Sei sulla pagina 1di 217



edenda curat








M O U T O N & CO.
© Copyright Mouton & Co., Publishers, The Hague, The Netherlands.

No part of this book may be translated or reproduced in any form, by print, photoprint, microfilm,
or any other means, without written permission from the publishers.

Second Printing 1985

ISBN 90 2790669 6

Printed on acid free paper

Printing: Kupijai & Prochnow, Berlin. - Binding: Dieter Mikolai, Berlin.

Printed in Germany
To Gunilla, Agneta, and Jesper

Voice in grammar is an aspect of English syntax which in recent years has attracted
considerable interest in discussions of linguistic theory. This book is not primarily
intended as a theoretical contribution (which of course does not exclude the possi-
bility that it might be used to such ends); it is a corpus-based discussion of some
grammatical categories that seem relevant to problems connected with voice in
English. There are, at the present time, diverse views on the value of a corpus.
Here, let it suffice to mention two reasons for using a corpus in the present inquiry:
firstly, one of the aims of this monograph is to describe the use of voice, in particular
the passive, in some varieties of present-day English; secondly, it is maintained that
corpus-studies will help to provide descriptively more adequate grammars. It is
interesting to note that in his latest book (which unfortunately arrived too late to be
discussed in the main body of this study) Noam Chomsky comments: 'Perhaps the
day will come when the kinds of data that we now can obtain in abundance will be
insufficient to resolve deeper questions concerning the structure of language' (Aspects
of the Theory of Syntax, Cambridge, Mass., 1965, 21). However far away the day is
when corpus-data are needed, it seems likely that, to get nearer our goal of under-
standing language, we need a multi-pronged approach.
Although presented as a Ph. D. thesis to Uppsala University, this book is essentially
a British product: it was conceived in Durham, and the work took place chiefly in
London where, for four years, I had the benefit of doing research in the stimulating
atmosphere of University College. · Most of the research was carried out under the
auspices of the Survey of English Usage, and I owe a very special debt to its Director,
Professor Randolph Quirk — for suggesting voice as a suitable field of study, for
helping to make available the necessary technical facilities, for setting an excellent
(albeit inimitable) example with his zest for work, and, above all, for providing
compelling stimulus and acute criticism at all stages of the work.
In spite of my long absence from Sweden, I have had the benefit of close connexion
with many scholars there. I am particularly indebted to the following Uppsala
professors: to Erik Tengstrand for guiding my first steps in English studies; to
H. W. Donner for constant support and encouragement; and to Johannes Söderlind
for supervising the final stages of my work. All of them have contributed to producing
a more finished product by making valuable suggestions.

The often very laborious tasks involved in this work were made lighter by the happy
camaraderie among Survey colleagues, who also contributed in many other ways.
In particular, 1 am obliged to Mr Henry Carvell, who lavishly and patiently provided
computational and statistical knowledge. Among other London colleagues who
have helped by criticism, discussion, informant responses, etc. are Mrs Judith Carvell,
Mr Derek Davy, Mr Norman Fairclough, Mr Sidney Greenbaum, Mrs Joan
Huddleston, and Mr Geoffrey N. Leech. My thanks are further due to Dr Sven
Jacobson for stimulating linguistic arguments and to Miss Berit Hallberg, Miss Ann
Helm, and Mr Bengt Odenstedt for assistance with proof-reading.
The research behind this book required both man and machine, and 1 have depended
heavily on the computational expertise of Mr J. C. Gower, Rothamsted Experimental
Station, Harpenden, Herts., and Mr A. J. Colin, University of London Institute of
Computer Science.
Finally, 1 want to thank my wife who has always taken an active interest in my
work, manifested by typing cards, providing consolation after computer breakdowns,
and in innumerable other ways.

Göteborg, February, 1966.





1.1 Views on voice 1

1.2 Mode of analysis 3
1.3 Using a corpus 5
1.4 Material 6
1.5 Aims 8
2.1 Simple and complex groups 10
2.2 Four types of complex groups 10
2.3 Type a 12
2.31 Criteria 12
2.32 Matrix 13
2.33 Classification 14
2.34 Concord valency classes 16
2.4 Type b 17
2.5 Type c 17
2.51 Criteria 18
2.52 Matrix 18
2.53 Classification 19
2.6 Type d 19
2.7 Lexical verbs 19
2.71 Phrasal and prepositional verbs 19
2.72 Prepositional verbs and prepositional phrases 20
2.73 Compounds 21
2.74 Verbal and nonverbal bases 22
3.1 Clause elements 25
3.2 Exponent classes of subject and complement 26

3.21 Form-class 26
3.22 Gender 27
3.23 Finitude 28
3.24 Person 28
3.25 Number 28
3.26 Modification 28
3.27 Coordination 28
3.3 Exponent classes of adjunct 29
3.31 Form-class 29
3.32 Position 30
3.33 Agent 30
3.4 Internal clause relation 31
3.5 External clause relation 31
3.6 Central and peripheral clause types 33
3.7 Extensive and intensive clauses 33
3.71 Copula 34
3.72 Gender/number class 34
3.73 Gender-selection 34
3.74 Voice transformation potential 35
3.75 Exponence 35
3.76 Deletion 35
3.8 Major clause types 36
3.9 Values of clause element C 37


4.1 Procedure 39
4.11 Corpus 40
4.12 Data 41
4.13 Input and output 41
4.14 Programs 41
4.15 Computation 43
4.16 Results 43
4.2 Type of finite verbal group 43
4.3 Exponent classes of subject and agent (S and Ag elements) 49
4.31 Subject gender 50
4.32 Agent gender 51
4.33 Subject form-class 52
4.34 Agent form-class 52
4.35 Coordination 52
4.36 Length of passive subject and agent 52
4.4 Exponent classes of complement (C element) 55
4.41 Complement distribution 55

4.42 Complement form-class 55

4.43 Complement gender 57
4.5 Exponent classes of adjunct (A element) 57
4.51 Adjunct frequency 58
4.52 Adjunct position 58
4.53 Adjunct form-class 59
4.6 Clause types 62
4.7 External clause relation 65
4.8 Internal clause relation 69
4.9 Summary of the results 69
4.91 Structure and frequency 70
4.92 Style 70
4.93 Voicu 71


5.1 Numerical taxonomy 72
5.2 The Taxonomic Passive Corpus 73
5.3 The criteria 74
5.4 Criteria relating to the clause 78
5.41 External clause relation 78
5.42 External subject relation 81
5.43 Internal clause relation 82
5.44 Active transformation potential 83
5.45 Permutation and transmutation 88
5.5 Criteria relating to the passive verbal group 91
5.51 The auxiliary 91
5.52 The lexical verb 95
5.6 Criteria relating to the subject 97
5.7 Criteria relating to the complement 101
5.8 Criteria relating to the adjunct 101
5.81 Adjuncts with agentive function 102
5.82 Adjuncts with nonagentive functions 109


6.1 Preparation of input data Ill
6.2 The classification program 113
6.3 Output analysis 115
6.31 Taxonomic analysis 115
6.32 Statistical assessment and linguistic interpretation 116
6.4 The diagnostic key 132
6.41 Classes α and β 132
6.42 Class β/γ 133
6.43 Class γ 134

6.44 Class δ 134

6.45 Class ε 135
6.46 Class ζ 137
6.5 The passive scale and voice relation 138


7.1 Five passive clause types 139
7.2 Three major passive classes 141
7.21 Agentive passives 141
7.22 Quasi-agentive passives 147
7.23 Nonagentive passives 148
7.3 Verbal group structure 150
7.4 The use of the passive voice in the texts 152

8.1 The concept o f ' T h e passive scale' 156
8.2 Transformational and serial voice relations 159
8.3 The description of the English passive 162

APPENDICES: Lexical verbs of the agentive clauses in the Major Passive Corpus 167



2:1 Verbal group types and some of their combinations 11

2:2 Matrix for Type a 13
2:3 Concord valency classes for some Type ο auxiliaries 17
2:4 Matrix for Type c 18
2:5 Diagnostic frames for simple, phrasal, and prepositional verbs . . . . 21

3:1 Gender classes 28

3:2 Major clause types (constituent structure) 36
3:3 Major clause types (transformational structure) 36

4:1 Distribution of verbal group types in the Voice Corpus 44

4:2 Voice distribution for the texts of the Voice Corpus 46
4:3 Distribution of verbal group types in the Voice Corpus (active only) and
in the Minor Passive Corpus 47
4:4 Contingency table for Types c/C and djD in the Voice Corpus . . . . 46
4:5 Occurrences of Type a/A auxiliaries 49
4:6 Voice and subject gender in the Voice Corpus 50
4:7a-c Voice and subject gender in Texts M l , M2, J1 51
4:8 Subject gender in the Minor Passive Corpus 51
4:9 Number of words in subjects and agents of the Minor Passive Corpus . 56
4:10 Distribution of complements 56
4:11 Form-class of simple complements (C 1 ) 57
4:12 Adjunct items, sets, and positions 59
4:13 Major clause types in the Voice Corpus 62
4:14 External clause relation 67
4:15 Internal clause relation 69

6:1-30 Contingency tables for categories in the Taxonomic Passive Corpus 118-131

7:1 The use of clause types in agentive passives 143

7:2 The use of verbal group types in agentive passives 150
7:3 The use of agentive passives in 28 texts 153
7:4 The use of passives in eight text sets 155

1:1 The material 7

2:1 Word elements and bases 23

4:1 Procedure of the voice data processing experiment 39

4:2 Verbal group types in the Voice Corpus 45
4:3 Verbal group types: active (Voice Corpus) and passive (Minor Passive
Corpus) 48
4:4 Subject-lengths in Class γ 53
4:5a, b Subject- and agent-lengths in Class α-β 54-5
4:6 A set distribution 60
4:7 Positions of A elements 61
4:8 Form-class of A elements 63
4:9 Major clause types in the Voice Corpus 64
4:10 C and A elements in Class γ 66
4:11 External clause relation 68

5:1 Criteria relating to the clause 75

5:2 Criteria relating to the passive verbal group 76
5:3 Criteria relating to the subject 77
5:4 Criteria relating to the complement 78
5:5 Criteria relating to the adjunct 79

6:1 Specimen of data input (OTUs 1-4) 113

6:2 Specimen of output: top part of the full similarity matrix 114
6:3 Non-proportional, shaded matrix representing mean similarities be-
tween and within the groups 115
6:4 Diagnostic key for finite passive clauses 117

7:1 Frequencies of the five passive clause types in the agentive classes. . . 142
7:2 Frequencies of the agentive passive classes in eight text sets . . . . . 154

8:1 Diagrammatic representation of some voice relationships 163




The study of the grammatical category voice has enjoyed considerable popularity
over the last few years.1 In fact, voice has probably received greater attention from
linguists than ever before in the history of English scholarship. Its sudden appearance
in the grammatical limelight can be attributed largely to the advent of transforma-
tional grammatical theory, where the active-passive relation has been used as a
prime illustration of the supremacy of the transformational model. 2 In his grammar,
Chomsky derives passive sentences from kernel active sentences: 'For every sentence
JVPj— V— NP2 we can have a corresponding sentence NPi-is-\-Ven—by-\-NP1.'
'Thus every sentence of the language will either belong to the kernel or will be derived
from the strings underlying one or more kernel sentences by a sequence of one or
more transformations.' 3 The idea of representing the English active-passive relation
in terms of transformations was not, however, revolutionary. Jespersen, for instance,
had spoken of the 'turning' 4 and Poutsma of the 'conversion' 3 of the verb form from
one voice to another; but it was only in the case of transformational theory that the

1 I n this study the term 'voice' will be used only with reference to the category 'grammatical voice'
or 'diathesis' in the verb.
2 See, f o r example, Chomsky 1957, 1962 and Lees 1957. Bach calls the passive 'a prototype of a
transformational relation' (1964, 62).
3 Chomsky 1957, 43, 45. I n later rules (1962, 140) the requirement 'transitive verb' ( K ( ) was added.
H o w e v e r , in the most recent versions o f T G rules, many transformations, such as passive, negation,
question, etc., have been partly o r wholly replaced by phrase structure or 'base' rules, since such
transformational rules were f o u n d to be t o o strong in 'expressive power'. T h i s modification o f the
theory 'excludes in principle certain kinds o f derivational pattern that were permitted by the earlier
version o f transformational theory, but never actually found' (Chomsky 1964b, 61). See the
reformulation by Lees (1964) below.
4 Jespersen 1924, 164: 'what was the object ... in the active sentence is made into the subject, and
what was the subject in the active sentence is expressed ... by means o f a prepositional group, in English
with by (formerly of)...' ' W e may express this in a formula, using the letter S f o r subject, Ο f o r object,
V f o r verb, a f o r active, ρ f o r passive, and C f o r "converted subject":
S V« Ο S ν» C
Jack loves Jill = Jill is loved by Jack,
Jack: S" = Cr>
Jill: O " =-- S " . '
; Poutsma 1926-9, 2.2.107.

use of transformation was extended, formalised and systematically incorporated into

a unified grammatical framework.
Chomsky took actives rather than passives as his kernel sentences, since this method
would lead to less complexity than deriving actives from passives (Chomsky 1957,
79-80). The same unidirectional transformation was also favoured by Lees because
'passives are less central than actives' (1957, 388). In a more recent article, how-
ever, Lees finds that Chomsky's rule 'is not only very a-typical of grammatical
transformations in general - it also fails to provide for the correct constituent-
structure of the resulting passive sentences, even though it does correctly serve to
derive [the] passive from an underlying active sentence.' An English passive sentence
'contains two components which do not appear in the underlying active sentence
from which it is derived. First, there is the special verbal auxiliary consisting of the
morpheme be and the participial suffix for the following verb base; ... Second, there
is a special "agentive" adverbial-like prepositional phrase in by plus a nominal
[expression].' His remedy is to 'permit in the expansion of the underlying constituent
trees preceding all grammatical transformation rules the optional selection in any
transitive-verb sentence of a special Agentive constituent. If not chosen, the deriva-
tion leads to active sentences; if chosen, the derivation leads obligatorily to passive
sentences.' Hence, the passive transformation rule is no longer optional but becomes
obligatory: 'it serves to permute in all trees containing the Agentive formative the
transitive-verb object nominal with whatever precedes it, and it substitutes the subject
nominal in [j/c] for the nominal object of by in the Agentive phrase' (Lees 1964,29-30).
Kruisinga had earlier pointed out that while 'it is usual to consider the passive as
a kind of secondary form of the verb, a derivative form dependent upon the "active"',
'this treatment, though supported by tradition and convenience, does not really
permit us to state the facts completely or correctly' (Kruisinga 1927-31, 2.1.335).
Nevertheless, his own classification is based on the active voice. W. S. Allen felt that
'a great deal of harm has been done by teaching the passive voice as if it were merely
another way of expressing a sentence in the active voice. Students are asked to put
such sentences as: John likes girls, Henry can read English and French, etc. into the
fantastic forms of Girls are liked by John, English and French can be read by Henry,
etc. We ought to stress the fact that the passive voice has an important and special
place in the language; most sentences that are good in the active voice are just grotesque
curiosities when put into the passive voice' (Allen 1959, 290).
McKerrow adopted a more radical attitude: 'If we were now starting for the first
time to construct a grammar of modern English, without knowledge of or reference
to the classics, it might never occur to us to postulate a passive voice at all. It seems
to me that it is questionable whether in spoken English of to-day there is really any
such thing, and though, as a matter of convenience, it may be well to retain it in our
grammars, 1 doubt whether it ought to occupy quite so prominent a position as it
sometimes does' (McKerrow 1922, 163).
Some of the diverse points of view advanced in these quotations may be paraphrased

and summarized as two statements that are largely contradictory in regard to voice
(a) There is a relation between the active and the passive voice, and it is therefore
economical to consider one voice in terms of the other. Passives, being less central
than actives, are then best derived from actives. Subject to certain conditions, such
as the verb being transitive, there is a passive sentence corresponding to every active
(b) There are indications that there is not a one-to-one relation between the active
and the passive voice. There seem to be considerable restrictions on the use of passive
sentences generated in this way, and considering the passive simply as a derivative
of the active will not yield a good grammatical description. In fact, the only reason
for keeping the category 'passive' is that it has come down to us as part of our
classical grammatical heritage, and there is little or no place for the passive in a
description of present-day English.


This book will discuss voice in present-day English from different standpoints. Most
attention will be given to the passive, and we shall try to view it in its own right by
reversing the customary procedure of analysis and making the passive our point of
This mode of analysis presupposes that we know what is meant by the term 'pas-
sive voice' in English. Any attempt to define its boundaries by reference to previous
work will soon reveal that there is no agreement among grammarians as to what
constitutes an English passive.® Numerous reasons might be given for this lack of
agreement. The name is certainly partly responsible. Grammarians do not generally
claim that the subject of a passive construction must necessarily be the 'sufferer' of
the action;7 yet there must be some such requirement present in the minds of those
grammarians who preoccupy themselves so much with concepts of 'action' and
'occurrence' as opposed to 'state' when they are setting up a definition of the passive

* There is no exhaustive treatment of the passive in present-day English. Apart from the works
already mentioned, discussions can be found for instance in Curme 1931, Erades 1950, 1958/9,
Francis 1958, Fries 1940,1959, van der Gaaf 1928,1929,1930, Hatcher 1949,1956, Hendriksen 1948,
A. A. Hill 1958, L. A. Hill 1964, Hockett 1958, Jespersen 1909-49,1937, Joos 1964, Kirchner 1936-7,
1951, Koumari 1956, Kruisinga 1927, Mihailovic 1963, Nida 1960, Owen 1914, Palmer 1965,
Strang 1962, Svartengren 1948. - We have chosen a formal synchronic approach applicable only to
present-day English, and we shall therefore not discuss here many works which, while dealing with
the passive in English, have little or no bearing on the present material and method of treatment.
The following list contains a selection of such works dealing with older periods of English: Akerlund
1914, Brose 1939, Curme 1913, Frary 1929, Fröhlich 1951, Green 1913, 1914, Jud-Schmid 1956,
Klingebiel 1937, Kurtz 1931, Meier 1953, Mustanoja 1960, Söderlind 1951, 1958, Turner 1962,
Visser 1941-56.
See, however, the controversy between Meyer-Lübke (1925, 1926) and Vossler (1925). Joos calls
the passive subject the 'victim' of the action (1964, 98).

voice: 'The forms of the verb conjugated with to be and the past participle of the verb
when it does not denote a state resulting from an action.' 8
The weakness of any such definition is not only that it makes agreement difficult
to reach because of reference to semantic criteria, but also that it excludes, apparently
quite arbitrarily, so many related constructions. The line of argument taken here is
that syntactic relationships can or should be expected to be multidimensional rather
than binary and that, in order to find and state this network of relations, it is best to
cast the net wide.
In this study, 'passive' will be considered as a technical term, used in a very wide
sense, for a formally defined construction. In the primary analysis, all the following
sentences will be considered 'passives':

The house was built by experts.

The house was built of wood *
His bills are paid.
His bills are paid regularly every month.
His bills are paid, so he owes nothing now.10
The snow was piled high by the wind.
The snow was piled high by the door.11
The village was {appeared, lay, looked, seemed) quite deserted.
He felt thoroughly disappointed.
The door remained locked,12

Whatever their differences in meaning, all these sentences have one formal feature in
common: they all have as verbs combinations of be (or auxiliaries commutable with
be) and a past participle. For the purpose of our discussion, this will be our simple
working definition of the passive in English.13
Scheurweghs 1959, 416. The following are also representative of this view: The 'participles of
transitive verbs can form a close group with verbs of little independent meaning to express an
occurrence or an action. The most important verb giving rise to such a purely verbal group is to be'
(Kruisinga 1927-31, 2.1.305). 'The subject signals either "that which undergoes the action" or
"that to or for which the action is performed" whenever the Class 2 word [i.e. the verb] to which
the subject is bound is the function word be (in its various forms) or get, with so-called past participle'
(Fries 1959, 180). A 'verbal group consisting of one of the forms of to be plus the past participle of
a transitive verb may denote an action undergone by the subject of the sentence', with the modification:
'in cases where "an action undergone by the subject" seems a somewhat forced definition, the passive
may be said to express what "happens" to the subject' (Zandvoort 1960, 53).
» Francis 1958, 336.
Jespersen 1909-49, 4.98ff.
Hill 1958, 323.
Zandvoort 1960, 49.
It may be objected that this is a definition of, say, the "be + ^erf-construction' rather than the
'passive'. The answer to this criticism is that, firstly, the name is too clumsy and, secondly, little
harm can be done by extending pro tem the domain of an established term to cover not only central
constructions but also those on the periphery. Disagreement on the meaning of grammatical
terminology is largely due to the fact that terms like 'active' and 'passive' are often applied indiscrimi-
nately to notional and formal categories alike, or are used loosely with reference to languages with
different voice systems, or are used without clearly separating diachronic and synchronic approaches.
Voice will be regarded as a grammatical system in the verbal group with two terms:
active and passive. The active term and the passive term are in formal binary opposi-
tion and will be studied both at the rank of the finite verbal group and at the rank of
the finite verbal clause, with regard to their internal relations as well as to their
external relations. The verbal group operates at place V in clause structure.14
This approach, starting the analysis with a certain formally defined structure and
ending up with a statement of its values, makes it necessary to exclude a concurrent
treatment of structures that are formally dissimilar but semantically similar to the
passive, as defined here. We do not want to argue that the converse procedure, i.e.
conducting the analysis from meaning to form, is impossible, but only that it is more
difficult to control in the case of voice. What is essential, however, is to keep the
semantic and nonsemantic approaches separate as far as possible.
Form and meaning may or may not coincide: 'Not only do structures usually signal
several different meanings but, what is more important, there is probably in present-day
English no structural meaning that is not signalled by a variety of structures' (Fries
1959, 203). Meaning may arise out of a pattern of distribution in the language under
description so that 'the grammatical "meanings" are determined by their inter-
relations in the systems set up for that language' (Firth 1957, 22). 'it is this distribu-
tional characteristic which above all others allows the investigator to discover a
morphemic class in the first place, and to know that such classes are relevant to the
language in question, whereas an attempt to make class meaning a basic starting
criterion for determining the classes is fatal to any structural analysis' (Pike 1954-60,


This investigation consists of a series of corpus-based studies. Some of them are

quite independent of each other, but they have all been designed to shed light on
problems of voice in English. There are several reasons for using a corpus. One
follows directly from our aim to describe the use of the passive voice in some co-
existing varieties of present-day English (see below). Another follows from the state-
ment quoted above that some passives automatically derived by rule from actives are
'grotesque curiosities'. Passive sentences recorded from actual speech and writing

On the other hand, a wholesale rejection of traditional terminology is of no advantage. Traditional

labels like 'active' and 'passive', 'subject' and 'object', are useful, provided each is defined in relation
to the particular formal system employed by a particular language at a particular time (cf. Vogt 1950,
137; Buyssens 1950,41; Zandvoort 1961). This is not the point at which we should try to justify our
somewhat unorthodox definition of 'passive'. Rather, its value should emerge from the discussion.
When it is convenient to make a distinction between the active and passive verbal groups, they
will be said to operate at places V and W, respectively, in clause structure (see Chapter 3). For the
use of 'system', 'structure', 'group', 'rank', and 'term', see Halliday 1961. Halliday, however,
recognizes a system of voice in his grammar only in the verbal group, not at clause rank (1964,14f.).

are assumed to be, on the whole, normal and natural uses,15 and a corpus-based study
should provide basic information about how and to what extent 'corpus-passives'
differ from 'rule-generated passives', and from actives. This raises a theoretical
problem: the value of 'data' for the construction of a grammar. If, in general,
'corpus-passives' turn out not to be in a one-to-one transformational relationship to
actives, it will seriously weaken arguments in favour of deriving passives from actives,
at least if our grammar is to be economical and relevant to the use of the language.


The material has either been taken from the files of the Survey of English Usage or
collected, as far as possible, on the principles of the Survey's method of text com-
pilation.18 The material is intended to represent some coexisting varieties of educated
present-day English, spoken and written. In this case, 'present-day English' means
English produced between 1950 and 1964; these are of course only arbitrary dates of
limitation. For practical purposes, it has been necessary to restrict the material to
British English. (There is however little reason to expect that an extension of the
material to cover the other major Standards would give significantly different results
from those arrived at for British English.)
The material consists of a number of 'texts', or samples of spoken and written
English which, with a few exceptions, are continuous stretches. In all, 28 texts varying
in length and totalling some 323,000 words have been analysed. (Text lengths are
stated in number of words, but are not intended to be anything more than approx-
imations. Hyphenated items count as single words.) The texts of 5,000 words each
were taken from the files of the Survey of English Usage, whereas the others were
collected separately.
17 texts of 5,000 words = 85,000 words
6 texts of 15,000 words = 90,000 words
1 text of 28,000 words = 28,000 words
4 texts of 30,000 words = 120,000 words
Total: 28 texts = 323,000 words

The texts are ordered in groups denoted by capital letters (A-M) and numbered (Bl,
This is not, however, a generally held assumption. Bach, for example, maintains that 'real
discourse - especially when spoken in a natural context - is always full of fits and starts and in-
congruities (This form is found in Homer, don't we?). In other words, we cannot identify the set of
grammatical sentence with the set of actually occurring sentence' (1964, 90). It is not possible to
discuss here the justification for equating, in principle, 'occurring' and 'grammatical' without entering
into the whole complex problem of grammaticality. Suffice it to say that the material so far analysed
in the Survey of English Usage seems to support the line taken here. For an investigation into
linguistic acceptability, see Quirk & Svartvik 1966.
See Quirk 1960, Crystal & Quirk 1964, Godfrey 1965.
Conversation (surreptious recording)
Spoken -TEXT Β
Discussions (impromptu unscripted speech)
Television advertising
Delivery in -TEXT D
Present-day spoken form Radio news
Sports reports
Press -TEXT G
Written Editorials
Informative -TEXT I
Popular Arts
Printed Learned -TEXT Κ
Imaginative -TEXT Μ
Fig. 1:1. T h e material.

B2, B3, etc.) where there is more than one text in a group. Figures after stops (e.g.
B1.6; F.26.5; 1.9; M l . 16, etc.) denote internal file references for Texts A, B, C, E,
dates (day and month) for Texts D, F, G, and page references elsewhere. The relation
between the text groups is shown in Figure 1:1. The imbalance between the spoken
and written material (40,000 as against 283,000 words) does not reflect the importance
attached to either variety but rather the difficulty of compiling spontaneously-spoken
material. First year of publication, if different from the year of the edition or im-
pression used, is given in square brackets.

A. CONVERSATION (surreptitious recording):

Text A: Conversation between two university teachers. 1963. 5,000 words.
B. DISCUSSIONS (impromptu, unscripted speech recorded from discussions on B.B.C.
Text B1: Any Questions? 30.1.1959. 5,000 words.
Text B2: Brains Trust. 14.11.1958. 5,000 words.
Text B3: Whafs the Idea? 16.6.1961. 5,000 words.
Text B4: Any Questions? 4.3.1958. 5,000 words.
Text B5: Any Questions? 28.10.1958 and
Any Questions? 16.1.1959. Together 15,000 words.
Text C: 617 advertisements broadcast over the Independent Television Network for the
first time 1.12.1960-31.5.1961. 255 different products were advertised in 7, 15, 30, 45,
60, or 75 sec. advertisements, totalling nearly 35,000 words and representing about
4 hours 30 minutes of continuous broadcasting. When 'repeats' are subtracted, the size
of the text is over 28,000 words.
Text D : B.B.C. Home Service news at 1 p.m. 18-20.3.1964. 5,000 words.

Text E l : The Times. 9.3.1964. 5,000 words.
Text E2: The Daily Express. 10.4.1964. 5,000 words.
Text F: The Guardian. 23-31.5.1960. 30,000 words.
Text G: The Times, not including 'the Fourth Leader'. 1-17.2.1960. 30,000 words.
Text H: Robert Thomson, The Psychology of Thinking (= Pelican Books A 453) (1959),
pp. 11-57. 15,000 words.
Text I: Simeon Potter, Language in the Modern World (= Pelican Books A 470) (1960),
pp. 9-52. 15,000 words.
Text J1: J. Z. Young, The Life of Vertebrates (London, 1962 [1950]), pp. 83-106.
5,000 words.
Text J2: G. E. Bacon, Neutron Diffraction (London, 1955), pp. 162-185. 5,000 words.
Text J3: Η. Ν. V. Temperley, Changes of State, A Mathematical-Physical Assessment
(London, 1956), pp. 22-34. 5,000 words.
TextJ4: G.H.Williams, Homolytic Aromatic Substitution (London, 1960), pp. 27-39.
5,000 words.
Text J5: 14 articles and letters to the editor on biochemistry, biology, genetics,
metallurgy, meteorology, physics, physiology, etc. published in Nature, a weekly
journal of science, vol. 186, 4721-2. 23 and 30.4.1960. 15,000 words.
Text K: P. F. Strawson, Individuals (London, 1961 [1959]), pp. 142-154. 5,000 words.
Text LI: Noel Coward, South Sea Bubble, a comedy (London, 1956), pp. 1-90.
15,000 words.
Text L2: Graham Greene, The Complaisant Lover, a comedy (London, 1959), pp. 1-77.
15,000 words.
Text M l : Rosamond Lehmann, The Echoing Grove (= Penguin Books 1262) (1958
[1953]), pp. 16-30. 5,000 words.
Text M2: Malcolm Bradbury, Eating People is Wrong (London, 1959), pp. 90-107.
5,000 words.
Text M3: Auberon Waugh, The Foxglove Saga (London, 1960), pp. 197-212. 5,000
Text M4: David Beaty, The Proving Flight (= Penguin Books 1318) (1958 [1956]),
pp. 45-59. 5,000 words.
Text M5: Angus Wilson, Anglo-Saxon Attitudes (= Penguin Books 1311) (1958 [1956]),
pp. 11-89. 30,000 words.
Text M6: Michael Innes, The Long Farewell (London, 1958), pp. 9-110. 30,000 words.

1.5 AIMS

The principal aims of this book are to study the values of the two terms within the
voice system of present-day English, with particular emphasis on the passive and its
affinities with the active; to set up a classification of finite passive clauses; to examine
the uses and frequencies of occurrence of the passive in some coexisting varieties of
contemporary English; and to consider ways of accounting for the production of
passive sentences. The discussion will be divided into four parts:
(a) A general discussion in Chapters 2 and 3 of the finite verbal group and the
structure of the rank next above in which it operates, i.e. the clause. Naturally, both
types of unit will be considered particularly from the point of view of voice. We
must exclude from our discussion nonfinite verbal groups and nonfinite clauses.
(b) A study in Chapter 4 of both active and passive clauses in a small sample.
This sample (Corpus I, or the Voice Corpus) consisted of two novel extracts (Ml
and M2) and one scientific text (Jl), together totalling some 15,000 words. In this
corpus all finite verb clauses, i.e. both active and passive, were collected except
equative with be as copula. (For a discussion of major clause types, see Section 3.8,
pp. 36 f.) This sample was chosen to give a general picture of voice in English and to
provide a background for the subsequent more detailed analysis of the passive.
Chapter 4 also includes a specific study of only passive clauses in a larger sample
(Corpus II, or the Minor Passive Corpus), consisting of agentive17 finite passive
clauses from eleven texts totalling some 55,000 words: Texts Bl, B2, and B3 (spoken);
Texts Jl, J2, J3, and J4 (learned scientific); Texts Ml, M2, M3, and M4 (novels).
This sample was chosen to provide more information on 'central' passive construc-
tions than could be ascertained from Corpus I.
(c) An attempt in Chapters 5 and 6 to set up a classification of various types of
passive clause, making use of multiple criteria and numerical taxonomy. A small
sample (Corpus III, or the Taxonomic Passive Corpus), consisting of 128 finite
passive clauses, was used for this experiment.
(d) A typological and quantitative examination in Chapter 7 of passive clauses in
some varieties of present-day English. The material (Corpus IV, or the Major Passive
Corpus) consisted of all finite passive clauses in 28 texts, i.e. the comprehensive
collection of material as described above, totalling some 323,000 words.

'Agentive', in contradistinction to 'nonagentive', will be used for finite passive verb clauses which
either have an agent (and hence are 'agentful') or are 'agentless' but may have extension with an
agent. These terms are discussed in Sections 3.33, pp. 30f., and 4.11, p. 40. Agentive passives are
considered central to the passive construction.



The English finite verbal group can be divided into two major categories: simple and
complex. The simple verbal group (Type 0) consists of a lexical verb without an
auxiliary (for example help-helps-helped, run-runs-ran, hit-hits-hit, etc.). The complex
verbal group, on the other hand, consists of an auxiliary plus some form of the lexical


It is convenient to divide the complex finite verbal group into four types: a, b, c, and d.

Type a ('modal'): λ + jto/ V, e.g. He may examine.

Type b ('perfective'): χ + Ved, e.g. He has examined.
Type c ('continuous'): χ + Ving, e.g. He is examining.
Type d('passive'): χ 4- Ved, e.g. He is examined.

Telescoped into one another, combinations of a, b, c, and d can form complexes of two,
three, or four types. With the system used here the combinations occur in alphabetical
order, so that a cannot follow b, nor b follow c: ab, ac, acd, bed, cd, etc.
In Table 2 : 1 , where combinations with Type d (passive) are contrasted with those
without d (active), the auxiliaries cited are may for Type a, have for Type b, and be for
Types c and d. Present forms, alone or as first elements in combinations, are indicated
by lower case; past tense forms by upper case. When no present/past distinction is
relevant, lower case letters will be used. Imperative and subjunctive are not included
in this classification. The table does not list all possible combinations, for instance
groups such as may be about to be getting fed, but, as we shall see later, all such strings
can be ultimately related to the basic types. Some of the combinations in the table are
of course extremely rare. Highly complex verbal groups such as the following abed-
structure have been heard in conversation, but there are no instances recorded in the
Among earlier treatments of the verbal group to which I am particularly indebted are Hill 1958,
Olsson 1961, Quirk 1962, Strang 1962, and Twaddell 1963.
present material (see Section 7.3, pp. 150ff.): 'Bynow the new cook will have been being
introduced to her duties for several weeks.'


Verbal group types and some of their combinations


0 eats d is eaten
0 ate D was eaten

a may eat ad may be eaten

A might eat AD might be eaten

b has eaten bd has been eaten

Β had eaten BD had been eaten

c is eating cd is being eaten

C was eating CD was being eaten

ab may have eaten abd may have been eaten

AB might have eaten A BD might have been eaten

ac may be eating acd may be being eaten

AC might be eating ACD might be being eaten

be has been eating bed has been being eaten

BC had been eating BCD had been being eaten

abc may have been eating abed may have been being eaten
ABC might have been eating A BCD might have been being eaten

In describing linguistic categories we can usually point to a central core which is

surrounded, as it were, by more or less peripheral subcategories. The exact nature of
the relationship between the various categories and subcategories is often very intricate.
This is certainly true of the complex verbal group in English. The mode of analysis we
use here is naturally prompted by the purpose for which it is intended, i.e. a study of
voice categories. The boundaries of the different types are not clear-cut, and it will be
the object of this book to treat one of them, Type d, in some detail. It does not
purport to deal exhaustively with Types a, b, and c.
It will, however, be necessary to try and define more closely the area occupied by
the auxiliaries. 'Auxiliaries' are here considered a special class of the general gram-
matical category 'verb', the other class being 'lexical' or 'full' verbs.
In order to arrive at a fairly comprehensive classification of Types a and c, a number
of criteria have been used, which are set out at the head of columns in a matrix table. 2
Some specimen verbs are entered in a column on the left of the table, and when a
particular verb satisfies a particular criterion, this is indicated in the cells by ' + ' ,
otherwise by '—'. 'Inapplicable' is denoted by '/'.
For the use of matrices in linguistics, see for example Pike 1962, 1963.

2.3 TYPE a

2.31 CritericP

CRITERION 1. Γο-less infinitive, for example, He must do it, as opposed to He has to

do it.
CRITERION 2 .Invariance for person and number in the present. Test-frame:
I/youjhe χ V now. Can but not be going to, for example, will satisfy this criterion:
I am/you are/he is going to do it.
CRITERION 3. Both the present and the past forms can be used, with modal distinc-
tion, in a present sequence. A test-frame like I think I χ do it now at once will accomo-
date for example either can or could, but only have/has to and not had to.
CRITERION 4. Only the position as first element is possible in verbal group structure,
i.e. the verb cannot occur in the frame: He may χ V. This criterion is satisfied for
instance by will but not by appear to.
CRITERION 5. Postverbal enclitic negation is possible. Do but not seem to, for
example, will fit the frame: You x'nt V.
CRITERION 6. No Jo-periphrasis is possible in negations and questions, i.e. the verb
cannot occur in the frame You do not χ V. This criterion is satisfied for example by
will, but not by want to.
CRITERION 7. This cannot serve as a substitute for (to) V, for example, He appeared
to drink beer and *He appeared this, as opposed to He attempted to drink beer and
He attempted this.
CRITERION 8. Voice transformation is possible of the form: Ν χ χ (to) V N2 «->
Ν2 x (to) be Ved. Both those finite verb clauses which satisfy Criterion 7, and those
which do not, may take voice transformation in the clause, but the transforms are
different and must be distinguished here. The two types may be illustrated by the
following pair of sentences, both symbolised as Ν χ χ to V Ν 2:

(i) He appeared to drink beer.

(ii) He attempted to drink beer.

Sentence (i), unlike (ii), may admit of a passive transform with χ as Type a auxiliary in
the passive verbal group and N2 as subject of the passive clause (JV2 χ to be Ved):

Beer appeared to be drunk (by everyone in the pub).

*Beer attempted to be drunk (by everyone in the pub).4
CRITERION 9. No ίΑαί-clause is possible. This criterion excludes, for example, hope to:
The slots in the frames where the verbs are tested are indicated by x.
* This should be distinguished from the passive transform This was attempted. Sentence (ii) does
not satisfy Criterion 7, and, in consequence takes passive transformation with χ as the lexical verb
in the passive verbal group and the factive substitute as subjects:
*This was appeared.
This was attempted.

He hopes to do it and He hopes that he can do it; it includes, for example, seem to: He
seems to do it but *He seems that he can do it.5
CRITERION 10. No Ν (to) V construction is possible. This criterion excludes, for
example, like, which admits both He likes to do it and He likes you to do it; it includes,
for example, attempt which admits only He attempts to do it but not *He attempts you
to do it.
CRITERION 11. No Ving construction possible. This excludes, for example, like:
He likes to do it and He likes doing it; it includes, for example, manage: He manages
to do it but *He manages doing it.

2.32 Matrix


Matrix for Type a


1 2 3 4 5 6 7 8 9 10 11

el can + + + + + + + + + + +
al do6 + — — + + + + / + + +
<z3 had better'' + + + + ± + + + + + +
a4 ought to8 - + + + + + + + + + +
a5 used to8 — + — + + + + + + + +
«6 be to — — — + + + + + + + +
al be going to — — — — + + + + + + +
c8 have to' — — — — + — + + + + +
«9 seem to + + + + +
alO start to ± + + —

all manage to + + +
al2 learn to + +
el3 expect to +
a!4 like to — — — — — — — — + —

The fourteen verbs listed in the matrix (Table 2:2) are different in that no two are
fully alike with respect to the selected eleven criteria. The poles of the scale are re-
presented by can, which satisfies all the criteria, and by expect to and like to, which
satisfy only one. It should be emphasized that the criteria were chosen fairly arbitrarily
Seem with proleptic it, as in It seems that he's leaving, is a different construction.
• Do does not cooccur with the auxiliary be in the verbal group except in emphatic imperatives
(such as Do be quiet!) and, consequently, does not cooccur in the lie-passive. Cf. however: Did he
get caught? The second do in Did he do it? is a lexical verb.
The negative He hadn't better do it is perhaps less common than He had better not do it.
Ought to, but less certainly use to, with <fo-periphrasis (Did he ought to/use to do it?) are probably
marginal to Standard English.
• Phonological and prosodic features may provide additional criteria: both have and used 'have
lost the juncture before to and consequently have undergone assimilation of the final consonant'
(Hill 1958, 198): ['haefta, 'hsesta, 'justs]. At least in British English, both the enclitic and periphrastic
negatives occur with have to.

from those that were felt to be relevant; a different choice of criteria would naturally
alter the classification. The decisive factor here in choosing a classification must be
the particular application for which it is intended. For a study of voice the obvious
division would seem to be one that enables us to make comparative statements about
the active and passive terms in the voice system. Hence we make our point of division
the one on the scale which separates those verbs which may operate as χ in the active
as well as in the passive verbal group from those which do not operate in this way.
In the table this point occurs between seem to and start to. The verbs in Classes al-9
will be called 'auxiliaries', and those in Classes alO-14 'lexical verbs'. All auxiliaries,
as the term is used here, should then in principle permit voice transformation, which
may be stated in the following formula:

(Type a) V (Type ad) be

Ν! χ (Type b) Ved N2*-+ N2X (Type bd) been Ved (by NT).
(Type c) Ving (Type cd) being

There are, however, considerable constraints in the use of auxiliaries which affect the
possibilities of voice transformation, for instance the alternation between shall and
will as future tense auxiliaries depending on the person of the subject. Such constraints
are not considered here.
Class alO (start, begin etc.) has ± for Criterion 8, since these verbs may take double
passive transformation, for example

They began to use a new method.

-» A new method began to be used.
-> A new method was begun to be used.(!)

Because of this and the fact that they do not satisfy Criterion 11, they are classed as
lexical verbs. This division will allow us to make comparative statements with Type c.

2.33 Classification

Type a auxiliaries can be subclassed in various ways. The most useful division for our
purpose seems to be one between those auxiliaries which can occupy the position
of first element only and those auxiliaries which cannot do so (as in Criterion 4), since
this has a bearing on the verbal group structure. Classes al-6, represented by the
verbs can, do, had better, ought to, used to, and be to, are termed 'closed class auxi-
liaries' (a'); and Classes al-9, represented by be going to, have to, seem to, 'open class
auxiliaries' (a").
Type a is formed either by closed class auxiliaries plus infinitive with or without to,
or by open class auxiliaries plus ίο-infinitive:

closed class auxiliary (a'): may eat, or,

open class auxiliary (a"): has to eat, or by a combination,
closed class + open class auxiliaries (a' + a"): may have to eat, or
open class + open class auxiliaries (a" + a"): seems to be about to eat.

The combinatory types occur because the open class auxiliaries can themselves form
substructured combinations with closed class auxiliaries, for example, may have been
having to eat ([a' + b + c +] a" + V), or with members of its o w n class, for example,
may seem to be about to eat ([a' + a" +] a" + V).w
The following list of auxiliary classes for Type a, set up on the basis of the criteria
in the matrix, is not intended to be exhaustive. There are, besides collocational 11
restrictions, considerable idiolectal, regional, and stylistic variations in the use of the
verbal group, and no general agreement can be expected on this listing.


a' 1 can, could a"l be (un)able to

may, might12 be about to
shall, should be due to
will, would be going to
ΊΙ, V 13 be sure to
must be (uri)likely to
dare11 have got to
need14 a" 8 have to
a'2 do a"9 appear to
a'2 had/'d better/best happen to
a'4 ought to seem to
a' 5 used to tend to
a' 6 be to

'The large number of these auxiliaries and the multiplicity of ways in which they may be
combined permit a very large repertory of verb structures to be built in English. From a historical
point of view, it is interesting to note that many of them are of quite recent development. This
seems to be an area of English grammar where change has been taking place rapidly. Indeed, change
seems to be still going on, and may continue to for some time to come' (Francis 1958, 259-60).
'Collocation' is used to mean 'the habitual association of a word in a language with other
particular words in sentences' (Robins 1964, 67).
" Mayn't is, according to Hill, 'very rarely used in the United States, though it is more frequent
in England.' Mightn't 'is commoner' (1958, 196).
The auxiliaries ΊΙ and V, which are themselves enclitic, cannot, however, take the enclitic
negation: *He'dn't do it.
Dare and need are typically marginal between open and closed class auxiliaries, as shown by
overlapping distribution and a profusion of alternative forms. Cf. for example He may dare do it
and *He may need do it. In an experiment involving the selection of negative forms with dare and
need, the following combinations were chosen (see Quirk & Svartvik 1966, 90ff.):

to 49 1
aux + neg + dare ^ 13 I

Lexical verbs, Classes a 10-14, may be arranged in valency classes. (Superscripts

denote valency as indicated by ' in Columns 8, 9, 10, 11 in Table 2:2.)
V-e care, fail, manage, serve, ...
ja8,-9 decide, learn, resolve, ...
F"8·11 begin, cease, continue, start, ...
y-β, »,-io expect, promise, want, wish, ...
Κ8-10·11 like, love, ...

2.34 Concord valency classes

Criterion 2, the formal variability criterion for person and number, can be further
refined if we adopt the idea of valency classes.15 These classes are set up on the basis of
concord between the subject and the finite verb. As concord-bearers there are three
classes of subjects:
(x) First person singular: I.
(.p) Second person singular and the entire plural:you, cats, thoughts, ...
(z) Third person singular: Tom, cat, thought,...
The subject classes are labelled differently from Olsson's in order to avoid confusion
with the terms used in the present treatment for describing the verbal group types.
Distinction is also made between present and past forms, indicated by lower and
upper case respectively. No account is taken here of the modal contrasts, which to
some extent cut across the temporal contrasts.
The valency for a particular verb form may then be stated as VALENCY \ Y for
were, VALENCY 1* for does, VALENCY 2X+Z for was, VALENCY 2 f o r do,
VALENCY for can, VALENCY 3X+Y+Z for could, VALENCY 6χ+ν+ζ+x+r+z
for must, and so on. Concord valency classes could be extended to all verbs, but it will
be sufficient for the present purpose if we restrict ourselves to Type a auxiliaries. On
the basis of Criterion 3 we may further divide auxiliaries into
(i) NON-TENSE-MARKED : auxiliaries with only one form, which may occur in both
present and past tense sequence ( = Concord Valency Class I).
+ to 61
+0 0

Need appears to be more common in negative constructions and need to in positive. Voice trans-
formation with dare is doubtful.
" See Olsson 1961, 29 ff.
(ii) SEMI-TENSE-MARKED: auxiliaries with two forms of which the past form is not
restricted to past tense sequence ( = Concord Valency Class II).
(iii) FULLY TENSE-MARKED: auxiliaries with one or more forms which do not share
the tense sequence properties of (i) and (ii) ( = Concord Valency Classes III-V).
This classification is illustrated in Table 2:3.

Concord valency classes for some Type a auxiliaries


X y ζ

fix+ll+z+X+Y+Z a'l must


a'l dare
I a'l need
a' 3 had better
fl'4 ought to

a'l can could



3 x+v+z/l a'l may might

II 3x+r+z a'l shall should
a'l will would
a'l '// 'd

III 3x+r+z a' 5 used to

a'2 do does did

<f 8 have to has, 's to had, 'd to
w IV 2χ*«/ I'// a'9 dare to dares to dared to
Ζ jx +r+z a"9 need to needs to needed to
Β a'9 seem to seems to seemed to
a"9 appear to appears to appeared to
u. ΐ χ / l v ß z (2x+v)ll am, Ίη | are, 're were
V 2x+z/\y pjr+y+z) a'6, 7 (aren't)1' is, 's was 1 was

2.4 TYPE b

There is one central auxiliary in Type b, viz. have. Other auxiliaries which can occupy
the place χ in the χ + Ved combination and commute with have might also be classed
as Type b auxiliaries. However, since there is no clear borderline between Types b and
d in many such cases, we shall be interested, for the present purpose, in considering
them in relation to Type d as a whole (see, for example, Section 5.44, pp. 83 ff.).

2.5 TYPE c

Type c has be as its central auxiliary, but there are other possible candidates, which
form a continuum similar to that described for Type a (see Section 2.3).
" In negative-interrogative clauses, aren't I occurs besides am I not.
Valency 3 J r + K + z refers to the subjunctive forms: Ifljyoujhe were ill, ...

2.51 Criteria

CRITERION 1. Postverbal enclitic negation is possible, for example, He isn't doing his
homework as opposed to *He startsnt doing his homework.
CRITERION 2. This cannot serve as a substitute for Ving, for example, He went on
singing and * He went on this, as opposed to He began singing and He began this.
CRITERION 3. Voice transformation is possible:
Nl χ Ving Ν 2«-» Ν 2 χ being Ved (by N^. He is doing the job and The job is being done,
but He avoids doing the job and * The job avoids being done.
CRITERION 4. No transformation is possible of the form N1 χ Ving ->• Ving is what
N1x \ He keeps walking but * Walking is what he keeps, as opposed to He likes walking
and Walking is what he likes.
CRITERION 5. No (to) V construction is possible, i.e. no change from Type c to
Type a: He keeps walking and *He keeps to walk, as opposed to He likes walking and
He likes to walk.

2.52 Matrix


Matrix for Type c

1 2 3 4 5

el be + H- + 4

<•2 keep — -i- + -L

<•3 begin - - -!• -

c4 avoid — — — — -f.

c5 like _ _ _ —

Five verbs are listed in the matrix (Table 2:4). Reapplying the argument put forth in
Section 2.32 for the classification of Type a, we call Classes cl-2 Type c 'auxiliaries' and
Classes c3-5 'lexical verbs' on the grounds that the former, unlike the latter, may
operate as χ in an active as well as in a corresponding passive verbal group. The reason
for dividing Type a auxiliaries into two classes, according to whether they can occupy
first place or not in verbal group structure, does not obtain in the case of Type c.
We may, however, recognize the difference between Classes 1 and 2 and call them
'closed' and 'open' respectively, although there is only one formal criterion which
separates them. Class c3 verbs are taken as lexical, since they do not so readily take
voice transformation: He continues playing the record -* (?) The record continues
being played; a Type a construction (continues to be played) is probably more likely.

' 6 The relation between He is to walk and He is walking is different from that between He likes to
walk and He likes walking.

2.53 Classification


c'l be c"2 keep (on (and on))
go on (and on)
Lexical verbs, Classes c3-5, may be arranged in valency classes. (Superscripts denote
valency as indicated by ' in Columns 4-5 in Table 2:4.)
V * avoid, remember, ...
V b begin, continue, stop, ...10
V *'-b like, enjoy, prefer, try, ...

2.6 TYPE d

Type d is the main subject of this study and, for the purpose of our discussion, the net
will therefore be cast wide to include instances which are more or less removed from
the construction centre (see Sections 1.2, 2.4). There are, for example, no specific
requirements of subject value for membership of Type d: at this point of the analysis
no account is taken of whether the grammatical subject is identical or not with the
logical subject (see Section 5.45, pp. 89 f.).


Lexical verbs may consist of one or more words. One-word verbs may be, for example,
simple words (dress); or complex words with bound stem + derivational suffix
(stupefy), with prefix + bound stem (intervene), with free stem + derivational suffix
(broaden) and with prefix + free stem which may be a simple word (untie, overflow,
outgrow, offset) or a complex word (reconvert).20

2.71 Phrasal and prepositional verbs

Verbs of two or more words consist of the verb proper (the 'nucleus') and adverbs or
adjectives or prepositions, with which they may form close syntactic units. We
distinguish the following three classes (example-centres are given in italics):
An experiment to test this theory was carried out on January 6, 1959. (J5.273)
" Stop has different meanings when followed by to V and Ving: cf. He stoppedfightingI to fight
and He started fighting!to fight.
" Francis 1958, 206ff.

But in the last six months attention has to some extent been drawn away from these
central themes to peripheral worries. (G.10.2)
Some raw shelving had been run up at one end to stack additional books - ...
Indeed, with his own salary and his wife's private income, they were really very
comfortably provided for. (M5.24)
'But in this country there are singularly few murders which are conceived of as
deliberately incident to a robbery.' (M6.38)
If a woman reads history for a degree, and then rejects a career for marriage, her
mind is not to be thought of as thrown away unless we count the family an arena of
no importance. (G.2.2)
He had in fact recovered for Packford some valuable documents which had been
made off with by a rather specialised sort of burglar. (M6.10)
There are two problems of delimitation here. One is between phrasal and preposi-
tional verbs, where two criteria have been used.21
Firstly, there is a difference in stress and intonation patterns. The particle com-
ponent of the phrasal verb normally (i.e. in unmarked use) bears full stress (and, if at
tone unit boundary, nuclear tone), whereas the preposition with the nonphrasal verb
is normally unstressed and does not carry a nuclear tone:
(Vph) He was taken Ίη. (The particle has a falling nuclear tone.)
(Vp) He was laughed -at. (The preposition has the 'tail' of the nuclear tone.22)
Secondly, the particle component of a phrasal verb has twofold positional privilege in
that it often takes either prenominal or postnominal position, whereas it is restricted
to postpronominal position. The diagnostic frames in Table 2:5 show that the two
classes of verbs, represented by call on and call up, structure differently.
Verb nuclei and adjectives may have close collocation and behave syntactically like
phrasal verbs, for example He pushed the door open. Adjectives in such superficially
similar constructions, as in He considered the man good, are however treated as com-
plements (C elements in clause structure; see Section 3.1, pp. 25 f.) for reasons that are
clear in Table 2:5.

2.72 Prepositional verbs and prepositional phrases

The other problem occurs in active clauses where prepositional verbs plus nominale
(Vp + N) have to be distinguished from simple verbs plus prepositional phrases as

This treatment draws heavily on Mitchell 1958. Cf. also Kennedy 1920, and Strang 1962, 156ff.
For the intonation system used here and elsewhere in the book, see Crystal & Quirk 1964.


Diagnostic frames for simple, phrasal, and prepositional verbs


(+) They call on the man. (+ ) They call up the man.

(+) They call early on the man. (—) *They call early up the man.
(+) They call on him. (-) »They call up him.23
(—) *They call the man on. (+) They call the man up.
(—) *They call him on. (+) They call him up.
(+) The man they call on. (+) The man they call up.
(+) The man on whom they call. (—) *The man up whom they call.



(+) They push open the door. (—) *They consider good the man.34
(—) *They push open it. (—) *They consider good him.
(+) They push the door open. (+) They consider the man good.
(+) They push it open. (+) They consider him good.
(+) The door they push open. (+) The man they consider good.
(—) 'The door open which they push. (—) "The man good whom they consider.
(—) *They push (that) the door is open. (+) They consider (that) the man is good.

adjuncts (V + pN; see Section 3.1). There is a scale of closeness and openness whose
poles may be illustrated by the following examples:

(Vp + N) She sent for the coat. -* The coat was sent for.
( V + pN) She came with the coat. *The coat was come with.

An attempt has been made elsewhere to classify such 'prepositional strings' compre-
hensively by the use of numerical taxonomy (see Carvell & Svartvik 1966). For the
present special purpose the chief criterion will be the voice transformation test, which
will help to distinguish prepositional verbs from verbs plus prepositional phrases. 25

2.73 Compounds

The past participle ( V e d ) in the passive verbal group presents special problems, as
compared with the active, since it may be morphologically isolated by compounding:
May possibly occur with emphatic (contrastive) stress on him, as in We rang up him, not his brother.
" Possible only with a heavy object, as in They would consider brave any man who could volunteer
for such a task.
" The acceptability of some such passive transforms varies considerably depending for example
on the amount of linguistic context. In a poll conducted in order to register informant reactions to
this kind of potential active-passive transformations, it was found, for instance, that the minimal
sentence The girl was turned to (from the active They turned to the girl) and the same sequence
differently contextualized The Prime Minister was turned to for help by the people suffering from
depression in the north-eastern industrial areas produced very different results.

unwanted, well-known, new-laid, etc. have no bases {to) *unwant, *well-know, *new-lay
(cf. Marchand 1960, 54).| The distinction between compounds and noncompounds
is not an absolute one, but one of degree. We shall consider some aspects which are
relevant to our problem.
Compounds often have single stress on the first element, or double stress: ['- - ] in
airborne, awe-struck, ['- - ] or ['-'-] in well-judged, etc. However, stress alone cannot
be used as a criterion, particularly since it is connected with position in the clause,
for example this well-known ['- -, '-'-] author but usually this author is well-known
It will for our purpose be convenient to let the term 'compound' embrace not only
word elements (for example all-admired) but also the prefix un- (as in undressed). The
««-prefix has two values (see Jespersen 1909-49, 6.464 ff.):
(i) REVERSATIVE un- which joins a simple word verb so as to form a complex word
verb, whose past participle form may occur in both the active and passive paradigms
(Types b and d): He has/is undressed; also undo, unpack, unbind, unbutton, uncover,
unfasten, etc.
(ii) NEGATIVE un- which joins an adjective, a noun, or a participle so as to form a
compound word (for example unsympathetic, unease, unwanted). This prefix is a
bound negative making the compound the antonym of its base (see Lyons 1963,61ff.):
This result may have been unexpected after Holland's fine golf on previous
days, ... (F.26.5)
-* This result may not have been expected ...
-* [they] may not have expected this result ...
The difference between the two types may be diagnosed by commutation tests within
the verbal group, for instance between Types d and b (see Section 2.1) :2e
Reversative prefix: She is/has undressed.
Negative prefix: She is/*has unexpected.

2.74 Verbal and nonverbal bases

Like the prefix un-, -ed has two values: it is noi only a verbal inflexion but also a
suffix. As a suffix it 'is now added without restriction to any sb. from which it is
desired to form an adj. with the sense "possessing, provided with, characterized by"
(something); e.g. in toothed, booted, wooded, moneyed, cultured, diseased, jaundiced,
" The difference between the two un- prefixes may be neutralized. The minimal structure She was
undressed may have the following expansion and transform: The chorus girl was undressed every
night before the show by the wardrobe mistress. -> The wardrobe mistress undressed the chorus girl
every night before the show. But whether the sentence She was/appeared undressed every night on
the stage should be analysed as She appeared denuded!stripped... or as She appeared not dressed... is
irrelevant, since the origin of unVedis unknown here and the distinction, in any event, syntactically

etc., and in parasynthetic derivatives, as dark-eyed, seven-hilled, leather-aproned, etc.'

'In mod. Eng., and even in ME., the form affords no means of distinguishing between
the genuine examples of this suffix and those ppl. adjs. in -ED1 which are ultimately f.
sbs. through unrecorded vbs.' (OED, s. v. -ed).
Three relevant ««-types will be analysed as follows (X stands for the first element):
(i) A complex word consisting of reversative prefix + verbal base + inflexion
[(X + V) ed], for example undressed.
(ii) A compound word consisting of negative prefix + verbal base + inflexion
[X(V + ed)], for example unexpected.
(iii) A compound word consisting of negative prefix + nominal base + suffix
[X (N + ed)], for example unskilled.
The distinction between verbal and nonverbal bases87 may be diagnosed by commu-
tation within the verbal group, in compounds with the first compound element
(i) He was undressed: He had undressed: He was dressed: He had dressed
(ii) He was unexpected: *He had unexpected: He was expected: He had expected
(iii) He was unskilled: *He had unskilled: He was skilled: *He had skilled

_—(V) simple taken

—{Vpk) phrasal looked up
(Ved) Noncompound —(Vp) prepositional looked upon
—(Vph-p) phrasal-prepositional got away with

Verbal base
—Ad] -η — —
—Adv - well-judged
-N - man-made
(X- Ved) Compound —un —
+ Ved
-self - self-appointed
-all - all-admired

{Ned) Noncompound wooded

Nonverbal base
—Adj -i — —
-Adv - down-hearted
-N - hunchbacked
(X-Ned) Compound —Sum— + one-eyed
—un — unskilled
-seif - self-willed
Fig. 2:1. Word elements and bases.
'Verbal base' naturally refers to present-day English. The fact that some of the Ved forms have
had a verbal base at some time in the history of the language is immaterial for a synchronic descrip-
tion. Ashamed, for instance, which has no verbal base today, is the past participle of a verb to ashame
(OE äsceamian), last recorded in 1826 by OED. In the case of alternative participial forms, both are
considered as having the same base, although there are often different restrictions on their use, such
as semantic (He was born in 1950 / air-borne), collocational (poverty-stricken / thunder-struck), or
paradigmatic (he is drunk / a drunken sailor).

By definition (see Section 1.2), the type illustrated by unskilled cannot enter into the
verbal group, but we shall eventually want to consider its relation to the voice system
as a whole (see Sections 5.52, p. 95 and 6.46, pp. 137 f.). Although ««-compounds
with verbal bases can have only restricted active-passive clause transformation
potential (see Section 5.44, p. 86), their constituent structure bears, nevertheless,
great resemblance to that of noncompounds with verbal bases. This is illustrated for
instance by their ability to take agents:

All his investigations were totally ungoverned by the slightest awareness of the actual
substance of the stuff he dealt with to such triumphant effect. (M6.11)

Figure 2 :1 exemplifies and displays diagrammatically the morphology of some com-

pounds and noncompounds with both verbal and nonverbal bases.



The finite verbal group operates at the place in structure called V in the rank next
above the group, viz. the finite verb construction, or, as it will be convenient to refer
to it here, the finite verb clause. At the highest level of abstraction the English clause
elements are V, S, C, and A.1
The V/W element (the verbal group) has been dealt with in Chapter 2. In the fol-
lowing sections we shall first discuss the other clause elements and then the major
clause types which are formed from them.
S (the subject) in English may be defined as that element in the clause with which
the finite verb is in grammatical concord (overt or covert).2 Since the finite verb is so
often morphologically invariable (see Section 2.34, pp. 16f.), a more useful working
definition, which holds good in most cases, is one based on position: S is that nominal
clause element which in affirmative clauses normally occurs before and in questions
immediately after the first element of the finite verbal group (cf. Strang 1962, 71).
5 may be zero, particularly in sequentially related clauses (He said good-bye and left),
and it is normally so with imperatives [Go away !).3
C (the complement) is that element besides S in the clause which is typically mani-
fested by the nominal group.
A (the adjunct) stands, broadly speaking, for the rest.
C may be said to be that nominal element which is usually essential to complete the
basic clause structure, 4 i.e. C is generally obligatory whereas A is optional. C and A
cover more indeterminate territories than S and V. They may be diagrammatically
For the term 'rank' see Halliday 1961, 250ff. In respect of this unit there is unique solidarity
between clause and group, and V is therefore used here both at the rank of the group for denoting
the lexical verb and at the rank of the clause for the place in structure at which the verbal group
operates. When a voice distinction at clause rank is needed, it will be convenient to use V for active
and W for passive.
Unless stated otherwise, 'subject' will be taken to mean 'grammatical subject'.
The fact that imperatives may have S realized (You go away!), albeit with restricted exponence,
makes a good case for considering imperatives as members of the finite verbal group. Another
reason for doing so is the fact that they take do-periphrasis (Don't go away!).
* See Quirk 1962, 190; and Firbas, who considers complements 'essential amplifications of the
respective verbal meanings' (1959, 46).

represented as two points, A being lower than C, on a scale of essentiality or indispen-

sability of clause structure elements:

V most
s essential clause elements
A least

Within the last two elements there are, however, subscales which may be so extensive
as to form functional overlap of the C and A elements. Prepositional phrases, for
example, which at the primary degree of analytical delicacy are all taken as A ele-
ments, may be like C elements in some respect, such as having voice transformation
potential (see Section 2.72, pp. 20f.).8
Dispensability seems to be correlated with the number of possible element items:
there can be only one S and one V element but either one or two C elements, whilst in
theory at least, there can be an unlimited number of A elements. In the material here
analysed, the maximal number of A items was four (see Sections 4.51, p. 58, and
5.82, p. 109).


The S and C elements can be treated together, although typically they have different
exponence. It is convenient to use the symbol Ν for any nominal and nominalized
element which operates at places S and C in clause structure, and also for the nominal
part of prepositional phrases which operate at place A.6

3.21 Form-class

We distinguish between nominal groups and nominalizations. The latter comprise

constituents other than nominal groups but which operate as such, for example finite
and nonfinite verb clauses.
Nominal group heads have three subclasses: nouns, pronouns, and names. Of these
only nouns regularly take determiners. Pronouns constitute a closed class, some of
whose members have oblique forms other than genitive (he/him, they/them, etc.).
Names may take the provided there is modification as well {the inimitable Jeeves, the
London I know); they do not admit of number change: if singular, they cannot be
pluralized (Agneta but *Agnetas); or, if plural, cannot be singularized (The Andes but
*an Ande). The in The Andes or The Thames is considered part of the name since it
For the concept of 'delicacy', see Halliday 1961, 272f. Subscales 'constitute small continua
within themselves* (Bolinger 1961a, 44).
Denoted ρ Ν, see Section 3.3. The classification here of the clause elements S, C, and A is,
naturally, less refined than that of V. See further Sections 5.6-8 (pp. 97 ff.).

'is not in determinative contrast with any entity, with *a Thames and *some Thames,
for instance. It follows that the is neither the definite article nor a generic determina-
tive' (S0rensen 1958, 150; cf. 133 ff.)·
Finite verb clauses are, for instance, ίΑαί-clauses, and conditional and relative
clauses, which in the following examples operate as C:
It was only men, Emma considered, who believed in things; women recognised
that being a woman was way of life enough. (M2.94)
'He asked me if I could look after him till he came out ...' (Ml. 18)
Faintly she frowned, contemplating the area of her labours, seeing what should have
been, what could be done. (Ml. 17)

Nonfinite verb clauses include, for example,

(1) To-full infinitive (to V):
To pass among the wooden stalls of the market, ... was a pleasant sensation.
(2) Γο-less infinitive (V):
they watched the apparition wave up and down ... (Ml.23)
(3) Present participle (Ving):
Ί found him cutting his hair in my drawing-room the other afternoon before the
departmental tea-party.' (M2.104)
(4) Past participle (Ved):
Ί don't want other people's humanity tied round my neck\ (M2.107)
There are other possible exponents of S and, particularly, of C, for example direct
speech and adjectives:
'You live alone?' said Madeleine, rather awkwardly. (Ml. 19)
/anything that makes the : army atträctive # . a is de/sirable and a good thing #

3.22 Gender

Gender distinctions are as far as possible set up on the basis of formal criteria, i.e.
patterning with substitution classes of personal and relative pronouns (see Table 3: l). 7
In addition to this, the category 'animate* also embraces nouns which collocate as S
with verbs like see or like (rat, snake, etc.). For the present purpose it has usually
been considered sufficient to recognize a two-term gender system with animate (Nan)
and inanimate (Nin) as terms. The former then includes the subterms 'personal' and

Cf. Francis 1958, 250f„ Fries 1959, 120ff., and Strang 1962, 95.

Gender classes


you Γhejit
he she/it [who ]
+ who ['' 1 + which
she he/she/it \whichj L they\
he/she Lit/they J
jthey _

boy, girl, doctor, horse, cow, baby, book, life, grammars,

mathematicians, ... committee, ... mathematics, ...

3.23 Finitude

Ν may be definite or indefinite. Nominal groups are taken as definite when having the
definite article, a genitive, or a possessive or demonstrative pronoun (the/John s/myl
this girl). Names and referential pronouns are definite. The latter include for example
personal he but not interrogative who (/ wonder who's coming?), proleptic it (It was
learnt that ...), or nonreferential you and one:
It was a showpiece of the unendurably modern - when you saw the modern like
that, it looked so dated that you couldn't believe it. (M2.97)

3.24 Person

Person comprises the traditional categories first, second, and third person as exem-
plified in Section 2.34.

3.25 Number

Number is divided into singular and plural according to pronoun patterning (see
Table 3:1). In the material where this category was used, no 'common number'
(The committe is/are ...) occurred.

3.26 Modification

Modification subsumes premodification (as in the little girl) and postmodification

(the man over there).
3.27 Coordination

Simple head (for example you) is contrasted with coordinate (for example you and I).


Exponent classes of adjunct (clause element A) are multifarious, and 'there is a

continuous spectrum of classes from those functioning as adjuncts to verbs to a host
of other kinds of adjuncts - the verb-adjuncts are at one end of the spectrum and
should be seen in relation to it as a whole' (Strang 1962, 161; see also Sections 4.5,
pp. 57ff., and 5.8, pp. 101 flf.). Our chief interest here lies naturally in that end of the
spectrum where we find the verb-adjuncts, and hence A elements which are taken to
have no such function are not considered in this study.
Consequently, A does not embrace adjuncts of S, C, and A elements, or sequential
relators and sentence adjuncts, such as accordingly, actually, however, in fact, in
general, of course, perhaps, probably. Sentence adjuncts are often characterized by
having (i) actual or possible individual intonation units, (ii) freedom of movement
within the clause, particularly if occurring with individual intonation units, or (iii)
no generic substitution class meaning8 as do most, but by no means all, adjuncts.
(This is indicated by the possibility of using them in response to questions with
where, when, how, etc., see Section 5.8.)
A, as the term is used here, does however embrace other types of adjuncts which
are more or less marginal to the V element, for example finite and nonfinite verb
clauses. Adverbial particles (up, down, etc.) are treated as part of the lexical verb
(see Section 2.71, pp. 19f.). Adjuncts have been classified according to form-class
and position in clause structure.

3.31 Form-class

A elements may be divided into the following form-classes:

(1) Adverbs, which include, for example, (la) central invariables ('He arrived now,
at once, here', etc.), (lb) central variables ('He works fastest'He runs slower), (lc)
de-adjectival ('He runs slowly'), (Id) other derivatives, compounds, etc. (adrift,
homewards, nowhere, publicity-wise, etc.). (See Strang 1962, 162 f.)
(2) Prepositional phrases:

professors nibbling cheese straws peered over the tops of them to see what was
happening. (M2.107)
(3) Finite verb clauses:
The sucker is bounded at the edges by a series of lips, which besides being sensory
serve also to make a tight attachment when the lamprey sucks. (J1.88)
(4) Nonfinite verb clauses (with or without conjunction), where the verb may be

» Cf. Francis 1958, 287f., Strang 1962, 164f.


(4a) Present participle (Ving):

She followed his course with a dreamy look, remarking that he was very
obedient. (Ml. 19)
(4b) Past participle (Ved):
'Tell me, has it arrived?' she asked, well pleased with the success of her little
gesture. (M3.212)
(4c) 7o-full infinitive (to V):
Ί heard you were coming with us to give a lecture on your findings to the New
York Canford Institute.' (M4.57)
(5) Verbless constructions (with or without conjunction):
it is unique among vertebrates in that the follicles have no ducts; when ripe they
rupture into the coelom, ... (J 1.96)

3.32 Position

There are three major places of A in relation to other clause elements: front-, mid-,
and end-position. The definitions given here apply to the order S V(C), the most
common order of clause elements in statements (for a detailed discussion of adjunct
positions, see Jacobson 1964).
(F) Front-position: pre-S, i.e. A before any other clause element.
after a while she stopped dead close to the church door ... (Ml.27)
(M) Mid-position: post-5 but pre-V, i.e. A between the subject and the lexical verb.
No distinction is made here between different mid-positions.
Ί have lived where you never could have survived;' (M2.97)
'Keats, Keats, Keats', he said spitefully. Ί don't care if I never see another keat
again.' (M2.101)
(E) End-position: post- V, i.e. after the lexical verb. If there is a C element present,
it makes no difference if A occurs before or after the complement.
The dog was no longer sitting in the porch. (Ml.23)

3.33 Agent

The different functions of A (such as indicating place, manner, time, etc.) are not
dealt with here, with one notable exception: the agent. 'Agent' is defined as that
element in passive clause structure whose nominal part may operate as S in an active
clause transform. The agent (denoted by Ag in contradistinction to Ad for nonagen-
tive adjuncts) is usually a prepositional phrase with by, but there are, as we shall see,
other prepositions with similar functions. This definition of 'agent' is based on
potential syntactic function alone and involves no semantic requirement such as
agent being 'actor' in contrast to subject being 'patient'. 8


Internal clause relation, i. e. linear ordering of clause elements, has been restricted to
the cardinal elements SVC, since A position is recorded in the description of the A
element (see Section 3.32). Among the major categories are the following (C elements
are italicized in the illustrative examples):
(1) SV order.
(la) SV(C), which is the normal, unmarked order:
Raleigh smokes
Socrates is wise. (K.143)
As the Prime Minister left, M.P.s gave him a massive demonstration of personal
loyalty. (E2.5)
(lb) CSV:
'Holliday knows what a glass of beer is, but champagne he does not understand
in the slightest degree.' (E2.53)
(2) VS order.
(2a) VS(C), where the finite verb precedes S:
'Can't you fetch your gardenerΤ (Μ 1.26)
(2b) CVS:
' You mean he's mad', said Treece. (M2.104)
(2c) Discontinuous C:
She smokes, thought Madeleine, like a chimney. (Ml.21)


External clause relation comprises three main types: syntactically bound, sequentially
related, and free clauses. These types, each with several subtypes, should be seen only
as convenient points of division on a scale of external clause relationships.
• The term 'agent* has been used in many different ways (see, for example, Green 1913, 1914;
Bloomfield 1933, 371; Hendriksen 1948); Koumari (1956, 51) goes as far as to deny its justification
as a category. The present use of 'agent' as a technical term, corresponds most closely to Jespersen's
'converted subject' (1909-49, 3.317ff.) and Sweet's 'inverted subject' (1955, 1.113). The agent will
be further discussed in Section 5.81 (pp. 102ff.).

(1) Syntactically bound clauses ('a' are immobile, 'b' are so with zero, and 'c' are
usually mobile within the sentence).

(la) Relative clauses:

Model (c) was immediately discounted by the neutron diffraction data since
there was no trace of the superlattice lines which would be required by the
enlarged unit-cell. (J2.175)

(lb) Conjunctional that- or zero ί/ιαί-clauses:

Still less do I say that grammatical substantives in general are such expressions.
(lc) Other conjunctional clauses (if, before, when, etc.):
When they reached cruising altitude, five minutes later, all the stars were already
there to meet them. (M4.49)

(2) Sequentially related clauses.

(2a) Fixed relator {and, but, etc.):

'Your soul rests easy, but nothing's solved.' (M2.105)

(2b) Fixed relator + elliptic subject:

He half-rose and gave an expansive gesture with his hand, ... (M2.96)

(2c) Fixed relator - f - elliptic subject + elliptic auxiliary:

but everybody had done something to the garden and made a mess of //.(Ml. 17)

(2d) Mobile relator (however, moreover, etc.):

'Are there no men in this village? Can't you fetch your gardener?'
'He's seventy-three,' said Madeleine. ''Besides, he's gone to the football match.
I think everybody has.' (Ml.26)

(2e) Elliptic subject:

/after all :mathematics . s :is thinking# . /is ldgic# /is reasoning# (B2.7)

(2f) Tag questions:

'He's a sporting dog, isn't he?' (M1.24)

(3) Free clauses or, strictly, 'more free than the previous types', including clauses

(3a) have a clause as C element:

he wished that he could tempt her to say the same thing about him. (M2.101)
(3b) have direct speech as C element:
Again she said: 'We'd better go away.' (Ml.24)
(3c) are followed by a sequentially related clause:
nothing was resolved and there were no firm rocks to settle on. (M2.91)
(3d) are 'fully free', i.e. other than 3a-c:10
What does one do with dustbins to make them look interesting? (M2.97)


Having surveyed the individual clause elements, we shall now consider the major
clause types which are formed from them. In dealing with the clause, we adopt the
same approach as for the verbal group, i.e. we first set up central types, to which
peripheral types can be related. For this reason we leave out prepositional phrases,
and indeed A elements generally, because they constitute a scale, the points of which
should ultimately be related to the central clause types.


A major dichotomous division is made between EXTENSIVE and INTENSIVE clauses.11

Intensive clauses (Type INT for active, int for passive) are those clauses between two
of whose nominal elements there is a relation of coreference. In extensive clauses
(Type EXT for active, ext for passive) no such SYSTEMIC coreference relation obtains.
Both types may, however, have EXPONENTIAL coreference, in which case the intensive
type displays three, and the extensive two, coreferent clause elements. Exponential
coreference occurs particularly with (i) reciprocal pronouns, (ii) reflexive pronouns,
and (iii) nominal groups with possessive premodification:

(i) Their eyes, like two advance posts of opposing armies, crossed each other ...
(ii) He pulled himself up with the little groan of protest allowed to the middle-aged
and successful. (M4.48)
He considered himself a good player.
(iii) He took his leave.
This point does not, however, represent the end of the scale of external clause relation. Other
features, such as finitude in a clause element implying grammatical reference, for instance anaphoric
the (cf. Smith 1964, 51), are also relevant, and will be considered in the taxonomic experiment in
Chapter 5 (see Criteria S 51-2, pp. 99f., and Ag85-6,p. 107). The lack of analytical delicacy in the
voice data processing discussed in Chapter 4, as compared with the taxonomic experiment, reflects
the natural progression of this enquiry as well as computational limitations and a desire for a
comprehensive initial approach.
The terms 'extensive' and 'intensive', but not the definitions given here, derive from Μ. A, K.
Halliday. For 'exponence', see Halliday 1961, 268ff.

The distinction between systemic and exponential will simplify the description at
various points. The following seven subsections will illustrate the difference between
extensive and intensive clauses.

3.71 Copula
Systemic coreference between two Ns may be demonstrated by the fact that they can
be connected by a copula (a closed list of verbs with the lexically unmarked be as its
central exponent, see Section 5.51, pp. 90ff.).
Type INT: He considered her a friend: She was a friend.
Type EXT: He gave her an apple: *She was an apple.18

3.72 Gender/number class

Consequently, the Ns in Type IN Τ must belong to the same gender/number class,
whereas there is no such constraint for Type EXT.13

Type INT: [ He is a teenager.

They are teenagers.
N&n/sg + N^an/sgl
N-ycmjpl + N2an/pl J
He likes the car. Νχαη/sg Njn/sg
cars. N2in/pl
Type EXT: • + · • + •
the girl. N^an/sg
They like girls. Νχαη/ρΙ N^anjpl

Deviations from this will be considered as metaphors, for example:

The computer is his best friend. N^n/sg + N^flnjsg.
There is however an exponential exception to this criterion: it does not apply to
'divisible' Ns, i.e. nouns which do not require agreement across the copula (cf.
Stockwell & Schachter 1962).
John's problem is headaches.
The flies are a nuisance here.
John and Mary are a couple /*a child/.

3.73 Gender-selection
In question transforms Type EXT clauses have what with Ν in and who with Nan,
whereas in Type IN Τ clauses the wA-form is not gender-selective in case of the latter
coreferent Ν:
It is structural ambiguity between Types EXT and INT which make possible such linguistic jokes
as * You are a cab, Sir Γ in reply to the order: 'Call me a cab Γ
an — animate, in = inanimate, sg = singular, pi = plural. Braces indicate that any com-
bination of pairs of sequences in the braces is possible (cf. Bach 1964,17-8). Subscripts are used for
labelling elements in clause structure: Nl VNt->N, (V, etc. Superscripts denote number of
element items: C' = two Cs, A*· = four As, etc.

Type EXT: He gave her a book. J Who/m/ did he give a book /to/?]
[What did he give her? J
Type INT: He considered her iWho/m/ did he consider a pretty girl?
a pretty girl. [What did he consider her?
In addition, Type INT often takes question-adverb here:
He considered Bleak House a great novel.

What did he consider Bleak House?
How did he consider Bleak Hornel

3.74 Voice transformation potential

Types EXT and INT may have different potential in regard to number of transforms:
(N, V N2 N3) He gave her a book.
(Ν1 VN3pN2) He gave a book to her.
Type EXT: (N2 WN3) She was given a book.
(Nz W Ν2) A book was given her.
W ρ N2) A book was given to her.
(NI VN2 N3) He considered her a pretty girl.
Type INT:
WNa) She was considered a pretty girl.

3.75 Exponence
Type INT, furthermore, takes a range of exponents different from EXT. (In the
N1 V Ν2 Ν3 structures below, Ν is used here for nominal groups and nominalizations
as well as other complements.)
Noun) He considered her a friend.
Adj) He considered her kind.
Num) He made it three.
V) He saw her leave.
to V) He asked her to leave.
to be Adj) He considered her to be kind.
Ving) He saw her leaving.
Ved) He found the paper neatly folded.

3.76 Deletion
Type IN Τ seems never to have N2 deletable without structural change:
Type EXT: He gave her a book: He gave a book.
Type INT: He considered her a nuisance: *He considered a nuisance.
(Some deletions do not come altogether naturally and may best be described as
ellipses. Even so, the criterion seems valid since ellipsis is not possible with the
Type INT example above.)


Returning to the clause types, we may, in addition to Types EXT/ext and INT/int,
distinguish clauses according to the number of cardinal structural elements other
than SV (i.e. in declarative clauses normally postverbal C elements): none (0), one
(C1), or two (C2). Tables 3:2-3 show the major clause types according to these two
parameters. Table 3:2 represents the major clause types from the point of view of
constituent structure. The same clause types may also be stated in terms of their
voice relationships. In Table 3:3 the S and C labels have been replaced by Nx, N2,
etc., in order to show the systemic voice transformation potential of the clauses.
Transformation is however closely linked with exponence, and the potential shown
in the table occurs only with certain Ν exponents. (Systemic coreference is indicated
by ΓΤ, active verbal group by V, passive by W. Agents, which are usually optional
clause elements, are given in round brdckets.)


Major clause types (constituent structure)



EXT-0 SV ext-0 S W (Ag)

ext-C SWC (Ag)
EXT-CC i K Q C ,

INT-C s~v~c int-C SWC (Ag)


Major clause types (transformational structure)



EXTENSIVE EXT-C ΛΤ, V Nt <—• ext-0 ΛΤ, W(p N,)
EXT-CC Νί V N% N, ext-C' Ν, W Nt (ρ Ν,)
ext-C" Nt W N3 (p Nr)
INT-CC Nr V Ν*Ν* int-C Nt W Nt (pNJ
(.EXT-0) Active intransitive : He departed.
(.EXT-C) Active transitive : He liked her.
(.EXT-CC) Active ditransitive :14 He gave her a book.
(.INT-C) Active equative : He was a tall man.
(INT-CC) Active factitive : He considered her a beauty.
{ext-0) Passive transitive : She was liked {by him).
(ext-C') Passive ditransitive' : A book was given her (by him).
(ext-C") Passive ditransitive" : She was given a book (by him).15
(,int-C) Passive factitive : She was considered a beauty (by him).
As the names imply, the clause labels have been chosen with a view to indicating
transformational rather than constituent structure, taking actives as bases.


Having established the highest degree of clause rank abstraction, we may go on to

assign different values to C.
In extensive clauses all C elements are 0 (object). If there is more than one O,
they are differentiated as O* (indirect object) and Od (direct object), the former being
the Ο which generally admits a transform with a prepositional phrase (and reversal
of the elements):
They gave him a book. -* They gave a book to him.
Oi is typically animate and the order of elements in the clause Ot + Od- Prepositional
phrases as transforms of indirect objects in active (They gave a book to him) or passive
clauses (A book was given to him) will not, however, be considered indirect objects.
'To call such expressions as to the boy an "indirect object" in the sentence the man
gave the money to the boy, leads to confusion. The expression to the boy does express
the same meaning as that of the indirect object, but this meaning is signalled by the
function word to, not by the formal arrangement which constitutes the structure,
"indirect object"'. The subject 'in the sentence the boy was given the money also
expresses the same meaning as that of the indirect object, but we rightly call it "sub-
ject", not "indirect object'" (Fries 1959, 185).
In intensive clauses, the latter of two systemically coreferent nominal elements will
be called Ρ (predicative; cf. Jespersen 1933, 321. 'Predicate', on the other hand, is
used for the entire predication). The former coreferent element is S in Types INT-C
and int-C; in Type INT-CC it will be called 0/S. In this last clause type, C1 functions

'Ditransitive', and 'equative' (or 'copulative') are terms used in tagmemics (see, for example,
Longacre 1964). For 'factitive', see Sledd 1959.
The two passive ditransitive transforms may be distinguished as 'primary' (ext-C) and 'secondary'
(ext-C"). These terms are adapted from 'primary' and 'secondary conversion' (see Poutsma
1926-9, 2.2.124ff.).

both as 0 of the previous V and S of C2, if this is a nonfinite verb clause (J saw him
come)', verbless constructions ( / f o u n d her unhappy) may be considered as truncated
verbful constructions (/ found her to be unhappy, etc.) which have equative relation
between the nominal elements.
Element C has been defined above as that nominal element which is essential to
complete the basic clause structure. The further distinction which is made here
between Ρ and Ο is, again, one between two poles of a scale: Ρ elements are generally
more essential than Ο elements. 'Copulas', which will be used for V in equative
intensive clauses, 'constitute a point in the structure of language at which the CD
[communicative dynamism] of the verb is at its weakest, and its need of a semantic
amplification at its greatest. They seem to set up one end of a gamut, one end at
which we find verbs with the lowest possible amount of CD. The other end of the
gamut would be made up by verbs performing the function of rheme proper' (Firbas
1959, 46).



The processing of voice categories comprised the following steps (see also Figure 4:1):

Text Ml 1 Class α Minor

CORPUS Text M2 | Voice Corpus Class β • Passive
Text J l J Class γ Corpus

DATA Linguistic categories abstracted from 2106 clauses in the corpus.

INPUT Data categories encoded into tape code characters in Columns 1-7 and
lexical verbs in Column 8.

Program 1 Program 2 Program 3 and Program 4
(inventory) (two-way correlation) (three-way correlations)
Program 5 Program 6
(tabulation) (alphabetization)

OUTPUT Tape code characters listed or tabulated for Columns 1-7 and lexical verbs
given in alphabetical order for Column 8.

RESULTS Interpretation of the output stated in terms of linguistic categories.

Fig. 4:1. Procedure of the voice data processing experiment.

(a) Selecting the corpus, (b) Abstracting the linguistic data from the corpus,
(c) Preparing the computer input from the data, (d) Writing the programs,
(e) Processing by computer, (f) Converting the output into linguistic categories,
(g) Analysing and presenting the results.

The input to the computer was not 'raw' linguistic data but carefully selected
categories which had been abstracted from the corpus. This experiment might be
described as 'linguistically sophisticated' but 'computationally trivial'.

4.11 Corpus

The analysis covered two sets of material described in Sections 1.4-5 (pp. 6 if.),
Corpus I (the Voice Corpus) and Corpus II (the Minor Passive Corpus), and the data
derived from them were processed by using the same computer programs though in
some cases with different significances attributed to the characters.
The Voice Corpus consisted of three texts, two novels (Ml and M2) and one
scientific text (Jl). This corpus was chosen to give a general picture of some voice
categories and to provide a background for the more detailed subsequent analysis
of the passive term. All finite verb clauses in these texts were analysed but, because
of the restricted storage of the available computer, we excluded from computer
processing the highly frequent equative clause type (INT-C; see Section 3.8, pp. 36f.)
with be as exponent of V, as in

The mouth is a small opening above the tongue and leads into a large buccal cavity.

The high proportion of active clauses in this limited material made it less useful for
a detailed analysis of the passive. For this purpose the Minor Passive Corpus was
used for an analysis conducted on similar lines. Since the texts making up the Voice
Corpus were also part of the Minor Passive Corpus, all detailed statements about the
passive term can best be made on the basis of this passive corpus. It consisted of
eleven texts, three spoken (Bl, B2, B3), four scientific texts (Jl, J2, J3, J4), and four
novels (Ml, M2, M3, M4). In contrast to the Voice Corpus, the data for the Minor
Passive Corpus were not processed text by text but were divided into three classes
according to whether the agent was expressed ('agentful passives') or not ('agentless
passives'), and according to the character of the agent. All the agentless passives
were, however, 'agentive', i.e. admitted agent-extension. The agentful category was
divided into one class with animate and into another with inanimate agents:
Class α = agentful with animate agent,
Class β = agentful with inanimate agent,
Class γ = agentless (but agentive).
For practical purposes, all expressed agents were usually treated together (Class
α-β). The taxonomic experiment (see Chapters 5-6) later confirmed that this was also
linguistically reasonable.
Excepting anacolutha (two cases in Text M l and one case in Text M2), the two
corpora yielded the following number of clause entries:
Text M l : 561 clauses Class a : 68 clauses
Text M2: 734 clauses Class β: 85 clauses
Text J l : 355 clauses Class γ: 497 clauses
Total: 1650 clauses Total: 650 clauses

4.12 Data

The following linguistic categories were abstracted from the two corpora and encoded
into eight columns as input (the maximal number of characters representing linguistic
categories in each column are given in square brackets):
Column 1: Type of finite verbal group [71 characters].
Column 2: Exponent classes of the subject and the agent (S and Ag elements) [26 characters].
Column 3: Exponent class of the complement (C element) [63 characters].
Column 4: Exponent class of the adjunct (A element) [62 characters].
Column 5: Clause type [28 characters].
Column 6: External clause relation [30 characters].
Column 7: Order of clause elements [23 characters].
Column 8: Ultimate exponent of the lexical verb.

4.13 Input and output

Each set of data input was an array of eight columns and at most 1024 (210) rows.
One set of input (Text M l ) began as follows:


Entries in the first seven columns could be any of about 80 characters drawn from the
Ferranti seven-track tape code. The entries in Column 8 were strings of upper case
letters and spaces. Spaces were ignored in input; the maximum number of letters
permitted was 16. These strings were the ultimate exponents of lexical verbs. They
were not stored in the computing store, but in the backing store, and could be re-
trieved when necessary. The input was read in by the four primary programs in the
same way, one row at a time, backspacing and erasing being permitted. The output
had to be on five-hole tape, and consequently many of the output characters were
different from the corresponding input characters, for example 'a' became Ά ' , ψ
became Ή Γ , and, to avoid confusion with the numerals used in totals, Τ became
*N1\ etc.

4.14 Programs1

Six programs were used, four primary (Programs 1-4) and two secondary (Programs
5-6). The primary programs had as input the data encoded in eight columns, whereas
I am greatly indebted to my colleague Η. T. Carvell and to A. J. T. Colin, University of London
Institute of Computer Science, for their help in producing the programs. On the use of computers
in linguistics, see for example Lamb 1961, Kuöera 1962, Garvin 1963.

the secondary programs had as input the output of the primary programs. Program
5 ('the Tabulation Program') was used for the output of Program 2, and Program 6
('the Alphabetization Program') was used for the output of Programs 3 and 4, for
copying headings and rearranging listed verbs in alphabetical order.
Program 1 produced an inventory of the characters in each of Columns 1-7. (It
was realized afterwards that it would have been substantially more convenient to
have the characters in each column listed in a pre-assigned and linguistically mean-
ingful order, including those characters that did not occur in all sets of the data.)
Program 2 listed the correlations between all 21 pairs of Columns 1-7, the output
order being governed by the input order. The output was tabulated in a linguistically
convenient order by Program 5, which also produced row and column inventories
and column totals, which were useful both for purposes of analysis and as a constant
check on the output.
Program 3 had as object to list the lexical verbs of clauses with certain specified
characteristics. For example, since the verbs with animate subjects were required,
part of the output (after alphabetization by Program 6) began as follows:
COL2 - N l , N2, N3,
For the program, this meant that ADMIRE in Column 8 cooccurred with one or
other of Nl, N2, N3 in Column 2 twice; that ADVANCE occurred similarly once,
and so on. Since the codings were
N l in Column 2 = pronominal animate subject,
N2 in Column 2 = nominal animate subject,
N3 in Column 2 = coordinate animate subject (of any form-class),
this gave the required information. It will be seen that, in effect, Program 3 needed
an additional set of data, namely a specification of all the sets of characters (for
example Nl, N2, N3) that were required, together with the appropriate columns.
In practice, the additional data were incorporated into the program itself.
Program 4 was similar to Program 3 but a little more ambitious. It caused the
computer to print what occurred in both Column 8 (lexical verbs) and a specified
column of the first seven when any one of some given list of characters occurred in
another given column. In fact there were a number of lists for Column 2 when the
cooccurring characters in Column 3 were asked for and similarly a number for
Column 3 when the cooccurring characters in Column 5 were asked for; and all
these lists were incorporated into the program itself as a matter of convenience.
The verbs were again arranged in alphabetical order by Program 6.

4.15 Computation
The computer used in these experiments was a Ferranti Mercury at the University of
London Institute of Computer Science.2

4.16 Results

The analysis of the results was made difficult and laborious by the profusion of de-
tailed output. For Programs 2 + 5 alone there were 105 tables, some of which had
more than 2000 cells each. It might in similar circumstances be useful to have a
program to select the significant correlations and, also, to have some method for
'conflation', i.e. forming larger groups from the small correlations. There is, however,
a danger of losing interesting information if this is done before the general picture
is known.
The presentation of the results is in no way complete, partly because this would
require too much space, partly because a large portion of the results are peripheral
or irrelevant to this study. Most of the present results are accounts of the major
linguistic categories in Columns 1-7. They derive from the inventories produced by
Program 1, except when active categories have been separated from passive in the
Voice Corpus, in which case the output of Programs 2 + 5 was used. 'Actives' refer
to the actives of Texts Ml, M2, and J1; 'passives', unless otherwise stated, to Classes
α-β and γ of the Minor Passive Corpus.
Since the texts and classes have unequal numbers of clause entries, it will be helpful
to use percentages for easy comparison. Percentages, given in round brackets under
the absolute numbers in the tables, are rounded off to the nearest integer, with the
single exception that nonoccurring categories are indicated by dots. Thus 1.7% is
given as (2%), 0.3% as (0%), and no occurrences as (.); also, 0.5%, 1.5%, 2.5%,
etc. are given as (1 %), (2%), (3%), etc.3


The categories have been outlined above in Chapter 2. 'Present' and 'past' refer to
all first elements in verbal group structure; 'simple present' and 'simple past' to 0 and
0 respectively.
It seemed generally true that frequency of occurrence decreased with increasing
The size of the computing store was 2,048 words of 20 bits, of which half could be used as a store
of 2,048 words of 10 bits. A larger computer would of course perform these tasks more conveniently.
For comparison it may be mentioned that the size at present (1965) of the computing store of the
Atlas at the Institute is over 32,000 words of 48 bits, and, in addition, Atlas is a very much faster
computer than Mercury.
* It was found convenient to use logarithmic graph paper for presenting some of the diagrams.
This has been indicated by 'log.graph'. The paper has been used in a slightly unusual way, in that 0
(per cent or occurrences) has been plotted against the base line.

Distribution of verbal group types in the Voice Corpus

STRUCTURE TYPE TEXT Ml t e x t M2 t e x t J1 total

Simple 0/0 340 434 215 989 989

(61%) (60%) (61%) (60%) (60%)

a/A 94 163 17 274

(17%) (22%) (5%) (17%)
b/B 38 40 7 85 543
(7%) (5%) (2%) (5%)
1 element (33%)
c/C 9 34 1 44
(2%) (5%) (0%) (3%)
d\D 24 31 85 140
(4%) (4%) (24%) (8%)

abf AB 8 9 0 17
(1%) (1%) (·) (1%)
actAC 0 3 0 3
(.) (0%) (·) (0%)
adjAD 6 7 19 32 80
2 elements (1%) (1%) (5%) (2%)
bc/BC (5%)
3 6 0 9
(1%) (1%) (•) (1%)
bd/BD 8 1 8 17
(1%) (0%) (2%) (1%)
cd/CD 0 0 2 2
(.) (.) (1%) (0%)

abc/ABC 0 0 0 0
(.) (.) (·) (.)
abd/ABD 2 0 1 3
(0%) (.) (0%) (0%)
3 elements (0%)
acd/ACD 0 0 0 0
(·) (.) (.) (.)
bed/BCD 0 0 0 0
(.) (.) (.) (.)

4 elements abcd/ABCD 0 0 0 0 0
(.) (.) (.) (.) (·)

Other Imperative 29 6 0 35 35
(5%) (1%) (·) (2%) (2%)

Total 561 734 355 1650

(100%) (100%) (100%) (100%)

structural complexity in the verbal group. Thus, the distribution between active and
passive forms was uneven, since the passive, being Type d in the verbal group, has
one more structural component than its corresponding active forms (cf. Types 0'.d,

60 ι


Fig. 4:2. Verbal group types in the Voice Corpus

(Percentages based on 1650 occurrences for all three texts. Log. graph.).

a :ad, b\bd, etc. in Section 2.2, pp. 10f.). The number of occurrences for both voices in
texts Ml, M2, and J1 are shown in Table 4:1 and Figure 4:2.
The distribution of active and passive forms varied considerably for the different
texts of the Voice Corpus. Table 4:2 provides evidence that the distributional differ-
ence between Ml and M2 as regards voice is real but very small. However, between
either of the Μ texts and J1 the degree of association is high, and the evidence is
extremely significant (ρ < 0.001, i.e. the chance of occurrence is less than 1/1000, as
determined by the χ 2 test; see Section 6.32, pp. 116ff.).
Voice distribution for the texts of the Voice Corpus

Active 521 695 240 1456

(93%) (95%) (68%) (88%)

Passive 40 39 115 194

(7%) (5%) (32%) (12%)

Total 561 734 355 1650

(100%) (100%) (100%) (100%)

In the Voice and Minor Passive Corpora only Types 0, 0, a, A, b, B, c, C, d, D,

ad, AD, bd, and BD, had frequencies of 2% or more in any one text or class. Types
ab, AB, AC, be, BC, cd, CD, abd, ABD had less than 2% occurrences. No instances
were recorded of the active types ac, abc, ABC, and their corresponding passives acd,
abed, ABCD, nor of ACD, bed, or BCD (see Table 4:3 and Figure 4:3).
There was in the material a negative correlation between Type c/C and the passive,
not only for cd/CD (both 1 % or less) but also for the combinations acd/ACD and
bed/BCD (all nonoccurring). The increasing effect of Types a/A and b\B with the
passive (see next paragraph) seems absent here. The number of cooccurring c/C
and d/D was low: 2/192 is less than 56/376 (see Table 4:4). The Minor Passive
Corpus had an even lower ratio for c/C + d/D as against ~ c/C + d/D (5/642).
Contingency table for Types c/C and d/D in the Voice Corpus
p = 6 x 10- 8

c/C ~ c/C TOTAL

d\D 2 192 194

~ d/D 56 376 432

58 568 626

~ = 'not'.

T A B L E 4:3

Distribution of verbal group types in the Voice Corpus (active only) and in the
Minor Passive Corpus



Ml M2 J1 α-β Y

0 47 136 213 396 d 49 156 205

(9%) (19%) (90%) (27%) (31%) (31%) (32%)
0 293 298 2 593 D 44 124 168
(56%) (43%) (1%) (41%) (29%) (25%) (26%)
a 49 87 15 151 ad 19 78 97
(9%) (13%) (6%) (12%) (12%) (16%) (15%)
A 34 56 1 91 AD 5 55 60
a%) (8%) (0%) (6%) (3%) (11%) (9%)
a-A* 11 20 1 32 ad-AD* 0 15 15
(2%) (3%) (0%) (2%) (·) (3%) (2%)
b 16 18 7 41 bd 29 43 72
(3%) (3%) (3%) (3%) (19%) (9%) (11%)
Β 22 22 0 44 BD 3 16 19
(4%) (3%) (.) (3%) (2%) (3%) (3%)
c 3 15 1 19 cd 1 3 4
(1%) (2%) (0%) (1%) (1%) (1%) (1%)
C 6 19 0 25 CD 0 1 1
(1%) (3%) (.) (2%) (.) (0%) (0%)
ab 2 0 0 2 abd 1 1 2
(0%) (·) (·) (0%) (1%) (0%) (0%)
AB 6 9 0 15 ABD 1 3 4
(1%) (1%) (.) (1%) (1%) (1%) (1%)
ac 0 0 0 0 acd 0 0 0
(.) (.) (.) (.) (.) (.) (.)
AC 0 3 0 3 ACD 0 0 0
(.) (0%) (.) (0%) (.) (.) (.)
be 3 4 0 7 bed 0 0 0
(1%) (1%) (.) (0%) (·) (·) (.)
BC 0 2 0 2 BCD 0 0 0
(.) (0%) (•) (0%) (.) (.) (·)
abc 0 0 0 0 abed 0 0 0
(.) (.) (.) (·) (.) (.) (·)
ABC 0 0 0 0 ABCD 0 0 0
(.) (.) (.) (.) (.) (.) (.)
Imperative 29 6 0 35 Imperative 1 0 1
(6%) (1%) (·) (2%) (1%) (.) (0%)
Subjunctive 0 0 0 0 Subjunctive 0 2 2
(·) (.) (·) (·) (.) (0%) (0%)

521 695 240 1456 153 497 650

Total Total
(100%) (100%) (100%) (100%) (100%) (100%) (100%)

* Hyphens between types denote non-tense-marked auxiliaries (see Section 2.34).


Percentages based Percentages
on 1456 occurrences. based on 650

Fig. 4:3. Verbal group types: active (Voice Corpus) and passive (Minor Passive Corpus).
Log. graph.
If we consider the passive and active verbal group types separately (i.e. with and
without Type d j D, as shown in Table 2:1, p. 11), we find that the passive have
distributions similar to the corresponding active. The exceptions were Types bd (19 %
in Class α-β and 9% in γ), and ad/AD in Class γ (30%), all of which were more
frequent in the passive material than the corresponding actives in the active material.
In interpreting the figures for the passive corpus, it must be remembered that one might
expect them to bear closer resemblance to those for the corresponding actives in Text
J1 than in Texts Ml and M2, since the passive is particularly frequent in scientific texts,
which yielded the majority of the clauses in the Minor Passive Corpus. This would help
to explain, for instance, the greater number of Type d than Type D occurrences, since
simple present was strikingly more common than simple past in Text J1 (213 as
against 2 occurrences). In the case of the categories just mentioned there is, however,
no reason for suspecting that the high passive frequencies were text-conditioned
rather than voice-conditioned: the percentage for Type bd was 19 in Class α-β, com-
pared with 3 for Type b in Text J l ; and it was 30 for Type ad/AD in Class γ,
compared with 7 for Type a/A in Text Jl. In fact, in this case the similarity was
greater with Texts Ml (18%) and M2 (23%). Discounting the auxiliary do, which
cannot occur in the 6e-passive (except in imperatives of the type Don't be deceived
into doing it!) the percentage for Type a/A was much higher in the passive than in
the active: 26% (172 out of 650 cases) for Type ad/AD, compared with 19% (274
out of 1456 cases) for Type a/A. In view of the fact that the number of active clauses
in the Voice Corpus was more than twice that of passive clauses in the Minor Passive
Corpus, the following auxiliaries in Table 4:5 seem particularly frequent in the passive
verbal group.


Occurrences of Type a/A auxiliaries


can 46 32
may 22 8
could 25 20
have to
and other open
class auxiliaries
(see Section 2.33) 28 22


(5 and Ag elements)

Subject and agent categories, only some of which will be distinguished here, have
been outlined in Sections 3.2 (pp. 26if.) and 3.33 (pp. 30f.). (In the code, the same .S
characters were used for both actives and agentless passives in the Voice Corpus, but

the distribution of the different categories within the two voices could be obtained
by correlating, for instance, Columns 2 and 5. Agentful passives were, however, given
specific characters for different combinations of S and Ag categories.)

4.31 Subject gender

There were about twice as many animate as inanimate subjects in the Voice Corpus
(1072 : 539). Taking the texts and voice terms separately, however, it appears that
the proportions are drastically different. The frequency of inanimateness in the sub-
ject displayed a certain parallelism with passive frequency in the texts:

Text M l : 21 % inanimate subjects and 7% passives

Text M2: 12% inanimate subjects and 5% passives
Text J1: 97% inanimate subjects and 32% passives

The ratios for passive clause subjects indicate that, compared with active, they have
a tendency towards inanimateness. When both text and voice factors combined, as
in the case of Text Jl, which had no animate but 115 inanimate passive subjects,
the result is particularly striking. The relative proportions of each gender remained
constant for both voices in the three texts: M2 had most, Ml less, and Jl least animate
subjects in both active and passive.

Voice and subject gender in the Voice Corpus
(p < 10""; for p, see pp. 116ff.)


Active 1035 383 1418

Passive 37 156 193

Total 1072 539 1611

The association was exceedingly high between passive voice and inanimate subject
as can be seen in Table 4:6 where all three texts in the Voice Corpus are brought
together in a contingency table. This association appears even more remarkable in
view of the fact that there is no significant or even remotely significant textual differ-
ence between Jl on the one hand and Μ1-2 on the other in the way in which they
associate San/Sin with active/passive (see Table 4:7a, b, c).
In the Minor Passive Corpus, more than f of the clauses had inanimate subjects
(see Table 4:8).


Voice and subject gender in Texts Ml, M2, J1


Active 406 85 491

p< 10- 1 1
(a) TEXT M l Passive 14 26 40

Total 420 111 531


Active 617 70 687

p< 10-'
(b) TEXT M 2 Passive 23 15 38
Total 640 85 725


Active 12 228 240

(C) TEXT J 1
ρ = 0.008
Passive 0 115 115
Total 12 343 355

4.32 Agent gender

In the Minor Passive Corpus there were more inanimate than animate expressed
agents ( = Classes β and α respectively), and animate subjects were even more infre-
quent in both classes (see Table 4:8).


Subject gender in the Minor Passive Corpus


Class α 9 59 68
(Agon) (13%) (87%) (100%)

Class β 5 80 85
(Ag in) (6%) (94%) (100%)
OO 00

Class γ 400 488

(82%) (100%)

Total 102 539 641

(16%) (84%) (100%)

4.33 Subject form-class

As was the case with gender, Texts Ml and M2 were very similar in respect of subject
form-class. Both had f pronouns and J nouns as subjects with actives and passives
taken together, whereas for Text Jl, these figures were reversed in the active and
there were | nouns and \ pronouns in the passive. For Class γ the proportions were
f pronouns and $ nouns. (Classes α and β were not fully distinguished with regard
to form-class.) The distribution of pronominal and nominal subjects was roughly
parallel in the active and passive clauses of Texts Ml, M2, and Jl.

4.34 Agent form-class

The predominant agent form-class was nouns. Only 3 out of 153 agents had pro-
nominal exponence.

4.35 Coordination

Coordination was rare in the subject (8 cases in Text Ml, 5 in Text M2, 2 in Text Jl),
but it was much more frequent in the agent. There were 39 cases there, 34 of which
occurred in Class a ; i.e. more than half the agents in this class were coordinate.

4.36 Length of passive subject and agent

Since there was little variation in Classes α, β, and γ for the categories described in
Column 7 (order of clause elements), this column was also used for registering the
length of the subject, measured in terms of number of words. (Hyphenated items
counted as single words.) Elements consisting of 1-10 words were given precise values,
those consisting of 11 or more words had to be grouped together in five groups. The
number of subject elements with more than 10 words was small: 18 in Class α-β and
26 in γ. The lengths of agent elements in Class α-β were counted manually. See
Figures 4:4 and 4:5.
The average length of the subjects in agentless passive clauses was just over 3
words (arithmetical mean = 3.2), and almost 5 words (mean = 4.96) in agentful
clauses. The mode, i.e. the most common word-length, was 1 word (45 % in Class
γ, and 28% in α-β; the next most common, 2 words, had percentages of 18% and
19%, respectively). Agents had an average length of 8.5 words. The mode was 2
words. For both elements, the norm seems to be: decreasing frequency with increasing
word-length. The figures for up to 10 words, covering 95% of the subjects in Class
γ, and 88 % of the subjects and 73 % of the agents in α-β, are shown in Table 4:9 (p. 56).
These figures suggest firstly, that there is considerable difference in weight, measured

2 3 4 5 6 7 8 9 1 0 11 12 13 1 4 1 5 1 6 17 18 19 2 0 2 1 2 2 2 3 2 4 ^ 5 jjj
NUMBER O F W O R D S '—·— « " §

Fig. 4:4. Subject-lengths in Class γ. (Log. graph.)


Fig. 4:5a. Subject-lengths in Class α-β. (Log. graph.)

in terms of word-length, between subjects and agents, in that the latter tend to be
much longer (the ratio of 1-word subjects to 1-word agents was 5 : 1 ) ; secondly,
that the subjects in agentful passive clauses are longer than subjects in agentless

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28
Fig. 4:5b. Agent-lengths in Class α-β.
(Log. graph.)


(C element)

60 categories besides C absent and anacolutha were encoded: 29 C 1 (simple comple-

ment) and 32 C 2 (double complement). For illustration of C elements, see Section 3.2
(pp. 26 ff.).
4.41 Complement Distribution

Table 4:10 shows the overall distribution of complements in both corpora. Text J1
differed from Ml and M2 in having fewer complements, in particular fewer C2.

4.42 Complement form-class

In active clauses exponents of simple complements were, in order of frequency:

nouns (38%), direct speech (22%), pronouns (20%), finite verb clauses (9%), non-

Number of words in subjects and agents of the Minor Passive Corpus


0 9 2 —

1 223 42 8
2 93 30 27
3 47 15 11
4 27 8 21
5 20 11 8
6 12 9 11
7 17 8 5
8 11 7 7
9 6 1 8
10 6 2 6

Total 471 135 112

Other 26 18 41
(i.e. over 10)
Total 497 153 153
Mean 3.2 words 5.0 words 8.5 words
Mode 1 word 1 word 2 words

TABLE 4:10
Distribution of complements



C-less 179 130 105 138 405

(34%) (19%) (44%) (90%) (81%)
Complement 307 531 134 15 92
(C) (59%) (76%) (56%) (10%) 09%)
Complement 35 34 1 0 0
( C ) (7%) (5%) (0%) (·) (·)

TOTAL 521 695 240 153 497

(100%) (100%) (100%) (100%) (100%)

finite verb clauses (6%), and verbless constructions (5%). Texts Ml and M2 had
different distributions from Text J1: in the latter nouns were about twice as frequent
as in Ml and M2, and pronouns and speech were very much more rare, whereas
verbless constructions were more common (see Table 4:11).

TABLE 4:11

Form-class of simple complements (C1)



Noun 107 166 98 371 1 9 10

(34%) (31%) (73%) (38%) (7%) (10%) (9%)
Pronoun 76 116 2 194 0 1 1
(25%) (22%) (2%) (20%) (·) (1%) (1%)
Finite verb 23 62 6 91 3 29 32
clause (8%) (12%) (4%) (9%) (20%) (31%) (30%)
Nonfinite verb 14 34 8 56 9 45 54
clause (5%) (6%) (6%) (6%) (60%) (49%) (51%)
Verbless 18 16 17 51 2 8 10
construction (6%) (3%) (13%) (5%) (13%) (9%) (9%)
Speech 69 137 3 209 0 0 0
(22%) (26%) (2%) (22%) (·) (·) (·)

Total 307 531 134 972 15 92 107

(100%) (100%) (100%) (100%) (100%) (100%) (100%)

In passive clauses the order of frequency was: nonfinite verb clauses (51%), finite
verb clauses (30%), verbless constructions and nouns (each 9 %), and pronouns (1 %).
There were no instances of speech. Nonfinite and finite verb clauses were particularly
common, with 54 and 32 occurrences respectively, hence making up 80% of C 1
exponents in the passive. The to-infinitive {to V) was the most frequent (50 occur-
rences) among nonfinite verb clauses, and ί/ιαί-clauses among finite verb clauses.
Active clauses in the Voice Corpus as a whole had more zero that- than ίΑαί-clauses as
C (21 versus 45 occurrences), whereas Text J1 had only ίΑαί-clauses (6 that- versus no
occurrences of zero that-). The passive material had predominantly /Aai-clauses (27
occurrences as compared with 3 zero ίΑαί-clauses).

4.43 Complement gender

Inanimate complements were by far the most frequent everywhere, and the predomi-
nance was complete in the scientific text. In the active, Texts Ml and M2 had 33:274
(11:89 %) and 73:458 (14:86 %) respectively for animate:inanimate simple complement,
as compared with 0:134 for Text Jl. Texts Ml and M2 were similar in this respect,
but the distribution in these texts in favour of animateness, which was apparent in
the case of subjects (see Section 4.31), did not obtain for the complement. More than
$ of their complements were inanimate.


(A element)

The categories relating to the adjunct have been described in Section 3.3 (pp. 29ff.).
Column 4 registered only Ad elements (i.e. nonagentive adjuncts), since Ag (agentive

adjuncts) in Classes α-β were encoded into Column 2 (see Sections 4.32ff.). In the
presentation below, however, A comprises both the subclasses Ag and Ad. A elements
are considered either in terms of Ά sets' or in terms of Ά items'. A sets refer to overall
adjunct structure. Thus we have M-less' clauses, which have no A item, and M-fulT
clauses, which have one or more A items. Hence A items refer to the number of
individual As in each clause. A sets can be given as A1, A2, A3, A4, the superscript
denoting the number of A items in the A set. Frequencies for Texts Ml, M2, and J1
refer only to their active clauses.

4.51 Adjunct frequency

Less than half the active clauses in Texts Ml and M2 had one or more adjuncts,
whereas most of the active clauses in Text Jl, and most of the clauses in the Minor
Passive Corpus, had A sets. Class α-β had of course by definition one A item (Ag)
per clause; in addition, Ad sets occurred in half the agentful clauses (see Table 4:12
and Figure 4:6).
The most common A set type was A1, the next most common was A2, then followed
A3, and finally A*. In the material, A4 was the maximum number of A items in any
one clause. It was very rare: only seven occurrences in 2106 clauses. The frequency
of A sets decreased with the increasing number of A items in the set. (Cf. the negative
correlation between frequency and structural complexity in the verbal group men-
tioned in Section 4.2.)

4.52 Adjunct Position (see Table 4\12 and Figure 4:7)

With regard to the three major adjunct positions in the clause, front, mid, and end
(outlined in Section 3.32, p. 30), remarkable similarity obtained in both corpora
except for Class α-β, whose clauses had fewer front- and more end-positions than the
others. The agent is almost invariably end-placed. For Class γ and actives in Texts
Ml, M2, Jl the approximate proportions were as follows:
Front-placed adjuncts: ^
Mid-placed adjuncts: ^
End-placed adjuncts: ^
It is interesting to note that the agentful class had considerably fewer Ad items than
the agentless class. The proportions of Ad items in relation to clauses were 95 : 153
(62%) for Class α-β and 470 : 497 (95%) for Class γ. This would suggest that it is
not entirely realistic to consider Ag as an optional, additional clause element which
is different from Ad. Rather, the figures suggest that Ag and Ad are to some extent
interdependent, in so far as the number of A items is remarkably constant, and that
there may be restrictions on the number of As generally in any one position.

TABLE 4:12
Adjunct items, sets and positions



Λ-less 0 284 (55%)

A1 35 8 140 183 183 (35%)
A% 25 5 58 88 44 (8%)
A3 5 1 21 27 9 (2%)
A* 0 1 3 4 1 (0%)

t Total 65 15 222 302 521 (100%)

4§ -less
A1 36 30 146
A' 15 7 74 96 48 (7%)

υ Total 53 40 238 331 695 (100%)
s Λ-less 0 81 (34%)
A1 23 14 77 114 114 (47%)
A 13 9 50 72 36 (15%)
J1 A» 0 2 19 21 7 (3%)
A1 2 0 6 8 2 (1%)
Total 38 25 152 215 240 (100%)

Λ-less 0 0 (.)
A1 0 0 74 74 74 (48%)
A% 8 23 97 128 64 (42%)
A· 5 7 30 42 14 (9%)
A» 1 0 3 4 1 d%)
8 153 (100%)
Total 14 30 204 248

Λ-less 0 142 (29%)

A1 34 28 187 249 249 (50%)
A" 33 25 138 196 98 (20%)
7 (1%)
1 (0%)
Total 72 56 342 470 497 (100%)

4.53 Adjunct form-class

Adjunct form-classes were registered for the most frequent A set types, A1 and A2
(see Figure 4:8). The prepositional phrase was by far the most common form-class,
accounting for just under half the A items in Texts Ml and M2, and for 70% in Text
J1 and 75 % in the passive clause classes. Adverbs, on the other hand, were more
frequent in the active than the passive (27 % : 19 %), and, in the active material, more
common in the novels than the scientific texts. Nonfinite verb clauses as A were




30% -

20% -

10% -

- A A 1 A 2 A 3 A4 - A A 1 A 2 A 3 A4 - A A1 A 2 A 3 A 4
Voice Corpus Class α-β Class γ

Fig. 4 : 6 . A set distribution in per cent.

(Log. graph.)

Active Class α-β Class γ

Fig. 4:7. Positions of A elements in per cent.

(Log. graph.)
F = front-position Μ = mid-position Ε = end-position

TABLE 4:13
Major clause types in the Voice Corpus



Ml M2 J1 Ml M2 J1

Intransitive 159 115 103 377

(EXT-0) (28%) (16%) (29%) (23%)
Transitive 310 522 117 949 Transitive 29 34 114 177
(EXT-C) (55%) (71%) (33%) (58%) (ext-0) (5%) (5%) (32%) (11%)
Ditransitive 15 14 0 29 Ditransitive 3 1 0 4
(.EXT-CC) (3%) (2%) (·) (2%) (ext-C) (1%) (1%) (·) (0%)
Equative 17 22 18 57
(INT-C) (3%) (3%) (5%) (4%)
Factitive 20 22 2 44 Factitive 8 4 1 13
(INT-CC) (4%) (3%) (1%) ( 3 % ) (int-C) (2%) (1%) (0%) (1%)

521 695 240 1456 40 39 115 194

(93%) (95%) (68%) (88%) (7%) (5%) (32%) (12%)

three times as common in active as in passive. Figure 4:8 shows that the form-class
distribution for A items was very similar in both Class α-β and Class γ. It suggests,
for example, that the number of prepositional phrases in a passive clause is usually
limited and fairly constant, regardless of whether there is an agent-phrase or not.


The categories presented in Section 3.8 (pp. 36 f.) were further refined. The clause type
active transitive (EXT-C), the largest one, was subdivided into 'fully transitive' and
'semitransitive'. Fully transitive clauses have nominal groups as complements and,
normally, systemic voice transformation potential:

He likes her. -*• She is liked.

Semitransitives have no such potential or have some kind of restriction on voice
transformation, such as exponential constraint, when they have nominalizations or
reflexive/reciprocal pronouns as exponents of C:
He likes to read. *To read is liked.
... , Γ *That he was leaving was said. 1
They said he was leaving. -» T
[It was said that he was leaving. J
He likes himself. -» ""Himself is liked.

Table 4:13 and Figure 4:9 show the frequencies of the major clause types in the texts
of the Voice Corpus. The totals were, in descending order:
FVC = finite verb clause

Fig. 4:8. Form-class of A elements in per cent.

(Log. graph.)

(EXT-C) Active transitive 949 (58%)

cEXT-0) Active intransitive 377 (23%)
(ext-0) Passive transitive 177 (11%)
(INT-C) Active equative 57 ( 4 % )
(INT-CC) Active factitive 44 ( 3%)
(EXT-CC) Active ditransitive 29 ( 2%)
Cint-C) Passive factitive 13 ( 1%)
(ext-C) Passive ditransitive 4 (0%)
Total 1650 (100%)

Active transitive, ditransitive, and factitive were more common in Texts Ml and M2
than in J l :
Text Ml Text M2 Text J1
(EXT-C) 55% 71% 33%
(.EXT-CC) 3% 2% 0%
(1NT-CC) 4% 3% 1%

Fig. 4:9. Major clause types in the Voice Corpus.

(Log. graph.)

The difference for active transitive was due chiefly to the disproportionately larger
number of semitransitive to fully transitive clauses in the novels as compared with the
scientific text. Speech as C was particularly frequent in Ml and M2. The number of
fully transitive clauses was very similar in the three texts: 31%, 25%, and 29%
respectively for Ml, M2, Jl. The two novel texts were much alike with regard to
clause type distribution, but differed in the following two categories (Text Jl also
given for comparison):
Text Ml Text M2 Text J1
{EXT-0) 28% 16% 29%
(EXT-C) 55% 71% 33%

Passives, as mentioned in connection with the verbal group (Section 4.2, p. 46), were
much more frequent in the scientific text than in the novels. The large number of
passive clauses in Text J1 was, however, unevenly distributed in the subcategories. In
fact, the difference in passive clause frequency among the Voice Corpus texts appeared
only in the passive transitive (ext-0), which totalled 5% in Text M l , 4% in M2, but
32% in Jl.
For the Minor Passive Corpus it was technically possible to register not only
passive clause type according to whether the clauses were C-full or C-less, but, in the
case of Class γ, also the number of A items. (The exponents of C and A could be
obtained by correlation of Column 5 with Columns 3 and 4.)
Classes α and γ showed similar proportions of C-less and C-full clauses, approx-
imately 8:1. Class β, on the other hand, was almost entirely C-less.
Figure 4:10 shows the A set type cooccurring with C-less and C-full clauses in the
large Class γ. A1 was by far the most common for C-less clauses, whereas C-full
clauses usually had no adjuncts. This suggests that passive clauses normally have
one clause element other than Sand V. This element may be either Cot A, less often
both. The most complex A set type was A3, which was rare in C-less (2%) and
nonoccurring in C-full clauses. For both elements taken together the occurrences
were as follows:

No CIA element (C-less, or Λ-less) 75 (15%)

One C\A element (C 1 , or A1) 294 (59%)
Two C/A elements (C 1 + A\ or A2) 119 (24 %)
Three C/A elements (C 1 + A2, or A3) 9 ( 2 % )


For a description of the relevant categories see Section 3.5 (pp. 31 ff.). No distinction
is made below between affirmative and interrogative clauses. Imperatives are included
in free clauses.
Table 4:14 and Figure 4:11 represent the distribution of the clauses among the
three main types of external clause relation: syntactically bound, sequentially related,
and free clauses. Sequentially related clauses, which showed little variation in the
different texts and classes, accounted for roughly Considering only actives in the
Voice Corpus, the percentages were: Text Ml 25%, Class α-β 24%, Text Jl 20%,
Class γ 21 %, Text M2 17%. The ratios of syntactically bound clauses were approx-
imately YQ for Text Ml, \ for Texts M2, and Jl, and Class α-β, but \ for Class γ.

240 -
220 -

200 -

180 -

160 -

140 -

120 -

100 -


I 80 -

60 -

40 -

20 -

Fig. 4:10. C and A elements in Class γ.

(Log. graph.)
Free clauses accounted for half or more in Texts Ml, M2, Jl, and Class α-β, but for
less than ^ in Class γ.
These distributional differences in the material are interesting because they seem
to cut across voice as well as text divisions, both of which have been found to be
significant elsewhere. The numbers here would suggest that, in respect of external
clause relation, agentful passives are more like actives than agentless passives. Half
the agentful clauses were free, i.e. had little or no external relationship, whereas, in
agentless passive clauses, less than ^ of the clauses were free and as many as half
were syntactically bound. However, the difference between syntactically bound and
sequentially related in Text Ml, on the one hand and in Texts M2 and Jl and Class
α-β on the other, would seem to be textually conditioned.

TABLE 4:14
External clause relation



Ml M2 Jl α-β Υ
(la) Relative clauses 19 55 44 23 98
(lb) Conjunctional that- or elliptic /fart-clauses 14 43 10 11 59
(lc) Other conjunctional clauses 25 87 11 5 89


(2a) Fixed relator 19 55 17 10 45
(2b) Fixed relator + elliptic subject 40 14 26 4 11
(2c) ± Fixed relator + elliptic subject + elliptic
auxiliary 8 15 0 5 7
(2d) Elliptic subject 24 2 0 2 2
(2e) Elliptic lexical verb 18 16 0 0 5
(2f) Mobile relator 17 12 6 15 35
(2g) Tag questions 2 2 0 0 1


(3a) Having clause as C 16 42 5 0 2
(3b) Having direct speech as C 74 137 0 0 2
(3c) Being direct speech 3 12 0 0 1
(3d) Followed by sequentially related clause 49 38 46 17 44
(3e) Imperatives 31 7 0 1 0
(3f) Fully free 162 158 75 60 96

(3) FREE CLAUSES 335 394 126 78 145

TOTAL 521 695 240 153 497

If we analyse the main types in greater detail, it appears that, in syntactically bound
clauses, Class γ had the highest proportions in all the three subtypes relative, conjunc-
tional that, and other conjunctional clauses.

In sequentially related clauses, Texts Ml and M2 showed affinities, for example 52

and 35 cases, respectively, for the categories ± fixed relator + elliptic subject + ellip-
tic auxiliary, elliptic subject, elliptic lexical verb, and tag question, as compared
with no occurrences at all in Text J1. On the other hand, more than half the clauses in
the latter text had fixed relator + elliptic subject. Although similar in other respects,
Text J1 and the passive classes differed here: the latter had only 11% such clauses.
In the passive, the most frequent sequentially related clause type was the one with
fixed relators for Class γ, and the one with mobile relators for Class α-β. The fre-
quencies of the two subtypes, representing close and open connection respectively on
the scale of external clause relation within the sequentially related type, reflect the
proportions of the main types of bound and free clauses in the passive classes.

Fig. 4:11. External clause relation.

(Occurence in per cent)
(See Section 3.4, p. 31, for examples of the categories)

This column emphasized stylistic differences. Only texts Ml and M2 had any note-
worthy deviation from the unmarked direct SV order. Considering only these two
cardinal clause elements, the ratios for SV and VS are given in Table 4:15.


In our selective inventory analysis of some voice categories of the total output, we
have arrived at numerous results of which a few seem significant, some suggestive,
and many trivial.

TABLE 4:15

Internal clause relation (Both actives and passives

are included in the Voice Corpus figures)


Ml M2 J1 α-β Y
SK 490 577 350 150 489
(87%) (79%) (99%) (98%) (98%)
VS 71 157 5 3 8
(13%) (21%) (1%) (2%) (2%)
Total 561 734 355 153 497

That a large part of the results should be 'trivial' was only to be expected with the
selected programs, and will not be further commented on here. It may be worth
emphasizing, however, that the difference between trivial and non-trivial results is far
from obvious. A statement such as 'there are more inanimate subjects in scientific
expository writing than in fictional as represented by novels' might well be called
trivial. Precise numerical results become less trivial, however, when considered in a
larger linguistic context, for example, in studying the style used in one field of scientific
research as compared with that of other fields. There are good reasons for believing
that the term 'scientific style' embraces diverse scientific styles. When it comes to
comparing large numbers of less strikingly dissimilar texts, we can no longer be sub-
jectively sure of the differences. Furthermore, although we appreciate that our scien-
tific text has more inanimate subjects and more passive clauses than the novels, we are
not as likely to know what is the relationship of gender and voice distribution. The
information that only 12 out of 355 subjects were animate in Text J1 becomes more
significant in the light of the additional statement that all those 12 subjects occurred in
active clauses, the ratio of animate to inanimate subjects in passive clauses of Text J1
being 0:115.
Some results which seem interesting but which were inadequately dealt with in the
present analysis cannot be used for further statements at this point. They include

voice transformation potential, passive subclasses, and the weight of clause elements,
and will be considered in greater detail in the two subsequent chapters.

4.91 Structure and frequency

As for the seemingly significant results, we shall now try to make a few points,
summarising some of the scattered remarks which were made earlier in this chapter.
In the following categories there appeared to exist a negative correlation between
frequency of occurrence and structural complexity.
(a) The finite verbal group (Section 4.2). Type d/D was more frequent than Types
bJB and c/C and should, with respect to frequency, be placed between a/A and b/B.
As was stated in Section 2.2 (pp. 10 f.), the reason for the adopted system is structural
order of the verbal group elements.
(b) Adjuncts (Section 4.51). Figure 4:6 shows a positively skewed distribution of
the sets A1*, which is to say that, whenever an A set was present in a clause of any
text or class, A1 was more frequent than A2, which was more frequent than A3, which
was more frequent than A*.
(c) Complements (Section 4.41). In all active clauses (there are no C 2 in passive
clauses), C 1 was much more frequent than C 2 .
(d) Subject length, i.e. if their length, measured in terms of number of words, is
considered comparable to structural complexity (see Section 4.36). The number of
words in expressed subjects varied from 1 to 37, the most common being 1-word
subjects, both in agentful and agentless passive clauses. In all classes the distributions
were, by and large, positively skewed (see Figures 4:4 and 4:5). This was also generally
true of agent length, but less clearly so, the mode being 2 words and the diagram
contour having peaks at 2, 4, 6, and 9 words.

4.92 Style

The choice of material turned out to be a rather interesting one. We have found cate-
gories where variations in distribution were obviously stylistically conditioned, and it
is sometimes striking how similar the two novel texts were to each other, on the one
hand, and, on the other, how different they were from the scientific text. This was
found to be the case for example with the following categories:
(a) Voice distribution, with more than four times as many passive clauses in Text J1
as in either Ml or M2 (Section 4.2).
(b) Subject gender. The proportions were 5:1 in Text Μ1, 9:1 in Text M2, but 1:19
in Text J1 for animate:inanimate subjects of active clauses (Section 4.31).
(c) Complement gender in terms of animate :inanimate was roughly 1:7 in active
clauses for Texts Ml and M2 as compared with 0:134 for Text J1 (Section 4.43).
(d) Subject form-class. Texts Ml and M2 had f pronominal and ^ nominal sub-
jects, whereas in Text J1 the proportions were reversed in active clauses (Section 4.33).
(e) Complement form-class. The preponderance of subject nouns in Text J1 ob-
tained also in the complement, whereas Texts Ml and M2 had proportionately more
pronouns and speech (Section 4.42).
(f) Clause types. Particularly rare in Text J1 were active transitives (EXT-C), and
there were no instances of ditransitives (EXT-CC) (Section 4.6).
(g) Internal clause relation. 99% of the clauses in Text J1 had direct SV order,
whereas there was more variation in the Μ texts (Section 4.8).
(h) Adjunct set distribution. A1 was more frequent than Λ-less in Text Jl, but the
proportions were reversed for the Μ texts (Section 4.51). There were however no
obvious stylistic differences in the case of, for example, external clause relation
(Section 4.7) and position of adjuncts (Section 4.52).

4.93 Voice

Since we shall henceforth concentrate on the passive voice, it might be interesting to

see in what respect voice terms may be taken to bear a relationship to the distributions
of other categories. The following are some observations which can be made on the
basis of the analysed material:
(a) No parallel active-passive distribution obtained in the verbal group, since the
passive, coterminous with Type d/D in the verbal group, has one more structural
component than its corresponding active form (Section 4.2).
(b) There was a negative correlation between Type c/C and the passive (Section
(c) Positive correlation existed between passive and Types b and a/A (Section 4.2).
(d) There was a tendency for inanimateness in the subject of passive as compared
with active clauses (Section 4.31).
(e) As passive clause complements there was a high proportion of finite and non-
finite clauses (80 % altogether) but fewer nouns and pronouns; more that- than zero
that-clauses (Section 4.42).
(f) The frequency of adjuncts was approximately the same for active as for agentless
passive clauses (Section 4.51).
(g) Agentful clauses had fewer front-posited adjuncts than agentless and active
clauses (Section 4.52).
(h) A difference was noted in the external clause relation of, on the one hand,
Classes α-β clauses which, like active, were usually free, and, on the other hand, Class
γ clauses, which were mostly syntactically bound (Section 4.7).
(i) Agents were longer than subjects and more than half the animate agents were
coordinate (Section 4.35). Since the nominal parts of agents are potential active clause
subjects, this may well be an important factor in the choice of voice.


The previous chapters have dealt with voice categories generally. From this point
onwards we shall narrow the scope of our inquiry and concentrate on the passive
voice. This does not mean that relationships between the two terms in the voice
system will be disregarded, but only that the passive will be our point of departure.
We shall first attempt to set up a classification of passive finite verb clauses and then
consider some of its uses in diverse texts.
The construction which, up to now, we have broadly called the 'passive' embraces
in fact a number of different subtypes. In the final analysis, it would clearly be un-
satisfactory to treat together, indiscriminately, all clauses with a Type d element in the
verbal group, thus ignoring different kinds of relationship within the clause. We have
in fact already in Chapter 4 classified some passive clauses with a division into
Classes α, β and γ. The foundation for this rudimentary classification of what may
be called 'central passives' according to present/absent and animate/inanimate agent
was however entirely subjective and its relevance is so far unproven. The following
two chapters will be devoted to an analysis in some depth of a small representative
sample in an attempt to establish a comprehensive basis for passive clause classifi-
cation. This chapter will discuss the criteria used in preparing the input data for an
experiment in numerical taxonomy. The next chapter will describe the coding, the
output analysis, and the setting up of a diagnostic key.


It is possible to state a number of desiderata for the classification of finite passive

clauses. We shall mention only four here.1 Such a classification should be, for
(a) Objective and empirically verifiable.
(b) Consistent with intuition. This is probably a basic requirement.
(c) Predictive, i.e. partial or complete knowledge of the relationship of a given
For a more detailed discussion of linguistic classification and numerical taxonomy, see Carvell &
Svartvik 1966. The principles outlined in that study are basic to the present experiment and will
not be repeated here. The techniques are developed for example in Sokal &. Sneath 1963, which
also gives further references.
clause to the classification should enable us to predict with some certainty many other
facts about the clause.
(d) Comprehensive, i.e. accounting for all the data, including the unclear, am-
biguous and marginal cases.
If a classification of passive clauses is to have all or most of these properties, it
seems essential to use multiple criteria. This requires some technique for dealing with
a large number of features, or characteristics attributed to each clause, and comparing
all the clauses in terms of these features. A rigorous technique falls under the heading
of numerical taxonomy. It will not be possible to discuss this technique here apart
from saying that it is based on the use of a large number of unweighted criteria.
Although we shall treat linguistically different types of criteria slightly differently, we
shall not consider some as a priori more important than others. The idea behind this
approach is that the important features are exactly those which will correlate highly
with other features and hence the groups formed will in fact be those that are con-
sistent with our intuition.


The classification program (CLASP, see Section 6.2, pp. 113ff.) classified 128 finite
passive clauses (called OTUs = operational taxonomic units) in terms of 108 criteria.
The size of the corpus used for this experiment, Corpus III (or the Taxonomic Passive
Corpus, see Section 1.5, pp. 8 f.), was dictated by the restricted storage of the available
computer for which the classification program had been written. It is also worth
mentioning that the large number of categories in terms of which each clause was
analysed posed also, of course, a practical, entirely human restriction. The 128 clauses,
which constituted the sample, were proportionately, and randomly, selected from six
groups into which the material had been provisionally ordered on subjective grounds.
The reason for this procedure was that where the use of numerical taxonomy was
expected to be most rewarding - and most interesting - was in the marginal, indeter-
minate, and classificatorily difficult cases. Hence steps had to be taken to ensure that
these clauses which are very often members of numerically small classes, should be
adequately represented in the sample. The OTUs came from the following sources:

22 OTUs from B-texts (B1 = 11, B2 = 5, B3 = 6);

71 OTUs from J-texts (J1 = 22, J2 = 18, J3 = 7, J4 = 24);
35 OTUs from M-texts (Ml = 10, M2 = 6, M3 = 15, M4 = 4).
The clauses were randomly ordered and numbered from 1 to 128.2 They will here be
labelled O T U 1', O T U 2', etc. in addition to their full normal text identification
('J 1.94', 'J2.169', etc.).
* The reason for using random ordering was to avoid routine reactions by the informant. Pilot
experiments had suggested that, when confronted with a sequence of similar linguistic items, the
informant tends to give identical responses to the criteria. Also, informants develop a certain skill
as the testing is going on, and this effect would be amplified if similar OTUs were together.


The criteria can conveniently be divided into five categories according to the aspect of
the clause to which they relate.

T H E CLAUSE: 2 0 criteria ( C L 1-20), described in Section 5.4 (pp. 78 if.) and Figure

(Clause element JV): 25 criteria ( W 21-45), described in

Section 5.5 (pp. 91 ff.) and Figure 5:2.

T H E SUBJECT (Clause element S): 24 criteria (S 46-59), described in Section 5.6 (pp.
97 ff.) and Figure 5:3.

T H E COMPLEMENT (Clause element C): 3 criteria ( C 60-62), described in Section 5.7

(p. 101) and Figure 5:4.

T H E A D J U N C T (Clause element A): 46 criteria ( A 63-108), described in Section 5.8

(pp. 101 if.) and Figure 5:5.

The criteria are of two main types: overt (or constituent) and covert (or transforma-
tional) criteria. The first type refers to features which are actually present in the
particular clause under analysis, for example expressed agent (Ag 75). The second type
refers to potential features, which the informant attributes to the clause, for example
potential agent extension (Ag 92). A male academic staff member gave his reactions
as a native informant to the clauses tested by covert criteria. It would of course be
preferable to have judgments from a number of informants, but this is difficult to
handle in an experiment requiring so many responses (see Section 6.1, pp. 11 Iff.; for
some problems connected with informant reaction tests, see for example Quirk &
Svartvik 1966). In the diagrams, overt criteria are indicated by solid lines, and covert
criteria by dotted lines.
Letters in round brackets (CL, W, S, C, and A, subsuming Ag and Ad) refer to the
category to which a given criterion relates. Numbers (1-108) following the letters are
identification numbers of the criteria. Numbers preceded by a colon denote alter-
native responses to a given criterion. Inventories, 'concocted' items, and initials
standing for names in the spoken material, are given in square brackets. Example-
centres, i.e. those clauses or parts of clauses which are being illustrated, are given in
italics. For the sake of clarity, clauses operating within illustrated passive clauses are
usually not italicized. Optional elements are in round brackets. No punctuation is
given with transforms.
The presentation of criteria for passive clauses and their illustration will not, how-
ever, be restricted to those used in the taxonomic experiment. In the appropriate

(CL2:2) Conjunctional (re) Ihat

(CL 1:1) Syntactically bound (CL2:3) Temporal

(CL ?;4) Other conjunctional
(CL 1:2) ~ Syntactically bound
(CL4:i) Fixed relator
(CL 4:2) Mobile relator
(CL 3:1) Sequentially related
(CL4:3) ± Relator + elliptic S + elliptic Mux
(CL3:2) ~ Sequentially related (CL4:4) Relator + elliptic S
(CL 4:3) hollowed by sequentially related

(CL 6:1) Parenthetic clauses, iJ-clauses

(CL 5:2) » Free
(CL 6:2) Other clause»

(CL 7:1) S coreferent with S of contiguous

active clause
(CL 7:2) S ~ coreferent with S of contiguous
active clause

(CL8:1) S coreferent with S of contiguous

passive dame
(CL 8:2) S - coreferent with S of contiguous
passive clause

(CL 9:1) S coreferent with Ο or Ρ of contiguous

THE CLAUSE (CL 9:2) S " coreferent with 0 or / of «on·
( C V M I A C L 1-20) tigoout cUuse

(CL 10:1) iDvereop ± Hurt

(CL 10:2) ~ Invenioo ± then J (CL 12:0) "Extensive
l(CL 13:1) Intensive
(CL !1:I) Potential |
active transformation ' J(CL13:0) *» Intensive
(CL 13:2)

!"(CL 15:1) Subject suppletion

TRANSfOR NATION J* {_(CL 15.·0) ~ Subject suppletioe"*"*""
t r ., . . _ - - ——· |(CL 16:1) Sequential constraint
I (CL 14:1) Constraint — —
J(CL 16:0) ~ Sequential constraint
I — _|(CL_17:1) Aspectual constraint
(CL 11:2) ~ Potential I J(CL 17:0) ~ Aspectual constraint
active transformation I
I (CM7:2) " ""

I(CL 1?:1) Potential m f(CL18:l) Miscellaneous constraint!

raw™** j «*«* permutation
CHANGE OF AND ^(CL 18:0) ~ Miscellaneous cmtraiote
_ CCLUDQ ~ ~ " ~
(CL 19:0) ~ Potential
(CL 14:0) ~ "constraint
active permntttioD

(CL 14:2)
j(CL 20:1) Potential
mtKnAL (CL2Q-.0) -"potential
active transmutation

Fig. 5:1. Criteria relating to the clause (CL 1-20).


1 ' 1 a! =

Κ Ύ . !|! J i ! ;bi
«is· « ι ί ί ι s ; ι;
°ι — I•=. r Ή m s J
R.RjS.SS?!;. ,
* fc Ϊ fe ί . fel* 7 +
° Liu l-^j w L i — j
ί ί I




2 1 <Λ
i ?i a
i a
ο" ο
a ά ä eq 2* υ
t £ £ u
1 ί 8.
I ι (Η (N
11 ίϊ
η a s ä <o
ί £ ϊ Μ

(S 46:0) Noun (S 47:1) Personal referential

(Ag 81:0) (S 47:2) Nonreferential
(S 46:1) Pronoun (S 47:3) Relative
FORM-CLASS (Ag 81:1) (S 47:4) Demonstrative
(S 46:2) Name (S 47:0)
(Ag 81:2)
(S 46:-) Zero

(S 4«: 1) Animate
(Ag 82:1)
(S48:0) Inanimate
(S 49:1) Concrete
(Ag 83:1)
(S 49:0) Abstract
(Ag 83:0)
(S 50:1) Countable
(Ag 84:1)
(S 50:0) Mass
(Ag 84:0) (S 52:0) Zero article
(Ag 86:0)
(S 52:1) Definite article
(Ag 86:1)
(S 52:2) Poss. pron./Genitive
(S 51:1) Definite (Ag 86:2)
FIN1TUDE (Ag 85:1) (S 52:3) Demonstrative pron.
(S 51:0) Indefinite (Ag 86:3)
(Ag 85:0) (S 52:4) Wft-pronoun
(Ag 86:4)
(S 52:5) Indefinite article
(S 53:1) First
(Ag 86:5)
(CRITERIA S 46-59)
(S 53:3) Third

(S 54:1) Singular
(S 56:1) Premodified
NUMBER (Ag 87:1) PREMODI- (Ag 89:1)
(S 54:2) Plural HCATION
(Ag 87:2) (S 56:0) Unpremodified
(S 55:1) Modified (Ag 89:0)
(Ag 88:1) (S 57:1) Postmodified
(S 5 5 : 0 ) Unmodified POSTMODI- (Ag 90:1)
ncA N
(Ag 88:0) ™ (S 57:0) Unpostmodified
(Ag 90:0)

(S 58:1) Coordinated
(S 58:0) Uncoordinated
(Ag 91:0)

(S 59) Log (Number of words)

Fig. 5:3. Criteria relating to the subject (S 46-59), and some subcategories of the agent
(see Fig. 5:5)

(C 61:1) to V
(C 61:2) Noun

(C 60:1/1) C-full (C 61:3) Adjective

(C 61:4) Ving

(C 61:0)

THE COMPLEMENT (C 62) Log (Number of words)

(C 60-62)

(C 60:0/0) C-less
Fig. 5:4. Criteria relating to the complement
(C 60-2).

places it will be convenient to extend the discussion to other criteria and examples,
partly in order to provide fuller illustration or to avoid duplication of the few OTU
examples and partly in order to discuss features which for some reason such as non-
occurrence or infrequency in the sample were not used in the experiment. Categories
which have been introduced earlier, in Chapters 2-3, need only be exemplified here,
whereas new categories will be discussed in some detail.


(Criteria CL 1-20, see Fig. 5:1)

5.41 External clause relation

The categories have been presented in Section 3.5 (pp. 31 ff.).

(CL 1) Syntactically bound clauses [52 cases] comprised the following four sub-
(CL 2:1) Relative clauses [23 cases].
... in order to obtain information concerning the relative rates at which various
reaction sites in various aromatic nuclei are attacked by free radicals.
(OTU 25, J4.27)

1 != 's I u !sI ο!ε
i 3 a !6 ä Iglü I c
.1s Ο'i ι·I®e
< !2 |3 I I |3
!! δ i'
's- i i i is .IJLIii:
11 • ο ι : i®
I» Λ s Js> 'S Ig; I» jg»
12 r ji S L-j-
< I I | ! < ί .<· Ι<Ί< I "
ι g ^ fj
I gf s "5.

~ Clause
L g

-· Front
o. »

- Mid
έ·° = £ Ü C
•c t U, ϊ 1βU
jju Τ- ο
Is ο τ τ ο"
ll j lU *c <•3 c
3S-.I ο ο ο ο 8 ε ο
Τ5 •ο
< J •α
<|,J1 Ό < < < <
•β <<
-Ό uÖ ϊI1*auSüS gC«
t _ jS£ls

(rt W, VIs VI
i er &L& &
Γ "Ι J ι
τ τ ^ is
4 - =! ι ι χ τ
» I
UJ C Ο 8.
f ?


ce ? Τ1?-®1 Τ T s T ι
υ V
•s i<Si<*!<lis = t ; - i=j
• V " Τ- isS|<. ' L sJji ' l i ' j s >·φζ > i s s j s .
ι I Τ ! Τ
.a «> — ι I ι j. I

c tf; .2 i ΖO oC« >
S S X · j j:
.2 < .<£ S-* «Sä
ι iiiill j s s· *


3 Η

Ζ >o

(CL 2:2) Conjunctional that- and so that-clauses [12 cases].

... it is a necessary corollary of this scheme that these radicals are formed only in
negligible amount. (OTU 52, J4.38)
Nevertheless, the recent work of Huisgen and Nakaten (1954) and Davies, Hey
and Williams (1956) has shown that nitrodiphenyls can be obtained from this
reaction, although in rather low yield. (OTU 71, J4.32)
(CL 2:3) Temporal clauses [8 cases].
... it follows that the unit cell found by X-rays is indeed the true unit cell when the
hydrogen atoms also are considered. (OTU 56, J2.163)
(CL 2:4) Other conjunctional clauses [9 cases].
... benzoic and anthranilic acids do not take part in the Gomberg reaction unless
they are first esterified, i.e. convertedfrom molecules which are ionised in aqueous
alkaline media, to typical organic esters. (OTU 93, J4.29)
(CL 3) Sequentially related clauses [55 cases] comprised the following five sub-
(CL 4:1) Clauses with fixed relator, such as and or but [17 cases].
The side walls are strong but the roof is composed only of a tough membranous
fibro-cartilage. (OTU 44, J1.87)
(CL 4:2) Clauses with mobile relator, such as however or then [11 cases].
However, values of (-0·40±0·02) χ 10 12 cm. and (0·57±0·08) χ 10 12 cm. were
deducedfor b for hydrogen and deuterium respectively. (OTU 45, J2.165)
(CL 4:3) Clauses with relator -f- elliptic subject + elliptic auxiliary [5 cases].
... he scuttled sideways ... into the old linen cupboard, where he was cornered,
wrapped in a blanket, and carried effortlessly by Trained Pig Percy-Scroop-
Beauchamp to the bathroom. (OTU 92, M3.209)
thus the forebrain is connected with smell, midbrain with sight, hindbrain with
acoustico-lateral and taste-bud systems. (OTU 112, J1.101)
(CL 4:4) Clauses with relator + elliptic subject [7 cases].
'No, I don't think we want him,' said Martin, but was disconcerted when his
mother said: O h , I am so glad.' (OTU 114, M3.204)
(CL 4:5) Clauses followed by sequentially related clauses [16 cases].
Martins rocking horse was discovered in an attic and installed in the day nursery,
(OTU 98, M3.210)
(CL 5) Free clauses [28 cases] comprised the following two subclasses.
(CL 6:1) Fully free clauses, i.e. with no external relationship of the types recognized
here in CL 1-4 [24 cases]. (Other external clause relationship, such as that denoted by
determiners, will be analysed in connection with fmitude in the subject.)
In this equation V is the volume of the crystal which is being radiated. The value
of ε usedfor calculating the integrated reflection pß is first correctedfor absorption.
[New paragraph] The structure factors determined in this way were ...
(OTU 43, J2.171)
(CL 6:2) Clauses with proleptic it, or parenthetic clauses [4 cases].
It would have been absurdly affected to expect a child of three to need lavatory
paper or towels, and in any case it would have embarrassed Lady Foxglove to
have to ask for them. (OTU 23, M3.200)
But it isn't that at all; he doesn't lack quality, I feel convinced.
(OTU 72, M2.105)

One of the suggestive results of the voice data processing (reported in the previous
chapter) was the high proportion of passive clauses with close external relation,
particularly relative clauses. This result was corroborated in the present analysis by
the inventory totals: 52 syntactically bound, 55 sequentially related, and 28 free clau-
ses. These numbers indicate that there were some overlapping cases, as in the fol-
lowing example with a bound conjunctional that-clause followed by a sequentially
related clause:
... the troops ... were so affected by the sights and the smells and the testimony to
human depravity that many of them were taken ill and had to be sent home on
leave. (OTU 67, M3.209)

5.42 External subject relation

The relationship of a passive clause to a contiguous clause may be denoted by other

categories than those mentioned in CL 1-6. There may, for instance, be identity of
subject, as in the following example with an active equative clause followed by a
sequentially related passive clause:
'He's only six; and not allowed to play with rough children' (OTU 60, Μ 1.26)
Following up the suggestion of the possibility of sequential clause relation as a factor
which might influence the choice of the passive voice, the subject-relation of the
passive clauses to clause elements of contiguous clauses was analysed in terms of
coreference. 'Contiguous' was taken to mean 'immediately preceding or immediately
following'. 'Coreference' implied either complete exponential identity, as in the exam-

pie above (OTU 60, which was of course also entered as a case of CL 4:3), or simple
coreference relation, as for example between pronouns and their antecedents. The
following three types of external subject relation were recognized.
(CL 7) The subject of the passive clause was coreferent with the subject of a con-
tiguous active clause [37 cases].

/they'd been at school # . /they'd been a /they'd been - a at the :depot # /they'd
been: trained # - /they hadn't "really # . / ' k n o w n and'had life # as it /is to be
had # (OTU 108, B2.23)
This [a well-developed heart] lies behind the gills and can be considered as a
portion of the sub-intestinal vessel, folded into an S-shape and divided into three
chambers. (OTU 85, J1.91)

(CL 8) The subject of the passive clause was coreferent with the subject of a
contiguous passive clause [19 cases].
Lead roofs could no longer be patched and would have to be renewed completely.
It [the photolytic decomposition of phenylazotriphenylmethane in benzene] has
been investigated by Horner and Naumann (1954) and Huisgen and Nakaten (1954),
and was found to involve a primary dissociation into phenyl and triphenylmethyl
radicals and nitrogen, in the manner indicated in equation (8).
(OTU 69, J4.34)
(CL 9) The subject of the passive clause was coreferent with the object (including
prepositional object) or predicative of a contiguous clause, active or passive [18
This is now being used instead of grain which is stored on British farms waiting
for buyers. (E2.29)

5.43 Internal clause relation

(For categories, see Section 3.4, p. 31.)

(CL 10) Inversion of S and W elements, with or without introductory there [3
[- in I fact in nineteen sixty # there were only jseven # . jhanged # - and as /long
ago as nineteen thirty # there were only /six -hanged # ]
(OTU 77, B3.4; [m=low, narrow]) 3

For the present purpose, it has been necessary to depart, in some cases, from the typographic
conventions given in Crystal & Quirk 1964. Stretches with modulation features, for example,
have been denoted by bracketing: [[allegro]], etc.

iri'lcluded in this class of crimes# . "jare as müch# . l_jinstances of ma:licious

wounding of :one thug or 'young man byj < an" lother
(OTU 79, B3.15, [low,narrow; piano] < allegro)

5.44 Active transformation potential [107 cases]

(CL 11) A closer analysis of these 107 clauses out of the total 128 proved, however,
that unrestricted passive-active transformation was possible in only 18 cases. 'Un-
restricted' transformation implies transformations without constraints or with con-
straints other than those recognized here and discussed later in this section. Hence,
the term 'free active transformation is simply the reversal of the passive transforma-
tion of transformational-generative grammars: 'If Sx is a grammatical sentence of
the form NP1 — Aux — V — NP2, then the corresponding string of the form NP2
— Aux + be + en — V — by + NPX is also a grammatical sentence' (Chomsky
1957, 43; cf. Section 1.1, p. 1). In the present terminology, the active transformation
formula may be stated as

N1 W ρ N2 -» N2 VNj,

The Ν before the verb is the subject in either voice, W is the passive verbal group, V
the active verbal group, and ρ is the agent-preposition, usually by.
A distinction was made between different clause types in the active transforms
(CL 12) Extensive clause type [104 cases] and
(CL 13) Intensive clause type [15 cases]. Normally the clause types in the active
transforms are those corresponding to the passives as outlined in Section 3.8.
Passive transitive (ext-0) active transitive (EXT-C), for example:

The relevant experimental data are summarised in the following conclusions.

(OTU 87, J4.37)
-* [we] summarise the relevant experimental data in the following conclusions
Passive ditransitive (ext-C) -*• active ditransitive (EXT-CC), for example:
if I"we had been offered the choice of being killed or not that möment # we would
have [/"got down # ] as /low as we could # into the /nearest shill hole #
(OTU 103, B2.19; [ staccato])
-* [someone] had offered us the choice of being killed or not
Passive factitive (int-C) active factitive (INT-CC), for example:

'[this bookshop] does cater for what's called the cultivated reading public - and
for specialists.' (OTU 106, Μ 1.20)
-* what [they] call the cultivated reading public

Some 'adjectival passives' as, for example, in

She was astonished, and said so. (OTU 113, Μ 1.23)
may take two active transforms, one extensive and one intensive:
-> [the dog's behaviour] astonished her / made her astonished
It is a property of this class of 'emotive verbs' that their Ved forms may operate as
C2 in active clauses of type INT-CC in addition to operating as Via active clauses of
type EXT-C. In the following example, transforms with two INT-CC types are
given, the latter of which may be analysed as a reduction of the former equative

but he was ... embarrassed by the implication of sentiment in his remark. (M5.25)
-»· the implication embarrassed him
-* the implication made him feel embarrassed
-» the implication made him embarrassed
Although Ved is most commonly the exponent of C2 in active intensive types, V and
Adj may also be considered in some cases:
I be/lieve in regard to :capital piinishment # [which /is of course a . a : the
Ideath · sentence # - for /murder # ] - < that it /should be applied # to jail
murderers # (OTU 17, B3.54; [low]; <allegro>)
->•... that [the courts] should apply it / make it apply / make it applicable to all
(CL 14) Transformational constraints. The analysis of the 107 clauses with
potential active transformation (see CL 11) proved that only one in six had 'un-
restricted' active transformation potential: as many as 88 clauses had some form of
transformational constraint, which was specified under four headings: (CL 15)
subject suppletion, (CL 16) sequential, (CL 17) aspectual, and (CL 18) miscellaneous

(CL 15) Subject suppletion. By far the most common constraint [75 cases] was
the necessity, in the case of agentless passives, of supplying active transforms with
subjects. Suitable active subjects were inferred from the context. Such 'concocted'
elements are given in square brackets:

The baby was installed in the old nursery wing at Bidcombe, and a very expensive
nanny was hired from London to look after it. (OTU 55, M3.198)
[Lady F.] hired a very expensive nanny from London to look after it

On the dorsal side is a single nasal opening, and behind this there is a gap in
the pigment layers of the skin through which the third or pineal eye can be seen as
a yellow spot. (OTU 86, J1.84)

-*• through which [one] can see the third or pineal eye as a yellow spot
(CL 16) Sequential constraints [14 cases] were due to external clause relation as
stated in CL 3 and CL 4. In the following example with elliptic passive subject the
active transform requires the suppletion of an object. (An additional constraint not
considered in the experiment is the unsatisfactory stylistic imbalance which is caused
by the suppletion of a light end-placed pronoun; see CL 18 below.)

Isokinetic relationships of this kind are not uncommon in organic chemistry,

especially in free radical reactions, and have been discussed recently by Leffler
(1955). (OTU 42, J4.33)
-* isokinetic relationships of this kind are not uncommon ..., and Leffler (1955)
has recently discussed [them]

If the passive clause is followed by a sequentially related clause with zero subject,
some suppletion will be necessary in the latter clause. The active transform of the
example in CL 4:5 might be as follows, with two active clauses,

[they] discovered Martin s rocking horse in an attic and installed [it] in the
day nursery

or, with an active clause followed by a passive clause,

... and [it] was installed in the day nursery

(CL 17) Aspectual constraint. In eight cases active transforms were only 'pos-
sible', or at least 'preferable', with a change of aspect or mode in the verbal group.
In the first example the change is from Type d to b, in the second from D to Β (see
Section 2.2, pp. 10f.):

Madeleine walked forward and stood at a little distance, near enough to see the
shape at Dinah's foot. 'Nearly dead,' said Dinah slowly. Ί think its back's
broken: (OTU 53, Ml.27)
-»· I think [the stroke] has broken (*breaks) its back

... Herring wept quietly and tearlessly all day long after her husband had told
her that he had sent the child to an adoption society. She was alone in a dark
world, and the only thing she had ever loved was removed from her.
(OTU 107, M3.197)
-* [her husband] had removed from her the only thing she had ever loved

Note on aspectual constraint: The illustrated passive type looms large in most handbooks.
It is in fact a moot point of classification whether it should be admitted at all as 'passive'.
Jespersen makes a distinction between two classes of verbs: CONCLUSIVE VERBS whose
'action is either confined to one single moment, e.g. catch, surprise, awake, leave, end, kill, or
implies a final aim, e.g. make, bring about, adorn, construct, beat'; and NON-CONCLUSIVE VERBS
'denoting feelings, states of mind, etc.: the activity, if any such is implied, is not begun in or-

der to be finished. As examples we may mention love, hate, praise, blame, see, hear' (Jespersen,
1909-49, 4.92-3). This dichotomy underlies his divisions of passives into two types: 'With
non-conclusive verbs... the participle is a predicative of being... and we may therefore speak
of a PASSIVE OF BEING' (He is admired). 'With conclusive verbs the time-relation is not so
simple: sometimes the participle is a predicative of being, sometimes one of becoming, and
therefore we sometimes have a PASSIVE OF BEING, sometimes a PASSIVE OF BECOMING'; "... in
such a sentence as His bills are paid sometimes the element of present represented by are,
sometimes the element of perfect implied in the participle is predominant. The sentence
thus may mean two things, either the present action as in His bills are paid regularly every
month = He pays, or the (present) result of a past action as in His bills are paid, so he owes
nothing now = He has paid' (Jespersen, op. cit., 98). Other terms that have been used
include STATAL or STATIC PASSIVE and ACTIONAL or KINETIC PASSIVE (Curme 1931, 443 ff.);
A view differing from that of Jespersen and Curme is held by many grammarians, among
them Kruisinga:4 the term "passive" 'is only applied to the verbal group when it expresses
an occurrence or an action. Thus, we have a group-passive in The book is sold for 5 s., but
when a second-hand bookseller informs a would-be purchaser that a book in the catalogue
is sold, the group expresses a state or condition and is not a passive. It is a curious and
naturally accidental result of this that the same verbal group comes to express two opposite,
and mutually exclusive, meanings'.
With the comprehensive approach adopted in this study, the term 'passive' embraces both
the actional and statal types.

(CL 18) Miscellaneous constraints [15 cases]. Several constraints which were
either too infrequent or unclassifiable as a group come under this heading,
(a) Introductory there and proleptic it are dropped in the active.

there's /also included in this class # . a /lot of !other -things # . such as] (pro\
curing abörtion # ) -and . enjdangering a ship at Isea #
(OTU 115, B3.16; [allegro]; <Iargo; narrow and rhythmic»
[the law] also includes in this class a lot of other things
It is seen that the hydrogen atom may be removed by one of the two processes
(17d) or (17e), that is, either by a radical R'· or by the solvent, and that for each
separate case the overall kinetic result is different. (OTU 7, J4.38)
-* [we] see that the hydrogen atom may be removed
(b) Compounds with the negative uw-prefix (see Section 2.73, pp. 21 f.) may sometimes
take active transformation if the bound negation is replaced by a free negation.
The details of the composition of the nerves of lampreys are still unknown, but
there are hints of considerable deviations from this plan. (OTU 95, J 1.97)
->· [we] still do [not] know the details of the composition
(c) In certain passive factitive clauses a copula is optional as part of C but impossible
or unlikely in the active transform:
Kruisinga 1925-31, 2:1.305. Cf. also Poutsma 1926-9, 2:2.98ff., Zandvoort 1960, 48ff., Francis
1958, 335 ff., Fries 1940, 189, Scheurweghs 1959, 156, 416.
Dusk had perceptibly deepened before Dinah was seen to be marching briskly
up the slope again, the bier across her shoulder. (OTU 109, M1.29)
-> before [they] saw Dinah (*to be) marching

(d) 'Adjectival passives', which were mentioned in CL 13, share many properties with
active equative clauses with adjectives as C. They can for example take modifiers like
very. Notice, however, that the active transitive transforms require 'verbal' modifiers.

I'm /very surprised actually # at /this . quotation of Riissell's # (B2.10)

this quotation of Russell's surprises me very [much]
(e) As was noted in connection with OTU 42 illustrating CL 16 above, stylistic
imbalance was not registered as a constraint in the present experiment. The following
example, however, shows that when the weight of clause elements becomes too dis-
proportionate the acceptability of the transform can be drastically reduced.

This has been neatly demonstrated by Grieve and Hey (1938), who showed that
benzoic and anthranilic acids do not take part in the Gomberg reaction unless
they are first esterified, i.e. converted from molecules which are ionised in aqueous
alkaline media, to typical organic esters. (OTU 101, J4.29)
-*• Grieve and Hey (1938), who showed that benzoic and anthranilic acids do
not take part in the Gomberg reaction unless they are first esterified, i.e. con-
verted from molecules which are ionised in aqueous alkaline media, to typical
organic esters, have neatly demonstrated this.
(f) Other types of constraint can be seen in the following examples (the first trans-
form is, however, not strictly synonymous with its passive original):
... she must have been motivated more by the desire for notoriety than Christian
charity in adopting, at her age, a baby from an unknown home and giving it her
name. (OTU 18, M3.197)
-> the desire for notoriety [rather] than Christian charity must have motivated
The variations in energy of activation are compensated by the variations in
entropy of activation, ... (OTU 61, J4.33)
the variations in entropy of activation ... compensate [for] the variations in
energy of activation
These 'follicles of Langerhans' were, appropriately enough, first seen by the
discoverer of the islets in higher forms, ... (OTU 125, J1.90)
-*• the discoverer of the islets in higher forms, appropriately enough, first saw
these 'follicles of Langerhans'
The following two transforms, however, bring out the sentence focus better, and are
generally more acceptable:

-»• the discoverer of the islets in higher forms was, appropriately enough, the
first to see these 'follicles of Langerhans'
it was, appropriately enough, the discoverer of the islets in higher forms who
first saw these 'follicles of Langerhans'

5.45 Permutation and transmutation

Voice transformation entails changes not only from passive to active verbal group
but also in the function and position of the nominal clause elements, for example
X is followed by Υ -* Yfollows X which may be represented by the following diagram
(cf. Jespersen 1924, 164):

The preposition is dropped when the postprepositional nominal part of the agent
becomes the subject in the active. In stating the transformational properties it is
generally more convenient to mark the nominal elements by subscripted Ns, the
first of which is the subject in either voice (see Sections 2.32, pp. 13f., and 3.8, pp. 36 f.):

In an attempt to describe relations obtaining among clause elements, another test

similar to transformation suggests itself: a change of voice in the verbal group without
inversion of the nominal elements; the result still being a grammatical clause, as in
X is followed by Y -+ Xfollows Y. In order to distinguish this test from transformation
we may call it PERMUTATION. In voice permutation the passive subject remains
subject in the active clause, but the agent changes, not, as in voice transformation,
to active subject but to object, with deletion of the preposition.
Indeed, the hydrogen atom is probably never actually free, but may be removed
by the aroyloxy radical from the α-complex formed between Ar'· and ArH.
(OTU 59, J4.39)
-* the hydrogen atom ... may remove the aroyloxy radical
As deduced from the X-ray results the nitrogen atom, taken as the origin, is
surrounded by eight chlorine atoms in positions such as \ £), giving a CsCl
structure. (OTU 19, J2.174)
-» the nitrogen atom ... surrounds eight chlorine atoms
Those two permutations are grammatical (but perhaps examples of 'ontological
nonsense'; see Bolinger 1961b, 371), whereas the following is non-grammatical:
The first investigations of the chloride by neutron diffraction were made by
Goldschmidt and Hurst (1951) ... (OTU 102, J2.174)
*the first investigations ... made Goldschmidt and Hurst
In agentless passive clauses the change under permutation is restricted to the verbal
[it's a consolation for the survivors] /rather than a igenuine consol-ation # for
/those who were -.killed # (OTU 81, B2.20)
-* for those who killed
He had a half-formed apprehension that Percy might pull his head off or some-
thing if he were bitten too hard, ... (OTU 22, M3.209)
-*• if he bit too hard
Permutation is however different from transformation in one respect other than
absence of inversion: it involves semantic change. This may be expressed by saying
that in transformation the grammatical subject changes but the logical subject
remains the same, whereas in permutation the grammatical subject is the same and
the logical subject changes.5 When some verbs are permuted, however, the result is
a grammatical sentence with no appreciable change of meaning, i.e. the logical
subject is felt to be the same, or nearly the same, in the active as in the passive. We
shall call this test TRANSMUTATION.
Her less fashionable friends piously closed their ears to the malicious rumours
which were being spread abroad, ... (OTU 16, M3.201)
-> which were spreading abroad
yet/these countries # . [/as we're !both a:gr£ed#J - /have#. <9 an in /some
cases an :equally serious or :even> "Imore serious crime situätion# than /we
do in this !!country# (OTU 35, B3.46; [low], <low, confident, sure))
-» as we both agree
There is no trace of any paired fins, but the tail bears a median fin, which is
expanded in front as a dorsal fin. (OTU 122, J 1.84)
-» which expands in front as a dorsal fin
... the curves do not again separate above the critical point, but are continued
as a single curve. (OTU 21, J3.31)
->• but continue as a single curve.

'Grammatical subject' is that element in the clause with which the finite verb is in grammatical
concord (overt and covert) (see Section 3.1, pp. 25 f.). 'Logical subject' may be defined as 'the person
or thing from which the predication is considered to originate' (Poutsma 1926-9, 2:2.7).

'Permutation' and 'transmutation' are introduced here as specific technical terms for
describing certain voice relations. If the analysis is carried farther, such criteria help
to distinguish subtypes with different voice relations, as in the following sentences
(cf. Olsson 1961, 180, and Erades 1950):
(a) He worried. = He was/felt very worried (about it).
(b) The door closed. = The door was closed (by someone).
(c) The house is building. (But *The house builds.) = The house is being built.
(d) The play reads well. (But *The play reads.) φ The play is read well.
The 'transmuted' sentence, as in The war began, has been called 'notional passive',*
and its subject 'neutral' since it is 'involved but not responsible' as in They began the
war. Long takes the extreme view that, since they 'are often used to predicate events
and states of affairs rather than actions', all active forms are best called 'common',
in contrast to 'passive' (1961, 113, 510). If one wants to let the semantic distinction
be reflected in the names of the categories, it seems that a more consistent classification
would be: 'active' for He began the meeting; 'common' for The meeting began;
'passive' for The meeting was begun.

Summarising, the three tests we have just described have the following typical features.

[different grammatical S,
same logical S ] N1 W ρ JV2
N2V N1
He was bitten by the dog.
-> The dog bit him.

[same grammatical 5","]

different logical S J
N1 W ρ Ν2
-* N1V Ν2
He was bitten by the dog.
->• He bit the dog.
same grammatical S, Nx W ρ Ν2 He was worried about the dog.
same logical S -* Nx V ρ Ν2 -* He worried about the dog.
Νι W ρ Ν2 He was opposed to the idea.
-» Νχν Ν2 He opposed the idea.
(CL 11) Potential active transformation [106 cases] seems to be by far the most
important test for denoting grammatical relationships between passive and active
(CL 19) Potential active permutation [19 cases] draws attention to such properties
as gender of clause elements (cf. Chomsky 1957, 78).
(CL 20) Potential active transmutation [19 cases] highlights the uses of verbs and
the relationship of clause types.

* Cf. Curme 1931, 440, Jespersen 1924, 164ff.



(Criteria W 21-45; see Figure 5:2, p. 76)

Two aspects of the verbal group were considered: auxiliaries and lexical verbs.

5.51 The auxiliary, tenses and types

The tense of the first element of the verbal group was analysed in terms of the follow-
ing categories (cf. Section 2.34, pp. 16f.):
(W 21:1) Present tense [75 cases].
It appears that the energy relationships involved are so balanced that many
reactions have a definite and observable probability dependent on the properties
of the radical and the substrate. (OTU 105, J4.30)
(W 21:0) Past tense [51 cases].
She thought it had had a number of owners; luckily the building itself had never
been touched·, but everybody had done something to the garden and made a
mess of it. (OTU 4, M1.17)
(W 21:-) Non-tense-marked [2 cases].
<5i a a [/[world] !wlde#] . {/increase in crime# - /notably :jüvenile crime#}
(must be /brought into this :matter as !well#> (OTU 76, B3.2; [forte]; {low};
<crescendo & allegro»
(W 22) Auxiliary Type a/A [21 cases] was further divided into three subclasses.
(W 23:l/0)7 Fully tense-marked [4 cases].
He was covered in snow and had virtually to be carried to the fireplace and
stripped of his outer clothing. (OTU 12, M2.101)
(W 23:0/0) Semi-tense-marked [15 cases].
This difficulty can be avoided in two ways: (OTU 48, J3.30)
(W 23:0/1) Non-tense-marked [2 cases]. (See W 21 for illustration.)
(W 24) Auxiliary Type b/B [17 cases].
there's a :rather "igreäter tendency for people to commit crimes a:gäin#] -
/äfter they'dbeenfl0gged# than /when they hädn't# (OTU 116, B3.19; [allegro])
(W 25) Auxiliary Type c/C [2 cases]. The infrequency of this type with the passive
' This criterion was technically treated as two tests with the three alternatives marked 'I/O, 0/0,0/1,

(noted for the Minor Passive Corpus, see Section 4.2) was corroborated by this
In this equation V is the volume of the crystal which is being radiated.
(OTU 82, J2.171)

Auxiliary Type d/D. All clauses had, by definition, type d/D auxiliaries; they will be
further analysed here by means of overt and covert criteria.

Note on auxiliary type d/D (the passive auxiliary): There is general agreement that be is a
passive auxiliary,8 but the status is less clear for other auxiliaries which occur with past
'Get and become are now increasingly common as auxiliaries for the passive of becoming;
with some verbs the distinction between them and be is particularly useful' (was/got married) ·,
get 'has (or had) a decidedly more colloquial colouring'; grow 'in the same function as
auxiliary of the passive of becoming is comparatively rare, apart from combinations like
grow accustomed'; stand is 'particularly frequent in judicial expressions like stand condemned';
rest, as in rest assured 'sometimes is approximately a mere auxiliary of the passive' (Jespersen
1909-49, 4.108 ff.).
'When connected with a past participle to get is apt to lose its character of a copula and
assume a function which differs little from that of to be as an auxiliary of the passive voice.
The altered function of to get, of course, postulates a change in the grammatical function
of the participle, which, from being mainly adjectival, becomes almost purely verbal ...
Some such modification of function may also be traced in to become and to grow, but with
these verbs the change is less pronounced, the participle being more retentive of its adjectival
features, and the combination suggesting a gradual process'; 'certain verbs that more or
less partake of the character of copulas, such as to feel, to stand, may be used to advantage
when no passiveness is intended ... Also copulas of the third kind, i.e. such as are used to
express the changing of a state into another ..., are not unfrequently combined with a past
participle to form a construction that bears a strong resemblance to the passive voice.
Naturally the participle is not so entirely devoid of adjectival characteristics in these com-
binations as it is in a pure passive voice. Instances with to get are quite common, especially
in colloquial style; to become and to grow being far less frequent in this function';'... the
combinations with to get ... are hardly distinguishable from a purely passive construction
with to be, [whereas those containing to become, and to grow] only vaguely express passive-
ness: in other words to get, when connected with a past participle, has lost almost entirely its
power of indicating incipient action and may, accordingly, be called an auxiliary of the
passive voice, but to become and to grow show this force in only a slightly weakened form'
(Poutsma 1926-9, 1.30; 2:2.99).
Get 'expresses the getting into a state or situation denoted by the participle; in other words,
it has a mutative meaning, which distinguishes it from the ordinary passive.' 'The difference
between to get and to become as auxiliaries of the passive may be expressed by the terms
PERFECTIVE and DURATIVE' (Zandvoort 1960, 57).
Get 'in spite of constant efforts of grammarians and teachers of English, has not been
abolished, but rather seems to be increasing in favor in the common idiom' (Frary
1929, 73).
Get 'occurs more frequently in vulgar English' (Fries 1940, 193).
But, as was pointed out in 'Note on aspectual constraint' under CL17 above (pp. 85f.), not even all
combinations of be with past participles are generally agreed upon as 'passive'.
The ^/-passive 'seems to be increasing in frequency, though grammarians are at present
not agreed as to its status' (Francis 1958, 335).
Hatcher argues against Jespersen who 'lists become (and even grow, stand, rest) along with
get as a passive auxiliary - without, of course, attempting to define the restrictions on its use.
These restrictions are such ... as absolutely to exclude it from consideration as a passive
auxiliary: for become may never be used to refer to any passive action performed by a human
agent (unless, of course, the participle is such as to invite an adjectival interpretation: this
tradition became accepted but not the present became accepted; ... unlike get, become is
found only in border-line cases' (got accustomed, drunk, lost, engaged). (Hatcher 1949,442f.).
Svartengren coins the term ACTIONAL-DURATTVE for the become-passive, which he considers
as a combination of the STATAL PASSIVE and the ACTIONAL PASSIVE since it 'describes the
action, i.e. the transformation and its lasting result' (During the first half of this century the
tribe had become christianized). The phrase had become christianized 'must undoubtedly be
regarded as a type of the passive, but by using the term ACTIONAL-DURATTVE, we admit that
there may be cases when ... the durative aspect so far outweighs the actional-verbal that
the past participle acquires an adjectival character ... and the expression can no longer be
regarded as a passive' (Svartengren 1948, 275).

The confused picture presented by these extracts is accounted for, largely, by lack of
formal criteria. Such concepts as 'pure passive voice' and 'adjectival interpretation'
are hard to agree on. Adhering to our intention to take a comprehensive view, we
shall initially admit as passive auxiliaries all verbs (other than Type bIB auxiliary have,
see Section 2.4, p. 17) that combine with past participles. Since only be is frequent in
this function, little can be said about the others on the basis of Corpus III; but it is
necessary to state at once how they will be handled.
Although we have made a distinction between auxiliaries in verbal groups (Types
a, b, c, and d) on the one hand and copulas in equative clauses on the other, it is
essential in setting up a classification of auxiliaries to see these two kinds of verbs in
relation to each other. The division proposed here for passive auxiliaries will there-
fore apply also to copulas.
The basic division is into LEXICALLY UNMARKED AUXILIARIES (be) and LEXICALLY
MARKED AUXILIARIES (auxiliaries other than be). The latter have the two subclasses
MUTATIVE and NONMUTATIVE. These terms have previously been used for the passive
(see Strang 1962, 146), but they have been extended here to cover all lexically marked
auxiliaries and copulas that are aspectually related. Be is formally unmarked but
may have either mutative or nonmutative value. The following list will illustrate the
two subclasses:
become (established) appear (delighted)
come (unstuck) feel (annoyed)
get (lost) lie (scattered)
go (sour) look (offended)
grow (tired) remain (divided)
fall (sick) rest (assured)
run (wild) seem (determined)

turn (sour) smell (sweet)

etc. sound (surprised)
stand (transfixed)
taste (better)
(W 26) Lexically unmarked auxiliary [120 cases] and lexically marked auxiliary [8
(W 27:1) Mutative [3 cases of become].
The tubules become longer and twisted after hatching and may perhaps serve
for salt-reabsorption. (OTU 1, J 1.95)
(W 27:0) Nonmutative [5 cases] (the auxiliaries were feel, look, remain, seem).
Then suddenly, a frown drew together the strong brows, but his eyes remained
tight shut. (OTU 20, M4.47)
Ί know what you're thinking, Doctor.' The florid face looked wounded.
(OTU 9, M4.56)
But still the stones seemed rocked, the unsterile mounds, reimpregnated, exhaled
dust's fever; a breath, impure, of earthbound anguish. (OTU 126, Ml.30)
The difficulty of classifying clauses with Type dJD verbal group is largely due to the
fact that be, its most common auxiliary, has the two values of passive clause auxiliary
and equative clause copula, and the distinction between them is by no means clear-
cut. In equative or near-equative clauses, be often commutes with other, lexically
marked copulas, for example He / was / seemed / sounded / felt / looked happy and
interested. In an attempt to plot some of the values of Type djD auxiliaries in the
corpus, commutation tests were used with seem to be, seem, and feel. Acceptance
varied considerably for the three tests as can be seen in the following table:
Auxiliary Yes Query No
(W 28) seem to be 88 7 33
(W 29) seem 45 6 77
(W 30) feel 15 2 111
if you /look at our !'present royal fämily# - am you can /see the thing which is
[seems to be] known in . genetics# as the /Habsburg :lip# (OTU 80, B2.50)
This change is [seems to be, seems] especially marked in the larva and is produced
by variation in the amount of a pituitary secretion (p. 107). (OTU 84, J1.84)
'No, I don't think we want him,' said Martin, but was [seemed to be, seemed,
felt] disconcerted when his mother said: O h , I am so glad'.
(OTU 114, M3.204)
his/property was [*seemed to be, * seemed, *felt] ordered !not to be confiscated#
but in /fact to be Isold # (OTU 65, Β 1.44)

5.52 The lexical verb

The morphological differences between the types represented by undressed, unexpected,

and unskilled have been dealt with in Section 2.73-4 (pp. 21 ff.). In the taxonomic
experiment a distinction was made between compounds and noncompounds and
between verbal and nonverbal (i.e. nominal) bases.
(W 31) Compound [4 cases] and noncompound [124 cases]. The following is an
example of a compound past participle:
These letters had been so pompous and ill-considered in tone, and so unrelated to
effective action, that it was impossible for Emma to think of them without either
annoyance or amusement. (OTU 40, M2.92)
(W32) Verbal base [126 cases] and nonverbal base [2 cases]. The following had
nonverbal bases:
The skin is mmy-layered ... (OTU 14, J 1.84)
The muscle-fibres run longitudinally and they are striped, but of a somewhat
peculiar fenestrated type. (OTU 91, J1.85)
Nonverbal bases were included in the sample because of their syntactic similarity
with passives. What is said in this presentation referring to the verbal group does not,
strictly speaking, include nonverbal bases, since nonverbal bases cannot by definition
enter into the verbal group. For the sake of simplicity, however, they will be included
here with verbal bases. There are verbs corresponding to the participles layered and
striped, but the -ed forms here seem to have closer relation with nominal bases;
be many-layered and have many layers
be striped and have stripes.
(W 33) Phrasal verbs. The lexical verbs were entered according to the classes set
up in Section 2.71 (pp. 19f.). There were no prepositional verbs in this corpus, but
three phrasal verbs, one of which was the following:
/I think [öi]. that Mr Dean :does -make the -army a-ttractive# or . there was a
/possibility he might# . a /for .[ the people who are going to be :called ύρ#\
(OTU 128, B1.8; [allegro & piano])
(W 34) Qualification [7 cases]. Ved premodified by 'qualifiers', i.e., typically, ad-
jective modifiers {highly, so, more, especially, etc.; cf. Roberts 1955):

It is recorded that when, the invading armies of the Allies reached Belsen, the
troops who witnessed the horrors of the camp for the first time were so affected
by the sights and the smells and the testimony to human depravity that many of
them were taken ill and had to be sent home on leave. (OTU 50, M3.209)
(W 35) Coordination with an adjective and/or a prepositional phrase [5 cases].
Flat fields unearthly green, dotted with grazing cattle, stretched into the distance
on the side they walked on; the other side was broken, hillocky, and patched
with unkempt plantations of smouldering beech and hazel.
(OTUs 41 and 124, M1.18)
(W 36) Potential coordination with an adjective [29 'yes', 3 'query', 96 'no'].
Here, it turns out that energy is needed [and indispensable] to create the 'regions
of misfit' between two domains, because these regions are of finite thickness.
(OTU 64, J3.27)
when ripe they [the follicles] rupture into the coelom, which becomes filled with
[and full of] spermatozoa ... (OTU 83, J1.96)
(W 37) Potential coordination with a prepositional phrase [33'yes', 4 'query', 91 'no'].
when there's / "so many variables and the mathematician is beaten # [or in
doubt/difficulties] . /then we :come !in # (OTU 90, B2.8)
Lady Foxglove ... had somewhere inside her a spring of the toughest steel which
was all the more disconcerting because it was hidden [and out of sight] for most
of the time, ... (OTU 104, M3.206)
Potential premodification by quite, much, rather, very.
Yes Query No
(W 38) quite 33 2 93
(W 39) much 26 5 97
(W 40) rather 18 1 109
(W 41) very 17 2 109
When news of Lady Foxglove's latest excursion in good works reached those
Christian circles in which she occupied so illustrious a place, opinion was [quite]
divided as to whether it was her best work so far, or whether it was conclusive
proof of her approaching insanity. (OTU 37, M3.197)
... the reaction of benzoyl peroxide with benzene -1 14 C, which was [much]
discussed in Section 3a, ... (OTU 49, J4.37)
She was [rather] astonished, and said so. (OTU 113, Μ 1.23)
he doesn't lack quality, I feel [very] convinced. (OTU 72, M2.105)
Potential modifying function of the past participle (Ved). The noun with which this
test was first made was the subject (Nx). When this was not possible for exponential
reasons, as in the case of personal pronouns, a noun was inferred from the context.
It appeared that the participle was much more readily accepted as a noun modifier
when it had some modification of its own. The participles were tested in both pre-
and postmodifying functions, in either case with and without individual adjuncts (A).
The following distribution was obtained:

Yes Query No
(W 42) Ved Ν 49 2 77
(W 43) A Ved Ν 101 1 26
(W 44) Ν Ved 41 3 84
(W 45) Ν Ved A 115 0 13
In the following example the participle could take any of the four functions:
Table 20 shows the good agreement obtained with these parameters for the dis-
ordered structure: for comparison the theoretical values of the intensities of the
decisive reflections (111), (211), and (221) for the ordered structure are also
given. (OTU 111, J2.177)
the given values
-*• the [previously] given values
the values given
-> the values given [by the analyst].
Other participles were preferably modified in some way, for example:
The intimate mechanism of this substitution step cannot be elucidated by exami-
nation of the reaction kinetics and products alone, and further discussion of it is
deferred to the next chapter. (OTU 120, J4.39)
->· *the elucidated mechanism
-*• the [carefully] elucidated mechanism
*the mechanism elucidated
-» the mechanism elucidated [by examination]
The formation of these unsymmetrically substituted diaryls was regarded as an
indication that ... (OTU 89, J4.32)
*the regarded formation
-• *the [previously] regarded formation
*the formation regarded
-» the formation regarded as an indication that ...


(Criteria S 46-59; see Figure 5:3, p. 77)

Most of the categories in terms of which the subject was analysed have been described

in Section 3.2 (pp. 26 ff.), and need only be exemplified here. Some new categories
will also be introduced.
Form-class. There were no nominalizations operating as subject in this corpus.
The nominal groups as S had the following types of head:
(S 46:0) Noun [74 cases].
When the observed background was corrected for multiple scattering and thermal
effects the residual value agreed with this expected amount of diffuse scattering,
within the experimental error. (OTU 2, J2.169)
(S 46:1) Pronoun [49 cases]. Examples will be given in connection with their
specification below under S 47.
(S 46:2) Name [3 cases].
but I /mean . !the real real reason why [ro] a Burns is :so . a :worshipped# is
/because of course he was a :self-made män# (OTU 39, B1.30)
(S 46:-) Zero [2 cases].
Thus a second-order transition of type Β is not, as [0] has been stated by some
authors, impossible [ital. in text] to realise, ... (OTU 57, J3.30)
Specification of pronouns.
(S 47:1) Personal referential pronoun [27 cases] (see, for example, OTUs 72 and
113 above under W 38-41).
(S 47:2) Nonreferential pronoun : proleptic it [3 cases].
Now, it can be shown without difficulty that, for an assembly containing a finite
number of molecules f should be differentiable any number of times with respect
to T, ... (OTU 29, J3.23)
(S 47:3) Relative pronoun [14 cases].
... the radical Ar'... ejects a hydrogen atom which is incorporated in the pro-
ducts N. (OTU 32, J4.27)
The experimental results for the various groups of lines which could be measured
are shown in Fig. 62... (OTU 33, J2.168)
(S 47:4) Demonstrative pronoun [5 cases].
... Davies (1958) has shown that a small but appreciable amount of the ester
ArCO.OAr'... is indeed formed, and this is taken as evidence of the fundamental
correctness of the approach expressed by Scheme III. (OTU 68, J4.38)
Gender, subsuming animate/inanimate, concrete/abstract, count/mass heads of
nominal groups.
(S 48:1) Animate [27 cases].
/those who lären't insäne# the /vast ma!jority#. are con/cerned with !some
personal quärrel# - or have been /driven to desperation by a :personal re:lä-
tion# . or /something of this klnd# (OTU 119, B3.59)
(S 48:0) Inanimate [99 cases].
The purified peroxide, which is generally crystalline, is added to the aromatic
solvent, and the decomposition is effected by heating, usually to about 70-80°.
(OTU 73, J4.34)
(S 49:1) Concrete [77 cases].
The olfactory capsule, imperfectly paired, is also almost detached from the
cranium. (OTU 66, J1.87)
(S 49:0) Abstract [49 cases].
... these investigations were carried out in 1948 (although they were not declassi-
fied and published until several years later) ... (OTU 74, J2.184)
This water must be removed without losing salt; accordingly in most freshwater
animals, including vertebrates, we find some system by which the separation can
be achieved. (OTU 88, J1.93)
(S 50:1) Countable [109 cases].
... the phenolic esters formed thereby are easily removed by hydrolysis and ex-
traction with alkali. (OTU 70, J4.35)
(S 50:0) Mass [17 cases].
The presence of a disordered structure was again substantiated by a marked hump
in the diffuse background of the powder diffraction pattern. (OTU 97, J2.178)
(S 51) Finitude. Definite [110], indefinite [16], zero [2 cases]. Finitude (for cate-
gories other than names and referential pronouns which were all taken to be definite)
was specified as follows:
(S 52:1) Definite article [46 cases].
... the hydrogen atoms are not restricted to the 0-0 bonds as they are in the other
three models. (OTU 27, J2.168)
(S 52:2) Possessive pronoun or 's genitive [5 cases].
The suprarenal tissue receives 'preganglionic' fibres from the spinal nerves. Its
cells sometimes seem to be connected with each other by fibres like those of
neurons ... (OTU 127, J 1.98)
(S 52:3) Demonstrative pronoun [7 cases].
Below the epidermis lies the dermis, a layer of bundles of collagen and elastin
fibres, running mostly in a circular direction. This tissue is sharply marked off
from a layer of subcutaneous tissue ... (OTU 118, J1.84)

(S 52:4) WA-pronoun [2 cases].

/what character, what :physical. body you häve# and [to /some extent] what
•.character you poss6ss# . [to the extent that your] . /"liver< shall we say) moulds
your chäracter#. /is inherited# (OTU 46, B2.44; [allegro]; (parenthetic))
(S 52:5) Indefinite article [3 cases].
A different explanation ... has been offered by Augood and Williams (1957).
(OTU 110, J4.37)
(S 52:0) Zero article [11 cases].
Direct control of the spinal cord from the brain is obtained through a number of
very large Müllers fibres, ... (OTU 13, J1.98)
(S 53:1) First person [4 cases].
/you and I and "äll of iis# are /human b6ings# we re /bora in the same coun-
try# (OTU 123, B1.15)
(S 53:2) Second person [O cases].
(S 53:3) Third person [122 cases] (including 2 zero subjects).
/most of the :gagslüsed# have al/reädy been iised# by the /two chäracters at the
other end of the :täble# (OTU 100, B1.35)
(S 53:-) Zero [2 cases] (see S 46).
(S 54) Number. Singular [80 cases], plural [46 cases].
(S 55) Modification [49 cases] was specified according to:
(S 56) Premodification [33 cases].
An early attempt was made ... (OTU 96, J2.166)
... so far as the heavier elements are concerned. (OTU 28, J2.163)
(S 57) Postmodification [34 cases].
... the formation of diaryls was observed ... (OTU 121, J4.28)
19 subjects had both types of modification, for example:
... the phenolic esters formed thereby are easily removed ... (OTU 70, J4.35)

(S 58) Coordination [6 cases].

... the unit cell and atomic positions of the elements other than hydrogen, or light
elements, are known from X-ray studies ... (OTU 63, J2.162)

... the stability of the transition state, and hence the activation energy for the
dissociation, are influenced ... (OTU 26, J4.32)
(S 59) The number of words in the subject ranged from 0 to 27 with an average of
words, the most frequent being 1 word.


(Criteria C 60-2; see Figure 5:4, p. 78)

(C 60) Presence of clause element C. There were only nine clauses with complements
as defined in Section 3.1-2, pp. 25if. They were the following form-classes:
(C 61:1) Nonfinite verb clause with to-infinitive, active or passive (to V) [5 cases].
[the diencephalon] is known to be concerned, throughout the vertebrate series,
with the integration of the internal activities of the animal. (OTU 94, J1.106)
(C 61:2) Noun [2 cases].
'[this bookshop] does cater for what's called the cultivated reading public - and for
specialists.' (OTU 106, M1.20)
(C 61:3) Adjective [1 case].
... many of them were taken ill ... (OTU 67, M3.209; for context see CL 6:2)
(C 61:4) Nonfinite verb clause with present participle (Ving) [1 case].
... one was left not knowing whether Viola was less innocent or Tanya more inno-
cent than each seemed. (OTU 38, M2.98)
(C 62) The number of words in the complement ranged from 1 to 21 with an aver-
age of words.


(Criteria A 63-108; see Fig. 5:5, p. 79)

Adjuncts were analysed in terms of 46 categories. The clause element A in passive

clauses has a particular interest to a study of voice, since the nominal part of A in its
function as agent is the potential subject of a corresponding active transform (see
Section 3.33, pp. 30f.).
Firstly, a distinction was made between '/4-full' and 'Λ-less' clauses, i.e. according
to whether the clauses had an A element present or not. If they were Λ-full, some
functions of A were established by tests using potential question-forms with wh-
pronouns and w/i-adverbs in Criteria A 66-74. Secondly, the values of the A element
with agentive function (Ag) and nonagentive function (Ad) were analysed separately in
Ag 76-98 and Ad 99-108 respectively.

The distribution in the clauses of Ag and Ad as sets was the following:

v4g-less but Ad-ΐυΜ [60 cases]
/4g-less and ^4<i-less [34 cases]
/4g-full and Ad-full [26 cases]
Ag-full but y4<i-less [ 8 cases]
(A 64) The 94 Ά sets' were the following types: A1 [53 cases], A2 [35 cases], A3 [5
cases], A4 [1 case],
(A 66) Potential question-form with wA-pronoun [51 cases], which was specified as
A 67 who/m/ [15 cases], and A 68 what [36 cases].
(A 69) Potential question-form with wA-adverb with or without preposition [66
cases], which was specified as (A 70) where [21 cases], (A 71) when [19 cases], (A 72)
how [25 cases], (A 73) why [4 cases], (A 74) other (excluding agentive function) [7

... he was ... carried effortlessly by Trained Pig Percy-Scroop-Beauchamp to the

bathroom. (OTU 92, M3.209; see CL 4:3 for context)
(A66) by whom (who ... by)
(A72) how was he carried?
(A70) where/ ... to/
Max Müller was subsequently convinced of the stupidity of his hypothesis and
recanted in no uncertain terms. (1-24)
(A67) of what (what ... of)
ι was Max Müller convinced ...?
(A71) when

5.81 Adjuncts with agentive function (Ag 75-98)

As a consequence of the function-based definition of 'agent' in this study (see Section

3.33), the application of the term will be considerably extended compared with pre-
vious treatments. A distinction will be made between the two classes grammatically
and lexically determined agents. The preposition introducing grammatically deter-
mined agents is by, and they will therefore usually be referred to as 'BY-AGENTS'.
Lexically determined agents are introduced by a variety of prepositions and will be
referred to as 'QUASI-AGENTS'. Unlike by in fty-agents, these prepositions are not
voice-conditioned, but are selected by collocation with particular verbs: be worried
about, surprised at, interested in, annoyed with, etc. Cf. the corresponding nouns and
their prepositions: worry about, surprise at, interest in, annoyance with. The distinction
between by- and quasi-agents may hence be stated in terms of colligation (for Ay-
agents) and collocation (for quasi-agents). (For the use of these terms, see for example
Robins 1964, 67if., 234.) It will also be convenient to let quasi-agents subsume a type
of agentive complements which are not introduced by prepositions. Being lexically

determined, quasi-agents may naturally display specific lexical meanings, such as

cause or instrument. Consider the following series with (a) Z>j>-agent, (b-c) quasi-
agents, and (d) non-agent:

(a) AGENCY: Abel was killed by Cain. -+ Cain killed Abel.

(b) CAUSE: Abel was killed out of hate. -* Hate killed Abel.
(c) INSTRUMENT: Abel was killed with a jawbone. -*• A jawbone killed Abel.
(d) PLACE : Abel was killed outside Eden. *Eden killed Abel.

Agent, cause, and instrument functions are related, and the distinction can be more
or less syntactically neutralized.9
However, in our case the difference between the by-agent (a) and the quasi-agents
(b-c) becomes clear if they are all in turn made active subjects with the other two
phrases retained as prepositional adjuncts:
(a) Cain killed Abel with a jawbone out of hate.
(b) *A jawbone killed Abel by Cain out of hate.
(c) *Hate killed Abel by Cain with a jawbone.
In an example like the following, however, there is neutralization in the active:

• '... just as the instrumental and causal domains border hard on each other and partly pass into
each other, so causality and authorship are distinctions but of degree and not of kind'.'It was from
the instrumental of means with an active verb that the instrumental of agency with the passive had
arisen, i.e. that to a great extent the personal agent owes its provenience to the material instrument
with which a deed is accomplished' (Green 1914, 551, 524f.). Green sets up a series of gradation
(his examples are Latin, but the gradient might well apply to English):
(a) 'material means, pure and simple'
(b) 'personal instrument, intermediary'
(c) 'personified means, often of causal tenor'
(d) 'personal agent'
In this connection it is interesting to note the considerable fluctuation in the development of
English in the use of agent-prepositions, 'first purh begins to lose ground, then fram follows suit,
and then of gives way to by, which becomes the modern preposition of agency through the gradual
supersession of all its competitors' (Green 1914, 552). Mustanoja cites six prepositions (by, from,
mid, o f , through, with) with agent-function in Middle English and points out that 'unambiguous
cases with by indicating the agent of a passive verb are rare until the end of the 14th century' (1960,
Some prepositions other than by occurring with agent-like functions are listed in the handbooks
for present-day English. Of 'is still in literary use, as a biblical, poetic, or stylistic archaism, or by
association with other constructions, e.g. 'on the part o f ' '(admired / loved / hated / ordained of\
abandoned / desirted / forgotten / forsaken of; unseen / unowned of)' (OED. sv. of: V. 15). With, 'the

We shall now illustrate the two agent-classes with examples from the material. Some
quasi-agents (2 g-i below) are peculiar in not being prepositional phrases but C
elements consisting of a nonfinite construction (fo-infinitive) or a finite (zero) that-
clause. The latter may however be regarded as truncated forms of prepositional
phrases of the type 'preposition + the fact + iAai-clause' or some other abstract
I was delighted jto hearI that he won.
To hear that he won
That he won
The fact that he won delighted me.
His winning
Hearing that he won
(1) 57-AGENTS.
(la) Animate by-agent:
Both Hungary's goals were scored by Albert, ... (F.23)
(lb) Inanimate Ay-agent:
The settlers have not been converted by the events of the past fortnight. (G.6)
(2a) About: We were all worried about this
(2b) At: I'm a bit surprised at [N.N.] you know (B5.3)
(2c) Before: Appelby was puzzled before this somewhat incoherent vein. (M6.27)
(2d) In: ... he's passionately interested in pandas. (L2.62)
(2e) To: A man prepared to go pretty high in the matter of buying a large library the
value of which could only be approximately known to him, ... (M6.115)
(2f) With: You won't be bothered with me any more. (L2.74)
(2g) To-infinitive: 'So I was surprised ... to learn that the poor chap had blown his
brains out. (M6.47)
(2h) 77ia/-clause: I'm not surprised actually that Mr [X.] has got on well. with a a
American industrialists and big businessmen (B5.6)
(2i) Zero that-clause: 'I'm surprised that Portway girl was so restless,...'' (M5.79)

JANUS-AGENTS. Some prepositional phrases may have twofold potential function

in active transforms: their nominal part may function as active clause subject, or the
entire prepositional phrase may function as adjunct (usually instrumental, 'by means
of') with some other nominal element as active clause subject. Such adjuncts which

typical preposition in Present English to denote instrumentality, is not unfrequently met with after a
passive participle in a function which is difficult to distinguish from that of by or of': met with strong
opposition, visited with a serious outbreak of typhoid fever (Poutsma 1926-9, 2:2.96f.). Jespersen gives
struck, seized, bit, eaten with, etc. (Part 3.319). In has 'a function approximating to that of agency'
in caught in the rain, covered in black leather (Poutsma, op. cit., 97). To is 'frequent' after known:
Ά man's intimate friends ... are quite unknown to, or by, his wife', and Jespersen asks 'What is the
difference?' (op. cit., 3.319).
permit two different active clause transforms, according to whether they are inter-
preted as Ag or Ad, will be called 'JANUS-AGENTS' (Ag/Ad).10
coal will replace oil
Oil will be replaced by coal.—
[they] will replace oil by coal
How much has in fact been achieved by the mergers that have been made!
Loss of inhibitory power by avenacin on enzymic hydrolysis was confirmed by
growth experiments ... (J5.325)
'... for his land when some official decided to take it from him, does not allow him
some compensation for a business that has been ruined by vacillation and muddle
when some other official has decided to give it back.' (M5.78)
Technically the French will need more tests if they are to develop the nuclear
device tested on Saturday into a weapon that can be carried in an aeroplane or
into a warhead that can be delivered by a missile. (G.15)
In some cases there exist more complicated transformational possibilities, as for
[a] collapse now threatens market prices through
a rush by fanners
Market prices are now threat- a rush by farmers to sell... now threatens market
ened with collapse through a prices with collapse
rush by farmers to sell before farmers now threaten market prices with collapse
the new crop starts in July. through a rush to sell
(E2.29) farmers rushing to sell ... now threaten market
prices with collapse
most languages express 'to whisper' by an imita-
'to whisper' is expressed in tive verb
most languages by an imita- in most languages [people] express 'to whisper' by
tive verb ... (1.52) an imitative verb
an imitative verb expresses 'to whisper' in most
In discussing potential transforms here we can say nothing about the probability of
their use. For the three active transforms of the last example, the probability of
occurrence may be 'likely', 'less likely', and 'least likely' in that order. .ßy-phrases,

This notion helps to distinguish John was frightened by the new methods which is Ag/Ad (Janus-
agent) from The picture was painted by a new technique which is only Ad (nonagentive adjunct). (See
Chomsky 1957, 89 and Lees 1957, 383.)

like other prepositional phrases, can of course also function exclusively as non-
agentive adjuncts:
, ι—• [this] took him by surprise
He was taken by surprise-] . ^ . ..
'-> »surprise took him
, „ , . , ι-» *[they] went him by 3 o'clock
He was gone by 3 ο clock-]
μ *3o clock went him
The last example accepts transmutation (see CL 20), He went by 3 o'clock, which
indicates its close relation to Type Β (see Section 2.4, p. 17).
Adjuncts with agentive function (Ag) were analysed with regard to both overt and
covert criteria.
(Ag 75): see p. 101.
(Ag 76) /lg-position in relation to Ved. It followed immediately after Ved in 23
out of 31 cases of ^g-full clauses.
(Ved + Ag):
The upper surface of the brain is covered by an extensive vascular pad, the choroid
plexus or tela choroidea ... (OTU 51, J1.101)
(Ved + Ad + Ag):
The green air ... was ... smashed to atoms, glowing round the wispy edges of
grey vapour-bright bits of lime-colour that grew smaller and smaller until they
were stamped out into nothing by the cold blackness of continuous cloud.
(OTU 6, M4.45)
(Ved + Ag + Ad):
This mechanism was supported by De Tar and Sagmanli (1950) in an attempt to
explain the products of the reaction. (OTU 10, J4.37)
(Ag 77) 'Janus-agent' [5 cases].
In principle the whole field of organic chemistry is filled with problems which
can be solved by neutron diffraction investigation of the position of hydrogen atoms.
(OTU 15, J2.162)
ι—• which neutron diffraction investigation ... can solve
which [we] can solve by neutron diffraction investigation
(Ag 78) The number of agent words varied from 2 to 40, the average being about
8 words. It is perhaps worth noting that the curve for agent-length is much more
irregular than the one for subject-length.
(Ag 79) Potential deletion occurred in 17 cases, plus 2 queries. This criterion
proved to be difficult to apply, and hence was of little value in its present formulation.
Nevertheless, it would seem true that, in the examples above, the agent in OTU 6 is
more dispensable than those in OTUs 51 and 15.
(Ag 80) Specific exponents of the agent preposition: by [30 cases], under [1 case].

The sacs form by evagination from the brain and remain connected with the
dorsal epithalamic or habenular region of the between-brain by two stalks.
(OTU24, J 1.103)
Presumably the osmotic flow of water into the body is relieved by the pressure
of the heart-beat forcing water out from the glomeruli into the coelomic fluid,
whence it is removed by the funnels, with the aid of their cilia.
(OTU 5, J 1.95)
The clause with under as 'quasi-agent' was the following:
'... on behalf of the National Society for the Prevention of Suffering in Helpless
Children, I am empowered, under the Act of 1948, to make an inspection of the
premises and circumstances of living of one Tarquin Foxglove, an adopted child
of this address.' (OTU 99, M3.206)
-»• the Act of 1948 empowers me to make an inspection
The nominal part of the agent, i.e. the potential active subject, was analysed in terms
of the same categories as the passive subject (see S 46-59): form-class, gender, finitude,
person, number, modification, and coordination (for other examples of this typical
agent-feature, see for example OTU 69, p. 82 and OTU 101, p. 87). The categories
need not be further illustrated here. Instead we shall compare some of the inventories
for both subject and agent in Ag-full clauses. (One of the ^g-full clauses had zero S
which accounts for the discrepancies between the totals of the 5 and Ag inventories.)


Form-class: Noun 20 23
(Ag 81) Pronoun 10 0
Name 0 8
Gender: Animate 5 11
(Ag 82) Inanimate 25 20
(Ag 83) Concrete 17 19
Abstract 13 12
(Ag 84) Mass 3 10
Countable 27 21
Finitude: Definite 26 19
(Ag 85-6) Indefinite 4 12
Person: First 1 0
Second 0 0
Third 29 31
Number: Singular 17 13
(Ag 87) Plural 13 18


Modification: 17 23
(Ag 88)
(Ag 89) Premodification 12 12
(Ag 90) Postmodification 14 19
Coordination: 1 10
(Ag 91)
No. of words (average): 8

(Ag 92) Extension with an agent was possible in 107 clauses, impossible in 17, and
was queried in 4. 'Possible' includes the 31 cases of expressed agents (see Ag 75 above).
(Ag 93-94) The most common potential agent preposition was by (102 'yes', 2
'no', 6 'queries'). Other elicited prepositions were at, of, and under (see Ag 80 above).
She was astonished [at/by the dog's behaviour] (OTU 113, Μ 1.23)
we're /born in the same country # [of English parents]u (OTU 123, Β 1.15)
Potential exponence of the nominal part of the agent was tested for the following
four categories (suggested agents are given in square brackets).
(Ag 95) Potential animate agent [70 cases].
Knowing the value of R® we can determine the structure factor F, when extinction
is negligible, by the equation (7.1) as has already been described [by the author]
in Chapter III. (OTU 30, J2.171)
(Ag 96) Potential inanimate agent [53 cases].
Now that Martin was a full lieutenant he was entitled [by tradition/his position,
etc.] to a Jhon [ital] by himself ... (OTU 75,M3.203)
(Ag 97) Potential concrete agent [75 cases]. See for example OTUs 34, 30, 24
(Ag 98) Concrete agent in the second of two possible agent types [2 cases]; abstract
only [15 cases]. Ag 98 was introduced as a very tentative criterion following the
observation that when a second type of agent was possible, this agent or near-agent
(it was usually a less likely agent than the first) was remarkably often abstract, as for
instance in the following two examples (where the second agents are of the Janus-
'In modern use the connexion with bear is no longer felt; the phrase to be born has become
virtually an intr. verb' (OED, .s.v. born). In the sense 'bring forth, give birth to' only borne is used
with by and the name of the mother, whereas born is restricted to the passive, when it has no by-agent,
and only in one sense: 'it has rather a neuter signification = "come into existence, sprung"' (OED,
s.v. bear, 44). Because of the tenuous relation of born to bear (*English parents bear us), the o/-phrase
with born, which was supplied in this example by the informant, has doubtful status, and indeed
does not qualify as agent in the definition of Section 3.33 (p. 30f.).
Faintly she frowned, contemplating the area of her labours, seeing what should
have been, what could be done. (OTU 36, Μ 1.17)
-»· what could be done [by her/by hard work]
'and if the truth is to be expressed I never shall [pass].' (OTU 54, M2.90)
-* if the truth is to be expressed [by me/by (in) a few words]

5.82 Adjuncts with nonagentive functions (Ad 99-108)

(Ad 99) The 86 Ad-iuYi clauses (see the beginning of Section 5.8) were analysed in
terms of the following categories:
(Ad 100) Number of Ad items in each clause (i.e. Ad set types, see Section 4.5,
pp. 57f.):
Ad1 [64 cases], Ad2 [18 cases], Ad3 [3 cases], Ad* [1 case].
(Ad 101) Closed class adverbs [11 cases], for example again, already, almost, first,
... the full mathematical details of the theory of non-equilibria, as will now be
shown. (J5.301)
Probably this reputation was always exaggerated. (G.6)
(Ad 102) Open class adverbs (usually formed by the suffix -ly added to an adjective)
[7 cases], for example absurdly, effortlessly, intimately, recently, neatly.
He tempted Cartwright with two half volleys which were royally punished ...
The invitation hadn't been very felicitously phrased, but was entirely cordial.
(Ad 103) Prepositional phrases [61 cases]. When the phrase had potential question-
form with wA-pronoun (see A 66), the exponents of the preposition were stated in
Ad 104. They were, in order of frequency, from [7 cases], with [6 cases], to [6 cases],
for [4 cases], as [4 cases], of [3 cases], and into [1 case].
the /only way of deciding [/whether it's a powerful deterrent#]# - is by [/look-
ing at 'what has happened#]. /in the leases to which it {:has in the p ä s t # been
appVlled#} (OTU 78, B3.26, [spiky], {rhythmic})
... the occurrence of an irreversible effect is so intimately connected with the
mechanism of the transition itself that ... (OTU 31, J3.34)
the [sei?] "sole offence# . for/which i t . :could be [applied#]
(OTU 58, B3.30; [low/creak])
The available evidence will be considered to lead to two interpretations and
these will be used as a framework for discussion. (J5.275)

Too much, however, should not be made of Griffins temporary embarrassment.

..., until an ancient parish church was made into the choir of a new cathedral for
Truro in the present century. (G-9)

(Ad 105) Finite or nonfinite verb clauses [12 cases].

Not until the sun was moving down over the blue hills of Donegal was the destiny
of these two matches decided, ... (F.28)
T o qualify for a record consideration round a curve the race must be made on a
track not more than 440 yards in circumference and started on some part of the
circumference.' (F.30)

Position (see Section 3.32, p. 30):

(Ad 106) Front-position [14 cases].
(Ad 107) Mid-position [14 cases].
(Ad 108) End-position [65 cases].
The first of the following two examples has a closed class adverb in mid-position;
the second has a finite clause in front-position and, in end-position, a prepositional
phrase plus a nonfinite clause whose subject is introduced by a preposition.

... this private enterprise Emperor would inevitably follow the same primrose
path that had already been blazed by other British long-range airlines designed
since the war. (OTU 11, M4.49)

... in the hydrogen distribution the average coherent cross-section per deuterium
nucleus would be 4nbb, but when there are twice as many possible sites as
atoms this will be reduced to 2nbt>, with the remaining cross-section of 2nb'b,
which amounts to 2.6 barns, appearing as diffuse background scattering.
(OTU 47, J2.169)


In Chapter 5 we introduced and illustrated a large number of criteria which were

applied to a small number of randomly selected finite passive clauses. This chapter
will describe the preparation of the input data, the operations of the classification
program, the analysis of the output, and the setting up of a diagnostic key.


The 108 criteria were applied to the 128 clauses (called OTUs', short for 'operational
taxonomic units'1), whose 'features' were entered into cells. Linguistic analysis and
informant reaction were used to determine, respectively, the features to be assigned
to overt and covert criteria. (For the two types of criteria, see Section 5.3, p. 74.) This
meant that 128 χ 108 = 13,824 cells had to be filled in.
The performance of an experiment on such a scale is a laborious process: to extend
the same detailed mode of analysis to a large corpus would be impracticable. What
we must hope for is that the results of such an experiment, or preferably a series of
such experiments, will give us sufficient insight into the problem to enable us to
detect certain 'key features'. Such criteria, once we have found them, can then be
economically applied to a very much larger corpus, and will thus provide a practical
means of arriving at a good classification.
In order to achieve such a sophisticated goal as was outlined in Section 5.1, pp. 72 f.
(that the classification should be, for example, objective and empirically verifiable, con-
sistent with intuition, predictive, and comprehensive), it would be necessary to follow
up pilot experiments on the lines proposed here with a series of related experiments,
each one successively making use of the previous results. There are many reasons
for such a procedure. The most important is that with a small corpus we have no
guarantee that the OTUs will include all passive clause types. The most that one
can expect is that a pilot experiment will provide a classification of only the most
frequent classes. An attempt to avoid this undesirable effect was made in this experi-
Here, it will be possible to give only a condensed outline of the operations of the classification
program and other technical aspects of the experiment. For a detailed discussion of relevant problems
in numerical taxonomy and for further references, see Carvell & Svartvik 1966.

ment by means of a provisional ordering of the material (as mentioned in Section 5.2,
p. 73).
The features were encoded in two ways, according to whether or not absence of a
certain feature counted. Criteria in the input data will be referred to as having
'levels'. For example, Criterion CL11 had two levels: 'potential active transforma-
tion' and 'no potential active transformation'. In the first class of criteria, absence
of the same feature in two OTUs is disregarded for purposes of comparison. This
class includes 'dichotomies' (in the input encoded as 1 and 0) and 'positives', or
'one-level qualitatives' (encoded as 0 and -), where 0 and - , respectively, denote
absence. An example of such criteria is the modification of the subject and agent
(S 55-7, Ag 88-90, pp. 100, 108).
In the second class of criteria, which may be called 'qualitatives' and 'quantitatives',
absence of the same feature in two OTUs counts as a similarity. Qualitatives take
levels ranging from 0 upwards (/, 2,3,4, etc.) referring to mutually exclusive features.
An example of a 3-level qualitative is subject form-class (see Criterion S 46, p. 98),
where 0 denoted noun, 1 pronoun, 2 name, and - zero. In quantitatives the levels are
an ordered series in a linear scale. This test type was suitable for the encoding of
element lengths, as in the subject (S 59, p. 101) and agent (Ag 78, p. 106). In order to
obtain what was thought to be a realistic scale, the logarithms of the element lengths
were entered in the data input.2
In principle, the first class of criteria was used for optional overt features, such as
modification and adjunct position (Ad 106-8, p. 110). The second class of criteria,
on the other hand, was used for obligatory overt features, such as element gender
(animate/inanimate, concrete/abstract, countable/mass, see S 48-50, Ag 82-4, pp. 98 f.;
107), and for covert features, such as agent extension potential (Ag 92, p. 108) and
potential active transformation (CL 11, pp. 83 if.).
The hierarchal relationship of some of the criteria (as is clear in Figures 5 :1-5, pp.
75 ff.) posed a special technical problem. In a tree with many sub-branches, the first
branches, which represented the primary distinction, tended to be considered more
and more similar as the number of sub-branches increased. In order to counteract
this undesirable effect in hierarchal criteria, 'dummy' levels were entered. For
Criterion S 47, p. 98, for example, which specified pronouns (personal referential,
proleptic it, etc., all of which were entered as 1 in S 46), a dummy 0 was encoded for
non pronouns ( = S 46:0, S 56:2, and S 46:-). 3
An example of the input form is given in Figure 6:1, which shows the first four
OTUs labelled by their finite lexical verbs: TWIST, CORRECT 1 (where 1 indicates that
this is the first of two or more occurrences of this verb in the corpus), EXCITE, and

The base of the logarithms was irrelevant. The effect of this was, for example, that the similarity
between elements of word lengths 1 and 2 was the same as that between elements of word lengths
2 and 4, or 5 and 10.
® Objections may be raised to this procedure on theoretical grounds. The problem has not, however,
been solved, and there was little choice in the programming facilities available.
0100 11110000 011111 110100 11 001123 3260 022250 0—
—01 -1111— 22 1010-1 -1--0- 0—7- —0 -00011 010
0122 11 6 6 - - 1101

0010 10110000 011002 100000 1100— 001113 3260 022203
00000— 22 1101-1 -1—0- 30 —0 -00011 0-0
1022 11 (10) (16) - - 1101
1000 11110000 111011 111001 11 111613 3260 022200
—1 01111— 22 1010-1 -1 7- -00011 010
0022 00 0 - - - 0000
0100 10110010 011002 000000 110-0- 001113 3260 022250
—0 00000-0- 22 1101-1 - 1 - 0 — 0—7- -0— -00001 010
0122 11 (10) 0 1101
Fig. 6:1. Specimen of data input (OTUs 1-4).

TOUCH. The clauses containing these verbs are cited on pp. 94,98,119, and 91 .respect-


The classification program (CLASP), which had been written for the ICT Orion com-
puter, classified the OTUs in terms of their resemblances, these being estimated on
the basis of their common features. The two principal operations performed by
CLASP are, very briefly, as follows: firstly, it calculates a coefficient of similarity
between each pair of OTUs; secondly, it sorts the OTUs into a nested hierarchy of
groups on the basis of the similarity coefficients. Furthermore, as a by-product of
this sorting process, it produces a linear ordering of the OTUs, such that the OTUs
of any group that has once been formed are never separated.
These operations produced the following output:
(a) THE FULL SIMILARITY MATRIX. This is a subdiagonal matrix of the similarity
coefficients printed to three significant figures. Both the names and the numbers of
the OTUs are printed in a column on the left hand side, against the corresponding
rows; the numbers only are printed at the bottom, at the foot of the corresponding
columns. The order chosen for the OTUs in both rows and columns is the order
referred to above, i.e. that obtained as a by-product of the sorting-process. Figure
6:2 shows the top part of the full similarity matrix.

* The data were actually prepared for a program called F39 written for an Elliott 401 computer.
Progress in this field is, however, extremely rapid, and by the time the data were ready for input a
new classification program (CLASP) was available for the Orion, a much larger computer than the
401. In addition to larger storage, permitting as many as 400 OTUs and 256 criteria to be used for
each OTU, CLASP provided many additional facilities, which were made use of. I am greatly
indebted to J. C. Gower, Rothamsted Experimental Station, for his cooperation in this experiment.

CORRECT 1 2 70.6
TOUCH 4 74.9 85.5
INSTALL 34 76.5 87.4 94.9
DISCOVER 98 74.0 84.7 92.6 94.2
HIRE 55 74.0 85.7 88.8 93.5 89.5
SPREAD 16 72.1 76.3 84.4 82.8 84.3 78.9
DESCRIBE 30 73.6 83.1 87.0 87.0 87.2 89.2 89.6
AVOID 48 72.4 80.1 91.1 90.1 88.8 85.1 82.9 86.8
ACHIEVE 88 76.7 79.9 87.1 88.6 85.9 84.7 78.6 80.5 93.1
DISCUSS 2 49 67.1 82.6 80.8 83.8 84.6 81.2 91.1 88.2 80.3 78.8
PUBLISH 74 69.7 85.2 82.3 83.0 84.6 81.9 89.6 91.4 81.6 78.6 91.3
HANG 1 77 72.0 80.8 87.6 88.7 87.7 90.7 78.9 85.6 80.7 80.0 77.8 83.5
INCLUDE 1 79 77.2 74.9 81.2 83.0 84.5 82.7 79.4 82.7 82.5 86.5 77.5 79.6 88.0 - - - -
SUMMARISE 87 78.9 83.5 88.3 90.0 90.4 87.2 85.2 91.7 88.5 87.0 81.7 86.9 87.5 87.4
GIVE 111 79.6 82.4 86.5 87.3 87.8 85.9 83.0 87.8 86.0 89.3 81.4 85.8 88.4 91.4
Fig. 6:2. Specimen of output: top part of the full similarity matrix
(printed to three significant figures).

(b) A B B R E V I A T E D SIMILARITY MATRICES. The full similarity matrix will normally be

found to contain more detail than is realistic in view of the number of criteria, and
more than is possible for the mind to comprehend, even after careful inspection. For
this reason the program has the additional facility of producing abbreviated similarity
matrices, which have coarseness as their characteristic feature. We may distinguish
two types of such matrices: the 'basic type' which is printed to one significant figure
only: i.e. using only the numbers 0-9; and the 'derived type', which is normally coarser
than the basic type.
The linear ordering between individual OTUs, as well as between groups of OTUs,
may be improved by the analyst. There are a number of options offered by CLASP,
some of which are in practice only applicable after the first output has been obtained.
On inspection of our first output, the OTUs appeared to form six major groups, and
they were rearranged accordingly. It is possible to define for the computer any
number of OTU groups with the members of the groups ordered according to the
analyst's preference. In one case this was done taking all the OTUs as an ordered
group, and in another case dividing the OTUs into six groups (I-VI). Needless to
say, this procedure cannot alter the similarity coefficients: it only provides the linguist
with a more readily analysable output, and also gives him an opportunity to ask for
the forms of abbreviated similarity matrices which are most realistic.
printed in the form of a small matrix. This is represented in Figure 6:3 as a shaded
similarity matrix.
(d) T H E MOST TYPICAL MEMBER OF A G R O U P is obtained by working out the mean
similarity of each member of a particular group to every other member of the group.
The OTUs are printed in descending order of mean similarity.
(e) G R O U P INVENTORIES can also be computed by CLASP. They are printed in the
form of two-way tables which state the number of OTUs in each group which attains
each level. In addition, the program gives the feature inventories for each individual


Per cent similarity

GROUP II 70-79

S5 ϊ ί ί




Fig. 6:3. Non-proportional, shaded similarity matrix representing mean similarities

between and within groups.


6.31 Taxonomic analysis

The abbreviated similarity matrices indicate that the OTUs fall into six groups of
varying internal and external relationship.
At the top of the triangle there are two closely related groups (I and II) with an
inter-group similarity of 70 per cent. Group I has an internal similarity of 73 per cent
and contains 19 members (the first and last of which are OTUs 70 and 99, going
downwards in the linear order of the matrix). Group II has a similarity of 79 per cent
and contains 12 members (OTUs 102-119).

Together, Groups III and IV comprise more than half of all OTUs. Group III
has 53 members (OTUs 34-38) and a similarity of 76 per cent;6 Group IV has 17 mem-
bers (OTUs 8-7) and a similarity of 78 per cent. These two groups are closely
associated: their mean similarity is 72 per cent, which is the highest between any two
Group V is the smallest cluster, comprising only 6 members (OTUs 3-23), with an
internal mean similarity of 75 per cent.
Finally, Group VI has 21 members (OTUs 117-44) and 76 per cent intra-group
mean similarity.
The mean similarities between and within the six groups are represented by different
shading in Figure 6:3. Groups III and IV are clearly central because of their fre-
quency and, above all, because of their external relationships. They are related to
Groups I and II as well as to Groups V and VI, whereas Groups I and II, on the one
hand, and Groups V and VI, on the other, have little affinity. For some purposes,
we may well be justified in conflating Groups III and IV, since they are so closely
related to each other and so similarly related to other groups. Groups I and II,
likewise, show great affinity. However, it is worth noting that Groups II and III are
more similar than I and III, than II and IV, and, particularly, than I and IV. Groups
V and VI are not highly related; Group V is more similar to III and IV than to VI;
Group VI is more similar to III than to V.
With the exception of Group VI, the groups conform most closely to a cline dis-
tribution, i.e. clusters which are serially related in a continuum.

6.32 Statistical assessment and linguistic interpretation

So far, the output analysis has been entirely taxonomic, since we have considered
the groups without any reference to their contents. We shall now give the output a
linguistic interpretation, basing our assessments, as far as possible, on considerations
of statistical significance. However small some of the numbers may seem, they are
nevertheless valid for statistical purposes, and the significances of their distributions
may be established by means of contingency tables and the χ 2 test.®
It is necessary to discuss briefly the nature of the statements we make about 2 x 2 contingency
tables. The situation will be that we have four entries, as for example in Table 6:1, which
shows the number of coordinate and non-coordinate subjects and agents (Criteria S 58 and
Ag 91, p. 100,108) in the Taxonomic Passive Corpus, 31 clauses (i.e. just under one quarter)
were agentful. Two clauses (one Λ^-full and one Λ^-less) had zero S which accounts for the
discrepancies between the totals of the S and Ag inventories.

' Considering that a large group tends to have a lower internal mean similarity than a small group,
the percentages for Group III, and for III/IV, are quite high.
• For testing significances in tables each one of whose marginal totals does not exceed 40, we have
used Finney et. al. 1963. Significances in tables with higher marginal totals have been established by
means of the χ2 test, using Yates' correction, or where some entries were small, by means of Fisher's
exact test (see, for example, Herdan 1964, 37ff.; Moroney 1956, 249ff.; Reichmann 1964, 335ff.

t> υ
> 'w
c Μ
α. cd
bO α ft
QO cd cd
ca υ Vi cd ft C
Γ"\>- Λ "
•α Μ
. >« 00 ο
ö 2 Κ e02-
a - . εE « S : J>
.• ?!
ii >
* - Ή
·π to 3 CO cd ΐ-Ρ ft
tn 3
Μ C > « 3 > CO
w ο c <
μ cd Έ η s SS §
< C
ο < eQ u
rt hJ ^ Λ
a u ft
< g
w Ζ

α c
υ -μ a>
60 cd βρ
c cd
υ 'S .2
SP ft 3 £ 3
n 3 Ο 3
ο σ· ο, σ1

£Λ Η
Β η
« <2 u
Ο c
.Ξ c ο
— ι cd
d «J 3 cd rt
4-» • 3 y
C Ο υ
e ω > cd
α •ts
c ao § Ο
α, &
υ cd

cd £
cd cd
ε c
6 e υ
cd Ο c
ι- υ
< cd Ο- Ο 60
— Λ

cd —
3 cd
3 ~ υ
Ι 2 c C
Η ΜΙ ο ω cd ω
Μ 2Ρ Ο Ο
< £ Ζ
ο 5
Ζ ft

Ε «
• κ AJ (ί-. ,ο
C Ο C/3
α .ä c ft e >
ο Χ a 2
β. ο a Ξ
ω cd
A ft
§ Ι
ζ 8

8 .δ 8
cn «a
cd i ä
U, ft 73

Criteria S 58, Ag 91

Coordinate Non-coordinate Total
Subject 1 29 30

Agent 10 21 31

Total 11 50 61

On the basis of Table 6:1 we may hypothesize that there is a greater tendency to co-
ordination in the agent than in the subject. It is unreasonable to hope for an estimate of how
likely it is that this hypothesis is true. Instead we proceed indirectly: assuming provisionally
that there is no connection, how unlikely is the table? The significance test used is based
on the principle of the Null Hypothesis, which 'postulates that, as between two samples,
there is no really significant difference and that such difference as occurs is attributable only
to random sampling errors'. 7 More precisely: we establish the chance - assuming the
Null Hypothesis - that this table, or one even more biased, would have occurred. The
situation we consider as given is provided by the row and column totals. In the example, this
would give the following 12 alternatives, not all equally likely (indeed, A is more likely
than L):


There are two reasonable interpretations of 'or one even more biased'. One applies when
we wish to test the bias in the form it has taken. This will be so if we have already suspected
that, for example, coordination was more likely to occur in agents than subjects. In this case
we shall be interested in the probability of one of A or Β on the Null Hypothesis. This is
usually known as the single-tail case.
On the other hand, we may only wish to know whether the table is sufficiently remarkable
to be worthy of investigation, in which case we shall have to consider bias in the opposite
direction as well. Suppose the Null Hypothesis is true: at the 95 per cent level, we may then
expect a deviation in one direction once in twenty times, and in the other direction once in
twenty times. (To work at the 95 per cent level means, roughly, that we shall on average
be wrong 5 per cent of the time.) Altogether, we may expect to be in error once in ten times
if we test for significance at the 95 per cent level, using the single-tail test, and we shall
therefore have to alter the significance level. Here, we shall use single-tail tests.
An example may make the situation clearer. Suppose a coin is tossed five times and
comes down heads every time. How remarkable is this? If we know that the tosser is a
cricket captain who has called 'heads', we should be justified in saying that the result is,
on our terms, significant, for the chance of five heads, with an unbiased coin, is If,
however, the tossing is done with no particular purpose in mind, we shall have to say that it
is not so remarkable: all that has happened is that the same result has been obtained five
times, and that five tails would have been equally surprising. The chance of five heads or

Reichmann 1964, 225.

five tails is of course not so small, being in fact The first situation we have described is
the single-tail case; the second situation is the double-tail case.
The probability of the tables is given as p, assuming the Null Hypothesis. The terms used
will be 'non-significant' for ρ > 0.05 (i.e. i ) , 'probably significant' for ρ < 0.05, 'significant'
for ρ < 0.01 (i.e. and 'highly significant' for ρ < 0.001 (i.e. ^g).

Before we go on to discuss the characteristic features of the groups, it will be helpful

to list their members. We shall first exemplify the most typical member of each group,
and then list the other OTUs in order of their mean similarities with the other members
of the group. (Owing to a computer printing error no value was obtained for OTU
62.) Percentages denote the mean similarity of that member to every other member
of the group. In the exemplification of the most typical members, the entire OTUs
are given in italics. In the list of OTUs, references are made for each OTU to place
of citation in Chapter 5 (criterion and page).
Most typical members:

Group I: Since aroyloxy radicals appear to be considerably less reactive entities

than aryl radicals in aromatic substitution, the extent of aroyloxylation
is generally small, and the phenolic esters formed thereby are easily
removed by hydrolysis and extraction with alkali. (OTU 70, J4.35; 77 %)
Group II: The first investigations of the chloride by neutron diffraction were made by
Goldschmidt and Hurst (1951) using the powder method for deuterated
ammonium chloride. (OTU 102, J2.174; 83%)
Group III: The baby was installed in the old nursery wing at Bidcombe, and a very
expensive nanny was hired from London to look after it.
(OTU 34, M3.198; 81%)
Group IV: there had been "/no sig nificant increase# of "/any -kind# in /crime#
. and there was /nö . pr£ssure# . for the /reintrolduction of xorporal
: punishment# . and "/no belief# that if it "/were introdüced# it would
[/do anything about the .'crime • w a v e # ]
(OTU 8, B3.46; [allegro]; 84%)
Group V: '... Truly, what do we live for?' He became excited·, vodka splashed in
his glass; professors nibbling cheese straws peered over the tops of them
to see what was happening. (OTU 3, M2.107; 80%)
Group VI: In later vertebrates the choroid extends only into the third and fourth
ventricles. Presumably the vascular membranes of the brain are highly
developed in lampreys because of the absence of cerebral blood vessels.
(OTU 117, Jl.101; 80%)

List of group members in order of internal group similarity. Forms with nominal
bases (layered, striped), compounds (unknown), and participles with tenuous relation
to verbal bases (bom, taken ill) are given in their forms of occurrence:

Group OTU OTU Mean Criterion Page

number name similarity citation citation
70 REMOVE-3 77% Most typical member
120 ELUCIDATE 76% W 42 (p. 97)
73 EFFECT 76% S 48 (p. 99)
51 COVER 75% Ag 76 (p. 106)
5 REMOVE-1 75% Ag 80 (p. 107)
19 SURROUND 75% CL 19 (p. 88)
61 COMPENSATE 75% CL 18 (p. 87)
11 BLAZE 75% Ad 106-8 (p. Ill)
97 SUBSTANTIATE 74% S 50 (P. 99)
25 ATTACK 74% CL 2 (p. 78)
59 REMOVE-2 73% CL 19 (p. 88)
6 STAMP OUT 72% Ag 76 (p. 106)
15 SOLVE 72% Ag 77 (p. 106)
127 CONNECT-4 71% S 52 (p. 99)
24 CONNECT-1 70% Ag 80 (p. 107)
26 INFLUENCE 69% S 58 (p. 100)
18 MOTIVATE 69% CL 18 (P. 87)
50 AFFECT-2 67% W 34 (p. 96)
99 EMPOWER 65% Ag 80 (p. 107)
102 MAKE-2 83% Most typical member
100 USE 82% S 53 (p. 100)
96 MAKE-1 82% S 56 (p. 100)
10 SUPPORT 82% Ag 76 (p. 106)
110 OFFER-2 81% S 52 (p. 100)
101 DEMONSTRATE 80% CL 18 (P. 87)
69 INVESTIGATE 79% CL 8 (p. 82)
125 SEE-3 78% CL 18 (p. 87)
42 DISCUSS-1 78% CL 16 (p. 85)
57 STATE 77% S 46 (P. 98)
92 CARRY 74% CL 4 (p. 80)
119 DRIVE 70% S 48 (PP. 98-9)
34 INSTALL 81% Most typical member
111 GIVE 81% W 42-5 (p. 97)
98 DISCOVER 80% CL 4 (p. 80)
87 SUMMARISE 80% CL 13 (P- 83)
30 DESCRIBE 80% Ag 95 (p. 108)
86 SEE-2 80% CL 15 (p. 84)
4 TOUCH 80% W21 (P- 91)
74 PUBLISH 80% S 49 (p. 99)

Group OTU OTU Mean Criterion Page

number name similarity citation citation

III 2 CORRECT-1 80% S 46 (p. 98)

III 88 ACHIEVE 80% S 49 (p. 99)
III 48 AVOID 80% W 23 (p. 91)
III 76 BRING 79% W 21 (P- 91)
III 55 HIRE 79% CL 15 (p. 84)
III 43 CORRECT-2 79% CL 6 (P- 81)
III 89 REGARD 79% W42-5 (p. 97)
III 71 OBTAIN-2 79% CL 2 (p. 80)
49 DISCUSS-2 79% W 39 (P- 96)
III 63 KNOW-1 78% S 58 (p. 100)
III 68 TAKE-1 78% S 47 (P- 98)
III 79 INCLUDE-1 78% CL 10 (p. 83)
III 52 FORM-1 78% CL 2 (p. 80)
III 80 KNOW-2 78% W 28 (p. 94)
III 45 DEDUCE 78% CL 4 (p. 80)
III 121 OBSERVE 77% S 57 (p. 100)
III 16 SPREAD 77% CL 20 (p. 89)
III 77 HANG-1 77% CL 10 (p. 82)
III 22 BITE 77% CL 19 (p. 89)
III 13 OBTAIN-1 77% S 52 (p. 100)
III 78 APPLY-3 76% Ad 103 (p. 109)
III 64 NEED 76% W 36 (p. 96)
III 27 RESTRICT 76% S 52 (p. 99)
III 95 UNKNOWN 76% CL 18 (p. 86)
III 118 MARK-2 76% S 52 (p. 99)
III 85 CONSIDER-3 76% CL 7 (p. 82)
III 115 INCLUDE-2 76% CL 18 (p. 86)
III 58 APPLY-2 75% Ad 103 (P- 109)
III 47 REDUCE 75% Ad 106, 108 (p. 110)
17 APPLY-1 75% CL 13 (p. 84)
III 29 SHOW 74% S 47 (p. 98)
III 93 CONVERT 74% CL 2 (P- 80)
III 107 REMOVE-4 74% CL 17 (P- 85)
III 83 FILL 73% W 36 (P- 96)
III 75 ENTITLE 73% Ag 96 (p. 108)
III 12 STRIP 73% W23 (P- 91)
III 104 HIDE 73% W 37 (P- 96)
III 90 BEAT 72% W 37 (p. 96)
III 53 BREAK-2 71% CL 17 (P- 85)

rroup OTU OTU Mean Criterion Page

number name similarity citation citation

111 1 TWIST 71% W 27 (P- 94)

111 94 KNOW-3 70% C 61 (p. ιοί)
III 60 ALLOW 70% CL 7 (p. 81)
III 103 OFFER-1 68% CL 13 (p. 83)
III 105 BALANCE 67% W 21 (P· 91)
III 38 LEAVE 67% C 61 (p. 101)
IV 8 INTRODUCE 84% Most typical member
IV 81 KILL 83% CL 19 (p. 89)
IV 36 DO 82% Ag 98 (p. 109)
IV 33 MEASURE 82% S 47 (P· 98)
IV 82 RADIATE 81% W 25 (p. 92)
IV 116 FLOG 81% W 24 (p. 91)
IV 56 CONSIDER-2 81% CL 2 (p. 80)
IV 54 EXPRESS 81% Ag 98 (p. 109)
IV 46 INHERIT 80% S 52 (p. 100)
IV 39 WORSHIP 80% S 46 (p. 98)
IV 128 CALL-2 80% W 33 (p. 95)
IV 109 SEE-3 79% CL 18 (p. 87)
IV 106 CALL-1 77% C 61 (p. 101)
IV 108 TRAIN 77% CL 7 (p. 82)
IV 65 ORDER 76% W28-30 (p. 95)
IV 7 SEE-1 74% CL 18 (p. 86)
V 3 EXCITE 80% Most typical member
V 113 ASTONISH 80% CL 13 (p. 84)
V 72 CONVINCE 77% CL 6 (p. 81)
V 9 WOUND 77% W27 (p. 94)
V 114 DISCONCERT 76% CL 4 (p. 80)
V 23 AFFECT-1 60% CL 6 (p. 81)
VI 117 DEVELOP 80% Most typical member
VI 84 MARK-1 80% W28-9 (p. 94)
VI 66 DETACH 80% S 49 (p. 99)
VI 21 CONTINUE 79% CL 20 (p. 89)
VI 112 CONNECT-3 79% CL 4 (p. 80)
VI 126 ROCK 78% W27 (p. 94)
VI 14 LAYERED 78% W 32 (p. 95)
VI 122 EXPAND 78% CL 20 (P· 89)
VI 41 BREAK-1 77% W 35 (p. 96)
VI 32 INCORPORATE 77% S 47 (p. 98)
VI 91 STRIPED 77% W 32 (p. 95)

Group OTU OTU Mean Criterion Page

number name similarity citation citation

VI 37 DIVIDE 76% W 38 (p. 96)

VI 28 CONCERN 75% S 56 (p. 100)
VI 31 CONNECT-2 75% Ad 103 (p. 109)
VI 40 CONSIDER-1 75% W 31 (p. 95)
VI 20 SHUT 75% W 27 (p. 94)
VI 35 AGREE 73% CL 20 (p. 89)
VI 124 PATCH 73% W 35 (P- 96)
VI 123 BORN 70% S 53 (p. 100)
VI 67 TAKEN ILL 69% CL 6 (p. 81)
VI 44 COMPOSE 69% CL 4 (p. 80)
Some of the groups correspond very closely to our previous intuitive classification
(see Section 4.11, p. 40): Group I is, roughly, coterminous with Class β, Group II with
Class a, and Groups III and IV with Class γ. We shall now consider all six groups
one by one as units. It will be convenient to say, for example, that 'Group II has
animate agents' instead of 'the members of Group II have animate agents'.
Groups I, II, III, IV, and V have as their common distinctive feature potential
active clause transformation (Criterion CL 11, pp. 83 f.). There is one exception: OTTJ
23 in Group V. It is clear from the list above that it is a misfit and should have been
placed in Group VI. Not only is this OTU the least typical member of Group V, but
the gap between it and the previous member (OTU 114) is as great as 16 per cent. This
difference in internal group similarity is far greater than that between any other two
adjacent OTUs in any group. For Groups I-IV, transformation is possible only into
the extensive active clause type (CL 12); Group V usually admits of transformation
into both extensive and intensive (CL 13) active clause type.8 Group VI (with OTU
124 as a doubtful exception) takes no active transformation. The first two groups have
expressed Zry-agents (Ag 75): Group I has inanimate and Group II animate agents
(Ag 82, p. 107). Groups III/IV may have agent extension (Ag 92, p. 108) with animate
(Ag 95, pp. 108f.) or inanimate (Ag 96) agents; Group V usually takes only inanimate
These will be our 'key criteria' since they are common to all, or almost all, the
members of the groups which have been formed on the basis of the total set of criteria.
Hence, these criteria may be used to define almost identical groups when we go outside
the small sample used in the experiment. In addition, there are a number of other
'Extensive' is used here in contrast to 'intensive' for distinguishing, for example, This remark
embarrassed him from This remark made him embarrassed as transforms of He was embarrassed by
this remark. These terms are convenient but somewhat inaccurate, since a transformation of a
passive factitive clause into an active factitive (for example He was called stupid [they] called him
stupid), which is a transformation from one intensive clause to another, is here called 'extensive'
(see Criteria CL 12-3, p. 83 f.).

Criterion S 55, Groups I/II, IV (p = 0.0018)

S modified S non-modified Total

Groups I/II 18 13 31
Group IV 2 15 17
Total 20 28 48

Criterion S 55, Groups / / / / , III/IV (p = 0.011)

S modified S non-modified Total

Groups I/II 18 13 31
Groups III/IV 22 48 70
Total 40 61 101

Criteria CL 7-8, Group I (p = 0.009)

Coreferent with Non-coreferent with Total

contiguous clause 5 contiguous clause S

active clause 8 11 19
passive clause 1 18 19
Total 9 29 38

Criterion W 29, Groups I, II (p = 0.0004)

Commutation with No commutation Total

seem with seem

Group I 12 7 19
Group II 0 12 12
Total 12 19 31

features which are not common to as many class members but are frequently con-
comitant with their key features (cf. what was said about 'predictive power' in Section
5.1, pp. 72f.).
Subjects of agentful clauses (Groups I/II) are, for example, more often modified
than other subjects (Criterion S 55, p. 100; see Tables 6:2-3 for the relevant contin-
gency tables). 9 Group I subjects are frequently coreferent with subjects of active, but
* This particular application of contingency tables must be interpreted with some caution, and is
further discussed in Carvell & Svartvik 1966, Section 7.
Criterion CL 19, Groups I, II (p = 0.019)

Potential No potential Total

permutation permutation

Group I 11 7 18
Group II 2 10 12

Total 13 17 30

not of passive, contiguous clauses (CL 7-8, p. 82; Table 6:4). Unlike Group II, its
passive auxiliaries commonly commute with seem (W 29, p. 94; Table 6:5), and
potential permutation (CL 19, pp. 88ff.)is more frequent than in Group II (Table 6:6)
and, in fact, in any other group.
GROUPS I/II have a number of complementary characteristics associated with
their key features (inanimate agent for Group I, animate agent for Group II). Unlike
inanimate agents, animate agents are usually coordinate (Ag 91, p. 108; Table 6:7) and
deletable (Ag 79, p. 106; Table 6 :8). Inanimate agents tend to be modified more often
Criterion Ag 91, Groups I, II (p = 0.019)

Coordinate Non-coordinate Total

Ag inanimate 3 16 19
Ag animate 7 5 12

Total 10 21 31

Criterion Ag 79, Groups I, II (p = 0.028)

Ag deletable Ag non-deletable Total

Ag inanimate 7 10 17
Ag animate 10 2 12

Total 17 12 29

than animate agents (Ag 88, p. 108; Table 6:9); furthermore, they tend to be indefinite
(Ag 85, Table 6:10), and countable (Ag 84, Table 6:11). Animate agents, on the other
hand, are more often definite and mass nouns. In order to find out whether there is,
normally, association in nominal heads (as was, in fact, found to be the case between
concrete and animate) between definite and mass or between indefinite and countable,
the χ 2 test was applied to the subject, the other nominal element in the same sample
which was similarly analysed. Table 6:12 shows that there is no evidence of associa-
tion between animate/inanimate gender and definite/indefinite in the subject through-
out the corpus (p — 19/20), whereas the distribution for the same categories in the

T A B L E 6:9
Criterion Ag 88, Groups I, II (p = 0.022)

Ag modified Ag non-modified Total

Ag inanimate 17 2 19
Ag animate 6 6 12

Total 23 8 31

TABLE 6:10
Criterion Ag 85, Groups I, II (p = 0.007)

Ag definite Ag indefinite Total

Ag inanimate 8 11 19
Ag animate 11 1 12

Total 19 12 31

TABLE 6:1 1
Criterion Ag 84, Groups I, II (p - 0.002)

Ag count Ag mass Total

Ag inanimate 17 2 19
Ag animate 4 8 12

Total 21 10 31

TABLE 6:12
Criteria S 48, S 51, Groups I-IV (non-significant)

S definite S indefinite Total

5 animate 23 4 27
5 inanimate 87 12 99

Total 110 16 126

agent has ρ — 0.007 (Table 6 :10). Similarly, Table 6:13 shows that the distribution
of animate/inanimate and count/mass in the subject provides no evidence of associa-
tion, whereas ρ = 0.002 for the same categories in the agent (Table 6:11). Tables
6:14-15, giving the two distributions expressed in percentages, show that the marginal
totals are only moderately different; the entries are much more so. Compared with
G r o u p I (Table 6:16), and indeed all other groups, G r o u p II clauses have notably
often free external clause relation (CL 5, p. 81) and verbal groups with a Type b
auxiliary (W 24, p. 91).
GROUPS I I I / I V have zero agents as their joint key feature: they may take agent

TABLE 6:13
Criteria S 48, S 50, Groups I-VI (nonsignificant)

Scount S mass Total

S animate 25 2 27
5 inanimate 84 15 99

Total 109 17 126

TABLE 6:14
Criteria S 48, S 50, Groups I-VI

S count 5 mass Total

S animate 19.8% 1.6% 21.4%

S inanimate 66.7% 11.9% 78.6%

Total 86.5% 13.5% 100%

TABLE 6:15
Criteria S 48, S 50, Groups I, II

Ag count Ag mass Total

Ag animate 12.9% 25.8% 38.7%

Ag inanimate 54.8% 6.5% 61.3%

Total 67.7% 32.3% 100%

TABLE 6:16
Criterion W 24, Groups I, II (p = 0.007)

Type b auxiliary No Type b auxiliary Total

Group I 2 17 19
Group II 7 5 12

Total 9 22 31

extensions with by (Ag 92, Ag 93, p. 108) and subsequent active transformation (CL
11, pp. 83f.)· Apart from its key features, Group III has hardly any overall individual
characteristics. Group IV, however, has the following additional features: its clauses
never have adjuncts (Ad 99, p. 109), and their external clause relationship is almost
invariably bound (CL 1, p. 78). In respect of these two features combined, Group IV
is at variance with all the other groups. Furthermore, its subjects are often pronouns
(S 46, p. 98), and just under half of them are coreferent with the object or predicative
of a contiguous clause (CL 9, p. 82). The participles of these clauses, more often than
those of other group clauses, can operate alone as nominal postmodifiers (W 44, p. 97);

TABLE 6:17
Criteria W 38, W 40, W 41, Groups IV, V (p = 0.00001)

Qualification No qualification Total

Group IV 0 17 17
Group V 6 0 6

Total 6 17 23

TABLE 6:18
Criterion W 30, Groups IV, V (p = 0.0002)

Commutation No commutation Total

with feel with feel

Group IV 0 17 17
Group V 5 1 6

Total 5 18 23

they do not admit of qualification with quite (W 38), rather (W 40), very (W 41)
(Table 6:17). In this respect Group IV is like Group II; Groups I and III have some
admissions of qualifiers, whereas Group VI has a scatter. Similarly, like Group II,
the passive auxiliaries of Group IV do not commute with seem (W 29, p. 94). Here,
Group I usually admits seem (12:7), Group III usually not (13:37 with 3 queries).
Group IV clauses are generally shorter than clauses in other groups, and their
functions are essentially those of postmodifying relative clause or clausal adjunct.
In view of their similar relations to other groups (see Figure 6:3), Groups III and IV
may be conflated, with Group IV as the 'marked' subgroup.
GROUP V has its typical features: the potential dual transformation into extensive
and intensive active clauses (CL 12-3, pp. 83 f.); most of its Type d auxiliaries commute
with seem (W 29), and with feel (W 30, Table 6:18); all its participles permit co-
ordination with adjectives (W 36) and qualification (W 38, W 40, W 41, Table 6:17),
but cannot usually operate alone as postnominal modifiers (W 44). This group differs
from all other groups in that its external clause relation is normally free (CL 5, p. 81),
and its subjects typically animate (S 48, p. 98). Subjects are, furthermore, frequently
pronominal (S 46), and they are not coreferent with an object or predicative in a
contiguous clause (CL 9).
GROUP VI does not usually admit of agent extension (Ag 92, p. 108). There are four
exceptions, all of them doubtful cases: OTUs 40,44,123,124. Nor do they take direct
active transformation (CL 11; one exception: OTU 124). The passive auxiliaries of
this group normally commute with seem (W 29); it is interesting that its complex
verbal group never has a Type a auxiliary (W 22, p. 91; Table 6:19). It has sometimes
potential transmutation (CL 20, p. 90; Table 6:20) and potential coordination with
adjectives (W 36). This is the only group besides Group V whose participles may often

operate alone as prenominal, but infrequently as postnominal, modifiers (W 42, W 44,

Table 6:21).
TABLE 6:19
Criterion W 22, Groups III, VI (p = 0.02)

Type α No Type a Total

auxiliary auxiliary

Group III 11 42 53
Group VI 0 21 21

Total 11 63 74

TABLE 6:20
Criterion CL 20, Groups Ijll, VI (p == 0.05)

Potential No potential Total

transmutation transmutation

Groups I/II 2 29 31
Group VI 6 13 19

Total 8 42 50

In addition to the characteristics of the agentful groups (I/II) with regard to each
other or to other groups, we may assess the probabilities of the subject and agent
distributions within each group. On the basis of our sample, there is good statistical
evidence to show that subjects and agents in agentful clauses differ from each other
significantly in many respects.
Agents tend to be longer than subjects. In the Taxonomic Passive Corpus, subject
lengths varied from 0 to 26 words, with a mean of 3.3 words and a mode of 1 word (in
the Minor Passive Corpus the lengths were 4.96 words and 1 word, respectively);
agent lengths varied from 2 to 40 words, with a mean of about 8 words, and a mode
of 6 words (8.5 and 2 words, respectively, in Corpus II). Curves showing frequency
and length of the two elements are quite different in that the subject curve decreases
more or less geometrically from 1 word on, whereas the agent curve is irregular (cf.
Section 4.36, pp. 52ff.).
As many as one third of the clauses in Corpus III and one half in Corpus II (see
Section 4.35) have coordinate agents, whereas coordination is rare in subjects of any
group (6 out of 128).
In agentful clauses, subjects tend to have definite and count nouns, and agents to
have indefinite and mass nouns (Tables 6:22-24). Subjects are usually nouns; agents
are usually nouns or names, but never pronouns in the sample (Tables 6:25-30). The
tables indicate that the chances of having, for example, a name as subject and a
pronoun as agent are very small indeed. The probability of Table 6:30 is 1/50,000!

TABLE 6:21
Criteria W 42, W 44, Group VI (p = 0.0009)

Possible Not possible Total

VedN 11 10 21
Ν Ved 1 19 20

Total 12 29 41

TABLE 6:22
Criteria S 50, Ag 84, Groups I, II (p = 0.03)

Mass Count Total

Subject 3 27 30
Agent 10 21 31

Total 13 48 61

TABLE 6:23
Criteria S 51, Ag 85, Groups I, II (p = 0.05)

Definite Indefinite Total

Subject 26 4 30
Agent 19 12 31

Total 45 16 61

TABLE 6:24
Criteria S 51, Ag 85, Group I (p = 0.0006)

Definite Indefinite Total

Subject 18 1 19
Agent 8 11 19
Total 26 12 38

TABLE 6:25
Criteria S 48, Ag 81, Group I (p = 0.01)

Pronoun Noun or name Total

Subject 6 13 19
Agent 0 19 19
Total 6 32 38

TABLE 6:26
Criteria S 48, Ag 81, Group II (p = 0.04)

Pronoun Noun or name Total

Subject 4 7 11
Agent 0 12 12
Total 4 19 23

TABLE 6:27
Criteria S 48, Ag 81, Groups / / / / (p = 0.0003)

Pronoun Noun or name Total

Subject 10 20 30
Agent 0 31 31
Total 10 51 61

TABLE 6:28
Criteria S 48, Ag 81, Groups / / / / (p = 0.003)

Noun or pronoun Name Total

Subject 30 0 30
Agent 23 8 31
Total 53 8 61

TABLE 6:29
Criteria S 48, Ag 81, Groups / / / / (p = 0.002)

Noun Pronoun Total

Subject 20 10 30
Agent 23 0 23
Total 43 10 53

TABLE 6:30
Criteria S 48, Ag 81, Groups I/II (p = 0.00002)

Pronoun Name Total

Subject 10 0 10
Agent 0 8 8
Total 10 8 18

It is of course impossible to give an estimate of the frequency, since there are no

occurrences of pronominal agents in the Taxonomic Corpus.


The diagnostic key is meant to be an economical method, arrived at via the OTU
classification, for obtaining a sufficiently detailed yet realistically coarse classification
of new instances of passive clauses. It is conceivable that several different key features
might give roughly the same classification. The diagnostic key proposed here should
therefore be seen only as a first step towards a better key, which can evolve only from
further experience.
Several factors must be taken into account when we set up a diagnostic key.
Besides fulfilling the basic requirement of providing the means of a 'good' classification
(as stated in Section 5.1, pp. 72 f.), a diagnostic key should be simple to use. The
selection of key criteria is largely conditioned by practical convenience. One factor is,
necessarily, the ease with which the criteria can be applied. A criterion which applies
very rarely or tends to produce different responses from different informants, or from
the same informant on different occasions, may not be very pernicious to the OTU
classification, but it may not satisfactorily be used for diagnostic purposes.
We found that the groups in the output had a small number of individual charac-
teristics (key features), which defined all, or almost all, the members of the groups.
Hence, these criteria have a high information content, and it is natural that we should
use them in the diagnostic key. Unlike the OTU criteria, which were unweighted (see
Section 5.1), key criteria are weighted and ordered.
The output of the taxonomic experiment showed that the OTUs formed clusters on
a scale. The diagnostic key proposed here is designed with a view to bringing out a
cline. However, it has not been felt necessary or even desirable to follow the output
order slavishly in constructing the key.
The proposed classification comprises six major classes (α, β, γ, δ, ε, ζ), each of
which may admit of varying degrees of further breakdown.

6.41 Classes a and β

Class α and Class β, which are at one end of the scale, have close voice relation, with a
systemic one-to-one correspondence between passive and active (see Section 3.7, p. 33).
The active transformation rule can be applied without systemic restrictions to generate
active transforms. It is important to emphasize the very considerable restrictions not
subsumed under 'systemic' here: they include, for example, sequential constraints
(see CL 16, p. 85), case alteration for pronouns with object forms (he.hirn, etc.),
adjunct placement, and stylistic imbalance (see CL 18e, p. 87). The key features of
Class α and Class β are, respectively, expressed animate by-agent and expressed
inanimate fry-agent. Since Class α corresponds to Group II and Class β to Group I,
see Section 6.32 (pp. 125f.) for 'concomitant features'.

Class α ('animate agent passives'):

He was given this puppy by a farmer in the Welsh hills. (Ml. 18)
Some work has also been done, notably by Broadhead and Pauson (1955), on the
arylation of ferrocene and phenylferrocene with various diazo compounds. (J4.27)
But more drastic action is to be demanded in Parliament by Mr. Fred Peart,
Socialist Shadow Minister of Agriculture. (E2.30)
if he'd been /introduced :properly and decently to :army life# he would /"not
have been followed by . '.press photographers and :press reporters# (Bl.l)
Class β ('inanimate agent passives'):
/mäy I -say# we've been "/well rewarded# [by our /visit to IBognor NlRegis#J
(B4.75; [narrow])
The removal of large quantities of water is an important problem in all freshwater
animals and is facilitated by a high pressure in the kidneys. (J 1.92)
... French, which, as already observed, remained the first language until after
Napoleon, when it was exceeded by German, which in its turn has recently been
surpassed by Russian. (1.31)
Pratt bowled rather longer than was justified by results, ... (F.23)

6.42 Class β/γ

One reason for reversing the output order (Class α = Group II, Class β = Group I)
is that there is frequently structural indeterminacy between Class β and Class γ
(which comprise clauses with potential fty-agents): prepositional fry-phrases with
inanimate nominal parts are often 'Janus-agents' (see Section 5.81, pp. 102ff.), i.e
they may be interpreted either as agents (Ag), in which case they are Class β clauses,
or as other adjuncts (Ad), in which case they are Class γ clauses. Such clauses will
be classified as β/γ.

'The purpose of the Government Bill to abolish resale price maintenance is to

ensure that the full force of competition is brought to bear and that profits and
dividends are not kept at an artificially high level by maintained prices', he said.
These sex differences, which develop shortly before spawning, can also be initiated
by injection of anterior pituitary extracts ... (J1.97)
Let us say that anything which is introduced, or can be introduced, into a remark
by an expression is a term ['term' in italics] (K.146)

6.43 Class γ

Class γ ('agentless passives') is the central and most frequent passive clause class. It
corresponds, roughly, to Groups III/IV, and has no expressed fry-agent, but may have
'direct' agent extension (which is usually animate) with subsequent systemic potential
active transformation. 'Direct' denotes that agent extension and active transformation
are possible within the same tense (cf. Class ε below).
No subdivision will be made here corresponding to Groups III and IV. The main
reason for this deviation from the output results is that these two groups are more
closely related than any other two (see Section 6.32, pp. 126ff.). This is all the more
remarkable in view of their large sizes. Furthermore, the one feature which distin-
guishes all Group IV members from Group III, and from any other group, is absence
of adjuncts, and we would feel reluctant to make absence of what is usually an optional
clause element a key feature in the final classification (see Section 3.1, pp. 25 f.). On
the other hand, further predominant features, such as bound external clause relation,
indicate that we may be justified in regarding Group IV as a functionally important
subclass of Class γ.

Ί could have got a job in the war. I was offered a decent one, in the B.B.C....'
The person convicted was fined only a pound.' (E2.45)
Ilife is lived. in the present # and /one . Iworks . for the füture# (B2.26)
The invitation hadn't been very felicitously phrased, but it was entirely cordial.
Many varieties of laterals are heard in English, ... (1-41)
Order has been restored without bloodshed and without concessions, ... (G.2)
The word or its synonyms mean quite different activities according to the context
and manner in which it is used. (H.13)
At last we have high-fashion knitwear that can be washed by hand and in the
washing-machine... (C.334)

6.44 Class δ

Class δ occupies an intermediate position on the 'passive scale' in that its members
have both verbal and adjectival properties. Their verbal character is manifested in
their potential transformation into extensive active clauses; their adjectival character
in their potential transformation into intensive active clauses (see Criterion CL 13,
pp. 83 f.), and in their ability to take coordination with adjectives, qualification, and
lexically marked auxiliaries {feel, seem, etc.). Members of this class have, typically,
animate subjects. Class δ is divided into two subclasses.
Class δ2 corresponds closely to Group V (see Section 6.32, p. 128). These 'emotive
passives' usually have, or can have, quasi-agents (see Section 5.81, p. 104), particularly
with the lexically unmarked auxiliary be (see p. 93).
Members of Class δ 1 (we may call them 'attitudinal passives') do not take quasi-
agents but are instead remotely related to Class β, and hence indirectly to γ, since
they may occasionally have fry-agents which are typically inanimate. This is reflected
in their position above δ2 in the classification.
Examples of Class δ 1 ('attitudinal passives'; cf. Bolinger 1961b, 379ff.):
We are encouraged, therefore, to use the radar data to obtain drop-size distributions.
... when Watson felt compelled to call up Pratt, the crisis appeared to have passed.
I've always been far too inclined to treat important objects as part of my own petty
existence. (M5.29)
Examples of Class δ2 ('emotive passives' with quasi-agents have also been illustrated
in Section 5.81):
Gerald was suddenly very annoyed. (M5.36)
I'm entranced with the whole idea. (LI. 14)
Mrs. Cressett ... was uncertain but interested. (M5.78)
For a moment Rood looked offended. (M6.40)
Indeed, there soon emerged a group of psychologists who were interested in
problem-solving and who were critical of Thoradike's methods and theory. (H.35)
... it struck him that the girl was now rather frightened. (M6.102)

6.45 Class ε

Class ε ('nonagentive passives') is the most multifarious class of all. This class repre-
sents the end of the 'passive scale' where the relation to active is most tenuous. When
asked if members of this class may have agent extension and active transformation,
native informants often vary considerably in their responses. Such reactions must
be taken as evidence that, towards this end of the passive scale, it is no longer realistic
to consider these clauses in terms of voice transformation potential. In Class ε
clauses, 'external agents' are either unlikely or impossible, because no 'agent' (used
in its non-technical sense, i.e. 'actor', 'doer', 'logical subject', etc.) is conceived of.
Many such clauses may be said to have their agents 'internalized' (cf. Bolinger 1961 b,
378), which is manifested by such actual or potential features as coordination of their
past participles with adjectives and other predicatives in equative clauses (W 36-7,
p. 96), lexically marked auxiliaries (W 26-7, p. 94), and qualification (W 34, pp. 95f.).
Moreover, ε-clauses may frequently have transmutation (CL 20, p. 90).
Class ε includes 'statal passives', which are related to Class γ in that they may take
agent extension and active transformation, but only 'indirectly', i.e. a change of
tense must take place. (See Criterion CL 17 and the subsequent discussion, pp. 85f.)

In the experiment, members of this subclass did not form a clear-cut group of their
own but were found among the least typical members of Group III (see, for example,
OTUs 107, 53, 105).
There he stood, in this difficult atmosphere, gleaming in his white coat, a tray
in either hand on which was tastefully arranged a bowl of turtle soup, a dish of
olives and celery and sliced tomatoes and a glittering array of tools to tackle
further delights to come. (M4.58)
Cavendish inquired with grave courtesy,'Is the thesis finished, Doctor?' (M4.57)
the /general# who /said my \flanks are :türned# my /centre is bröken# /1 shall
attäck# (B4.16)
'Sandra - listen'
'Not another word. The die is cast.' (LI. 17)
In the following example, the precise passive value of was posed is indicated by the
aspectual contrast of the active had posed:
She was posed - for it was much as if her companion had deliberately posed her
for his own pleasure and Appleby's - in a small shifting dapple of sunlight and
shadow. (M6.97)
The subclassification of Class ε will not be further pursued here, apart from making a
distinction between clauses with mutative and nonmutative values. The proper
context of a detailed description of Class ε (as well as of δ) would include equative
and factitive clauses (see Section 3.8 and Chapter 8).
Class ε 1 has mutative value, and the auxiliary may be unmarked {be) or marked
(become, get, etc., see Section 5.51, pp. 93f.).
This should be compared with Fig. 2, which shows the diffraction pattern when
a stearic acid monolayer has been retracted on the silver, with the crystal in the
same azimuth. (J5.301)
Not until the sun was moving down over the blue hills of Donegal was the destiny
of these two matches decided, and of all the long day the last hour must come first.
As the modern world becomes more highly industrialized and mechanized,... (1.33)
'Think of the episode of supreme savage comedy when Hamlet jumps into her
grave and gets jammed in it.' (M6.16)
As a consequence the Order Paper gets silted up, questions get stale, and the
intervals between a particular Minister's appearance at the Dispatch Box become
longer and longer. (G.9)
The significance of the mystery, however, was lost on Clarissa. (M5.21)
Lexically marked auxiliaries (notice gets silted up, get stale, become longer in three
sequentially related clauses in the last but one example) emphasize the decreased voice
relationship, since the agentive α-, β-, γ-classes normally take only be as passive auxiliary.
Many of the lexical verbs in these clauses are mutatively marked by their suffixes,
as in specialize, departmentalize, etc.
Class ε2 has nonmutative value, and the auxiliary may be unmarked (be) or marked
{look, remain, etc.).
It was not their sense of patriotism that was involved. It was their sense of smell.
... Nordic, Mediterranean, and Alpine, once discernible races, are now intermingled
everywhere to a greater or less degree. (1.21)
'As far as poetry and that sort of thing is concerned, Lewis's taste simply didn't
exist.' (M6.65)
Often an older firm is disproportionately strong in foreign business, while a
younger firm may be concentrated rather in the home industrial financing ... (G.10)
They are most contented as they are. (LI.57)
'... suppose that she was simply mistaken about the ink.' (M6.81)
She was not normally given to calling people 'dear' or to leering at them, ...

6.46 Class ζ

Class ζ comprises compounds generally, where the isolation from the lexical verb is
morphologically marked (see Section 2.73, pp. 21 f.). Although compounds are
morphologically isolated, they are not necessarily syntactically isolated with regard to
voice, in so far as they may occasionally have agents (cf. example in Section 2.74 and
below). One compound in the Taxonomic Passive Corpus (OTU 95) was in fact
grouped with Group III. But most compounds in the material are nonagentive, and
have, frequently, lexically marked auxiliaries and coordination with equative predica-
tives. The lexically marked auxiliary and the compound participle are often in close
collocation (pass unnoticed, come unstuck, etc.).

But Cavill was unimpressed by this sally. (M6.54)

The significance of these photomechanical changes is unknown but they demon-
strate that the pineal cells are sensitive to light. (J 1.103)
'Not all the parts of a thought can be complete; at least one must be "unsaturated"
or predicative·, otherwise they would not hold together.' (K.152)
In relying upon the grammatical phrases, 'substantival expression' and 'expres-
sion containing a verb in the indicative mood', the distinction seems both parochial
and unexplained ... (K. 148)
Exodus flights are, no doubt, made also by individuals from the smallest possible
populations but pass unnoticed. (J5.349)
But her question remained unspoken. (M5.45)
As long as clause four stands unaltered and as long as the variably influential section
of the party which subscribes to a full-blooded interpretation of it goes undefeated,
so long will the contradiction remain. (G.l 1)

... the South Africans' final anxieties had been self-inflicted. (F.25)
Old Professor Wali-Anu will be heart-broken. (LI.8)
Pig breeders have not been as panic stricken as some of their spokesmen, ... (G.5)
It was hoped ... that recent agitation on those lines was ill-founded and artificial.


The classes we have set up are not clearly delimited, and they differ in size and degree
of interrelation: α and β have much in common; so have δ and ε, but the latter two
classes also accommodate clearly distinct subclasses; δ, for example, has a cline
within itself, one end of which is ß-like, the other ε-like, and so on. A representation
in a single dimension can only inadequately depict the complexity of their relation-
The key in Figure 6:4 (p. 117)is a diagrammatic representation of the classification. It
is designed to be used in applying the key criteria, but it is not intended to be anything
more than a coarse guide to the major classes. Progression may take place only in
the direction of the arrows, along an itinerary which is determined by a series of
binary choices at the nodal points. The diagram should be self-explanatory, except
for the double entries leading to Class δ: entry may be made either via 'potential
extensive active transform' or via 'no potential extensive active transform'. The
reason for this is that the criterion of potential extensive active transformation applies
less readily to some members of Class δ than the small number of OTUs in Group V
led us to believe. It seems therefore realistic to admit this fact by providing two
paths leading to the same goal.
This weakening of our prime criterion - potential extensive active transformation -
is only to be expected, as we proceed down the scale, away from the pole with close
transformational voice relation. Passive clauses (in the wide sense that we have used
the term 'passive') may be seen as related to active clauses but in different ways and
to varying degrees. Agentive clauses, at the top end of the passive scale, are trans-
formationally related to active clauses (transitive, ditransitive, factitive, see Section
3.8), whereas non agentive clauses, at the bottom end of the passive scale, have
syntagmatic affinity with active equative clauses. In between the two extremes of the
scale there are intermediate classes with mixed properties. Generally speaking, we
may say that, as the degree of transformational voice relationship decreases, the
degree of syntagmatic relationship increases.


This chapter will be devoted to a study of the use of the passive voice in a fairly
extensive selection of contemporary English, the Major Passive Corpus (see Section
1.5, pp. 8 f.). Passives that occurred in this material will be classified according to the
system set up in Chapter 6, and discussed with regard to clause types, verbal group
types, and texts.


In Chapter 3 we discussed some major clause types, particularly in the light of their
voice relations. Having narrowed the scope to the passive, we shall now introduce
a more detailed clause typology for this term in the voice system. Although the
FACTITIVE (int-C) are unsatisfactory in that they are modelled on the names of their
corresponding active clauses and hence imply, misleadingly, that there is always a
direct active-passive relationship, it seems nevertheless useful to have an indication
in easily convertible terms of potential clausal voice relation, provided the transforma-
tional constraints are also borne in mind.
We can achieve an increase in the delicacy of our analysis by subdividing passive
transitives into clauses which have and clauses which do not have a complement-like
prepositional adjunct in close collocation with the verb. The clauses which do not
have such collocation retain the original clause label, whereas the others are further
distinguished according to their internal extensive and intensive clause relations, as
these terms have been defined in Section 3.7 (pp. 33 ff.). The three types may be
illustrated by the following examples.
Many famous American and English philologists were trained in Germany in
the nineteenth century, ... (1.16)

'How are we going to jockey Middleton into accepting the editorship that will
undoubtedly be offered to him?' (M5.27)


... the London School of Economics and Political Science has come to be regarded
as a thoroughly English academic institution. (G.6)
The semifactitive type of clause differs from the sem'ditransitive in having an equative
relation between its subject and the head of the prepositional phrase.
-* the LSE is a thoroughly English academic institution
Consequently it has, for example, the characteristic number restriction, which does
not apply to extensive clauses.
*the LSE has come to be regarded as thoroughly English academic institutions
-*• the editorship ... will undoubtedly be offered to them
The close linking of verb and preposition is perhaps particularly striking in those
verbs which have the alternative possibilities of retaining or omitting the preposition,
as for example:

(ext-C) The letter was given her.

(ext-C") She was given the letter.
(ext-A) The letter was given to her.
(int-C) This was considered a duty.
(int-A) This was considered as a duty.
The first two transforms represent the primary and secondary passive ditransitive
clause types; the third is like the primary except for the preposition to. Despite
obvious parallelism between the two, such a phrase will be called A and not C, i.e.
it will not be considered an indirect object (see Section 3.9, p. 37).
The fact should not be overlooked that the ίο-phrase also has affinity with other
prepositional phrases which are not similarly related to complement-structures.
Compare, for example,
The letter was addressed to her.
which has no transforms
""The letter was addressed her.
*She was addressed the letter.
On the other hand, these prepositional phrases should be distinguished from adjuncts
with clear adverbial value and less cohesion between verb and preposition, as in
The letter was addressed in the post-office.
There is no ready way of establishing the diverse values of prepositional phrases, and
for the present purpose we shall have to be content with a simple dichotomy into
what may be called DOLOSE (as in addressed to her) and ΛΌΡΕΝ (as in addressed in the
post-office). The criterion used for this separation will be the test with wA-questions
(see Criteria A 66 and A 69 in Section 5.8, p. 102). Class Λ CLOSE has question-forms
with wA-pronoun (whojm/ or what) + preposition, whereas Class Λόρεν had question-
forms with wA-adverb (where, when, how, etc.) ± preposition. The question-form
should have unmarked tonicity and falling tone, as in
/who(m) was the letter addressed -to#
/where was the letter addressed#
In case of overlap, i.e. when either question-form seems equally natural, the guiding
principle has been to consider the pronominal criterion as superordinate to the
adverbial criterion. This applies, for example, to such phrases as
She was considered as a friend.
which may have either of the following question-forms (cf. Section 3.73, pp. 34f.):
What was she considered as!
How was she considered?
Λ c l o s e is intermediate between C and Λ ό ρ ε ν . In clause designation, this will be
indicated by '-A' attached to the extensive (ext-A) and intensive (int-A) clause labels;
in the clause name, by the prefix 'semi' added to the corresponding major clause type
('semiditransitive' and 'semifactitive'). Λόρεν (which may occur in any clause type)
will not be specially registered here, and hence no distinction is made between clauses
with or without such adjuncts.


Excluding compounds from the present study, we may conveniently deal with the
following three major passive classes separately: agentive, quasi-agentive, and non-
agentive passives.
7.21 Agentive passives
Agentive passives (Classes α, β, β/γ, γ; see Sections 6.41-3, pp. 132if.) probably
correspond fairly closely to the general concept 'passives' in English. The centrality of
agentive passives to the passive construction as a whole is evidenced by their high
frequency in our material: 2696 occurrences, i.e. almost three quarters of all collected
passive clauses. Of our agentive passives, 80 per cent are agentless (Class γ), and the
remaining 20 per cent of agentful clauses are equally divided between the animate
(Class a) and inanimate (Class β) by-agent classes. This proportion of γ-passives is
only slightly higher than for the Minor Passive Corpus, which had over 76 per cent
(497 out of 650 occurrences; see Section 4.11, p. 40). It conforms with the figures given
by Jespersen: 'between 70 and 94 per cent' (1924, 168), and 'over 70 per cent of
passive sentences found in English literature contain no mention of the active subject'
(1933, 121). All the agents in our a- and ß-passives are fey-phrases. The figure for
ß-passives also includes the ambivalent class of Janus-agents (β/γ, see Section 6.42,

Fig. 7:1. Frequencies of the five passive clause types in the agentive classes.

p.133), which accounts for over one fifth of the 265 occurrences with inanimate
In respect of clause type, 73 per cent (1970 occurrences) of the agentive passives
are transitive, 9 per cent (249) semiditransitive, 3 per cent (74) ditransitive, 5 per cent
The use of clause types in agentive passives

^v class Class α Class β Class β/γ Class γ Total
type \
Passive 216 195 55 1504 1970
transitive (8%) (7%) (2%) (56%) (73%)

Passive 16 7 0 226 249

semiditransitive (1%) (0%) (·) (8%) (9%)

Passive 8 0 0 66 74
ditransitive (0%) (•) (·) (2%) (3%)

Passive 9 1 0 125 135

semifactitive (0%) (o%) (.) (5%) (5%)
Passive 17 7 0 244 268
factitive (1%) (0%) (.) (9%) (10%)

Total 266 210 55 2165 2696

(10%) (8%) (2%) (80%) (100%)

(135) semifactitive, and 10 per cent (268) factitive (see Figure 7:1). We shall now
illustrate some of the passive clause types in relation to the passive classes. The
frequencies are given in Table 7:1; the verbs are listed in the Appendices (pp. 167ff.).
The predominance of the PASSIVE TRANSITIVE clause type (ext-0) is particularly
notable in agentful constructions, where as many as 88 per cent (466 occurrences) are
of this type (see Appendix la, pp. 167ff.).
(a) The children - who have not been told the seriousness of their father's
illness - were met by their mother, ... (E2.18)
(a) So far I have been concerned solely to set out certain associated distinctions,
or aspects of one distinction, which have historically been made or recognized
by philosophers. (K.142)
(β) The obscurity of the words baffled the audience; and they were not much
helped by Mrs Middleton's demonstration of a bird in flight. (M5.87)
(β) The sucker is bounded at the edges by a series of lips, ... (J1.88)
(β/γ) Loss of inhibitory power by avenacin on enzymic hydrolysis was confirmed
by growth experiments ... (J5.325)
(γ) 'You wish that Packford had been murderedΤ (M6.60)
(γ) Four decisive ideas in the Nathan Report have been introduced. (G-8)
(γ) 'Her parents were killed in the Cafi de Paris raid,' said John. (M5.79)

(γ) In situations in which the functional characteristics are grasped, the solution
can be transferred and applied in a variety of situations. (H.50)
(γ) ... if the academical world insisted on its narrow limits, then other means
of disseminating the truth must be found. (M5.17)
Passive transitive clauses include not only clauses with referential subjects but also
those with proleptic it anticipating a finite or nonfinite clause, which operates as
complement in active clause structure. Hence, proleptic it as passive clause subject
disappears in transformation to active:
-*• numerous workers have demonstrated that ...
(a) It has been demonstrated by numerous workers that the shape of the pW
curve of myosin adenosine triphosphatase at 25° C. is not altered from that
shown in Fig. 1 at higher ionic strengths, ... (J5.296)
(γ) Yesterday it was learnt that a rocket had, for the first time, been destroyed
in mid-flight by another rocket. (G.13)
(γ) Biologically it may be found that there are few differences between an African
negro and a white Scandinavian. (1-23)
We also include proleptic //-clauses which are followed by direct speech, and those
which are 'parenthetic' (see Appendix lb, pp. 183ff.).
(γ) It might be asked at this point: 'Why doesn't an examination of logic help us
to discover the methods we use when we think out problems?' (H.47)
(γ) The first [point] was how the greatly increased number of Africans which it
was envisaged would be admitted to the new legislature should be elected.
Clauses with it anticipating a nonfinite verb clause are rare (5 γ) as compared with
those anticipating finite verb clauses (4 α and 116 γ). See Appendix Ic, p. 185.

(γ) 'it may be decided not to rely exclusively on fixed-site missiles.' (G.17)

PASSIVE SEMIDITRANSITIVE clauses (ext-A) have prepositional phrases as adjuncts

where the prepositions collocate closely with the verb. The collocations are listed in
Appendix Ha, pp. 185 ff. There are 16 α-, 7 β- and 226 γ-passives in the corpus.

(a) '... Packford ... was then robbed of it by someone who killed him in the
process?' (M6.43)
(a) It [this word] was also applied by Shakespeare to human beings ... (1.24)
(β) Adults, however, were found mostly on the upper surfaces of stones, shells
and glass, to which they are confined by their need for a smooth surface.
(γ) Bureaucratic clerks in all their hideous, inhuman behaviour were charged
with the deed·, (M5.12)
(γ) It is surprising that more has not been heard of him, although shortly after
the war he was almost in the Walker Cup side. (F.28)
In this clause type, there are only two occurrences of proleptic subject (it) anticipating
a finite verb clause (see Appendix lib, p. 188).
(γ) ... it hadn't in fact been mentioned to him that she was a domestic employee
of the genteel variety. (M6.86)
PASSIVE DITRANSITIVE clauses (ext-C) constitute the least frequent type in the material
(8 a, 66 γ). There are three exponent classes for C in the corpus: nominal, nominal
+ preposition, and finite clause. Nominal as complement (see Appendix ilia,
pp. 188 f.):
(α) He was given this puppy by a farmer in the Welsh hills. (Ml.18)
(γ) We aren't allowed a tragedy nowadays without a banana-skin to slip on and
make it funny. (L2.42)
(γ) By using removable blocks different problems can be set the animal. (H.32)
Nominal + preposition as complement (see Appendix IHb, p. 189):
(a) But it [the house] had been made a mess of, comparatively late in its history,
by some owner with a taste for the Gothic. (M6.69)
(γ) ... she might judge she was being made fun of. (M6.61)
Finite verb clause as complement, including the parenthetic type (see Appendix IIIc,
p. 189).
(α) Ί am assured by Packford's physician that our friend had no rational
occasion to fear for his health.' (M6.35)
(γ) Ί was told you wanted to see me,' Gerald said. (M5.32)
(γ) 'Suicide is a crime, I have been told.' (M6.85)
In common with semiditransitive clauses, PASSIVE SEMIFACTITIVE clauses (int-A) have
close collocation of verb and preposition, but whereas the former display a variety
of prepositional exponents, semifactitive verbs collocate almost exclusively with as.
There are 135 occurrences and three kinds of prepositional adjuncts: preposition +
nominal, preposition + present participle, and preposition + adjective, or past
preposition + nominal as close adjunct (see Appendix IVa, pp. 189f.):
(α) Before the learned journals could lumber into reasoned appraisal, the whole
thing had been accepted as gospel by the common reader and become estab-
lished as a plain fact of literary history. (M6.17)
(γ) 'You'// be classed as a second-rater.' (M5.38)
(γ) ... he could see no reason why Radford's time over the longer distance
should not be accepted as a world's record. (F.30)

Preposition + adjective or past participle as close adjunct (see Appendix IVb,

p. 191):
(a) they [those who were killed in the First World War] are < remembered#
. /with grief . as young# . and /gäy# and /beautiful# . by /those who
survived -them#> (B2.16; <narrow and rhythmic»
(γ) It is, for example, possible to argue that in these regions Communist policy
should no longer be treated as united or consistent. (G.10)
Preposition + present participle as close adjunct (see Appendix IVc, p. 191):
(a) Haemopoietic tissue occurs in the intestinal wall of the larva and this has
been regarded by some as representing the spleen. (J 1.93)
(β) ... and Barber was deceived by one of Goddard's faster balls into attempting
a too ambitious cut. (F-25)
(γ) [the three people] /have been ar!rested# . and /chärged# with /[trying to
Nsmüggle] a : [Czech Nnätional] out of the ci>untry# (D.9)
(γ) Every few months the gap at Geneva is reported as being narrowed. (G.10)
PASSIVE FACTITIVE clauses (int-C) have C exponence similar to A in Type int-A apart
from absence of preposition: nominal, present participle, adjective, past participle,
etc. An additional, and highly frequent, exponent class is the fo-infinitive. In all,
there are 17 a, 7 β, and 244 γ.
Nominal as complement (see Appendix Va, p. 191):
(γ) He used to be called'Crossbones'. (L1.34)
(γ) The 'return' flight has been considered a necessity without which a species
would lose its breeding sites. (J5.349)
Adjective or past participle as complement (see Appendix Vb, p. 192f.):
(α) ... only a limited amount of money will be made available by the British
Transport Commission (or the Government) for pay increases, ... (G.8)
(β/γ) It [this cheese] is kept fresh ... by the special wrapping (C.96)
(γ) The animal is placed hungry at R with the direct and previously learned
pathway R-X blocked at Β. (H.42)
To-infinitive as complement (see Appendix Vc, pp. 192f.):
(α) ... and the primary unimolecular dissociation which it represents was shown
by a number of workers, by the use of free radical 'traps' to inhibit any
subsequent reactions, to give rise to the formation of two benzoyloxy radi-
cals ... (J4.36)
(γ) You're a famous author and all famous authors are expected to give lectures
as a matter of course. (LI .5)
(γ) ... Latin grammar was thought to embody universally valid canons of logic.
Present participle as complement (see Appendix Vd, p. 193):
(γ) There was still a mass of stuff to delight any authentic student of bibliopegy
who should be set browsing in the place; (M6.83)
(γ) ... the low-temperature product of rhesus rhodopsin can be seen regenerating
following irradiation with λ500 ηιμ light. (J5.293-4)

7.22 Quasi-agentive passives

Quasi-agentive passives (Class δ) are intermediate between agentive and nonagentive

passives and have mixed verbal and adjectival properties. Potential active transforma-
tion demonstrates their verbal nature, and among the adjectival features are actual
or potential qualifiers and lexically marked auxiliaries (see Section 6.45, pp. 135ff.).
Quasi-agentive passives, which are further characterised by animate subjects, have two
subclasses: δ 1 ('attitudinal passives') and δ 2 ('emotive passives'). The first of these is
the more closely related to agentive passives in often having the possibility of extension
with 6y-agent. The second class may have quasi-agents, which differ from ordinary
by-agents in that they are, typically, inanimate and embrace not only phrases with
prepositions other than by but also ίΑαί-clauses and to-infinitives (see Section 5.81,
p. 104).
Compare, for example, the following series:
(δ1) He was (felt) entitled to do this.
[his position] entitled him to do this
•to do this entitled him
*it entitled him to do this ) ^ ^ ^
*it made him /feel/ entitled to do thisj
(δ2) He was (felt) embarrassed to do this.

*[his position] embarrassed him to do this

to do this embarrassed him
it embarrassed him to do this
it made him /feel/ embarrassed to do this

Attitudinal passives, totalling 56 occurrences in the material, occur in a limited set

of clause types: in semiditransitive clauses with to + nominal, in factitive clauses
with to + infinitive, and in semifactitive clauses with to + present participle.

The Africans were heavily committed to the principle of'one man, one vote'. (G.12)
'For one thing, you haven't been obliged to earn your living ...' (Ml.20)
In plural societies where we can still mould the future we are deeply committed
to promoting the experiment of partnership. (G.4)

Emotive passives constitute a homogeneous class, and it is interesting that its defining
syntactic characteristics are matched by semantic similarity (as between for example
flummoxed, non-plussed, perplexed, staggered, and startled) or even 'macrosemantic'
similarity (as between flummoxed, etc. and interested, delighted, dissatisfied, obsessed,
overwhelmed, relieved, worried, and so on). The verbs of this class have the property
of being 'connected with a specific human "reaction"' (Nowell-Smith 1954, 85; cf.
Chomsky 1965a, 50f.). Many of its verbs may of course appear in other passive
classes, but then with different syntactic and semantic attributes. Compare the
following two sentences:
(γ) ... the sting was taken out of Wakefield by an injury to Fox after only thirteen
minutes' play which left him hobbling on the left wing for the rest of the
game. Rollin, too, was hurt, and these misfortunes certainly greatly reduced
Wakefield's striking power. (F.23)
(δ2) Hali would have been bitterly hurt if I had suddenly abandoned him and gone
home with Chris. (LI.87)
Unlike the second clause with hurt, the first will not, for example, admit of a lexically
marked auxiliary and a qualifier: * Rollin felt very hurt ... (but Hali would have felt
very hurt ...); nor of transformation to an intensive active clause: *It made Rollin
hurt ... (but It would have made Hali bitterly hurt ...). Semantically, too, the two
verb uses are different: in the first clause, hurt refers to physical injury; in the second,
to emotional injury. Other verbs, which are similar in having both these meanings,
and corresponding grammatical correlates, are depress, exhaust, upset, wound, etc.
There are 227 occurrences of emotive passives in the material. In addition to the
discussion in Sections 5.81 and 6.45 (pp. 135ff.), the following examples will adequately
illustrate this class.
... it was the bowler rather than the batsman who was surprised. (F.27)
... at the moment I am more than disconcerted·, I am profoundly worried. (LI. 19)
... only on occasions were they [the umpires] not entirely satisfied with the
absolute fairness of a delivery. (F.30)
She was delighted with the phrase - it would serve so well to interpret the academic
world to her literary friends. (M5.34)

7.23 Nonagentive passives (Class ε)

With quite a large proportion (18 per cent, 666 occurrences) of the passives in our
material, agent extension is unlikely or impossible (see Section 6.46, pp. 137 f.). As we
might expect, the distinction between nonagentive and other passives is far from clear-
cut, and a more refined analysis than ours would arrive at further subclasses at this end
of the passive scale. Most of the marginal cases have be as auxiliary with mutative
value; when the auxiliary is marked, the possibility of agent extension appears to be
even more remote than when it is unmarked.
The only serious contender to be as agentive passive auxiliary is get. But it is rare:
in the entire corpus there are not even half a dozen agentive gei-passives, none of
which is agentful. Although it is probably symptomatic that the occurrences below
are either from speech or written dialogue, there is no indication in our material that
the ge/-passive is common in colloquial English. However, as we shall see in Section
7.4, the passive as a whole is by no means infrequent in the spoken material.
'But Cintio's yarn may [orig. ital.] have got translated into English, you know,
without the translation's having survived.' (M6.20)
this is a free country and criticisms should be made and they get treated with the
contempt that they deserve (B5.6.8)
'Is it that place that got started in the thirties - I used to see it - called The
Socialist Bookshop or something like that?' (Ml. 19)
The agentiveness becomes more doubtful in:
People ... splash about with washleather and water, exhaust themselves and get
thoroughly soaked. (C.542)
'Think of the episode of supreme savage comedy when Hamlet jumps into her
grave and gets jammed in it.' (M6.16)
We know very well how some thinking sticks to the point and moves steadily
towards its conclusion while other thinking runs round in circles or drifts off
into blind alleys or gets bogged down\ (H.ll-2)
Become usually has the specific aspectual function of indicating gradual change,
which is often enhanced by modification with more and more, increasingly, etc. and
suffixation with -ize (conventionalize, industrialize, mechanize, organize, etc.).

Taking the argument still farther, it is becoming increasingly widely recognized

that African numerical superiority in Kenya will have ultimately to be reflected
in the constitution. (G-l)
... people who now speak one and the same birth tongue may, in fact, become
so highly specialized and departmentalized in all their words and ways that they
will no longer be able to understand one another ... (1.9-10)
In many simple problems these four phases may become telescoped in a single,
swift coordinated phase of activity - but in complex situations all phases may be
distinctly observed. (H.54)

The majority of the lexically marked auxiliaries in Class ε are mutative {get and
become), and there is only a scatter of nonmutative auxiliaries (remain, lie, look, etc.).

The post lay scattered on the table where Yves had searched it eagerly for possible
evidence of cheques. (M5.73)
... long before half-time Wakefield's forwards, still tired perhaps after their
efforts against Hull, looked well beaten. (F.23)

The use of verbal group types in agentive passives

Verbal group type Class α Class β Class γ Total

d 38 115 588 741

(14%) (43%) (27%) (28%)
D 117 61 536 714
(45%) (23%) (25%) (26%)
ad/AD 38 58 700 796
(14%) (22%) (33%) (30%)
bd 56 22 225 303
(21%) (8%) (10%) (11%)
BD 9 5 72 86
(3%) (2%) (3%) (3%)
cd 4 2 17 23
(2%) (1%) (1%) (1%)
CD 1 0 8 9
(0%) (·) (0%) (0%)
abd/ABD 3 0 14 17
(1%) (.) (1%) (1%)
other 0 2 5 7
(subjunctive, imperative) (.) (1%) (0%) (0%)

Total 266 265 2165 2696

(100%) (100%) (100%) (100%)

Compare this last example with beat as a γ-passive, where be does not commute with

With the exception of the men's and women's doubles all the top seeds were
beaten in the finals of the St. Annes lawn tennis tournament on Saturday. (F.23)

Just as get and become have the functions of indicating incipient action and gradual
change, so these nonmutative auxiliaries mark the aspect of state. Be can also be used
in 'statal passives' (see Section 5.51, pp. 92f.), but since it is aspectually unmarked,
auxiliaries like lie, look, and remain may serve to indicate the aspect unambiguously.
Nonagentive passives will be further discussed in the next chapter.


The verbal group structure of all the agentive clauses in the Major Passive Corpus
was analysed in terms of the types set up in Section 2.2 (pp. 10 f.). Conflating present
and past forms (denoted by lower and upper case letters respectively), we find that over
half the passive verbal groups are Type djD, and that the next commonest are Types
ad/AD and bd/BD; there is only a scatter of other types. The frequencies of the major
types were as follows (for a detailed breakdown into agentive classes, see Table 7:2):

d/D (is/was eaten) 1455 (54%)

ad/AD (may/might be eaten) 796 (30%)
bd/BD (has/had been eaten) 389 (14%)
cd/CD (is/was being eaten) 32 (>1%)
abdjABD (may/might have been eaten) 17 «1%)
acd/ACD (may/might be being eaten) 0 ( · )
bed) BCD (hasIhad been being eaten) 0 ( · )
abcdjABCD (may/might have been being eaten) 0 ( · )
This overall distribution is very similar to the one for the Minor Passive Corpus
discussed in Section 4.2 (pp. 43 if.). Since even in our present much larger material,
there are no instances of the last six types, it is evident that they have a very low
probability of occurrence. Our observation of a negative correlation in the Voice
Corpus between Type c/C and the passive is strengthened by the rarity of this type in
the present material (just over 1 per cent). We also noted a positive correlation
between Types b, a, A and the passive: again, the high frequency of bd, ad, and AD
in the present results would seem to support this observation. For the last two, the
associative tendency is particularly striking in agentless clauses, and for Type bd in
The a/A-type auxiliaries are predominantly 'closed-class' (here defined as not taking
a preceding auxiliary; see Section 2.33, pp. 14ff.). The frequencies for individual
auxiliaries in Type ad/AD are as follows:


can 177 have to 52

may 102 appear to 7
could 84 be (uri)likely to 6
will 79 be going to 5
should 69 have got to 4
must 54 be bound to 2
would 48 seem to 2
be to 45 need to 2
might 27 be able to 2
ought to 14 come to 2
shall 4 tend to 1
'd 3
7/ 2 85
used to 2
need 1


There are about a dozen combinatory a/A-types, where open class auxiliaries have
substructured verbal group combinations (see Section 2.33), as for example:
(A'+A" +D) He could remember only that he was sixty-four, could wonder only
whether his growing lust was a simple case of enlarged prostate
that would 'have to be dealt with'. (M5.13)
(A'+B+A"+D) To have detected the larger drops recorded by the radar in their
observed concentrations the sampling volume would have had to be
increased by a factor of 102-104. (J5.272)

Summing up, we may emphasize that the use of the passive verbal group is very much
restricted to a fairly small number of simple structures: Types d/D, ad/AD, and bd/BD
account for 98 per cent of all agentive occurrences in the Major Passive Corpus.


The use of the passive in the texts (listed in Section 1.4, pp. 6ff.) varies considerably,
in respect of classes as well as overall frequencies. Taking the number of passive
clauses per thousand words as the unit, the frequency ranges from 32 in one scientific
text (J4) to 3 in the sample from television advertising (Text C), with an average of
11.3 for the whole corpus. At the same time, the frequencies for most of the texts of
a text group are remarkably similar (see Table 7:3), and we can therefore achieve
greater generality and a broader basis for subsequent statements by making confla-
tions, not only of the texts of one group (such as El and E2) but also of related text
groups (such as I and J). Instead of 28 individual texts, this will give us, for example,
the eight text sets in Table 7:4. They are arranged in order from most to least frequent
in their use of the passive voice, measured in terms of occurrences per thousand words.
The samples taken from scientific exposition have a clear lead in frequent use of
passive constructions, with 23.8 and 21.5 clauses per thousand words for the learned
(Text J) and popular (Text H) varieties, respectively.
Next follow the mass media in their covering of topical events in press and radio
news (Texts Ε and D) and in editorials (Text G) with 16.4,14.2, and 15.8, respectively.
The two samples from prose dealing with arts subjects show rather different uses: the
more popular introduction to linguistics (Text I) has 13.8 as against 9.4 for the learned
philosophical discussion (K). Since the latter is a much Shorter text, their joint average
remains as high as 12.7 passives per thousand words.
All the samples discussed so far belong to the broad category 'informative prose'.
Next in our list follows a rather more disparate collection of material: speech, sports
reports, and novels. With one exception (Text B3), the spoken material is surprisingly
homogeneous in its use of passives, and the surreptitious recording (A) does not differ
from unscripted discussions (B) in this respect. Their combined average frequency is
9.2, which is slightly higher than that for press sports reports (Text F).

The use of agentive passives in 28 texts



A 5,000 40 8.0

B1 5,000 40 8.0
B2 5,000 44 8.8
B3 5,000 80 16.0
B4 5,000 45 9.0
B5 15,000 117 7.8

C 28,000 83 3.0

D 5,000 71 14.2

El 5,000 93 18.6
E2 5,000 71 14.2

F 30,000 269 9.0

G 30,000 474 15.8

Η 15,000 322 21.5

I 15,000 207 13.8

J1 5,000 116 23.2

J2 5,000 140 28.0
J3 5,000 79 15.8
J4 5,000 160 32.0
J5 15,000 337 22.5

Κ 5,000 47 9.4

LI 15,000 82 5.5
L2 15,000 76 5.1

Ml 5,000 40 8.0
M2 5,000 43 8.6
M3 5,000 55 11.0
M4 5,000 26 5.2
M5 30,000 231 7.7
M6 30,000 257 8.6

28 texts 323,000 3645 11.3

In view of their size and number, the samples from novels (M) should provide the
most reliable basis of all for a frequency study. The figures for most of the individual
texts are close to the average of 8.2 passives per thousand words. Novels and plays,
which together make up the category 'imaginative prose' in our corpus, have similar
passive frequencies. It is interesting to note that 'imitated speech' as represented in
two comedies (Texts El and E2) has far fewer passive occurrencies than our recorded

Class α-passives

[ Class ß-passives

Class γ-passives

Fig. 7:2. Frequencies of the agentive passive classes in eight text sets.

The use of passives in eight text sets



'Science' (Texts H, J) 50,000 967 (19.3) 9 (0.2) 178 (3.6) 1154 (23.1)
'News' (Texts D, E, G) 45,000 611 (13.6) 21 (0.5) 77(1.7) 709 (15.8)
'Arts' (Texts I, K) 20,000 200 (10.0) 6 (0.3) 48 (2.4) 254 (12.7)
'Speech' (Texts A, B) 40,000 261 (6.5) 25 (0.6) 80 (2.0) 366 (9.2)
'Sports' (Text F) 30,000 211 (7.0) 13 (0.4) 45 (1.5) 269 (9.0)
'Novels' (Texts M) 80,000 358 (4.5) 141 (1.8) 153 (1.9) 652 (8.2)
'Plays' (Texts L) 30,000 37 (1.2) 64 (2.1) 57 (1.9) 158 (5.3)
'Advertising' (Text C) 28,000 51 (1.8) 4 (0.1) 28 (1.0) 83 (3.0)

Eight text sets 323,000 2696 (8.3) 283 (0.9) 666 (2.1) 3645 (11.3)

Frequencies in terms of occurrences per thousand words are given in brackets.

'actual speech'. The smallest passive voice consumer is television advertising with only
3 passives per thousand words.
The frequencies in the eight text sets and the proportions within each set of the three
major passive classes (agentives, quasi-agentives, and nonagentives) are shown in
Table 7:4. It shows that the absolute occurrence of nonagentives is rather similar
in all the sets, and hence proportionately greater for the text sets with low overall
passive frequency. There is a notably larger proportion of quasi-agents in the samples
from imaginative prose than in any other set.
Figure 7:2 shows the distribution in the eight text sets of agentive clauses only, and a
slight consequential reordering of the sets: sports reports and speech on the one hand,
and advertising and plays, on the other, have changed places. The graph suggests that
the proportions of Classes α, β, and γ remain largely constant throughout the sets.
There is, however, a notable preponderance of Class β over α in science, and of Class
α over β in sports and speech.
Summing up, the results of this inquiry into the use of the passive voice emphasize
the centrality of the agentive classes in the majority of the samples, and particularly
of the agentless class. Furthermore, the results indicate, firstly, that there is consider-
able variation in overall passive frequency in the individual texts, the use in the most
frequent being over ten times that in the least frequent; secondly, that in their use of
the passive there is notable consistency among the several texts of a particular cate-
gory; thirdly, that the major stylistic determining factor in the frequency of its use
seems to lie in a distinction such as that between informative and imaginative prose,
rather than in a difference of subject matter or between the spoken and written lan-
guage. However, within the category of informative prose, passives are most com-
monly found in scientific exposition. Speech occupies an intermediate position be-
tween the most extreme passive users.


In this study of the English verb we have discussed, from various standpoints, some
problems connected with voice: in Chapters 2 and 3, the typology of the finite verbal
group and clause; in Chapter 4, the use of the voice terms and their cooccurrence with
other grammatical categories in the Voice Corpus and the Minor Passive Corpus; in
Chapters 5 and 6, a more delicate analysis of a small but more representative number
of passive clauses (the Taxonomic Passive Corpus) in terms of a large number of
criteria and an attempt at a classification of passive clauses using numerical taxonomy;
in Chapter 7, a quantitative study with the classification system applied to all the
passive clauses in the Major Passive Corpus, consisting of 28 sample texts and totalling
almost a third of a million words. We have now reached a point in our inquiry when
it should be possible to draw some general conclusions.


The concept of'the passive scale' (see Section 6.4, pp. 132ff.) is particularly important
for the direct bearing it has on the relation of the two voice terms. This scale consists
of a number of passive clause classes that have different affinities with each other and
with actives.
At one end of the scale are the agentful classes, α (with animate agents) and β (with
inanimate agents). Clauses belonging to these two classes have close systemic trans-
formational relation with the active voice, that is to say, possess all the necessary prop-
erties in the clause elements for the application of a transformational rule to convert
them into their corresponding active clauses. The word 'systemic' is, however, an
important qualification here, since it does not subsume transformational constraints
that are imposed by, for example, clause sequence (see Criterion CL 16, p. 85),
adjunct placement, and in general the requirements of balance between clause ele-
ments, case alteration in pronouns with oblique forms (as in he : him), and auxiliary
change between shall and will determined by the person of the subject.
The question of balance is not simply a stylistic point but can be a most important
consideration (see, for example, Criterion CL 18e, p. 87). Although 1:40 in this
example is an extraordinary proportion for the number of words in the subject and
in the agent, it is indicative of many significant differences between these two nominal
elements in passive clauses. In the material that was analysed for element length,
agents had about twice as many words as subjects. We found that the average for the
subject was 3.3 words in the Taxonomic Corpus and 5 words in the Minor Passive
Corpus, but for the agent 8 and 8.5 words respectively (see Sections 4.36, pp. 52ff.;
and 6.3). Unlike subjects, agents are frequently coordinate, but are rarely expounded
by personal pronouns (see Sections 4.34 and 6.32, p. 129). Hence, we may con-
clude that one of the motivating factors in selecting the passive in favour of the active
is the preference for placing heavy nominal groups at the end of sentences.
Differences between the two agentful classes include, beside agent gender, possibility
of deletion and frequency of coordination (which are greater for animate than inani-
mate agents); frequency of modification (which is greater for inanimate agents);
finitude and noun class (animate agents, more often than inanimate agents, are
definite - as opposed to indefinite - and mass nouns - as opposed to countable nouns;
see pp. 125 f.). The positioning of Classes α and β on the scale has been determined by
the fact that there is often structural indeterminacy between β and γ, since preposi-
tional phrases may be interpreted as either agents or other adjuncts (see Section 6.42,
p. 133). Such phrases, which have been called 'Janus-agents', occur in Class β/γ,
This class is quite large (see Section 7.21, pp. 141f.), and constitutes a bridge between
the agentful and agentless classes.
There are also other indications that there is not an abrupt separation between the
classes: phrases with prepositions other than by may function as agents, or near-
agents, as in
And now all tendency to laughter was submerged in parental admiration as the
little girls advanced singing "Es ist ein' Ros' entsprungen" [ital]. (M5.85)
Furthermore, similar functions may be carried not only by prepositional phrases but
also by adverbs, as in

The immigrant problem is widely admitted to be the most important election issue.

Adverbs have, of course, no nominal head which can operate as active subject, but
their agent-like nature is made plain by the tautologous effect of at least certain fry-
phrases in addition to the adverb, as in

The immigrant problem is [*widely] admitted by everyone to be ...

and by the inadmissibility of the adverb in the active transform:

-+· everyone admits [""widely] that the immigrant problem ...

In some cases, whole clauses may have agent-like function (this is a regular feature of
emotive passives):

Where there was a second sale at a higher price, and the agency decided the
increase was caused because planning permission had been granted, they would
collect the 75 per cent tax. (El.16)
The 'agent', which is here rather the cause (cf. Section 5.81, pp. 103f.), is felt to be the
following clause. It is perhaps worth noting the similarity between the because-
clause and an inanimate factive by- agent:
... the granting of planning permission ] . ,
, ° , , · . . . . . . ι caused the increase
... the fact that planning permission had been granted j

In our descent of the scale, we are now already in Class γ, whose key feature is absence
of, but possible extension with, 6j-agent. Judging by the frequency of occurrence in
all text samples, this class is central to the passive construction. In the corpus, it
covered some 80 per cent of all agentive passives (i.e. Classes α, β, γ). The proportions
of the agentful classes varied somewhat in the texts, but Classes α and β (including
β/γ) had an average of about 10 per cent each (see Section 7.21, p. 141).
There are also features cooccurring with 'absent agent' that further isolate Class γ
from α and β. Most interesting is perhaps the distribution for external clause-relation
commented on in Section 4.7 (pp. 65 fF.), where we found a remarkable difference
between Class γ, on the one hand, and Classes α and β as well as active clauses, on the
other, in that the ratio of syntactically bound to free clauses was very much higher
in Class γ than the rest (see Figure 4:11, p. 68). In the output of the Taxonomic
Experiment, Group IV (which together with Group III roughly corresponds to Class
γ) was at variance with all the other groups in that its clauses had no adjuncts and
were almost invariably bound (pp. 127f.). These two sets of results, which were
arrived at via different techniques and different corpora, further testify to the differ-
ences between the agentive classes on the passive scale; whereas agentful and active
clauses are similar in usually being free (as well as in having potential direct trans-
formation relation), proportionately many more agentless agentive clauses are bound
in their external clause-relation. Since other characteristics of Group IV include high
frequency of pronoun subjects and a large proportion of coreference between the
passive subject and the object or predicative of a preceding contiguous clause, we may
assume that an important function of many Class γ-passives is that of postnominal
modifier, clausal adjunct, etc. (cf. p. 128):

He tempted Cartwright with two half volleys which were royally punished... (F. 28)
... Lewis Packford himself had contrived a situation which could be exhibited as
entirely ludicrous. (M6.84)

Furthermore, we noted for the Voice Corpus that the use of tense forms in the verbal
group was not parallel in the active and passive (see Section 4.2, pp. 43 ff.): the passive
Types ad, AD, and bd (see Section 2.2, pp lOf.) were more common than their active
counterparts (a, A, b), whereas other tense forms were more frequent in the active; this
was particularly true for Type c/C as against cdjCD and for the more complex verbal
group structures (ac/AC and bcjBC as against acd/ACD and bedjBCD). However, the
high ratio of Type ad/AD in the passive applied to Class γ but not to Classes α and β;
Type bd, on the other hand, was particularly frequent in the agentful classes.
Class δ consists of two subclasses, δ 1 ('attitudinal passives') and δ2 ('emotive pas-
sives'), and a chief characteristic of the latter (its range of possible agents) has given
the whole class the name of 'quasi-agentive passives'. The two subclasses demonstrate
various points of similarity and dissimilarity (see Sections 7.22, pp. 147f.; and 8.2
below). Emotive passives are unique in our classification in having a striking macro-
semantic correlate to their defining syntactic attributes.
Class ε is rather more heterogeneous than δ, and is interesting because of its low
place on the passive scale and its complex voice relations. These last two classes
should be dealt with more fully using a different set of criteria and together with the
related intransitive and equative clause types. Nevertheless, it will be of interest to
indicate here their place in the overall description of voice relations in English.


As we proceed down the passive scale, it becomes increasingly difficult to consider the
passive in terms of a transformational voice relation, since this is being gradually
replaced by a different relation which we call 'serial'. The theory of serial relationship
states that sentences can be produced paradigmatically: "There is surely every reason
to suppose that the production of sentences proceeds by a complex interplay involving
'transformation' and what Miller calls 'a sentence-frame manufacturing device'."1
The concept of serial relation will help us to account for many voice features which
cannot easily be explained otherwise. Bolinger, whose idea of 'syntactic blending' is
similar to serial relation, draws attention to the 'stepwise relationship to the passive' of
adjectives like prone and loth in a passive series such as the following:2

I am forced (led, induced, etc.) to agree with you.

(Active transformations normal.)
I am inclined (disposed, disinclined) to agree with you.
(Active transformations less usual.)
I am minded to agree with you.
(No active transformation.)
The illustrated three types correspond to our γ-, δ-, and ε-passives. We may add that
for Class δ, there is a greater likelihood of inanimate than animate agent and a possi-
bility of transformation to active intensive clause type:
This makes me inclined to agree with you.
*This makes me forced to agree with you.
Quirk 1965, 213.
Bolinger 1961b, 378.

Since verbs in the quasi-agentive class can also occur in other classes, passive construc-
tions with such verbs may be potentially ambiguous in minimal structures (although
in the corpus they were hardly ever found to be so, and this is unlikely to be the case
in actual context). For example, the sentence
He was interested in linguistics.
may be interpreted either as a γ-passive (serially related to Class a) or as a 52-passive
(serially related to active equative clauses):
(a) He was (very much) interested in linguistics by his professor.
(b) He was (very much) interested in linguistics.
(c) He was (very) interested in linguistics.
(d) He was (very) keen on linguistics.
The corresponding active transforms are:
(a) His professor interested him in linguistics.
(b/c) Linguistics interested him.
Although informants find the second of these the right transform of the 62-passive, it is
worth noting that, unlike the passive, the active cannot take verb qualification with
Structures of the favoured passive factitive type W to F(see p. 145), have a range of
active transform structures; in some cases there are no active analogues at all, and in
many more cases, transforms are dubious. The following series may illustrate a
postulated serial relationship for the passive voice in the left-hand column:
allowed -* They allowed him to do it.
seen They saw him do it.
said -» They said that he did it.
He was to do it.
calculated They calculated that he would do it.
scheduled -* (?) They had scheduled him to do it.
reputed -* *They reputed him to do it.
The ambivalence of Class δ2 can be satisfactorily accounted for by the concept of
serial relation since verbs of this class have transformational as well as serial voice
His sisters behaviour annoyed him very much.
His sisters behaviour made him feel very annoyed and angry.
He was (felt) very annoyed (and angry) at his sister s behaviour.
There is an aspectual serial relation operating when a change of tense produces a
passive which permits transformation to active. The ε-passive
The house is already sold.
is indirectly related to the active

The estate agent has already sold the house.

via the γ- and α-passives:
The house has already been sold.
The house has already been sold by the estate agent.
Other examples of 'statal passives' (see pp. 92 f.) with indirect transformational
relation are the following:
Cavendish inquired with grave courtesy, 'Is the thesis finished, Doctor?
'Not quite, Captain.'
Ί heard you were coming with us to give a lecture on your findings to the New
York Canford Institute.' (M4.57)
'We must leave it [the rat]', repeated Madeleine.
'We can't. It's all mauled - don't you see? We cannot [ital.] It's got to be killed.'
In principle, statal passives do not have expressed agents, but serial relation will help
to explain occasional examples of statal passives with agents, as in
This memorial is placed here by his friends and neighbours in testimony of respect,
affection, and gratitude*
which has no active transform
*His friends and neighbours place this memorial here ...
but is serially related to
His friends and neighbours have placed this memorial here ...
and its passive transform
This memorial has been placed here by his friends and neighbours ...
as well as to
This memorial is placed here in testimony of...
Lexical marking in the auxiliaries of nonagentive passives may be taken to denote one
further remove from the agentive class, as in the following series leading to an active
The door remains bolted and barred.
The door is bolted and barred.
The door has been bolted and barred.
The door has been bolted and barred by the shopkeeper.
The shopkeeper has bolted and barred the door.
» Erades 1958/9.

This view of statal passives enables us not only to dispense with a rigid, unrealistic, and
arbitrary dichotomy into 'passive' and 'nonpassive', but also to account satisfactorily
for those extremes of serial relation that may be called 'serial isolates', as in
The rat reared up full stretch in a supreme convulsion, as if about to spring. But
it was done for. It rolled over, twitching. Immediately its open eyes began to
film. (Ml.28)
Here, only the last part of the series remains, since in accepted usage we now have no
corresponding γ-, and α-passives or, indeed, active correlates:
*The rat had been done for {by the dog).
*The dog had done for the rat.
OED has an entry do for (s. v. do 38), both active and passive, which is labelled
'colloq.', but the last quotation for the active form is dated 1876; interestingly, Web-
ster (1961) acknowledges the serial isolate by making no reference to do for under do
but having a separate entry done for, which is labelled 'adj . A similar serial isolate is
the colloquial done in ('exhausted').
Since we no longer have to account for all passives in terms of transformation,
serial relationship also bridges the gap, in a natural way, between noncompounds and
compounds (Class ζ) and between verbal and nonverbal bases (see Sections 2.73-4,
pp. 21 ff.; and 6.46, pp. 137f.). Agentful compound-passives are then related serially
to agentful noncompound-passives instead of transformationally to actives, as in:
Everyone rejects me.
I am rejected by everyone.
Ί am unloved and rejected by everyone'... (E2.27)
But Cavill was unimpressed by this sally. (M6.54)
Generally speaking, transformational voice relation operates most strongly at the top
end of the scale, and serial relation most strongly at the bottom end, with varying
degrees of both relations in between these two extremes, as may be illustrated by the
set of sentences in Figure 8:1. Sentence (A) is the direct transform of α-, and the in-
direct transform of the γ-passives; (B) is the direct transform of β. The δ-passives
have two active analogues each, one in transformational and one in serial relationship.
The ε-passive is serially related, on the one hand, to γ-passives, and, on the other hand,
to both the active intransitive (G) and the equative (H) sentences. The particular
ε — intransitive relation, which involves type-change in the verbal group, has been
referred to as 'transmutation' (see Section 5.45, pp. 88 ff.).


We must now conclude by attempting to answer some of the questions raised in

Section 1.1, in particular whether the passive in English should be derived from actives

This law has been applied PASSIVB active The president has applied
to all murderers by the ^A^ this law to all murderers
president. (by/in an edict).

This law has been applied

to all murderers by an An edict has applied this
edict. law to all murderers.

This law has been applied

to all murderers in an

This law has been applied

to all murderers.
[The president/An edict]
obliges everyone to obey
this law.
Everyone is obliged to
obey this law.
Everyone has to obey this

It pleases everyone to have

this law.
Everone is pleased to have
this law.
Everyone is glad to have
this law.

This law now (already)

This law is now (already) applies to all murderers.
applied to all murderers.
This law is now applicable
to all murderers.

Fig. 8:1. Diagrammatic representation of some voice relationships.

<><><> = transformational relation
= serial relation

by transformation or whether it should be treated as a clause type in its own right.

There exists, in certain respects, a clear connection between the active and passive
which can be stated by transformational rules. At the same time, when native
speakers are faced with a choice, in actual sentences, they almost invariably prefer one
of the voice terms to the other. The preferred one is usually the active. The following
extreme result, obtained as a by-product in an experiment designed to measure lin-
guistic acceptability, may serve as an illustration of both these points. Asked to
estimate its deviance, 73 informants rejected, two queried, and one accepted the
following sentence:
A nice little car is had by me.
On the other hand, 74 informants accepted, one queried, and one rejected an active
I have a black Bentley.
Before giving a direct judgment of the acceptability of the sentences, the informants
were asked to perform a simple grammatical operation on each of them, in this case a
change from positive to negative. Whereas a single informant failed in the case of the
active sentence, only 36 managed to produce the successful operation on the passive
sentence (which, as it happened, was one of the lowest results for any of the 50 sen-
tences in the test battery). From our point of view, it is interesting to note that as
many as 23 informants transformed the passive sentence into the active without being
asked to do so, whereas no one transformed another passive sentence in the battery,
Clothing was needed by the poor.
Such a telling demonstration could surely be possible only if there was a feeling of
great discomfort with the passive in this sentence, as well as a clear transformational
voice relationship, by means of which the informants could achieve the necessary
rectification of the deviant sentence.4
Once the existence of this basic voice relation has been stated, we must draw atten-
tion to the considerable drawbacks of a procedure by which one voice term is derived
from the other by means of transformation. The favoured direction has been from
active to passive (see Section 1.1). The advantage of this particular unidirectional
derivation (instead of one from passive to active) is largely supported by experience
in the present work. We may mention the following points:
(a) The low proportion of agentful passives (in fact only some 20 per cent of agent-
ive clauses and less than IS per cent of all passive occurrences in our material; see
pp. 141 ff.).
(b) The apparently general preference for the active and lower overall frequencies
for the passive. Even in a text (Jl) of the group 'scientific exposition' (which was
found to be the group with the highest occurrence of passives), the proportion of pas-
See Quirk & Svartvik 1966,74
sives to actives did not exceed 1:3, even discounting active equativefee-clauses(p. 46).
(c) The differences in the use of verbal group types, where the restrictions are largely
on the passive (see pp. 46ff.). Compare, for example, the following pair:
These Conservatives have not been winning seats lately. (El. 14)
Seats have not been being won by these Conservatives lately.
(d) The greater restrictions on lexical verbs in the passive than the active. Apart from
intransitive and equative verbs, examples include some uses of the verbs have, lack;
contain, hold; become, fit, suit; mean, resemble·, befall·, cost, last, take, walk, weigh.6
(e) The limited and as yet not clearly defined range of prepositional verbs in the
passive. These are verbs and prepositions that have sufficient cohesion to be able
to operate together as transitive verbs in the passive (see Sections 2.71-2, pp. 19fT.),
as in
But in this country there are singularly few murders which are conceived of as
deliberately incident to a robbery. (M6.38)
Prepositional verbs in the material also include account for, callfor, care for, call upon,
deal with, depend upon, dwell upon, fire on, go into, look at, preside over, provide for,
talk of, think of", the phrasal-prepositional verbs include get away with, and make off
with. Some of the collocations (like look at) are very close and can occur in the passive
with few or no restrictions, but others (like go into) will do so only under certain
conditions. Compare, for example, the following pair of sentences:
I think a lot of that [problem] can be gone into and gone into pretty . pretty
thoroughly - (B5.6.14)
*The room was gone into at once.
The active analogue of the latter is, however, fully acceptable:
He went into the room at once.
The difference in acceptability between the two passives must, it seems, be accounted
for by some concept like concrete/abstract subject or literal/metaphorical verb
(compare The town/conclusion was arrived at).
(J) The restrictions on the relations between clause elements, such as coreference
between subject and object. This can be manifested by reciprocal and reflexive
pronoun objects or by secondary concord (see Section 3.7, p. 33; and Olsson 1961,
111ff.)manifested by possessive pronoun:
' Cf. Gleason 1965, 307f. and Twaddell 1963, 9. A much mote acceptable passive of have than
A nice little car is had by me (discussed above) is the clichd A good time was had by all (cf. Hill 1958,
203). Although this sentence appears to contradict the general rule that have in the sense of 'own',
'enjoy' does not occur in the passive, it is, rather, the exception that proves the rule. In fact, it is
precisely because of its deviance (and institutionalization) that this sentence achieves its special effect
of facetiousness. Have ('obtain', 'get') has no similar restrictions in the passive: Taxis can be had
anywhere outside the station.

There is a further bonus for which most M.P.s are rubbing their hands - though
Blackpool hotels are upset. (E2.7)
The doctor shook his head, and applied himself once more to his notebook.
A general difficulty with using transformations of sentences in context is the imbalance
that may result. It may be caused by discrepancies in the lengths of clause elements,
such as personal pronouns and other short agent exponents occurring with long sub-
ject exponents. We have found that agents are generally heavier than subjects.
Pronouns are very frequent as subjects but extremely rare as agents. As many as one
third of the clauses in Corpus III and one half in Corpus II had coordinate agents,
whereas coordination was rare in the subject (see Sections 4.34-36, pp. 52 ff.; 6.32,
pp. 129 ff.).
Stylistic imbalance can further be attributed to disruption of the natural develop-
ment of the discourse, as this has been stated in the principle of functional sentence
perspective (FSP), which 'causes the sentence to open with thematic and close with
rhematic elements. Very roughly speaking, 'thematic elements are such as convey
facts known from the verbal or situational context, whereas rhematic elements are
such as convey new, unknown facts'.® The following sequence may serve as illustration
of how the passive may be obligatory both to avoid imbalance and to accord with the
principle of FSP:
'Thinking ... may be provisionally defined as what occurs in experience when an
organism, human or animal, meets, recognizes, and solves a problem.' This
provisional definition was also favoured by Dewey, who pointed out that so long
as our activity glides along smoothly, or as long as we allow our imagination to
entertain fancies, there is no reflection. (H.30)
There are, as we have seen, many characteristic features of the passive voice that
cannot easily be accounted for in terms of unidirectional active — passive trans-
formation. Rather, as we proceed down the passive scale, in the direction away from
the agentful classes, it becomes increasingly realistic, and economical, to consider the
production of passive sentences in terms of serial relationship with equative and
intransitive active clause types.

• Firbas 1964,112.




(see pp. MSf.)1

accompany accept
address adopt
advise announce
aid appreciate
appreciate arrange
beat attempt
butcher await
call call
carry carry on
catch catch
convince commission
corner confer
defeat confirm
dismiss confiscate
entertain construct
examine create
finance cut
follow decide
force decipher
govern define
join demand
knock demonstrate
meet describe
no-ball detect
nobble determine
pay develop

Verbs with animate subjects in the first column; verbs with inanimate subjects in the second

penalize devise
punish discuss
receive disturb
reject do
ride dominate
stump drive
support emphasize
surround establish
undo examine
wrap explain
lay down
make off with
pick up
preside over

put up
sort out
think of

Class β: affect abstract

attract accomodate
catch accompany
convert affect
corner arylate
cover assist
COW attack
defeat bias
deter blaze
embarass block
fascinate blur
help bound
hold bring about
impress carry
incense cause
interrupt circumscribe
motivate compensate
muddle complicate
overwhelm confirm
precondition connect
reduce consolidate
retain control
reward cover
save create
shock define
split demonstrate
startle destroy
surprise determine
take in dictate
trouble discount
uproot disguise

set off
stamp in

take over
take up

Class β/γ: accompany

break down


Class γ: accost abandon

amuse abolish
arrest absorb
assassinate accept
beat accomodate
believe accomplish
bite account for
blackmail achieve
bowl acquire
bring add
bring up adduce
bury adjourn
buy out administer
cage adopt
call up advance
capture affect
carry alert
cast out allow
catch alter
caution analyse
censure answer
chain applaud
cheer apply
conquer approach
create arrange
cut articulate
defeat ascertain
deplore ask
detain assess

detect assume
discredit attempt
dismiss avoid
dissect await
distribute backdate
drop balance
educate bear
elect bind
employ block
enter blunt
exclude bring
execute bring about
expose bring in
fear bring out
find broadcast
fire on build
flog bungle
foil bunker
garland bury
give out buy
hang calculate
have call for
hear cancel
help care for
hire carry
hit carry out
hunt cause
hurt cauterize
ignore cement
injure change
install characterize
invite chase
keep check
kill clamp down
knock clarify
know clear away
leave clear up
look after close
massacre coin
mention combine
mislead commit
mollify comprehend

move compress
murder compute
name concede
note condemn
oppose conduct
outplay confirm
outvote connect
ovariectomize consider
overrule construct
pay continue
persecute control
place cool
post correct
press count
provide for cover
punish create
pursue criticize
put curtail
put away cut
recall cut down
represent cut off
respect dangle
restrain deal with
sack debate
satisfy declassify
scrutinize deduce
see defer
select define
send delay
shoot demonstrate
show up depict
smuggle describe
stump design
summon destroy
take detect
take in determine
talk of develop
teach dig up
test diminish
thrash discard
threaten discipline
train discover

transfer discuss
underpay disentangle
usher disperse
warn display
withdraw distribute
worship divide
draw up
dwell upon

fence in
find out
get away with
go into

lay off

pick up
pin down


run up

set off
set up
sort out
take up

try out
turn down
use up
work out


(see p. 144)

a: demonstrate

Class γ: accept
point out



(see p. 144)

Class γ: agree


(see pp. 144f.)

Class a : convict (of) apply (to)

drive (to) associate (with)
rob (of) bequeath (to)
take (to) do (on)
draw up (for)
give (to)
mill (for)
put across (on)
subject (to)
submit (to)

Class β: bully (into) develop (into)

confine (to) impose (upon)
stamp out (into)
transform (into)
Class γ: admit (to) accept (for)
ask (about) adapt (to)
banish (from) add (to)
bring (into) address (to)

celebrate (for) apply (for)

charge (with) apply (from)
convict (of) apply (to)
convince (of) ascribe (to)
crowd (out of) associate (with)
deprive (of) attach (to)
discharge (from) attribute (to)
introduce (to) blame (on)
leave (with) blend (for)
let in (on) bring (into)
let out (of) bring (to)
mesmerize (into) build up (on)
pair (with) choose (for)
pay (for) combine (with)
punish (for) commit (against)
put (into) commit (to)
recompense (for) commute (for)
remind (of) compare (to)
rouse (to) compare (with)
sentence (to) compile (on)
sew (into) complete (into)
show (to) conduct (against)
strip (of) confuse (with)
supply (with) consign (to)
suspect (of) convert (from)
take (for) convey (to)
tip off (about) deduce (from)
treat (to) defer (to)
deny (to)
direct (at)
direct (to)
direct (towards)
discuss (with)
do (about)
do (on)
do (to)
draw (from)
efface (from)
entrust (to)
equip (with)
exempt (from)
expose (to)

extend (for)
extend (to)
extract (from)
focus (on)
give (of)
give (to)
gouge (out of)
grant (to)
hear (of)
impose (on)
induce (in)
infer (from)
introduce (into)
know (about)
know (from)
know (of)
leave (for)
leave (to)
liken (to)
make (about)
make (for)
make (into)
make (of)
make (on)
make (to)
motivate (towards)
notify (to)
offer (to)
pass (into)
pay (to)
permit (to)
place (in)
place (on)
postulate (for)
prefer (to)
present (to)
promote (to)
propose (for)
provide (for)
put (into)
put (on)
put (to)

raise (about)
refer (to)
relate (to)
reserve (for)
restrict (to)
reveal (to)
say (about)
say (of)
send (to)
separate (from)
set (against)
shield (from)
show (to)
snatch (from)
spend (in)
spend (on)
subject (to)
submit (to)
substitute (for)
take (into)
tell (about)
throw (on)
tie (to)
transform (into)
translate (into)
turn (into)
use (of)
voice (about)
vouchsafe (to)
write (to)


(see p. 145)

Classy: bear (upon)

mention (to)


(see p. 145)
Class a: give give

Class γ: allow give

ask permit
deny set


(see p. 145)

Class α: make (a mess of)

Class γ: make (fun of)


(see p. 145)

Class a: assure

Class γ: advise


(see p. 145)

Class a: accept (as)

describe (as)
regard (as)

take (as)
use (as)

Classy: accept (as) accept (as)

acknowledge (as) acclaim (as)
assess (as) base (as)
bring (as) bring (as)
class (as) calculate (as)
describe (as) categorize (as)
employ (as) characterize (as)
know (as) choose (as)
select (as) conceive (as)
try (as) conceive of (as)
consider (as)
define (as)
designate (as)
develop (as)
display (as)
do (as)
express (as)
give (as)
hear (as)
identify (as)
illustrate (as)
intend (as)
interpret (as)
know (as)
look at (as)
need (as)
predict (as)
read (as)
recognize (as)
regard (as)
see (as)
summarize (as)
take (as)
treat (as)
use (as)


(see p. 146)

Class a : remember (as) accept (as)

Class γ: describe (as) accept (as)
define (as)
describe (as)
exhibit (as)
regard (as)
take (as)
think of (as)
treat (as)


(see p. 146)

Class a : regard (as)

Class β: deceive (into)

Class γ: accuse (of) prevent (from)

charge (with) recognize (as)
describe (as) regard (as)
inhibit (from) report (as)
preclude (from) see (as)
prohibit (from) single out (as)
report (as) take (as)
restrain (from)


(see p. 146)

Class a : call

Class γ: appoint call

call consider
make make
name name


(see p. 146)

Class a: keep make

Class β: make keep

Class γ: consider account
keep assume
kill call
place consider


(see p. 146)

Class a: ask report

invite show
press think

Class β: bind

Class γ: advise admit

allow allow
appoint assume
ask believe
believe bring
call upon calculate
depend upon claim
expect consider
intend declare
invite design
know expect

leave find
make guarantee
mean hear
observe hold
prove intend
recommend know
report make
repute mean
require observe
say order
see regard
teach report
think say
use see


(see p. 147)

Class γ: leave see


Akerlund, A. (1914), "A Word on the Passive Definite Tenses", Englische Studien 47.321-37.
Allen, W. S. (1959), Living English Structure. London.
Bach, E. (1964), An Introduction to Transformational Grammars. New York/Chicago/San Francisco.
Bloomfield, L. (1933), Language. New York.
Bolinger, D. L. (1961a), Generality, Gradience, and the All-or-None. The Hague.
(1961b), "Syntactic Blends and Other Matters", Language 37.366-81.
Brose, B. (1939), Die englischen Passivkonstruktionen vom Typus "I am told a story" und "I am sent
for". Würzburg.
Buyssens, E. (1950), "La conception fonctionnelle des faits linguistiques", Grammaire et Psychologie
(= Journal de Psychologie 43.37-53).
Carvell, Η. Τ. & Svartvik, J. (1966), Linguistic Classification and Numerical Taxonomy. TheHague.
Chomsky, N. (1957), Syntactic Structures. The Hague.
(1962), "A Transformational Approach to Syntax", Third Texas Conference on Problems of
Linguistic Analysis in English in 1958 (Studies in American English). Austin, Texas.
(1964a), Current Issues in Linguistic Theory. The Hague.
(1964b), "Topics in the Theory of Generative Grammar" (to appear in Current Trends in
Linguistics 3, ed. by T. A. Sebeok).
Crystal D. & Quirk, R. (1964), Systems ofProsodic andParalinguistic Features in English. The Hague.
Curme, G. O. (1913), "The Proper Subject of a Passiv Verb", Modern Language Notes 28.97-101.
(1931), Syntax (=Curme, G. O. & Kurath, H., A Grammar of the English Language 3). Boston.
Erades, P. A. (1950), "His books don't sell", English Studies 31.155-7.
(1958/9), "Points of Modern English Syntax", English Studies 39.286, 40.63-4.
Finney, D. J. et al. (1963), Tables for Testing Significance in a 2 x 2 Contingency Table. Cambridge.
Firbas, J. (1959), "Thoughts on the Communicative Function of the Verb in English, German and
Czech", Brno Studies in English 1.39-63.
(1964), "From Comparative Word-order Studies", Brno Studies in English 4.111-26.
Firth, J. R. (1957), "A Synopsis of Linguistic Theory, 1930-1955", Studies in Linguistic Analysis
(Special Volume of the Philological Society). Oxford.
Francis, W. N. (1958), The Structure of American English. New York.
Frary, L. G. (1929), Studies in the Syntax of the Old English Passive with Special Reference to the Use of
'wesart and 'weordan". Language dissertations published by the Linguistic Society of America,
No. 5. Baltimore.
Fries, C. C. (1940), American English Grammar (English Monograph No. 10, National Council of
Teachers of English). New York.
(1959), The Structure of English. London.
Fröhlich, J. (1951), Der indefinite Agens im Altenglischen (= Swiss Studies in English 25). Bern.
Gaaf, W. van der (1928), "The Predicative Passive Infinitive", English Studies 10.107-14.
(1929), "The Conversion of the Indirect Personal Object into the Subject of a Passive Con-
struction", English Studies 11.1-11, 58-67.
(1930), "The Passive of a Verb Accompanied by a Preposition", English Studies 12.1-24.
Garvin, P. L., ed. (1963), Natural Language and the Computer. New York.
Gleason, H. A. Jr. (1965), Linguistics and English Grammar. New York.
Godfrey, J. (1965), "The Survey of English Usage", English Language Teaching 19.98-103.
Green, A. (1913), The Dative of Agency. New York.
(1914), "The Analytic Agent in Germanic", The Journal of English and Germanic Philology
Halliday, Μ. A. K. (1961), "Categories of the Theory of Grammar", Word 17.241-92.
(1964), "Syntax and the Consumer", Monograph Series on Languages and Linguistics 17.11-24.
Harris, Z. S. (1957), "Co-occurrence and Transformation in Linguistic Structure", Language 33.
Hatcher, A. G. (1949), "To Get/Be Invited", Modern Language Notes 64.433-46.
(1956), "Syntax and the Sentence", Word 12.234-50.
Hendriksen, H. (1948), "The Active and the Passive", Sprakvetenskapliga Sällskapets i Uppsala
förhandlingar 1946-8. Uppsala Universitets Arsskrift 1948:13.61-76.
Herdan, G. (1964), Quantitative Linguistics. London.
Hill, A. A. (1958), An Introduction to Linguistic Structures. New York.
Hill, L. A. (1964), "The Passive Again", English Language Teaching 18.69-74.
Hockett, C. F. (1958), A Course in Modern Linguistics. New York.
Jacobson, S. (1964), Adverbial Positions in English. Stockholm.
Jespersen, O. (1909-49), A Modern English Grammar on Historical Principles. Parts 1-7. London/
(1924), The Philosophy of Grammar. London.
(1933), Essentials of English Grammar. London.
(1937), Analytic Syntax. Copenhagen.
Jones, D. (1963), Everyman's English Pronouncing Dictionary. London.
Joos, M. (1964), The English Verb. Madison/Milwaukee.
Jud-Schmid, E. (1956), Der indefinite Agens von Chaucer bis Shakespeare (= Swiss Studies in
English 39). Bern.
Kennedy, A. G. (1920), The Modern English Verb-Adverb Combination (Stanford University
Publications, University Series, Language and Literature 1:1).
Kirchner, G. (1936-7), "The Verbs with Direct and Indirect Object Re-Examined", English Studies
18.1-16, 206-22; 19.97-112.
(1951), "Past Participles prefixed by 'un- Used in true Verbal Function to Form an Emphatic
Passive of Negation", English Studies 32.218-9.
Klingebiel, J. (1937), Die Passivumschreibungen im Altenglischen. Berlin.
Koumari, A. (1956), Certain Uses of the English Passive Voice and their Teaching to Foreign Students.
Kruisinga, E. (1927-31), A Handbook of Present-Day English. Part I in 4th ed.; Part 2 in 5th ed.
(1927), "Retained Accusatives in Passive Sentences", English Studies 9.38-40.
Kuiera, H. (1962), "A Note on the Digital Computer in Linguistics", Language 38.279-82.
Kurtz, G. (1931), Die Passivumschreibungen im Englischen. Berlin.
Lamb, S. M. (1961), "The Digital Computer as an Aid in Linguistics", Language 37.382-412.
Lees, R. B. (1957), Review of Chomsky, Syntactic Structures in Language 33.375-408.
(1963), The Grammar of English Nominalizations. Bloomington.
(1964), "On Passives and Imperatives in English", Gengo Kenkyu 46.28-41.
Long, R. (1961), The Sentence and its Parts. Chicago.
Longacre, R. E. (1964), Grammar Discovery Procedures. The Hague.
Lyons, J. (1963), Structural Semantics (Publications of the Philological Society, Vol. 20). Oxford.
Marchand, Η. (1960), The Categories and Types of Present-Day English Word-Formation. Wiesbaden.
McKerrow, R. B. (1922), "English Grammar and Grammars", Essays and Studies 8.148-67. Oxford.
Meier, Η. H. (1953), Der indefinite Agens im Mittelenglischen (1050-1350) (= Swiss Studies in
English 34). Bern.
Meyer-Lübke, W. (1925), "Vom Passivum", Neusprachliche Studien (= Die Neueren Sprachen,
Beiheft 6.158-71).
(1926), "Vom Wesen des Passivums", Die Neueren Sprachen 34.161-84.
Mihailovid, L. (1963), "Some Observations on the Use of the Passive Voice", English Language
Teaching 17.77-81.

Mitchell, Τ. F. (1958), "Syntagmatic Relations in Linguistic Analysis", Transactions of the Philological

Society 1958.101-18. Oxford.
Moroney, M. J. (1956), Facts from Figures (Pelican Book A 236).
Mustanoja, T. (1960), A Middle English Syntax. Part I: Parts of Speech ( = Mimoires de la Societi
Neophilologique de Helsinki 23). Helsinki.
Nida, E. A. (1960), A Synopsis of English Syntax. Norman, Oklahoma.
Nowell-Smith, P. (1954), Ethics (Pelican Book A 293").
Olsson, Y. (1961), On the Syntax of the English Verb with Special Reference to HAVE A LOOK
and Similar Complex Structures (= Gothenburg Studies in English 12). Gothenburg.
Owen, Ε. T. (1914), "The Relations Expressed by the Passive Voice", Transactions of the Wisconsin
Academy of Sciences, Arts, and Letters 17:1.77-148. Madison, Wisconsin.
Oxford English Dictionary. Oxford.
Palmer, F. R. (1965), A Linguistic Study of the English Verb. London.
Pike, K. L. (1954-60), Language in Relation to a Unified Theory of the Structure of Human Behavior.
Parts 1-3. Glendale, California.
(1962), "Dimensions of Grammatical Constructions", Language 38.221-44.
(1963), "A Syntactic Paradigm", Language 39.216-30.
Poutsma, H. (1926-9), A Grammar of Late Modern English, Parts 1 and 2:2. Groningen.
Quirk, R. (1960), "Towards a Description of English Usage", Transactions of the Philological Society
1960. 40-61. Oxford.
(1962), The Use of English. London.
(1965), "Descriptive Statement and Serial Relationship", Language 41.205-17.
Quirk, R. & Svartvik, J. (1966), Investigating Linguistic Acceptability. The Hague.
Reichmann, W. J. (1964), Use and Abuse of Statistics (Pelican Book A707).
Roberts, P. (1955), "Fries's Group D", Language 31.20-4.
Robins, R. H. (1964), General Linguistics. London.
Scheurweghs, G. (1959), Present-Day English Syntax. London.
Sledd, J., A Short Introduction to English Grammar. Chicago.
Smith, C. S. (1964), "Determiners and Relative Clauses in a Generative Grammar of English",
Language 40.37-52.
Söderlind, J. (1951, 1958), Verb Syntax in John Dryderis Prose, Parts 1 and 2. Uppsala.
Sokal, R. R. & Sneath, P. H. A. (1963), Principles of Numerical Taxonomy. San Francisco/London.
Sorensen, H. S. (1958), Word-Classes in Modern English with Special Reference to Proper Names with
an Introductory Theory of Grammar, Meaning and Reference. Copenhagen.
Stockwell, R. P. & Schachter, P. (1962), "Rules for a Segment of English Syntax" (mimeographed).
Strang, Β. Μ. H. (1962), Modern English Structure. London.
Svartengren, H. (1948), " 'Become' as an Auxiliary for the Passive Voice", Moderna Sprdk 42.272-81.
Sweet, H. (1955), A New English Grammar, Part 1. Oxford.
Turner, G. W. (1962), "The Passive Construction in English Scientific Writing", AUMLA 18.181-97.
Twaddell, W. F. (1963), The English Verb Auxiliaries. 2nd ed. Providence, R. I.
Visser, F. Th. (1941-56), A Syntax of the English Language of St. Thomas More. Louvain.
Vogt, Η. (1950), "Un aspect du problfeme actif-passif dans le verbe", Journal de Psychologie 43.130-8.
Vossler, Κ. (1925), "Das Passivum, eine Form des Leidens oder des Zustandes?", Die Neueren
Sprachen 33.401-7.
Webster's Third New International Dictionary of the English Language (1961). Springfield, Mass.
Zandvoort, R. W. (1960), A Handbook of English Grammar. London.
(1961), "Grammatical Terminology", English Studies Today, Second Series, 283-94. Bern.

A, see adjunct aspectual constraint 85 f.

Λ-item, /I-set 58 'attitudinal passives', see Class δ 1
abstract (nominal group head) 99, 108, 112, 165 auxiliaries 11 ff., 9Iff.
acceptability, linguistic 6, 164f. .elliptic42,67, 80
actional passive 86, 93 , lexically marked/unmarked 93, 134ff.,
actional-durative passive 93 147 f.
active, see transformation et passim , mutative/nonmutative 92ff., 136, 148
adjective 27, 96,101,128,135,145,191f. , Type a 10,12ff., 91,128,151f.; closed/open
adjunct (A) 25f., 70, 97, lOlff., 127, 158 class 14ff., 151, 168; fully tense-marked, non-
, agentive (Ag) 102ff., see also agent tense-marked, semi-tense-marked 16f., 91
, close M C L O S E ) 140ff„ 189ff., , Type b 10,17, 91,126,168
, criteria relating to the adjunct 74,79, lOlff. , Type c 10,17ff„ 91f., 168
, exponent classes 29ff., 41, 57ff., 71,128 , Type d 10,19,92ff., 168
, nonagentive (Ad) 57 ff., 108f.
, open ( Λ Ό Ρ Ε Ν ) 140ff. Bach, Ε. 1, 6,34
balance 156, 166
, position 30, 58ff„ 110,112,132,156
, sentence adjuncts 29 base, verbal/nonverbal 22,95
, verb-adjuncts 29 become-passive 92ff., 135f., 149f.,
Bloomfield, L. 31
adverb 29, 108, 157
agent (Ag) Bolinger, D.L. 26, 89,135, 159
, Ay-agent 102ff., 123,125,133,141ff. Brose, B. 3
, definition 30 Buyssens, E. 5
Ay-agent, see agent
, exponent class 41, 49 ff., 52, 116, 129ff.,
157, 166 C, see complement; C-full/less 65
, extension 108,112,123,126ff„ 134f., Carvell, Η. T. 21, 82, 111, 124
147ff., 158 case alteration 132,156
, external 135 Chomsky, N. If., 83,90,105,148
, internalized 135 CL, see clause
, Janus-agent 104ff., 133, 141, 157 'CLASP' (classification program) 113
, length 52ff., 112, 129ff., 156, 166 Class α ('animate agent passives') 40, 123, 132f.,
, quasi-agent 102ff., 134, 147 136, 138, 141 ff., 156ff.
, zero 126 Class β ('inanimate agent passives') 40,123,132f.,
agentful/agentless 9, 40, 81, 129, 134, 141, 151, 136, 138, 141ff.,154f.
156 ff. Class β/γ ('Janus-agent passives') 133,138,141 ff.
agentive/nonagentive 9, 40, 137f., 140ff., 147ff., Class γ ('agentless passives') 40, 123,134ff., 138,
164 141ff.,159
'agentless passives', see Class γ Class δ 1 ('attitudinal passives'), Class δ® ('emotive
Akerlund, A. 3 passives') 134f., 138, 157ff.
Allen, W. S. 2 Class ε ('nonagentive passives') 135ff., 138, 159
animate/inanimate 27f., 34, 50, 108, 112, 123f., Class ζ ('compound passives') 137f.
141 classification of passive clauses 72-138
'animate agent passives', see Class α clause 13, 25-38

clause, criteria relating to the clause (CL) 74,75, dispensability 26

78 ff. ditransitive clause 37, 62ff., 83, 138ff., 188f.
, elements V, W, S, C, A 25f., 41 do-periphrasis 12, 25
, finite/nonfinite verb clause 27,29, 55ff., 71, double-tail test 119
101, 104, 110, 144f., 147,183ff., 188 durative passive 92
, parenthetic 81,144
, relation: external 31 ff., 41, 65ff., 71, 78ff., 'emotive passives', see Class δ2
126ff., 158; internal 31, 69, 70, 82 emotive verbs 84
, sequence 156 equative clauses 37,62ff, 84,86,93,135ff., 159ff.
, types 33,36f., 41,62ff., 71,139ff.; extensive equative relation 140
(EXT, ex/)/intensive(INT, int) 33ff., 83f., 123, Erades, P. A. 3, 90,161
128, 134, 139 exponential coreference 33
cline distribution 116, 132 extensive, see clause
colligation 102
factitive clause 37,62ff.,83,86,136, 138ff, 160,
collocation 15, 102, 138, 165
191 ff.
common 27 f.
feature 11 Iff., 123 ff., 132ff, 158
communicative dynamism (CD) 38 finite verb clause, see clause
commutation 22f., 124f., 128 finitude 28, 99,125f., 157
complement (C) 20, 25ff., 41, 64f., 188f. Finney, D. J. etal. 116
, criteria relating to the complement 74, 78, Firbas, J. 25, 38,166
101 Firth, J. R. 5
, exponent classes 26ff., 55ff., 70f., 145 Francis, W. N. 3,4,15,19, 27, 29, 86,93
compounds 21, 86, 95 Frary, L. G. 3,92
'compound passives', see Class ζ free clauses 31ff., 65 ff., 81,128,158
conclusive/non-conclusive verbs 85 f. Fries, C. C. 3,4,5,27,37,86,92
concord, covert/overt 25; valency classes 16ff. Fröhlich, J. 3
concrete 99,108,112,125,165 functional sentence perspective (FSP) 166
conjunctional clauses 32, 67, 80
constituent structure 36 Gaaf, W. van der 3
contextualization 21 Garvin, P. L. 41
contingency tables 116ff. gender 27f„ 34,50f., 57,70,112,125,157
continuum 26 ^/-passive 92ff, 136f., 149f.
conversion 37 Gleason, H. A. Jr 165
coordination 28, 52, 96, 100, 125, 128f., 134f., Godfrey, J. 6
137, 157, 166 Green, A. 3, 31,103
copula 34, 38, 93 f. group, linguistic 5 et passim
coreference 33,37. 81 ff., 124,127f., 158,165 Groups I-VI (taxonomic) 115f., 119ff.
corpus 5ff.
, Corpus I, see Voice Corpus Halliday, Μ. A. K. 5,25,26, 33
, Corpus II, see Minor Passive Corpus Hatcher, A. G. 3,93
, Corpus III, see Taxonomic Passive Corpus Hendriksen, H. 3, 31
, Corpus IV, see Major Passive Corpus Herdan, G. 116
countable 99,112,125 ff., 129,157 Hill, A. A. 3,4,10,13,15,165
criteria 73, 74,123 Hill, L. A. 3
Crystal, D. 6, 20, 82 Hockett, C. F. 3
Curme, G.O. 3, 86, 89
imperfective passive 86
data: preparation of input data 111ff.;voice data 'inanimate agent passives', see Class β
processing 39-71 indefinite 28,99,125f„ 129,157
definite 28,99,125 f., 129,157 indeterminacy, structural 133
deletion 35,106,125,157 infinitive
delicacy, analytical 26, 138 , to-full (to V) 14f., 27, 30, 57, 101, 104,
determiner 26 146f., 185,192f.
diagnostic frames 21 , to-less (K) 12,14f.,27
diagnostic key 132ff. input 39, 41
'dichotomies' 112 intensive, see clause
intransitive clause 37, 62ff., 159 ff. Null Hypothesis 116f.
introductory there 86 number 26, 28, 34,100,140
invariance for person and number 12 numerical taxonomy 72f., I l l
Jacobson, S. 30 object, (O) 37f„ 127f„ 158
Janus-agent, see agent , direct {Od), indirect (O.) 37,140
'Janus-agent passives', see Class β/γ obligatory 25, see feature
Jespersen, Ο. 1, 3, 4, 22, 31, 37, 86, 88, 89, 92, Olsson, Y. 10, 16, 90
102, 141 operational taxonomic unit (OTU) 73, 111
Joos, M. 3 optional 25, 58, see feature
Jud-Schmid, E. 3 OTU, see operational taxonomic unit
output 39,41
Kennedy, A. G. 20
Owen, Ε. T. 3
kinetic passive 86
Oxford English Dictionary 23, 103,108, 162
Kirchner, G. 3
Klingebiel, J. 3
Koumari, A. 3, 31 p, see probability
Kruisinga, E. 2, 3,4, 86 P, see predicative
Ku£era, H. 41 Palmer, F. R. 3
Kurtz, G. 3 participle: present (Ving), past (Ved) 27, 30, 97,
101, 145, 191ff.
Lamb, S. M. 41 passive of 'being' and 'becoming' 86
Lees, R. B. If., 105 'passive scale' 134ff., 138, 156ff.
levels (in computer program) 112; (in significance passive voice: definitions 3ff., use of 139-55
test) 118 perfective passive 86, 92
Long, R. 90 permutation 88 ff., 125
Longacre, R. E. 37 person 28,100,112
Lyons, J. 22 phrase, prepositional 20f., 29,96,104,109f., 140,
'macrosemantic' similarity 147,159 Pike, K. L. 5,11
Major Passive Corpus (Corpus IV) 9,139ff., 151, plural (pi) 34
167ff. 'positives' 112
Marchand, H. 22 Poutsma, Η. 1, 37,86,89,92,102
mass nouns 99,112,125ff., 129,157 predicate 37
material 6ff. predicative (P) 37f., 127f„ 137,158
McKerrow, R. B. 2 probability (p) 116ff.
Meier, Η. H. 3 program (computer) 39, 41f., 112
Meyer-Lübke, W. 3 proleptic it 13, 81,86,112,144ff., 183ff„ 188
Mihailovic, L. 3 pronoun 26,28,52, 55ff., 98,112,127,128f„ 157
Minor Passive Corpus (Corpus II) 9, 39ff., 129,
157 qualification 95,128,134f., 147 f.
Mitchell, T. F. 20 'qualitatives' 112
modification 26, 28, 96,112,124ff., 127 ff„ 157 f. 'quantitatives' 112
Moroney, M. J. 116 quasi-agent, see agent
Mustanoja, T. F. 3,103 'quasi-agentive passives', see Class δ
mutative, see auxiliaries question-adverb 35,102

N, see nominal element Quirk, R. 6,10,15 f., 20, 25, 74, 82,116,159, 164
name 26,28,98,112,129
negation, postverbal enclitic 12,15,18 rank 5, 25
Nida, E. A. 3 referential pronoun 28,98,112
nominal group 26f., 98,157 Reichmann, W. J. 116,118
nominal element (N) 36, 145ff., 187f., 191 f. relationship: internal/external 115 ff.; serial 159 ff.
nominalization 26 f. relative clauses 32,67,78,128
'nonagentive passives', see Class ε relator: fixed, mobile 32, 67, 80; sequential 29
noun 26,52,55ff., 98,101,112,129,157 rheme 38, 166
Nowell-Smith, P. 148 Roberts, P. 95
Robins, R. H. 15,102

S, see subject system 5 et passim

scale 26, 33, 38 systemic coreference 33,36,37
Schachter, P. 34 systemic voice transformation potential 36, 132,
Scheurweghs, G. 4, 86 134, 156
semiditransitive 139ff., 185 ff.
semifactitive 139ff., 189ff. tag question 32, 67
semitransitive 62 Taxonomic Passive Corpus (Corpus III) 9, 73 ff.,
sequential constraints 85,132 157
sequentially related clauses 25, 31 f., 65 if., 80 taxonomy, see classification
serial isolate 162 term 5 et passim
serial voice relation 159ff. theme 166
significance, statistical 116ff. transformation
similarity , constraints 84,139
, coefficient 113 , dual 128
, inter-group, internal, intra-group 115 ff. , potential 24, 26, 34, 36, 83ff., 88ff., 112,
, matrix 113ff. 123, 127 f., 134ff., 147f., 158
, mean 114ff. , theory 1
single-tail test 118 , voice relation 159ff.
singular (sg) 34 transformational structure (of major clause types)
Sledd, J. 37 36
Smith, C. S. 33 transitive clauses 37, 62ff., 83, 86, 138, 167ff„
Sneath, P. H. A. 72 183ff.
Söderlind, J. 3 transmutation 88 ff., 128f., 135,162
Sokal, R. R. 72 Turner, G. W. 3,86
Serensen, H. S. 27 Twaddell, W. F. 10,165
speech, direct 27, 55ff.,64,144,183ff. Types a, b, c, d, see verbal group
statal passive 86,93,135,150,161 f. un-, negative/reversative 22, 86
static passive 86
statistical assessment 116ff. V, see verbal group
Stockwell, R. P. 34 valency classes 16ff.
Strang, Β. Μ. H. 3,10,20,25,27, 29,93 verbs
structure 5 et passim , auxiliaries; lexical verbs 11, 19ff., 41, 67,
style 70, 87,132 95ff.,112, 165, 167
subject (S) 25ff., 37,118ff., 158 , phrasal/prepositional 19ff., 95, 165
.converted 1, 31 verbal group (K, W) 10-24, 25f„ 41, 43ff„ 70f.,
, criteria relating to the subject 74, 77, 97 ff. 150, 158
, elliptic 32, 67, 80 , criteria relating to the passive verbal group
, exponent classes 26ff., 41, 49ff„ 70f., 116, 74, 76, 91 ff.
124ff., 134,147, 156ff., 166 , simple 10, complex 10-19
, external relation 81,158 verbful construction 38
, grammatical 89 verbless construction 30, 38, 55 ff.
, inverted 31 Visser, F. Th. 3
, length 52ff.,70,112,128 ff. Voice Corpus (Corpus 1) 9, 39ff, 158
, logical 89 voice
, referential 144 ff., 167ff., 185 , data processing 39-71
, suppletion 84 , distribution 70
subscript 34 , relation 138
substitution classes 27 , transformation, see tranformation
superscript 34 Vogt, Η. 5
Survey of English Usage 6 Vossler, Κ. 3
Svartengren, H. 3, 93
Svartvik, J. 6, 15f., 21, 72, 74, 111, 116, 124,164 W, see verbal group
Sweet, H. 31 Webster's Third New International Dictionary 162
syntactically bound clauses 3If., 65ff., 78ff., 127,
158 Xä-test 46,116ff.
syntagmatic affinity 138 Zandvoort, R. W. 4,5,86,92