so far considered it.
Rules relating to various matters of word order are taken up again
in Part three.
The arguments in the preceding paragraph lead to the
conclusion that conjoined sentences are not to be considered
part of the grammar at all, but are derived by text formation
rules. This conclusion is not wholly comfortable, but the logic of
the argument seems to force us into this position. It at least has
the advantage that it permits us to adopt the definition of the
sentence with which we began the chapter: the sentence is an
abstract unit, established in order to account for distributional
regularities of its constituents. The question of whether there
are further distributional regularities between sentences in some
larger unit, say the paragraph, is discussed in Part three.

Technical terms

co-ordinate sentence text formation rules

co-ordinating conjunction subordinate sentence
sentence subordinating conjunction
sentence fragment

Exercises

The Introduction suggested that much of the data actually used

by linguists to establish grammar is ‘regularized’. In informal
conversation we do not typically speak in units that can easily
be recognized as grammatical sentences. Here is an extract from
a transcription of part of a radio phone-in discussion on police
pay. The speaker is a serving policeman. The items in brackets
indicate interpolations from the Chairman of the programme.
 The sentence 157
The symbol + indicates a pause. The symbols eh um indicate
‘filled pauses’.

First of all + I would say that + I don’t think any (1)

policeman wants a + wants a medal for + for eh the
profession which he’s in + (No) every person + eh
decides + how his eh life is going to + be run (um hum)
We we’d work seven days shifts we’d be going to have (5)
our two days off (Yes) + and + I don’t think + eh +
the people realize about shift allowance in England we
will receive no shift allowance for this (um hum) um I
feel that the press eh eh in a lot of cases give us a bad
publicity (Yes) +eh the article in fact that you have (10)
written + you + read this morning about the + the
Strathclyde eh (Yes) the press reported that + the +
this would + mean a loss of wages about the overtime
+ about thirty five pounds a week (Yes) well in our
local force here in the county area + eh we’re not (15)
actually allowed to be paid for overtime we’re told that
if we want to do overtime then we must + take time
off for it.

1
Make a transcription of the text for publication in a periodical.
You should make sure that your transcription follows the usual
orthographic conventions about punctuation, etc. and that you
write what, in the orthography, are considered to be
grammatical sentences.

2
Make a note of what sort of things you have left out of your
‘cleaned up’ version (pauses, filled and unfilled; repetitions;
false starts; . . .) and of the sort of things you have found it
necessary to add. See if you can formulate these as a set of
instructions to a secretary who was asked to do a similar task.

What is the relationship between the sort of units that are

usually held to be sentences and any sort of unit you can discern
in the text?
Part two

Morphology
11 Words and morphemes

Hitherto the word has been the smallest constituent with which
we have been concerned. It is clear, however, that many words
are susceptible to further analysis. Thus in the sentence:

1 The cats attacked the dog

the words cats and attacked can usefully be further analysed as

cat-s and attack-ed. In other words, words themselves appear to
have a constituent structure.
The study of the internal structure of words is called
morphology. We do not, however, regard morphology as an
independent study, but rather as a bridge between the syntax of
a language and its phonology. It is part of syntax insofar as the
syntactic study of a language involves the labelling and
bracketing of sentence constituents. Since we have regarded
words as constituents, the study of the constituent structure of
words is logically part of the study of syntax: there is, in
principle, no difference between the analysis of the structure of
the noun phrase, and the analysis of the structure of the noun.
This chapter extends our grammatical description beyond the
level of the word, down to the smallest constituents a grammar
seeks to account for. On the other hand, morphology is
connected to a phonological study of a language insofar as it
involves a study of the phonological shapes of words. In this
sense morphology cannot be regarded as an independent study,
but rather as a bridge between syntax and phonology.
Why, then, should morphology be treated separately? This
question has to do with the status of the word as a linguistic
unit. Literate people in western cultures do not in general have
any difficulty in recognizing the words in their language.
Children as they learn to read quickly come to recognize the
established practice of the orthography, and for adults the word
seems to represent a ‘natural’ unit. To be sure there are
162 Part two: Morphology
marginal difficulties. In some cases it is not clear whether we
should write one word or two (all right or alright?)-, in the case
of ‘compound words’ like windmill, water-mill, flour mill there is
some indeterminacy as to whether these should be written as a
single orthographic unit, a hyphenated unit or as two discrete
units - an indeterminacy reflected in the practice of dictionary
compilers. On the whole, however, literate societies do not find
word divisions a great difficulty. In illiterate societies, speakers
seem capable of segmenting their utterances into ‘word-sized’
units; when such languages are eventually reduced to writing
there is generally, once more, agreement about the size of the
unit established as a word. These facts seem to indicate the unit
has some linguistic reality. It is not, however, easy to lay down
the criteria by which words are to be identified.
Three sorts of criteria are commonly called on - semantic,
grammatical and phonological. A semantic criterion would be
that a word is ‘a unit of meaning’. This belief is widely held, but
how can the definition be made to apply in any straightforward
way? For example, reheat is a word, yet it appears to have at
least two relatively independent units of meaning. The first is
the meaning attached to the prefix re-, perhaps something like
‘again’, and found in other words like rewrite and so on; the
second is the meaning associated with the stem heat, and also
found as a component of words like heater. So if reheat is a unit
of meaning, is not the string heat again also a unit of meaning,
though it is agreed to be a sequence of two words? The word
heat itself seems to have a meaning we can represent as ‘make
hot’, but make hot again is a string of two words. Clearly the
semantic criterion by itself cannot be made to stand up. On the
other hand, the notion that words do have a kind of intuitive
semantic unity is not totally to be discarded, particularly for
compound words. A blackboard, as used in the classroom, is
often a piece of ‘board’ which is ‘black’ (though with modern
technology this is less and less the case!), but the meaning of
blackboard as a unit word is more than a simple conjunction of
the meanings of the two stems from which it is composed, and is
different from black board, in the sense ‘a piece of board which
is black’. We shall not call upon a semantic criterion for defining
words, but it is not possible, and probably not desirable, to
eliminate semantic considerations altogether.
Phonological criteria are more satisfactory. A common
 Words and morphemes 163
criterion is that of ‘potential pause’: there is the potentiality for
pause between word boundaries, but not within a word. This
criterion indicates those points within an utterance where it is
possible to insert a ‘filled pause’ - of the form usually
represented orthographically as er or um. In general this works.
The reader should test it by reading, say, this paragraph in this
fashion. In most cases it is difficult to conceive of an er being
interpolated into what are orthographically represented as words
- poss-er-ible or fill-er-ed are unlikely. On the other hand the
criterion is not infallible. One may well find, say,
un-er-grammatical, particularly if a speaker were hesitating
between un-grammatical and un-acceptable. There is no
suggestion, of course, that speakers do in fact pause between
words in normal speech. On the contrary, particularly in rapid
speech, words are ‘run together’. This is sometimes recognized
in the orthography, regularly in forms like can’t, isn’t, and also
in forms like dunno (don’t know) wanna (want to). It is a
feature of those humourous books dealing with particular
accents of English (Jew asbestos to believe . . . Do you expect us
to believe. . .). Cases of this sort are usually recognized as
instances of two or more words being ‘fused’ together by quite
regular processes in rapid colloquial speech (see the discussion
in Brown (1977)). The notion of potential pause tries to capture
the generalization that if speakers are to pause within their
utterances, they are most likely to pause between what we
recognize as words. However, another difficulty is raised: it
suggests that the recognition of places at which a pause is
possible rests on a prior knowledge of word boundaries. This is
a valuable clue to the recognition of word boundaries, but is not
in itself criterial.
In some languages other phonological features correlate with
word units. Thus in some languages the placement of accent is
related to word structure. In Spanish, words ending with a
consonant are typically accented on the final syllable {avion
‘aeroplane’; general ‘general’; etc.), and words ending with a
vowel are typically accented on the penultimate syllable
(muchacho ‘boy’; bianco ‘white’; etc.). Other languages have
different patterns of accent placement (Czech words are
normally accented on the initial syllable; Turkish words on the
final syllable; etc.). Some languages have rules of ‘vowel
harmony’, usually associated with word units: in such languages
164 Part two: Morphology

the vowels are divided into two or more ‘sets’ and in any word
all the vowels are chosen from one set: Akan and Turkish
are instances. In these cases, however, the criteria for deciding
the extent of the unit over which the rules (of accent placement,
or of vowel harmony, or whatever) hold is not a simple
phonological decision - grammatical criteria also come into
consideration (the affixation of suffixes to a word, for example).
Such criteria, then, like potential pause, are useful indications,
but do not by themselves suffice to identify words.
The most satisfactory criteria are syntactic criteria, among
them the ‘internal cohesion’, and ‘external distribution’ of a
word, and its ability to stand alone as a ‘minimal free form’.
Together these criteria show that a word has a unity that
justifies its treatment as a unit. Consider, for example, the
structure of a word like uninterruptability. It is probably agreed
that we can segment this word as un-interrupt-abil-ity. The basis
of this segmentation is the recognition that each of these
segments can recur in other words, words given a related
meaning by the common segment (note that meaning slips in
here - it is hard to exclude!): thus we find U^-intelligent;
corrupt-ABIL-ITY; stupid-ITY, etc., and the form interrupt can
itself stand as a word. As far as the ‘internal cohesion’ of the
word is concerned, it is not possible to interpolate any other
segments between the segments we have isolated, nor is it
possible to rearrange the order of the segments we have
isolated. In other words the item is bound together into a unit.
This is not generally the case with strings of words - where it is
almost always possible to add new items or rearrange the items
within the sentence and still maintain a grammatical string. We
can see this if we compare black hat (black top hat, hat which is
black, etc.) with blackcap (in the sense ‘type of bird’): and we
can compare blackcap with black cap (black corduroy cap, cap
which is black, etc.).
Similarly in its external distribution a word operates like a
unit. Either the whole word must be moved round the sentence
or none of it can be moved round the sentence - you cannot
move part of a word about. Thus, as the example black hat
shows, each word is capable of a certain amount of independent
movement, but in the case of blackcap either the whole word is
moved or nothing is moved.
These two criteria are reflected in the ability of a word to
 Words and morphemes 165

stand on its own as an isolate - the definition of a word as a

‘minimal free form’, which derives from Bloomfield. So,
uninterruptability, our example, is as an appropriate one-word
response to some such request as Name a syntactic criterion for
the identification of a word. It is impossible to think of a
comparable question which yields abil as a response (except
perhaps Can you give an example of a segment of a word that
cannot stand as an isolate?). This leads to a distinction between
‘free’ and ‘bound’ forms explored further on page 176).
These criteria are in general satisfactory for an identification
of words in English. Like all linguistic criteria, they apply to a
greater or lesser extent with individual items. For the most part
forms that we wish to classify as verbs, nouns, adjectives and
adverbs meet the criteria outlined. Some cases are less
determinate, particularly compound words like water-mill. Other
form classes do not so readily fulfill all these criteria. This is
particularly the case with articles like the or a, as Bloomfield
himself recognized; these are not usually found as isolates, and
they are not usually detachable from the nouns with which they
are in construction in an NP (as noted on pages 24-5). On the
other hand they are not, in English, bound to their nouns in the
way that a prefix like un- is bound: we can, for instance,
interpolate other items (THE great grey green greasy Limpopo
RIVER : the river). For this reason they are usually considered
separate words. Having isolated those units that do fulfill all the
criteria noted, other segments, which may fulfill only some of
the criteria, must also be treated as words.
The foregoing discussion suggests a certain indeterminacy
about the definition of a word: this is indeed the case. On the
other hand it does seem to be a unit that is unavoidable, and on
to which a number of criteria converge.
At this point we leave discussion of criteria for words, and
consider another problem about the use of the term word itself.
As Lyons points out (1968, 196 ff.), the word is used in a number
of different senses. We need to distinguish between them.

One use is to refer to the actually occurring physical forms. In

this sense of the term we can say that in the sentence:

2 John loved Mary once, but he doesn’t love her any more

there are eleven words. We shall use the term word form for
166 Part two: Morphology

this usage: thus we shall say there are eleven word forms in 2.
From time to time it is convenient to make the further
distinction between a phonological and an orthographic word
form. Orthographic word forms, when used for exemplification
are italicized; phonological word forms will be represented in
‘phonemic brackets’ e.g. word forms, /w3d fomz/.
A different usage is illustrated by saying that in 2 loved and
love are different forms of the word love. This use of the term
‘word’ applies when we say we are going to look up the meaning
of a particular word in the dictionary: we do not expect to find
an entry for each of the different inflectional forms such as
loved, loves, loving, etc. We use the term lexeme to distinguish
this usage of the term word. To distinguish lexemes from word
forms we always use small capital letters to refer to lexemes, a
usage silently followed in previous chapters.
The status of these two terms is rather different. Word forms
can, in some sense, be regarded as substantial units, as actually
occurring forms; the words written on this page are word forms.
The notion of lexeme, on the other hand, is an abstract notion.
Word forms like love, loves, loved and loving can all be related
to the lexeme LOVE, but none of them can be considered actu¬
ally to be the lexeme love. If we associate the head word of a
dictionary entry with the notion lexeme, the lexicographer must
choose some word form by which to represent the lexeme in
question; the choice of word form is a matter of convention in
different languages. In English dictionaries, typically, verbal lex¬
emes are represented by the ‘infinitive’ form: i.e. LOVE is rep¬
resented by love. The same is true in French and Spanish, so we
find aimer in a French dictionary and amar in a Spanish dic¬
tionary. In Latin or Greek dictionaries, however, verbal lexemes
are typically represented by the ‘first person singular present
indicative’ form of the verb; thus we find amo rather than amare
(the ‘infinitive’ form) in a Latin dictionary, and cfiZXea) in a
Greek dictionary.
We might also note instances of ‘homography’: when two dif¬
ferent words (= lexemes) are orthographically represented by
the same word (= word form). Thus the same word form mace
may realize either the lexeme MACE (1) with the sense ‘spice’
or the lexeme MACE (2) with the sense ‘staff of office’ - a fact
usually recognized in dictionaries by according each lexeme a
different entry.
 Words and morphemes 167
The difference between these two senses of word is clearly at
issue if the question is asked: ‘How many different words are
there in this chapter?’ If we mean word in the sense of word
form, then we count each instance of word forms like is, are, be,
was, were, as ‘different words’, and each instance of word forms
like word as ‘the same word’. If, on the other hand, we mean
word in the sense of lexeme, then all the forms is, are, etc. are
different forms of the same word (= lexeme) BE, and we need
to distinguish between WORD (1) (= word form) and WORD
(2) (= lexeme). It is by no means always clear in word counts
which of the two senses is at issue, but they need to be
distinguished.
There is, finally, yet a third use of the term ‘word’ that it is
sometimes useful to distinguish; it can be illustrated by com¬
paring the forms I wrote and I have written with the forms I
laughed and I have laughed. Viewed as word forms, wrote and
written are different word forms, but laughed and laughed are
instances of the same word form. However, from a different
point of view, we can describe wrote as the ‘past tense of
write’ and written as the ‘past participle of write’ and
analogously, the first instance of laughed as the ‘past tense of
LAUGH’ and the second instance of laughed as the ‘past partici¬
ple of laugh’. Syntactic descriptions such as these can be use¬
ful; they are called morphosyntactic words. Typically, but not
invariably, a morphosyntactic word consists of a lexeme and
some associated grammatical description, as in ‘past tense of
LAUGH’, and there is a direct correspondence between a mor¬
phosyntactic word and the word form by which it is realized, as
in the examples given.
So now we have identified three different senses: in different
circumstances we say that laughed and laughed in the examples
given are the same word (= word form); are different forms of
the same word (= lexeme); or are different words (= mor¬
phosyntactic word)! Unless the context makes it clear which
sense is at issue, we need to make a terminological distinction
between these three senses of word.
This brings us to another difficulty. The forms laughed and
wrote may, as morphosyntactic words, be described as, respec¬
tively, the ‘past tense of laugh’ and the ‘past tense of
write’. And from a strictly syntactic point of view we wish to
be able to describe them in this way. However, whereas laughed
168 Part two: Morphology
may, without disagreement, be segmented as laugh-ed, no com¬
parable segmentation is possible for the form wrote. Further¬
more, in the case of laugh-ed the segment laugh can be iden¬
tified with the lexeme LAUGH and the segment -ed with the
description ‘past tense’. The principle that permits us to do this
is that other word forms can be related to the lexeme laugh;
for example in laugh, laugh-s, laugh-ing we find the same seg¬
ment laugh; and other past tense forms contain the segment -ed,
e.g. kill-ed, walk-ed, attack-ed. This suggests the constituent
structure analysis of laughed shown in Figure 69.

Verb

V Tense

LAUGH past

I
laugh ed
Figure 69

But what now do we do with wrote? For syntactic description it

is clearly advantageous to postulate a similar structure, as in
Figure 70

Verb

Figure 70 wrote

since this enables us to account in a systematic way for syntactic

descriptions like ‘past’. But there is no very satisfactory way to
identify part of the word wrote as realizing the lexeme write and
some other part of the word as realizing the syntactic description
‘past tense’: rather, the whole form wrote, as a unity, realizes the
description ‘WRITE + past’. For such reasons we will find it
advantageous to identify two different sorts of units: one is a set of
abstract units, like laugh, write and past, which we need for
 Words and morphemes 169
syntactic descriptions, and one a set of units, like laugh, write,
wrote and -ed, which we need to account for the actual structure
of word forms.
To distinguish between these two different sorts of units, we
call the grammatical units morphemes, and the word forms, or
segments of word forms, morphs as shown in Figure 71.

V Tense V Tense
1
1
Morpheme: LAUGH past WRITE past
1 1
1 1
Morph: laugh ed wrote
Figure 71

The relationship between morpheme and morph is called re¬

alization. Thus, the string of morphemes {WRITE + past} is
realized by the single morph wrote; the morpheme {LAUGH}
is realized by the morph laugh; and the morpheme {past} is
realized (except in cases like those mentioned above) by the
morph -ed.
The morpheme is an abstract unit, a unit that is part of the
syntactic description; from the point of view of the syntactic
description it matters little whether a single morpheme is real¬
ized by a single morph, as is the case of the relationship be¬
tween {LAUGH} and laugh; or whether a string of morphemes is
realized as a single morph, as is the case of the relationship
between {WRITE + past} and wrote.
The morph, on the other hand, is a substantial unit, a word or
part of a word form. When a word form can be segmented into
smaller units, which cannot in their turn be segmented, these
smaller units are morphs. Within words we show morph bound¬
aries by using hyphens (e.g. laugh-ed). Thus word forms may
either consist of single morphs (as wrote, which we consider to
be unsegmentable) or of a string of morphs (as laugh-ed).
In these terms a morpheme is not part of a word form at all:
it is an abstract element whose distribution is determined by the
syntactic description: the relationship between a word form, or
170 Part two: Morphology

the morph(s) that compose it, and a morpheme or string of

morphemes, is one of realization: a morph realizes a morpheme
or a string of morphemes. To distinguish morph from morpheme
in discussion we enclose morphemes in curly brackets, and we
show morphs, like word forms, of which they may be part, by
italic letters: thus the morpheme {past} and the morph -ed.
Further reasons for this distinction will emerge. We may offer
here a number of examples that show the general principle: that
the morpheme is a unit that is accounted for in the syntactic
description, and that the morph is a unit that accounts for the
formation of words; that the morpheme is an abstract and the
morph a substantial unit. This enables us to free syntactic
descriptions from an unnecessarily close identification with word
forms or morphs, and allows us to postulate in the syntactic
descriptions those units, morphemes, that are useful for a syn¬
tactic description. Conversely it allows us to study the structure
of words without necessarily seeking a syntactic correlation for
every morph we may wish to postulate.
We have already discussed examples like:

3 walk-ed; laugh-ed; ran; wrote; went

Syntactically speaking they all represent the past tense of the

verbal lexeme concerned - {WALK + past} ... {GO + past} -
irrespective of the fact that in the first two examples the mor¬
phemes concerned correspond directly with one of the morphs
we wish to identify; whereas in the last three examples the
words are not readily segmentable.
Next consider the forms:
4a large larg-er
4b bad worse
4c foolish more foolish

Forms like those in the left-hand column have traditionally been

called the ‘simple’ forms of the adjectives concerned (LARGE,
bad, foolish), and the forms in the right-hand column the
‘compared’ forms. Simple adjectives have a different distribution
from compared adjectives:
5a My house is large, but yours is larger than mine
5b My essay is bad, but yours is worse than mine
5c My opinions are foolish, but yours are more foolish
than mine
 Words and morphemes 171
It would not be possible to interchange the simple and compared
forms:

6 *My opinions are more foolish, but yours are foolish

than mine

From a syntactic point of view we describe all of the compared

forms as having the same structure - let us say {comp(ared) +
Adj} although in word formation all three forms are different:
larg-er consists of two morphs, worse is a single unsegmentable
word form, and more foolish is two separate word forms - and
furthermore the morph realizing {comp} in this case precedes
the adjective stem, whereas in the case of larg-er it follows. For
syntactic description we need one type of statement, but for
word formation we need another type.
This brings us to our final example for the present. Consider
the word foolish in example 5c. foolish itself may be considered
as segmentable into morphs fool-ish on the analogy of other
forms, like fool, fool-hardy, on the one hand, and boy-ish,
girl-ish on the other. This brings us to a distinction often
drawn between ‘derivational’ and ‘inflectional’ morphology.
Derivational morphology is concerned with those rules by which
new lexemes can be formed; thus from the noun fool we can
derive the adjective foolish by the addition of the morph -ish;
similarly, from the noun morpheme we can derive the adjective
morphemic or the verb morphemicize (= to analyse into
morphemes). Typically each of these forms is represented by a
separate dictionary entry - FOOL, FOOLISH, MORPHEME,
MORPHEMIC, MORPHEMICIZE - and for this reason we
consider them separate lexemes, and derivational morphology to
be the set of rules by which new lexemes can be formed.
Inflectional morphology, on the other hand, is concerned with
the syntactic rules by which a lexeme and its associated
grammatical morphemes are realized. Thus inflectional
morphology relates the lexeme FOOL to the forms fool and
fools, the lexeme morphemicize to the forms
morphemicizes, morphemicizing, morphemicized and so on.
Typically, as we have observed, these latter forms are not found
in a dictionary, and we do not consider them lexemes. But to
return to foolish. From the point of view of the syntax, it is
irrelevant whether foolish is a simple, underived, adjective, like
large, or whether it is derivationally a complex form foolish. For
172 Part two: Morphology
syntax we only need to know that foolish is an adjective, since
the syntax can then assign to it the additional morpheme {comp}
to yield the appropriate compared form. This last example, then,
is a word which can quite legitimately be analysed further into
morphs, but where this morphological analysis is irrelevant to
the syntactic description. In other words, foolish realizes the
single morpheme FOOLISH. Its compositional analysis into
fool-ish is a matter for the derivational morphology and not for
the syntax. (We discuss the status of word forming morphs like
-ish on pages 222-7.)
We conclude that the compositional structure of words is to
some degree independent of the syntactic description of a
sentence. It is thus necessary to distinguish between the
morpheme - an abstract unit whose distribution is accounted for
by the syntax - and the morph - a substantial unit whose
distribution is not wholly accounted for by the syntax.

Technical terms
derivational morphology morphosyntactic word
homographs orthographic words
inflectional morphology phonological words
lexeme potential pause
morph realization
morpheme word form
12 Morphemes and morphs

Morphemes

In the preceding chapter we described the morpheme as an

abstract unit whose distribution is accounted for in the syntax.
We described strings such as {LAUGH 4- past}, {BOY + plur} and
{LARGE + comp} as strings of morphemes. There are two
distinct types of morpheme: items like {laugh}, {boy}, and
{LARGE}; and items like {past}, {plur}, and {comp}. The former
type we call lexical morphemes, or lexemes, and the latter
grammatical morphemes. There is an important distinction
between them.
Lexical morphemes, lexemes, are listed in the lexicon, and are
inserted into syntactic derivations by a lexical insertion rule, in
the manner described on page 94 (though these rules need
some amendment now that we have embarked on a
morphological description). Lexical morphemes, we have seen,
are inserted in tree diagrams under terminal nodes labelled
Noun, Verb, Adjective, etc. Terminal nodes of this sort are
called lexical category nodes, since lexemes are inserted beneath
these nodes, and, as we have seen, lexemes can be categorized
into classes of Noun, Verb, etc. Classes like Noun, Verb, etc. we
call lexical categories; in the lexicon each lexeme needs to be
characterized in terms of the lexical category to which it
belongs.
Lexical categories are, in principle, open; it does not seem
possible to list exhaustively the membership of any major lexical
category, since it is always possible to introduce new items.
Thus, in English, the category of Noun can be extended by the
addition of new members as cultural needs dictate; a new
artifact, for example, is often given a name, and this name,
syntactically, is a noun. Nor is it the business of a syntactic
description of a language to determine the total distribution of a
174 Part two: Morphology
particular lexeme: other considerations are important,
particularly semantic restrictions of the sort discussed (pages
85-8) as selection restrictions. Thus, while the syntax
determines the distribution of the category Noun, the total
distribution of a particular noun is only in part determined by
the syntax.
Grammatical morphemes, items like {past}, {non-past}, {sing},
{pi}, etc., are somewhat different. These morphemes too can be
grouped into categories, such as tense or number, which we call
grammatical categories. The important distinction between
lexical and grammatical categories is twofold. The first
distinction is that whereas the membership of a lexical category
is in general open, the membership of a grammatical category is
closed: it is possible to list exhaustively the membership of
grammatical categories. Thus, in English, the category number
has two terms - we shall refer to members of grammatical
categories as ‘terms' in the particular category at issue - singular
and plural. These two terms are established by the fact that
there is, in English, just this opposition possible in this category
and no other. It is not a necessary characteristic of any language
that there should be just two terms in the category number;
some languages have a third term, dual, characteristically
applied to objects that occur typically in pairs. Thus in classical
Greek we find:

Singular Dual Plural

6 7TOU<? ta)7ro8e oi nodes

‘the foot’ ‘the (2) feet’ ‘the feet’

Note the formal contrast between the noun forms, echoed by

the concordial forms of the article (6 - tea - ot). Such a formal
contrast is not available in English, so we say that English has
only two terms in its number system. The number of terms in a
grammatical category is determined by formal characteristics of
the language in question and cannot be assumed in advance.
The second distinction between grammatical morphemes and
lexemes is that the distribution of purely grammatical categories
is, in principle, describable purely within the syntax. This
distinction can best be illustrated by the following fragment of a
grammar:
 Morphemes and morphs 175
Constituent structure rules:

NP + VP

VP^ V + NP

V —> Verb + Tense

Tense —» I ^aSt 1
1 non-past j

NP^ Art + N

N —» Noun + Number

Number

Lexical rules:
Noun —» {boy, dog, CAT, HOUSE, ■}

Verb -» {frighten, amuse, . . .}

In this grammar the distribution of morphemes like {past} or

{sing} is totally determined by the constituent structure rules. On
the other hand, whereas the distribution of the class Noun or
Verb is determined by the constituent structure rules, the
distribution of an individual noun — HOUSE, say - is deter¬
mined in part by the distribution of the class Noun, but
also by semantic considerations which are external to the syntax
proper.
Having drawn this distinction, we can observe that in many
languages certain grammatical categories are typically associated
with certain lexical categories. In English the category of tense
is typically associated with verbs. So the morphs associated with
the morpheme {past}, orthographically realized as -ed are found
in verb words (kill-ed, walk-ed, etc.) not in noun words.
Conversely the grammatical category of definiteness is realized
in English by articles, which, although they are not affixed to
nouns, only occur in association with nouns, (the man, a walk,
etc.).
A further distinction commonly used in grammatical
discussions is between grammatical and lexical words.
Grammatical words are those words that realize only
grammatical morphemes; lexical words are those items whose
176 Part two: Morphology

stems realize lexemes. This is why it is possible to recognize in

nonsense poetry the fact that the poem is, in some sense, in
English:

’Twas brillig and the slithy toves

Did gyre and gimble in the wabe.

The grammatical words are all clearly English - and, the, did - it
is the lexical words that have ‘nonsense’ stems - tove, gyre,
gimble, wabe. Take slithy for instance; it is in the syntactic
position for an adjective (Art Adj N), and it has the morph -y, a
common marker of a derivational process deriving adjectives
from nouns (cf. slime: slimy, grime: grimy, etc.). There are
lexical classes truly open to new membership. On the other
hand it would probably be impossible to interpret ‘nonsense’
where the lexical stems were related to ‘regular’ English
lexemes, but where the ‘grammatical’ words were nonsense.

Morphs

When a word form can be analysed into smaller forms, which

cannot themselves be analysed into yet smaller forms, the
smaller forms are morphs. A morph is thus a word form or part
of a word form. Word forms that can be analysed into a string
of two or more morphs we call morphologically complex. Word
forms that consist of a single morph we call morphologically
simple: walk-ed is morphologically complex; went or ran is
morphologically simple.
A morph that can stand alone as a word is a free morph: boy,
tea, time, half, back, happy, long are free morphs. A morph
incapable of standing alone as a word is a bound morph: -ish,
un-, -ed, -ness in boy-ish, un-happy, walk-ed, good-ness are
bound morphs. We show bound morphs preceded or followed
by a hyphen, e.g. -ish, -un.
By definition a morphologically simple word must consist of a
single free morph. A morphologically complex word may consist
of a free and a bound morph (boy-ish, un-happy, good-ness,
etc.), two free morphs (tea-time, half-back, etc.), or any
appropriate combination of free and bound morphs.
In discussing morphologically complex words it is convenient
to have some terminology for various different types of
constituent. That part of a word form to which derivational or
 Morphemes and morphs 177
inflectional affixes can be added is a stem. A stem may be
morphologically simple or morphologically complex; a
morphologically simple stem is often called a root. An affix is an
inflectional or derivational morph added to a stem. Affixes
which precede their stem are prefixes', affixes which follow their
stem are suffixes; and affixes inserted within their stem are
infixes.
So, for example, in the forms duck, duck-ling and duckling-s:
duck consists of just a simple root; duck-ling consists of the
stem duck, which is also a root, and the derivational suffix -ling;
and duckling-s consists of the stem duckling, which is not a root
because it is morphologically complex, and the inflectional suffix
-s.1

Prefixes and suffixes can easily be illustrated in English, im-,

un- and pre- are prefixes in the words im-possible, un-likely and
pre-fix; -ed, -ment and -ly are suffixes in walk-ed, conceal-ment
and ungentleman-ly. Infixes do not regularly occur in English,
though one might consider im-bloody-possible as a peculiar
example of infixation. Infixation does, however, occur in
children’s ‘secret’ languages. A quite widely spoken secret
language, sometimes called “aigy-paigy”, is formed by infixing
the syllable aig (pronounced /eig/ as in the proper name Craig)
before every vowel nucleus of ‘regular’ English: Thaigis aigis
haigow yaigou taigalk saigecraigetlaigy - th-is -is h-ow y-ou t-alk
s-ecr-et-ly. In this language aig is an infix in a word like th-is,
yielding th-aig-is. Many Semitic languages are analysed as
containing infixes: in modern Arabic, a root like k-t-b ‘write’
appears in words like katab ‘he wrote’, kita.b ‘book’, and, with
prefixes, ma-ka:tib ‘places for writing, studies’, ma-ktab ‘places
for writing, study’, je-ktab ‘he is writing’; similarly the root g-l-s
appears in galas ‘he sat’, gatlis ‘sitting person’, ma-ga:lis
‘councils’ ma-glas ‘council’ (Bloomfield 1935, 243-4).
Hitherto many of our examples have been recorded in the
traditional orthography; we should, however, also consider
examples involving phonological as well as the orthographic
form. Consider the following plural forms in English:

1 Sometimes a further terminological distinction is drawn between stems to

which inflectional elements can be added, which are called stems, and stems to
which derivational elements are added, which are called bases. Thus duck in
duck-ling is a base, since -ling is a derivational affix; and duckling in duckling-s
is a stem, since -s is an inflectional affix.
178 Part two: Morphology

buses /bAS-iz/ hats /hat-s/ hags /hag-z/

cheeses /tjiz-iz/ sacks /sak-s/ bags /bag-z/
bushes /buf-iz/ maps /map-s/ knobs /nDb-z/
garages /gara3-iz/ proofs /pruf-s/ eaves /iv-z/
patches /patj-iz/ moths /mD0-s/ lathes /lei5-z/
judges /d3Ad3-iz/ bees /bi-z/
paws /po-z/

Three distinct morphs realize plural in English: /-iz/, /-z/ and

/-s/. When a particular morpheme is realized by a series of
different morphs, the alternative realizations are called
allomorphs. In the example above the distribution of these
allomorphs is determined by phonological characteristics of the
stem final segment, thus:

/-iz/ occurs after stems ending in /s, z, J, 3, tf, d3/ i.e.

after stems ending in a sibilant, voiced or voiceless
/-z/ occurs after stems ending in a voiced segment other
than a sibilant
/-s/ occurs after stems ending in a voiceless segment
other than a sibilant

When the distribution of allomorphs can be described in

phonological terms, the allomorphs are said to be phonologically
conditioned.
In the example quoted there is some phonological similarity
between the appropriate allomorph and the stem to which it is
affixed: /s/ and /z/ are voiceless and voiced alveolar fricatives,
affixed to stems ending in voiceless and voiced segments. There
is no need for phonological conditioning to involve phonological
similarity. Thus in the following data from Tsotsil (taken from
Nida 1949) we find dissimilation - i.e. the affixes are phono¬
logically dissimilar in a regular fashion from the stem vowels:

-k'uf wedge la -k'ufi to put a wedge

ITT

-Jik' prop used beneath 2a -Jik'u to put a prop

an object under
-Jon prop used against 3a -Joni to put a prop
an object against
-vov crazy 4a -vovi to go crazy
t'uf wet 5a -tufi to become wet
 Morphemes and morphs 179
The suffix is /-i/ (a front vowel) when the stem vowel is /u/ or
/o/ (both back vowels), and /-u/ (a back vowel) when the stem
vowel is /i/ (a front vowel). There may indeed be no
phonological similarity at all between stems and phonologically
conditioned affixes (Tojolabal data from Nida 1949):

1 -man to buy 7 -al to say

2 hman I buy 8 awal you (sg) say
3 -il to see 9 -lap to dress
4 kil I see 10 -slap he dresses
5 -k'an to want 11 -u to drink
6 ak'an you (sg) want 12 yu he drinks

Given this limited amount of data the affixes appear to be as

follows:

‘1st person’ (examples 2, 4) /h-/ preceding consonants

/k-/ preceding vowels
‘2nd person’ (examples 6, 8) /a-/ preceding consonants
/-aw-/ preceding vowels
‘3rd person’ (examples 10, 12) /s-/ preceding consonants
/y-/ preceding vowels

If this analysis accurately represents the situation in the

language as a whole, we see that in the case of the ‘1st’ and
‘3rd’ person pronouns there appears no obvious phonological
similarity between the preconsonantal and prevocalic
allomorphs: the distribution, however, is clearly phonologically
conditioned.
The selection of allomorphs is not always phonologically
conditioned. There may be other conditioning factors. Consider
the following data, from English:

house /haus/ houses /hauz-iz/

calf /kaf/ calves /kav-z/
elf /elf/ elves /elv-z/
sheaf //if/ sheaves /Jiv-z/
wife /waif/ wives /waiv-z/
bath /ba0/ baths /ba6-z/
path /poG/ paths /pad-z/

The distribution of the plural affixes (in the right-hand column)

is entirely regular, and follows the rules for English plurals given
180 Part two: Morphology
above. The stem in this case shows allomorphic variation:
/haus/ : /hauz-/ etc. There appears to be no phonological
reason for this alternation, since comparable forms are entirely
regular:

piece /pis/ pieces /pis-iz/ not */piz-iz/

reef /rif/ reefs /rif-s/ not */ riv-z/

and indeed the ‘possessive’ forms are entirely regular:

wife /waif/ wives /waiv-z/ wife’s /waif-s/

youth /ju0/ youths /ju5-z/ youth’s /ju0-s/

If you check, the allomorphs for the possessive (e.g. wife’s), are
identical to those for plural.
We are thus forced to say that with items like (HOUSE) the
allomorph /hauz-/ is chosen in the environment {_ plural}.
When the selection of the appropriate allomorph is determined
by circumstances which cannot be described phonologically, then
these allomorphs are said to be grammatically conditioned.
Grammatically conditioned affixes in English include the
plural affixes -im and -en in:

kibbutz kibbutz-im
seraph seraph-im
ox ox-en

These irregular forms include loans (kibbutzim) and linguistic

‘fossils’ (oxen).
Occasionally morphs may be in free variation: in such cases it
does not matter which of two (or perhaps more) allomorphs is
chosen. Thus the forms burned and burnt seem to be free
variants.

I burned ... I burnt.. .

I have burned ... I have burnt. . .

If this is the case, then /-d/ and /-t/ are in free variation as
realizations of the morphemes past and past participle in
construction with {BURN}.
The grammar may sometimes have a morphemic distinction
when there is no overt marker of this particular distinction in
actual word forms. In such cases a ‘zero’ morph or allomorph is
sometimes postulated. This is usually represented by 0. Zero
 Morphemes and morphs 181
allomorphs are often postulated when the structure of a series of
related forms is such that there is a ‘significant absence’ of a
formal marker at some point in the series: e.g., with English
noun plurals we find:

boy - boys; dog - dogs; ox - oxen, etc.

but

sheep - sheep; salmon - salmon; grouse - grouse, etc.

cf.
I shot two dogs yesterday
I shot two grouse yesterday
*1 shot two grouses yesterday

We can postulate that the plural form of GROUSE is grouse-0.

Zero morphs are sometimes postulated when the general
structure suggests a zero element. For example, in Totonac the
subject pronouns are as follows:

k- first person singular -tit second person plural

-wi first person plural 0- third person singular
-ti second person singular -'qu third person plural

The third person singular is never indicated overtly; i.e. it has

no obvious form. The absence of some other form indicates the
third singular. Structurally this is a type of significant absence
(Nida 1949, 46).
Zero should be used sparingly since it can lead to abuses.
Consider the following ways of looking at the structure of
English singular and plural nouns.

1 Suppose we analyse the singular forms of nouns as consist¬

ing of the simple stem with no marker at all: then we may say
that boy differs from boys in that the former represents the
single morpheme {boy} and the latter the two morphemes
{BOY + pi}: boy consists of a single morph, and boy-s of two
morphs. Under this analysis grouse (singular) represents the
single morpheme GROUSE and grouse (plural) the two mor¬
phemes {GROUSE 4- pi}: grouse (singular) consists of a single
morph, and grouse (plural) of the two morphs grouse-0.

2 Suppose now that we consider the absence of a marker of

singular in nouns as a form of ‘significant absence’ (cf. above).
182 Part two: Morphology

We might then say that the form boy represented the two mor¬
phemes {BOY + sing}: and the form boys the morphemes
{BOY + pi}. Now {sing} is represented by the morph 0. Thus boy
consists of the two morphs boy-0, and boys of the two morphs
boy-s. Analogously grouse (singular) consists of the two morphs
grouse-0 and grouse (plural) of the two morphs grouse-0. In the
former case 0 is the invariant morph representing {sing} and in
the latter case 0 is an allomorph of {pi}!

Realization rules

The relationship between morpheme and morph is described as

‘realization’. Thus in the form walk-ed, walk realizes the lexeme
WALK and -ed realizes the grammatical morpheme {past}. This
entails a new set of rules, ‘morphological realization rules’. These
are those rules that realize the various morphemes, lexical and
grammatical, introduced by the grammar.
Our total description now has three sets of rules:
i Constituent structure rules. These develop constituent
structures as far as the terminal symbols of the grammar. The
terminal symbols are either lexical category symbols (N, V, Adj,
etc.) or terms from grammatical categories (past, sing, pi, etc.).
ii Lexical rules, or a lexical insertion rule and an associ¬
ated lexicon. We have already noted that, in the grammars we
are dealing with, these produce the same effect, but that we
would prefer to use a lexical insertion rule. In the example
which follows, however, it is convenient (for lack of space) to
use a lexical rule.
iii Morphological realization rules. These rules introduce
the appropriate morphs that correspond to the grammatical or
lexical morphemes introduced in i and ii. Realization rules
can be formulated either as rules which introduce phonologically
specified items (i.e. items specified in phonological terms) or
orthographically specified items. We use both kinds at various
stages of our account. The rules that follow, to exemplify the
principle, are formulated as rules which introduce phonologically
specified morphs.
The data we will seek to account for represents the formation
of some singular and plural, and possessive forms of English
nouns. Our data is as follows:
 Morphemes and morphs 183
BOY CAT BUS GROUSE
N boi kat bAS graus
N + poss boiz kats bASIZ grausiz
N + pi boiz kats bASIZ graus
N + pi + poss boiz kats bASIZ grausiz
SHEEP WIFE DWARF
N Jip waif dwDf
N + poss Jips waifs dwDfs
N + pi Jip waivz dWDVZ
N + pi + poss Jips waivz dwDVZ

Constituent structure rule:

1 N NS (+ pl)(+ poss) NS = Noun stem
pi = plural
poss = possessive

Lexical rule:
2 NS ^ {BOY, cat, bus, grouse, sheep,
WIFE, DWARF, . . .}

Morphological realization rules:

3 DWARF -» /dwav-/ in the environment_pi
/dwDf/ elsewhere
4 WIFE -> /waiv-/ in the environment_pi
/waif/ elsewhere
5 SHEEP -» /Jip/
6 GROUSE—> /graus/
7 BUS /bAs/
8 CAT /kat/
9 BOY -*/boi/
10 pi -* 0 in the environment /graus/_
or /Jip/—
11 poss —>• 0 in the environment pi_
/-iz/ in the environment sibilant C_
/-s/ in the environment voiceless C_

/-z/ elsewhere
12 pi /-iz/ in the environment sibilant C_
/-s/ in the environment voiceless C_
/-z/ elsewhere
184 Part two: Morphology

Some typical derivations:

Rule Rule
1 NS + pi + poss
2 boy + pi + poss 2 wife + pi + poss
9 /boi/ + pi + poss 4 /waiv/ + pi + poss
11 /boi/ + pi 11 /waiv/ + pi
12 /boi-z/ 12 /waiv-z/
Rule
2 GROUSE + pi + poss
6 /graus/ + pi + poss
10 /graus/ + poss
11 /graus-iz/

Rules 3, 4 and 10-12 introduce allomorphs: where the envi¬

ronment is specified in phonological terms these are phonologi-
cally conditioned allomorphs; where the environment is gram¬
matically conditioned they are grammatically conditioned
allomorphs.
Rules 10 and 11 introduce zero elements.
Rules 3-9 introduce bases.
Rules 11, 12 introduce suffixes.
You might like to think of the rules as formulaic statements
of ordinary language statements, with which we hope you agree,
like: 3 dwarf has two stems /dwDv-/, which is a bound form,
occurs in plural forms, and /dwDf/, which is a free form, else¬
where; 11 the genitive is not marked separately in the plural
form of nouns (except with nouns like GROUSE, SHEEP, which
have no plural marker accounted for by rule 10); where it is
marked the marker is an affix /-iz/, /-z/ or /-s/ as appropriate.
Note that the rules must be applied in the order stated. For
instance rules 10-12 must be in this order. Try using them as in
the derivations above in the reverse order - you will get wrong
answers:

1
2 BUS + pi + poss SHEEP + pi + pOSS
7 /bAs/ -I- pi + poss 5 /Jip/ + pi + poss
12 /bAS-iz/ + gen 12 /Jip-s/ + poss
11 /bAS-IZ-lz/ 11 / fip-s-iz/
(rule 10 cannot apply since the
environment in which it ought
to have applied is gone)
 Morphemes and morphs 185
Technical terms

allomorph lexical morpheme

base lexical word
bound morph morph
category morpheme
derivational affix morphological realization rule
derivational morphology morphologically complex
free morph morphologically simple
free variation phonological conditioning
grammatical category prefix
grammatical conditioning realization rule
grammatical morpheme root
grammatical word stem
infix suffix
inflectional morphology term (in a grammatical category)
lexical category zero
13 The analysis of words

We have distinguished between the morpheme (a distributional

unit of the syntax) and the morph (a distributional unit connected
with word formation). The relationship between the two we have
called realization, and we have seen that the relationship between
morpheme and morph is not always one-to-one. In this chapter
we look at how the sort of analysis we establish can be accommo¬
dated in a grammar containing the three types of rules discussed
in the last chapter: constituent structure rules, lexical rules, and
morphological realization rules.
Our data will be extremely limited, but are sufficient to
illustrate the problems. The data are the forms of the nouns and
verbs in simple sentences like

la The boy yawned

lb The boys yawned
lc The boys yawn
Id The boy yawns

Note that we must consider these forms in sentences, rather than

in isolation since the morpheme, as a syntactic unit, has implica¬
tions that go beyond purely word formation.
Let us first consider the apparently straightforward description
of the singular and plural forms of English nouns. The ‘regular’
paradigm for English nouns is illustrated in such pairs as:

2 singular plural
boy boys
girl girls

The paradigm is called regular since it applies to the vast majority

of English nouns, and applies to all new borrowings (e.g. sputnik :
sputniks). For simplicity we shall take our examples from the
written rather than the spoken form of the language and do not
consider questions of allomorphy: so we consider the regular
 The analysis of words 187

plural morph to be simply -s. Since we have distinguished between

morph and morpheme, in analysing forms such as those in 2 we
need to distinguish a morphological analysis (= analysis into
morphs) from a morphemic analysis (= an identification of the
morpheme or morphemes realized by a particular morph).
The principal criterion for a morphological analysis is a formal
difference between a set of words. Boy and hoys differ in that the
former lacks and the second contains the morph -5. This formal
distinction is shared by other nouns (girl : girl-s; cat : cat-s, etc.)
and leads us to analyse boys into two morphs boy-s, and boy into
a single morph boy. The formal analysis also has a semantic
justification: the form boy is found in sentences appropriately
applied to situations describing a single young male human; and
the form boys in sentences appropriately applied to situations
describing more than one young male human. Since boy is com¬
mon to both words, we assume that this is the form associated
with the semantic notion of ‘young male human’; and since boys
contains the additional morph -5 we assume that this can be
associated with the semantic notion of ‘more than oneness’. The
semantic criterion reinforces our formal analysis. It is convenient
to refer to boy as a lexical stem and -s as the ‘plural marker’.
As far as a morphemic analysis (= an identification of the
relevant morphemes) is concerned, we again find formal and
semantic criteria. The boy part of either word can be related to
the lexeme BOY. The alternation between the simple form boy
and the complex form boy-s is accounted for by postulating a
category of number with the two terms {sing(ular)} and {pl(ural)},
such that boy realizes the morphemes {BOY + sing}, and boys
realizes the morphemes {BOY + pi}. The formal identification of
these morphemes, however, can involve considerations beyond
those which apply to single words, since there are wider
considerations involving the structure of whole sentences. We
note, for example, that in sentences like lc, Id the difference in
form between boy and boys is matched by a difference in form
between the verbs yawns and yawn. This difference is systematic
in that it is replicated in other similar pairs of sentences (the cat
walks; the cats walk, etc.); the systematic difference between the
subject nouns is matched by a difference in the verb word. We
shall later identify this too as an alternation in number, and it is
convenient to have the same system operating in the verb as in
the noun. To facilitate this, it is covenient to match morphemes in
188 Part two: Morphology
both words. This lends additional support to the analysis of boy as
realizing the morphemes {BOY + sing}, and boys as realizing the
morphemes {boy + pi}. As far as semantic considerations are
concerned, with the particular pairs of items in 2 we have no
further observations to add to those made for the morphological
analysis since we need to account for the same distinction.
Using our analysis, we can formalize our description into the
following little grammar for the NP:
3 Constituent structure rules:

NP Art + Noun
Noun N + Number
f sing}
Number
IPI J
Lexical rule:
N -> {BOY, GIRL . . .}

This grammar generates the trees in Figures 72 and 73.

Art Noun

N Number

BOY sing BOY pl

(the boy) (the boy-s)

Figure 72 Figure 73

Already there is a problem. In Figure 73 each of the morphemes

can be related to a single morph (Art: the; BOY: boy; pl: -5) but
what about Figure 72? We appear to have too many morphemes.
Before we examine this problem further, consider some of the
‘irregular’ noun forms in English:

4 sing pl
man men
mouse mice
goose geese
 The analysis of words 189

There appears to be no acceptable segmentation of the plural

forms here: i.e. no morphological analysis is possible. A mor¬
phemic analysis, however, is clearly possible, and exactly the
same considerations that applied to the morphemic analysis of
2 applies to the forms in 4. We want to say that man realizes
the morpheme string {man + sing} and men the morpheme
string {man + pi}. This implies that we can expand our lexical
rule in 3 to:

5 N—>{boy, girl, man, mouse, GOOSE, .. .}

Let us now turn to morphological realization rules. Consider
first the plural forms. As far as the rules for forms like boys are
concerned, no great problem arises. We can have quite straight¬
forward rules of the form:

6 BOY —> boy

pi —> -s

Informally we call rules of this sort ‘agglutinative’ rules: such

rules have a single morpheme on the left-hand side of the rule
matched to a single morph on the right-hand side. As far as rules
for forms like men are concerned, we can, since we have decided
that these forms are unsegmentable, have rules of the form:

7 MAN + pi —> men

We call rules like this ‘fusional’: fusional rules have two mor¬
phemes on the left-hand side of the rule matched to a single
morph on the right-hand side. A rule like this seems to capture
well the fact that men is unsegmentable, and that it realizes two
morphemes.
Rules like 7 are no longer simple ‘rewrite’ rules, as defined on
pages 40-3, since there are two items on the left-hand side of
the rule. A rule like this is in fact a transformational rule. We
shall find that morphological realization rules are of a ‘mixed’
nature, but this will not detain us here.
Let us now turn to the singular forms. There are a number of
possible solutions. We can suggest a fusional type of realization
rule here too:

8 B O Y + sing —> boy

We might alternatively say that BOY corresponds to boy and that

there is no singular morph: the absence of any marker is in itself
190 Part two: Morphology
distinctive (you will remember the story where Sherlock Holmes
perceived the significance of the fact that the dog did not bark).
We can adopt an analysis involving the use of a ‘zero’ morph, as
discussed in the preceding chapter. However, postulating zero
morphs brings problems (and if we follow this solution we will
need a lot of zero morphs to account for the relevant facts of
English, even in this small area). Instead, we propose the follow¬
ing rule:

9 N + sing —» N

This rule can be understood to mean: for any string {N + sing},

the morpheme {sing} has no realization. In other words singular
nouns occur in their simple stem form. We call rules like this
‘null realization rules’: note that this type of rule is not a rewrite
rule either. Note too that the implications of this type of rule are
different from the use of zero. Zero implies a realization, but that
the realization is 0. Null realization rules imply no realization at
all.
Another solution is more radical. Since only plural forms have
a marker, we might consider plural to be an optional category;
singular forms have no mark of number either at the morpheme
or at the morph level. This implies that the constituent structure
rules shown in 3 need to be amended to:

10 NP -»> Art + N
N -» N (+ pi)

These rules yield the tree diagrams of Figures 74 and 75:

NP NP

Art Noun Art Noun

N N pi

BOY BOY

(the boy) (the boy-s)

Figure 74 Figure 75
 The analysis of words 191
Consider each solution in turn. The last solution, 10, has the
attraction that we need only postulate morphemes when we find
morphs that correspond to them: but we have already seen that
this is unsatisfactory for all forms, e.g. the irregular plurals. A
much greater problem lies in the fact that with rules like those of
10 the grammar does not systematically account for the category
of number: this is a drawback at a theoretical level since we
expect our grammar to give a systematic account of such
categories; even at a practical level a systematic account of the
category of number makes easier the description of the facts of
such grammatical characteristics as number agreement between
subjects and verb (illustrated in (lc,ld).
The solution of 8 is perfectly possible; its drawback is that listing
every possible noun in a rule of this sort makes the grammar
extremely cumbersome and anyway misses a proper generalization
about the language: that singular nouns appear in the base form
alone. This leaves solution 9: it has the advantage of being able to
postulate a morpheme {sing} and allows us to state the correct
generalization that singular nouns are never marked by any
morph: they occur in their simple stem form. We can summarize
our rules as:

11 Constituent structure rules:

NP -* Art + N
N —» N + Number

Lexical insertion:
N -» {BOY, GIRL, man, mouse, . . .}

Morphological realization rules:

MAN + pi -» men
MOUSE + pi —> mice
N + sing —*■ N
MAN —> man
mouse —> mouse
BOY —> boy
GIRL girl
-5
pl
Art the
192 Part two: Morphology
Note that the morphological realization rules must be ordered as
we saw in the last chapter; or at least partially ordered: clearly
the first rule shown must precede the fourth and the rule realizing
plural as -s: otherwise we shall produce forms like mans (children
indeed sometimes produce such forms).
Consider now the analysis of the verb forms in the sentences:

12a The boy yawns

12b The boys yawn
12c The boy yawned
12d The boys yawned.

The verb forms shown are often called the ‘regular’ forms of the
verb, since these forms are most common, and are used with new
borrowings or coinings. As far as a morphological analysis is
concerned, there are three distinct word forms, which can be
analysed as yawn, yawn-s and yawn-ed. The formal identification
of these morphs can be reinforced by semantic criteria. The forms
12c, 12d are appropriately applied to a situation which happened
at some time previous to the time of utterance, whereas the forms
12a, 12b are more appropriately applied to a situation describing
either a general truth or habit (it is the habit of boys to yawn,
etc.) or possibly to a situation actually happening at the moment
of utterance - rather in the fashion of a commentary.
Let us characterize this distinction as one between past 12c, 12d
and non-past time, 12a, 12b. A morphemic analysis (an analysis
into morphemes) needs to postulate the same distinction, which
we can, following tradition, identify as the category of tense, as
we have already noted.
Now consider the pair of forms yawns and yawn in 12a, 12b.
We have already identified the distinction between boy and boys
as involving the grammatical category of number: we now note
that the distinction between yawn and yawns correlates with this.
We find sentences like 12a, 12b, but we will not find:

13 *the boy yawn

*the boys yawns

and this relationship is, of course, shared by other different

sentences similar to 12a, 12b:

14 the cat yawns; the cats yawn

the boy walks; the boys walk
 The analysis of words 193'
The selection of the appropriate verb form depends on the
number of the subject NP: if the subject is singular, then the verb
has an -5 suffix; if the subject is plural, then the verb has no affix.
This suggests we should characterize the distinction in the verb
forms too as one of number. Agreement between subject NP and
verb form is called concord.
Note at this point that the criteria for setting up these mor¬
phemes of number in the verb involve syntactic considerations
(i.e. the need to account for concord) that go beyond strict
morphological considerations (i.e. an analysis into morphs).
We are now able to postulate that the morphemes realized by
the forms yawns and yawn are:

15a yawns', {yawn + non-past + sing}

15b yawn: {yawn + non-past + pi}

This morphemic analysis might be represented in constituent

structure trees of the form shown in Figures 76 and 77.

Verb Verb

V Tense Number V Tense Number

I I
1 1
YAWN non-past sing YAWN non-past Pi
(yawns) (yawn)

Figure 76 Figure 77

Note the same problem as was encountered with respect to the

analysis of the noun: we have too many morphemes for the
number of morphs we can identify. Once again a number of
possible solutions suggest themselves. We can have fusional rules
of the form

16 YAWN + non-past + sing -> yawns

yawn + non-past + pi —» yawn

but this is open to the same objection as the comparable rule for
the singular forms of nouns discussed in 8: it misses an obvious
generalization. More realistically we could have a rule for the
lexeme yawn:

17 YAWNyaw/T
194 Part two: Morphology
and then consider realization rules for the grammatical
morphemes. Here a fusional rule may seem appropriate for the
singular form:

18 non-past + sing -> -s

but what of the plural form? Neither {non-past} nor {pi} have any
realization. We can have a rule:

19 V + non-past + pi —» V

analogous to rule 9: i.e. a ‘null realization’ rule. Yet another

solution is to say that {non-past} has no realization in either the
singular or plural forms, that singular is realized by -s and plural
has no realization. This suggests rules like:

20 V + non-past —> V
sing —» -s
V + pi -» V
YAWN —> yawn

We provisionally adopt this last solution as the best, though we

shall have cause to return to examine it later. Before leaving
these rules, observe that the rules in 17-19 and those in 20 make
somewhat different generalizations about the forms in question.
The first set of rules, 17-19, asserts that -5 is a marker of both
tense and number, i.e. that it is a singular tense marker. The
second set of rules, 20, assert that there is no marker of tense and
that -5 is merely a singular marker.
Now consider the morphemic description of yawned in 12c, d.
These forms have already been characterized as being {past} in
comparison with the forms in 12a, 12b which are {non-past}. We
need to consider whether yawned in 12c is {sing} and yawned in
12d is {pi}, paralleling the singular/plural description of yawn and
yawns. We may take either of two opposite positions. We can
assert that verb forms always agree, or concord, with their subject
nouns in number, but that in past tense forms there is no overt
marker of this agreement: yawned in 12c realizes {yawn + past +
sing}, and yawned in 12d realizes {YAWN + past + pi}.
Alternatively, we can say that verbs only concord with the subject
NPs in the non-past, as suggested by 20, and there is no concord at
all in the past. So yawned in both 12c and 12d simply realizes
{YAWN + past], and the question of number agreement is
irrelevant, since verbs in the past do not concord.
 The analysis of words 195
A problem for either of these analyses is just how to capture
the relevant generalization in terms of rules. There are two
problems: one related to the realization of the tense morphemes
and the other related to the number of morphemes. Let us
consider them in turn.
Tense is a category that, in English, is associated with the verb.
We have suggested that this category has two terms: {past} and
{non-past}. We have further hypothesized that {past} can be
associated with the morph -ed and {non-past} has a null
realization. It may be argued then that we can dispense with a
morpheme {non-past} altogether. But this raises the same sort of
objections as we raised to a comparable situation with singular
nouns, discussed with respect to 10: it misses an important
generalization about English verb groups. Furthermore we find
the morphemes {past} and {non-past} useful for other purposes,
for example co-occurrence restrictions with adverbs of time,
which can be accounted for, at least in part, in terms of
co-occurrence restrictions with tense morphemes: the tense
morphemes have a syntactic function irrespective of the fact that
one of them need not be realized by any overt morph. We
therefore retain tense morphemes in the verb.
Number morphemes pose a different problem. We can argue,
for English, that number is primarily a category associated with
the noun, only secondarily a category associated with the verb:
the subject noun controls the number morpheme on the verb. For
example, the verb cannot be in the singular form if the subject
noun is plural (*the boys yawns). In traditional terms this
relationship is called concord and the noun considered to be the
‘controller’ of the concord. But does the subject noun always
require the verb to show concord? As we have seen, this question
can be answered in two ways. We can say that the verb always
agrees in number with the subject (i.e. always has a number
morpheme) but that this agreement is not always marked with an
overt morph of number agreement. This solution can be
diagrammed, informally, as:

21 {N + sing}— {V + non-past + sing} (yawns)

{V + past + sing} (yawned)
{N + pi} - {V + non-past + pi} (yawn)
{V + past + pi} (yawned)
196 Part two: Morphology
At the other extreme, we can say that the verb only concords
with the subject in number when the subject is singular and the
verb is non-past. This solution we can diagram as:

22 {N + sing}— {V + non-past + sing} (yawns)

{V + past} (yawned)
{N + pi} - {V + non-past} (yawn)
{V + past} (yawned)

As far as morphology is concerned, the solution shown in 22 is

simpler in that we have fewer morphemes to account for that
have no overt realization: the solution shown in 21 leaves us with
a number of morphemes with no overt realization. Solution 22
also enables us to say that a form like yawned is simply the past
tense form of YAWN (there being no number agreement in the
past), where the solution 21 entails our having to say that yawned
is either the past tense singular form of YAWN or the past tense
plural form of YAWN. A further advantage of 22 is that it enables
us to make clear in the grammar that number is primarily a
category of the noun, and that it is a verbal category only
secondarily, and by concord.
How, then, can we express this generalization in terms of
rules? Constituent structure rules for these phenomena suffer two
drawbacks: firstly, they are extremely clumsy to write (the reader
is invited to try), but more importantly they do not express the
generalization so clearly. More appropriate is to formulate a
transformational rule that has the effect of copying the
morpheme of number from the NP subject on to the verb. This
means that in the constituent structure we only need to develop
number as a category of the noun (which is what we want); the
transformational rule then copies it on to the verb (also what
we want). The rules in 23 do this.1
Constituent structure and lexical rules develop the tree shown
in Figure 78 opposite.

1 In the rules of 23 we treat tense as a category of the verb, and further as a

category that follows the verb. On page 110 we made a different proposal: that
tense should be treated as an auxiliary and that auxiliaries should be considered
a category of the VP rather than of the verb. The more straightforward analysis is
preferred at this point, since we shall then not have the further complication of
re-ordering Tense + Verb to Verb + Tense, to accommodate the facts of morph
order. The analysis adopted does not affect the argument in any way.
 The analysis of words 197
23 Constituent structure rules:
S -* NP + VP
VP^ V
NP—* Art + N
N —> N + Num

nHp"8}
V —> V + Tense

Tenser | Past 1
[non-past J

Lexical rules:
V —> {yawn, . . .}
N {boy, girl, . . .}
Art -* the

Subject-verb concord:

SA: jvj p(Art + N + sing) “ yerb(V + non-past)

SC: n p^Art + N + sinS) _ Verb^V + non_Past + sin§)

the BOY YAWN

Figure 78

The concord rule operates on this structure. The structure meets

the structural analysis specified in SA: there is a singular N
followed by a non-past verb. The structural change (SC) specifies
that we must add a morpheme {sing} to the verb. The resultant
structure is shown in Figure 79:
198 Part two: Morphology

NP VP

Art Noun Verb

N Num V Tense sing

sing non-past

the BOY YAWN

Figure 79

Realization rules operating on Figure 79 yield The boy yawns.

Consider the advantages of the analysis proposed. To begin
with, we have identified the categories of number and tense as
being primarily categories of, respectively, the noun and the
verb: this is shown by the constituent structure rules. We have
identified the category of number as only secondarily a category
of the verb (shown by the fact that it is derived by the concord
rule) and then only a category under certain conditions (i.e. when
the verb is non-past and singular).
Before closing the discussion of the morphology of the verb,
note that, just as we find irregular noun plurals in English (e.g.
goose: geese), so too we find irregular past tense forms of verbs -
run: ran; think: thought, etc. It is clear that the relevant past
tense forms can be analysed as realizing morpheme sequences
such as {run + past}, {think + past}, etc.
Morphological realization rules for the verb forms discussed
are shown in 29 below. In form they are similar to those
introduced for the noun forms discussed at the beginning of the
section.
At this point it is sensible to illustrate the operation of concord
rules with a further example. Consider the following NPs:

24a the boy

24b the boys
24c a boy
24d some boys
 The analysis of words 199
The opposition the : a has been traditionally described as one of
definiteness, the category of definiteness having the terms
definite and indefinite. This is done on the grounds that in
uttering a sentence like:

25 I want the boy to come and see me this evening

the speaker typically has some particular individual in mind,

whereas in uttering a sentence like:

26 I want a boy to come and see me this evening

no particular boy is in the mind of the speaker and any boy will
serve. This characterization suffices for our present purposes, and
we postulate morphemes {def(inite)} and {indef(inite)} to account
for the distinction. What then of the plural forms 24b and 24d?
Consider first the forms 24a, 24b. The article morph is the same
in both cases: the. We encounter the problem discussed with
respect to agreement in the verb: do we wish to analyse the in the
a sentence as ‘the singular form of the definite article’ and the in
the b sentence as ‘the plural form of the definite article’. This
analysis suggests that the forms correspond, respectively, to
morpheme strings {def + sing} and {def + pi}. Alternatively we
can say that the is simply the definite article, realizing the mor¬
pheme {def} and no question of number arises. This seems a more
satisfactory solution since, as before, it leads us to postulate
fewer morphemes that have no overt realization.
In the forms of 24c, 24d, it is possible to argue that the forms a
and some are, respectively, ‘the singular form of the indefinite
article’ and ‘the plural form of the indefinite article’. Evidence for
such an analysis is shown if we compare 26 with:

27 I want some boys to come and see me this evening

In 26 we can assert that the speaker has no particular boy in mind

and any boy will serve - the speaker clearly has only a single boy
in mind; in 27 the speaker has a number of boys in mind, but no
particular number and no particular boys. The opposition appears
to be between a single unspecified individual and a number of
unspecified individuals. This leads us to postulate that a and some
realize, respectively, the morpheme strings {indef + sing} and
{indef + pi}.
If we accept this analysis, we need to consider how it can be
incorporated in our grammar; and once again a concord rule
200 Part two: Morphology
seems to be an appropriate mechanism. The following will serve:

28 Constituent structure rules:

N P —» Art + N
N —» N + Num

Article-Noun concord:
sing
SA:NpWindef>-N"[pl
sing sing
Lpl J )
)—N—
SC: NP<Art(indef + Lpl J

The concord rule is understood exactly as the previous rule,

except that here we have an additional notational convention: an
item in one pair of square brackets is matched with the
corresponding item in the other pair of square brackets. For
example the constituent structure rules generate the trees in
Figures 80 and 81:

NP NP

Art Noun Art Noun

1
1 /
indef N Num def N Num
1
1
sing Pi
Figure 80 Figure 81

The concord rule now operates on the first of these strings to

yield the derived tree shown in Figure 82. Note that the rule
cannot operate on the second of these trees (Figure 81) since it
does not meet the structural analysis: this corresponds to our
observation that the definite article does not concord with the
noun in number.
 The analysis of words 201

NP

indef sing N Num

sing
Figure 82

We can summarize the rules we have proposed and add

appropriate realization rules:

29 Constituent structure rules:

S-> NP + VP
VP^ Verb
NP—» Art + N
N —> N + Num
| sing \
Num —>•
l Pi i
fdef
Art —*
{indef }
Verb —» V + Tense
| past |
Tense -*
[non-past j

Lexical rules:
N —> {boy, girl, man, goose, . . .}
V -> {yawn, walk, run, . . .}

Concord rules:
Subject-verb concord:
SI: Np(Art + N + sing) -
Verb(V + non-Past)
SC: Np(Art + N + sing) - _ r , (V + non-past + sing)
Verb
Article-noun concord:
smg
SA: NP(Art(indef) " N Pl
smg sing] x
)-N-
SC: NP<Art<indef - pl Pl V
202 Part two: Morphology

Morphological realization rules:

1 MAN + pi -*■ men
2 GOOSE + pi —»geese
3 N + sing N
4 BOY boy
5 GIRL girl
6 MAN man
7 GOOSE goose
8 pi -s
9 RUN + past ran
10 V + non-past —> V
11 RUN run
12 YAWN yawn
13 WALK walk
14 sing -s
15 past ■ -ed
16 def ■ the
17 indef + sing —» a
18 indef + pi -* some

This grammar produces trees like Figures 83 and 84.

NP VP

Verb

V Tense

YAWN past
| |

(12) (15)
1 1
1 1
some men yawn -ed

Figure 83 (some men yawned)

 The analysis of words 203

1 Subject-verb concord does not operate since verbs only

concord in number, according to our rule, when they are
non-past, and when the subject is singular: neither condition is
satisfied in our tree.
2 Article-noun concord does operate since the article is
indefinite and the noun is plural - article concord is shown in the
tree.
3 The relevant morphological realization rules are identified
in the tree by the number of the rule that has operated.

NP VP

Art Noun Verb

def N Num V Tense sing

BOY sing WALK non-past

I I I I
(16) (4) (3) (13) (10) (14)

I I
walk
I
-s
the boy

Figure 84 (the boy walks)

1 Subject-verb concord has operated since the subject NP is

singular and the verb non-past.
2 Article-noun concord has not operated since this rule only
applies to indefinite articles, and the article here is definite.
3 The relevant morphological realization rules are identified
by the number of the rule that has operated. Note that {sing}
in the NP (rule 3) and {non-past} in the V (rule 10) have no
realization.

Before closing this chapter it is useful to point out some

implications of our grammar. There is nothing ‘given’ about our
model: it is the result of analytic decisions taken about various
204 Part two: Morphology

aspects of the structure of sentences and their constituents.

Different decisions would have resulted in a different grammar.
Note also that the formalization of the grammar makes totally
overt those analytic decisions we have made: for instance our
observations on the scope of concordial relations, and to which
category the categories of number and tense ‘primarily’ apply.

Technical terms:
agglutinative rules null-realization rules
concord rule ordering
fusional rules tense
number
14 The order of morphs and the
order of morphemes

Hitherto we have assumed that we should represent the

morphemes in a morphemic description in the same order as the
morphs which realize them. Thus we have represented the
morphemes realized by a form like walked as {walk + past}.
Morphs need to be ordered in the sequence in which they
actually occur, since in a form like walk-ed the morphs always
appear in that order, never in the impossible order *ed-walk.
However, if the morpheme is an abstract unit, why should we not
order the morphemes in a way that makes our grammatical
statement most revealing and economical? Thus in the grammar
we may order the morphemes that are realized by walked as {past
+ walk} if there seems good reason to do so: we have already
seen that there is good reason in the discussion of pages 110-11.
The realization rules that relate morphemes to morphs ensure
that the morphs occur in the correct order. We will look at the
morphology of English verb forms in a moment. First let us
consider a different, and more straightforward example. Consider
the verb forms in the following Akan sentences.1 The second
word in each sentence is the verb word, which has been
morphologically analysed. The verb stem is fura (FURA ‘wrap
round, put on, wear’). The descriptions to the left of each
sentence are the names used to refer to the various verb forms,
and we suppose that we can regard these as descriptions of the
relevant morphemes: for example in 1 we regard the verb word as
realizing the morphemes {hab + FURA}.

1 hab(itual) Kofi fura ntoma ‘Kofi wears a cloth’

2 prog(ressive) Kofi re-fura ntoma ‘Kofi is putting on a
cloth’

1 The sentences are represented in the orthographic form. Tone markings have
been omitted, although these are also relevant to a complete description of the
verbal system. This does not affect the point at issue.
206 Part two: Morphology
3 fut(ure) Kofi be-fura ntoma ‘Kofi will put on a
cloth’
4 pret(erite) Kofi fura-a ntoma ‘Kofi put on a cloth’
5 perf(ect) Kofi a-fura ntoma ‘Kofi has put on a
cloth’

The habitual verb form has no affix; there is simply the bare
root jura. For all the other forms there is an affix; it is a prefix in
all cases except the preterite, where it is a suffix. Also note that
all these affixes are mutually exclusive: no form can have more
than one of these affixes - there are no forms *a-re-Jura,
* be-fura-a, etc. It appears that we have here a five-term category,
which we call the category of Aspect. It is clearly most
illuminating to have a rule of the following kind:

6 Verb Asp + V (Aspect)

hab
prog
Asp fut •
pret
perf

This rule places the {pret} morpheme before V. We therefore

need a rule to reorder these two morphemes into the correct
order for the realization of the relevant morphs. Such a rule
might be:

7 Preterite hopping:
pret + V -» V + pret

It is not easy to devise a set of simple rewrite rules that would

achieve the correct ordering, and still maintain the generalization
that all five terms belong to the same category of Aspect: the
reader is invited to try. The rules we have proposed, on the other
hand, allow us to make the correct generalization in the grammar
6, with a subsequent rule 7 to accommodate the facts of morph
order in the actual structure of word forms.
The principle is to enable the grammar to make the
appropriate generalizations about the classification of various
grammatical morphemes into appropriate categories, in this case
that {hab, prog, pret, . . .} are all members of the category of
Aspect, independent of the quirks of word formation.
Let us now consider the morphology of English verb forms. We
 The order of morphs and the order of morphemes 207
have already discussed some aspects of this (pages 105-11). There
we established the description shown schematically in Figure 85.

Of the auxiliaries, only Tense is obligatory (since all verb groups

in simple sentences are tensed) and this category has two terms
{past} and {non-past}. The other auxiliaries are optional, but if
they occur they must occur in the order shown. The category
Modal can be realized by any one of the modal auxiliaries (a
form of CAN, WILL, MAY, etc.). The category Perf is realized by
a form of HAVE, which form depending on what other auxiliary
immediately precedes it, and is followed by a verb in the past
participle form: this we represented by the notation pp. Similarly
the category Prog is realized by a form of BE, which form
depending on what precedes, and it is followed by a verb in the
-ing form. The realization of Perf and Prog is thus discontinuous:
they are realized by two non-contiguous morphs. We can
represent this, diagrammatically, as in Figure 86, for the verb
group has been walking.

non-past Perf

Figure 86
208 Part two: Morphology

has is formed by the fusion of non-past + HAVE, been and

walking are formed agglutinatively by the suffixation of -en and
-ing to the stems be and walk. Perf is realized discontinuously by
(ihas . . . -en), and Prog by (be .. . -ing).
The most economical grammar to deal with these facts
supposes that a constituent structure rule introduces those
elements that are later to be discontinuous as a single constituent;
the discontinuity is handled by a later reordering rule. Let us see
how this might be managed for the examples we already have.
We propose the rules in Figure 87 below:

Constituent structure rules:

S -» NP + VP
NP Art + N + Num
| sing)
Num -
iPl 1
VP Aux + V + NP
Aux Tense (+Modal) (+Perf) (+Prog)
Modal {can, will, may, . . .}
{past
Tense
{non-past J
Perf have + pp
Prog B E + -ing

Lexical rule:
Figure 87 V -* {walk, .. .}

These rules produce strings like:

8a past + walk
8b past + Perf + walk: past + have + pp + walk
8c past + Perf + Prog + WALK: past + HAVE + pp +
BE + -ing + WALK
8d non-past + Modal + Prog + walk: non-past + will
+ BE + -ing + WALK etc.
To account for the discontinuities we propose a rule:

9 Affix hopping:
Any affix which is found to the left of a verb is to be
moved to the right of that verb. An affix may hop a verb
only once, past, non-past, pp and -ing are affixes; any
 The order of morphs and the order of morphemes 209

modal have, BE and any item categorized as V and

introduced by a lexical rule are verbs.
Applying this rule to the strings in 8 we get:

10a past^WALic => walk + past (walk-ed)

10b past + HAVE + pp + WALl > HAVE + past +

WALK + pp (had walk-ed)

10c pa^T+^HAVE + ppfT~BE^ + WALlC =>- HAVE

+ past 4- BE +pp+ WALK + -ing (had be-en
walking)

lOd non-pasTT^wTul + BE + -ing^r WALK => WILL +

non-past + BE + WALK + -ing (will be walk-ing)

(Since 9 is formulated as a transformational operation, we have

used the symbol => in the structures in 10.)
We now need morphological realization rules. For the data we
have been considering we can propose rules of the following sort
(which are formulated as ‘orthographic rules’):
11 HAVE + past had
HAVE + non-past have
B E + past was
BE + non-past is
BE + pp been
HAVE have
BE be
can + past could (and similarly for
CAN + non-past can will and may)
PP -ed
past -ed
WALK walk

Our examples now become:

12a walk-ed (the past form of walk)
12b had walk-ed (the past perfective form of walk)
12c had been walking (the past perfective progressive form
of walk)
12d will be walking(the non-past modal (will)
progressive form of walk)
210 Part two: Morphology

The reader is invited to try other combinations as permitted in

the rules.
We now try to accommodate within this type of description the
facts of number agreement between subject and verb. We have
discussed some aspects of this problem already, pages 186ff., but
two amendments are needed to bring our account into line with
our new model. The first is needed to deal with number
agreement as it affects all the verbal auxiliaries (our previous
account dealt only with verb groups with a single verb, past or
non-past - forms like walk, walks, walked). The second has to do
with the order of morphemes: our previous account supposed
that tense was introduced following the verb, and we have now
decided to place it as the first auxiliary: the changes that this new
analysis requires are relatively trivial.
The facts of English number agreement are complex and can,
as we have seen, be analysed in a number of different ways. In
line with our previous discussion, we will suppose that verbs only
agree in number when there is an overt marker of number on
one of the auxiliaries or on the main verb. With this assumption,
the facts can be summarized as follows:

i Number agreement is only shown on the first item in the

verb group, whatever that is:

13 The man walks/the men walk

The man is walking/the men are walking
The man has walked/the men have walked

Note that the first item in the verb group is also the item that
shows tense. Tense and number go together, even in
interrogatives:

14 Does the man walk/do the men walk?

Is the man walking/are the men walking?
Has the man walked/have the men walked?

We say that tense and number are cumulated, i.e. realized

together, on the first verb in the verb group.

ii Modal verbs never show number agreement:

15 The man can go/the men can go

The man could go/the men could go
 The order of morphs and the order of morphemes 211
iii With non-past verb forms, when the subject is singular,
HAVE, BE and the main verb show number agreement:

16 The man has gone

The man is going
The man goes

iv With non-past verb forms, when the subject is plural, only

BE shows number agreement (but not HAVE or the main verb,
which are in their stem forms)

17 The men have gone

The men are going
The men go

v With past verb forms, only BE shows number agreement in

both singular and plural:

18 The man was going/The men were going

The man had gone/The men had gone
The man went/The men went

We can summarize these facts in terms of the three rules:

19 Verb Number agreement (obligatory)

19a S A: X - N - sing - non-past - V - Y

SC: X - N - sing - Tense(non-past + sing) - V - Y

(on condition V is not a modal)

19b S A: X - N - pi - non-past - BE - Y
SC: X - N - pi - rr0„co(non-past + pi) - be - Y
Tense
sing
19c SA: X-N — past — BE Y
Pi
sing" sing Is
SC: X - N- BE - Y
Pi J Tensed + pi Y
The summary i-v above is expressed in these rules as
follows, i is shown by the fact that the number morpheme is
adjoined as a sister to tense (and dominated by tense) showing
that the two are cumulated. The generalization is expressed like
this so that in formulating the question rule we can say that tense
hops over the subject NP, and tense brings number along with it
(cf. 14). ii is shown by the fact that modals are specifically
212 Part two: Morphology

excluded from the scope of any of the rules proposed, iii-v

are directly represented in the rules 19a-19c respectively.
The number agreement rule, expressed this way, simplifies our
morphological realization rules, though it clearly leads to a
complex statement of the number agreement rule. It is possible to
state the number agreement rule in a different way. We can, for
instance, assert that all verbs agree in number with their subjects:
but this complicates the morphological realization rules since we
then must say, of for example modals, that whereas they agree in
number with their subjects (as stated by the rule), number
agreement is never realized in any way. It seems preferable to
formulate the rules as we have done in 19 so number agreement
is only held to occur when there is an overt marker of this
agreement.
We can also accommodate in this description the facts of
interrogative and negative sentences as discussed on pages
112-18. There we proposed (cf. discussion on page 127) that the
morpheme {neg} should be adjoined to the tense morpheme.
Note that just as number cumulates with tense, so too does the
negative morpheme; as can be seen most clearly in interrogative
negative sentences.

20 The man hasn’t gone/the men haven’t gone

The man doesn’t go/the men don’t go
Didn’t the man go?

Finally observe that the introduction of a passive auxiliary by a

passive transformation of the sort envisaged in chapter 8 (cf. page
126) falls quite naturally within the scope of our rules. We can
consider this by the derivation of the sentence:

21 Weren’t the dogs bitten by the cat?

We start from the active structure (corresponding to the cat bit

the dogs) shown in Figure 88 (some details are omitted). From
this we derive the passive structure (by the passive rule page
125) corresponding to the dogs were bitten by the cat, as shown
in Figure 89. From this we derive the negative structure (by the
rule on page 126) corresponding to the dogs weren’t bitten by
the cat, shown in Figure 90.
 The order of morphs and the order of morphemes 213

CAT sing Tense BITE Art N Num

Figure 88 past DOG pi

Figure 89 CAT sing

NP VP

Art N Num Aux V PP

DOG pi Tense Pass BITE

A A
past neg BE pp Art N Num

Figure 90 CAT sing

214 Part two: Morphology
We apply the number agreement rule 19c (the one that applies
here) to derive the structure shown in Figure 91:

Figure 91 CAT sing

Finally we apply the interrogative rule (page 127) and derive the
structure in Figure 92:

Aux NP VP

Tense Pass Art N Num Aux V PP

Al
past neg pi BE
II
DOG pi
I I Ax
pp BITE by NP

Art N Num

sing

weren’t the dogs bitten by the cat

Figure 92

The effect of the morphological realization rules is shown below

the structure in Figure 92.
 The order of morphs and the order of morphemes 215
Technical terms

affix hopping
cumulation
number agreement

Exercises

1 Noun classes in Luganda (a language from Uganda)

In Luganda, as in other Bantu languages, nouns can be analysed

as having the structure prefix + stem (e.g. omu-kazi). The
alternation between singular and plural is shown by different
prefixes (e.g. omu-kazi ‘woman’, aba-kazi ‘women’). Nouns can
be assigned to ‘gender classes’ depending on the form of the
singular and plural prefixes. Four different noun classes are
illustrated in the data. They have been assigned arbitrary
numbers for ease of reference.

singular gloss plural stem

omu-kazi woman aba-kazi -kazi
omusajja man abasajja -sajja
umuntu person abantu
omwana child abaana
omwezi sweeper abeezi
omwozi washerman aboozi
egkoko chicken egkoko -koko
erjgo oil erjgo
ente cow ente
endiga sheep endiga
embogo buffalo embogo
emmese rat emmese
omugga river emigga
omuggo stick emiggo
omwaka year emyaka
omwezi moon emyezi

ekisolo animal ebisolo

ekintu thing ebintu
ekyalo village ebyalo
ekyenyi forehead ebyenyi
216 Part two: Morphology
Segment the words into morphs. Write the stem form in the
right-hand column above. Fill in the matrix below with the prefix
forms. When a prefix is realized by two or more allomorphs,
there is a space for each allomorph. State the appropriate
environment that conditions the allomorphs. Make your
statements as general as possible. In the worked example, _C
means ‘in the environment preceding a consonant’ and_V means
‘in the environment preceding a vowel’.

class singular plural

in the environment in the environment

1 omu- _C

omw- _V

2 The verb word in Luganda

In Luganda the verb word is usually marked with a prefix which

represents the subject, and with a morph which shows the ‘tense’
of the verb; it may also, in certain circumstances, be marked with
a morph which represents the object. Each of the verb words
below could stand as an independent sentence, with the meaning
glossed. In this exercise you are asked to analyse the verb words
below:

1 ndigenda I shall go
2 yagenda He went
3 nkyagenda I am going
4 twagenda We went
5 aligenda He will go
6 wagenda You(sg) went
7 baligenda They will go
 The order of morphs and the order of morphemes 217
8 mukyagenda You(pl) are going
9 mwagenda You(pl) went
10 tukyagenda We are going
11 baagenda They went
12 nnagenda I went
13 akyarjkuba She is beating me
14 baamukuba They beat him
15 bakyakulaba They are looking at you(sg)
16 yatukuba She beat us
17 ndibagoba 1 shall chase them
18 akyandaba He is looking at me
19 tukyabalaba We are looking at them
20 olitugoba You(sg) will chase us
21 twabalaba We saw you(pl)
22 tulibalaba We shall see you(pl)
23 nnabakuba I beat them
24 mulindaba You(pl) will see me

(a) Give the morphs associated with the following translation

meanings: where a given morpheme is realized by two
allomorphs, state the environment (i.e. what conditions the
choice of one allomorph rather than another).

subject pronouns
in the environment
I

you(sg)

he/she

we

you(pl)

they
218 Part two: Morphology

object pronouns
in the environment

me

you(sg)

him/her

us

you(pl)

(hem

tense/aspect affixes

in the environment
future

past

progressive

Verb stems
in the environment

‘go’
‘beat’

‘chase’

‘see /look
at’

(b) What is the order of affixes within the word?

(c) Supply the probable forms for the following meanings:

I shall beat them You (pi) saw me

 The order of morphs and the order of morphemes 219

3 Syntactic analysis

Here is an extensive morphological and syntactic exercise. The

data comes from a Xhosa (a Bantu language of Southern Africa)
folk tale.
The data are presented in three lines. In the first line is the
language text; the second line is a fairly literal ‘word-by-word’
gloss into English; the third line contains a more idiomatic
English translation. You should beware of taking the glosses
literally, since the grammatical structure of the language differs
from English and a full word-by-word translation is not possible
without a lot of further explanation.
Following the text are a number of questions designed to help
you towards an analysis. You will not be able to make an
exhaustive grammatical description of Xhosa on the basis of the
amount of data you have here. The texts do, however, have
enough data to give some idea of the structure of the language as
a whole.

Amasela (Thieves)

1 Ngenye imini, udyakalashe nomvolufu babona inqwelo yeentlanzi

One day Jackal and Wolf they saw a wagon of fishes
One day Jackal and Wolf saw a wagon offish
2 Kwa oko udyakalashe wathi “Mvolufu, ndilambile. Ndifuna intlanzi.”
Then Jackal he said Wolf I hungry I want a fish
Then Jackal said, ‘‘Wolf I am hungry, I want a fish.”
3 Mvolufu wathi “Uya kuzifumana njani iintlanzi?” Udyakalashe
Wolf he said You will find them how the fishes Jackal
Wolf said, ‘‘How are you going to find the fish?” Jackal
4 wathi “Uya kubona’’. Udyakalashe yena wabaleka waya kulala
he said You will see Jackal he he ran he went to he down
replied, ‘‘You will see.” Jackal, he ran and lay down in the road
5 endleleni phambi kwenqwelo. Umqhubi wasibona esi silo
in the road in front of the wagon Driver he saw this creature
in front of the wagon. The driver saw this creature
6 wacinga ukuba sifile. Wasiphakamisa ke wathi “Ndiya kumhlinza
he thought that it dead He picked it up then he said l will flay him
and thought it was dead. Then he picked it up and said, “/ will flay
1 lo dyakalashe ekhaya. Ndiya kwenza umnweba ngesikhumba.
this Jackal at home I will make a caross with the skin
this Jackal at home and make a caross with the skin
8 ndiwuthengise, ndifumane imali.” Kwa oko
I sell it I find money Then
and sell it to make money.”
220 Part two: Morphology

9 wamphosa enqwelweni waya kuhlala esihlalweni sakhe

he threw him on the wagon he went to sit down on the seat of him
Then he threw him on the wagon, went and sat down on his seat
10 waqhuba inqwelo. Wavuka udyakalashe, wathatha iintlanzi
he drove the wagon He got up Jackal he took the fishes
and drove on. Jackal got up and took the fish
11 waziphosa phantsi emva kwenqwelo, watsiba ke enqwelweni,
he threw them down behind the wagon he jumped then from the wagon
he threw them down behind the wagon. Then he jumped off it
12 wahlala phantsi, watya. Umvolufu wayibona le nto. Wabaleka
he sat down he ate Wolf he saw it this thing He ran
and sat down and ate. Wolf saw this happening and ran
13 kudyakalashe, wafika wathi “Dyakalashe, nam ndilambile”.
to Jackal he arrived he said Jackal, I also 1 hungry
to Jackal. On arrival he said, “Jackal, I am hungry too.”
14 Waphendula udyakalashe wathi “Mvolufu, ukuba ulambile, yiya
he answered Jackal he said Wolf that you hungry you go
Jackal answered, “Wolf, if you are hungry, go
15 kulala phaya endleleni. Umqhubi uya kucinga ukuba ufile.
to lie down there in the road The driver he will think that you dead
and lie down there in the road. The driver will think that you are dead.
16 Uya kukuphakamisa, akuphose enqwelweni aye kuhlala esihlalweni
He will pick you up he throw you on the wagon he go to sit on the seat
He will pick you up, throw you on the wagon and go and sit on his seat
17 sakhe aqhube inqwelo. Wena ke uya kuvuka, uthathe iintlanzi,
of him he drive the wagon You then you will get up you take the fishes
and drive the wagon. Then you get up and take the fish
18 uziphose phantsi, utsibe ke enqwelweni, uhlale phantsi,
you throw them down you jump then from the wagon you sit down
and throw them down. Then you jump down from the wagon, sit down
19 utye. Umvolufu walithatha eli cebo, wabaleka waya kulala
you eat Wolf he took it this advice he ran he went to lie down
and eat. Wolf took this advice, and ran and went and lay down
20 endleleni. Kodwa umqhubi waqonda ezi zilo
in the road But the driver he understood these creatures
in the road. But the driver now knew that these creatures
21 zamqhatha. Wambetha kakhulu umvolufu emzimbeni.
they were cheating him He beat him hard Wolf on the body
were cheating him. So he beat Wolf hard on his body.
22 Umvolufu wakhala, wavuka, wabaleka. Udyakalashe sisilo
Wolf he cried out he got up he ran away Jackal it a creature
Wolf cried out, got up and ran away. Jackal is a creature,
23 nomvolufu sisilo naye. Zizilo zonke. Kodwa udyakalashe
and Wolf it a creature too They creatures both But Jackal
and so is Wolf. They are both creatures. But Jackal
24 uhlakaniphile, umvolufu sisidenge.
he clever Wolf he stupid,
is clever and Wolf is stupid.
 The order of morphs and the order of morphemes 221

(a) Make a list of all the verb stems in the data, and gloss each
item (e.g. bona ‘see’). List the stems in the form in which they
occur, or in which you think they would occur, in the simple past
tense form. Do not try to account for forms which are translated
in the free English translation by ‘I am —’, ‘They are —’, ‘I want
—etc.: i.e. forms with the ‘present’ tense (e.g. ndilambile 2,
ndifuna 2 and the forms in 22-4).

(b) Diagram the structure of past tense verb forms in terms of the
morphs that obligatorily and optionally occur. Make a list of
subject and object morphs, and gloss them in some way so that
the relevant concordial relationships are clear.

(c) Diagram the structure of those verb forms that are apparently
(judging from the glosses provided) used to refer to future events
etc. Note that there are two types of form (e.g. in sentence
beginning line 16). Describe their distribution.
(d) There appear to be three different types of locative
expression (they translate such phrases as ‘on the road’, ‘in front
of the wagon’, etc.). Diagram their structure. (Note the following
nouns: umzimba ‘body’; isihlalo ‘seat’; indlela ‘road’.)
(e) Make an analysis of the sentence Uya kukuphakamisa . . .
aqhube inqwelo (lines 16-17) into morphs. Separate the morphs
in each word by hyphens (e.g. u-dyakalashe) and identify each
morph (e.g. Personal name class prefix-jackal). Ignore the word
sakhe.
15 Lexical morphology

On page 171 a distinction was drawn between lexical or

derivational morphology and inflectional morphology. Processes
for the formation of new words, in the sense of new lexemes, is
the field of lexical or derivational morphology. Thus from write
we can derivationally derive writer and rewrite. On the other
hand the formation of forms like writes, wrote is a matter of
inflectional morphology. In these terms the word writing is a
derivational formation in the sense ‘handwriting’ (Your writing is
illegible), but an inflectional formation when it is part of a verb
group (He is writing a letter to his girl friend). Inflectional
morphology is a matter for the grammar as we have seen, but we
shall argue that derivational morphology is not. The boundary
line between the two is not entirely clear-cut but in this brief
account we will assume that a distinction of the sort drawn above
can be made.
It is obvious that many lexemes are morphologically complex,
as writ-er, re-write above. It also seems to be the case, as noted on
page 171 that, from the point of view of the grammar, we
need to treat items like this as units. This suggests that we should
not treat -er and re- in the examples as morphemes, in the sense
in which we have defined morpheme. We call them lexical
formatives. The stem to which lexical formatives are affixed we
call a lexical stem. Lexical stems may be morphologically simple,
as in write, or may themselves be complex, for example the stem
in un-realistic. The derivational process by which this latter
example is formed proceeds as follows: to the adjective stem real,
which is a root, and morphologically simple, we affix the
derivational suffix -ist to form the noun realist. To this stem,
which is complex, we affix the derivational suffix -ic to form the
adjective realistic. Finally we prefix the derivational affix un- to
form the adjective unrealistic. This suggests that the word has the
internal constituent structure represented in Figure 93.
 Lexical morphology 223

Adj

un- Adj

Adj
A -ist

Figure 93 un real
I I
ist ic

The fact that words can be analysed as having an internal

constituent structure of the sort shown in Figure 93 suggests some
general ‘word formation rules’, of this form:

la Adj + -ist => N real-ist

lb N + -ic => Adj angel-ic
lc un- + Adj => Adj un-real
Id N + -ish => Adj girl-ish
le Adj + -en ^ V rip-en
If V + -al => N refus-al
lg V + -ion => N confus-ion
lh N + -hood => N man-hood
Rules of this sort need to be supplemented by further rules
relating to the phonological and orthographic changes that
accompany them. Phonological changes can be illustrated by lb
and lg. The noun angel is stressed on the first syllable, but in the
corresponding adjective, angelic, the stress has ‘shifted’ to the
second syllable. The final segment of the verb confuse /kanfjuz/
is ‘changed’ from /z/ to /•$/ in the noun confusion /konfju3n/.
Orthographic changes can be illustrated by le-lg. In these cases
the stem form loses its final -e before the affix is added:
refuse :rejus-al. Many such phonological and orthographic
‘changes’ are subject to rules, but we shall not explore these
here.
Rules of the sort shown in 1 capture some valuable
generalizations about the structure of lexical items. We have not
formulated the rules as rewrite rules, which rules do not here
seem appropriate. A rewrite rule of the sort:

2 N —> Adj + -ist

224 Part two: Morphology
captures the wrong generalization: we do not want to say that a
noun has the constituents Adj + -ist (which is not generally true)
but rather that in appropriate cases a noun may be formed from
an .adjective stem by the suffixation of -ist: the generalization
captured in rule la. We consider rules like those in 1 as rules of
word formation belonging to the lexicon (i.e. they are rules for
forming new lexical items) rather than rules that belong to the
grammar as we have previously considered grammar. There are a
number of reasons for this.
In the first place, formation rules like 1 are of limited rather
than general productivity. By productivity we mean the general
applicability of a rule. Inflectional rules are of general
productivity - the rule that affixes {past} to a verb stem is one
that can apply to all verb stems. On the other hand rules like If
or lg are of limited productivity, because it is not the case that
any V undergoes these rules. The rules are, however, productive
insofar as they apply to a range of stems (arrival, rebuttal,
perusal, acquittal, . . .; expectation, action, procession, edition,
. . .). Not only are such rules of limited productivity, but they also
apply to lexical stems in a somewhat arbitrary fashion. While rule
If applies to many verb stems (arrive, rebut etc.), it clearly does
not apply to all (there are no forms expect :*expectal; act:*actal).
Stems like expect and act must instead derive nouncs by rule lg
(expectation, action). In general, if a verb stem undergoes one
formation process (like refuse), then an alternative process is not
open to it (*rejusion). In a few cases both forms are available
(remit, remittal, remission) - and here we note that each of the
derived forms has a different sense. Different rules are
productive to different degrees. The suffixation of -y (length-y,
fox-y) is still an active productive process, as in contemporary
spoken forms as trendy, or a very linguisticy sort of solution. So
too is the suffixation of -ish, established in words like girlish,
bookish, and also found in spoken forms like five o’clockish, a
very smokerish reaction (‘the typical reaction of a smoker’), etc.
Conversational forms like smokerish, linguisticy suggest that a
rule like Id is an active part of some speaker’s word formation
rules. By contrast the derivational process that relates long
and length (Adj + -th => N (with appropriate vowel change) -
strong .strength', wide '.width) seems no longer productive.1

1 In some cases words appear to have been derived by a productive process,

 Lexical morphology 225
Word formation rules, then, are of limited rather than general
productivity; some are still productive, other fossilized; and they
have a somewhat arbitrary application to particular stems. In all
of these respects they contrast with the centrally grammatical
rules, such as the rule forming noun plurals, which are generally
productive and of wide application.
We have already noted a second reason (page 171): from the
point of view of the grammar, the process by which a lexeme is
formed is not relevant. For the adjectives STUPID and FOOLISH
in the environment a_boy, it is not relevant that STUPID is a
root and FOOLISH a derived form. No grammatical rules need to
have access to the information that FOOLISH is derived from a
noun root. Thus in Figure 93 grammatical rules operate only on
the topmost Adj node: the subordinate N and Adj nodes are not
subject to grammatical rules. The rule ascribing number to a noun
cannot apply to the string realist when it is part of the adjective
realistic (there is no form * unrealistic), nor can the string realistic
be compared when part of the form unrealistic (*unmorerealistic).
Indeed it is only confusing the grammar to include such
information in a string to which grammatical rules apply, since we
would need to put special restrictions on our rules. For example,
we would need to state that the rule ascribing number to a noun
does not apply when a particular noun stem is itself dominated, as
in Figure 93, by an Adj node. This observation is confirmed by the
fact that grammatical affixes are usually more ‘peripheral’ to a
word than ‘derivational’ affixes. Schematically, we find strings like
GA + DA + STEM + DA + GA (where GA = grammatical
affix and D A = derivational affix) but not strings like *D A + G A
+ DA + STEM + GA + DA, where the grammatical affixes are
less peripheral than the derivational affixes. In English, for
instance, we find morph-eme-ic-iz-ed (STEM + DA + DA +
DA + GA) but not say *morph-eme-s-ic-iz-ed (STEM + DA +
GA + DA + DA + GA) where the number affix (-5) is
sandwiched between two derivational affixes. This generalization
appears true of many, perhaps all, languages, and demonstrates
the point: grammatical and word formation rules need to be

when this is not in fact the case. Thus we have aggression, but no * aggress-, butler
and vicar, but no *buttle or *vic (though in fact all of the starred forms can be
found — the latter two in novels by P. G. Wodehouse). The derivational process by
which they are assumed to have been formed is called ‘back formation : the
analogy of think-.thinker suggests the comparable *vic:vicar.
226 Part two: Morphology
separate. The grammatical rules are part of the grammar, and it is
not relevant to the grammar whether a given lexeme is a root or a
derived form. It seems more appropriate to put the lexical
derivation rules in the lexicon.
Thirdly, some of the rules in 1 involve recategorization. Thus,
la, lb, ld-g take a stem of one category and produce a lexeme
that is assigned to another category: so the ‘class changing’ rule
la derives nouns from adjectives. Others of the rules, like lc, lh,
are ‘class maintaining’: the derived category is the same as the
original category. We do not in general wish to have rules like
this within the grammar that change items from one major lexical
category to another.
A fourth reason has to do with the semantic properties of rules
such as those in 1. A derived lexeme generally has some semantic
relation to the lexeme from which it is derived, usually through the
root which both lexemes share. It is not clear, however, what
generalizations can usefully be made about which aspects of the
meaning of the stem are reflected in the derived form - compare
forms like boy '.boyhood; knight'.knighthood; sister'.sisterhood
etc. Similarly the derivational affix often has some general
meaning, but any statement of the meaning of an affix will have
to be stated in extremely general terms, to which there are
exceptions. For example, nouns formed by rule lh are typically
abstract {man'.manhood) the connotations of -hood being ‘state,
quality or rank’ (Partridge 1958, 848), but Partridge also
notes the “resultant secondary senses, ‘concrete instance’ -
as in falsehood, and, from the nuance ‘rank’, ‘collectivity’
as in brotherhood and sisterhood”, A similar example is
remit'.remission'.remittal, noted earlier in the chapter. The two
noun forms clearly share some aspect of the meaning of the
corresponding verb, but the particular aspects of shared meaning
differ in the pairs remit'.remission and remit'.remittal. Such
changes of meaning are more appropriately accounted for in the
lexicon than in the grammar.
The examples considered so far have been examples of word
formation, where a lexical formative has been affixed to a lexical
stem. Another common source of new lexemes is composition or
compounding, illustrated by words like bookmark, teatime,
halfback. The distinction between compounding and derivation
rests on the fact that each part of the compound is itself a lexical
stem, and is typically a free morph - as is the case with book,
 Lexical morphology 227
mark, bookmark. By contrast, formatives are always bound
morphs.
In compounding there is often a considerable problem about
word division, and orthographic conventions offer little help:
sometimes compounds are written as a single word (waterfall),
sometimes hyphenated (water-drop), sometimes as two words
(water bottle) - the examples are all from the Shorter Oxford
English Dictionary. A recent history of English (Strang 1970, 39)
treats as ‘morphological units’: middle distance, market research,
cost effectiveness, standing ovation. The criteria used have been
discussed (pages 162-4): typically no further item can be inserted
between the two elements; they cannot be transformationally
separated; one element cannot be pronominalized or be taken as
the antecedent for pronominalization; etc. However, there is
little clarity as to what counts as a compound word and what is a
modifier-head construction where the modifier is a noun. Thus it
is curious that in the Shorter Oxford English Dictionary, of the
forms field mouse and harvest mouse only the latter has an entry
(under harvest)\ Compounding is an extremely productive source
of new lexical items in English, even though its boundaries are
unclear.
As a final consideration, we may ask how far one would wish
to carry the process of derivational analysis within words. Clearly
the basis for analysis is the fact that words share common
phonological material and a common element of meaning. Thus
strawberry is analysed as straw-berry on the analogy of dew-berry,
logan-berry, rasp-berry, etc. One can easily justify the berry
morph, and in the case of strawberry, the straw morph: logan-
and rasp- are more difficult and it would perhaps be best to
consider loganberry and raspberry as simple non-derived forms.
Another problem area involves ‘phonaesthemes’ (formatives with
phonaesthetic properties) in onomatopoeic words. Some analysts
have wished to make an analysis of such words. Thus Bloomfield
(1933, 245) sees ‘intense, symbolic connotations’ associated with
the ‘prefixes’ and ‘affixes’ in such words as:

3 fl- ‘moving light’: flash, flare, flame, flicker, . . .

gl- ‘unmoving light’: glow, glare, gloom, glint, . . .
si- ‘smoothly wet’: slime, slush, slip, slide, . . .
-are ‘big noise or movement’: blare, glare, flare, . . .
-ash ‘violent movement’: bash, crash, flash, slash, . . .
228 Part two: Morphology

An analysis of this sort can, with some ingenuity, be made quite

extensive. A similar and more problematical area involves the
etymology of words. Thus in the following set of words:

4 re-ceive re-cur * re-duct re-duce

de-ceive * de-cur de-duct de-duce
con-ceive con-cur con-duct con-duce
*in-ceive in-cur in-duct in-duce,

some analysts recognize ‘etymemes’ (formatives with etymo¬

logical relevance). As with phonaesthemes, the problem with
this sort of analysis is how far to go; the analyst with a
background in the historical study of a language might well wish
to go further than an analyst without such a background. For
ordinary speakers of the language there is no perception of the
etymology of many words (orchard deriving from Old English
ort-^eard, the first element probably deriving from Latin hortus
‘garden’; caterpillar said to derive from Old French chate ‘cat’ +
piller, pilour ‘pillager’!). Probably no one would wish to analyse
such extreme examples as these. Items like receive are
more problematical, partly because re- is still productive
(re-electrification) and partly because of the possible analogies
with deceive, etc., as shown above. On the other hand the ‘stem’
-ceive is no longer usable in any productive way: its status is
rather like logan- and rasp- mentioned above. It is probably best
to treat items like receive (whose internal structure relates to
etymological considerations which are no longer productive) and
items like flash (whose internal structure relates to phonaesthetic
considerations) as simple, underived lexemes.
Our conclusion then is this: it is useful to distinguish be¬
tween inflectional and derivational morphology. Inflectional
morphology deals with the distribution of categories introduced
by the grammar, and is general and productive. Derivational
morphology deals with the production of new lexemes, is less
productive and often arbitrary in application: as such it belongs
with the lexicon rather than the grammar. It can also be extended
to make generalizations about the internal structure of some
lexemes that otherwise one would wish to treat as units (cf. the
discussion of phonaesthesia and etymology).
 Lexical morphology 229
Technical terms

back formation lexical formative

class changing lexical morphology
class maintaining lexical stem
composition onomatopoeia
compound word phonaestheme
derivational morphology productivity
etymeme recategorization

Exercises

The relationship between kind and unkind (and other similar

pairs of words like helpful -.unhelpful, etc.) can be described by
the lexical rule un- + Adj Adj; prefixing the lexical formative
un- to an Adj(ective) stem yields another Adj. This is a
‘class-maintaining’ rule in that the base form and the derived
form both belong to the same class, in this case Adj. Consider
these words:

class purify simplification equality

legality glorification equalize simple
classify legalize popular centrality
glory beautify classification centralize
sterilization beauty popularity sterile
purity popularize legal pure
legalization central equal glorify

Analyse these words into morphs. Devise lexical rules to account

for the derivational relationships between the various words.
Which of the rules are class changing and which are class
maintaining? How productive are the rules? Make a note of any
orthographic regularities in the derivational processes. Can you
assign any general ‘meaning’ to the various lexical formatives you
identify (as, for example, un- in the example at the top of the
exercise is ‘negative’).

2 Nationality nouns and adjectives

A number of different morphological paradigms in English relate

country names (e.g. England, China); the names given to
230 Part two: Morphology

nationals of that country (e.g. Englishman, Chinese); the

adjective applied to that country (e.g. English, Chinese)', and the
language spoken in the country (e.g. English, Chinese). There are
other grammatical differences between paradigms: sometimes,
for instance, the name given to nationals has a distinct plural
form {one Englishman', two Englishmen) and sometimes it
doesn’t {one Chinese', two Chinese). What are the relevant forms
relating to the following countries?

England, Wales, Scotland, Ireland, America, Poland, Russia,

France, Hungary, Switzerland, Thailand, Portugal, Denmark,
Turkey, India, the Netherlands, Brazil.
16 Form classes and grammatical
categories

Previous chapters discussed the principles by which form classes

are established. The discussion assumed that the application of
the principles was relatively straightforward and would yield a
determinate set of classes. In this chapter we re-open the
question.
Earlier chapters identified two different types of
morpheme : lexical morphemes or lexemes, and grammatical
morphemes. Lexical morphemes were grouped together into
classes, called lexical categories, to which we assigned names like
Noun, Verb, Adjective, etc. We saw that in principle such classes
are ‘open’, in that new items can be added to them. So, for
example, sputnik has been added to the lexical category of Noun
in English. Grammatical morphemes too were grouped together
into classes, called grammatical categories, to which we gave
names like Tense, Number, etc. Each of these categories has a
finite number of terms; they are ‘closed’ in that new items cannot
arbitrarily be added to them. So, for example, the English
category of Number has the two terms singular and plural; it is
difficult to see how a new term, say ‘dual’, could be added since it
would have extensive repercussions for the whole grammar of the
language.
In discussion hitherto we have supposed that the distribution of
a grammatical category, like number, and of the terms of the
category, like singular and plural, is totally definable within the
grammar. This is shown by the fact that our constituent structure
rules develop structures which include such items as singular and
plural. By contrast, whereas the grammar can define the
distribution of some lexical category as a whole, it is unable to
define the distribution of individual members of the category. So,
for example, we can use the grammar to define the distribution of
the class Verb, but we cannot in the grammar define the total
distribution of some particular verb - this would require, amongst
232 Part two: Morphology
other things, a statement about the selection restrictions for the
verb in question, and we have not included such information in
the grammar.
A further distinction between lexical and grammatical
categories is this: lexical categories are realized either as free
forms (e.g. dog, walk) or as the stems of complex forms (e.g.
dog-s, walk-ing). Grammatical categories, on the other hand, are
very often realized as bound forms (e.g. s, -ing, -ed) though they
may on occasion be realized as individual words (e.g. the
auxiliary verbs). These facts are reflected in oppositions
commonly found in grammars between lexical (or ‘full’) words
and grammatical (or ‘form’) words - lexical words are typically
those that contain a lexical stem, and grammatical words are
those that realize grammatical categories alone. The distinction is
quite useful but not entirely clear-cut, and is not always easy to
draw. The fact that grammatical categories are often realized as
bound morphs leads, in many languages, to a typical association
between a lexical category and a grammatical category or
categories. So, for example, in English the category of Number is
realized on the Noun and the category of Tense is realized on the
Verb - an important point to which we return.
Considerations of this sort lie behind the establishment of a
system of ‘parts of speech’ for a language. Discussions of parts of
speech often distinguish between the ‘major’ (Noun, Verb,
Adjective, Adverb) and the ‘minor’ (Article, Preposition,
Conjunction, etc.). The major parts of speech are, or have stems
which are, lexical categories. In descriptions of particular
languages linguists have usually (but not invariably) been able to
identify classes to which they assign the names of the major parts
of speech. The minor parts of speech often realize grammatical
categories, and descriptions of different languages vary as to the
number of minor categories identified and the names given to
them. For example English is usually described as having a
category Article, but in many languages, Latin for instance, no
such category is identified.1

1 Such observations raise a host of interesting questions that we do not have

the space to explore. For example: to what extent is there a similarity between the
classes named Verb in Akan, Gaelic and English which justifies their being called
by the same name? Do all languages have a syntactic class to which we can
appropriately apply the name Verb? Those interested in pursuing this question
are directed to Lyons (1968, chapter 7) and the references there. An interesting
review of word classes in a variety of languages is Lingua, vol. 17, 1967.
 Form classes and grammatical categories 233

In traditional grammars, which do not recognize the morpheme

as a relevant linguistic unit, parts of speech are classes of words
(the traditional word corresponding to the orthographic unit). In
some more recent treatments, parts of speech are regarded as
classes of stems - a classification that works best with the major
parts of speech, with a lexical stem, but which also works with
other stems realizing only grammatical categories. The two
approaches are obviously somewhat different, but not necessarily
in conflict. Thus Hockett (1958, 221) defines a part of speech as:
A form class of stems which show some similar behaviour in
inflection, in syntax, or both. The part of speech system of a language
is the classification of all its stems on the basis of similarities and
differences of inflectional or syntactical behaviour.

Compare this definition with the definitions given for some of the
parts of speech in classical Greek by Dionysius Thrax, a Greek
grammarian of the first century BC - he is often regarded as the
forerunner of the western tradition of grammatical description
(see Robins 1967, 33):

The noun is a part of speech inflected for case, signifying a person or

thing.
The verb is a part of speech without case inflection, but inflected for
tense, person and number, signifying an activity or process performed
or undergone.
The preposition is a part of speech placed before other words in
composition and in syntax.

Both Hockett and Thrax mention two types of criteria:

inflectional and syntactic. As we have seen, inflection involves
the internal structure of a word form, perhaps by affixation of a
bound morph, etc. In the case of the major parts of speech
inflection typically involves the affixation, etc. of bound morphs
realizing grammatical categories to a lexical stem - as in Thrax’
definition of the Noun and Verb in Greek. Syntactic criteria are
distributional criteria, reflecting the external distribution of a
form. It was these two criteria that we used to establish form
classes on pages 34-5. Nowadays syntactic criteria are usually
held to be primary: in those languages that do not inflect at all
they are the only criteria available; in the case of languages in
which some but not all parts of speech inflect, they are the only
possible criteria for the non-inflecting parts of speech - as in
Thrax’ definition of the Preposition in Greek. When languages do
234 Part two: Morphology

inflect, however, inflections will obviously be an important

criterion.
Thrax’ definitions also include a semantic criterion lacking
in Hockett’s-(. . . ‘signifying a person or thing’ . . ., etc.). We
have already observed that we cannot rely on notional definitions
(see pages 60-6). A moment’s thought uncovers many forms
that are syntactically nouns but do not ‘signify a person or thing’
- ACTION, ACTIVITY, MOVEMENT, etc. Indeed nouns like this
‘signify an activity’, supposedly the criterion for verbs. We do not
take semantic criteria as criterial in establishing form classes. It is a
different matter when we come to naming a class; since it is
indeed the case, in for example English, that the formally defined
class of Noun includes many items that do ‘signify a person or
thing’, it seems sensible to retain the traditional name for this
class rather than inventing a new arbitrary name. The question of
naming (which may as well relate to some central semantic
feature, and will yield an easily remembered name) is a different
matter from the formal establishment of a class (which ignores
such criteria).2
To exemplify inflectional and syntactic criteria, consider how
we might identify the class Noun in English. English is not
particularly rich in inflections and the only inflectional criterion
we can apply is that members of this class co-occur with the
plural morph -s (or inflect for number by fusion man :men, etc.).
The inflectional criterion that they co-occur with the genitive
morph -s is only marginally useful in English since this item might
more profitably be considered to be an affix to the N P as a whole
rather than to the noun specifically (the woman next door’s cat =
the cat of the woman next door).3 Distributional criteria reflect the
ability of the noun to operate as head of the NP, and hence to
co-occur with articles, numerals, etc. within the NP. Then, as
NPs they have a distribution as subject, object etc. of the verb.
2 At one time it was fashionable to discard even the names Noun, Verb, etc.
since it was thought that these might smuggle in undesirable semantic
considerations. Thus Fries (1957) conducts a rigorous distributional analysis of
English and labels the classes he establishes with numbers and letters. It is no
surprise to discover that these classes, although given arbitrary labels, turn out to
be the familiar classes of Noun, Verb and so on. It is worth consulting Fries to see
how far such an analysis can be pursued.

3 Various derivational affixes are also limited to noun forms - like the -ness in
goodness, helplessness, etc. But since -ness, and others like it, is a lexical formative
and not an inflectional affix, we do not consider such items further here.
 Form classes and grammatical categories 235

If we apply these criteria to a set of items we find a class,

including man, boy, dog, CAT, etc., that will meet them all.
We call these the ‘central members’ of the class. Other forms
meet some but not all of the criteria. Some are ‘defective’ with
respect to the inflectional criteria: abstract nouns (SINCERITY,
warmth, beauty) and proper nouns (JOHN, FIDO) do not
occur in the plural; a few nouns occur only in the plural
(SCISSORS, TROUSERS, suds, etc.). Others are ‘defective’
with respect to some of the distributional criteria: proper nouns
do not typically occur with articles, and so on. However defective
their morphology may be, and however restricted their
co-occurrence possibilities with the NP are, all the items we have
mentioned occur as the head noun of an NP, and hence as
subject, object, etc. of the verb. These latter are perhaps the most
important criteria - note that they are syntactic. There are,
however, a few items that we would probably wish to identify as
nouns that barely meet even this minimal criterion: they are
items like HEADWAY, TABS (They kept tabs on him) that occur
only, or mainly, in idioms.
The preceding paragraph illustrates the principle used to
establish form classes - a given set of criteria identifies a
particular class. It also shows up a problem - particular items
meet these criteria to a greater or lesser extent. We can establish
a class of ‘central members’ that meet all or most of the criteria.
There remain various sub-classes that meet the criteria to a
greater or lesser extent, or, on the margins, hardly at all! To some
extent we can turn this fact to our advantage since it allows us to
identify sub-classes of the major category - in a previous chapter,
for instance, we established the class of Proper Nouns in terms of
the fact that they do not co-occur with the definite article. It
does, however, have a drawback: the categories have no ‘hard,
definable edges’. Let us give a non-linguistic example to define
the problem. Individuals may be assumed to have criteria by
which they assign a particular plant to the category ‘tree’ or ‘bush’.
There seem to be central members of these classes: we speak of
pine or beech trees, not bushes, and similarly of rhododendron or
lavender bushes not trees. But what of holly or hawthorn? There
is a comparable sort of indeterminancy between form classes.
We will consider this question by looking at the distinction
between adjectives and nouns, and then at the distinction
between adjectives and verbs in English.
236 Part two: Morphology

We have already outlined criteria for the class Noun.

Adjectives are usually defined in terms of four criteria, one
inflectional and the other three syntactic. The syntactic criteria
are: i that adjectives occur attributively as modifiers of the
noun in NP (a poor man)-, ii that they occur predicatively after
the copula (the man is poor)-, and iii that they can themselves be
modified by intensifying expressions (very poor, rather poor). The
inflectional criterion is iv: they occur in comparative and
superlative forms (poor-.poorer-.poorest). On the evidence
POOR is a central member in that it satisfies all four criteria. Less
central members include adjectives: that do not occur attributively
(*an afloat boat, *an ablaze building)-, that do not occur
predicatively (*The difficulty is same, *The reason is principal)-,
that cannot be modified by intensifying expressions (frather
afloat, *very major)-, and that do not compare (*more major,
*most major). These restrictions need not in themselves worry us.
The criteria for adjectives and nouns obviously do not overlap.
Now we come to the problem. We identified POOR as centrally
an adjective. Yet we find that it occurs in most of the
environments taken to be criteria for nouns: it co-occurs with
articles (the poor, those poor, etc.), and it can be subject or object
of a verb (The poor are always with us; He pities the poor, etc.).
There are, to be sure, restrictions - it does not for instance occur
in the plural (*The poors). But the ‘adjective’ POOR seems to
occur in more of the typical noun environments than a peripheral
‘noun’ like tabs! The converse is also true. CONCRETE, which
appears to be a noun, occurs in some of the adjective
environments: a concrete floor, the floor is concrete. It does not
meet all the criteria, however - it does not compare (*more
concrete, *most concrete) and can hardly occur with modifiers
(frather concrete). Here we find the ‘noun’ CONCRETE
occurring in more of the typical adjective environments than a
peripheral ‘adjective’ like afloat. How are we to tackle this
problem?
A number of possible solutions have been mooted. We discuss
four, and exemplify them with POOR.
The first is to say that we have two distinct, though
homophonous, items: poor(1) is an adjective and poor(2) is a
noun. One drawback to this solution is that POOR(2) still retains
some adjective-like properties - it can be intensified (He only
pities the very poor, cf. *He only pities the very man) and it can,
 Form classes and grammatical categories 237

just, be compared {He only pities the poorest, cf. * He only pities
the most man). Nouns cannot be intensified or compared.
A second solution is to say that POOR is ‘basically’ an
adjective, but can be ‘recategorized’ as a noun in certain types of
structure. A drawback of this solution is that we must then say
that some adjectives can recategorize easily (e.g. POOR,
beautiful, good), that others do not recategorize at all
(e.g. major, ablaze), and that yet others recategorize in
some circumstances (e.g. BLACK, PINK). This seems to involve
adding a statement in the lexicon for each adjective to indicate
the ease with which it can be recategorized.
A third solution is to assert that POOR is an adjective, but that
in certain structures additional material is ‘understood’; with this
solution the poor might derive by a transformational deletion rule
from a structure like those who are poor or those people who are
poor. This solution is attractive in that it sets out a semantic
reading of the usage of POOR concerned, but it has problems if
we extend the principle widely: I want the black might be related
to I want the black dress, I want the black ball (in billiards), I want
the black man, etc.: the ‘understood’ items depend on the context
of utterance.
A final solution is to say that in the poor are always with us
POOR functions as the head of an NP, and to relax our rules on
the structure of NPs to allow adjectives in certain circumstances
to take on this syntactic function. This solution involves
distinguishing the syntactic function of an item (as head of an
NP, etc.) from its syntactic class (as adjective, etc.). This final
solution seems attractive in the case of items like CONCRETE,
noted above, where, in a structure like a concrete floor, it
functions as the modifier of a noun head. Many items we wish to
class as nouns can occur as modifiers (a picnic basket, a bus
station, iron railings, etc.); it does not seem sensible in these cases
to identify either two items (PICNIC(I) an adjective, and
picnic(2) a noun) or to say that picnic has been
recategorized.4 This solution is also attractive in the case of
structures like the down line (a preposition modifier), the very
man (an adverbial modifier). The discussion seems to suggest that

4 This solution is not, however, open to the sort of grammar that we have been
considering so far: we have no satisfactory way of identifying grammatical
‘functions’ like modifier. The grammar is set up to account for the distribution of
categories. We return to this problem on pages 254-60.
238 Part two: Morphology

there is no unique satisfactory solution. Some instances are best

handled in one way, others in another.
We now turn to consider the relationship between adjectives
and verbs. We have already seen (pages 60-2) that the distinc¬
tion drawn in English between adjectives and verbs is not
drawn in the same way, if at all, in some languages: in these
languages items that are syntactically verbs translate into English
adjectives. What is the position in English? The inflectional
criterion for a class Verb in English is that the category of tense
and the ‘aspectual’ categories of progressive and perfect may be
realized on verb stems. The situation is complicated by the fact
that in English all these categories cannot be realized
simultaneously on the verb stem, instead auxiliary verbs are
introduced to carry them (see discussion on pages 107ff.).5
Syntactic criteria for verbs would include their distribution within
the VP, their ‘strict subcategorization’ possibilities, and so on.
We should also note that verbs can be subcategorized in terms of
inherent features (rather in the manner in which we
subcategorized nouns by features like [±animate], etc. on pages
85ff.). The feature most usually discussed in this connection is
[±state] (which we discuss in more detail in pages 288-92).
For now we note that some verbs do not readily co-occur with the
progressive auxiliary *He is knowing Chinese, *He is seeming
intelligent, or in the imperative (as a command) *Know Chinese,
*Seem intelligent. These, usually known as ‘stative’ verbs, can be
characterized as [+state]. Other verbs that do occur with the
progressive and in the imperative are known as ‘active’ or ‘action’
verbs; they can be characterized as [—state]: He is turning off
the light, Turn off the light, etc.
At first glance the criteria for verbs seem to be distinct from
those for adjectives. A few items can be used either as adjectives
or verbs {the grass is yellow, The grass is yellowing), and there are
a number of systematic derivational relationships between
adjectives and verbs (ripe: ripen, clear: clarify, etc.).
On the other hand the two classes do not overlap inflectionally.
The category of tense for example can be marked on verb stems
{He walks, He walked) but not on adjective stems {*He poors,

5 The situation is different in other languages where the verb stem can carry
both tense and aspect categories - see for example the brief exemplification of
Akan on page 60. Such languages often do not have a category of auxiliary verb;
Akan doesn’t.
 Form classes and grammatical categories 239

*He poored). But then we come to auxiliary verbs. Suppose we

say that the verb BE is introduced as a ‘dummy verb’ with
adjectives to carry the tense distinction (this sort of analysis
seemed appropriate to account for the occurrence of the ‘dummy
verb’ DO in negative and interrogative constructions on pages
116-17), giving *He walksn’t but He doesn’t walk; *Walks he? but
Does he walk? Perhaps, by analogy, BE is a ‘dummy verb’ for
adjectives - *He poors but He is poor, *He poored but He was
poor. If we adopt this analysis, we can treat adjectives in English as
another subclass of verb, distinguished by the fact that they require
the introduction of the dummy BE!
Such a proposal is not entirely fanciful (and indeed is adopted
by some linguists). If we adopt it, how do the other distinguishing
criteria for verbs work out? To begin with, adjectives, like verbs,
have a distribution within VP - both are ‘predicators’. We can
use much the same kinds of strict subcategorization frames for
adjectives as we can for verbs. Some verbs take sentential
complements (/ believe that the world is flat), others (RUN,
PAINT) do not. Similarly some adjectives take sentential
complements (It is odd that the world is flat), and others (red,
intelligent) do not. Some verbs are transitive and others are
not. Adjectives can be considered in a similar light if we note that
non-subject NPs in this case require prepositions: I like reading:
Reading is easy for me; I delight in syntactic analysis: Syntactic
analysis is delightful to me; I can’t understand the argument: The
argument is incomprehensible to me, and so on. A further similarity
involves the characterization [±state]. Just as we have stative
and action verbs, so we have stative and action adjectives: red
and POOR are [+state] (*/ am being poor, *Be poor);
CAREFUL and CLEVER may be [-state] (/ am being careful.
Be careful). Some verbal forms occur attributively like adjectives -
singing bird, roaring fire. We could derive NPs like these by
exactly the same rules as we used for attributive adjectives on
pages 142-3.
We note that verbs and adjectives are related; earlier we
showed that adjectives and nouns were related. Does this mean
that nouns and verbs are also related? Well, we could extend the
classification [±state] somewhat. Nouns like PERSON and
VIRGIN seem to be [+state] (*She is being a virgin, *Be a
virgin) in contrast to HERO (He’s being a hero, Be a hero). Simi¬
larly we note that some nouns take sentential complements, like
240 Part two: Morphology
RUMOUR, INSINUATION (.Have you heard the rumour that the
world is flat?), and some nouns do not, like CAT and BOY.
The position reached in the last paragraph does not seem very
helpful. If we group all these items into one large class, called,
say, ‘contentives’, then we would still need to distinguish between
those contentives that typically introduce BE to carry tense (the
former class of adjectives), those that permit tense to be shown
on the lexical stem (the former class of verbs) and those that may
occur with articles (the former class of nouns)! For English at
least it seems more sensible to retain the original classes and
recognize that the classification of stems is not wholly without
problems.
In any given sentence it is usually clear whether a particular
lexical stem is being ‘used as’ a noun or adjective or verb. We
were after all able to set up three separate and non-overlapping
criteria for these classes. The typical problem is whether we are
able to assign a particular stem to one or other class independent
of its use in a sentence, and this is not always straightforward. A
speaker, by using a given lexical stem in more than one class, can
make subtle distinctions of meaning; English may be unusually
liberal in the latitude it allows to individual stems to occur in
different classes. Other languages do not seem to allow quite such
freedom!

We now turn to a different though related problem: the re¬

lationship between sub-classes of a lexical category and the number
and type of grammatical categories that we may wish to establish
in a language. As a preliminary, we distinguish between ‘overt’
and ‘covert’ categories. Some categories are overt in that at least
one term of the category is identified by a formal marker of some
sort: in the English category of tense, the term past is overtly
marked by the morph -ed. Similarly, the term plural in the cate¬
gory of number is overtly marked by the morph -s affixed to noun
stems. Even though the singular term is not also overtly marked
we say that the category is overt. Such categories have other syn¬
tactic implications. Thus, for example, there are co-occurrence
restrictions between nouns and numerals that can, at least in part,
be related to the singular-plural distinction. The numeral one
cannot co-occur with plural nouns, and numerals other than one
cannot co-occur with singular nouns. Some verbs require non¬
singular subjects or objects (so we can have he gathered the men
 Form classes and grammatical categories 241

into a group but not *he gathered the man into a group), etc. The
category of number of English controls verb concord and, to a
limited extent, concord within the NP. We saw on pages 186ff.
that co-occurrence restrictions and facts like concord were impor¬
tant in establishing the category of number. Other categories may
be ‘covert’, in that no term in the category can be identified with
a formal marker: on the other hand co-occurrence restrictions or
other grammatical phenomena enable us to establish the exis¬
tence of the category. Thus the French category of gender is a
covert category: nouns, which are assigned to gender classes, are
not themselves usually marked for gender, but the gender is clear
from co-occurrence with, for example, article forms. There is no
overt marker of gender in a noun like chaise but it must co-occur
with the article la rather than le: although the category of gen¬
der is inherent in the noun, the actual formal marking of gender
occurs not on the noun but on the article which it is in construc¬
tion with. Covert categories are no less important in a description
than overt categories.
This particular distinction also brings us into direct confronta¬
tion with problems concerned with the relation between
‘notional’ and ‘formal’ categories. This can be illustrated by con¬
sidering the category of number in English. At first sight it seems
a fairly straightforward matter. The grammatical category of
number is an overt category with two terms, singular and plural,
shown by the alternation between pairs of forms like cat: cats.
The grammatical category correlates, more or less, with a
notional category which we may suppose also has two terms,
‘oneness’ and ‘more-than-oneness’ - we use cat to refer to a
single ‘cat’, and cats to refer to more than one ‘cat’. So far as it
goes this is correct, and the names given to the grammatical
category reflect these semantic facts. Note that the particular dis¬
tinction drawn in English is not universal; we have already noted
that some languages, classical Greek and Arabic for example,
have a three-term number category, with an additional term,
‘dual’, used to refer to things that typically go in pairs (arms, legs).
English has no grammatical category of dual. English can and
does have the means to refer to pairs of items - for example the
lexical items both, neither -but the distinction is not ‘gram-
maticalized’, realized as a formal system in the grammar, but
is rather ‘lexicalized’.
We return to the question of number in English. It can be
242 Part two: Morphology
argued that, prior to any singular:plural distinction, a massrcount
distinction is needed, the number distinction of singular:plural
only applying to count nouns. Consider the following. Count
nouns in general are used to refer to discrete objects that
can be counted - thus two chairs, twenty cigarettes, a hundred
pipers, and so on. Mass nouns, on the other hand, are character¬
istically used to refer to items perceived not in terms of individual
members, but in some mass terms: items like water, sand, butter,
ink, etc. The count:mass distinction is covert, in that nouns in
English are not marked in any way to show their membership of
either sub-class. The distinction can, however, be established on
the basis of the co-occurrence relationships of the two classes of
noun (see discussion in pages 90-1). Mass nouns do not co¬
occur with numerals {*one sand, *two waters, etc.) nor in the
plural (*butters, *inks etc.); they do not co-occur with the indefi¬
nite article (*an ink, *a water), but do occur with the determiner
some (pronounced /sm/) {some butter, some ink). Count nouns
are the opposite: they occur with numerals {one boy); in the
plural form {boys); with the indefinite article {a boy); but not, in
the singular, with /sm/ {*sm boy). We also observe that, for the
purposes of concord, both subject-verb concord and concord
within the NP, mass nouns operate like singular count nouns
{This butter is . . .; * These butter are . . .). There are, of course,
idiosyncracies: a few mass nouns are plural in form and take
plural concord {clothes, guts, suds, etc.); a few are not overtly
plural but take plural concord {gentry, cattle, clergy, etc.); and so
on. These idiosyncracies apart, they offer no particular problems.
The count:mass distinction may be held to be a reflex of a
notional distinction which we call a distinction between ‘individu¬
ated’ and ‘non-individuated’ referents.
The descriptive problem begins to arise when we find nouns
that we might wish to regard as ‘basically’ mass nouns occurring
with the syntax of count nouns:

1 I always have a coffee at this time of the morning

2 Two milks, please
3 I don’t like either of these two wines

The converse also occurs:

4 The scrum was not producing enough ball for the backs
5 He is collecting worm for his fishing
6 This room smells of cat
 Form classes and grammatical categories 243

As with our discussion of poor (pages 236-7), we can adopt

one of a number of solutions, each seeming attractive for some
items and less attractive for others. We can suppose that two
items are involved - coffee(I), the mass noun, and
COFFEE(2), the count noun, and so on. Or we can suppose that
COFFEE is basically a mass noun, and has been recategorized in
1. Alternatively, we can suppose that some relevant quantifying
expression has been elided — I always have [a cup of] coffee . . . .
The first solution seems attractive when the phenomenon is
well established in usage, and this fact is often reflected in
dictionaries. So for example lamb is given two senses in the
dictionary, LAMB(l) the ‘animal’ and LAMB(2) the ‘meat of the
animal’, reflected in sentences like:

7 I would like a lamb

8 I would like lamb

This solution is reinforced by the fact that in some cases the

animal and its meat receive two different lexical realizations
PIG : PORK; COW : beef, etc. A similar solution is open to
items like BEAVER(l), the ‘animal’ and BEAVEr(2) the ‘skin of
the animal’; OAK(l) the ‘tree’ and OAK(2) the ‘wood of the
tree’. Such usages seem well established. There are, however, two
points to be made. Firstly, the type of understood categorization
depends to a large extent on context - in a draper’s shop 8 would
typically refer to the ‘skin’ rather than the ‘meat’ of the animal
(and the converse is a restaurant). Secondly, the extent to which
the process is established in usage (if dictionaries are any guide)
is variable. Thus we might not expect to find POTATO(l) the
‘vegetable’ (Give me a potato) and POTATO(2) (Would you like
some (=sm) potato?), and similarly APPLE, BANANA, etc.
The second solution seems attractive for examples like 4-6.
Usages like this do not seem to be established in the language -
though 4 is becoming established in television commentary on
rugby football, in Britain at least - and are clearly heavily
context-dependent for their interpretation.
The third solution seems attractive in cases like 1-3, though it
is not without problems. In a case like two milks we need to
account for the resiting of the plural marker (compare two pints
of milk), but we can doubtless get round this. More difficult is
that the precise quantifying expression relies once again on
context: 2 relates to two [pints of] milk in a note to the milkman,
but to two [cups/glasses etc. of] milk in a cafe.
244 Part two: Morphology

The count:mass distinction appears to rest on a perceptual

distinction, at least to some degree. Some things are
characteristically perceived as non-individuated (like ‘sand’ and
‘water’), but they may be individuated in quanta {a bucket of
water, a heap of sand) and then expressed like count nouns (two
waters). Conversely, some things are typically perceived as
individuated (like ‘ball’ and ‘cat’), but may, or some of their
attributes may, be perceived as non-individuated, and then be
expressed as mass nouns (cf. 4-6).
A similar problem arises with the class of ‘collective’ nouns
(COMMITTEE, GOVERNMENT, HERD, etc.). These occur with
either singular or plural concord:

9 The committee has agreed to the appointment

10 The committee have agreed to take no further action

The usual account of such sentences is that the number choice

depends on whether the collective is thought of as acting as a
single body (the committee has . . .) or as a collection of
individuals (the committee have . . .). With such nouns, even
though they take plural concord on the verb, concord within the
NP remains in the singular (this committee have . . .)! As in the
count:mass distinction, the assignment of number appears to
rest on a perceived or imputed semantic perception.
To account for this, we might suppose that in some community
some ‘typical’ perception such as ‘individuated:non-individuated’
and ‘oneness:more-than-oneness’ is ‘grammaticized’ into a formal
grammatical category - count:mass, singular:plural. The
grammatical category is reflected in a range of syntactic
behaviour; once established as a grammatical category, it is open
to exploitation for semantic effect. (Not all communities
grammaticize the same ‘typical’ percepts.)
An interesting way of looking at this problem is proposed by
Jespersen (1929, 46). He suggests that we can regard language
as involving three inter-related levels of description: a level of
forms (the actually occurring morphological markings); a level of
grammar (with formally established categories like count:mass,
singulanplural, etc.); and a level of meaning (with notional
categories like individuated :non-individuated; oneness:more-
than-oneness etc. The grammatical level faces both ways:
towards the level of form and towards the level of meaning as in
Figure 94.
 Form classes and grammatical categories 245

forms grammar meanings

stem form singular non-individuated

oneness
affix

vowel change
-5

plural — more-than-
> individuated

(man :men) oneness

Figure 94

A perceptual characterization is reflected in a particular

grammatical choice, which is realized as a particular form. This
approach allows us both to maintain the principle of a grammar
established on formal principles (the formal opposition of
number, etc.) and to understand how this formal system is related
to, and can on occasion be manipulated by, the semantics.
We can incorporate it into our previous description in the
following manner. Recall that we suggested (pages 86-9)
that nouns should be characterized in the lexicon in terms of a
set of syntactic features, [±count], etc. We also saw that
a characterization, say [+count], had consequences for
co-occurrence with articles, etc. Suppose now we say that
whereas a particular lexical item may be characterized as
‘typically’ [Fcount] or [—count], this characterization may be
altered by a semantic consideration. However, once they have
been altered, the grammatical consequences are predicted by the
grammar. We can also accommodate within such an account those
irregularities we have noted during the discussion - SUDS for
example can be characterized as [-count], [+pl].
A further advantage is that this way allows us to accommodate
both the situation described above for [±count] nouns in English,
where a semantic characterization can influence a grammatical
choice, and the sort of situation found with French gender, where
the grammatical characterization is relatively uninfluenced by
semantic considerations. In different languages different
categories are influenced to different degrees by semantic
considerations. In French, and other European languages, nouns
cannot be switched from gender class to gender class for com¬
municative effect (one might think, for instance, that French would
allow une homme as a characterization of an effeminate man -
246 Part two: Morphology
but this sort of thing is not permissible). Many languages do
permit it: so in Lugisu, a Bantu language from Uganda, we find
mu-twe ‘head’ ‘recategorized’ as a diminuitive ka-twe ‘pin head’, or
as an augmentative gu-twe ‘big head’ by changing gender class -
the effect is derogatory.
Let us now turn to another grammatical category in English,
that of tense. How should our description account for the
relationship between notions of time and the formal opposition
between ‘past’ and ‘non-past’ established and discussed on
page; 109? The category is an overt two term category, called
tense, because it often corresponds to notions of time, but the
correspondence is far from direct. There is a frequent corres¬
pondence between past time and past tense (7 was in London last
week, We went to America last summer, etc.) but there are
circumstances when a past tense form is used, with no implica¬
tion of past time. If the main verb in reported speech is in the
past tense, the verb in the reported speech may also be in the past
tense, irrespective of the time reference. If my daughter says
to me at breakfast one morning ‘7 am going to the pictures this
evening’ (using a non-past tense verb form), I can report this to my
wife at lunch time with the words Lane said she was going to the
pictures this evening'1 (using a past tense verb form). Past tense is
also often used for ‘unreal conditionals! - as in If my daughter
were to go the pictures . . ., If only I understood what it was all
about . . ., 7 wish I knew . . ., and so on. In yet other circumstances
the distinction between the past and non-past tense does not reflect
any time distinction at all, as in: 7 wonder/wondered if you would
like a cup of coffee; I think /thought you would be interested to
know that . . .; etc. The most striking cases of non-correspondence
occur with the modal auxiliary verbs, where very frequently the use
of a morphological past tense form bears no relation at all to past
time: 7 might go if you asked me; I could stop smoking if I tried; I
wouldn’t have the slightest idea; etc.6

6 There are some cases where past tense and past time do correspond: I can
speak French quite well now: I could speak German when I was younger, but this
is not the regular pattern for modal verbs. There is an argument that some
so-called past tense forms among modal verbs are better thought of as simply
separate items. An example is should when used in the sense of ‘duty’, e.g. You
should always give up your seat on the bus to an old lady. Replacing should by
shall produces a sentence of doubtful acceptability! The interested reader is
referred for further discussion to Leech (1971) and Palmer (1971).
 Form classes and grammatical categories 247
Next consider sentences with a non-past verb form. They can
sometimes be found with a ‘present’ time reference: Dalglish
shoots and it’s a goal, but, as we have already observed, this is
not the usual way to refer to actions taking place at the time of
utterance. More frequently the simple non-past verb form is used
in a ‘timeless’ sense - Honesty is the best policy, War solves no
problems', to refer to ‘habits’ - I cycle to work every morning, I
always get drunk at Hogmanay; to refer to ‘states’ - I belong to
Glasgow, I speak Chinese', and so on. Non-past verb forms may
also be used to refer to future events - I leave for Paris in the
morning; and indeed to past events (the so-called ‘historic present’)
- Earlier this morning I come into the kitchen and what do I find?;
and past states of affairs - Shakespeare draws his characters from
real life. The correspondence between ‘present time’ and the
non-past form of the verb is even less straightforward than the
correspondence between ‘past time’ and the past form of the verb.
What, then, do we make of characterizations like ‘timeless’,
‘habit’, ‘state’ and so on, characterizations frequently met with in
grammars. There are, to be sure, occasions when the
correspondence between some such category and a term in the
category of tense is reasonably direct: thus ‘timeless’ is typically
realized by a non-past verb form (God was good; Christmas was
on 25 December; Cocks crowed at dawn seem anomalous if they
are to be understood as statements of ‘general truth’ which hold
regardless of time!) Similarly ‘past instantaneous actions’ are
usually realized by a simple past verb form (/ bought a coat this
afternoon; I gave my wife a diamond ring for Christmas). But
usually the relation is less direct, ‘present habits’ usually correlate
with a non-past verb form, but past habits frequently occur, not
with a simple past form, but with a form of USED TO (/ cycle to
work every morning; I used to cycle to work every morning).
Similarly, as we have observed before, ‘present instantaneous’
actions are often realized by the progressive rather than the
simple non-past verb form (I am typing this sentence now). The
relationship between notions like ‘habit, etc. and some formal
overt grammatical category is often indirect, but such notions can
play an important part in controlling other features of the
sentence. Thus, for example, ‘timeless’ statemeftts usually require
a noun of general reference as subject, and are restricted as to
the type of adverbial modifiers they permit: Cocks crow at dawn
can be held to be a timeless statement, but My favourite red
248 Part two: Morphology
rooster crows at dawn most mornings cannot. Similarly ‘past
habits’ impose restrictions on adverbial expressions - one can
hardly say */ used to cycle to work this morning\
Notions like ‘timeless’, ‘habit’, etc. seem to have relevance for
more than the single category of tense. Grammatical features that
can be correlated with such notions extend throughout the sen¬
tence and may be relevant for grammatical selection at various
places. A valuable discussion of some of these notions is Crystal
(1966) to which the reader is referred (see also Leech (1971)).
The difficulty may be summarized thus: categories like
‘timeless’ ‘habit’ ‘unreal condition’, etc. have syntactic
implications in relation to such matters as co-occurrence
restrictions relating to adverbs etc.; on the other hand there is no
single overt category to which they can be tied in any
straightforward way. There is a further problem with respect to
the notional categories - it is not clear how many such categories
we need to postulate, or what the distinctions are between them.
For example, we have suggested a category of ‘state’ - but this is
clearly not a unitary category at all, there are all sorts of states.
We might recognize: inherent states (I am a man); resultant
states (My coffee is cold (now, but it was hot ten minutes ago));
transitory states (It is fine now (but will be raining again in ten
minutes); and so on. It is far from clear that reliable distinctions
can be drawn clearly between them.
This is at least true for English where such categories are not
relatable to a formal distinction in the grammar. In terms of our
former discussion, they are not ‘grammaticalized’. The situation is
different in some other languages. In Finnish the distinction
between ‘transitory’ and ‘inherent’ states is reflected in a
difference in the case form adopted by the noun. In Akan the
distinction between ‘states’ and ‘actions’ is reflected in the
existence of a special ‘stative’ verb form, open only to verbs that
describe states; the distinction between ‘inherent’ and ‘resultant’
states is typically marked by different aspect markers on the verb.
As far as English is concerned, it does not seem profitable to
set up categories like ‘timeless’ as grammatical categories, on a
par with the grammatical category of tense - though attempts
have been made. It would seem more appropriate to account for
such distinctions in terms of a model of description like that
hinted at on page 245, but we shall not pursue this further here.
 Form classes and grammatical categories 249
Technical terms
closed category lexical category
collective noun lexical word
count noun mass noun
covert category open category
form word overt category
full word part of speech
grammatical category recategorization
grammatical word

Exercises
In the grammatical analysis of languages words are assigned to word
classes on the formal basis of syntactic behaviour, supplemented and
reinforced by differences of morphological paradigms, so that every
word in a language is a member of a word class. Word class analysis has
long been familiar in Europe under the title "parts of speech’, and since
medieval times grammarians have operated with nine word classes or
parts of speech: noun, verb, pronoun, adjective, adverb, preposition,
conjunction, article and interjection. (R. H. Robins, General Linguistics,
Longmans, 1964)

1
Can you think of some of the principal syntactic environments in
English in which you would find members of the classes usually
named noun, verb and adjective?

2
Can you describe some of the principal morphological charac¬
teristics of these word classes?

To what word classes (in the list given by Robins above) would
you assign the words in the following and why?

’Twas brillig, and the slithy toves (1)

Did gyre and gimble in the wabe;
All mimsy were the borogroves,
And the mome raths outgrabe.
250 Part two: Morphology

“Beware the Jabberwock, my son! (5)

The jaws that bite, the claws that catch!
Beware the Jubjub bird, and shun
The frumious Bandersnatch!”

He took his vorpal sword in hand:

Long time the manxome foe he sought - (10)
So rested he by the Tumtum tree,
And stood awhile in thought.

And as in uffish thought he stood,

The Jabberwock, with eyes of flame,
Came whiffling through the tulgey wood, (15)
And burbled as it came.

4
You will find that the membership of some classes is entirely
English, while that of other classes is partly English and partly
invented. Which classes are open to invented words and which
are not? Why do you think some classes have no invented words?
Part three

Functional relations
-

'
17 Syntagmatic and paradigmatic
relations in syntax

Part one looked at the constituent structure of simple sentences.

This part widens the discussion by looking at other aspects of the
structure of sentences that have been traditionally the concern of
grammarians. This chapter examines some of the different types
of relationships that constituents bear to each other, considers
how these relations are marked in the grammars of different
languages, and whether all such relationships can be satisfactorily
described in the type of constituent structure grammars we have
been looking at. Chapter 18 examines notions like ‘agent’ and
‘patient’ as they might be applied to a sentence like The
policeman (‘agent’) arrested the burglar (‘patient’), and sees how
the relevant notions might be captured in a grammar. Chapter 19
examines the terms ‘subject’ and ‘object’. Chapter 20 is
concerned with some questions of word order as they apply to
connected text.
This chapter is concerned with those grammatical relations that
have been discussed in the linguistic literature under the general
title of ‘syntagmatic’ and ‘paradigmatic’ relationships.1
In straightforward terms, paradigmatic relationships are those
contracted between items that are mutually substitutable in some
context. Form classes are paradigmatic classes. So, in the English
NP, we have a paradigmatic class of determiners including a, the,
some, etc., and a paradigmatic class of nouns, including man, boy,
house, etc.
Syntagmatic relations are those contracted between forms or
form classes within some structure. These may include relations

1 Other terms that relate to the same area are structure and system (cf. Aber¬
crombie 1967), chain and choice (cf. Halliday 1963). These relationships are very
general, and can be applied at all levels of linguistic description: they apply
equally to phonological and semantic descriptions as they do to syntactic ones. We
shall, however, only concern ourselves with such relationships in the grammar.
254 Part three: Functional relations

of order (thus in the English NP the determiner must precede the

noun: the man not *man the) or relations of dependency (thus the
English NP contains a ‘head’ usually a noun, and may optionally
contain an adjective ‘modifier’: we find big men and men as NPs,
but not simply big: the adjective is dependent on the noun within
the NP).
These relationships are interdependent. The most general
statement of an environment (a syntagmatic description)
characterizes some form class (a paradigmatic description) in the
widest terms. A closer specification of the environment specifies
a sub-class of the major class and so on. In the most general
terms the distribution of the classes determiner and noun is
defined by a rule NP —» Det + N. But the sub-class of mass
nouns only co-occurs with a sub-class of determiners - so we have
the butter but not *a butter. Our principal concern is with various
kinds of syntagmatic relationship.

Syntagmatic relations of dependency and exclusion

We will examine four types of relation - mutual or bi-lateral

dependency; unilateral dependency; co-ordinate dependency and
mutual exclusion - but will not discuss them in detail since they
are extensively covered in introductory textbooks. There are
two fundamental things to recognize about such relationships:
first, that they are abstract relationships between sentence
constituents, and second, that they are functional relationships
indicating the syntactic and semantic function of constituents in
their relationships with other constituents. Obviously a language
needs to find ways of marking a particular relationship, and
different languages select different ways of marking: the
relationships themselves, however, are in principle independent
of any particular aspect of word or sentence form. Two examples
illustrate the point. The unilaterally dependent relation
modifier: head can be exemplified by constructions involving
adjectives and nouns. The relationship is typically realized in
English in the order Adj 4- N (the red book), but in French in the
reverse order, N + Adj (le livre rouge), without affecting the
dependency. The unilaterally dependent relation possessor:
possessed (which is itself a type of modifier: head construction
with the possessed as the head) can be realized in English by two
constructions — the son of the king and the king’s son. The
 Syntagmatic and paradigmatic relations in syntax 255

dependency relation is the same in both constructions, but the

word order is different and each construction requires different
‘markers’ (of and -5) to show the relationship.
The fact that these relationships are functional means that we
need to find names to label the type of relationship involved -
as modifier: head, possessor: possessed, etc. Sometimes these
relationships are definable in terms of constituent structure
configurations, but this is not always the case, as we shall see.
This is not surprising if we consider that constituent structure
grammars operate on categories like NP, V, etc, which are
distributionally and not functionally defined. For example, in
English, both subject and object (functional descriptions), are
realized by NPs (a categorial class). In English we can use
constituent structure configurations up to a point to define the
functional relation of the relevant N Ps - as we did in footnote 2
on page 52, where we defined ‘subject’ as the NP dominated by
S, and ‘object’ as the NP dominated by VP. The discussion on
pages 261-84 will use languages where this definition is less easily
workable, but where the functional relationships of dependency
hold as in English. In our initial discussion of the relationships
mentioned at the beginning of the chapter we will, however, use
examples where the relations can be related to distributional
characteristics.
Constructions involving bi-lateral or mutual dependency are
called exocentric constructions. The definition of an exocentric
construction is usually given in distributional terms: the
distribution of either constituent in such a construction is
different from the distribution of the construction as a whole.
Constituent structure rules introducing mutually dependent
constituents are usually of the form X—>Y + Z, e.g.
PP—» Prep + NP. In a prepositional phrase the distribution of a
PP as a whole is different from the distribution of either a
preposition or of an NP: the two constituents of a PP. Many of
the major syntactic relationships in all languages are mutually
dependent. If we make the fundamental syntactic construction of
English sentences NP + VP, we assert that all full sentences in
English are of this construction type. In English most VP
constructions are also bi-laterally dependent: these include the
transitive verb and its object (V + NP); the copula and its
complement (V + Pred); locative verbs and their locative
complements (V + PP); and so on. Hockett (1958) contains a
256 Part three: Functional relations
taxonomy of many different kinds of such construction. All the
constructions noted are bi-laterally dependent, but it is clear that
they are not all of the same kind. This is reflected in the different
functional labels given above; part of the reason why a set of
functional labels like this were developed is precisely to
differentiate between different types of dependency. The
importance of this can be appreciated by noting that the three
sentences John beat the dog, Fido is a dog and John went home
can at some level of structure all be described as strings of the
form NP + V + NP: but the functional relationship between the
second NP and the verb is different in each case - they are
respectively transitive verb + object; copula + complement; and
locative verb + locative complement. It will be recalled from
pages 50-1 that we introduced the category Pred precisely in
order to be able to distinguish between NPs introduced as objects
of transitive verbs and NPs introduced as complements.
In many bilaterally dependent constructions, one constituent
requires the other constituent to assume a particular grammatical
form. In English, in prepositional phrases where the NP is
realized as a pronoun, the pronoun must be in the ‘oblique form’
- we find to him; to her, etc., but not *to he\ *to she, etc. This
relationship is known as government; in this case, we say that the
preposition is the governor, and governs the NP with which it is
in construction. In many languages with morphological case
systems, like Latin, German or Old English, the relationship
between a verb and its object(s), or between a preposition and
the NP with which it is in construction is marked by the fact that
the NP in question must occur in a particular case; in these
instances the verb, or preposition, is said to govern the NP in
some particular case. Thus, in Latin, prepositions like ad ‘to,
towards’, intra ‘within’, etc. governs the N Ps with which they are in
construction in the accusative, while prepositions like ab ‘from’,
sine ‘without’ govern an NP in the ablative. Some preposi¬
tions may govern an NP in either the accusative or ablative;
this typically correlates with a systematic difference in
meaning: in + Acc. ‘into, towards’; in + Ablative, ‘in. among’;
sub + Acc, ‘up to’, sub + Ablative, ‘under’. When such
prepositions govern the accusative there is a general sense of
‘directional movement’; when they govern the ablative there
is a general sense of ‘position’. Further examples can be
found in any Latin grammar. We have noted that not all
 Syntagmatic and paradigmatic relations in syntax 257
bi-laterally dependent constructions in English are the same: for
the examples we discussed we might say that transitive verbs
govern a direct object, and that copula verbs govern a
complement, adjectival or nominal. In English there are few
markers of this relationship, but in a language like Latin the
object of the transitive verb is usually governed in the accusative
(i.e. objects are marked by the accusative case), and
complements are in the nominative; they agree with the subject
in case and number, and in gender too if the complement is an
adjective.
In unilaterally dependent constructions, one constituent is
typically obligatory and the other, the dependent constituent, is
typically optional. These are called endocentric constructions;
and a distributional definition states that the distribution of the
construction as a whole is parallel to that of the obligatory
constituent. The obligatory constituent is called the head of the
construction and the optional constituent the modifier of the
head. Constituent structure rules introducing unilaterally
dependent constructions typically take the form X—»(Y+) Z,
e.g. NP —> (Det+) N. In English NPs the distribution of
unmodified nouns is parallel to that of nouns modified by
determiners: the determiner is thus the modifier and the noun the
head of the construction. In English, unilaterally dependent
constructions occur in a variety of different orders; the modifier
may precede, follow or be discontinuous about the head (in small
capitals in the following examples): a red CAR. PEOPLE that live
in glass houses, the earliest TIME possible. Such constructions in
English include: Adj + N ((big)man); Intensifier + Adj {{very)
important)-, subordinate clauses of various kinds (/ feel tired
{when I get up)), {He closed the window {in order to keep out the
rain))-, and constructions involving sentence adverbs (see pages
261-4).
It is frequently the case in unilaterally dependent constructions
that modifiers must agree, or concord, with the head with respect
to some particular grammatical category (concord is discussed in
more detail on pages 278-84). Thus in English NPs demon¬
stratives {this-.these-, that-.those) concord in number with the
head noun: this man, these men, and not * these man, *this men.
In many European languages articles and adjectives must
concord in number and gender with the head noun: so in
German we find ein junger Mann ‘a young man’; eine junge
258 Part three: Functional relations

Frau ‘a young woman’; ein junges Madchen ‘a young girl’. In

such cases the head noun controls the form of the modifiers -
and we say that the head is the controller of the concord.
In constructions of co-ordinate dependency neither constituent
depends syntactically on the other: in distributional terms each
constituent has the same distribution as the construction as a
whole. A constituent structure rule we used to indicate
co-ordinate dependency is Adj—»Adj* (see page 79). This
introduces strings of adjectives like a little old lady with the
interpretation ‘a lady who is both little and old’. Many
co-ordinate constructions include a marker of co-ordination like
and or or (My husband AND I). (you OR your wife). Not all
co-ordinate constructions show such a marker, as we have noted
with co-ordinate modifying adjectives above. Another case of
co-ordinate dependency with no marker is appositive structures:
President Kennedy; Professor Miller; Mr Bun, the baker; Our
father, which art in Heaven. These examples show that the
distributional definition for co-ordinate constructions is less
convincing than with the other types of construction we have
mentioned.
Relations of exclusion are important insofar as they are useful
in defining particular grammatical categories. Thus, for example,
verbs of state in English do not in general occur with progressive
auxiliary verbs, and this is one of their defining characteristics.
We find no sentences like */ am knowing Chinese-, *Bill is
seeming ill. Proper nouns in English do not typically co-occur
with the definite article (*the Susan), except when they co-occur
with a relative clause (the Susan I used to go out with twenty years
ago). Mass nouns do not co-occur with plural expressions (*these
butter, *Rice are grown in India).
The discussion has made clear the importance of dependency
relations. Such relations lay behind the classification of verbs on
pages 50-60: the strict subcategorization frames we established
were statements of different types of bi-lateral dependency.
Similarly, unilateral dependency relations lay behind our
discussion of modifier head relations among NP constituents
(pages 75-9) and in the discussion of adverbs (pages 69-75).
These relations can be captured up to a point in constituent
structure grammars, as our discussion of different rule types for
different dependency relations has indicated in the foregoing
discussion, and as our discussion of the ‘scope’ of modification on
 Syntagmatic and paradigmatic relations in syntax 259
pages 69—85 showed. But this is not always entirely successful.
We give three examples. First: our discussion of STAND on pages
53-4 showed that such verbs can occur either with or without an
object NP (The lamp stood on the table-, He stood the lamp on the
table). We captured this by expanding the VP by a rule VP-* V
(NP) PP. But note that the ‘optional’ object NP stands in a
relation of bi-lateral dependency to the verb, and is not in any
sense a ‘modifier’. Second: as discussed above, we introduced the
node Pred on page 51 precisely in order to distinguish between
NPs which are complements of copula verbs- and NPs which are
the direct objects of transitive verbs. Here additional structure
was necessary to capture a dependency relationship. Even
supposing that we were able to capture all the relevant
dependency relations by postulating additional structure, or other
additional machinery (perhaps transformations), we might ask
whether such a complication of the grammar is the best way to
capture these relations. Third: in our discussion of different
adverb types (pages 69-75) we in fact smuggled in functional
labels by differentiating between adverbs of time, adverbs of
place, etc. Such labels are not, strictly speaking, categorial labels.
The same applies to some extent to the category Pred noted
above.
We need not be surprised that dependency relations cannot
always be captured in a straightforward fashion in constituent
structure grammars: constituent structure grammars are
constructed in terms of formally established categories and the
constructions they form, whereas dependency relations are, as the
name suggests, relations. Furthermore, even if we can adequately
represent these relations in constituent structure terms, we need
additional interpretative machinery to interpret the relations
from constituent structure trees, or constituent structure rules or
both.
In early constituent structure grammars (Bloomfield 1933,
Nida 1960) information about dependency was shown in one of
two ways. Sometimes trees were marked to indicate the type of
dependency involved, illustrated in Figure 95.2 Alternatively, a
statement of the types of relationship involved was appended to a

2 It was fashionable at the time for trees to be rooted in the ground rather than
for them to depend from an S node, as is more fashionable today: either way the
information is the same.
260 Part three: Functional relations
particular tree. Such statements of dependency relations are lost
in some more recent descriptive approaches, as they are to a
large extent in earlier transformational grammars.

the little girl cried marks an ‘exocentric’

(bi-laterally dependent)
construction

marks an ‘endocentric’
(unilaterally dependent)
construction
Figure 95

There are approaches to description in which dependency

relations are more central, but to discuss them here would take us
too far out of our way. The interested reader is referred to the
brief discussion of ‘Categorial Grammar’ in Lyons (1968,
227-31).
At the beginning of this section we noted that different
languages employ different formal means of marking dependency
relations - word order, the use of ‘markers’, case inflections, etc.
In the sections that follow we discuss some of these.

Word order and grammatical markers

All languages use word order to a greater or lesser extent as a

marker of various functional relationships, but different
languages impose different ordering restrictions and within any
one language some ordering restrictions are strict and others
admit of a greater or lesser degree of latitude. It is usually
possible to determine, for some particular area of some language
that a particular order is the ‘normal’ or ‘unmarked’ order:
deviations from this, where they are permitted, usually have some
particular stylistic or communicative effect.
In addition, all languages use a number of forms, free or
bound, as markers of particular relationships: these markers
often have little or no lexical meaning - their function being
entirely to mark relationships. Thus, in English, we find the -5
used to mark the genitive relation (John’s book; the old man’s
 Syntagmatic and paradigmatic relations in syntax 261

dog, etc.) and the by found in passive sentences to mark the

agent (The village was destroyed by an earthquake).
We will consider word order and the use of markers in four
constructions in English, and then look briefly at comparable
constructions in three other languages: Akan, Scots Gaelic and
Turkish.

Noun phrases in English

The unmarked order for NP constituents in English is:

Determiner, Numeral, Adjective, Noun. The noun is the head of
the construction and the other constituents are unilaterally
dependent on it. As far as determiners are concerned the
ordering is strict: they can only occur initially in the NP, and no
reordering is permitted at all. Numerals almost invariably
precede adjectives and nouns. Old King Cole, it may be
remembered, called for his fiddlers three, but this particular
ordering seems dictated by the rhyme scheme and is not an
option usually available in contemporary spoken English: I can
hardly feed my cats two\ Occasionally we find NPs like the big
four, the silent three of St Botolph’s, but the numeral here is
perhaps best treated as a nominalization. the big four derives
from the four big powers, where powers is elided and may be
contextually recovered, just as in your two will be quite grown up
by now, / expect the word two is held to refer to ‘children’, ‘dogs’,
etc. as the context dictates (see discussion on pages 236-7).
In English the unmarked position for adjectives is preceding
the noun. There are a few exceptions. When the head noun is an
indefinite pronoun, ending in -body, -one, -thing, etc., then
adjectives follow the noun: somebody clever, rather than * clever
somebody (a typical ‘goon show’ usage!). A few adjectives
characteristically follow rather than precede the head noun, or
regularly have a slightly different meaning from ‘the same’
adjective when it precedes the head noun:

1 Any proper university these days will have a flourishing

department of linguistics
2 The school of shorthand is not part of the university
proper, but is closely associated with it

More usually, re-ordering involves a particular stylistic effect:

262 Part three: Functional relations
Some postposed adjectives, especially those ending in -able or -ible
retain the basic meaning they have (when they precede the noun) . . . but
convey the implication that what they are denoting has only a temporary
application. Thus, the stars visible refers to stars that are visible at a time
specified or implied while the visible stars refers to a category of stars
that can (at appropriate times) be seen [Quirk et al. 1972. 249].

In sentences like

3 He prefers his mistresses unmarried

his mistresses unmarried is not an NP containing a postposed

adjective. Unmarried is, rather, a different constituent from the
NP his mistresses: it may be called a complement. Note the dif¬
ferent paraphrase relations between 3 and

4 He prefers his unmarried mistresses

Paraphrases like:

5a He prefers his mistresses to be unmarried

5b He prefers that his mistresses should be unmarried

relate to 3, and a paraphrase like

6 He prefers those of his mistresses who are unmarried

relates to 4.
In English, while adjectives precede the noun, there are also
ordering restrictions on adjectives themselves. Thus, for example,
we will find a nippy little red sports car, but hardly *a red little
nippy sports car. The restrictions in this case relate to sub-classes
of adjectives which can to some extent be defined semantically -
see the exercise on page 48. Sub-classes defined in terms of order
are known as order classes. We noted (pages 73-5) that adverbs
also fall into order classes, though restrictions there are less
rigorous.

Prepositional phrases in English

English prepositional phrases are bi-laterally dependent construc¬

tions. The preposition must always precede the NP, sometimes
referred to as the ‘object’ of the preposition. A preposition can
sometimes be separated from its object, as in relative clauses like

7 The knife which he cut the salami with

 Syntagmatic and paradigmatic relations in syntax 263
which we examined on pages 139-41), but the general pattern is
for them to be closely connected and strictly ordered.3

Genitive constructions in English

English has two genitive constructions, both unilaterally depen¬

dent. Genitive constructions can be used for a variety of func¬
tional relationships in English and in other languages: consider
the variety of meanings that can be attached to an expression like
Myron’s statue, which may be: ‘possessive’ (the statue which
Myron owns); ‘subjective’ (‘the statue made by Myron’); objec¬
tive’ (‘the statue of Myron’); and so on. We shall consider only
the ‘possessive’ sense, both for English and the other languages
we look at. One genitive construction uses the marker -s, as in:

8 Charlie’s book; Pharaoh’s daughter

Here the NP marked by -s is distributionally equivalent to a pos¬

sessive pronoun (his, her, my etc.). This NP, which we call the
‘possessor’ is the modifier of the ‘possessed’ NP, which is, dis¬
tributionally, the head of the construction, and must always fol¬
low the possessor. The other structure involves a prepositional
phrase with of:

9 The daughter of Pharaoh; the roof of the car

This time the NP in the genitive construction is distributionally

equivalent to a pronoun like mine, hers, etc., and the possessed
precedes the possessor. However, it is still the possessor that is
marked, here by the use of of. In both cases the head of the
construction is the possessed noun - the obligatory noun and the
one which controls number concord when such phrases are used as
subjects in a sentence. To a certain extent these structures are
interchangeable, as an old riddle shows:

3 We observed on page 140 that in this respect prepositions must be distin¬

guished from homophonous items that are verbal particles:

He added up the bill but He ran up the hill

He added the bill up *He ran the hill up
He added it up *He ran it up
*He added up it He ran up it

Verb + particle constructions are bi-laterally dependent.

264 Part three: Functional relations

10 Is the daughter of Pharaoh’s son the son of Pharaoh’s

daughter?

but this is not always the case. Whichever construction is chosen,

there is a strict ordering restriction, though the direction of the
ordering is different in each case.

Sentence constituents in English

In English the unmarked word order of sentence constituents is

Subject-Verb-Object, which we abbreviate as SVO. Circum¬
stances when this particular order can be changed are examined
on pages 363-73.

The ordering relations found in English for these four construc¬

tions do not apply universally. We will examine such structures
briefly in three other languages: Akan, Gaelic and Turkish.

Noun phrases in Akan

In Akan, as in English, NP constructions are unilaterally

dependent with the noun as the head of the construction. The
normal order of constituents is Noun, Adjective, Determiner (the
reverse of the English construction).

11 Gyinamoa ketewa bi
cat small a a small cat
Onipa kesee no
man big the the big man

Even in NPs which contain a relative clause the Determiner is

frequently the last NP constituent:

12 Onipa aa >te ha no (ye me nua no)

man relative he-lives here the (is my brother)
marker
Np(N s(relative clause )§ Det)Np

the man who lives here (is my brother)

Prepositional phrases in Akan

Prepositional phrases are bi-laterally dependent constructions.

Akan has a distributional class of items similar in sense and
 Syntagmatic and paradigmatic relations in syntax 265

function to English prepositions, but this class always follows

rather than precedes the NP with which it is in construction: they
are often called ‘postpositions’:

13 Ofie no ho
house the exterior outside the house

PP^NP^ ^ ^et)^pPreP)pp

Ahina no mu
pot the inside in the pot

opono no so
table the top on the table

Genitive constructions in Akan

In genitive constructions, the possessor precedes the possessed.

No separate morph shows this relationship, but, in most genitive
constructions, the possessed undergoes a tonal modification

14 Kofi Kofi (personal name) [_-] me my [_]

ponkb horse [_~] nantwie cow [_]

Kofi ponko Kofi’s horse [_

Kofi nantwie Kofi’s cow [_ _ ]

me ponkb my horse [_ ~ _ ~]

As will be seen, the possessed noun acquires a high tone on its

first tone-bearing unit.4 The tonal modification is a necessary
marker of the genitive construction since mere order alone does
not suffice: this can be seen by considering the following
examples, which contrast a genitive construction and an
adnominal construction.

4 High tones are marked with an acute accent; low tones are unmarked.
Tone-bearing units are; every consonant vowel sequence, and every consonant
and vowel left over (tw is. phonologically. a single consonant; compare sh in
English). Thus Kofi has two tone bearing units, and ponkS has three (po-n-kS).
etc. High tones are realized at a higher pitch than low tones but in a sequence
HLH the second high tone is usually at a slightly lower pitch than the first. The
pitch pattern for the examples is as shown after each example, except that this
‘lowering’ effect is not represented.
266 Part three: Functional relations

15 nsa beer [_"] tumpan bottle [____]

esiam flour [_] kotokuo bag [— ]

genitive
Nsa tumpan beer bottle [_ _ ]
esiam kotokuo flour bag [_ _ ]

adnominal
Nsa tumpan bottle of beer [_~_]
esiam kotokuo bag of flour [_ _]

Sentence constituents in Akan

The unmarked order for sentence constituents is, as in English,

SVO:

16 Kofi resua Twii

Kofi is-learning Twii Kofi is learning Twii

Kofi kumm gyata no

Kofi killed lion the Kofi killed the lion

Constituent order in Gaelic is different from both English or

Akan.

Noun phrases in Gaelic

The normal order for NP constituents is Determiner, Noun,

Adjective, as we noted in Part one (pages 62-6): the noun is the head
of the construction.

17 An iolair mhor
the eagle big the big eagle
An coineanach beag
the rabbit little the little rabbit

Am fear beag
the man little the small man

A’ chaileag mhor
the woman big the big woman

5 The variant forms of the article depend on the initial consonant or vowel of
the noun stem they precede.
 Syntagmatic and paradigmatic relations in syntax 267
Prepositional phrases in Gaelic

Within the prepositional phrase the preposition precedes its NP

object:

Anns an sporran
in the purse in the purse
(Prep
NP
(Det
NWW
Aig a’ chladach
at the shore at the shore

Air a’ chreig
on the rock on the rock

Genitive constructions in Gaelic

In genitive constructions, the possessed precedes the possessor.

The possessor is marked by appearing in the genitive case.
Rules for the formation of cases in Gaelic are extremely compli¬
cated (the reader is referred to a Gaelic grammar, e.g.
Mackinnon 1971) but the principle is clear from the examples;
note that in Gaelic the article, as well as, or instead of, the noun
may inflect for case:

19 an cat the cat (nominative, singular)

a’ chait of the cat (genitive, singular)
nan cat of the cats (genitive, plural)
an ceann the head (nominative singular)
na cinn the heads (nominative plural)
Ceann a’chait (head the cat) the cat’s head
Cinn nan cat (heads the cats) the cats’ heads

(The article is not used with the possessed noun.) Possession is

often expressed not with a construction of the sort illustrated
above, but in constructions of the following sort:

20 Bha peann aig Mairi

Was a pen at Mary Mary had a pen

Tha cu dubh aig Calum, ach tha cat ban aig Mairi
Is dog black at Calum but is cat white at Mary

Calum has a black dog, but Mary has a white cat

268 Part three: Functional relations
These may be analysed as V + NP + PP, where the PP is
expanded as Prep + NP. Structures of this sort may equally be
used to express location:

21 Tha Calum aig a’ chladach

Is Calum at the shore Calum is at the shore

(Many languages, including English, show relations between

‘locative’ and ‘possessive’ constructions, see discussion, pages
303-9.)

Sentence constituents in Gaelic

The unmarked word order for sentence constituents is VSO:

22 Reic Seumas an car

Sold James the car James sold the car

Ghlas Mairi an dorus

Locked Mary the door Mary locked the door

and we have already seen that in structures like 20 the subject

immediately follows the verb.

Typical word order patterns in Turkish differ from those we

have already examined.

Noun phrases in Turkish

The typical order of constituents in an NP in Turkish is

Adjective, Determiner, Noun:

Adj Art N
Buyiik bir ev
big a house a big house

Zeki bir kiz

intelligent a girl an intelligent girl

Prepositional phrases in Turkish

Turkish, like Akan, has postpositions, rather than prepositions -

these, obviously follow the NP:
 Syntagmatic and paradigmatic relations in syntax 269
24 Karakola yakin
police station near near the police station
pp(NP Prep)pp

Sizler igin
you (pi) for for you

Many expressions that in English translate into prepositional

phrases are represented in Turkish by case suffixes, as in Latin.
Prepositions in Turkish govern particular case forms: in the
examples yakin governs the dative, and igin the absolute, case.

Genitive constructions in Turkish

There are a variety of genitive constructions, depending on such

features as whether the possessor is definite or indefinite,
whether it is a ‘full NP’ or pronominal, etc. (The interested
reader is referred to a Turkish grammar, e.g. Lewis 1953.) In
all cases the possessor precedes the possessed. If the possessor is
definite, both nouns are marked for the relationship: the
possessor is in the genitive case and the possessed is marked
with a possessive suffix:

25 Miiduriin odasi
director his-room the director’s room
(genitive)

Kizlann odalan
girls their-rooms the girls’ rooms
{gen. plur.)

Sentence constituents in Turkish

Word order in a Turkish sentence is usually SOV. If the

subject is a personal pronoun it is usually omitted since the verb
is marked for person. Thus:

26 Ahmet otomobili aldi

Ahmet the car took Ahmet took the car

Mehmet biraz para istedi

Mehmet some money wanted
Mehmet wanted some money
270 Part three: Functional relations

The preceding discussion has exemplified four dependency

relations in four different languages. The relations remain the
same from language to language, but different languages mark
the dependencies in different ways. In all of them word order is
an important marker of the relation, but word order alone is not
always sufficient. In most cases a formal marker of some kind is
necessary in addition to word order: sometimes an additional
word (like of in the English genitive construction); sometimes a
morph (like -5 in the other English genitive construction); some¬
times case marking (as in Turkish and Gaelic genitive construc¬
tions); sometimes a tonal modification (as in the Akan genitive
construction). Sometimes the marker is the only feature differ¬
entiating between constructions - as with the genitive and
adnominal constructions in Akan.
In many of the constructions exemplified, the dependency
relations can be captured in constituent structure terms. For
example, prepositional phrases are formed by rules like:

27a PP—» Prep + NP (for English and Gaelic)

27b PP^ NP -I- Prep (for Akan and Turkish)

When the preposition governs the noun in a certain case, then

this too can be captured by an interpretation of these rules.
In other examples the dependency relations are not so easily
captured in constituent structure terms. We have already seen
that rules like:

28a S —> NP + VP
28b VP —» V + NP

can account for the constituent structure of English sentences,

and through such rules we can reasonably, though not entirely,
successfully capture the relevant dependency relations: subject
and predicate (28a); transitive verb and object (28b); etc. Indeed,
we carried out the strict subcategorization of verbs by appealing
to different predicate types - different expansions of VP. Such
rules seem reasonably adequate for a S VO language like English.
Similarly rules like:

29a S —» NP + VP
29b VP —> NP + V

seem satisfactory for SOV languages like Turkish. In these

terms, Gaelic is more difficult, as it has VSO word order: the
 Syntagmatic and paradigmatic relations in syntax 271
copula and its complement and the transitive verb and its object
are discontinous around the subject.
In Gaelic, just as in English, verbs can be classified in terms of
their predicate type. Thus we find copula + complement:
30 Bha - an cu - dubh
V NP Adj
(Was - the dog - black) The dog was black
(The sentences are analysed into their constituents) copula +
locative complement:
31 Bha - Tearlach - anns an Fhraing
V NP PP
(Was-Charlie-in the France) Charlie was in France
intransitive verb:
32 Bhasaich - Tearlach (Died-Charlie)
V NP Charlie died
transitive verb + object:
33 Ghlac - Calum - breac (Caught-Calum-a trout)
V NP NP Calum caught a trout

The above characterizations, like those established previously for

English, rely on ‘predicate’ types - reflected in co-occurrence
relations between a verb and some other constituent in the
sentence (an intransitive verb in Gaelic, as in English, occurs with
one NP, the subject; a transitive verb occurs with two NPs,
subject and object; etc.). However, do we want in Gaelic to
establish a constituent VP? If we do so, then we need a
reordering rule to take the verb and place it at the beginning
of the sentence in front of the subject NP; do we really want
such a drastic structure-changing rule to be involved in the
generation of simple sentences? It seems that in Gaelic
dependency relations are not as readily interpretable from
constituent structure configurations as they may be in English.
A similar problem arises in ‘free word-order’ languages. Latin
is often said to be such a language since the translation
equivalents of a sentence like ‘The boy loves the girl’ include:
34a Puer puellam amat (the boy-the girl-loves) SOV)
34b Puer amat puellam (SVO)
34c Amat puer puellam (VSO)
272 Part three: Functional relations
and any other permutation of the three constituents involved.
(Latin is not a language in which anything goes with respect to
word order - prepositions and their objects are strictly ordered,
for instance.) The grammatical relations of subject and object are
marked in Latin by case (the subject takes the nominative case
and the object, for most verbs, the accusative case - as in the
example) rather than order, but the relations remain as relevant
as they are in English! AMO is a transitive verb, and in a relation
of bi-lateral dependency with its object NP. Clearly we can
decide that one of these orders is basic, either a priori or on some
other grounds, frequency of occurrence for instance. So we might
decide that the order shown in 34a is ‘basic’ and postulate rules
like:

35 S —» NP + VP
VP NP + V

etc. This analysis enables us to identify in constituent structure

terms which NP is the subject and which the object and makes
the assignment of the appropriate case relatively straightforward,
but involves postulating a transformation ‘scrambling’ the
constituents into any of the permissible orders. But such a
description may only be a convenience, and one which derives
from an ‘English’ view of word order.6
The point at issue is this. Since grammatical relations of
dependency etc. are crucial in the description of a language, we
need to know which constituents are dependent on which other
constituents, and just what type of dependency relation is
involved. On the one hand such relations seem, to an extent
which varies between different languages, independent of the
facts of word order etc.; on the other, constituent structure is
also important. Both need to be accommodated in a linguistic
description. It happens that in English many dependency
relations can be described in terms of constituent structure
relationships. In some other languages this does not seem the
case. Clearly it is possible to force a description of dependency in

6 There is also the problem that we can find sentences like puellam amat ‘he
loves the girl’, where no separate word is identifiable as the subject. Since the
Latin verb is marked for ‘person’ (amo ‘I love’, amas ‘you love’, amat ‘he loves’,
etc.) the subject is not usually realized as a separate constituent when it is ‘pro¬
nominal’. We do not consider this complication further.
 Syntagmatic and paradigmatic relations in syntax 273
constituent structure terms on any language (as by the use of
rules for Gaelic and Latin like those suggested above), but is this
legitimate or appropriate?
A final comment on word order is in order. There appears to
be a tendency in languages for certain types of ordering
restrictions to occur together. In an influential article Greenberg
(1963) writes:
Linguists are. in general, familiar with the notion that certain
languages tend consistently to put modifying or limiting elements before
those modified or limited, while others just as consistently do the oppo¬
site. Turkish, an example of the former type, puts adjectives before the
nouns they modify, places the object of the verb before the verb, the
dependent genitive before the governing noun, adverbs before adjectives
which they modify, etc. Such languages, moreover, tend to have post¬
positions for concepts expressed by prepositions in English. A language
of the opposite type is Thai [or. as we have seen in this section. Gaelic]
in which adjectives follow the noun, the object follows the verb, the
genitive follows the governing noun, and there are prepositions. The
majority of languages, as for example English, are not well marked in
this respect . . .. More detailed consideration of these and other
phenomena of order soon reveals that some factors are closely related to
each other, while others are relatively independent.

These facts about order lead Greenberg to postulate a number of

universal tendencies, some of which are:

Universal 1: In declarative sentences nominal subject and

object, the dominant order is almost always one in which the
subject precedes the object. (This means that although six orders
are possible (SVO, SOV, VSO, VOS, OSV, OVS), only three
orders occur frequently (SOV, VOS and SVO) and the other
three are rare.
Universal 2: In languages with prepositions, the genitive almost
always follows the governing noun, while in languages with
postpositions it almost always precedes.
Universal 3: Languages with dominant VSO order are always
prepositional.
Universal 4: If a language has dominant SOV order and the
genitive follows the governing noun, then the adjective likewise
follows the noun.

The interested reader is referred to Greenberg’s book for

further discussion and exemplification of these universals.
274 Part three: Functional relations

Linkage

Under the general heading of linkage (a term deriving from

Hockett 1958) we may group together a number of grammatical
constructions whose function is to show that two constituents are
grammatically or referentially related. We will briefly exemplify
two types of linkage, ‘pronominal cross-referencing’ and
‘concord’ or ‘agreement’.
Pronominal cross-referencing can be illustrated in English with
the sentence

36 John thinks that he is intelligent

The sentence has two possible readings: ‘John thinks that he

(John) is intelligent’ and ‘John thinks that he (someone else) is in¬
telligent’, depending on whether John and he are used to refer to
the same individual, or co-refer, or refer to different individuals.
We indicate reference by subscripts: if two items co-refer, then
they will have the same subscript; if they do not they will have
different subscripts. Thus the two readings of our sentence are:

37 John! thinks that he, is intelligent (he = John)

38 John! thinks that he2 is intelligent (he ¥= John)

We suppose that a sentence like 36 is produced by the grammar

and that later rules are able to identify co-referentiality. Exactly
how these rules operate is complex and not yet fully understood,
but the following two informal generalizations account for many,
though by no means all, of the curious features of English
cross-reference:

i If a ‘full’ NP precedes a pronoun, then they may be

co-referential.
ii If a ‘full’ NP commands a pronoun, then they may be
co-referential.

The relation of ‘command’ can be most straightforwardly

illustrated with respect to ‘main’ and ‘subordinate’ or embedded
sentences. An NP in the main sentence commands an NP in a
subordinate sentence if the S that immediately dominates the
main sentence also dominates the subordinate sentence.7

7 The notion derives from Langacker (1969), who formulates the relation
more precisely. The interested reader is referred to this article for further
discussion of this interesting area of English grammar.
 Syntagmatic and paradigmatic relations in syntax 275

S,

Adv NP took off his hat

Time

S,

Figure 96 After NP came into the room

S,

NP took off his hat Adv

Time

S2

Figure 97 After NP came into the room

The first generalization is met in 36: John precedes he, so they

may be co-referential, as is marked in 37. The qualification ‘may’
is introduced since the two do not have to be co-referential, as is
shown by 38. The second generalization is illustrated in Figures
96 and 97. In Figure 96 the NP in the subordinate sentence, S2,
precedes but does not command the NP in the superordinate
sentence, SI, so by generalization i:

39 After John! came into the room, he! took off hisj hat

is well formed. In the same figure the NP in SI commands,

though it does not precede, the NP in S2, so by generalization ii:

40 After hej came into the room Johni took off hisj hat

is also well formed. In Figure 97 the NP in SI both precedes and

commands the NP in S2, so:

41 John] took off hisj hat after hej came into the room

is well formed. But the fourth possibility:

276 Part three: Functional relations
42 *He, took off hist hat after John! came into the room

is not well formed since the, NP in S2 neither precedes nor

commands the NP in SI. In general ‘forwards’ cross-referencing,
as in 39 and 41 is well formed, since the full NP precedes, and
may or may not command, the pronoun, but ‘backwards’
cross-referencing is only possible when the pronoun is
commanded by a full NP; thus 40 is well formed, but 42 is not. So
to return to our original example: 36 is well formed since John
precedes and commands he, but:

43 *He, thinks Johni is intelligent

is ill formed since John does not command he. (Both 42 and 43
are, of course, well formed if he and John are not understood to
be co-referential.) Note that in co-ordinate structures like that
illustrated in Figure 98 forward cross-referencing is possible, by
generalization i; but since neither NP commands the other,
backwards cross-referencing is not:

NP came into the room NP took off his hat

Figure 98

44a John, came into the room and he, took off his hat
44b *He, came into the room and John! took off his hat

Our two generalizations will account for the well-formedness or

otherwise of the following:

45a Mary, married someone she, met at college

45b *She, married someone Mary, met at college

46a John, said that he! wanted to go

46b *He, said that John, wanted to go

47a The insinuation that John, was stupid angered him,

47b The insinuation that he, was stupid angered John,
47c John, was angered by the insinuation that he, was stupid
47d *He, was angered by the insinuation that John, was
stupid.
 Syntagmatic and paradigmatic relations in syntax 277

We will not pursue the mysteries of English forward and back¬

ward pronominalization further. We must, however, make one
further observation about generalization i, since it needs a
modification to cover instances like

48 John! injured himselfj

49 John, injured him2

In a simple sentence, when all the constituents are dominated by

the same S, as in 48 and 49, a full NP and a following non¬
reflexive pronoun cannot be co-referential (there are a few
exceptions to this generalization in sentences like John kept the
book near himiself)). In such circumstances co-referentiality
depends on the presence of a reflexive pronoun, as in 48.
The types of cross-referencing shown for English seem also to
occur in many other languages. Some languages, however, can
draw distinctions that English does not. Thus, several of the
Kwa languages, spoken in West Africa, employ morphologically
distinct pronominal forms in some constructions involving ‘verbs
of reported speech’ to distinguish between reference to the
speaker and reference to other parties. Thus in Efik, a Nigerian
language, the two senses of the sentence:

50a Okon, agreed that he, would pay the debt

(Okon = he)
50b Okon, agreed that he, would pay the debt
(Okon * he)

are differentiated by the use of two different pronouns.

51a Okon, enyime ete imoj eyekpe ison oro

Okon agreed that he would pay debt the (= 50a)
51b Okon, enyime ete enye2 eyekpe ison oro
Okon agreed that he would pay debt the (= 50b)

The ‘obviative’ pronoun imo is only found in embedded sentences

like 51a where it cross-refers to the subject of a verb of
reported speech.
In many languages the main verb may be marked for
cross-reference to the subject. Thus in the Latin sentence:

52 Puer puellam amat

boy girl he loves the boy loves the girl

the form of the verb indicates cross-reference to the subject puer.

278 Part three: Functional relations
In such languages, a sentence need not contain an NP subject
expression. Thus

53 puellam amat
girl he loves he loves the girl

is well formed. In such a case we say that the verb has ‘deictic
reference’. This may be to an NP elsewhere in the linguistic
context - in a preceding sentence perhaps - or to an individual in
the non-linguistic context. In those languages in which verbs are
marked in this way, intransitive verbs may, of course, form a
complete single-word sentence - this is the case with many
modern European languages, like Italian. (This causes problems
for a constituent structure grammar of the sort we examined in
Part one.)
We now turn our attention to ‘concord’ or ‘agreement’. Two,
or more, constituents are said to be ‘in concord’ when they are
both, or all, marked for the same grammatical category. Thus in
English demonstratives and the head noun are said to be in
concord within the NP: this book, these books', that book, those
books. The singular form of the demonstrative can only co-occur
with the singular form of the noun, and the plural form of the
demonstrative with the plural form of the noun - there are no
NPs like * these book. In such constructions one constituent is the
‘controller’ of the concord (in the case illustrated above it is the
head noun), and the other constituents are ‘in concord’ with the
controller: so we say that demonstratives in English concord in
number with the head noun within an NP.8 The reasons why the
head noun in English is held to be the controller were explored
on pages 193-6.

8 It is possible to take a different view. We have taken the view that number is
a category of the noun in English, and that other constituents are in concord with
the head noun for number. We might instead assert that number is a category of
the NP as a whole, and that all relevant constituents within the NP must be
marked for this category. To some extent these different views can be accommo¬
dated in different rules. Thus a rule like:

NP—» Art + Noun

Noun —» N + Number

shows number unambiguously as a category of the Noun. But a rule like:

NP—» Art + N + Number

can imply that Number is a category of the NP as a whole.

 Syntagmatic and paradigmatic relations in syntax 279
In cases like these books the concord category is overtly
marked on the controller (book-s has a plural marker). This is
not always the case: sometimes the controller is not overtly
marked for the category which controls the concord. Thus in
French GARMON is not overtly marked for gender, but it is
‘inherently’ masculine. This covert category controls concord on
other NP constituents, like the article -le gargon, not *la gargon.
Similarly in German MANN and FRAU are not overtly marked
as masculine and feminine respectively, yet this covert
categorization controls concord on articles, adjectives etc. within
the NP - ein junger Mann, eine junge Frau.9
In many languages concord is particularly to be noticed in the
following constructions:

i NP constructions: articles, adjectives and other noun

modifiers often concord with the head noun: concord may
be for number and/or gender and/or case, and possibly
other categories as well.
ii subject + main verb constructions: the main verb is often
in concord for number with the subject; often other
categories, like gender in Arabic and Hebrew, can also
concord.

iii copulative constructions: the complement is often in

concord with the subject, often in number and gender.

English has a limited degree of concord. Some, but not all, NP

constituents show concord as we have seen. Subject-main verb
concord does operate:

54a My father (sing) lives (sing) in Somerset

54b My parents (pi) live (pi) in Somerset

but only in a patchy way (see discussion on pages 210-11). There

is number concord in copulative constructions with NP
complement:

9 Some writers distinguish between ‘concord’ and ‘governmental concord’

depending on whether the concord category is overt (concord) or covert (govern¬
mental concord) on the controller. Thus in French, gender is an instance of gov¬
ernmental concord (since gender is covert), but number is a matter of simple
concord (since number is overt -les gar(ons). The term governmental concord is
clearly derived from the notion of government, discussed on page 256.
280 Part three: Functional relations

55a This man (sing) is a fool: (sing)

55b These men (pi) are fools (pi)

Many languages show no concord at all, or very little. Thus in

Akan we find:

56 Onipa kesee no (man big the) the big man

Nnipa kesee no (men big the) the big men
Onipa no rekasa (man the is-speaking)
the man is speaking
Nnipa no rekasa (men the are-speaking)
the men are speaking

Onipa no ye kesee (man the is big) ithe man is big

Nnipa no ye kesee (men the are big) the men are big

(Number is shown on the noun by the alternation in the form of

the prefix -o-nipa (man) n-nipa (men).) These examples show no
concord at all.
By contrast some languages show very extensive systems of
concord. An example is Luganda, a Bantu) language spoken in
Uganda. All nouns in this language have a structure prefix +
stem. The prefix shows two things: the gender class to which the
noun belongs, and the number of the noun, singular or plural.
Luganda, like other Bantu languages, divides nouns into a large
number of gender classes: we note only three and refer to them
as the omu-aba, en-en and omu-emi classes, since these are the
forms of the singular and plural prefixes taken by the classes in
question. First we illustrate concord within NP, which has the
constituent order N + Adj + Demonstrative + Numeral. As you
will see Adjectives, Demonstratives and Numerals concord for
number and gender class with the head noun:

57a Omu-kazi omu-lungi o-no

woman pretty this this pretty woman

57b Aba-kazi aba-lungi ba-no aba-satu

woman pretty this three
these three pretty women

58a En-jovu en-kadde ey-o

elephant old that that old elephant
 Syntagmatic and paradigmatic relations in syntax 281
58b En-jovu en-kadde ez-o es-satu
elephant old that three
those three old elephants

59a Omu-ti omu-tono gu-no

tree small this this small tree

59b Emi-ti emi-tono gi-no e-satu

tree small this three
these three small trees

Two observations are in order: firstly, while the concord prefixes

for the noun head and the adjective modifier are the ‘same’, dif¬
ferent forms of concord prefix may be found on the demonstra¬
tive and numeral modifiers. Secondly, while the singular noun
and adjective concord prefix is the same for the omu-aba and the
omu-emi classes, the demonstrative and numeral concord prefixes
are differentiated for each class. Similarly the singular and plural
noun and adjective prefixes are the same for the en-en class, the
demonstrative and numeral prefixes differ. This sort of pattern is
very common in Bantu languages.
In sentence constructions there is cross-reference between the
subject NP and the main verb. The verb word always carries a
subject prefix, which is always the initial morph. This subject pre¬
fix not only cross-refers to the subject of the sentence, but it must
also be in number and gender concord with the subject NP. In
the following examples the verb has the structure subject pre¬
fix + tense marker + verb stem:

60a Omu-kazi o-no a-li-fa

woman this she-future-die
this woman will die

60b Aba-kazi ba-no ba-li-fa these women will die

61a En-jovu ey-o e-ri-fa

elephant that it-will-die that elephant will die

61b En-jovu ez-o zi-ri-fa those elephants will die

62a Omu-ti gu-no gu-li-fa

tree this it-will-die this tree will die

62b Emi-ti gi-no gi-li-fa these trees will die

282 Part three: Functional relations

As in Latin (see discussion of examples 52 and 53, pages

277-8), and for the same reasons, the verb word in Luganda can
form a one-word sentence. In this case we say that the pronoun
prefix has deictic reference. So erifa (see 61a) is appropriate
if the speaker refers to a single elephant, cow or other object
which nouns of the en-en class are used to refer to. Similarly
gulifa (see 62a) is appropriate if the speaker refers to a single
tree or some other item which omu-emi class nouns are used to
refer to.
We now turn to consider patterns of concord and cross-
reference in transitive sentences. Basic constituent order is Sub¬
ject-Verb-Object (SVO). In SVO sentences the verb carries a
pronoun prefix which cross-refers to the subject and is in concord
with it for number and gender class. There is no cross-reference
shown in the verb to the object:

63a Omu-kazi a-li-goba en-koko

woman she-will-chase chicken
the woman will chase the chicken

63b Aba-kazi ba-li-goba omu-sota

women they-will-chase snake
the women will chase the snake

(-koko ‘chicken’: class en-en; -sota ‘snake’: class omu-emi.) Now

consider sentences like:

64a Omu-kazi a-li-gi-goba

woman she-will-it-chase
the woman will chase it (sc. chicken)

64b Aba-kazi ba-li-gu-goba

woman they-will-it-chase
the women will chase it (sc. snake)

Comparing the sentences in 63 and 64, we see that when a transi¬

tive verb has no object NP expression, the verb carries an object
pronoun between the tense marker and the verb stem. This has
deictic reference to some appropriate object in the context, and
must be in the appropriate form for the gender class and number
of the noun which is used to refer to such an object. Object pro¬
nouns do not occur within the verb word in sentences of basic
SVO constituent order. Let us give some more examples:
 Syntagmatic and paradigmatic relations in syntax 283
65a Omu-sota gu-li-goba omu-kazi
snake it-will-chase woman
the snake will chase the woman
65b Omu-sota gu-li-mu-goba
snake it-will-her-chase
the snake will chase her (sc. woman)

66a En-koko e-li-goba omu-sota

chicken it-will-chase snake
the chicken will chase the snake
66b En-koko e-li-gu-goba
chicken it-will-it-chase
the chicken will chase it (sc. snake)

In Luganda, unlike English, object pronouns in such sentences

are not found as separate constituents, but are rather infixed into
the verb word.
Both subject and object may be represented by pronouns
within the verb word: then we get one-word sentences like:

67a A-li-gi-goba
she (sc. woman) will chase it (sc. chicken)

67b A-li-gu-goba
she (sc. woman) will chase it (sc. snake)

67c Gu-li-mu-goba
it (sc. snake) will chase her (sc. woman)

67d Gu-li-gi-goba it (sc. snake) will chase it (sc. chicken)

67e E-li-gu-goba it (sc. chicken) will chase it (sc. snake)

and so on.
Finally, we discuss a case where the system of concord and
cross-referencing can be used for a particular communicative
effect. When the order of sentence constituents is SVO we have
seen there is no object pronoun within the verb word. Now note
these sentences, where this basic word order is disturbed: pro¬
noun referencing within the verb allows the speaker and his
hearer to keep track, as it were, of ‘who is doing what to whom’.
The English translations are an attempt to indicate the emphasis
etc. that these sentences carry (small capitals indicate stress in
English):
284 Part three: Functional relations
68a Omu-kazi a-li-goba en-koko (SVO)
the woman will chase the chicken

68b Omu-kazi a-li-gi-goba (SV(O))

the woman will chase it

68c Omu-kazi en-koko a-li-gi-goba (SOV)

the woman will chase the CHICKEN

68d En-koko omu-kazi a-li-gi-goba (OSV)

it's the|CHICKEN the woman will chase

68e En-koko a-li-gi-goba omu-kazi (OVS)

it’s the CHICKEN that will be chased by the woman

Technical terms
agreement exocentric
bi-lateral dependency government
command linkage
concord marker
controller (of concord) mutual dependency
co-ordinate dependency mutual exclusion
co-reference paradigmatic
cross reference pronominal cross-reference
deictic reference reflexive
dependency syntagmatic
endocentric unilateral dependency

Exercises
These two exercises illustrate concord systems in two languages,
Luganda and Gaelic. Both build on previous exercises: Luganda
on pages 215-17 and Gaelic on pages 67-8.

1 Luganda

All the strings in any one group of sentences have the same
structure, indicated at the head of the groups. All the strings are
morphologically analysed. The numbers 1,2,3 correspond to
classes 1,2,3 in the previous exercises; sg = singular and pi =
plural. As before the noun word can be analysed as consisting of
prefix + stem, the form of the prefix depending on the class of the
 Syntagmatic and paradigmatic relations in syntax 285
stem. Adjectives, Demonstratives etc. all bear a prefix concording
with that on the noun stem. The verb also, where appropriate,
has a prefix concording with the subject noun. Fill in the matrix
provided.

A & B NP(N + Adj + Demj (+Num))

Alsg omu-kazi omu-lungi o-no (this pretty woman)
A2sg en-te en-nungi e-no (this nice cow)
A3sg omu-ggo omu-lungi gu-no (this nice stick)
Blpl aba-kazi aba-lungi ba-no These three pretty
aba-satu women)
B2pl en-te en-nungi zi-no es-satu (these three nice
cows)
B3pl emi-ggo emi-rungi gi-no e-satu (these three nice
sticks)

C&D NP(N + Adj + Dem2) + Vb

Clsg omu-sajja omu-kadde oy-o (that old man will
a-li-fa die)
C2sg en-jovu en-kadde ey-o e-ri-fa (that old elephant
will die)
C3sg omu-ti omu-kadde ogw-o (that old tree will
gu-li-fa die)
Dipl aba-sajja aba-kadde ab-o (those old men will
ba-li-fa die)
D2pl en-jovu en-kadde ez-o zi-ri-fa (those old elephants
will die)
D3pl emi-ti emi-kadde ogy-o gi-li-fa (those old trees will
die)

E & F NP(N + Adj + Dem3 (+Num))

Elsg omu-sajja omu-tono o-li (yonder small man)
E2sg en-te en-tono e-ri (yonder small cow)
E3sg omu-ggo omu-tono gu-li (yonder small stick)

Flpl aba-sajja aba-tono ba-li (yonder five small

aba-taano men)
F2pl en-te en-tono zi-ri et-taano (yonder five small
cows)
F3pl emi-ggo emi-tono gi-ri e-taano (yonder five small
sticks)
286 Part three: Functional relations

class noun adjective numeral demonstrative

'this' 'that' 'yonder'

lsg omu- omu- o- oy- o-

Pi

2sg

Pi

3sg

Pi

2 Gaelic

Previous exercises examined NPs of the structure NP —> Art +N:

the following NPs contain adjectives as well:

an taigh ban ‘the white house’

an ramh gearr ‘the short oar’
an tunnag bhan ‘the white duck’
an nabaidh math ‘the good neighbour’
an duilleag mhor ‘the big page’
an sgadan beag ‘the little herring’
an ramh gearr dubh ‘the short black oar’
an sgoil bheag dhona ‘the small bad school’
an tunnag bheag bhan ‘the little white duck’
an sgadan beag math ‘the good little herring’
an luinneag mhath ‘the good song’
an la fada ‘the long day’
an sgoil bheag ‘the little school’
an nighean ghearr ‘the short girl’
an duine mor ‘the big man’
an oidhche fhada ‘the long night’

(a) Write a constituent structure rule that generates the NPs in

this data.

(b) How many gender classes are there in Gaelic? Assign the
Nouns to your gender classes.

(c) In the data given which word class inflects for gender?
 Syntagmatic and paradigmatic relations in syntax 287
(d) You will discover that inflexion for gender (as illustrated in
the data) involves changing the initial consonant of some words.
This process is generally known as lenition. Orthographically
lenition is often marked by ‘adding an h’ (e.g. ban:bhan).
Phonologically lenition is manifested in various ways. In the data
the correspondences are as follows - the orthographic forms are
shown first (and in italics), the phonological realizations are
shown in phonemic brackets:

b : /b/ bh : /v/ m : /m/ mh : /v/ /: /f/ fh : 0

g : /g/ gh : /u/ d : /d/ dh : /u/

(e) Describe the rule for gender concord within the NP.

Previous Gaelic exercises examined some sentences using the

‘attributive copula’ verb form bha. Please refer to that data and
your analysis. Here are four more similar sentences:

bha an taigh beag ‘the house was little’

bha an duine math ‘the man was good’
bha an oidhche fada ‘the night was long’
bha an nighean dona ‘the girl was bad’

(f) What is the rule for concord in these sentences between the
subject noun and the adjective?
On the basis of your analysis to earlier questions, analyse the
following sentences:

bha an tunnag bheag ban ‘the little duck was white’

bha an tunnag beag ban ‘the duck was little and white’
bha an sgoil mhor dona ‘the big school was bad’
bha an sgoil mor dona ‘the school was big and bad’
bha an duine mor dubh ‘the big man was black’/
‘the man was big and black’
bha an taigh beag ban ‘the little house was white’/
‘the house was little and white’

(g) Notice that no ambiguity arises in the first two pairs of

sentences, but that the last two sentences are ambiguous. How do
you account for this?

(h) Draw two tree diagrams for the final sentence which will
show how the ambiguity arises.
18 Processes and participants

This chapter is concerned with what we shall refer to as the

‘propositional structure’ of the sentence. By this we mean a
specification of:

i the type of ‘state’ or ‘action’ described by the sentence - this

we call the ‘process’ of the sentence, and it is largely
associated with the verb.
ii the ‘participant roles’ involved in the state or action - these
are typically associated with NPs in construction with the
verb, and describe their functional relation to the verb.

A description of the process and participants forms the

‘propositional nucleus’, which it is convenient to describe as ‘one
place’ (e.g .John ran), ‘two place’ {John sharpened the knife), ‘three
place’ {John gave a book to Mary), etc., depending on the number of
participants we identify. From time to time it is useful to refer to
‘circumstantial’ roles associated with the propositional nucleus:
these are typically adverbial expressions of time, place, manner,
etc., which we have already observed to be optional constituents
of a constituent structure description, and are, in general,
optional also to the propositional structure, for much the same
reasons. John sharpened the knife in the woodshed last night
involves a two-place propositional nucleus with circumstantial
constituents of place and time. Our major concern is the
propositional nucleus. The sorts of processes and participants we
identify are probably common to all languages, though their
sentential realizations differ from language to language, and our
discussion focuses on English.
Let us begin by considering the sentences:

1 John sharpened the knife

2 The knife was sharp

In simple constituent structure terms, discussed earlier, these

 Processes and participants 289

sentences are dissimilar: 1 is a transitive and 2 a copular

sentence. In propositional terms they are more closely related: 1
describes an action (‘sharpening a knife’) that brings about the
state of affairs (‘the knife being sharp’) described in 2. The link is
reflected in the lexical relations between the sentences. The NP
the knife that is the object of the action sentence is the subject of
the state sentence; furthermore the verb SHARPEN in 1 is
morphologically related to the adjective sharp in 2.
The distinction between sentences describing actions and those
describing states is a fundamental one that we will find in other
pairs of sentences we will examine.1
We need to recognize different types of [action] and [state], so we
must add a further specification to each characterization. We say
that 1 expresses a [directed], [causative], [action]. It is [directed]
because it involves the action of an [agent] (John) on a [patient]
(the knife): the [agent] is the participant responsible for
performing the action, the ‘doer’; the [patient] is the participant
which, to use traditional terminology, ‘receives or suffers the
action of the verb’. Not all two-place [action] propositions involve
the [directed, action] of an [agent] on a [patient]:

3 John played tiddly winks

4 John ran a mile

In these cases, the [agent] cannot be seen as ‘doing anything’ to

the other participant. In 1 the [action] is [causative] because the
action of ‘sharpening the knife’ causes the state of affairs where
the ‘knife is sharp’ as may be seen from the somewhat stilted
paraphrases:

5a John made the knife sharp

5b John caused the knife to be sharp

Not all two-place propositions are [causative]: those in 3, 4 are

clearly not, and no paraphrase like 5 is open to them. Nor are all
[directed action] propositions [causative]:

1 To avoid confusion we call sentences like 1 [action] sentences, rather than

‘active’ sentences, reserving the form ‘active’ for the opposition ‘active: passive’.
Thus we may say the 1 is an [action] sentence in the ‘active’ form. The corre¬
sponding ‘passive’ form of this [action] sentence is The knife was sharpened by John.
We enclose characterizations of processes and participants in square brackets, to
indicate that these are syntactico-semantic ‘features’ of the relevant propositions
(‘syntactico-semantic’, since they have both semantic and syntactic implications).
290 Part three: Functional relations

6a John beat the dog

6b John built a house

The [state] proposition in 2 can be further specified as

[descriptive]: the knife is described as having the [attribute]
‘sharpness’. This contrasts with other [state] specifications
examined later - a [possessive, state] 7 and [locative, state] 8:

7 John has a knife

8 John is in bed

Turning now to the participants involved, note first that both

sentences contain the NP the knife. Since this participant is
common to both propositions we call it a [neutral] participant
(other reasons for choosing this characterization will emerge).
We have already seen a further characterization of the [neutral]
participant in 1 as a [patient], and we can characterize the object
of 6a also as [neutral, patient]. The characterization as [patient] is
clearly not appropriate to the object of build in 6b - this we
characterize as [neutral, result], since the house can be seen to be
the result of the building (in a way that the dog in 6a can hardly
be described as the result of the beating). The participants tiddly
winks and a mile in 3 and 4 cannot appropriately be specified as
either [patient] or [result]; we characterize them as [range], thus
further characterizing the type of process by further specifying it,
limiting it in some way, expressing a measure, etc. (This
participant type could be further subclassified, since it covers a
number of different semantic notions.) None of the [range]
participants of this description can be considered in any way the
‘object’ of the verb concerned - whereas [patient] and [result]
participants can be classified appropriately in this way. Some
verbs that can be characterized in these terms are shown in
Figure 99. (Note that both [patient] and [result] participants can
expect to become the subjects of passive sentences, but this is in
general impossible or infelicitous for [range] participants.)
The remaining participant in 1, (John), we have already
specified as [agent], and we offer no further sub-classification of
[agent] in this chapter (but see footnotes 8 and 13 below). We
specify the participant realized by John in sentences 3-6 also as
[agent]. Such a characterization is clearly inappropriate for 7-8,
and we return to discuss these sentences later. Turning now to 2,
we have already characterized the knife as a [neutral] participant,
 Processes and participants 291

Verbs involving [range] participants (see 11)

agent action range

John is playing tiddly winks
Mary sang a song
Harold ran a mile

Verbs involving [range] participants, but not [agent] subjects

neutral state range

John weighs twelve stone
This cost £100
Her bust measures 42 inches

Non-[causative] verbs with [neutral, patient] participants

(see 12)

action neutral
agent directed patient
John beat his wife
The cat chased the dog
The vandal slashed the priceless painting
The cat has scratched me
I painted the sitting room

Verbs involving [result] participants (see 13)

agent action neutral result

Wren built St Paul’s Cathedral
Peter is digging a hole
John burnt a hole (in the carpet with
his cigar)
Velasquez painted the Rokeby Venus

Note that some verbs can take either a [result] or a [patient]

participant:
Velasquez painted the Rokeby Venus [result]
I painted the sitting room [patient]
John burnt a hole in the carpet [result]
John burnt his old love letters [patient]

Figure 99
292 Part three: Functional relations
and no further characterization seems necessary; we have noted
that we can describe sharp in 2 as an [attribute].
These observations are summarized in 9-13: we refer to such
characterizations as propositional structures.

9 agent action neutral (=1)

directed patient
causative
descriptive
10 neutral state attribute (=2)
descriptive
11 agent action range (=3, 4)
12 agent action neutral (=6a)
directed patient
13 agent action neutral (=6b)
directed result

(Propositional structures for 7-8 are discussed later.)

We will represent propositional structure, as here, in a form

which corresponds to the order of constituents in an active
declarative sentence. They should however be regarded as
abstract representations of the propositional relations within
sentences and will consequently correspond to a variety of actual
sentences, where the form depends on considerations other than
simple propositional structure. Thus 9 is the propositional
structure appropriate for both:

14 John sharpened the knife

The knife was sharpened by John

The relationship between propositional and constituent structures

is discussed briefly at the end of this chapter.
So far our characterization has been entirely in semantic terms,
but the footnote on page 289 said that features like [action] are
syntactic-semantic, and we now enquire whether there are
syntactic correlates to these features. We will see that there are
no absolute one-to-one correlations, but a number of
generalizations can be usefully made.
The [state] : [action] distinction is reflected in the following
(we exemplify using the propositional structures 9 and 10: the
reader is invited to verify the observations with respect to other
 Processes and participants 293

examples given here and later). The verb in [state] sentences

cannot occur in progressive form: (but see 110, page 311, for an
exception to this)

15a John is sharpening the knife

15b *The knife is being sharp
Only [action] propositions containing an [agent] can be put in
correspondence with imperative sentences:

16a Sharpen the knife

16b *Be sharp!

Only [action] propositions are typically associated with

[instrument] 17 and [benefactive] 18 roles:

17a John is sharpening the knife with a whetstone

17b *The knife is sharp with a whetstone

(The [instrument] role is realized by with a whetstone-, note that

[instrument] roles are frequently identifiable by the use of the
preposition with - we note the frequent association of roles with
particular prepositions.)

18a John is sharpening the knife for his mother

18b *The knife is sharp for his mother

(The [benefactive] role is realized by for his mother - she is the

participant who ‘benefits’ from the action: it can frequently be
identified by the preposition for. [State] propositions cannot
usually be associated with circumstantial roles of manner, place
and intention:
19a John is sharpening the knife carefully
19b *The knife is sharp carefully (manner)
20a John is sharpening the knife in the woodshed
20b *The knife is sharp in the woodshed (place)
21a John is sharpening the knife intentionally
21b *The knife is sharp intentionally (intention)

Circumstantial roles of manner and intention seem

characteristically tied to the occurrence of [agent], as John was
careful in sharpening the knife. John acted intentionally in
sharpening the knife. One might attest 21b, but only if it has the
sense ‘it was someone’s intention (agentive) that the knife should
be sharp’: this sentence can hardly be construed with the sense
294 Part three: Functional relations

‘the knife had the intention of being sharp’. Place participants, on

the other hand, appear tied to the proposition as a whole: Where
John sharpened the knife was in the woodshed. * Where the knife
was sharp was in the woodshed is nonsense — it implies that it
wasn’t sharp elsewhere!
Let us now turn to some distinctions that can be drawn
between the various types of [action] proposition. They can
typically be questioned by:

22 What happened?

Thus:

23a What happened?

23b John sharpened the knife
23c John ran a race

etc. This is not possible for [state] propositions. [Action]

propositions with [agent] can typically be questioned by:

24 What did [agent] do?

Thus:

25a What did John do?

25b He sharpened the knife
25c He built a house

etc. We have not yet encountered [action] sentences without

[agent], but this generalization can be tested when we do. [Action]
propositions with [agent] and [patient] can typically be
questioned with:

26 What did [agent] do to [patient]?

Thus:

27a What did John do to the knife?

27b He sharpened it
28a What did John do to the dog?
28b He beat it

but hardly:

29a What did John do to the house?

29b He built it

etc. Similarly we find the question:

 Processes and participants 295

30a What happened to [patient]?

as in:

31a What happened to the dog?

31b John beat it

but not:

32a What happened to the race?

32b John ran it

Note also that [action] sentences with [agent] and [patient] or

[result] correspond to passive sentences:

33 The dog was beaten by John

The house was built by John

but less felicitously:

34 Tiddlywinks was played by John

Further observe that the preposition by marks [agent]

participants (as with marks [instrument] and/or [benefactive]).
Propositional structures with [state] do not lend themselves to
question-and-answer sequences like those illustrated above: a
proposition like 10 can only typically be questioned by questions
like:

35 What was the knife like?

36 Can you describe the knife?

Another propositional structure seems to be, as it were,

intermediate between the structures of 9 and 10. The relationship
between these two structures and our new structure can be
illustrated in the three sentences:

37 The shirt is torn (= 10)

38 The shirt is tearing
39 John is tearing the shirt (=9)

We characterize 38 as

40 neutral action
patient inchoative
descriptive

The subject participant we characterize as [neutral, patient]

296 Part three: Functional relations
since it is affected by the action of the verb, just as the object
participant is in 39. The characterization [inchoative] in 40 is the
traditional term for ‘change-of-state’ verbs: the sentence 38
describes a shirt changing from a state where the shirt is not torn
to one where it is torn. These characterizations can be checked by
using the tests outlined in 15 onwards.
A very large number of sentences in English can be related to
each other in the manner shown in 37-39: [state], [inchoative,
action] and [causative, directed, action]. Some lexical items that
are related in this way are shown in Figure 100, and the reader is
invited to check this by constructing appropriate sentences. In
Figure 100 the three columns are to be understood as relating,
respectively, to the characterizations in 10, 40 and 9.

action
action directed
state inchoative causative
descriptive descriptive descriptive
(= 10) (=40) (= 9)
a RIPE RIPEN RIPEN
WIDE WIDEN WIDEN
b YELLOW YELLOW YELLOW
COOL COOL COOL
c TORN TEAR TEAR
CLOSED CLOSE CLOSE
d MOLTEN MELT MELT
ROTTEN ROT ROT
e SHARP — SHARPEN
MOIST — MOISTEN
f LARGE — ENLARGE
BOLD — EMBOLDEN
g PASSIVE — PASSIVIZE
BEAUTIFUL — BEAUTIFY
h DEAD DIE KILL
HIGH RISE RAISE
i — VANISH —

Figure 100

A number of observations can be made about the sets of items

shown in Figure 100. The relationship typically involves a form that
is syntactically an adjective (left-hand column), a form that is an
intransitive verb (central column), and a transitive verb (right-
 Processes and participants 297

hand column). Sometimes (a, e, f, g) there is a morphological

relationship of derivation (RIPE : ripen, etc.), sometimes (b)
the ‘same’ form is used in all three columns (yellow); some¬
times (c) the ‘adjectival’ form is the same as the passive participle
of the corresponding verb (torn : tear); sometimes (d) the
adjectival form is a former passive participle (molten : melt);2
sometimes (h) there is a suppletive relationship between adjective,
and one or more of the verb forms (dead: die: kill). There
are also cases where there is a causative form, but no corre¬
sponding inchoative (sharp : sharpen).3
Some other asymmetries are shown in Figure 100.
In some of the cases involving asymmetries an alternative expres¬
sion, as it were, ‘fills the gap’. Thus, for example, corresponding
to:
41a The gap is narrow
41b The gap is narrowing
41c John is narrowing the gap
we find:
42a The gap is large
42b The gap is getting larger/the gap is becoming larger
42c John is enlarging the gap

but there is no sentence:

43 *The gap is enlarging
Sentences like 42b can also be constructed corresponding to, and
paraphrasing, 41b:

2 This fact may lead one to question the analysis of the passive adopted so far.
Note that the ‘short’ passive form, where the agent is deleted (The shirt was torn {by
John)) is identical to the adjective form. It is usually held that the adjectival and
passive forms can be distinguished by the possibility of inserting an adjective mod¬
ifier in the one but not the other: thus, the shirt was rather torn but *the shirt was
rather torn by John.

3 The sentence The knife is sharpening is not typically found in the inchoative
sense ‘The knife is becoming sharper’. Note that we can find sentences like The
knife sharpens easily. Jespersen (1927) describes such sentences as ‘medio-
passives’, intermediate between an active and a passive use. They are typically
understood in a sense like ‘I find it easy to sharpen this knife’: i.e. a sense involving
an understood [agent]. Such sentences usually require an adverbial expression.
We will not consider them further here.
298 Part three: Functional relations

44 The gap is getting narrow

The gap is becoming narrow

What is the relationship between 41b and the paraphrases in 44?

We postulate a propositional structure as in 45 for such sen¬
tences:4
45 neutral action attribute
patient inchoative
descriptive

We may compare 45 with the characterization already given for

[state] sentences, like 41a:

46 neutral state attribute

descriptive

and [inchoative] sentences like (41b)

47 neutral action
patient inchoative
descriptive

These propositional structures are related in interesting ways.

The structure in 46 is that realized by a [state] copulative sen¬
tence, which we have seen typically involves the copular verb BE,
though other copular verbs like SEEM are equally appropriate
in such structures. The structure in 45 is also that of a copular
sentence, but this time involves an [action, inchoative] copular verb,
become, get, and we may add TURN (see pages 50-1 for a dis¬
cussion of copular sentences). The structures in 45 and 47 both
involve an [action, inchoative, descriptive] process, but in 45 we
find a copular verb (become, etc.) in construction with an
adjective, whereas in 47 we find the [inchoative] verb that mor-

4 We may also observe that the attribute tends to denote a ‘result’ rather than a
‘state’. This is particularly noticeable if we consider such sentences with the perfect
verb form: My tea has got cold implies that it is now cold, but once was not cold -
such sentences are sometimes said to involve a ‘resultant state’ (a state reached
after some inchoative process), in contrast to a simple ‘state’, as in the tap water is
cold, where one might suppose either that it never was hot, or that the speaker is
not concerned to make this implication. The sentence the tap water has got cold is
appropriate if the speaker refers to the ‘hot’ tap which is now running ‘cold’ as the
‘hot water’ is no longer ‘hot’, or if the water is colder than usual. This is
perhaps why sentences like the ice has got cold are anomalous, since there is an
apparent and curious implication that ‘the ice’ was once ‘not cold’.
 Processes and participants 299

phologically corresponds to the adjective appropriate to 45. This

suggests that sentences like:

48 The apples are getting ripe

are in some sense ‘analytic’: the [inchoative] feature of the pro¬

cess is overtly ‘spelled out’ by the verb GET; while sentences like

49 The apples are ripening

are in some sense ‘synthetic’, the [inchoative] feature is as it were

‘incorporated’ into the adjective, which now becomes morphologi¬
cally a verb. An idiosyncracy of the lexical structure of English is
that this process is inhibited for LARGE, so we find no sentence
like 43. (It is possible to consider increase in the sentence the
gap is increasing as a suppletive, inchoative form - there is no
adjective corresponding to INCREASE.)5
We can make similar observations with respect to the [causa¬
tive] structures we have identified. Thus, for example, VANISH is
only found as an intransitive [inchoative] verb:

50 The rabbit is vanishing

and there is no corresponding [causative] (though it would be

comprehensible if it did occur):

51 *The magician vanished the rabbit

There are, however, the [causative] constructions:

52 The magician made the rabbit vanish

The magician caused the rabbit to vanish

5 We may also observe that in some languages what in English is realized as a

construction be + Adj. is realized as a verb (page 60). In Akan there is no
adjective corresponding to cold in English; instead we find the verb dwo to be,
or become, cold’. Furthermore, there is in this language a special ‘stative’ verb form
that is appropriate for such verbs. So we find:

nsuo no dwo (water the be-cold (stative)) ‘the water is cold’

as the [state] sentence corresponding to the English translation given. In this lan¬
guage an [inchoative] sense is realized, much as in English, by the progressive verb
form:

nsuo no redwo (water the is-cooling) ‘The water is getting cold’

In this respect Akan is a ‘synthetic’ language in a way that English is not: since both
the Akan examples involve intransitive verbs,.
300 Part three: Functional relations

and we have already seen (see 5) that such sentences paraphrase

causative sentences. Here again we can consider that MAKE and
cause function as overt markers of the causativity of the sen¬
tences involved, and that sentences like 52 or 53b, 53c are
‘analytic’, while sentences like 53a are ‘synthetic’.

53a The sun ripened the apples

53b The sun made the apples ripen
53c The sun caused the apples to ripen

By exactly the same sort of argument we can analyse DO and

happen discussed in 22 to 32, as items that spell out the [action]
nature of sentences which they paraphrase:

54 What the magician did was make the rabbit vanish

So an analytic sentence like

55 What the sun did was make the apples become ripe

which paraphrases the synthetic

56 The sun ripened the apples,

overtly marks the [action] (do), [causative] (make) and

[inchoative] (BECOME) senses underlying 56.
Two questions immediately present themselves: why does
English have these two opposing tendencies?6 Does this fact
affect our analysis?

6 Both tendencies are very common in English, and may be attested in a much
wider range of examples than we have space for. Thus, for example, there is a
paraphrase relation between the complex expressions in the left-hand column
below, and the corresponding single verbs in the right-hand column.

fix with a nail nail

feed with a spoon spoon feed
hit with the hand slap
poke with the finger finger
make a hole hole
travel by foot walk
make a presentation to present
give a battering to batter

The process shown is still extremely productive, as can be seen from sentences
like:

She eyeballed the student’s essays

I train to work every morning
 Processes and participants 301

The chief communicative value of the analytic forms appears to

be that they enable a speaker to make some particular distinctions
that are either neutralized in the synthetic forms, or are syntacti¬
cally impossible. A distinction which is neutralized with the
inchoative verbs is that between a comparative and an absolute
usage. Thus:

57 The blackberries are ripening

does not distinguish between:

58 The blackberries are getting riper (comparative)

and:

59 The blackberries are getting ripe (absolute)

In a sentence like:

60 Let the blackberries ripen before1 you pick them

it is not clear whether the sense is:

61 Let the blackberries get riper before you pick them

or

62 Let the blackberries get ripe before you pick them

Cases which involve syntactic impossibilities are those where the

analytic form permits one constituent to be modified, say by an
adverbial expression, but where no comparable synthetic form is
possible. Thus:

63 The blackberries are getting somewhat too ripe

cannot be paraphrased by a sentence using the corresponding

inchoative verb.7

The derivational relationships that can be established are various, and work both
as a ‘verbalizing’ process (e.g .fix with a nail - nail) and as a ‘nominalizing’ process
(e.g. batter - give a battering to). For an interesting and wide-ranging discussion
of such examples see Liefrinck (1973).

7 This feature is even more noticeable with the analytic expressions mentioned
in the previous note. There are no synthetic expressions that can correspond to
‘poke with the forefinger’; ‘travel by the nine o’clock train’; ‘fix with a four inch
rusty nail’; etc.
302 Part three: Functional relations
The problem that examples like 55 pose for the analysis
adopted hitherto is that such sentences clearly involve
embedding. An analysis sometimes suggested looks like:
64 (Sun CAUSE (apples BECOME (apples BE ripe)))
or indeed, if we wish to indicate, the action nature of the
sentence:
65 (Sun DO (sun CAUSE (apples BECOME (apples BE
ripe))))
A formula of this sort quite neatly captures the relationships we
have been discussing, since the successive bracketings represent
the [state] proposition ((apples BE ripe)); the [inchoative]
proposition ((apples BECOME (apples BE ripe))); the [causative]
proposition; etc. We can then suppose that the minimum
amalgamation of the various elements of the representation
would involve the most analytic form - perhaps:
66 The sun did make the apples come to be ripe
(if this is acceptable!) - and the maximum amalgamation of the
various elements results in the most synthetic form:
67 The sun ripened the apples
Such an analysis obviously has its attractions, and many of the
structures we discuss can usefully be examined in this light: some
indeed involve embedding more obviously than the structures we
have been considering.8 We will not, however, pursue such

8 Such analyses also raise questions. For example, how widely can the para¬
phrases involving become, get, make, cause be applied, and what restrictions
are there on such paraphrases? One example will suffice. We have analysed sen¬
tences like the following as containing an [agent] participant:

a John split the wood

b The sun split the wood
However, to group together ‘animate’ expressions like John, and ‘natural forces’,
like the sun, begs questions. In a John is a ‘direct’ agent in that he actually splits the
wood - thus we can add instrumental phrases like with an axe; in the b sentence the
sun seems to be an ‘indirect’ agent and instrumental phrases seem rather odd
(unless we see the sun as ‘personified’):

The sun split the wood with its great heat

Furthermore, the paraphrase with make only seems to apply felicitously when we
are dealing with the indirect agent:
 Processes and participants 303
analyses further here since our purpose is to take a more
straightforward and taxonomic look at some propositional
structures in English and classify verbs in terms of the
propositional structures, they can enter.
We now turn to some further propositional structures, which,
for reasons of space, are discussed in less detail. We begin with:
68 John had a book
69a Mary gave John a book
69b Mary gave a book to John
Again note a distinction between a [state] process 68, and an
[action] process 69. The [state] process is further characterized as
[possessive], for obvious reasons; and the [action] process ^s
[directed] and [causative], in the senses already discussed.9
As to the participant roles involved, we say that in both
sentences a book realizes a [neutral] role. In 69a John realizes the
participant role of [goal], the book changes hands from Mary, the
[source] of the book, to John, the [goal]. We also say that Mary
realizes an [agent] role, since she is the actor responsible
for the transfer. This description contrasts with the description
appropriate for a sentence like:

The sun made the wood split

If we have:

John made the wood split

John, it seems, must be understood as an indirect agent - he might have made it

split by leaving it in the sun, forgetting to apply teak oil, etc., but we hardly under¬
stand this sentence as involving direct agency:

?John made the wood split with an axe

One reason why vanish is not used as a transitive verb may be that it involves
indirect, rather than direct agency. After all, it is ‘the rabbit’ which ‘does the
vanishing’! A possible answer is to make further subclassifications of the partici¬
pant role [agent], but it is not clear how far this process would have to be taken, or
even if there would be an end to it! We shall continue to recognize a single partici¬
pant agent.

9 The feature [possessive] must be understood in a wide sense since, as will

become clear, not all the verbs involved are ‘possessive’ in the ordinary language
sense. However, all the verbs do fall into a semantic class which involves possession
and other notions related to this. The same is true, to a greater or lesser extent, of
all our characterizations.
304 Part three: Functional relations

70 Mary took the book from John

where John is now to be identified as [source] and Mary as both
[actor] and [goal]. In 69b the [goal] participant is identified by the
preposition to, the typical [goal] preposition, and in 70 the
[source] participant is identified by the preposition, from, the
typical [source] preposition. What now of the participant realized by
John in the [state] sentence 68? This role we identify as [dative];
John here is neither [source] nor [goal], if these bear the
connotations of movement ‘to’ and ‘from’ suggested by our
previous description. [Dative] is typically found in [state]
propositions, and it too, when not in subject position, may be
associated with the preposition to, as in:
71 The book belongs to John

where the order of participants is [neutral]-[dative], rather than

[dative]-[neutral] as in 68.
There is one further characterization to be made. In 69 are two
alternative versions of the sentence, one in which John and the
other in which a book is the object of GIVE. Further, when the
[goal] participant becomes the object of the verb, it loses its
preposition. We account for this by supposing that the further
characterization [patient] can be added to either [goal] or
[neutral]. One advantage is that this enables us to say that the
[patient] participant may become the subject of the
corresponding passive sentence: we find both:

72a John was given a book by Mary

72b A book was given to John by Mary

Some support for this analysis may be gained by observing that

this additional specification may be made only to [goal] or [neutral]
participants: the [source] cannot become also [patient]. Thus the
sentence:

73 Mary took John a book

can only be understood as:

74 Mary took a book to John

not as:

75 Mary took a book from John

 Processes and participants 305
In 73, according to our analysis, John is [goal], and the
generalizations made above hold; in 75 John is [source] and this
cannot become a [patient] participant. Similarly the passive
sentence:

76 John was taken a book by Mary

can only be understood in a sense where John is [goal] not

[source]. Finally, [source] participants do not usually lose their
preposition from.
The propositional structures we propose, then, are:10

77 dative state neutral

possession

78 agent action neutral goal

source directed X\ y/
causative '•patient
possession

D St N A/S Ac N G A/G Ac - N/P S

\pf

HAVE GIVE TAKE

OWN TAKE BUY

SELL BORROW

LEND RENT

HIRE HIRE

RENT

N St D

BELONG PRESENT SNATCH

GRAB

GET

Figure 101 (see footnote 11, page 306)

10 In 78 we indicate by the arrows that [patient] may be attached either to the

[neutral] or to the [goal] participant.
306 Part three: Functional relations

19 agent action neutral source

goal directed patient
causative
possessive

Some of the verbs that can occur in these structures are set
out in Figure 101.11 Some are in pairs (BUY : SELL;
BORROW: lend) that contrast as to whether they have [agent,
source] or [agent : goal] subjects; some (e.g. take) can occur
in either structure; and others (e.g. GRAB, present) can only
occur in one or the other structure.
This analysis can be extended to cover a number of other
verbs, but a problem is that [neutral] participants are sometimes
obligatorily marked with prepositions: so, for example:

80 John robbed Mary of £10

We suggest here a propositional structure of the form:

81 agent action source neutral

goal directed patient (of)
causative
possessive

Note the source participant is marked as necessarily [patient],

indicating that it must be the object of the verb (as discussed
above), and may be the subject of the corresponding passive
sentence: we have also shown the [neutral] participant as marked
by the proposition of. This is the only possible structure for
ROB since there is no sentence:

82 *John robbed £10 from Mary

This ‘sense’ must be expressed as:

83a John stole £10 from Mary

84 agent action neutral source

goal directed patient
causative
possessive

11 Here and in subsequent figures, we use the following abbreviations: D(ative);

A(gent); S(ource); G(oal); P(atient); N(eutral); St(ate); Ac(tion) - other process
characterizations can to be supplied from the text.
 Processes and participants 307

where this time the [neutral] participant must be [patient] (it

must be object of the active sentence, and may be subject of the
corresponding passive sentence): we have not specified the
preposition for [source], since this is the typical preposition from.
A final instance may be illustrated with the sentences

85 The Chinese supplied arms to the Vietcong

86 The Chinese supplied the Vietcong with arms
87 agent action neutral goal
source directed (with)\^ y/
causative patient
Here, either the [neutral] or [goal] participant may be [patient]:
as such it becomes the object of the verb in an active sentence,
and appears with no preposition, or may be the subject of a
passive sentence. The typical [goal] preposition, is to and is not
therefore specified. Other verbs that can enter a propositional
structure like 87 include FURNISH, ISSUE, FEED, AWARD.
We now turn from propositions related to possession to
consider some locative propositions. Consider the sentences:

88 The lamp is on the table

89 John put the lamp on the table
90 John took the lamp off the table

We characterize these as:

91 neutral state Lplace
locative
92 agent action neutral Lgoal
directed patient
causative
locative
93 agent action neutral Lsource
directed patient
causative
locative

Some verbs that can enter the propositional frames shown in

91-93 are shown in Figure 102.
There are two points to make about these structures. First,
note that [state, locative] verbs occur in the progressive form,
even though they are [state] sentences (see remarks on 15, page
308 Part three: Functional relations

[ St LP A Ac N/P LG A Ac N/P LS

STAND STAND REMOVE

SIT SIT TAKE

LIE LAY EXTRACT

HANG HANG EJECT

PUT

PLACE

TAKE

Figure 102

293). We will see later (see 109) that sentences like John is sitting
down have two senses: one is a [state] sense (in the state of being
sat down) as is required in the present examples: the other an
[inchoative] sense (in the process of sitting down), which we
discuss later. The other remark we may make is with respect to
the specification [Lgoal] (locative goal) and [Lsource] (locative
source). These specifications remind us of the close connections
between [possessive] and [locative] propositions. Semantically,
the relevant constituents are as much [goal] and [source] here, as
they were in the [possessive] sentences, and some verbs can
enter both [locative] propositional structures and [possessive]
propositional structures:

94a John took the lamp from Mary

94b John took the lamp from the table

There are differences. [Source] and [goal], we have seen, are

characteristically identified by the prepositions from and to
respectively. [Lsource] and [Lgoal] have a wider range of
prepositions open to them - we can replace from in 94b by off
(and in other sentences by out of, etc.) The prepositions in
[locative] sentences have a more significant semantic role to play
than the simple marking of propositional function. Furthermore,
as we have seen, [source] and [goal] may combine with [agent] -
see Figure 101 and associated discussion - but [Lgoal] and
[Lsource] do not characteristically combine with [agent] in
this fashion. On the other hand there are interesting inter¬
relationships. Here we mention only one. The typical [state,
 Processes and participants 309

possession] verb is have; one of the commonest [state, locative]

verbs is be, often referred to as a locative copula:

95 John has a book

96 The lamp is on the table

Now consider the sentences:

97 The book is John’s

98 The table has a book on it.

HAVE and BE have ‘changed places’. One thing peculiar about

this relationship is that it appears to be related to the definiteness
of the [neutral] participant. Thus in the [possessive] sentences, if
the [neutral] participant is indefinite, we typically use have
rather than BE:

99a John has a book

99b ?A book is John’s

and the same is true of the [locative] sentences:

100a The table has a book on it

100b ?A book is on the table

Indeed, many [possessive] meanings appear to be realized as

apparently [locative] sentences:

101a Have you any matches on you?

101b The book is in my care
101c The evidence is in my possession

For now we need merely note that there is some justification for
an apparently similar analysis.
Another set of [locative] sentences, that once again show a
similarity to [possession] propositional structures, can be
illustrated with:

102a John planted roses in the garden

102b John planted the garden with roses
103 agent action neutraL Lgoal
directed (wiTHjN^ /
causative patient
locative

Like the similar [possessive] structures discussed in 86-87, we find

that either the [neutral] or the [goal] participant may become the
310 Part three: Functional relations

object of the verb (or the subject of the corresponding passive

sentence); hence, once more, we identify [patient] as a role
associated with either participant. And, as before, if the [neutral]
participant is not also [patient] then it occurs with the preposition
with. Some other verbs that enter this structure are shown in
Figure 103. (They fall roughly into semantic ‘sets’, and are set out
in this way in Figure 103: the reader is invited to supply further
examples.) A peculiarity of many, but not all, verbs that enter this
propositional frame is that when [patient] is associated with
[Lgoal], the sentence tends to have an implication of
‘completeness’: thus 102b tends to imply that the whole garden, or
the whole of the relevant parts of the garden, has been ‘planted
with roses’. The corresponding sentence, where [patient] is
associated with the [neutral] participant, has no such implication.
This implication can be confirmed by comparing the following:

104a John planted a rose in the garden

104b *John planted the garden with a rose
105a John planted the garden all over with roses
105b *John planted roses all over in the garden

This implication may account for the fact that fill, which
necessarily implies completion, can only occur in a sentence of the
form 102b:

106a John filled his house with flowers

106b *John filled flowers in his house

in contrast with pour, which can only occur in a sentence of the

pattern 102a:

107a John poured petrol over the floor

107b *John poured the floor with petrol

It does not however explain why some verbs do not have any
particular implications associated with either order:

108a He marked his name on his shirt collars

108b He marked his shirt collars with his name

Finally a few verbs occur in a structure like 103, but, instead of an

[Lgoal] participant, we find an [Lsource] participant:

109a John stripped the bark off the tree

109b John stripped the tree of its bark
 Processes and participants 311

Constructions like these lend themselves very interestingly to an

analysis in terms of embedded predications.

^ Ac N (with) LG A Ac N(of) LS A Ac N LG/P

\p/ \p/

PLANT MARK STRIP POUR

CRAM ENGRAVE EMPTY PUT

PILE INSCRIBE

STACK A Ac N/P(with) LG

DAUB FILL

PAINT COVER

SPRAY

SMEAR Figure 103

Thus far we have considered [state] and [action, causative]

propositions involving [possession] and [locative] verbs. We may
now enquire whether there are also [inchoative] verbs in these
classes. We first discuss [locative] verbs. Consider the sentence:

110 John is sitting down

We have already noted (page 308) that 110 is capable of a

[state] interpretation and the verb is in the progressive form, even
though it realizes a [state] proposition. We have also noted that
110 is capable of an [action] interpretation. The two
interpretations we might represent as:

111a John is (in the state of) sitting down

111b John is (in the process of) sitting down

The latter interpretation may be analysed as [inchoative]. We

propose a propositional structure:

112 agent action Lplace

patient inchoative
locative

The only comment we make here on this frame is with respect to

the characterization of the subject as both [agent] and [patient]:
i.e. both the agent responsible for the action, and the patient who
312 Part three: Functional relations

is affected by the action. This analysis implies that such structures

are ‘reflexive’.12

We also find sentences with a non-agentive subject expression:

113 The tree is falling into the river

which we analyse as:

114 neutral action Lgoal

patient inchoative
locative

and corresponding to this we find the [causative] sentence:

115 John felled the tree

agent action neutral
directed patient
causative

These two structures lead us to a large class of locative verbs

which can be called verbs of ‘motion’. They can be illustrated by:

116 The pebble rolled to the bottom of the hill

117 neutral action Lgoal

patient motion

118 John rolled the pebble to the bottom of the hill

119 agent action neutral Lgoal

causative patient
motion

Once again we note a non-causative:causative relationship. In such

12 Reflexive structures are those where the agent of the verb carries out the
action of the verb upon himself. This is overtly shown in sentences like John hurt
himself, John killed himself, etc., where the constituent himself is described as a
reflexive pronoun. There are a few verbs in English that permit the reflexive pro¬
noun to be deleted, and yet sentences with these verbs are understood reflexively:
John shaved, John washed, etc. In this latter sense we describe sentences like John
sat down as reflexive. In English such verbs do sometimes appear as overt reflex¬
ives: thus we find Sit yourself down, The man came into the room and sat himself
comfortably in an armchair. We may also observe that in some languages such verbs
are often syntactically ‘reflexive verbs’ cf. French, se raser ‘to have a shave’,
S’asseoir ‘to sit down’.
 Processes and participants 313

sentences there is also the possibility of not only an [Lgoal] but also
an [Lsource]; and the locative participant is not obligatory.

120a John pushed the car from his house to his garage
120b John swam from Dover to Cap Gris Nez

A selection of such propositional frames and the verbs which can

enter them is shown in Figure 104. We have not discussed all of
the structures shown in this figure, but they are self-explanatory.

1 A Ac 1(LP) 2 A Ac (LG)
l(LG)(LS)

JUMP DESCEND
WALK ENTER
RUN DISAPPEAR
SKIP
LEAP

3 A Ac (LS) 4 N Ac LP

DEPART LIVE
LEAVE STAY
EMERGE

5 N/P . Ac (LG)(LS) 6 A Ac N/P (LG)(LS)

* *

ROLL ROLL
SLIDE SLIDE
FALL DROP
DROP PUSH
DRAG
SHOVE

Examples

1 John jumped on the table/onto the table/off the table

2 He descended into Hell
3 The butterfly emerged from its chrysalis
4 I live in Edinburgh
5 The apple fell from the tree onto Newton’s head
6 The policeman dragged the prisoner from the cell into the dock.

Figure 104
314 Part three: Functional relations

Three comments may be made. Firstly, many verbs of motion may

take either an [Lgoal] or an [Lsource], or both, or neither. This is
represented by the linked brackets in Figure 104. It is also
appropriate, for many verbs of motion, to postulate a participant
[Lpath] to indicate the path followed by some other participant in
the sentence, when this is clearly neither [Lsource] or [Lgoal].
Thus for example:

121a The ball rolled across the room

121b John went from London (LS) to Edinburgh (LG) via
Leeds (LP)

Prepositions appropriate to [Lpath] include across, through, via,

etc. We do not pursue this suggestion further here. This range of
choice is not characteristically available for verbs of possession - as
Figure 102 makes clear, and as the reader is invited to verify.
Secondly, note that some verbs of motion are analysed as
occurring either with an [Lgoal] or an [Lsource] or with an
[Lplace] expression. The latter may be illustrated in sentences like:

122a John is walking in the park

122b John is jumping on the table

Both of these have an interpretation where the motion is being

carried out in some place. The second sentence has another
interpretation more unambiguously expressed by:

123 John is jumping onto the table

where the locative expression is clearly an [Lgoal].13

13 jump can be used both non-causatively, as in the example, or causatively as

in:

John jumped the horse over the fence

This sentence, however, poses an analytic problem. John may quite appropriately
be analysed as realizing an [agent] role. The horse may also appropriately be
analysed as realizing a [patient] role (the object, etc. affected by the action). In this
case, since it is the horse that does the jumping, it should perhaps also be con¬
sidered an [actor]. This suggests we should distinguish between the [agent] (the
instigator or initiator of the action) and the [actor] (the performer of the action).
It might further be observed that for many of the verbs of motion shown in Figure
104, a description as [factor] rather than as [agent] may seem appropriate. The
problem with jump may be more appropriately solved by considering it to realize
an embedded proposition.
 Processes and participants 315

This leads us to a final comment on Figure 104: it is possible to

sub-categorize the prepositions that occur in these sentences as to
whether they relate to [Lplace], [Lgoal] or [Lsource]. Thus,
typically in, on, near, beside are [Lplace] prepositions; in (to),
on(to), to, towards, are [Lgoal] prepositions. Some prepositions,
like under, over, by, may be either [Lplace] or [Lgoal]
prepositions.
We now turn again to [possessive] processes. A few verbs may
be analysed as realizing an [inchoative] process. These include
those shown in:

124a John has acquired a new PhD student

124b John obtained a distinction (in his exam)

125 dative action neutral

inchoative
possessive

providing they are understood in a non-agentive sense as ‘come

to have’, not in an agentive sense as ‘actively sought and got’. In
the latter sense we analyse them as occurring in a proposition
with an [agent] subject. One further possessive structure is
illustrated by:

126 John collected the books for Mary

127 agent action neutral benefactive

possessive patient

Note that in such structures the [agent] is not analysed as also

being either [source] or [goal] (as the structures in Figure 101 were)
and a new participant role, [benefactive], is introduced. This role
appears to differ from the role [goal] introduced earlier: it may
for example be paraphrased by on behalf of , an option not open
to goal expressions.
The last set of sentences we will concern ourselves with in this
chapter can be illustrated by:

128a John is the chairman

128b The chairman is John

129a The committee elected John chairman

129b The committee elected John as the chairman
129c The committee elected John to be the chairman
316 Part three: Functional relations

Our sentences are once more related as [state] to [action]. The first
two sentences we may describe as involving [identification]: John
is identified as the person who occupies a certain position, or plays
a certain role, that of chairman. The second set of sentences all
describe an action that can cause such a state.
Let us look first at the sentences 128: it is suggested that these
are [identification] sentences. In order to discuss some of the
peculiar characteristics of such sentences we can best compare
them with sentences like 2 (the knife is sharp) where we were
dealing with the attribution of a quality to an object; in the
particular case of 2 we were concerned with the attribution of
‘sharpness’ to ‘the knife’. Here we are concerned with the
identification of ‘John’ as filling the particular role of ‘chairman’.
The distinction between [attribute] and [identification] can be
seen more clearly in a slightly different example:

130 My wife is tall and beautiful

131 My wife is that tall and beautiful woman over there

130 describes my wife; 131 identifies her. Note that 131, but not
130, reasonably answers the questions:

132 Who is your wife?

Which one is your wife?

Questions like 132 are asking for an identification, not a

description.
[Identification] sentences have certain syntactic peculiarities.
We concern ourselves with two of them. Firstly, both NPs are
characteristically definite: they are usually either proper names,
like John, or are NPs with the definite article, like the chairman.
Definite NPs usually refer uniquely to some individual in the
context of utterance. With descriptive sentences, the subject NP
is often definite, but the other constituent is usually either an
adjective or, if it is an NP, usually does not contain the definite
article:

133 My wife is a good cook, a teacher, etc.

My wife is tall and beautiful, etc.

In these examples, the adjectives and non-definite NPs do not

refer, they rather describe. Thus, for example, in 128 both John
and the chairman are NPs that could be used to refer; whereas in
 Processes and participants 317

133 the NP my wife can be used to refer to a unique individual,

but a good cook cannot.
The second characteristic is that the NPs in identification
sentences are reversible round the verb. Thus, for example, in
answer to the question:

134 Which one is John?

we might find:

135 John is the CHAlRman

The CHAlRman is John

The capitals indicate that CHAlRman is the word that appropriately

in this context receives the sentence stress - it is the centre of the
intonation contour. Note that whichever word order is used, the
stress centres on CHAlRman - the NP that answers the identifying
question. This particular characteristic of such sentences com¬
plicates our description, since the question can be phrased the other
way:

136 Which one is the chairman?

137 JOHN is the chairman

The chairman is JOHN

where the intonation placement is exactly reversed. We clearly

need to know what is intended by a particular sentence before we
can properly assign a propositional structure to it: thus we say
that with respect to the sequence 134, 135 chairman is the NP
that is identified with the function we describe as [role]; whereas
in 137, it is clearly John that has this function. The other NP
we describe as realizing a [neutral] function.
What now of the sentences in 129? Here the committee can be
assigned the role [agent] and John the role [neutral, patient].
Chairman we describe as [role] as before. These characterizations
are summarized as:

138 neutral state role (=128)

identity

139 agent action neutral role (=129)

causative patient
directed
identification
318 Part three: Functional relations
Sentences like 129 are particularly good candidates for a
description in terms of an embedded predication, and this is
reflected in the paraphrases involving as, to be.
If there is a relationship between propositional frames like 138
and 139, there is a comparable relation between pairs of
sentences like:

140 John is a fool

141a I consider John a fool
141b I consider John to be a fool
141c I consider John as a fool

Our propositional description here is of the form:

142 neutral state attribute

description
143 agent action neutral attribute
description

The proposition structure 142 is not characterized as [causative]

since there is clearly no causative relation between sentences like
135 and 141: just because ‘I consider John to be a fool’, it does
not mean that ‘he is one’! This structure, like that in 139, also
lends itself to an analysis in terms of an embedded predication.
We may also observe that the [role] and [attribute] participants
in 139 and 143 may be characterized in terms of the ‘state’ and
‘result’ distinction we observed in footnote 4 on page 298. The
[causative] sentences have a ‘result’ implication, and the
non-causative sentences a ‘state’ interpretation. Some other verbs
that can enter these frames are shown in Figure 105.

A Ac N/P Ro A Ac N Att

ELECT CONSIDER

NOMINATE THINK

CHOOSE ESTEEM

APPOINT

Figure 105 MAKE

We have obviously described only a part of the propositional

structure of English - some exercises at the end of the chapter
 Processes and participants 319
invite the reader to explore some of the many other areas that
lend themselves to a description of this sort - but the part we
have described includes some of the most important and
fundamental relationships among propositions that are to be
found in English. Relationships of this sort seem to be found in
many languages. They are summarized in outline in Figure 106.

state action

inchoative causative

N V Att (=10) N/PV (=40) AVN/P (=9)

descriptive
N V Att (=10) — A V N Att (=143)

possessive D VN (=77) DVN (=125) AVN/PG (=78)

locative N V LP (=91) N/PV LG (=114) A V N/P LG (=92)

identifying N V Ro (=138) — AVN/P Ro (=139)

The bracketed figures refer to the specifications of propositional

structures in the text

Examples

St N V Att The fruit is ripe

Inch N/P V The fruit is ripening
Caus A V N/P The sun is ripening the fruit
St N V Att John is a fool
Caus A V N Att I consider John a fool

St D V N John has a car

Inch D V N John has acquired a car
Caus A V N/P LG Mary gave a car to John

St N V LP The pencil is on the floor

Inch N/P V LG The pencil is falling onto the floor
Caus A V N/P LG John put the pencil on the floor

St N V Ro John is the chairman

Caus A V N/P Ro They elected John chairman

Figure 106

The sentence types defined in this figure form a matrix where

sentences which can be categorized as [descriptive], [possessive],
[locative] or [identifying] are cross-categorized in terms of the
characterizations [state], [inchoative] and [causative]. Certain
broad generalizations can be made. The [state] column involves
320 Part three: Functional relations

[state] verbs, the other two columns [action] verbs with the syntactic
consequences described earlier in the chapter. The [state] and
[inchoative] columns involve one or two participants, neither of
which is [agent]; the [causative] column always introduces a third
participant, which is always [agent]; this too has syntactic
consequences. We can also make some generalizations
about the order of constituents in active declarative sentences.
The verb is always second constituent. There is a hierarchy
among the roles in terms of their ability to occur as subject,
object, etc. The hierarchy is: A:D:N: etc. (where etc. represents
the other roles). Thus, if there is an A it is subject; if there is no
A, then D is subject; otherwise N . . . etc. Similarly in sentences
with an A subject, N takes precedence over other roles as object.
Stated like this, the hierarchy is too crude (and subject to a
number of exceptions), but the notion of a hierarchy seems evi¬
dent, and is also reflected in other languages that can be analysed
in similar terms.
The symmetry of the figure is also somewhat deceptive, and
only seems to apply systematically in the cases noted, and in a
few others. Many propositions do not lend themselves to a com¬
parably neat classification, for example the verbs of motion dis¬
cussed in 117. These do not seem to correlate, in any very straight¬
forward way, with either a [state] or a non-embedded [causative]
proposition. In some cases we can postulate correlations like:

144a He is coming (active)

144b He is here (stative)

145a He is going (active)

145b He is gone (stative)14

But this does not seem to be true also of sentences like:

146 He is running

which have no [state] counterparts. Similarly, with causatives we

can certainly construct [causative] sentences like:

14 Many English speakers make no distinction between He has gone (resultant

state) and He is gone (state): see footnote on page 298. This is perhaps because in
speech they would be homophonous He’s gone. The speech form He’s come is
usually rendered in writing as He has come, although He is come is sometimes also
found. How far this can be carried is unclear - which of the following are accept¬
able? He is arrived, He is departed. He is disappeared. He is appeared.
 Processes and participants 321
147 John made me run
but this overtly involves an embedded predication, and such a
causative can be formed on any sentence with an [agent]:
148 John made me put a lamp on the table
149 Sir Harold made the committee elect his friend Charles
as chairman

In terms of a single predication, verbs of motion do not appear to

correlate with either [state] or [causative] propositions, without
embedding. Further examples can easily be found.
Figure 106 has one further important feature. An approach of
this sort leads to a classification of verbs in terms of the propo¬
sitional structures they may enter. This suggests that we can use
such a classification in lexical entries. Thus, for example, we can
enter in the lexicon information like:

150 STAND: St., N LP

Ac, A N/P LG
put: Ac, A N/P LG
TAKE: Ac, A N/P LG
Ac, A N/P LS
REMOVE: Ac, A N/P LS

(the characterizations are taken from Figure 102). In these entries

we have placed first the characterizations [state], [action], and
followed them by a specification of the participant roles involved
(we noted previously — page 292 — that the statement of elements
in propositional structures was abstract, and, in principle, unor¬
dered). Characterizations like those in 150 bear an intimate rela¬
tionship to strict sub-categorization frames discussed on pages
55-7. The difference is that the strict sub-categorization frames
were there stated in purely categorical terms, i.e. in terms of
categories like Adj, NP, PP, etc., whereas these categorizations
are stated in functional terms, [agent] [goal], etc. We can com¬
bine the two types of statement, perhaps like:
151 STAND: V, St;_PP; N LP
V, Ac;,_NP PP; A N/P LG
where the first characterization shows categorial information
(stand is a stative verb), the second strict subcategorization
information (STAND occurs in the environment _PP), and the
322 Part three: Functional relations

third indicates the functional role of the non-verbal constituents

(the subject of stand is [neutral] and the PP indicates a [loca¬
tive place]).
Either of these ways of characterizing verbs enable us to relate
propositional information to constituent structure, and to make
use of it to control such things as the applicability of transforma¬
tions.
We briefly mention two ways in which this could be done. One
is associated with a school of grammar that has come to be
known as ‘Case grammar’. This school of grammar suggests that
the ‘most basic’ representation of a sentence should be not in
‘categorial’ terms (i.e. using categories like NP, VP, V etc.), as in
earlier chapters of this book, but ‘functional’ terms. Thus:

In the basic structure of sentences . . . we find what might be called the

‘proposition’, a tenseless set of relationships involving verbs and nouns
. . . separated from what might be called the ‘modality’ constituent. This
latter will include such modalities on the sentence-as-a-whole as
negation, mood, and aspect. . . . The first base rule, then, is . . .

152 Sentence —> Modality + Proposition

(Fillmore 1968, 24)

The proposition constituent is further expanded as a verb

together with a number of nodes specified in terms of such labels
as [agent], [instrument], etc.15

Each functionally labelled node is then developed as Preposition

+ NP - we have already seen the typical association between
prepositions and roles and this formalizes it. Such rules would
develop structures like those in Figures 107-9. The grammar
would then contain rules of the following sort:

Every English sentence has a surface subject, if only formally so. . . .

In general the ‘unmarked’ subject choice seems to follow the following
rule: If there is an [agent] it becomes subject; otherwise, if there is an
[instrument], it becomes subject; otherwise the subject is [patient]. . . .
The normal choice of subject for sentences containing an [agent] ... is
the [agent], (Fillmore 1968, 33)

15 For Fillmore the specification of the participant roles, which he refers to as

‘cases’, is somewhat different from that we have adopted, although the basic idea
is very much the same. In order not to confuse the reader we have adapted
Fillmore’s scheme to our terminology, and silently amended the quotations.
 Processes and participants 323

M P

V agent patient

A A
Prep NP Prep NP

past OPEN 0 the door past OPEN by John 0 the door

Figure 107 Figure 108

M p

past

Figure 109

This rule is clearly in line with our previous remarks about a

hierarchy of roles (page 320). This approach can deal with such
things as the active:passive relationship, by deriving passive sen¬
tences directly from a propositional structure rather than by a
transformational rule. The following is proposed: A verb like
OPEN permits [patient] to occur as subject

as long as this ‘non-normal’ choice is ‘registered’ in the V. This ‘register¬

ing’ of a ‘non-normal’ subject takes place via the association of the
feature [ +passive] with the V. (Fillmore 1968, 37)

The effect of these informally stated rules is shown in Figures

110-12. Note that a further rule is required deleting preposi¬
tions from roles that occur in subject position (cf. Figure 111).
There are many problems about formulating a grammar in this
324 Part three: Functional relations
way, but we do not explore them further here (but see the discus¬
sion in Brown and Miller, in press). The only point at issue is
that there are those who would wish to exploit functional
information in the grammar of a language.

S instrument M P

Prep NP V patient

/\
0 the door past OPEN
IA Prep

I
with a key past OPEN 0 the door
NP

(the door opened) (a key opened the door)

Figure 110 Figure 111

patient

/\
Prep NP

Figure 112 (the door was opened by John)

Another way in which functional information can be exploited

is as follows. We still use a constituent structure grammar formu¬
lated as in earlier chapters, and use an appropriately modified
lexical insertion rule to mark the various categorial nodes for the
relevant information. Thus the grammar generates trees such as
those shown in outline in Figures 113 and 114. The first lexical entry
for stand, shown in 151 lexicalizes the tree in Figure 113, and
 Processes and participants 325
the second entry for STAND lexicalizes Figure 114. We have
shown this in the tree, and have annotated the nodes with the
relevant functional information.

the lamp pres STAND on the table

Figure 113

Figure 114

Application of the relevant realization rules produces:

153a The lamp is standing on the table

153b John stood the lamp on the table

The functional information can now be used to control transfor¬

mations. Suppose we annotate the passive transformation along
the following lines:

154 SA: NPl - Aux - V - NP2

SC: NP2 - Aux + PASS - V - by + NP,
Restriction: NP2 must be [neutral, patient] or [neutral,
result] V must be [action]
326 Part three: Functional relations
The SA is met by the configuration in Figure 114, and it does not
offend any of the restrictions, so we may derive the passive sen¬
tence:

155 The lamp was stood on the table by John.

Figure 113 does not meet the S A. In the same spirit we can derive
and block the following active and passive pairs:

156a The sun ripened the fruit

156b The fruit was ripened by the sun

The fruit realizes [neutral, patient], RIPEN is an [action] verb, so

the passive can be derived.

157a John built this house

157b This house was built by John

This house realizes [neutral result]; BUILD is an [action] verb, so

the passive can be derived.

158a John has a book

158b *A book is had by John

A book is not [neutral patient] or [neutral result]; have is

[state] so the passive cannot be formed.

159a John is playing tiddlywinks

159b *Tiddlywinks is being played by John

Tiddlywinks is a [range] participant, so although PLAY is [action]

the passive cannot be formed. The reader is invited to try other
derivations.
We mention one final transformation not discussed previously
- the imperative, which is responsible for deriving sentences like:

160a Put the book on the table!

160b Build a house!
160c Play tiddlywinks!

Imperative sentences can only be formed from propositional

structures that contain an [action] verb and an [agent] participant.
These two facts have to be stated in restrictions on the formation
of the imperative to account for the acceptability of sentences
like those in 160 and the unacceptability of sentences like:
161a *Have a book (where HAVE is [state])
 Processes and participants 327
161b *Know Chinese (where know is [state])
161c * Ripen (oh, fruit) (where ripen is [action], but fruit,
does not realize an [agent] partici¬
pant)
161d *Cost six pounds (where book does not realize an
(oh, book) [agent] participant)

Such restrictions are difficult to formulate in straightforward

categorial terms; we also need access to functional information,
however this is captured in the grammar.

To close, we briefly consider two questions. How redundant

are the specifications? How many propositional roles, etc. are we
to recognize? As to the first question, it is clear that there is a
certain amount of redundancy: for example, all [causative]
structures involve an [agent]. The redundancy can be removed,
but we prefer to keep it, partly for the sake of clarity, and partly
because role features cannot always be predicted on the basis of
process types, or vice versa-, and until a full classification is
developed in these terms it is unclear exactly what redundancy
can be removed. The second question is more difficult. It seems
the case that we can choose either to reduce the number of roles,
thus achieving an apparently ‘simpler’ system, or to multiply the
number of roles, thus achieving a descriptively more accurate
system. For example, we can collapse together the ‘locative’ roles
([Lplace], [Lgoal] and [Lsource]) and the ‘dative’ roles [dative],
[goal] and [source]. A justification for this is their very close
relation, both semantically and grammatically. Typically we chose
a ‘dative’ characterization when an animate is involved, otherwise
a ‘locative’ characterization. Furthermore, it seems that we do
not wish any nuclear proposition to contain both a ‘dative’ and a
‘locative’ participant (a dative proposition may, of course, be
associated with a locative circumstantial role - John gave Mary the
money in the park - but that is a different matter). On the other
hand there are grammatical differences between the two types of
proposition, and they define different classes of verb (e.g.
‘locative’ and ‘verbs of motion’, and ‘dative’ and ‘verbs of giving
and receiving’). A case where we might choose to multiply roles is
[agent], which as we have observed involves a number of notions
that might be stranded out - ‘initiator’, ‘actor’, ‘cause’, etc. As so
often in linguistic descriptions, some balance must be sought
328 Part three: Functional relations

between maximum generality (the ‘reductionist’ approach, which

leads to numerous instances requiring some special treatment -
as with our [agent] role) and maximum explicitness (which leads to
a proliferation of roles and might eventually lead to identifying a
unique structure for every use of every verb, since presumably all
verbs differ, however slightly, in their grammatical and semantic
behaviour). We justify our description by its offering a degree
of both generality and particularity. This conclusion is un¬
comfortable, since it has no easily defended validity, but, there
seems to be no alternative in the current state of knowledge. We
can only take refuge in the fact that it is a position shared by anyone
who attempts a description along these lines, and it seems to reflect
the somewhat indeterminate nature of this aspect of language.

Technical terms
analytic; circumstantial roles; incorporation; proposition;
propositional nucleus; propositional structure.

Process terms
descriptive, possessive, locative, identification, state, action,
inchoative, directed, causative

Participant role terms

agent
neutral, patient, result, attribute, role
dative goal, source
locative place, locative goal, locative source, locative path
range, instrument, benefactive

Exercise

Consider the following ‘roles’ A(gent); I(nstrument); L(ocation);

Dir(ection); S(ource); G(oal); D(ative); N(eutral) illustrated in
the following sentences:

1 John (A) beat the dog (N) with a stick (I)

2 Mary (D) knows Chinese (N)
3 Charlie (A/S) gave the book (N) to Mary (G)
4 Mary (A/G) took the book (N) from Charlie (S)
5 The cow (A) jumped over the moon (Dir)
 Processes and participants 329
We can represent ‘role structures’ for particular verbs in the
following way:

beat: AVbN(I) (the man beat the dog with a stick)

open: N Vb (the door opened)
A Vb N (the man opened the door)
give: A/S Vb N to-G (John gave the book to Mary)
A/S Vb G N (John gave Mary the book)
rob: A/G Vb S of-N (John robbed Mary of £10)
steal: A/G Vb N from-S (John stole £10 from Mary)

The formulae are intended to indicate the particular role structure

available to a given verb when the verb is active (i.e. it does not
take account of passive structures). So, for example, BEAT, in
active sentences, requires both an A(gent) and a N(eutral) noun
(The man beat the dog) but cannot occur with only an A (The man
beat) or only an N (*The dog beat): it may optionally occur
with an I(nstrument). With some verbs, like GIVE, there are two
alternative active sentences, as shown in the examples; with other
verbs, like ROB, there is only one possible order of constituents -
there is no sentence John robbed £10 of Mary or the like.
Note that in the examples given, a passive can be formed with
the NP immediately following the verb as subject: Mary was
robbed of £10 by John.
In the groups of verbs given below:

(i) Write a statement, or statements, of possible role structures for

each verb, as in the examples above. Note the prepositions that
occur.

(ii) Start by considering only active sentences - can passive

sentences always be formed from NPs that can become ‘object’ of
the verb (and immediately follow it without a preposition)?

1 ROB, STEAL, PLUNDER, LOOT, RANSACK

2 SEND, TAKE, BRING, FETCH, CARRY
3 SUPPLY, ISSUE, FURNISH, DELIVER
4 GIVE, PRESENT, TAKE, OFFER, ACCEPT, RECEIVE
5 KNOW, TEACH, LEARN, INSTRUCT
6 TELL, ANNOUNCE, INFORM, MENTION, REPORT,
NOTIFY
7 SMEAR, PAINT, FESTOON, SPREAD, ENGRAVE,
MARK, SPATTER, SPRAY
19 Subjects, objects, complements
and adjuncts

The traditional apparatus of grammatical description includes the

set of terms: subject, object, oblique object, indirect object,
complement and adjunct. Their definitions are typically an
amalgam of features deriving from their grammatical character¬
istics, from syntactico-semantic characterizations as discussed in
the preceding chapter, and from characteristics of their behaviour
in texts. Criteria of the first two sorts, grammatical and
syntactico-semantic, are chiefly involved in this chapter. We can¬
not entirely avoid textual criteria, but largely leave these until the
next chapter. We shall be concerned to establish whether any or
all of these notions has a special role in a grammatical descrip¬
tion. The discussion concerns itself with English, since the gram¬
matical characteristics of these notions vary from language to
language: the general principles involved, however, do seem to
be widely applicable.

Subject
A distinction is frequently drawn between the ‘grammatical’ sub¬
ject (characterized by grammatical considerations, such as being
the controller of number agreement), the ‘logical’ subject
(characterized by syntactico-semantic considerations - in sen¬
tences with an [agent] NP this is usually held to be the logical
subject) and the ‘thematic’ or ‘psychological’ subject (character¬
ized by textual considerations - ‘this is what the sentence is
about’). These can be illustrated in the sentences la, b, c where
G,L and T indicate respectively the grammatical, logical and
thematic subjects:

la John (G,L,T) took the largest kitten

lb The largest kitten (G,T) was taken by John (L)
lc The largest kitten (T), we (G,L) gave away
 Subjects, objects, complements and adjuncts 331

As can be seen the three characterizations do not necessarily

coincide. We defer consideration of the thematic subject until the
next chapter, and concentrate here on the notions of grammatical
and logical subject.
We first consider the grammatical subject. All full declarative
sentences must have a subject. This is an NP. With few excep¬
tions the grammatical subject immediately precedes the main
verb and is in number concord with it. Grammatical subjects are
never marked with a preposition. NPs that fulfill these character¬
istics we call grammatical subjects. As we have seen before, for
some actual sentence the structural configuration ‘NP that is
immediately dominated by S’ typically identifies the grammatical
subject.
The subject is frequently a ‘full’ lexical NP, like John or the
■largest kitten in la, lb, or a pronoun, like we in lc, but it may also
be a nominalized sentence or sentence-like constituent as in:

2a That Edinburgh’s New Town is magnificent is undeni¬

able
2b For you to run off with Mary would be madness

Sometimes when no such constituent is available to act as subject

a ‘dummy’ subject is supplied: this is the case with ‘weather’
expressions:1

3 It is raining

and in cases like 4b where a nominalized sentence is ‘extraposed’

(see pages 372-3):

4a That Edinburgh’s New Town is magnificent is

undeniable
4b It is undeniable that Edinburgh’s New Town is
magnificent

The main verb concords in the singular with the dummy subject
it. It does not, however, have the full range of syntactic possi¬
bilities of a full subject - questioning is impossible; no sense is
made by asking:

1 In terms of the terminology introduced in the preceding chapter, we might

refer to the proposition which 3 realizes as a ‘no place’ proposition. In traditional
grammars they are often realized by what are referred to as ‘impersonal verbs’. It
may be observed that not all languages permit no place propositions: in Akan the
sense of 3 would be realized as nsuo to ‘water falls’, ‘it is raining’.
332 Part three: Functional relations

5a *What is raining?
5b *What is undeniable that Edinburgh’s New Town is
magnificent?

Another item that operates like a dummy subject is there in

sentences like:

6a There is a glass on the mantelpiece

6b There are glasses in the drinks cupboard

Sentences like these are called ‘existential’ because they are typi¬
cally used to assert the existence of something. We call this use of
there the ‘existential’ there, distinct from another use of there in
sentences like:

7 There is the glass

which we call the ‘deictic’ there because sentences like 7 can be

used to point to something. Existential there has several curious
features. It does not control number agreement on the verb, as
can be seen in 6a and 6b. Unlike the deictic there, it cannot be
stressed:

8a There is the glass (deictic)

8b * There are glasses in the drinks cupboard
(existential)

Connected with this is the fact that it is almost invariably

pronounced in a ‘reduced’ form: / 5oz/ ‘there is’, / Sara/ ‘there are’:
the deictic there is usually non-reduced: / Seariz / ‘there is’ / Seara /
‘there are’. Thus / Saz a glas / ‘There is a glass’, with the reduced
form, is understood as existential and not deictic. Finally we note
that existential there is always in subject position:

9a There are glasses in the drinks cupboard

9b * Glasses in the drinks cupboard are there

(It may, of course, occur as subject of an embedded sentence: It

appears that there are a number of problems.) By contrast, deictic
there need not necessarily be subject:

10a There is the glass

10b The glass is there

Existential there, unlike deictic there cannot be questioned:

11a Where is the glass? (deictic)

 Subjects, objects, complements and adjuncts 333

lib *Where are the glasses in the drinks cupboard?

(* existential)

Finally, existential there is typically restricted to sentences with an

indefinite NP:

12a The glass is on the table

12b ? A glass is on the table
12c ?There’s the glass on the table
12d There is a glass on the table

(12c is to be understood as the reduced, existential there: it is

acceptable as the stressed deictic there.2)
Some arguments that tend to the view that subject has a special
status in the grammar of English can be derived from a
transformational approach to description. In this approach we
distinguish an underlying from a surface level of description (see
pages 123-5). Suppose that the sentence:

2 The total distribution of existential there is 'fuzzy'. It occurs with the locative
copula BE, and also with verbs like:

There appears to be some difficulty about the tickets

There seem to be some people who still believe the earth is flat
There should be an answer to that question
There happens to be a party this evening
There still exist some problems in the analysis of there

In all these examples the NP immediately following the verb is indefinite. If we

replace this by a definite NP, then the sentences are often of dubious acceptability:

? There appears to be the difficulty about the tickets

Nor is it exactly clear how far this construction can be extended. The following
seem only marginally acceptable with the existential there:

? There hung a picture on the wall

? There lay a man in the gutter

Another item that operates rather like existential there in some constructions is
here'.

Here comes the procession

Here’s a pretty kettle of fish

Like existential there, this item is usually unstressed and reduced phonologically. It
needs to be distinguished from a ‘locative’ here:

Here lies John Smith

The procession comes past here at ten o’clock
334 Part three: Functional relations
13 Everyone believes that Charlie is handsome

is derived from an underlying structure which we represent as:

14 Everyone believes (Charlie is handsome)

Suppose too that the sentence:

15 Everyone believes Charlie to be handsome

is an alternative realization of this underlying structure. We now

observe that whereas in 13 Charlie is clearly the subject of the verb
is in the embedded sentence, in 15 Charlie appears to operate
syntactically like the object of the verb believe in the matrix sen¬
tence. Three pieces of evidence seem to support this observation.
First, if we replace Charlie by a pronoun, then we must have
he, the subject form, in 13, but must have him, the object form
in 15:

16a Everyone believes that he (*him) is handsome

16b Everyone believes him (*he) to be handsome

Secondly, consider the following sentences with reflexive pro¬

nouns:

17a Charlie, believes that he, (*himself) is handsome

17b Charlie, believes himself, to be handsome
17c Charlie, believes him2 to be handsome

We have followed here the convention of marking subscripts on

co-referential items (pages 274-7). The distribution of the reflex¬
ive and non-reflexive pronouns here strongly suggests that him¬
self is the object of believe. Thirdly, we note that the passive
sentence:

18 Charlie is believed by everyone to be handsome

is well formed. The passive is largely restricted to objects - and this

suggests that Charlie can be considered to have been at some stage
the object of believe. There is, of course, no passive corresponding
to 13:

19 * Charlie is believed by everyone that is handsome

If we accept that such sentences are related to the structure

shown in 14, then it seems that this structure has been ‘re-analysed’
along the lines shown in 20:
 Subjects, objects, complements and adjuncts 335
20a Everyone believes (Charlie is handsome)
20b Everyone believes Charlie (to be handsome)

Transformational grammarians postulate a rule of ‘subject raising’

to account for this.3 The rule only applies to subjects, as its name
implies. Consider the following sentences:

21a Everyone believes (Charlie has killed Mary)

21b Everyone believes Charlie (to have killed Mary)
21c Charlie is believed by everyone (to have killed Mary)
22a Everyone believes (Charlie has killed Mary)
22b Everyone believes (Mary has been killed by Charlie)
(Passive in embedded S)
22c Everyone believes Mary (to have been killed by
Charlie)
22d Mary is believed by everyone (to have been killed by
Charlie)

Note that Mary, the object of KILL in 22a cannot be raised:

23 *Everyone believes Mary (Charlie to have killed)

but once it has become the subject of the embedded sentence by

passivization, then it can be raised - yielding 22c, and subsequently
22d.
Returning briefly to there, we note that this item also operates
like a subject for these purposes:

24a Everyone believes (there is a solution to this problem)

24b Everyone believes there (to be a solution to this
problem)
24c There is believed (to be a solution to this problem)

The argument suggests that the grammatical subject has an

important role. Another argument offered by a transformational
approach can be illustrated by the following sentence:

25 John, expected that the committee would choose him,/

him.

The subscripts indicate two possible readings, one in which him

3 As we have remarked before, this volume does not attempt to present a full
account of the formal machinery involved in a transformational grammar. For a full
discussion of this transformation the reader is referred to one of the books listed on
pages 386-7.
336 Part three: Functional relations
cross-refers to John, and another where him refers deictically to
some other individual in the context (see pages 276-7). Let us
suppose that 25 is related to the underlying structure:

26 John expected (the committee would choose him)

after the fashion already established in discussing 13-15. In the

same spirit as before we consider the sentence:

27 John, expected the committee to choose him,/him2

is another way of describing the same situation. As before,

him can have two interpretations. Let us now passivize the
embedded sentence:

28 John expected (he would be chosen by the committee)

which is realized as:

29 John, expected that he,/he2 would be chosen by the

committee

then he has two readings as in 25. However, if we passivize the

embedded sentence in the structure corresponding to 27 we get:

30 John, expected him2 (*him,) to be chosen by the

committee

him can only have one reading, with deictic reference. It cannot
cross-refer to the subject of expected. But the sentence:

31 John expected to be chosen by the committee

has the other reading: i.e. John is understood not only as the
subject of expected but also as the subject of be chosen. There is,
however, no overt subject present. For the purposes of discussion
we represent this situation like this:

32 John, expected 0, to be chosen by the committee

Once again we see that the subject, in this case an ‘understood’

subject, has a special role.4 This relationship is quite general:

4 The phenomena briefly outlined here are treated by transformational

grammarians as the result of another transformational operation known as ‘Equi
NP Deletion’ or ‘Co-referential subject deletion’. Once again the reader is referred
to Further reading, pages 386-7. This transformation is believed by many to be
somewhat dubious, but the phenomenon it is intended to account for lends some
further support to the status of ‘subject’, which is our particular concern.
 Subjects, objects, complements and adjuncts 337

33a I, expect 0, to go to America this summer

33b Gill, wants 0, to come with me
33c Gill, expects her2 (*her,) to go with me

Our concern is not with the machinery of the putative

transformational rules involved, but with the observation that the
grammatical subject seems from the discussion to have a special
status in English grammar.
So far all our discussion has involved declarative affirmative
sentences. Let us briefly consider interrogative sentences. These
can be illustrated by the b sentences here:

34a Katie is writing a French essay

34b Is Katie writing a French essay?
35a Sarah plays the trombone
35b Does Sarah play the trombone?

We have discussed the formation rules for interrogative sentences

on page 127. In outline the rule is: invert the first auxiliary verb
(is in 34) and the grammatical subject; if there is no auxiliary
supply an appropriate form of DO, as in 35. Note that after the
operation of this rule, the subject NP still precedes the main verb,
and it still controls number agreement on the auxiliary, or DO.
Subject-auxiliary inversion is also typical of certain sentences
involving certain ‘negative adverbs’:

36a Never have I heard such nonsense

36b Seldom have I tasted a more magnificent claret

Our final observation on the grammatical subject is that there

are a few constructions in which it follows rather than precedes the
main verb. Thus we find:

37 Down comes the flag, off goes the gun and away go the
boats

Such sentences have a distinctly literary flavour, and we consider

them to be a stylistic device, discussed further in the next chapter.
Note that number concord between subject and verb is preserved
even though the subject follows rather than precedes the verb.
The arguments strongly suggest that, for English at least, the
grammatical relation of subject has a special role in grammatical
description.
We now turn to the notion of logical subject. This, as we
338 Part three: Functional relations
observed on page 330, usually relates to the question of the
participant role a subject realizes. As the preceding chapter made
clear, the grammatical subject can realize any of a number of such
roles. Perhaps the most typical role for the subject is agentive:

38a John beat the dog

38b William invaded England in 1066

Instrumental subjects can be seen in:

39a The axe smashed the door

39b A brush could clear those drains

The fact that such subjects are instrumental is established by the

passive forms of the sentences; where the instrumental preposition
with occurs:

40a The door was smashed with an axe

40b Those drains could be cleared with a brush

Dative subjects are shown in:

41a Harry knows that his wife is unfaithful

41b I am interested in linguistics

and this description is supported by paraphrases with to:

42a That his wife is unfaithful is known to Harry

42b Linguistics is interesting to me

where the dative preposition to occurs.

Goal subjects are to be observed in:
43a Harry received a gold medallion from the Royal Society
43b Charlie obtained a licence from the local authority

and this analysis is supported by the fact that the preposition to

occurs in sentences like:

44a The Royal Society presented a gold medallion to Harry

44b The local authority issued a licence to Charlie

Source subjects are to be seen in 44; and note the paraphrase

with the preposition from in 43.
Place locative subjects occur in sentences like:

45a Edinburgh is cold, wet and windy

45b This box contains fifty-two matches
 Subjects, objects, complements and adjuncts 339

where alternative formulations of the proposition contain a

locative proposition:

46a It is cold, wet and windy in Edinburgh

46b Fifty-two matches are contained in this box

Patient subjects are found in the sentences:

47a The butter is melting

47b The ice is cooling

Finally we note neutral subjects in:

48a Mary is very tired

48b Harry is the Professor of Linguistics

All the participant roles identified in chapter 18 may occur in

some sentence as the grammatical subject. The traditional notion
of logical subject is usually related to sentences involving an
[agent] participant. Thus in:

49 William invaded England in 1066

William, the [agent] participant, is referred to as both the logical

and the grammatical subject. In the corresponding passive
sentence:

50 England was invaded by William in 1066

William remains the logical subject, but England is now the

grammatical subject. The notion of logical subject appears to
relate to the unmarked subject choice in some particular
propositional structure. By analogy the axe and a brush are the
logical subjects in 39; Harry and / in 41 and so on. We have
already noted, page 322, that given the participant roles [agent],
[instrument] and [neutral], the active declarative sentence chooses
the [agent] participant as subject over the [instrument] and the
[neutral], and the [instrument] over the [neutral]: these choices are
reflected in 38 and 39. Similarly, given [dative] and [neutral], the
most unmarked order, all other things being equal, chooses
[dative] as subject, and this is reflected in 41. These facts suggest
that if we introduce the notion of participant roles together with
appropriate statements about unmarked subject choice we can
dispense with the notion of logical subject, since this new
machinery gives a more precise characterization of the notions
involved.
340 Part three: Functional relations

Object
The grammatical object, like the subject, is realized by an NP. In
active declarative sentences with unmarked word order four
grammatical features characterize the object: i it directly follows
the verb; ii it is not in construction with a preposition; iii it can
become the subject of the corresponding passive sentence; and iv
it is an obligatory constituent with transitive verbs.
The most clear-cut cases of objects are those constituents
traditionally referred to as the direct or affected object. These are
NPs that realize what we called (page 290) [neutral, patient]
participant roles in two place propositions where the verb is a verb
of directed action, causative or non-causative, and the other
participant, which must become subject in active sentences is
described as [agent],

51a Samson smote the Philistines

51b The forester split the log

In such constructions no possible paraphrases involve

prepositions, the relevant NPs can become the subject of the
corresponding passive sentences and the direct object is an
obligatory constituent. We do not find:

52a *Samson smote

52b *The forester split

where only an [agent] participant is realized.

Traditional grammars also recognize other types of object in
two-place sentences. The names usually given to them reflect the
type of propositional role that the object realizes. All of them are,
to a greater or lesser extent, unable to meet the four criteria out¬
lined above. Thus, the following sentences involve an ‘object of
result’ (also called an ‘effected’ or ‘factitive’ object):

53a Mary wove that blanket

53b The workmen are digging a hole

Objects of result can typically become the subject of a passive

sentence, and there are no paraphrases involving prepositions.
Many verbs that occur in such sentences can also appear in
one-place sentences with an agent subject but no object:

54a Mary is weaving

54b The workmen are digging
 Subjects, objects, complements and adjuncts 341
Sentences like:

55a Mary sang a song

55b I dreamed a strange dream (last night)

are sometimes described as involving ‘cognate objects’, since the

relevant NP usually contains a noun morphologically derived from
(and hence cognate with) the verb stem. We identified such
participants as [range] participants but noted that, while there are
no paraphrases involving prepositions, they do not always
felicitously become the subjects of passive sentences.
Some grammarians have identified the object in sentences like:

56a Harold is eating his lunch

56b Patricia is reading War and Peace

as an ‘object of concern’: they are clearly neither affected (direct)

or effected (resultant) objects. Such objects do not always appear
as subjects of passive sentences with the same degree of
acceptability as the other objects we have looked at, and, like
objects of result, are often omissible.
The cases examined so far suggest that there is a hierarchy of
‘objecthood’. The prime exemplar is the direct object. This has a
particularly close tie to the main verb; it is an obligatory sentence
constituent; it immediately follows the main verb; it will not occur
in a paraphrase involving a preposition; and it can be the subject of
the corresponding passive sentence. Other objects (of result,
concern, etc.) do not meet one or more of these criteria.
We now consider sentences like:

57a The Russians supplied arms to the Vietnamese

57b The Russians supplied the Vietnamese with arms

In the previous chapter we described these as realizing three place

propositional structures with [agent] (the Russians), [goal] (the
Vietnamese) and [neutral] (arms) propositional roles. The [agent],
as we anticipate, becomes subject of the active sentence. There is,
however, a choice as to whether the [goal] or the [neutral]
participant immediately follows the verb. Whichever is chosen, that
participant is realized as an NP without a preposition, and the
other participant is realized as a prepositional phrase. We refer to
the NP that immediately follows the verb as the object. Note that
it operates syntactically like an object. It cannot occur with a
preposition in this position:
342 Part three: Functional relations
58a *The Russians supplied with arms to the Vietnamese
58b *The Russians supplied to the Vietnamese with arms

It may become the subject of the corresponding passive:

59a Arms were supplied to the Vietnamese by the

Russians
59b The Vietnamese were supplied with arms by the
Russians

And furthermore it is no longer omissible:

60 *The Russians supplied with arms

whereas the PP is omissible:

61a The Russians supplied arms

61b The Russians supplied the Vietnamese

We call the NP in the prepositional phrase an ‘oblique object’,

recognizing that the NP in the prepositional phrase might, as it
were, have become the object, had the other NP not done so. The
oblique object is omissible, as we have observed, and cannot
generally become the subject of a passive sentence:

62 *The Vietnamese were supplied arms to by the

Russians

We furthermore suppose that both sentences realize the same

propositional structure. In chapter 18 we represented this by the
formulation:

63 agent action neutral goal

directed /*
patient

The arrows from [patient] indicate that this role could be

associated with either the [neutral] participant (arms) or the [goal]
participant (the Vietnamese): the patient participant becoming the
object, and the other participant being realized as an oblique
object.
In the case of verbs like SUPPLY neither of the non-agent
participants is, as it were, the ‘designated’ object: one of the
participants may become an Qbject, but it may be either. The effect
of becoming an object is important. The syntactic effect we have
already discussed; but there is also a semantic effect. The semantic
implications vary from verb to verb as we saw in the discussion on
 Subjects, objects, complements and adjuncts 343
pages 309-10. It varies from cases like 57 where there seems to
be little semantic effect to cases with considerable semantic
implications:

64a An archer shot at William with an arrow

64b An archer shot William with an arrow
64c An archer shot an arrow at William

SHOOT may occur with two oblique objects, or with a direct and
an oblique object. The semantic effect may be so great here that
the reader may be disinclined to treat all three sentences as
deriving from the same propositional structure.

Indirect object
We now turn to ‘indirect objects’: their status is very ambiva¬
lent. The term is used when a verb is followed by two NPs,
neither of which is associated with a preposition. In these
structures the first NP is the indirect object, and the second the
direct object:

65 John gave Mary (IO) a book (DO)

A sentence like 65 can be paraphrased as:

66 John gave a book to Mary

Here the direct object directly follows the verb; what was the
indirect object is now introduced in a prepositional phrase. If the
direct object immediately follows the verb, then the other NP
must be introduced by a prepositional phrase"' - there is no
sentence:

67 *John gave a book (DO) Mary (IO)

5 There are some exceptions to this generalization. Normally the IO object is

animate and the DO inanimate (reflecting that normally one gives ‘things’ to
‘people’). This need not necessarily be the case:

John gave the school a library (a library to the school)

John gave the nunnery his daughter (his daughter to the nunnery)
John gave the nurse the child (the child to the nurse)

In such cases the paraphrase with to seems preferred to the structure involving IO
and DO: doubtless in order that the semantic relations should be clearly marked.
Further observe that if we pronominalize the DO, then only the first of the
following sentences is grammatical:
344 Part three: Functional relations

Frequently the prepositional phrase in a sentence like 66 is also

referred to as an indirect object. We shall not, however, do this,
reserving the term indirect object for the first of two NPs in
sentences like 65. Instead we call it, as before, an oblique object.
From a semantic point of view, the indirect object frequently
realizes a [goal] participant, as in 65. But not always. We find a
[benefactive] in:

68a Mary baked me a cake

68b Mary baked a cake for (*to) me

a [comitative] in:

69a Mary played John tiddlywinks

69b Mary played tiddlywinks with (*for, *to) John

and perhaps a [source] in:

70a John asked Mary a favour

70b John asked a favour from (*to, *for) Mary

Sometimes the [goal]:[benefactive] distinction is unclear (perhaps

not surprising in view of the close relations between these roles -
page 327):

71a Mary sang me a song

71b Mary sang a song to/for me

We call all the participants realized in prepositional phrases

John gave it to the school

*John gave the school it

If, however, we pronominalize the IO then both:

John gave it a library

John gave a library to it

are possible. If we pronominalize both NPs then, if both are animate or both
inanimate sentences like-

*John gave it it
*John gave her him

do not seem acceptable. If however one NP is animate and the other inanimate,
then we can have:

John gave it her

John gave her it

and whichever order is used the animate is understood as the IO and the inanimate
as the DO!
 Subjects, objects, complements and adjuncts 345

oblique objects, and, if necessary, distinguish between a

‘benefactive oblique object’, a ‘comitative oblique object’, and so
on.
The peculiarity about structures of this sort seems to be that one
participant is, as it were, the ‘designated’ direct object. It retains
this function even when the NP in the oblique object ‘becomes’ an
object too - thus giving rise toaV + NP + NP structure. Before
demonstrating this, let us enquire how widely the process of what
we may call ‘indirect object formation’ applies. We illustrate with
‘benefactive oblique objects’, and invite the reader to determine
which of the following are possible:

72a Cash me a cheque (a cheque for me)

72b Buy me a beer (a beer for me)
72c Catch me a butterfly (a butterfly for me)
72d Do me a favour (a favour for me)
72e Change me a fiver (a fiver for me)
72f ?Run me an errand (an errand for me)
72g ?Purchase me a beer (a beer for me)
72h *Feed me the cat (the cat for me)
72i *Beat me the carpet (the carpet for me)
72j * Pursue me the postman (the postman for me)

(We indicate our judgements, as before, with ? for questionable,

and * for impossible.) Some of these structures were acceptable
in earlier stages of English, and some, for some speakers, still are
acceptable. This variability seems to occur with other indirect
object constructions.
We now apply our criteria for objecthood to the direct and
indirect objects in these sentences. We have already seen that the
indirect object may occur as an oblique object, and that this is not
possible for the direct object. We now observe that the indirect
object, or its corresponding oblique object, can usually be omitted
without affecting the grammaticality of the sentence, but the direct
object cannot be omitted:

73a John gave Mary a book

73b John gave a book
73c *John gave Mary (except in another sense)
74a John bought me a beer
74b John bought a beer
74c *John bought me (except in another sense)
346 Part three: Functional relations
Consider now the problems of passive formation in these
sentences. When the direct object immediately follows the verb, it
can become the subject of the corresponding passive:6
75a John gave that book to Mary
75b That book was given to Mary by John
Similarly, the indirect object, the NP immediately following the
verb, can usually become the subject of the corresponding passive:

76a John bought Mary a new car

76b Mary was bought a new car by John
77a Mary baked me a cake
77b I was baked a cake by Mary
78a John bought me a beer
78b I was bought a beer by John
but not always:
79a John asked Mary a favour
79b ?Mary was asked a favour by John
80a John caught me a butterfly
80b *1 was caught a butterfly by John
81a Mary played John tiddlywinks
81b *John was played tiddlywinks by Mary

Once again the bounds of acceptability are not entirely clear.

A further difficulty is that in indirect object constructions
(V + NP + NP) the direct object can still sometimes become the
subject of the passive sentence:

82a My grandfather gave me this watch

82b This watch was given me by my grandfather
83a John bought me that bottle of Glen Grant
83b That bottle of Glen Grant was bought me by John

6 All sorts of problems arise in this area if we try to passivize indefinite N Ps.
Thus:
A book was given to Mary by John
seems intuitively less acceptable than:
That book was given to Mary by John
Here we deal only with definite NPs. Indefinite NPs as subjects are discussed on
pages 370-1.
 Subjects, objects, complements and adjuncts 347

The NP in the oblique object cannot, however, become subject of

the passive sentence (unless, of course, it has first ‘become’
indirect object):

84a My grandfather gave this watch to me

84b *1 was given this watch to by my grandfather
The NP that immediately follows the verb has a privileged
status, both syntactically and semantically. When only one NP is
available for this role (i.e. in two place propositions) there would
seem to be a hierarchy of ‘objecthood’ (page 341). When two NPs
are available for this role (i.e. in three place propositions), the
situation is more complex. Sometimes one NP seems to have what
we have called a ‘designated’ object role, and retains many of its
object-like features even when displaced from the position
immediately following the verb (i.e. in indirect object + direct
object constructions). In other circumstances neither of the NPs is
a designated object, and then either one of them must become an
object (e.g. with SUPPLY - pages 341-2) or neither of them need
become an object (e.g. with SHOOT - page 343). The hierarchy of
objecthood can be extended to accommodate this. Direct objects
are most ‘object-like’, in terms of the criteria established at the
beginning of this section, followed by objects of result, indirect
objects, objects of concern, etc., eventually arriving at oblique
objects.
We should not leave this discussion of objects without
mentioning briefly ‘phrasal’ verbs, since their syntax clearly
intersects with our observation on oblique objects. Phrasal verbs
are items like BLOW UP in:

85 The commandos blew up the bridge

In such cases UP seems to be related to BLOW rather than to be a

constituent with the bridge. We have already seen (page 140)
that the syntax of these constructions supports such an analysis.
Furthermore the constituent the bridge operates syntactically like
the direct object in such constructions. It is an obligatory
constituent; it can occur immediately following the verb; and it can
be the subject of the corresponding passive:

86a *The commandos blew up

86b The commandos blew the bridge up
86c The bridge was blown up by the commandos
348 Part three: Functional relations

The interested reader is referred to Quirk et al. (1972) for general

discussion; to Bolinger (1971) for a perceptive analysis of many
different constructions of this type; and to Jespersen (1927) for his
many examples.

Complement
Traditional grammars distinguish between a number of different
complement types.7 The names given to them, like the names given
to different types of object, relate to the propositional roles that
they realize. In two place sentences, the primary distinction is
between ‘attributive’, ‘identity’ and ‘locative’ complements: we
discuss each in turn.
The italicized constituents in 87 are attributive complements, so
called because they describe the class membership of the subject
noun, or ascribe an attribute to it:

87a Roses are red

87b Roses are flowers

Because such complements relate back to the subject noun, they

are called ‘subject complements’, a relationship sometimes
described as ‘intensive’ to the subject. Further distinctions can be
profitably drawn. In 87b the noun flowers is a ‘nominal
complement’. The complements in 87 are ‘state complements’
since they are found in stative sentences and describe states, in
contrast to the complements in inchoative sentences like:

88a The leaves are turning brown

88b Sir Charles became a soldier

which are known as ‘result complements’. The variety of

terminology in this area relates to the complex propositional status
of such constituents, discussed in the last chapter. It also reflects
syntactic differences between the various types of complements,

7 Sometimes all NPs that follow the verb are known as complements. In this
usage, what we have called the ‘direct object' is called an ‘extensive complement’,
to distinguish it from the ‘intensive complement’, which relates back to the subject.
We do not follow this usage.
The term complement is also applied to the sentential objects that are associated
with particular verbs. Thus in the sentence John believed that he was clever the
constituent that he was clever is a sentence complement. The description of that as a
‘complementizer’ (cf. page 134) derives from this usage.
 Subjects, objects, complements and adjuncts 349

and between the complements and the direct object. The

complement is an obligatory constituent and always follows the
verb. The nominal complement is almost always an indefinite NP,
and concords in number with the noun to which it is intensive (He
became a soldier; They became soldiers). The complement cannot
become the subject of a passive sentence. Some copular verbs (BE,
SEEM) are restricted to stative sentences, others (BECOME,
turn) to non-stative sentences.
The identity complement can be illustrated in:

89 Harold is the man with the arrow in his eye

and is always a definite NP. The two constituents in such a sen¬

tence can be reversed round the copular verb:

90 The man with an arrow in his eye is Harold

In these circumstances, which constituent is to be identified as the

complement depends on textual considerations (pages 316-17). In
sentences in isolation the NP following the verb is most usually
identified as the identity complement:

91 Harold is (to be identified as) the man with the arrow in

his eye
The prepositional phrase in a sentence like:

92 Mary is in bed

is called the locative complement. The locative complement is

usually prepositionaPphraseT^as in 92, or a place adverb like
upstairs, over there, etc. The locative complement is typically, but
not invariably, an obligatory constituent, and can be found with a
variety of verbs in state sentences:

93 The newspaper is lying on the floor

Your coat is hanging in the cupboard
The Scott Monument stands on Princes Street

(See discussion on pages 307-10.) Corresponding to the locative

complements in 92 and 93, we can also recognize a ‘directional
complement’ in sentences like:

94 The ball rolled under the bed

The car skidded across the road

If we distinguish between locative and directional complements,

350 Part three: Functional relations
then we may also note that the latter typically occurs only in
non-state sentences.
These three complement types correspond to three of the basic
types of propositional structure identified in the preceding chapter
(see Figure 106, page 319):

95a attributive John is a fool (state)

John is clever
The litmus paper is turning red (result)
95b identity John is chairman
95c locative Mary is in London (place)
The ball rolled under the bed (direction)

It is interesting, and somewhat curious, that the fourth basic type

we identified:

96 possessive Mary has a little lamb

Mary has red hair

is not also described as involving a complement - but it isn’t. (It

may be observed that Mary is a redhead does involve a
complement!). We return briefly to this point below.8
We now turn to sentences like:

97a I consider John a fool

97b I consider John clever
97c Acid will turn litmus paper red
97d The committee elected John chairman

8 In English the copula verb be is used in all three types of complement

construction. This is not the case in all languages. In Akan, for instance, we find:

attributive (Ye) Kofi ye ohene hi (Kofi is chief a) Kofi is a chief

Kofi ye kesee (Kofi is big) Kofi is big
identity (Ne) Kofi ne ohene no (Kofi is chief the) Kofi is the chief
ohene no ne Kofi The chief is Kofi
locative (Wo) Kofi wo ofie no mu (Kofi is house the in) Kofi is at home

This can cause, and has caused, translation problems: in a Bible translation, is
the copula in God is love to be translated with ye (has the attribute) or ne (is to
be identified as) ?
We should also observe that in this language we also find:

possessive (Wo) Kofi wo ofie yi Kofi owns this house

ofie yi wo Kofi This house belongs to Kofi
 Subjects, objects, complements and adjuncts 351

The italicized constituents are ‘object complements’ - so called

because they are intensive to the object. The distinctions ‘state’,
‘result’, ‘nominal’, etc. that we drew for the subject complements
in 87 and 88 can also be drawn here. In the sentences illustrated
the complement is usually obligatory; constituent order is usually
as shown; the nominal complement agrees in number with the NP
to which it is intensive; and so on.
We have already noted (page 318) that many of the sentences in
97 correspond to paraphrases like:

98 I consider John to be a fool

This suggests that such sentences might be seen as derived from an

embedded predication. We can illustrate this by the following
sequence:

99 I consider (John is a fool)

I consider that John is a fool
I consider John to be a fool
I consider John a fool
100 They elected John (John is chairman)
They elected John to be chairman
They elected John chairman

In the case of a sentence like 97c a more abstract representation

might be more suitable:

101 Acid CAUSE (litmus paper turn red)

where CAUSE is a representation of the causative status of TURN.

Such representations remind us of a similar analysis discussed on
page 302. They also make very obvious the parallel between
the two place constructions set out in 95, and the corresponding
three place sentences in 97.
Before turning to three place sentences with locative comple¬
ments, we should note that the italicized constituents in the
sentences in 102 are also often called complements:

102a Jane returned home safe

102b Harriet danced the tango naked
102c I always drink my coffee cold
102d I always buy my meat fresh
102e They painted the fence yellow
 352 Part three: Functional relations
Some of these items are intensive to the subject, 102a, 102b, and
others to the object, 102c, 102d; the state:result distinction can be
drawn. In these sentences the complements seem to have a looser
syntagmatic tie to the verb than do the complements previously
discussed. They are, to begin with, both omissible and not relevant
for strict subcategorization. Furthermore their status seems to be
more ‘adverbial’ than that of the object complements of 97: in
some cases they can be substituted by adverbs {safely instead of
safe in 102a). Other cases relate to a variety of different
paraphrase constructions - I always drink my coffee after it has
become cold; Harriet was naked when she danced the tango; and so
on.
Consider now the following three place locative sentences:

103a John put the book on the table

103b John rolled the ball under the bed

The prepositional phrase constituent is still described as a locative

complement, and the place: direction distinction observed in 95c
still applies. As before, the relationship between the two and three
place sentences is a causative one. We can represent this with the
abstract formulation:

104a John CAUSE(PUT) the book (the book BE on the

table)
104b John C A US e( ROLL) the ball (the ball GO under the
bed)

BE and GO represent, respectively, the place and directional

nature of the embedded predications concerned. These abstract
representations are rather distant from the actual surface form of
the sentence; it is not suggested that the representations might be
underlying structures, in the sense that we used this term (page
124).
We now close this section by returning to the possessive
construction mentioned earlier:
105 Mary has a little lamb

106a The shepherd gave Mary a little lamb

106b The shepherd gave a little lamb to Mary

The constituent a little lamb is not usually described as a

‘possessive complement’ or the like, either in 105 or in 106. Indeed
in 106 this constituent is called the direct object, and in 106a Mary
 Subjects, objects, complements and adjuncts 353
is the indirect object, as we have seen (pages 343-7). We also saw
that the status of the indirect object is less clear than that of the
direct object. It is now clear that the relationship between the sen¬
tence in 105 and that in 106 is not very dissimilar from the relation¬
ship between the two-place locative sentences in 95c and their
three place counterparts in 103. We might represent this with
abstract representations like:

107 Mary HAVE a little lamb

108a The shepherd CAUSE(GIVE) Mary (Mary have a
little lamb)
108b The shepherd Cause(GIVe) alittle lamb (alittlelamb
GO to Mary)

The different orderings in the embedded sentence correspond to

the two different surface orderings in 106, and perhaps correspond
to a different nuance of meaning. An analysis of this sort may also
go some way towards explaining the apparent difference in
meaning surrounding the choice of object with verbs like supply
and SHOOT discussed on pages 341-3, but to pursue this matter
further would take us too far out of our way (it is discussed further
in Brown and Miller, in press).

Adjuncts
Subjects, objects and complements form the nucleus of a sentence.
They are nuclear in at least two senses, both of which are
complementary. They are the constituents which, in constituent
structure terms, are either obligatory or are introduced by an
expansion of the VP, and hence relevant to strict subcategoriz¬
ation. They are also to be identified with nuclear participants
in the sense discussed in the preceding chapter: many of the names
given to different types of subject and object reflect this.
All other sentence constituents we call adjuncts. Adjuncts are
usually adverbials, whether they are adverb phrases, prepositional
phrases, adverbs or subordinate clauses of time, place, manner,
etc. that distributionally function like adverbials. They are
typically optional sentence constituents, and have a degree of
mobility within the sentence denied to the nuclear constituents.
Adjuncts are clearly a rather ‘mixed bag’, in that syntactically
there are numerous subclasses which have different and over¬
lapping distribution, and they fill a variety of semantic roles, as
354 Part three: Functional relations
can be seen by consulting the discussion of adverbs in any standard
grammar of English - for example Quirk et al. (1972).

Conclusion
In the preceding chapter we characterized sentence constituents in
terms of their syntactico-semantic function within a sentence as
[agent], [patient], and so on. We suggested that a propositional
structure characterized in these terms is an abstract structure
which could be realized by any one of a number of actually
occurring sentences. A given propositional role may occur in a
number of syntactic positions within the sentence without affecting
the way in which the role is characterized, though there will be
other syntactic phenomena that reflect the different sentence
positions the role occupies. We also noted that the
syntactico-semantic characterization can be used to control various
grammatical features, like the ability of some roles to occur with
the progressive auxiliary, its ability to occur in imperative
sentences and so on.
In this chapter we have looked at the functional roles of subject,
object and complement, and at the various syntactic and semantic
properties that can be associated with these functions. Of
particular interest has been the various circumstances that permit
an NP to take on the function of subject, object, etc. From a
transformational point of view, deriving the various different
sentence orders from a single underlying structure, the discussion
may be seen to consider the ways in which sentences can be
provided with subjects and objects.9 Suppose, for example, we
establish a hierarchy of functional roles in outline like this:

109 subject - object - indirect object - oblique object

and suppose the underlying order of constituents in the sets of

sentences in 110 and 111 is as shown in the first of each set of
sentences, then the italicized constituent as it were ‘climbs up’
the hierarchy:

9 The transformational account can of itself only deal satisfactorily with the
actual formal operations of providing subjects and objects, etc. It does not handle
the semantic consequences of these operations very satisfactorily, unless additional
interpretive machinery is provided somewhere in the total description to take
account of the semantic consequences of the operations. As we have seen, these
vary from verb to verb and may sometimes be considerable.
 Subjects, objects, complements and adjuncts 355
110 John gave the book to Mary (oblique object)
John gave Mary the book (indirect object)
Mary was given the book by John (subject)

111 The Russians supplied arms to the Vietnamese (oblique

object)
The Russians supplied the Vietnamese with arms
(object)
The Vietnamese were supplied with arms by the
Russians (subject)

If we also consider subject raising to be within the grammar, then a

constituent can, as it were, climb to the front of its own sentence,
hop up into a superordinate sentence, and then climb up that:

112 si (‘People’ allege ^(John gave the book to Mary))

By the process illustrated in 110 we get:

113 §i (‘People’ allege §2(Mary was given the book by John))

Raising produces:

114 si (‘People’ allege Mary^ (to have been given the

book by John))
Passivization of SI, and the deletion of the indefinite ‘people’,
then yields:

115 Mary is alleged to have been given the book by John.

In this account, while the functional relation of the NP Mary

changes from derivation to derivation, its propositional role does
not - it is still understood as having a [goal] relation to GIVE.
The conditions under which Mary climbed up its own sentence,
hopped into the superordinate sentence and then climbed up that
have little to do with its propositional role, but are instead related
to its functional role as successively oblique object, indirect object,
subject, and so on.10

10 We observed at the end of the preceding chapter that there is a school of

‘case grammarians’ who take propositional roles to be basic. We may now observe
that there is a developing school of ‘relational grammar’ that takes functions such as
subject, object, etc., to be basic in much the same way. There is, so far,
comparatively little publicly available explaining the advantages of this approach,
but see some of the articles in Cole and Sadock (1977).
356 Part three: Functional relations
There is very obviously a large degree of overlap between the
two types of characterization - of propositional and of functional
role. Are both types of characterizations necessary? Or could one
type or the other not be ‘enriched’ so as to accommodate both sorts
of description? We have attempted this to some degree: we used
an ‘arrow’ notation in the preceding chapter (see, for example
page 307). and have remarked that the different names given to
different types of object and complement reflect their
propositional status. On the other hand the two different
characterizations do seem to control rather different sorts of
syntactic phenomena, and in our present state of knowledge it
seems best to retain both.

Technical terms
subject - grammatical, logical, thematic
object - affected, cognate, of concern, direct, effected, factitive,
indirect, oblique, of result
complement - attributive, identity, locative, nominal
adjunct
existential sentence
raising
20 Theme, rheme and end focus;
topic and comment;
given and new

Notions like subject in the previous chapter were discussed by

examining sentences in isolation. We now consider another use of
the term subject: what can loosely be called the ‘thematic subject’,
the ‘psychological subject’ or the ‘subject of discourse’. To avoid
confusion, we do not use the term subject in this connection
(reserving it for the grammatical subject discussed in the preced¬
ing chapter), but instead introduce the terms theme, rheme and
end focus, topic and comment, and given and new. We are con¬
cerned with those features of word order within sentences that are
related to the communicative function of sentences within texts.
The relevant notions clearly lead off into a study of the structure
of text and it is not our intention to pursue this matter in detail.
‘Theme’, ‘rheme’ and ‘end focus’ refer to structural positions
within the sentence. The two focal points in English sentences are
the beginning and the end; the language has a number of pro¬
cesses that position a constituent either initially or finally. The
constituent that' occurs in initial position is the theme, and
processes used to make some constituent initial are processes of
thematization. The theme is italicized in the following:

1 Someone parked a large furniture van right outside our

front door last night
2 A large furniture van was parked right outside our front
door last night
3 Right outside our front door someone parked a large
furniture van last night
4 Last night someone parked a large furniture van right
outside our front door

The propositional structure of each of these sentences is

identical: they differ in terms of which constituent is thematized.
Constituents other than the theme are the rheme. Obviously
358 Part three: Functional relations

some constituent must be sentence initial, and the word order

shown in 1 is usually thought of as being the ‘most neutral’ or
‘unmarked’ word order in a set of sentences like 1-4: it is an
active declarative sentence with place and time adverbs in end
position. Such sentences we call thematically unmarked. The
other sentences are thematically marked in one way or another,
and some involve what one intuitively thinks of as a greater
degree of markedness than others - 3 for instance. We can also
say that there is an unmarked order of constituents in the rheme.
What the unmarked order is depends on which constituent is
thematized; and we must use a somewhat intuitive notion of what
we consider the unmarked or ‘most neutral’ order of rhematic
constituents. It will probably be agreed that the rhemes in 1-4
show an unmarked constituent order, given that some constituent
has been thematized. Now, just as some constituent must be
thematized, so some constituent must occur in final position. We
call the final position the ‘end-focus’.1 If all the rhemes in 1-^4 are
in an unmarked word order, then all the end-focuses are also
unmarked. However, just as there are thematizing processes-
which produce marked themes, so too there are end-focusing
processes that produce marked end-focus. Consider:

5 Someone parked a large furniture van last night right

outside our front door
6 It was parked right outside our front door last night, a
large furniture van
7 Parked right outside our front door last night it was, a
large furniture van
8 A large furniture van, right outside our front door last
night, parked\

1 The term end-focus is taken from Quirk et at. (1972). Their use of the term is
derived from the fact that, in spoken language, the unmarked position for the
intonation centre in an English sentence is the last lexical item in that sentence.

The theme is the last conSTiTuent

If the intonation contour falls on any other constituent, as:

The theme is the last constituent

The theme is the last constituent

then the sentence is taken to be in some sense marked - the effect is often
emphatic or contrastive. We do not discuss spoken language, except incidentally,
and use the term end-focus simply for the last relevant sentence constituent.
 Theme, rheme and end focus 359

Some of these end-focuses are clearly more marked than others

as the reader can confirm by reading them aloud - they involve a
successively more indignant intonation pattern!
Considerations that govern which constituent is to be chosen as
theme and which as end-focus are related to communicative
processes within text.
The terms given and new can only be understood in terms of
text. In their most straightforward sense, these terms can be
understood as information that has literally been ‘given’ in the
preceding text and information that is ‘new’ to the sentence
immediately under consideration. A number of linguistic features
correlate with this. New information is characteristically spelled
out in full - otherwise there is no way for the hearer or reader to
get access to it. Given information is typically either assumed and
not referred to at all, or is referred to by the use of proforms or
other cross-reference expressions. For example, suppose you
received a letter which contained:

9 I must tell you the news about John and Mary

10 They have just got married

9 introduces the new information that there is some ‘news’ about

‘John and Mary’. When we come to 10 this can be treated as
given: hence they cross-refers to John and Mary, and no mention
is made of ‘news’, but a piece of new information is added - that
the ‘news’ is that ‘John and Mary have got married’. It would be
unnatural if the letter read:

11 I must tell you the news about John and Mary

12 The news I have to tell you is that John and Mary have
got married

where all the given information is spelled out in the second

sentence. In a long text such a process eventually becomes
impossible, since new information is constantly being added.
Given information need not always be overtly referred to in
text. It may be information ‘given’ in the sense that both
participants share it as speakers of the same language, it may be
cultural information shared between members of a linguistic
community, or it may even be information privately shared
between two individuals. To illustrate this, suppose the ‘letter’ in
9-10 continued:
360 Part three: Functional relations

13 I must tell you the news about John and Mary

14 They have just got married
15 A very flashy reception with lots of extravagant presents
on display.
16 Dear old Charlie gave a toast-rack!
17 He does have a flair for the original!

The shared background of writer and reader includes the

knowledge that marriages may involve receptions and the
provision of presents. The way ‘dear old Charlie’ is introduced
implies a certain type of relationship between the writer and
‘Charlie’, which is given to the reader. The last sentence can only
be appreciated if one realizes that a ‘toast-rack’ is a totally
conventional wedding present (not to say a literary convention).
From a linguistic point of view, we note the pronominalization of
‘given’ information, and the elision of They had in 15, and of the
goal participant in 16 . . . gave (them) a toast-rack.
The importance of the notions of given and new to a study of
word order lies in the fact that what is given is very frequently the
constituent that is thematized. The unmarked structure for text
might be schematically represented as:

18 given new. given new. (given) new. etc.

where the given is pronominalized and thematized, and the new

information occurs in the rheme. We see several examples of this
in the letter: they in 14, he in 17. What is given may be elided: in
15 we find (they had) a very flashy reception ... so that the
resultant sentence is all new information. We might regard a
toast-rack in 16 as an example of end focus (note the prominence
given to this constituent by the exclamation mark). Another way
of presenting the same information is: A toast-rack from dear old
Charlie, where the ‘given’ (and understood) They received, or the
like, is elided and A toast-rack is thematized.
The notions of topic and comment can be illustrated in the
following way. By topic we mean the ‘perspective’ from which a
sentence is viewed, what the sentence is ‘about’. Topic seems to
correspond to what people mean by ‘psychological subject’. The
comment then is something said about this topic. With sentences
in isolation, we normally understand a sentence like:

19 John patted the dog on the head

 Theme, rheme and end focus 361

as being a sentence which takes John as its topic and tells us

something about what ‘John’ did. By contrast the sentence:

20 The dog was patted on the head by John

seems to focus attention on ‘the dog’, and tells us something that

happened to it. Frequently topic and theme coincide, as they do
in 19 and 20. Frequently, too, topic, theme and given coincide, as
in 14 and 17. So, in 14 for example, they is topic and theme and
cross-refers to John and Mary in the previous sentence: the
comment have just got married is then made on this topic.
Topic and theme need not, however, coincide. Consider:

21 There has been a lot of bullying in the school this term

22 The one who is always being picked on is Tom Brown

We might reasonably identify Tom Brown as the topic of 22, yet

here this constituent is in end-focus. The effect of putting Tom
Brown in end-focus is precisely to delay identification of the
topic, thus creating a particular communicative effect.
Before considering each of these three notions in more detail
we should briefly mention the question of what ‘meaning’
attaches to them. At this point we enter a peculiarly difficult and
shadowy area. Clearly sentences like 1-4 or 5-8 have an
‘emphatic’ or ‘contrastive’ effect, but it is difficult to characterize
this with any precision. Similarly, the ability to recover ‘given’
information, as for example in the ‘letter’ in 13-17, has
communicative importance - but this again is difficult to
characterize. Thematizing processes do not in general appear to
affect what we may refer to as ‘propositional meaning’ - the
underlying propositional roles of the various constituents: though
even this statement needs some qualification since, as we have
seen, subject and object forming processes are often significant.
In one area thematization has a particular effect. This involves
sentences where one or more of the NPs involved contains a
quantifying expression (like every, few, many or a numeral) or
the sentence contains a negative element. So, for example, the
sentence:

23 Everyone in this room speaks two languages

can be understood as asserting either that ‘everyone’ speaks the

same two languages (say French and German); or that everyone
speaks two languages, but they may be different for each
362 Part three: Functional relations
individual (say French and German, Chinese and Hindi, and so
on). The latter is the more usual interpretation: the sentence is
understood as a comment on the linguistic ability of ‘everyone in
the room’, which is the topic. By contrast, in the sentence:

24 Two languages are spoken by everyone in this room

the most usual interpretation is that just two languages, say

French and German, are at issue and everybody speaks them.
Some support for this may be gathered by considering ways in
which the sentences may be extended:

25 Two languages are spoken by everyone in this room and

I understand neither of them
26 Everyone in this room speaks two languages and I
understand neither of them

The first of these sentences seems immediately comprehensible.

In the second the reader may be momentarily confused by the
suspicion that ‘everyone’ comprises only two people (neither of
them). In some cases this effect is so strong that pairs of sentences
appear to differ in meaning; sometimes one sentence of a pair is
almost incomprehensible:

27a Many men love few girls

27b Few girls are loved by many men
28a Nobody speaks seventeen languages
28b Seventeen languages are spoken by nobody
29a A few men marry two women
29b Two women are married by a few men

The ‘focusing’ effect of thematization is very noticeable in

sentences like these, but it varies from case to case. 23 and 24
may fairly easily be understood in either sense, though one sense
is probably preferred for each sentence. Each sentence in 28
appears to have a distinct interpretation, though with a little
effort the reader can contextualize the other interpretation. In
cases like 29, the b sentence is almost incomprehensible.

Theme, rheme and end-focus

Of the notions just discussed, theme, rheme and end-focus offer

least descriptive and conceptual difficulty, because they are
 Theme, rheme and end focus 363
formal terms identifying structural positions within the sentence.
The initial, thematic, constituent is where the sentence starts; and
the final, end-focus, constituent is the culmination of the
sentence. In a sentence like:

30 Ice cream, that’s the pudding I like best in the world

the speaker, as it were, announces,what the topic of his sentence

is and then goes on to make a comment on it. By contrast, in
a sentence like:
31 The pudding that I like best in the world is ice cream

the speaker keeps his hearer in suspense as to what he likes until

the very end. We can reasonably successfully specify the formal
operations involved in forming sentences like 30 and 31; it is
much more difficult to say why a speaker might prefer 30 to 31,
or vice versa, since this involves the more difficult considerations
of what sort of effect the speaker thinks he will achieve by using
one or other sentence.
In this section we examine some of the formal machinery
available in English to enable a speaker to thematize a
constituent or bring it into end-focus. We begin with declarative
sentences: interrogative and imperative sentences function
somewhat differently, so we defer consideration of these.
First we consider thematization processes. These fall into four
very general types.
The first type we call ‘subject selection rules’. These are cases
where any one propositional role within a given propositional
structure may be selected as grammatical subject, and the
sentence remains an active declarative sentence:

32 Blood flowed in the gutters

33 The gutters flowed with blood

The second we call ‘promotion to subject rules’: these are cases

where a particular propositional role, which in an active
declarative sentence would not be grammatical subject, is
promoted to subject with some consequent alteration in the verb
group (as is the case with the passive) or by the introduction of a
pro-verb (like HAVE):

34a The managing director sacked the strikers

34b The strikers were dismissed by the managing director
364 Part three: Functional relations

35a My auntie knitted a pair of gloves for me

35b I had a pair of gloves knitted for me by my auntie

35b is to be understood as having the same general sense as 35a,

rather than in the causative sense ‘I got my auntie to knit me a
pair of gloves’.
The third general type is called ‘left movement rules’: these
involve the thematization of a particular constituent without any
consequent change of grammatical function: thus the subject
function of I and the object function of Christmas do not alter in:

36a I hate Christmas

36b Christmas, I hate it

The fourth type we call ‘clefting rules’: these involve the

distribution of the constituents of some proposition into a
copular sentence:

37a I am very fond of marzipan

37b It’s marzipan that I’m very fond of
37c Marzipan is what I’m very fond of

Before discussing these processes, some general comments can

be made. We cannot possibly list all possible types and subtypes
exhaustively in this chapter, nor are these four general types
mutually exclusive - they are not and they interact in various
ways. The characterization is merely a useful ad hoc typology.
Nor do we claim a strict paraphrase relation between pairs of
sentences examined, as for example any of the pairs of sentences
noted above: it does seem, however, that the propositional
relations involved remain the same. The thematization processes
themselves may add emphases of meaning, or predispose the
reader to a particular reading in preference to some other
possible reading.
We can group a number of processes together under the
general heading of ‘subject selection rules’. The most obvious
cases involve verbs like flow, illustrated in 32, 33, which are
associated with two propositional roles, either of which may
become subject. Another example is
38a His face streamed with blood
38b Blood streamed down his face

and similar sentences can be constructed with verbs like CRAWL,

BENEFIT.
 Theme, rheme and end focus 365

A large class of verbs and related adjectives that refer to

psychological states occur in structures like:

39a I am bored with writing

39b Writing is boring to me
40a I am excited at the prospect of Christmas
40b The prospect of Christmas is exciting to me

and similarly AMUSE, CONFUSE, DISGUST, EXCITE,

FRIGHTEN, AMAZE, PUZZLE, etc.2 Under the
WORRY,
general heading of subject selection we can also include cases of
lexical suppletion, since these too permit one rather than another
constituent to become grammatical subject, and the resulting
sentences are, in general terms, paraphrases. We have explored
some converse relations (pages 305-6).

41a I lent my daughter five pounds

41b My daughter borrowed five pounds from me

and similarlywith BUY : SELL; TEACH : LEARN;

amuse : please; etc. We may also, under this general heading,
note a number of verbs where an ‘indefinite and non specific’
agent role may not be realized at all, and some other constituent
assumes the role of grammatical subject:

42a They produce a lot of whisky in Scotland

42b Scotland produces a lot of whisky

Similar verbs, permitting this construction include GROW,

MAKE, MANUFACTURE.
Perhaps the paradigm case of a ‘promotion to subject’ rule is
the passive:

43a A bus knocked Mary down

43b Mary was knocked down by a bus

2 The usage we are concerned with here is the non-passive one illustrated in
40. Sentences like these offer some analytic interest: although a sentence like 40a
appears to contain a passive verb group, we note the preposition at rather than by.
We can also observe that the item excited here operates more like an adjective
than a ‘true’ passive since we can modify it with ‘intensifiers’ etc.:

I am extremely/very/quite excited at the prospect of Christmas

Passives formed with transitive verbs do not operate like this:

*John was very killed by Bill

366 Part three: Functional relations

A very characteristic use of the passive is illustrated by:

44a Someone has eaten all the cheese

44b All the cheese has been eaten

where in the b sentence the indefinite and non-specific agent

expression has been elided. Indeed a sentence like:

45 All the cheese has been eaten by someone

is often held to be unacceptable, and is certainly stylistically

marked. Such structures are also often used when the speaker is
anxious not to mention a specific participant, or when it is
contextually recoverable. Sentences involving modal verbs often
lend themselves to loose paraphrases involving a promotion to
subject:
46a You might buy a second-hand motor-bike for £200
46b £200 might buy you a second-hand motor-bike

The same sort of process is observable with many other verbs

with a modal or aspectual meaning:

47a You need to have your car serviced

47b Your car needs to be serviced
48a This novel begins to interest me
48b I begin to be interested in this novel
‘Raising’ transformations (see discussion on pages 334-5) can
also have the function of creating thematic constituents. Thus
consider sentences like:

49a (John will go) is certain

49b That John will go is certain
49c John is certain to go

The subject of the constituent sentence is raised to become the

subject of the matrix sentence: similar sentences can be formed
with likely, seem, happen. A similar operation can raise the
object of a constituent sentence to become the subject of the
matrix sentence:

50a (to please John) is difficult

50b John is difficult to please

and similar sentences with verbs and adjectives like hard,

TOUGH, IMPOSSIBLE (but not POSSIBLE), EASY.
 Theme, rheme and end focus 367

Other promotion to subject rules involve the introduction of

pro-verbs like have, get, etc. have seems involved in a
number of constructions involving locative or benefactive roles,
or participants that occur in genitive constructions:

51a There is a pond in my aunt’s garden

51b My aunt’s garden has a pond in it
52a There is an article by the Bishop of London in
today’s Times
52b Today’s Times has an article by the Bishop of
London (in it)
52c The Bishop of London has an article in today’s
Times
53a The policeman twisted the criminal’s arm
53b The criminal had his arm twisted (by the
policeman)
54a Someone bought a gold watch for the dustman
54b The dustman had a gold watch bought for him

In 54b we see once again the non-realization of an indefinite and

non-specific-agent participant, paraphrases with take and get
often seem to involve goal roles:

55a The waves battered the ship

55b The ship took a battering from the waves
56a Parliament didn’t approve the Devolution Bill last
year
56b The Devolution Bill didn’t get the approval of
Parliament last year

(Note in these cases the nominalization of the verb - see footnote

6 on page 300.)
We turn now to our third set of rules - ‘left-movement rules’.
Many constituents can be thematized by simply moving them to
the front of the sentence. We have already observed (page 129)
that this can be done with sentence adverbs:

57 Last night I proposed to Mary

58 On Hogmanay in Scotland few people go to bed before
midnight

When constituents are thematized in this way it does not affect

their status as grammatical subject, object, etc.:
368 Part three: Functional relations

59 Mary, I proposed to
60 That soup, I find totally disgusting
61 The kittens, we drowned

Sometimes a pronoun is ‘left behind’ in the slot from which a

constituent is moved:

62 Jane, she’ll be late home this evening

63 That soup, I find it totally disgusting

Rules of this sort often permit a constituent to be thematized

from quite ‘deep down’ in an embedded sentence:

64 That man, I thought you told me you were never going

to see him again

Our last general type of thematization operation is ‘clefting’.

Under this head we mention two types of operation. The first is
called simply clefting:

65a John gave the book to Mary

65b It was John who gave the book to Mary
65c It was the book John gave to Mary
65d It was Mary John gave the book to.

The syntax of cleft sentences is far from straightforward, and the

reader is urged to do a little personal research into the extent to
which the operation can be extended. It seems that almost any
constituent can be clefted in English except the verb, and even
this is possible in some dialects of English:

66a ?It’s singing John is

66b ?It’s delivering the mail the postman is

The distribution of tense, mood and aspect features in cleft

sentences is also worth investigation:

67a John might have given the book to Mary

67b It might have been John who gave the book
to Mary
67c It is John who might have given the book
to Mary

Another structure of this general type is pseudo-clefting,

illustrated by sentences like:
 Theme, rheme and end focus 369
68a John loves Mary
68b The one who loves Mary is John
68c John is the one who loves Mary
68d The one who John loves is Mary
68e Mary is the one who John loves
69a John bought a screwdriver
69b What John bought was a screwdriver

In such sentences the participants from a simple

proposition - e.g. that relating to 68a - are distributed into an
equational sentence like 68b. Equational sentences can be
reversed round the copula (68b and 68c), and some of the
conditions for this reversal were discussed on page 317. Simple
cleft sentences, like 65, cannot be reversed. Also note the forms
the one, and what in 68b and 69b: these form a particularly
puzzling feature of pseudo-clefts, since we can also find sentences
like:

70 The one/person/girl who John loves is Mary

71 What/the thing/the tool John bought is a screwdriver
We have discussed four types of operation that produce, or may
produce, themes. The first two - subject selection and promotion
to subject - have the consequence of making some constituent
the grammatical subject. This frequently, but not inevitably,
means that this constituent is also the theme. The third
type - left-movement rules - necessarily produces themes, but
has no effect on the functional relation (subject, object, etc.) of
the thematized constituent. Clefting produces a theme which is
the grammatical subject of the copular verb be but does not
affect the functional relations of the ‘original’ sentence, which
now appears in a construction that resembles a relative clause.
Several syntactic consequences flow from these differences.
Both subject-forming rules and left-movement rules can occur in
the same sentence:

72a Blood flowed in the gutters

72b The gutters, blood flowed in them!
73a I liked the novel, but I was disgusted with the musical
73b The novel pleased me, but the musical disgusted me
73c The novel I liked, but the musical I was disgusted with

Secondly, left-movement rules apply to a wider range of

370 Part three: Functional relations

constituents than subject forming processes. Many constituents

cannot become grammatical subjects, but they can be thematized
by left-movement rules. Thus adverbial expressions:

Stealthily the cat stalked the mouse

74b One dark and stormy night the robbers met

constituents like up in:

75 Up jumped John

or negative adverbs like never ox seldom:

76a Never have I heard such magnificent playing!

76b Seldom have I tasted a more magnificent Stilton!

These particular constructions involve inversion as well as

thematization. Perhaps most strikingly, however, note that
subject-forming processes can only apply within the simple
sentence (as, for example, the passive) or from a constituent
sentence to the matrix sentence (in the case of raising rules).
Movement rules and clefting rules seem to be able to apply,
within ill-defined limits, across sentence boundaries. Thus in:

77a I am anxious that my dog will not keep on wanting to

mate with the mongrel next door
77b It’s the mongrel next door that I am anxious that my
dog will not keep on wanting to mate with

the clefted constituent comes from a deeply embedded sentence:

no subject-forming rule could extract a subject for a higher verb
from so deeply embedded a sentence.
A further difference between subject forming and other
thematization rules concerns the category of definiteness in the
NP. From a statistical point of view, subject expressions are more
likely to be definite than indefinite, not surprising in view of our
previous observations about the structure of given and new
information: given information characteristically precedes new
information. Given information, since it is given,
characteristically involves a definite NP. This is not, of course, to
say that subjects cannot be indefinite NPs, merely that they are
more frequently definite. Themes created by left-movement
rules, on the other hand, are characteristically definite, again not
surprising, in view of the emphatic nature of many such
sentences. Consider subject NPs first. Both of the sentences:
 Theme, rheme and end focus 371
78a My wife gave me that camera
78b That camera was given to me by my wife
seem perfectly acceptable; but of the sentences:
79a My wife gave me a camera
79b A camera was given to me by my wife

the latter seems much less acceptable, though it could be

contextualized. If we now try thematizing the same constituents
by a left-movement rule we find that:
80 That camera my wife gave me
is quite acceptable, but:
81 A camera my wife gave me
seems deviant. This is particularly true with indefinite expressions
like anybody, somebody, etc.:
82a Somebody has stolen my socks
82b * Somebody, he’s stolen my socks
82c * It’s somebody who has stolen my socks
83a Anybody can ride a bicycle
83b * Anybody, he can ride a bicycle

Just as a number of types of rules make some constituent

thematic, so too there are rules which bring a particular
constituent into end-focus. It is not so easy to characterize these
into general types, and many of them turn out to be, as it were,
the obverse of theme-forming rules - clearly, since if some
constituent is chosen as theme, then some other constituent must
be placed in end-focus. As before, we suggest four general types.
Firstly, ‘end-focus selection rules’, the simple obverse of subject
selection rules:
84a The changes in the income-tax laws will benefit the
lower-paid worker
84b The lower-paid worker will benefit from the changes in
the income-tax laws
We do not discuss these further.
Secondly, what we may call ‘postponement rules’:
85a Getting hold of a plumber these days is difficult
85b It’s difficult to get hold of a plumber these days
372 Part three: Functional relations
Many of these rules, like that illustrated, involve changing the
grammatical function of a constituent: thus Getting hold of a
plumber these days is the subject of 85a, but not of 85b.
Our third class of rules we call ‘right movement rules’:

86a Your mother has gone out shopping

86b She’s gone out shopping, your mother

These rules involve moving a particular constituent to the end of a

sentence, but do not change its grammatical function. Finally
pseudo-cleft sentences, discussed on page 369, can be as easily
considered rules producing an end-focus, as they can be
considered rules producing a theme (see examples 68 onwards
on page 367). We do not discuss these again here.
Postponement rules transport a constituent to the right of the
sentence. The most widespread such rule is ‘extraposition’, which
takes a constituent, typically from subject position, and moves it
to a position after the verb, leaving a dummy it subject behind:

87a That you enjoy reading linguistics books is amazing

87b It is amazing that you enjoy reading linguistics books

This operation is obligatory for a number of verbs - thus

88a It happens that I am related to the Russian Royal

Family
88b * That I am related to the Russian Royal Family
happens

Even when it is not obligatory, it is particularly frequent when

the NP involved is a ‘heavy’ NP: i.e. when it is, in a literal sense,
long - often the case when the constituent in question is an
embedded sentence. Thus, while both 87a and 87b are
acceptable, 87b seems stylistically more acceptable. Why this
should be is unclear, but it may have to do with ‘processing’
difficulties in understanding sentences. The point can be
appreciated better if we make the subject constituent of is
amazing even heavier:

89a That you enjoy reading linguistics books even when

you are lying on the beach at St Tropez surrounded by
all those beautiful topless girls is amazing
89b It is amazing that you enjoy . . .

The b sentence is undeniably less difficult! The shifting of heavy

 Theme, rheme and end focus 373
constituents can also be seen in sentences like the following:

90a They presented a copy of the works of Bloomfield

which was hand-printed on parchment and bound in
red morocco to the retiring Professor
90b They presented to the retiring professor a copy of the
works of Bloomfield which was hand-printed on
parchment and bound in red morocco

where the heavy direct object moves to the end of the sentence.
Or again:

91 They presented a copy of the works of Bloomfield to the

retiring professor which was hand-printed on parchment
and bound in red morocco

where a heavy relative clause has been extraposed to the end of

the sentence. (The reader is invited to consider the extent to
which relative clauses can be extraposed, and what conditions
permit or inhibit this.) In all cases the movement seems the
consequence of a heavy constituent.
Right-movement rules typically operate on the subject con¬
stituent, and typically a pronoun copy is left behind:

92 It’s disgusting, soup

Sometimes in addition to the constituent which is dislocated, a

pro-verb copy of the verb is introduced:

93 He jilted her, John did

and frequently constituent and pro-verb are inverted:

94 He jilted her, did John

These processes are available in independent declarative

sentences. We should briefly comment on thematization
processes in non-declarative sentences and in dependent
sentences. It can hardly have failed to strike the reader that one
effect of thematizing operations is to bring some particular
constituent into prominence - this usually involves making it
appear as the topic of the sentence. So it is not surprising that the
most usual thematic element in imperative and interrogative
sentences is not the grammatical subject. Interrogative sentences
are very frequently used to ask questions and in ‘wh questions’ it is
the wh word that is thematic:
374 Part three: Functional relations
95a Who did I see you with last night?
95b Where have you put my slippers?
95c How are you?

The main thrust of a question is to discover the identity of the

participant signalled by the wh word. With wh questions, if the
wh word is not thematic, then, in speech, it usually bears heavy
stress:

96 You put my slippers WHERE?

and is typically no longer a straightforward request for

information, but rather a request for confirmation, an expression
of surprise or the like:

97a A: Where are my slippers?

97b B: I put them in the dustbin
97c a: You put them where?
Note in passing that in terms of the four-way distinction drawn
earlier, the formation of wh questions involves a left-movement
rule: and the wh element can be thematized from a deeply
embedded sentence:

98 Who do you suppose Mary told me she met yesterday?

‘Yes-no’ interrogatives are marked by the thematization of an

auxiliary verb:

99a Didn’t we meet in Marakesh?

99b Have you got a match?

In imperative sentences the thematic element is normally the

main verb, perhaps not surprising if we think that imperative
sentences are characteristically used as commands, and
commands are typically requests for action:

100a Go away!
100b Shut up!

Sometimes a ‘vocative’ element is thematized:

101 Roger, give me a cigarette will you?

presumably to attract the attention of the person being

addressed, or to single him out from among a number of possible
addressees.
 Theme, rheme and end focus 375
Finally we mention thematic processes in subordinate clauses.
Almost invariably subordinating conjunctions are thematized
within their sentence: from a functional point of view, this marks
both the fact and the type of subordination. Thus:

102a If you don’t stop, I’ll scream

102b When Charlie comes home, we’ll have a party
102c John gagged her, so she wouldn’t scream
102d I won’t stop until you ask me nicely

Complementizers occur initially in their complement sentence,

and relative pronouns and other relative markers are initial in
their sentences:

103a I know that my redeemer lives

103b The man who believes that will believe anything

In the same connection, though not with respect to subordinate

sentences, binding elements in general (thus, for example, in
conclusion, etc. see page 149) are usually thematic - again to
mark the nature of the relation between sentences.

This section has looked at a number of syntactic processes

available in the language that secure initial or final sentence
position for some particular constituent. The function of such
operations appears to be tied up with communicative processes of
various kinds. It is not always easy in well-formed text to identify
precisely what effect a given thematization has: it is,
paradoxically, much easier in an ill-formed text to appreciate that
the clumsiness springs, or can spring, from a failure to make an
appropriate thematization. At the end of the chapter is an
exercise based on this premise - to which the reader is invited to
devote his ingenuity after reading the other sections.

Given and new

As was suggested in the introductory section to this chapter, a

principal syntactic consequence of the distribution of infor¬
mation into that which can be held to be given for some
sentences and that which is new to the sentence, lies in the fact
that given information is ‘reduced’ in some way or not mentioned
at all. Reduction processes that can be observed in written text
include most obviously the use of proforms of various kinds - not
376 Part three: Functional relations

only pronouns, but also such proforms as the various forms of

DO. The auxiliary verbs, including the modals, can be used with a
proform type of effect in sequences like:

104a A: Are you going to Alan’s party tonight?

104b B: Must I?
104c A: I think you should

a usage referred to by Palmer, following Firth, as ‘code’ (Palmer

1965). In spoken discourse reduction is even more evident -
given information is typically un-stressed, and may be
phonologically reduced in various ways, many of which are
discussed in Brown (1977).
A lot of given information is not mentioned at all. This is
illustrated in the example above, where the complements of the
modal verbs in 104b, 104c have been elided, and can be
recovered contextually. Elision results in sentence fragments,
which, as we noted (pages 150-2), can only be understood by
reference to ‘fuller’ forms, which can be recovered contextually.
When given information is mentioned, the effect is frequently
contrast of some sort, or emphasis. Thus, for example:

105a a: I’ m going to play squash this afternoon

105b B: Squash! I didn’t know you played
106a A: Would you like a drink? There’s whisky or gin
106b B: I think I’d like a whisky

Topics
It is appropriate to complete this chapter by returning briefly to
the notion of topic. The discussion must be brief since a full
examination of the issues involved leads to a consideration of the
structure of text as a whole, and this is not the place for such a
discussion.
Of all the notions discussed topic is the most difficult to come
to grips with. We have suggested that we can view the topic as
being ‘what the sentence is about’, the ‘perspective from which
the sentence is viewed’, etc. In many simple sentences we can
identify the topic by asking what is the implicit request or
question which the sentence would be an appropriate answer to.
Thus, for example we might identify ‘Edinburgh’ as the topic in a
sentence like:
 Theme, rheme and end focus 377

107 Edinburgh is the most beautiful city in Scotland

if we assume it answers the implicit request:

108 Tell me something about Edinburgh

On the other hand if 107 were to answer the implicit question:

109 Which is the most beautiful city in Scotland?

then the topic would be ‘the most beautiful city in Scotland’. We

have used the phraseology ‘implicit request or question’ since
107 would not be a normal answer to either 108 or 109 if it were
an actual question. In such circumstances we would rather find:

110 It’s the most beautiful city in Scotland

as the answer to 108, and:

111 Edinburgh is

or:

112 It’s Edinburgh

as the answer to 109 - illustrating the process of reduction

consequent on the given new distinction mentioned earlier.
This approach to the identification of topic is all right so far as
it goes. Our observations suggest, correctly, that the notion of
topic cannot be divorced from considerations of given-new (with
the associated grammatical consequences of pronominalization,
elision, etc.) and from processes of thematization. It also
suggests, correctly too, that there is no simple one-to-one
correspondence between topic and theme {Edinburgh in 111 is
theme, but not topic, if 111 is the answer to the implicit question
109). Nor is there any necessary connection between topic and
given, since 107 might be the initial contribution to a
conversation, itself introducing a topic.
The approach seems reasonably satisfactory for simple
exchanges, but it meets with some difficulties when applied to
longer passages in text; some of them relate to the difficulties
experienced in trying to identify a sentence unit (pages 149ff.).
In the first place, a complex sentence can often be analysed as
having either a complex topic, or several topics. Related to this is
the problem of the units within which we wish to identify topics:
for example, one might wish to say of this book that as a whole it
378 Part three: Functional relations
has a topic, defined in general terms, that each part has a topic,
and so on down through chapters, sections, paragraphs, etc. To
regard topic as a notion restricted in its application to a single
sentence is unrealistic, and the technique of asking ‘implicit
questions’ can be applied as well to groups of sentences,
paragraphs etc., as to single sentences. Let us briefly examine
each of these two problems.
Consider first the problem of the identification of a topic in a
complex sentence:

113 In everyday speech ‘fitness’ means suitability or,

adaptedness or being in good condition; ‘evolution’
means gradual change, with the connotation of un¬
folding; and as for ‘inheritance’, we may hope to
inherit money, rights or property; we might inherit
too a mother’s eyes or a grandfather’s gift for fiddling.
114 These are the meanings ... of fitness, evolution,
and inheritance - the meanings for which scientists
chose them when they were struggling to put their
conceptions into words.3

The first orthographic sentence 113 might answer the implicit

question:

115 What do scientists mean by ‘fitness’, ‘evolution’ and

‘inheritance’?

But this seems to introduce three topics, each in turn topic for
one of the succeeding clauses separated by semi-colons. We can
resolve this difficulty by supposing that the sentence could be
dissolved into independent simple sentences (which would
involve some amendment to the text and would destroy its
stylistic characteristics!) each answering one implicit question
(What do scientists mean by fitness? What do scientists mean by
evolution? etc.) We then have the difficulty of identifying an
appropriate topic for the final clause in 113: perhaps it answers
the implicit question, ‘What else can you inherit?’ But this does
not seem helpful, since the function of this sentence is further to
exemplify what is meant by ‘inherit’. The second sentence, 114,
also presents a problem. It seems better to regard both of these

3 The quotations here and in examples 116-118 and 120-121 are from
P. B. Medawar, The Future of Man (1960).
 Theme, rheme and end focus 379

sentences as relating to a single, complex topic: an elementary

initial definition of what is meant by ‘fitness’, ‘evolution’ and
‘inheritance’. The same sort of difficulty arises if we try to
continue this method with the rest of the paragraph:

116 In the course of time those conceptions have become

clearer . . . but the words which embody them have
remained the same. 117 The change that has gone on is
sometimes described by saying that scientists give the
words a new precision and refinement. . . .
118 The idea scientists now have in mind when
they speak of ‘fitness’ can be explained like this. . . .

It is perhaps appropriate to say that the topic of the two

sentences 116 and 117 together answered the implicit question,
‘How have scientists refined the definition of these terms?’ The
next paragraph, the beginning of which is shown in 118, goes on
to offer a contemporary definition of ‘fitness’ that itself extends
over a number of sentences.
Even with declarative sentences in text, then, there are
problems in identifying topics in any straightforward way. Further
difficulties arise if we try to identify the topic of interrogative
sentences. Take the case of 109 for example. It seems hardly
appropriate to ask what question this answers since it is itself a
question. We might say it answers some implicit question like:

119 What question would you like answered?

but this is tautologous. A more appropriate approach is to

suggest that instead of ‘having a topic’, overt interrogative
sentences like 109 ‘introduce a topic’. This certainly seems the
function of the ‘rhetorical question’ often found in expository
texts:
120 ‘Inheritance ’ was the second of the three words of which I
said that biologists use in special or unfamiliar ways. 121
Just what is inherited when geneticists speak of
inheritance?

120, a declarative sentence, reintroduces as a topic something

previously mentioned, and 121 establishes that the author is
about to define what he means by this topic. The author is using
the ‘implicit question’ technique we started this section with as
an explicit technique for defining a topic - it then hardly seems
380 Part three: Functional relations
sensible to enquire what the topic of this topic-introducing
question is!
This approach to the identification of topics leads to the
question of what unit it is within which we wish to identify topics;
it is suggested above that it is appropriate to consider there to be
a hierarchy of such units. The hierarchy starts, if this is
appropriate, with the clauses of which a complex sentence
consists (see discussion of sentence 113), goes through the
orthographic sentence if this is relevant, and it may not always be
so (see discussion of sentence 114), to groups of sentences, then to
the paragraph, and so on. If this approach is appropriate it will be
noted that the orthographic sentence may not necessarily be a
unit that needs to be attended to, though it generally will be.
To illustrate what is meant by groups of sentences serving as a
unit for the identification of topics, consider for instance the
structure of the second paragraph in this section. The first
sentence is introductory, the second offers an elementary
definition, the third suggests a technique of identification, the
forth and fifth exemplify this technique, and so on. The implicit
question technique works well on this paragraph, providing it is not
applied to each individual sentence, but rather to groups of
sentences as appropriate:

What is this section about? (SI)

What do you mean by ‘topic’? (S2)
How would I identify one? (S3)
Can you give me an example? (S4, 5)

and so on. Note that this set of questions refers as much to what
we might call the rhetorical structure of the paragraph as to the
structure of individual sentences. Insofar as it relates to rhetorical
structure, it helps account for a number of facets of sentence
construction, notably the distribution of the binding expressions
noted on pages 149-50.
This approach to the identification of topics may assist in
untangling some of the mysteries of word order. We have already
noted that what is topic is frequently also what is thematic, but
this statement needs to be qualified. The thematic element may
relate to the topic of some unit larger than the sentence (as is
perhaps the case when binding expressions are thematic - they
often identify the function of the sentence (for instance, thus, in
conclusion, etc.)). Equally the thematic element may have
 Theme, rheme and end focus 381
nothing to do with the topic, but be some other element of the
message that the speaker wishes to foreground.

Concluding remarks

The phenomena considered in this chapter pose a peculiar

difficulty for constituent structure grammars. Much of it,
particularly the observations relating to theme and end-focus,
clearly has to do with constituent structure, but some of it,
particularly that relating to topic, while it may have implications
for constituent structure, cannot usefully be brought into a
constituent structure grammar of any sort that we have discussed.
Even the material clearly relevant to constituent structure is of a
curious sort. Some of it, particularly the thematization operations
that have to do with subject selection, seems relevant to what we
may call centrally grammatical processes - processes like subject
selection with verbs like RUN, STREAM (page 364) and raising
(pages 344-5). Other operations, particularly those like left-
movement rules (page 367ff.), seem outside the centrally
grammatical processes and have more to do with communicative
effects in texts. Yet it is not as simple as that. We have noted
(page 374) that question formation resembles a left-movement
rule (in that it can take a constituent embedded deeply in the
sentence and thematize it), and we surely want to think of
question formation as a centrally grammatical process.
Conversely the passive operation, often thought of as a central
grammatical rule (transformational grammars often treat it as the
first transformation to be taught) seems to have as much to do
with communicative effects as anything else (see the very
perceptive remarks on the functions of passivization in Jespersen
(1924).
One might think it convenient to be able to distinguish the
centrally grammatical rules, whatever they may be defined to be,
from those rules that are concerned with text-forming processes,
but this is clearly impossible. Nor should this surprise us.

Technical terms
cleft sentence comment left dislocation
given left-movement rules
‘heavy’ NPs new
382 Part three: Functional relations

pseudo-cleft sentence thematically marked/unmarked

right-movement rules thematization
subject selection rules theme
postponement rules topic
promotion to subject rule

Exercise
Each of the following two texts contains much the same
propositional matter; the propositions follow each other in the same
order. Each text too exhibits thematization devices of various sorts.
Identify the types of thematization operations involved. Account
for the fact that while the text A seems to be fairly coherently
structured, text B is decidedly odd.

1 It was in 1960 that the cave was discovered. 2 An old shepherd

thought he heard ‘water running underground’ at the foot of a
near-by mountain. 3 He told a friend of his who lived in the
near-by village of Petralona. 4 The villager removed a few stones
and literally fell into the cave. 5 A quick search showed that the
floor was littered with animal bones. 6 What he had found was a
prehistoric habitation site. 7 On a subsequent visit he found an
entire human fossil, lying on its right side with its legs drawn up.
8 He told other villagers that night that he had found the remains
of a ‘big monkey’ in a cave.

1 What was discovered in 1960 was the cave. 2 A shepherd who

was old thought that ‘water running underground’ had been
heard by him at a near-by mountain’s foot. 3 A friend of his who
lived in the nearby village of Petralona was told by him. 4 A few
loose stones were removed by the villager and it was the cave he
literally fell into. 5 That animal bones littered the floor was
shown by a quick search. 6 He had found a prehistoric habitation
site. 7 Lying on its right side with its legs drawn up, he found an
entire human fossil on a subsequent visit. 8 He told other
villagers that night that the remains of a ‘big monkey’ had been
found by him in a cave.
Postscript

Part one of this book was devoted to a consideration of

constituent structure grammars. We paid particular attention to
this way of describing the structure of language because it has
been the dominant approach since the 1930s, first with the
‘structuralist’ grammarians and more recently with the
‘transformationalist’ grammarians. Part three looked at some
other equally important aspects of the structure of sentences -
the function that particular constituents have within the sentence.
This has traditionally been the concern of grammarians, and in
recent years interest in this area has revived. We have been led to
enquire whether the sorts of matters discussed in Part three can
be accommodated within a constituent structure grammar of the
sort developed in Part one: in many cases it cannot without the
addition of new descriptive machinery.
The reason for this is clear. The elements used in the
construction of constituent structure grammars are distribu-
tionally defined form classes and construction types. There
is no simple way of introducing considerations of syntactic
function into such a grammar - whether the function at issue is
dependency, participant type, grammatical relation or textual.
Such functions are relational: they specify the type of relation
between two or more constituents. To take an example: we have
identified, for English, the prepositional phrase as a particular
construction type, with the constituents Preposition + Noun
Phrase. How should we now identify the functions of such a
construction in a particular sentence? Prepositional phrases serve
many different functions: they may be adverbial modifiers (a
dependency relation); they may be nuclear or circumstantial
participants (a participant role relation); they may be oblique
objects (a grammatical relation); or they may be themes (a
textual function). They may be, and they typically are, more than
one of these simultaneously: an adverbial modifier is typically
384 Postscript

also a circumstantial participant; an oblique object may be a

nuclear participant; and so on. A functional characterization
of a particular constituent and its constituent structure
characterization interact in many complex ways. Part one
attempts to include some functional information in our grammar
by enriching the structure, as for example by the introduction of
the node labelled Pred (page 51), or by smuggling in function¬
ally labelled nodes, like adverb of place (page 73). But we
cannot accommodate all the functional information we might
need in this way without considerable modifications to the
straightforward constituent structure grammars we started off
with. We saw (page 259) that some of the earlier approaches to
constituent structure grammars included some functional
information - by labelling the branching structures in various
ways, or by adding notes commenting on the constituent
structures. These ways of describing functional relations were not
perpetuated in the main stream of linguistic description, perhaps
because they could not be formalized in the way constituent
structure itself can, and because many linguists became interested
in the formal and mathematical properties of constituent
structure grammars, and later transformations, themselves. More
recently attempts have been made to re-introduce this functional
information, either directly into grammars that look rather like
constituent structure grammars (see, for example, the sort of
grammar described briefly on page 323ff.), or by enriching the
lexicon (see, for example, pages 321ff.).
A more radical departure from constituent structure grammars
is found in the grammatical theories known as ‘Tagmemic’ and
‘Systemic’ grammar. It is not possible to do more than mention
these alternative approaches here, and the reader is referred to the
Further reading on page 386. We may, however, mention one
feature. Most of the basic notions of distribution, form classes,
constructions, and so on that we discussed in Part one are used in
these grammars too, though sometimes in a slightly different way.
One way they differ is that each structural position within the
Sentence, the Noun Phrase, etc., is identified both in terms of its
function and of its constituent structure realization. So in a
sentence like The sad girl wept, the constituent the sad girl is
labelled both as a subject (of some particular kind) and as being
realized as an NP. The constituents of the NP are then labelled
as a head, realized by a noun, girl, preceded by two modifiers,
 Postscript 385

each of a different type and realized by items of a particular form

class. Grammatical function and grammatical form are described
simultaneously. This approach may, and in systemic grammar
does, involve the recognition of several different layers of
structure simultaneously; the structures not necessarily being
coincident. Such a description does not lend itself to
representation in the form of a tree diagram.
The problem lies in the fact that language serves a variety of
functions simultaneously. The functions are all closely interknit
and relate in a variety of ways to the actual forms of language.
We have chosen to look at the forms of language first and then
consider functions because we believe this to be the most readily
comprehensible introduction to the complex structure of
language.
Further reading

Introduction
We have not discussed many issues concerned with the nature of
language in general; the reader is directed to Sapir (1921),
Jespersen (1922) and (1929), Bloomfield (1935), Langacker
(1968), Nash (1971) and the introductory chapters of books like
Lyons (1968), Hockett (1958).

Part one
The study of constituent structure (Chapter 1) and of form
classes (Chapter 2) is so basic to any approach to syntax that all
introductory books discuss it. Bolinger (1968), Gleason (1969),
Hockett (1958), Lyons (1968), Palmer (1971), and Robins
(1964) - consult the indexes for the relevant pages - are among
those we have found the most useful. Each has a slightly different
presentation of essentially the same view of constituent structure
that we take in this book. Householder (1972) is a useful
collection of articles, many of which bear on these questions.
While the view of constituent structure expressed in this book is
representative of the views most commonly held in contemporary
linguistics, there are ‘schools’ of linguistics that take a somewhat
different position: we may mention three - ‘Tagmemics’ (see
Cook (1969), Longacre (1964)); ‘Scale and Category grammar’
(see Sinclair (1972)); ‘Systemic grammar’ (see Berry (1975), Muir
(1972)).
Lyons (1968) contains a good, though sometimes difficult,
account of the matters discussed in Chapters 3 and 5-7. The most
interesting discussion of these questions is to be found in works
on Transformational Generative grammar. As noted in the
introduction, this book does not present a fully formalized
account of such a grammar, though many of the ideas it presents
 Further reading 387

derive from work on transformational grammar. The companion

volume (Brown and Miller, in press) presents a formal account of
this approach. The transformational approach derives from the
work of Chomsky (1957). A useful overview is Lyons (1970).
Introductions we find helpful include Akmajian and Heny (1975);
Huddleston (1976) and Koutsoudas (1966).
Problems concerning the sentence (Chapter 10) are discussed
in the introductory books mentioned in the first paragraph. The
view presented here is much influenced by Lyons (1977).

Part two
The view of morphology presented here derives in the main from
Lyons (1968). A comprehensive and useful discussion is found in
Matthews (1974). The introductory books noted in Part one can
be consulted with profit, but the terminology differs to some
extent from author to author.
Form classes and grammatical categories (Chapter 16) are
discussed in the introductory books. A useful discussion of form
classes in a number of different languages is in Lingua (1967).

Part three
For the subjects discussed in Chapter 17 see particularly Hockett
(1958), Lyons (1968), Palmer (1971), and Robins (1964).
On ‘Processes and participants’ (Chapter 18) see particularly
Brown and Miller (in press), Fillmore (1968), Halliday (1970).
Liefrinck (1973) contains an interesting discussion of some
problems in this area. The works on Tagmemics cited earlier
draw together notions of constituent structure and process and
participants; a comprehensive account is Longacre (1976).
For the matters discussed in Chapters 19 and 20 an interesting
non-technical discussion is Jespersen (1929). Halliday (1970) is
also very approachable. There is some discussion in Lyons
(1968).

Exercises
The following contain useful collection of exercise materials:
Gleason (19556), Langacker (1972), and Nida (1971).
References

AKMAJIAN, A., and HENY, F. (1975), An Introduction to the

Principles of Transformational Syntax, MIT Press

BERRY, M. (1975),Introduction to Systemic Linguistics, Batsford

BLOOMFIELD, L (1935), Language, Allen & Unwin
BOLINGER, D (1968), Aspects of Language, Harcourt & Brace
BOLINGER, D (1971), The Phrasal Verb in English, Harvard
University Press
BROWN, E. K., and MILLER, J. E. (in press), Syntax: Generative
Grammar, Hutchinson
BROWN, G. (1977), Listening to Spoken English, Longman

CHOMSKY, N, (1957), Syntactic Structures, Mouton

COLE, P., and SADOCK, J. M. (eds.) (1977) Syntax and
Semantics, vol. 8: Grammatical Relations, Academic Press
COOK, w. A. (1969), Introduction to Tagmemic Analysis, Holt,
Rinehart & Wilson
CRYSTAL, D. (1966), ‘Specification and English tenses’, Journal
of Linguistics, vol. 2, 1-34

Fillmore, C.J. (1968), ‘The case for case’, in E. Bach and

R. T. Harms (eds.), Universals in Linguistic Theory, Holt,
Rinehart & Winston
FRIES, C. C. (1952), The Structure of English, Harcourt Brace

GLEASON, H. A. (1955), Workbook in Descriptive Linguistics,

Holt, Rinehart & Winston
GLEASON, H. A. (1969), An Introduction to Descriptive
Linguistics, Holt, Rinehart & Winston
GREENBERG, J. H. (ed.) (1963), Universals of Language, MIT
Press
 References 389

HALLIDAY, M. A. K. (1970), ‘Language structure and language

function’, in Lyons, J. (ed.), New Horizons in Linguistics,
Pelican
HOCKETT, C. F. (1958), A Course in Modern Linguistics,
Macmillan
HOUSEHOLDER, F. w. (1972), Syntactic Theory I, Penguin
HUDDLESTON, R. (1976) An Introduction to English
Transformational Syntax, Longman

JESPERSEN, O. (1922), Language, Its Nature, Development and

Origin, Allen & Unwin
JESPERSEN, O. (1929), Philosophy of Grammar, Allen &
Unwin
JESPERSEN, O. (1961) (reprint), A Modern English Grammar,
Allen & Unwin

KOUTSOUDAS, A. (1966), Writing Transformational Grammars,

McGraw-Hill

LANGACKER, R.W. (1968), Language and its Structure,

Harcourt Brace
LANGACKER, R. w. (1972), Fundamentals of Linguistic
Analysis, Harcourt Brace
LANGACKER, R. W. (1969), ‘On pronominalization and the
chain of command’ in D. A. Reibel and S. Schane, Modern
Studies in English, Prentice Hall
LEECH, G. (1971), Meaning and the English Verb, Longman
LIEFRINK, F. (1973), Semantico-Syntax, Longman
LINGUA (1967), Word Classes, North Holland
LONGACRE, R.E. (1964), Grammar Discovery Procedures,
Mouton
LONGACRE, R. E. (1976), An Anatomy of Speech Notions,
Peter de Ridder Press
LYONS, J. (1968), An Introduction to Theoretical Linguistics,
Cambridge University Press
LYONS, J. (1970), Chomsky, Fontana
LYONS, J. (1977), Semantics, vol. 2, Cambridge University Press

MACKINNON, R. (1971), Gaelic, Teach Yourself Books

MATTHEWS, P. H. (1974), Morphology, Cambridge University
Press
390 References
MUIR, J. (1972), A Modern Approach to English Grammar,
Batsford

NASH, w. (1971), Our Experience of Language, St Martin’s Press

NIDA. E. (1966), A Synopsis of English Syntax, Mouton
NIDA, E. (1971), Morphology, University of Michigan

PALMER, F. R, (1965), A Linguistic Study of the English Verb,

Longman
PALMER, F. R. (1971), Grammar, Penguin
partridge, (1958), Origins, a Short Etymological
E.
Dictionary of Modern English, Routledge & Kegan Paul

QUIRK, R., GREENBAUM, S., LEECH, G., and SVARTVIK, J.

(1972), A Grammar of Contemporary English, Longman

ROBINS, R. H. (1964), General Linguistics, an Introductory

Survey, Longman
ROBINS, R. H. (1967), A Short History of Linguistics, Longman

SAPIR, E. (1921), Language, an Introduction to the Study of

Speech, Harcourt Brace
SINCLAIR, J. MCH.
(1972), A Course in Spoken
English-Grammar, Oxford University Press
STRANG, B. M. H. (1970), A History of English, Methuen

THOMAS, L. V. (1967), Elementary Turkish, Harvard University

Press
Index

action (verb), 238, 288 ff. 72; square, 118-19, 200

active (v. passive), 102—1, 112, branching, 28
121, 289, 339
adjective, 48-9, 50-1, 61-2, 76-9,
143-4, 223, 236—10, 316, category, 173-4, 231-2; closed,
348-9; attributive, 236; 231; formal, 60 ff., 234, 241;
predicative, 50-1, 236, 348-9; grammatical, 97, 174-5, 187-8,
shift rule, 143 191, 231-2, 240 ff.; lexical, 173,
adjunct, 353—4 231; open, 231
adverb, 70-5, 128-9, 145, 288, causative, 289 ff., 319-20, 364
353 class: changing, 226; maintaining,
affix, 177 226
affix hopping, 208 clause, relative, 137-47
agent, 102, 289 ff., 338 command, 274
agglutinative rule, 189 complement, 52, 136, 348-53
agreement, see concord complementizer, 134, 153, 348
Akan, 60, 68, 265-6, 280, 299 component, base, 124;
allomorph, 178 transformational, 123-4
ambiguity, 81-3, 145 concord, 192-203, 211, 257,
analytic (v. ‘synthetic’), 299-302, 278-80; controller of, 195, 258,
352-3, 367 278
article, 26, 33-1, 90-1, 198-200, conditioning, grammatical, 180,
232 184; phonological, 178
aspect, 105, 206, 247 conjunction, subordinating, 153
attribute, 298, 318-19, 348, 350 constituent, 21, 28, 40; immediate,
auxiliary, 105 ff., 207-14; modal, 28; optional, 56, 69-83, 257;
106 ff., 210, 246 see also structure
construction, 28; exocentric, 255
back formation, 225 content, level of, 10
base, 177 co-ordinating conjunctions, 154
benefactive, 293, 344 copula, see verb, copular
bracketing, 26-7 co-reference, 274-7, 334
brackets: curly, 41-2; round, 56, correctness, 14-17
392 Index

count (v. mass), 88, 90-1, 96-7, feature, 86 ff., 289 ff.; syntactic,
242-3 91, 245
covert (v. overt), 240-1 form, bound, 176, 227, 232
creativity, 146 form, free, 176, 226, 232;
cross-classification, 55, 88 minimal, 164—5
cross-reference, 274-7, 281-4, form, word, 166-7, 176
336-7; pronominal, 274-8, form class, 32-9, 60, 231-49, 253
280—4 formation rules: text, 152; word,
cumulation, 210 223 ff.
cycle, 76, 136 formative, lexical, 222
fusion, 189
dative, 304 ff., 338; movement of,
130, 343-7 Gaelic, 62 ff., 232, 266-8, 286-7
daughter, 28 gender, 98, 241, 279, 280
declarative, 112 given, 359-60, 375-6
definite (v. indefinite), 199 goal, 304, 339, 344
deictic reference, 278, 282, 336 government, 256
dependency, 254-60; bi-lateral, grammar: case, 322; formal, 40-7,
255-6; co-ordinate, 258; 60-6; generative, 45;
mutual, 255-6; relations, 154, transformational, 122-9
254-60; unilateral, 257 grammaticality, 45
derivation, 43, 121-3
distribution, 23 ff., 34, 40, 60, head of construction, 137, 237,
164, 233-5 254
do support rule, 116, 127-8 heavy NP. 372-3
domination, 28 hierarchies, 88, 320, 322, 339,
341, 354; see also structure
elision, see ellipsis homography, 166
ellipsis, 151, 237, 243, 336,
359-60, 376 identification, 316, 348-50
embedding, 30, 134-46 imperative, 128, 326, 374
emphasis, 128, 358-9 incorporation, 299
end-focus, 358, 371-3 inchoative, 295 ff., 311, 315
endocentric, 257 infix, 177
entry, lexical, 22, 40-1, 55, 89, insertion, lexical, 43, 94, 96, 97
136, 321 instrument, 139, 293, 322, 338
environment, 34 intensifier, 79, 236
etymeme, 228 interrogative, 112-16, 212-13,
etymology, 228 373-4
exclusion, 258
expansion, 23, 41 labelling, 27, 33
expression: binding, 150, 156, language, figurative, 86
375; level of, 10 left-movement rules, 364, 367-8,
extraposition, 372 369-70
 Index 393

lexeme, 166, 174, 231 parse, 21

lexicalization, 43, 94, 96 participants, 288-328
lexicon, 40-3, 55, 88-90, 97-8, participle, past, 106, 107
222 ff„ 244-5 particle, 131, 140, 347
linkage, 274-84 passive, 102-4, 107-8, 112,
linking verb, see verb, copular 125-6, 325, 346-7, 365-6
locative, 307-14, 338, 349 pause, potential, 163
Luganda, 215-17, 280-4, 285-6 perfect (aspect), 106, 207
personification, 86
marker, 153, 255, 260-73 phonaestheme, 229
mass (v. count), see count phonology, 11
modifier, 71, 227, 237, 254 possessive, 254, 263, 303, 350
morph, 169 ff., 176-82 postponement rules, 371-2
morphemes, 161-72, 173-6; predicate, 51, 255
grammatical, 174, 231; see also prefix, 177
lexeme preposition, 53, 233, 256, 262-3
morphology, 12, 161 ff.; process, 288-328
complexity in, 176, 222; productivity, 224
derivational, 171, 177, 222-30; pro-form, 150, 375-6
inflectional, 171, 222; simplicity progressive (aspect), 107, 207
in, 176 promotion to subject rules, 363,
365-6
negative, 116-19, 212-13 proper (v. common), 37-8, 88-9
nesting, see embedding proposition, 288
new, 359-60, 375-6 pseudo-clefting, 368
node, 28
notional (v. formal), 60 ff., 234, raising, 334-5, 355, 366
241 realization, 168, 182—4, 189;
noun, 26, 33-4, 233-5; collective, discontinuous, 207-8, 257
51, 244; phrase, 23-5, 48, recategorization, 227, 237, 243
137 ff., 234, 254 recursion, 76, 134^16
nucleus, 71-2, 288, 353 reflexive, 277, 312
null realization rule, 190 relations, paradigmatic and
number, 96, 174, 186 ff., 240-2 syntagmatic, 253-84
result, 290 ff., 340, 348
object, 29, 53 ff., 103, 340-43; rheme, 358
indirect, 53, 342-8; oblique, right-movement rules, 373
342 role, 317; circumstantial, 288
optional rule, 125, 257 root, 177
order, 205-15, 260-73; word, rules (of language): constituent
260-73 structure, 40 ff.; lexical, 43, 55;
order class, 49, 262 lexical insertion, 43, 56, 94, 96;
ordering, of rules, 184, 192 phase structure, 44; recursive,
overt (v. covert), 240-1 76; rewrite, 41;
394 Index

rules-contd tags, 128, 132-3

subclassification, 91; tense, 105, 109-10, 168-9,
transformational, 44 195 ff., 207, 246 ff.
term (in category), 174
theme, 357, 362 ff.
scope, 72, 80, 258 time adverb fronting, 129
selection, 85-101, 144, 174 topic, 360, 376-9
sentence, 12-13, 22-3, 119-20, transformational rule, 44, 102 ff.,
149-56; basic, 72, 119-21; cleft, 125, 189
364, 368-9; complex, 152-6; tree, 27, 44
co-ordinate, 154-5; existential, Turkish, 268-9
332, 335; fragment of, 150;
matrix, 124; see also embedding ungrammaticality, 23, 45
sister, 28 universals, 273
speech, parts of, 232 ff. utterance, 12-13
stative, 238-9, 248, 288 ff., 348
stem, 177, 228; lexical, 226
variation, free (of allomorphs),
string, 21, 25; constituent analysis
180
of, 29
verb, 25, 33,232,238 ff.; classes of,
structure: analysis of, 125; change
47, 50-7; copular, 48, 50, 69,
in, 125; constituent, 21 ff.;
349; di-transitive, 53; dummy,
derived, 124-5; hierarchical, 28;
116, 239; group, 105;
internal, 23-4, 34, 164, 233;
intransitive, 37, 48, 51;
surface, 124; underlying, 124,
intransitive locative, 53; main,
137, 333-4
103, 105, 110; phrasal, 131,
subcategorization: inherent, 56,
140, 347; transitive, 37, 48, 52;
85; strict, 55-6, 71, 85, 136,
transitive locative, 54
258
vowel harmony, 163
subclassification, 47, 69, 91
subject, 29, 52, 103, 330-9, 357;
dummy, 372; grammatical, word, 14, 22, 161-72; classes of,
331-6; logical, 339; 33 ff.; compound, 162, 226;
psychological, 357, 360; form, 232; full, 232;
selection rules of, 363-71 grammatical, 175, 232; lexical,
substitution, 23, 32, 63; table, 32, 175, 232; morphosyntactic, 167,
64 232
suffix, 177
symbol: boundary, 38; constituent, Xhosa, 219-22
21 ff.; initial, 43; non-terminal,
43; terminal, 43, 173 zero, 181—2, 190
synthetic, see analytic zero morph, 180-2