(Prof. Peter Collins, David Lee (Eds.) ) The Clause

THE CLAUSE IN ENGLISH
STUDIES IN LANGUAGE COMPANION SERIES (SLCS)

The SLCS series has been established as a companion series to
STUDIES IN LANGUAGE, International Journal,
sponsored by the Foundation “Foundations of language”.
Series Editors
Werner Abraham Michael Noonan
University of Groningen University of Wisconsin-Milwaukee
The Netherlands USA
Editorial Board
Joan Bybee (University of New Mexico)
Ulrike Claudi (University of Cologne)
Bernard Comrie (Max Planck Institute, Leipzig)
William Croft (University of Manchester)
Östen Dahl (University of Stockholm)
Gerrit Dimmendaal (University of Leiden)
Martin Haspelmath (Max Planck Institute, Leipzig)
Ekkehard König (Free University of Berlin)
Christian Lehmann (University of Bielefeld)
Robert Longacre (University of Texas, Arlington)
Brian MacWhinney (Carnegie-Mellon University)
Marianne Mithun (University of California, Santa Barbara)
Edith Moravcsik (University of Wisconsin, Milwaukee)
Masayoshi Shibatani (Kobe University)
Russell Tomlin (University of Oregon)
John Verhaar (The Hague)
Volume 45
Peter Collins and David Lee (eds)
The Clause in English

THE CLAUSE
IN ENGLISH
In honour of Rodney Huddleston
Edited by
PETER COLLINS
University of NSW
DAVID LEE
University of Queensland
JOHN BENJAMINS PUBLISHING COMPANY

AMSTERDAM/PHILADELPHIA
TM The paper used in this publication meets the minimum requirements
8
of American National Standard for Information Sciences — Permanence of

Paper for Printed Library Materials, ANSI Z39.48–1984.
Library of Congress Cataloging-in-Publication Data

The clause in English : in honour of Rodney Huddleston / edited by Peter Collins, David Lee.
p. cm. -- (Studies in language companion series, ISSN 0165-7763 ; v. 45)
Includes bibliographical references and index.
1. English language--Clauses, I. Huddleston, Rodney O. II. Collins, Peter, 1950- . III. Lee, David
(David A.) IV. Series.
PE1385.C57 1998
425--dc21 98-39788
ISBN 90 272 3048 X (Eur.) / 1 55619 931 7 (US) (alk. paper) CIP
© 1999 – John Benjamins B.V.
No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any other
means, without written permission from the publisher.
John Benjamins Publishing Co. • P.O.Box 75577 • 1070 AN Amsterdam • The Netherlands
John Benjamins North America • P.O.Box 27519 • Philadelphia PA 19118-0519 • USA
Table of Contents
Contributors vii
Introduction ix
Peter Collins and David Lee
Curriculum Vitae of Rodney Desmond Huddleston xvii
The semantics of English quantifiers 1
Keith Allan
Language, linear precedence and parentheticals 33
Noel Burton-Roberts
The English modifier well 53
Ray Cattell
The deictic-presentation construction in English 67
Peter Collins
Relative clauses: Structure and typology on the periphery of standard
English 81
Bernard Comrie
Post nominal modifiers in the English noun phrase 93
Peter H. Fries
Elliptical clauses in spoken and written English 111
Sidney Greenbaum and Gerald Nelson
On the nature of ?I believe Jack to arrive tomorrow 127
Hisashi Higuchi
Intransitive prepositions: are they viable? 133
David Lee
vi TABLE OF CONTENTS
Sentences, clauses, statements and propositions 149

John Lyons
Some interactions between tense and negation in English 177
James McCawley
The English accusative-and-infinitive construction: A categorial
analysis 187
John Payne
On the boundaries of syntax: Non-syntagmatic relations 229
Peter Peterson
Gerund participles and head-complement inflection conditions 251
Geoffrey Pullum and Arnold Zwicky
Isolated if-clauses in Australian English 273
Lesley Stirling
Functional and structural: the practicalities of clause knowledge in
language education 295
Lynn Wales
Subject Index 323
Contributors
KEITH ALLAN BERNARD COMRIE

Department of Linguistics Max-planck-institut für
Monash University evolutionäre Anthropologie
Clayton VIC 3168 Inselstaße 22
Australia D-04103 Leipzig
keitha@silas.cc.monash.edu.au Germany
comrie@eva.mpg.de
NOEL BURTON-ROBERTS PETER FRIES
Department of English Literary Box 310
and Linguistic Studies Mount Pleasant
University of Newcastle upon Michigan MI 48804
Tyne NE1 7RU USA
U.K. 104462.1706@compuserve.com
N.Burton-Roberts@newcastle.ac.uk
RAY CATTELL HISACHI HIGUCHI
48 Roy Ave 206 Sun-Green Arimochi
Bolton Point NSW 2283 2–29–7 Yagumo-Kita
Australia Moriguchi, Osaka 570-0008
LNNRC@cc.newcastle.edu.au Japan
PETER COLLINS DAVID LEE
Linguistics Department Department of English
University of NSW University of Queensland
Sydney 2052 QLD 4072
Australia Australia
p.collins@unsw.edu.au dalee@lingua.cltr.uq.edu.au
viii CONTRIBUTORS
JOHN LYONS GEOFFREY PULLUM

The Master’s Lodge Stevenson College
Trinity Hall University of California at Santa
Cambridge University Cruz
Cambridge CB2 1TJ 1156 High Street
U.K. Santa Cruz, CA 95064
pullum@ling.ucsc.edu
JAMES McCAWLEY LESLEY STIRLING
Department of Linguistics Department of Linguistics and
University of Chicago Language Studies
5801 Sth Ellis Ave University of Melbourne
Chicago, Illinois 60637 Parkville VIC 3052
USA Australia
jmccawle@midway.uchicago.edu stirling@ariel.ucs.unimelb.edu.au
GERALD NELSON LYNN WALES
Survey of English Usage, Centre for Language Teaching and
University College London, Research
Gower St, University of Queensland
London WC1E 6BT, UK QLD 4072
uclegen@ucl.ac.uk Australia
wales@lingua.cltr.uq.edu.au
JOHN PAYNE ARNOLD ZWICKY
Department of Linguistics Department of Linguistics
University of Manchester Stanford University
Manchester M13 9PL Stanford CA 94305
U.K. zwicky@csli.stanford.edu
John.Payne@Manchester.ac.uk
PETER PETERSON
Department of Linguistics
University of Newcastle
NSW 2308
Australia
lnpgp@cc.newcastle.edu.au
Introduction*
Peter Collins David Lee
Rodney’s name first came to the attention of one of us (David Lee) in 1956 at
the Manchester Grammar School, since he had just won a Major Scholarship to
Corpus Christi College, Cambridge. In those days the Modern Sixth at MGS was
presided over by a legendary schoolmaster by the name of Albert Hyslop,
universally known as “Alf”. It is appropriate to begin by saying a few words
about Alf, since he was undoubtedly one of the major influences on Rodney’s
early life and indeed on his subsequent career.
The success achieved by Alf’s pupils in the Oxbridge entrance examinations
over many years was based on some unusual pedagogic practices. One of these
involved handing out detailed notes on the French seventeenth century classic
tragedies and comedies, particularly the plays of Corneille, Racine and Molière.
These notes flowed constantly from Alf’s pen and were passed to his favourite
pupils in an almost conspiratorial manner at the end of many a school day. They
were written in minute, spidery handwriting, yet it was nothing to receive three
or four pages of exegesis on just a few lines of text. Most of these were
obviously written on the bus from Manchester to Wilmslow (where Alf lived),
since they turned up on scraps of paper of all sizes and conditions. Alf’s notes
were the fruit of years of musings on the qualities and flaws of the heroes and
heroines of the classical dramas and on the significance within the context of the
play as a whole of a particular speech or word. Any student who worked through
them ended up by being able to recite large sections of the plays by heart and
discuss crucial passages in some depth. This clearly went down very well with
the Oxbridge examiners. But it also inculcated in Alf’s students the habit of
paying attention to detail that was to stay with them in many cases for the rest
of their lives.
x INTRODUCTION
Clearly, the credit for Rodney’s extraordinary analytic gifts is entirely his
own. But there seems little doubt that Alf played a major role in fostering the
early development of these gifts, not only through his notes but through the
demonstration in his (sometimes wildly) enthusiastic teaching that the enjoyment
of literature could be immeasurably enhanced by close textual analysis. It was in
fact a perfect method not only for developing an appreciation of literature but
also for fostering a love of language, since the subtleties of Alf’s interpretations
often hinged on the finest linguistic nuances.
This book is a celebration of Rodney’s career but it seems not inappropriate
to include this expression of deep appreciation for a schoolmaster who had a
major impact on Rodney and on the lives and careers of many others over
several generations.
When Rodney left Manchester Grammar School, military service was still
compulsory, so he deferred his Cambridge scholarship for two years to learn
about war. It somehow seems typical that what he in fact did with his military
service was learn to type. This skill came in useful much later with the advent
of word processing, which he embraced with great enthusiasm. He loves the
precision of the PC and the advent of computer corpora has added a new
dimension to his work on grammar.
After graduating from Cambridge with a First Class Honours degree in
Modern and Mediaeval Languages (1960), Rodney enrolled for a PhD at
Edinburgh, supervised by Michael Halliday. He completed this in 1963 and was
appointed to a lectureship in linguistics. He moved with Halliday to University
College London in 1965, where he led a research project on the linguistic
properties of scientific English.
In 1967 he took up a lectureship at the University of Reading, where he
worked with Frank Palmer, Peter Matthews and David Crystal. He moved to a
temporary lectureship at University College, London in 1968.
In 1969 he moved to the University of Queensland, where he has spent the
rest of his career. By this time Rodney had already acquired a strong reputation
in English grammar. He quickly established himself at Queensland not only as an
outstanding researcher but also as a fine teacher. He himself was surprised by the
quality of the Queensland students. In his first undergraduate class, he had a
number of students who produced original work in syntax, including one who
presented an excellent critical analysis of Jacobs and Rosenbaum’s English
Transformational Syntax. Later generations of students did not always quite live
up to his first impression that Queensland was a land of geniuses but there is no
INTRODUCTION xi
doubt that he taught a significant number of outstanding students. Of course the

standard of their work owed an enormous amount to the quality of his teaching.
Not only did they respond to the meticulous care with which he presented
grammatical description and theory but they were also conscious of his deep
personal commitment to them. One student was overheard in the library confess-
ing to a friend her concern about not having managed to complete the worksheet
for Rodney’s next class. “He’ll be so upset,” she wailed.
In 1989 Rodney conceived the daunting plan of writing a major new
grammar of English, after reviewing Quirk et al’s Comprehensive Grammar of
the English Language. Rodney had a great deal of respect for this work but felt
that there was a need for a grammar that was more theoretically rigorous and
more internally coherent. The grammar workshops that have been held twice or
three times a year in Sydney and Brisbane have been an unforgettable experience
for those of us fortunate to have been part of the team. It was in these work-
shops that the depth of Rodney’s astonishing knowledge of grammatical theory
and description came into play in full force. It was always something of an
ordeal to present a draft, since we were all quite well aware of the fact that it
was now to come under the scrutiny of one of best minds in the field. Yet there
was also a spirit of great goodwill and humour in the proceedings. Intellectually,
these were intense and exhausting sessions but they were associated with
extremely enjoyable social gatherings. In some ways it is the social side of these
events that lingers in the memory long after the details of linguistic discussion
are forgotten. We remember particularly dawn jogs to Alexandra Beach from
Rodney’s house at Sunshine Beach, pool volleyball and table tennis games
fought with great ferocity, and walks through Noosa National Park with spectac-
ular sunsets over Noosa Bay.
The fact that a number of workshops have been held at Rodney’s house at
Sunshine Beach, Noosa is just one example of the generous hospitality that he
has extended over the years to many friends and colleagues. There has, however,
sometimes been a price to pay. Those unwary enough to react positively to an
innocent suggestion that they might enjoy “a walk” in the Glasshouse Mountains
near Noosa soon found themselves clinging desperately to minute bumps and
depressions in a sheer rock face that guards the access to Mount Beerwah.
Linguists like Grev Corbett, Geoff Pullum, Geoff Leech (and his wife Fanny)
have fallen into this trap but luckily they all survived (by the skin of their shins,
in some cases).
xii INTRODUCTION
Although he is undoubtedly one of the leading international researchers in

English grammar, Rodney has always assigned the greatest importance to
teaching. It is therefore appropriate to conclude with some comments on this
crucial aspect of his work. In 1988 the University of Queensland inaugurated a
system of awards for excellence in teaching and Rodney was one of the first
group of awardees. One particularly impressive document relating to this process
was a letter which had appeared two years earlier in the Newsletter of the
Queensland Association of Teachers of English as a Second Language
(QATESOL), August 1986, by J. McCormick. McCormick noted that, as a
student in an introductory course in linguistics taught by Rodney, he had learned
a great deal about linguistics but he had learned even more about teaching. He
commented on the careful grading in the presentation of material, on the fact that
knowledge was never assumed unless Rodney had taught it himself, on the fact
that he always corrected worksheets to see whether students had mastered the
material, retaught areas that were clearly causing major difficulty, handed out
solutions to worksheets, answered questions on the worksheets, dispelled doubts,
re-iterated principles, used simple language when explaining a point, didn’t show
off or use unnecessary linguistic jargon and gradually built up a blackboard
display working to a preconceived plan, so that by the end of the class the board
presented a clear picture of the path of the lesson. He commented on Rodney’s
patience, courtesy and humour in dealing with all types of question and on the fact
that he gave out revision questions at regular intervals throughout the semester
so that students could assess their own learning. He summed up as follows:
Because he expects them to learn, they learn. His continual application to the
job and the integral feedback and follow-up approach are an example that
could well be followed by teachers in any field. I was inspired.
This will give some indication of the admiration and affection that Rodney
engendered in his students. It is a eulogy that will probably give him as much
pleasure as any number of favourable reviews of the Grammar.
The contents of the present volume reflect Rodney Huddleston’s primary
interest, in English grammatical theory and description. Rodney is unquestionably
— though modesty would undoubtedly prompt him to vigorously deny it — one
of the outstanding grammarians of the present century. Building on his detailed
knowledge of work in the major frameworks, traditional, structural/generative
and functional, he has continued to produce careful and original grammatical
descriptions that constantly advance our understanding of English grammar.
INTRODUCTION xiii
The arrangement of papers presented us with a formidable challenge as

editors, given that they lend themselves to subclassification along a diversity of
parameters. Some are based on corpus data (e.g. Greenbaum and Nelson, Fries,
Stirling); others on introspectively-derived examples (e.g. McCawley, Peterson,
Higuchi). Some have a chiefly descriptive orientation (e.g. Cattell, Greenbaum
and Nelson, Collins); others chiefly theoretical (e.g. Allan, Lyons). Some take as
their point of departure contentious or allegedly false earlier claims (e.g. Pullum
and Zwicky, Peterson, Lee). Some explore “minor” or “fringe” structures (e.g.
Stirling, Comrie). In the end we decided to order the collection simply in terms
of the alphabetical ordering of authors’ names. There follows a brief survey of
the contents.
• Allan uses a formal metalanguage to relate number, countability, quanti-
fication and definiteness, and to demonstrate their dependence on clause seman-
tics. • Burton-Roberts adopts a “representational” approach to the distinction
between language and speech. This enables him to tackle problems associated
with the analysis of parentheticals, particularly non-restrictive relative clauses.
• Cattell defines a continuum of expressions containing the modifier well, from
those with a verb-like passive participle (e.g. a well fielded ball) to adjective-like
(e.g. a well typed essay). • Collins presents arguments for recognising a distinct
“deictic-presentation construction” in English, one that has developed special
features such that it cannot be merely subsumed under “locative inversion”.
• Comrie adduces data from the periphery of standard English which suggest that
relative that is sometimes a subordinator (e.g. when the NP head is not taken up
in the relative clause), and sometimes a pronoun (for example when inflectionally
genitive, as that’s). • Fries examines the arguments for a distinction between
post-head complements and modifiers in NP-structure (a number of which are
based on the comparable behaviour of these categories in the clause). He
concludes that the evidence is less clear cut than has been suggested by previous
writers. • Greenbaum and Nelson find that ellipsis conditioned by coordination
is, in their British corpus, more common in writing than speech, with the reverse
finding for ellipsis which is not conditioned by coordination. • Higuchi explores
the restrictions imposed by verbs such as believe on their infinitival complements
(which cannot be “eventive”: *I believe her to die). • Lee explores recent propos-
als that the preposition category be extended to include words such as aboard,
downstairs, here and then, which do not take a complement (“intransitive
prepositions”). Finding the supporting arguments unconvincing, he suggests that
a promising framework for conceptualising the problems here is “emergence
xiv INTRODUCTION
theory”, in which no assumption is made that all the words in a given language
will necessarily slot into a restricted set of lexical categories. • Lyons explores
the semantic justification for postulating the categories sentence, clause, state-
ment, and proposition. In the process he vigorously attacks a number of com-
monly made assumptions (e.g. that declaratives are necessarily more basic than
non-declaratives; that propositional content is universal throughout the world’s
languages). • McCawley revises his own earlier proposal that tenses and auxiliary
verbs are deep structure predicate elements taking clausal complements, noting
that while this proposal allowed for both deep structures in which a tensed clause
is negated and for those in which the complement of a tense is negated, only the
former type were actually made use of. • Payne uses a version of categorial
grammar to provide a new solution to various syntactic problems raised by
accusative and infinitive constructions in English, a solution which involves the
treatment of objects as infixed functors. He explores the consequences of the
proposed analysis for heavy NP shift, coordination and extraction. • Peterson
provides evidence that “juxtaposed” elements do not form syntactic units with the
“host” clause. In this regard he finds himself essentially in agreement with
Burton-Roberts, who speaks of a relationship of semantic/pragmatic coreference
rather than syntactic coindexing between a parenthetical element and its “host”.
Both Peterson and Burton-Roberts comment on the inadequacies of conventional
tree diagrams and labelled bracketings in representing the relationships in
question. • Pullum and Zwicky claim that Milsark’s reformulation of Ross’s
“doubl-ing constraint” is a misguided attempt to treat a syntactic condition in
morphophonemic terms. They present and argue for a version of the constraint
that is not only consonant with the facts of usage but quite conservative in the
theoretical concepts that it invokes. • Stirling considers whether it is necessary to
posit ellipsis in the analysis of “isolated if-clauses” (i.e. those without an
accompanying main clause), or whether they have become sufficiently conven-
tionalised to count as distinct sentence types. • Wales discusses the importance
of structural knowledge — of the clause in particular — in language education,
arguing for instance that in the later years of secondary school students encounter
texts containing a variety of syntactic structures not commonly found in everyday
language. She observes that while Halliday’s functional model declaredly
provides the inspiration for much recent curriculum material in Australia, the
influence of descriptivist structural models is in evidence, although this is
insufficiently acknowledged.
INTRODUCTION xv
Notes
* We most gratefully acknowledge the assistance given to us by Peter Peterson in assessing and
commenting upon papers, and by Maria Oujo in the preparation of the manuscript for
publication, and by Kathy Collins in the preparation of the index. We are also grateful to Joan
Mulholland for providing some of the information included in this introduction.
Curriculum Vitae of Rodney Desmond Huddleston
DATE OF BIRTH: 4 April 1937.
Academic qualifications and awards

B.A. Honours Class I (Modern and Medieval Languages), Cambridge, 1960.
(Awarded Corpus Christi College “Bishop Green” Cup for the best B.A. result.)
Graduated M.A. 1964 (without further examination).
Ph.D., Edinburgh, 1963; thesis: “A descriptive and comparative analysis of texts
in French and English: an application of grammatical theory”.
Fellow of the Australian Academy of the Humanities; elected 1984.
Senior Research Fellow, Australian Research Council. 1993–98.
Positions held
University of Edinburgh: Lecturer (Oct 1963 – Dec 1964) in the Department of
English Language and General Linguistics.
University College London: Research Assistant (Jan 1965 – Sept 1967) in the
Department of General Linguistics; Principal Investigator in the research
programme “The Linguistic Properties of Scientific English”, supported by the
Office of Scientific and Technical Information.
University of Reading: Lecturer (Oct 1967 – Sept 1968) in the Department of
Linguistic Science.
University College London: Lecturer (Oct 1968 – June 1969) in the Department
of General Linguistics.
University of Queensland: Lecturer in Linguistics (from June 1969), Senior
Lecturer (from Jan 1971), Reader (from Jan 1975), Professor (Personal Chair
from Oct 1990) in the Department of English.
xviii CURRICULUM VITAE OF RODNEY DESMOND HUDDLESTON
Teaching
Winner of one of the three inaugural “Excellence in Teaching” awards, Universi-
ty of Queensland, 1988.
Service
Foundation Editor, Australian Journal of Linguistics (journal of the Australian
Linguistics Society), 1980–1985; member of the Editorial Board from 1985.
Vice-President, Australian Linguistics Society, 1986–1988.
Member of the Editorial Board, Studies in English Language (Cambridge
University Press), from 1987.
Member of the Editorial Board, Cambridge Studies in Linguistics and Cambridge
Textbooks in Linguistics (Cambridge University Press), from 1990.
Chairperson, Linguistics Section, Australian Academy of the Humanities,
1991–94.
Publications
1. Monographs
The Sentence in Written English: A Syntactic Study Based on a Study of Scientific
Texts. (Cambridge Studies in Linguistics, 3.) Cambridge University Press.
1971. Pp. viii + 344.
An Introduction to English Transformational Syntax. (English Language Series.)
London: Longman. 1976. Pp. xiii + 213. Translated into Japanese as Hen kei
too go ron gai setsu (English Grammar Methodology, 4), Nanundoo. 1980.
Introduction to the Grammar of English. (Cambridge Textbooks in Linguistics.)
Cambridge University Press. 1984. Pp. xvi + 483.
English Grammar: An Outline. Cambridge University Press. 1988. Pp. xii + 212.
English Grammar in School Textbooks: Towards a Consistent Linguistic Alterna-
tive. Occasional Paper No 11. Applied Linguistics Association of Australia.
1989. Pp. iv + 99.
2. Articles
“Rank and Depth”. Language 41 (1965). 574–586. Reprinted in Readings in
Systemic Linguistics ed. by M.A.K. Halliday and J.R. Martin. London:
Batsford (1981), 42–53.
“More on the English Comparative”. Journal of Linguistics 3 (1967), 91–102.
CURRICULUM VITAE OF RODNEY DESMOND HUDDLESTON xix
“Some Observations on Tense and Deixis in English”. Language 45 (1969),

777–806.
[With O. Uren] “Declarative, Interrogative and Imperative in French”. Lingua 22
(1969), 1–26. Reprinted in Readings in Systemic Linguistics ed. by M.A.K.
Halliday and J.R. Martin. London: Batsford (1981), 237–256.
“Some Remarks on Case-Grammar”. Linguistic Inquiry 1 (1970), 501–511.
“Two Approaches to the Analysis of Tags”. Journal of Linguistics 6 (1970),
215–222.
“The Syntagmeme”. International Journal of American Linguistics 37 (1971),
39–44.
“A Problem in Relative Clause Reduction”. Linguistic Inquiry 2 (1971), 115–116.
“A Comparative Tautology”. Linguistic Inquiry 2 (1971), 252–254.
“The Development of a Non-Process Model in American Structural Linguistics”.
Lingua 30 (1972), 333–384.
“A Note on Order”. Journal of Linguistics 9 (1973), 251–260.
“Embedded Performatives”. Linguistic Inquiry 4 (1973), 539–541.
“Further Remarks on the Analysis of Auxiliaries as Main Verbs”. Foundations of
Language 11 (1974), 215–229.
“Componential Analysis: The Sememe and the Concept of Distinctiveness”.
Canadian Journal of Linguistics 19 (1974), 1–17.
“Homonymy in the English Verbal Paradigm”. Lingua 37 (1975), 151–176.
“The Linguistic Study of Language”. Language Performance and Exceptional
Children ed. by R.J. Andrews. St Lucia: Fred and Eleonar Schonell Educa-
tional Research Centre (1976), 17–32.
“Past Tense Transportation in English”. Journal of Linguistics 13 (1977), 43–52.
“In Defence of Parasitic Base Structures”. Studies in Language 1 (1977),
245–254.
“The Futurate Construction”. Linguistic Inquiry 8 (1977), 730–736.
“On the Constituent Structure of VP and Aux”. Linguistic Analysis 4 (1978),
31–59.
“A Survey of the Crossing–Coreference Controversy”. Papers in Linguistics 11
(1978), 295–319.
“Would have become: Empty or Modal will?”. Journal of Linguistics 15 (1979),
335–340.
“On Palmer’s Defence of the Distinction between Auxiliaries and Main Verbs”.
Lingua 50 (1980), 101–115.
xx CURRICULUM VITAE OF RODNEY DESMOND HUDDLESTON
“Criteria for Auxiliaries and Modals”. Studies in English Linguistics for Randolph
Quirk ed. by S. Greenbaum, G. Leech and J. Svartvik. London: Longman
(1980), 65–78.
“Systemic Features and their Realization”. Readings in Systemic Linguistics ed.
by M.A.K. Halliday and J.R. Martin. London: Batsford (1981), 58–73.
“A Fragment of a Systemic Description of English”. Ibid., 222–236.
“The Treatment of Tense in Indirect Reported Speech”. Folia Linguistica 23
(1989), 335–340.
“What Ought Teachers to Know about English Grammar?”. Responding to
Literacy Needs: Implications for Teacher Educators and Training Consultants
ed. by Board of Teacher Registration. Queensland. Toowong. Queensland
(1991), 75–86.
[With M.A.K. Halliday] “Theory of Language”. Linguistics in Australia: Trends
in Research ed. by M. Clyne. Canberra: Academy of the Social Sciences in
Australia (1991), 15–34.
“On Exclamatory-Inversion Sentences in English”. Lingua 90 (1993), 259–269.
“Remarks on the Construction You won’t believe who Ed has married”. Lingua 91
(1993), 175–184.
“Sentence Types and Clause Subordination”. The Encyclopedia of Language and
Linguistics ed. by R.E. Asher, R.E. Oxford: Pergamon Press (1993), Vol 7,
3845–3857.
“The Case Against a Future Tense in English”. Studies in Language 19 (1995),
399–446.
“The English Perfect as a Secondary Past Tense”. The Verb in Contemporary
English: Theory and Description. ed. by B. Aarts and C.F. Meyer. Cam-
bridge University Press (1995), 102–122.
“The Contrast between Interrogatives and Questions”. Journal of Linguistics 30
(1995), 411–439.
“What is an Appropriate Model of Grammar for Teachers?” Papers in Language
and Linguistics 1 (1996), 59–70. Centre for Language Teaching and
Research, University of Queensland.
“Commutation and English Infinitival to”. Glossa 17 (1997), 61–76.
“Complementation in English”. Encyclopaedic Dictionary of Linguistic Terminol-
ogy (in press) ed. by V. Prakasam.
CURRICULUM VITAE OF RODNEY DESMOND HUDDLESTON xxi
3. Review articles
“Predicate Complement Constructions in English”. [P.S. Rosenbaum, The Gram-
mar of English Predicate Complement Constructions.] Lingua 23 (1969),
241–273.
“Some Theoretical Issues in the Description of the English Verb”. [F.R. Palmer,
The English Verb.] Lingua 40 (1976), 331–383.
“On Classifying Anaphoric Relations”. [M.A.K. Halliday and R. Hasan, Cohe-
sion in English.] Lingua 45 (1978), 333–354.
“Constituency, Multi-Functionality and Grammaticalization in Halliday’s Func-
tional Grammar”. [M.A.K. Halliday, Introduction to Functional Grammar.]
Journal of Linguistics 24 (1988), 137–174.
Quirk, S. Greenbaum, G. Leech and J. Svartvik, A Comprehensive Grammar of
the English Language. Language 64 (1988), 345–354.
Reviews (selection)
S. Lamb. “Outline of Stratificational Grammar”. Lingua 22 (1969), 362–373.
M. Ehrman. “The Meanings of the Modals in Present-Day American English”.
Lingua 23 (1969), 165–176.
E. Roulet. “Syntaxe de la Proposition Nucléaire en Français Parlé”. Journal of
Linguistics 6 (1970), 277–280.
E. Buyssens. “Les Deux Aspectifs de la Conjugaison Anglaise au XXe Siècle”.
Lingua 27 (1971), 382–392.
W.A. Cook. “Introduction to Tagmemic Analysis”. Journal of Linguistics 7
(1971), 291–293.
P.A.M. Seuren. “Operators and Nucleus”. General Linguistics 12 (1972), 96–105.
D.G. Lockwood. “Introduction to Stratificational Linguistics”. Journal of Linguis-
tics 9 (1973), 349–356.
G. Sampson. “Stratificational Grammar”. Linguistics 122 (1974), 79–85.
D. Bolinger. “Degree words and that’s that”. Journal of Linguistics 11 (1975),
316–319.
R.M. Brend. “Advances in Tagmemics”. Linguistics 181 (1976), 86–104.
M. Berry. “An Introduction to Systemic Linguistics”. Vol 1. Language 53
(1977), 190–192.
H. Chr. Wekker. “The Expression of Future Time in Contemporary British Eng-
lish”. Journal of Linguistics 13 (1977), 357–358.
E. C. Davies. “On the Semantics of Syntax: Mood and Condition in English”.
Journal of Linguistics 17 (1981), 121–124.
xxii CURRICULUM VITAE OF RODNEY DESMOND HUDDLESTON
I.A. Melčuk and N.V. Pertsov. “Surface Syntax of English: A Formal Model
within the Meaning–Text Framework”. Linguistics 29 (1991), 156–161.
C.L. Baker. “English Syntax”. Journal of Linguistics 27 (1991), 239–250.
F. Stuurman. “Two Grammatical Models of Modern English: The Old and New
from A to Z”. Journal of Linguistics 28 (1992), 278–280.
R. Declerck. “A Comprehensive Descriptive Grammar of English”. Lingua 91
(1993), 349–354.
The semantics of English quantifiers*
Keith Allan
My purpose in this essay is to use just one metalanguage to provide a compre-

hensive account of number, countability, quantification, and (in)definiteness in
English and to show semantic interrelations between them. A secondary purpose
is to establish a clear link between morphological form and compositional
semantics. Some aspects of the semantics of quantification are dependent on clause
semantics, and that is the (tenuous) link my paper has to the theme of this book.
We need a semantic representation for countable and uncountable NPs (cf. Allan
1980) to capture the naive intuition that countables denote individual entities of
a similar kind, and uncountables an undifferentiated (comm)unity — contrast
knives and forks with cutlery. Compare (1) with (2).
(1) Harry bought some cream cakes
(2) Harry bought some cream cake.
Note that in other respects these sentences have similar logical implications:
there was someone called Harry, Harry bought something (with cream on top),
etc. Perhaps even more telling is a sequence such as
(3) Do have some cake; which one would you like to try?
Here the uncountable ‘some cake’ is followed by the countable ‘which one’
meaning “which [kind of] cake”. The pronoun one(s) indicates lexical/semantic
identity and not coreference; by implication, then, cake heading a countable NP
is lexically and semantically identical with cake heading an uncountable NP. Set
theory is useful to capture the semantics of countable NPs because its primitive
notion is that a set takes individuals as members; thus, the denotata of some
cakes can readily be conceived as members of the set K={x:x is a cake}.
Uncountables such as the NP some cake do not denote individuals, however; they
2 KEITH ALLAN
‘refer to entities as having a part − whole structure without singling out any
particular parts and without making any commitments concerning the existence
of minimal parts’ (Bunt 1985: 46). Exactly the contrary is presupposed for
countable NPs. What we need is a metalanguage that captures the properties of
both countables and uncountables using something very like set theory but whose
primitive notion is a part-of relation (meronym) for uncountables and a member-
ship relation for countables. Recall that, in set theory, the relation A ⊆ B, “A is
a subset of B”, is defined by all members of A being also members of B, and
not because A forms a part of B — even though it appears to be represented
meronymically in a Venn diagram such as Figure 1. What we need is meta-
language which uses the notion ‘part of’ as primitive, but defines anything which
is ‘the smallest part’ as the member of a set. There is such a metalanguage,
 , defined by Bunt (1976, 1979, 1985).
Figure 1. A ⊂ B
We will consider only essential characteristics of ensemble theory, details are

presented in Bunt’s works. Perhaps borrowing from conventions in mereology
(cf. Leśniewski 1988; Moravscik 1973; Ojeda 1991; Schein 1993), Bunt uses
lower case x, y, z as symbols for ensembles; the motivation is to differentiate
them formally from the usual notation for sets. An individual (particular) is
denoted by a unit set, and a set is just a special kind of ensemble (1985: 57); this
seems to reflect very well the way that languages express countability, judging
from (1–3). In ensemble theory the primitive relation for any ensembles x and y is
x⊆ y “x is part of y”.
The other relevant relations are:
∀ x,y[x=y iff (x ⊆ y) ∧ (y ⊆ x)]
∀ x,y[x ⊂ y iff (x ⊆ y) ∧ ¬ (y ⊆ x)]
x⊆ y ⊆z → x ⊆ z
THE SEMANTICS OF ENGLISH QUANTIFIERS 3
x∩ y is the  of ensembles x and y and is defined in terms of suben-

sembles (parts) and not members:
∀ a,x,y[a=x ∩ y iff (a ⊆x ∧ a ⊆ y) → ∀ z[(z ⊆ x ∧ z ⊆ y) → z ⊆ a]]
x∪ y is the  of ensembles x and y and is defined in terms of sub-
ensembles (and not members):
∀ b,x,y[b=x∪ y iff (x ⊆b ∧ y ⊆b) → ∀ z[z⊆ b → (z ⊆ x ∨ z ⊆ y)]]
A non-null ensemble with no proper parts is . If x is atomic, then x
has exactly one member (for ‘ ∃!z’ read “there is exactly one z”):
∃ !z[ ∀x,y[x={z} ∧ [x ⊆ y ≡ z ∈y]]
A merge (union) of atomic ensembles x ∪ y is a discrete ensemble or 
(1985: 71)
x:x ⊆ u is the     of u. The merge of all sub-
ensembles of u is uniquely equivalent to u. Thus,
∃ !u[∀ x,y[ x⊆ u=u iff (y ⊆x ≡ y ⊆u) ∧ x ∪ y=u]]
We earlier defined a set abstractly using the formulation
K={x:x is a cake} “K is the set of x’s that are cakes”
We define ensembles in a comparable way. For instance,
g= x:x is gold “g is the ensemble of parts that are gold”1
Other symbols used are: 1 means “one” (not “true”) and 0 means “zero” (not
“false”), “is very much greater than”, “is much greater than”, > “is greater
than”, ≥ “is greater than or equal to”, ≤ “is less than or equal to”, < “is less
than”, “is much less than”, “is very much less than” (the negative counter-
parts of these are, e.g. “not greater than”, / “not very-much-greater-than”,
etc.). A semiformal metalanguage is used as a tool to focus on the semantics of
quantifiers within English NPs and to simplify other aspects of clause structure
and meaning; consequently, and perhaps perilously, the properties of the formal
system are ignored.
We shall treat familiar English quantifiers as ‘generalized quantifiers’
(Barwise and Cooper 1981). A singular countable NP such as a cat can be
represented by the quantified subformula (4), where the notation derives from
May (1985; Neale 1990; Schein 1993).
(4) [a(n) x: Cx]
4 KEITH ALLAN
The quantifier a(n) binds x and (4) is read “an x such that C of x”; however, x
is an ensemble and not an individual (it is a(n) that determines x is an atomic
ensemble). We shall assume that in formulae such as [a(n) x: Cx], the predicate
C denotes an ensemble c . We define it c= x:x is cat , using the uncountable
predicate NP ‘is cat’ (undifferentiated catness) rather than the countable is a cat.
Many languages have morphological and/or syntagmatic marking of countables,
but no language systematically marks uncountables while leaving countables
unmarked. (Some languages, e.g. Sinhalese and noun class 6 in many Bantu
languages, have similar syntactic marking on uncountables and plurals —
suggesting that the prime motivation for countability is to identify the individual
from the mass.) Furthermore any noun lexeme can occur in environments which
are not countable, e.g. travel by car, car-obsessed, scissor-movement, etc. We
shall therefore assume that the default item for semantic purposes is the noun
lexeme unmarked for countability.
(4) sets a pattern we shall see throughout the analysis of quantifiers, but it
ignores the fact that the quantifier a(n) defines a singular countable NP. The
quantifier determines a   which takes the ensemble x as its
domain and outputs a quantity, x . We need to capture the fact that (4) tells us
x is atomic, i.e. a unit set such that x ≥1, i.e. there is at least one x, hence line
(ii) in (5) in which X Y symbolizes “X conventionally implicates Y”.2
(5) [a(n) x: Cx] (i)
[a(n) x: Cx] x =1 ∧ x ⊆ c (ii)
We make the general assumption that when the measure function yields a
number n ≥1, the NP is countable and the ensemble is a set. The conventional
implicature ( ) appears in line (ii) because the subformula to the left of it, [a(n)
x: Cx], is not a proposition but a  . On the lefthand side
in line (ii), ‘x’ is quantified by a(n) and this is indicated by ‘ x =1’ on the
righthand side. Obviously a(n) is being treated as synonymous with one, which
is preferred to a(n) in order to focus on the quantity. Note the existential claim
implicit in line (ii); of course, like any claim, this may turn out to be false; but
nonetheless, the claim — in the character of a presupposition — is made.
The sentence a cat is sick is represented by the formula in line (i) of (6),
where S= x:x is sick .
(6) [a(n) x: Cx](Sx) (i)

[a(n) x: Cx] x =1 ∧ x ⊆ c (ii)
x⊆ s (iii)
[a(n) x: Cx](Sx) → x =1 ∧ x ⊆c ∩ s (iv)
[a(n) x: Cx](Sx) ≡ c ∩ s ≥1 (v)
Line (i) is read “an x such that C of x, and S of (that) x (which is C)”; when the
parenthetical parts are included, this gloss captures the ‘lives on’ or ‘conser-
vativity’ property of quantifiers, cf. Barwise and Cooper (1981; Gamut 1991;
Cann 1993). Line (i) identifies a NP with an indefinite article, which makes x
denumerable (countable). The NP head denotes a cat (C) which is sick (S) at
some index. Line (ii) needs to be interpreted as purportedly true at a certain
location a within the world spoken of (mental space) wi at time ti: there are
occasions when this is crucial: e.g. in A cat is sick and another is badly injured
there are two cats in the world (mental space) and each NP refers to only one of
them, because the cats are in different locations.3 x⊆ s in line (iii) follows from
our interpretation of Sx in line (i); x remains bound by a(n). In line (iv) x ⊆ c ∩ s
“x is a subensemble of the overlap of c and s” follows from lines (ii-iii), (x ⊆ c)
∧ (x⊆ s). Truth functors appear in lines (iv) and (v) because the formula to their
left is a proposition. Line (v) follows from line (iv), and means “a cat is sick is
true if and only if the number of sick cats ≥ 1”. c ∩ s symbolizes the measure
function on the overlap (intersection) ensemble of cat and sick entities; its value
is >1 if x ⊂ c ∩ s “x is proper subset of c∩ s” but exactly 1 if x is an improper
subset of c ∩ s, i.e. x=c∩ s. For example, A cat is sick is true if (but not only if)
three cats are sick:
(7) Three cats are sick A cat is sick
It is a pragmatic fact that, in normal language use, quantifiers form a class of
scalar implicatives (cf. Horn 1972, 1989: 232, Levinson 1983: 134). Relevant
scales for present purposes include:
〈 all, most, many/much, some, a few/little, a(n)〉
〈 n≥ 6, 5, 4, 3, 2, 1〉
〈 no, not all, few/little〉 .
We need to distinguish the truth conditions from the standard implicatures of
quantified NPs, but both are required for their proper interpretation. Consequently,
(6) must be revised to include the standard implicature (symbolized X Y “X
conversationally implicates Y”) has c ∩ s =1.
6 KEITH ALLAN
(6) [a(n) x: Cx](Sx) ≡ c ∩ s ≥1 (v)

[a(n) x: Cx](Sx) c ∩ s =1 (vi)
Note that the implicature adopts the strongest, most specific interpretation from
line (iv) of (6), namely that x ⊆c ∩ s x=c∩ s.
There is a quantity scale, roughly like that in Figure 2.
many several ≥ a few few a(n) / one

all > most > > > > ? > no
much a little little a (tiny) bit of
Figure 2. The quantity scale
(Possible alternatives to a (tiny) bit of are very little and almost no.) Any quanti-
fier logically implies all quantifiers downscale to its right except for no; e.g.
all N many N one N / no N.
(X / Y symbolizes “X does not conventionally implicate Y”). Consequently,
scalar implicatures work in a predictable way: the selection of any quantifier
standardly implicates that the quantifier upscale to its left (and any quantifier to
the left of that also) is inappropriate. The reason that some does not appear in
Figure 2 is that it simply doesn’t fit neatly anywhere; it is in fact compatible
with any quantifier to the right of all and to the left of no.
Consider the zero surface quantifier of the italicized NPs in (8–10).
(8) We bought vegetables and fruit and milk.
(9) Cherries are delicious.
(10) Salmon is very tasty.
When there is no surface quantifier/determiner and the head noun is plural, the
NP is countable (cf. Allan 1980). Where the NP has no surface quantifier/
determiner and the head noun is in the N0 (citation) form, the NP is uncountable.
We represent the semantics of the plural noun
[Q x: Fx]
for any predicate F, and of the N0 form as
[ ∅ Q x: Fx]
In (9–10), but not in (8), the NPs are generic. The generic is an inference based
not on the semantics of the NP, but on the fact that the proposition is ‘law-like’;
the superficial zero quantifier is semantically consistent with both interpretations,

and we should ignore the difference between generic and nongeneric NPs,
though not between generic and nongeneric sentences.
(11) P 
[Q x: Fx](Gx) (i)
[Q x: Fx] x >1 ∧ x ⊆ f (ii)
[Q x: Fx](Gx) → x >1 ∧ x ⊆ f∩ g (iii)
[Q x: Fx](Gx) ≡ f ∩ g >1 (iv)
Because x is quantified by Q, line (ii) says that the NP denotes more than one
(thing that is) F at location a in wi at time ti — a consequence of the plural
quantifier; therefore, ensemble x denotes a countable NP. The existential
presupposition carries through to line (iv), although it can be overridden by
various mechanisms, cf. (18–19). Line (iii) x ⊆ f∩ g “x is a subensemble of the
overlap of f and g” follows from lines (i-ii), x ⊆f ∧ x ⊆ g. Line (v) sums up,
saying that at a in wi at ti “things which are F are G”, cf. (9).
(12) Z 
[ ∅ Q x: Fx](Gx) (i)
[ ∅ Q x: Fx] x ≠0 ∧ x⊆f (ii)
[ ∅ Q x: Fx](Gx) → f ≠ 0 ∧ x ⊆ f∩ g (iii)
[ ∅ Q x: Fx](Gx) ≡ f∩ g ≠0 (iv)
Line (ii) says that there is an existential presupposition: at this location a in wi
at ti, x is nonnull. The existential presupposition carries forward to line (iv).
There is nothing in (12) to tell us that the overlap of f and g, f ∩ g, denotes
countables and the default interpretation is that the NP is uncountable, cf. (10).
(12) is the proper representation for the N0 form which is found not only in
uncountables but also in singular countables (their countability is determined by
other means); compare (13) with (6).
(13) [a(n)/one y: y ⊆ [ ∅ Q x: Fx]](Gy) (i)
[ ∅ Q x: Fx] x ≠0 ∧ x⊆f (ii)
[a(n)/one y: y ⊆ [ ∅ Q x: Fx]] y ≥1 ∧ y ⊆x ⊆ f, ∴ y ⊆ f (iii)
[a(n)/one y: y ⊆ [ ∅ Q x: Fx]] y =1 (iv)
[a(n)/one y: y ⊆ [ ∅ Q x: Fx]](Gy) →
y ≥ 1 ∧ (y⊆ f ∧ y ⊆ g), ∴ y ⊆ f∩ g (v)
[a(n)/one y: y ⊆ [ ∅ Q x: Fx]](Gy) ≡ f ∩ g ≥ 1 (vi)
[a(n)/one y: y ⊆ [ ∅ Q x: Fx]](Gy) f∩ g =1 (vii)
8 KEITH ALLAN
Line (ii) recognizes the effect of the N0 NP head; it identifies the NP as nonnull.
Line (iii) illustrates the    : leftmost quanti-
fiers impose interpretations that may override the interpretations otherwise
appropriate to quantifiers in their scope — i.e. those that the left-quantifier
c-commands;4 scope is indicated by the [ to the left of Q and the paired ] to the
right. (iii) shows that a(n) determines a countable NP: y is a discrete ensemble
consisting of at least one atomic subensemble; y is a subensemble of x which is
a subensemble of f, so y is a subensemble of f. Because y is denumerable, so are
f and g. The implicature in (iv) is properly an implicature deriving from the
quantifier; naturally it carries through to (vii). Note the similarity between (v-vii)
and lines (iv-vi) of (6).
Now consider a plural NP with a numerical quantifier. Let m= x:x is man
in this room and p= x:x is secret policeman ; the simplified representation m
for ‘men in this room’ conceals the fact that ‘in this room’ identifies the relevant
location a for the NP referents.
(14) Three men in this room are secret policemen.
(14 ′ ) [three y: y ⊆ [Q x: Mx]](Py) (i)
[Q x: Mx] x >1 ∧ x ⊆m (ii)
[three y: y ⊆ [Q x: Mx]] y ≥3 ∧ y ⊆ x ⊆m, ∴ y ⊆ m (iii)
[three y: y ⊆ [Q x: Mx]] y =3 (iv)
[three y: y ⊆ [Q x: Mx]](Py) → y ≥3 ∧ y ⊆m ∩ p (v)
[three y: y ⊆ [Q x: Mx]](Py) ≡ m ∩ p ≥3 (vi)
[three y: y ⊆ [Q x: Mx]](Py) m ∩ p =3 (vii)
Line (ii) indicates that there is more than one (thing that is) M, thus determining
that the NP is countable; in line (iii), m ≥3 more specifically identifies at least
three subensembles which denote things which are M; line (iv) implicates that
there are exactly three. In line (v), y in the formula y ⊆ m ∩ p is quantified by
three, this is indicated by the preceding y ≥3.
The left quantifier control condition is what makes collectivized NP such as
three giraffe semantically plural (cf. Allan 1986, fc). It also yields the preferred
interpretation for an ungrammatical NP such as one men singular. It is worthy of
note that, in processing spoken or written language, the leftmost quantifier is
encountered before the elements in its scope; hence, a processing model would
not work from the inside out in the manner of this compositional semantics. For
that reason we revise it to mirror language processing, and to establish left quantifier
control from the beginning. Thus we revise (13) to (13′ ), and (14 ′ ) to (14″ ).
(13 ′ ) [a(n)/one y: y ⊆ [ ∅ Q x: Fx]](Gy) (i)

[a(n)/one y: y ⊆ [ ∅ Q x: Fx]] y ≥1 (ii)
[a(n)/one y: y ⊆ [ ∅ Q x: Fx]] y =1 (iii)
y⊆ [ ∅ Q x: Fx] y⊆ x⊆ f, ∴ y ⊆ f (iv)
[a(n)/one y: y ⊆ [ ∅ Q x: Fx]] y ≥1 ∧ y ⊆f (v)
[a(n)/one y: y ⊆ [ ∅ Q x: Fx]](Gy) → (y ⊆f ∧ y ⊆ g), ∴ y⊆ f ∩ g (vi)
[a(n)/one y: y ⊆ [ ∅ Q x: Fx]](Gy) ≡ f ∩ g ≥ 1 (vii)
[a(n)/one y: y ⊆ [ ∅ Q x: Fx]](Gy) f∩ g =1 (viii)
The effect of the leftmost quantifier is evident in lines (ii) and (iii); the existen-
tial import of zero quantification over x is redundant and therefore omitted from
line (iv). Line (vii) follows from lines (ii) and (vi):
( y ≥1 ∧ y ⊆ f ∩ g) → f ∩ g ≥ 1
Obviously line (viii) is validated by (iii), (vi), and (vii).
(14 ″ ) [three y: y ⊆ [Q x: Mx]](Py) (i)
[three y: y ⊆ [Q x: Mx]] y ≥3 (ii)
[three y: y ⊆ [Q x: Mx]] y =3 (iii)
y⊆ [Q x: Mx] y⊆ x⊆ m, ∴ y⊆ m (iv)
[three y: y ⊆ [Q x: Mx]] y ≥3 ∧ y ⊆ m (v)
[three y: y ⊆ [Q x: Mx]](Py) → y ≥3 ∧ y ⊆m ∩ p (vi)
[three y: y ⊆ [Q x: Mx]](Py) ≡ m ∩ p ≥3 (vii)
[three y: y ⊆ [Q x: Mx]](Py) m ∩ p =3 (viii)
Let’s now turn to universal quantifiers. Whereas all ranges over an ensemble that
may be countable or uncountable, each and every identify the NP as countable
(they differ in that each involves a distributive function; each N can be roughly
glossed “distributed among every N”, but will not be further discussed here; cf.
Allan fc). [all x: Fx] identifies ensemble f as the     -
 such that ∃ !f[ x:x ⊆ f=f], from which it follows that f is the quantity of
all f’s subensembles. For example,
(15) All gold is valuable.
(15 ′ ) [all y: y ⊆[ ∅ Q x: Gx]](Vy) (i)
[all y: y ⊆[ ∅ Q x: Gx]] y:y ⊆ [ ∅ Q x: Gx]]=[ ∅ Q x: Gx]] (ii)
[ ∅ Q x: Gx]] x ≠0 ∧ x ⊆ g, ∴ g ≠0 (iii)
y⊆ [ ∅ Q x: Gx] y ⊆x ⊆ g, ∴ y ⊆ g ∧ g ≠ 0 (iv)
10 KEITH ALLAN
[all y: y ⊆[ ∅ Q x: Gx]] ∃!g[ y:y ⊆ g=g] ∧ g ≠0 (v)

[all y: y ⊆[ ∅ Q x: Gx]](Vy) →
∃ !g[g= y:y ⊆g] ∧ y ⊆ v, ∴ ∃ !g[ y:y ⊆ g∩ v=g] (vi)
[all y: y ⊆[ ∅ Q x: Gx]](Vy) ≡ ∃ !g[ y:y ⊆ g∩ v = g] ∧ g ≠0 (vii)
Line (ii) establishes that the quantified NP denotes the merge of all subensembles, y,
of gold, g. Line (iii) establishes the ensemble g and quantifies it; in effect, it
translates gold. Line (iv) identifies y as a subensemble of g. Line (v) brings together
lines (ii), (iv) and (iii). Formula ∃ !g[ y:y⊆ g=g] in line (v) means “there is exactly
one ensemble g such that the merge of all subensembles of g is identical to g”
(i.e. everything that is y is G or, alternatively, [all y: y ⊆ [∅Q x: Gx]] denotes
g ); this is why (15) is virtually synonymous with Gold is valuable, the
difference being that the latter refers to the ensemble of gold whereas (15)
focuses on the quantity by referring to all the subensembles whose merge is this
ensemble. In line (vi) the conclusion ∃ !g[ y:y ⊆ g∩ v=g] is justified by the fact
that y in the subformula Vy (line (i)) is quantified by all, i.e. everything that is
y is V; and we have established in line (v) that [all y: y ⊆ [∅ Q x: Gx]] denotes
g , so it follows that g (“everything that is y”) is a subensemble of v. There is
nothing in (15 ′ ) to show that the NP is countable, therefore, by default it is
uncountable.
Next consider all quantifying a countable NP.
(16) All men in this room are secret policemen.
(16 ′ ) [all y: y ⊆[Q x: Mx]](Py) (i)
[all y: y ⊆[Q x: Mx]] y:y ⊆[Q x: Mx]]=[Q x: Mx]] (ii)
[Q x: Mx] x >1 ∧ x ⊆m, ∴ m >1 (iii)
y⊆ [Q x: Mx] y⊆ x⊆ m, ∴ y⊆ m ∧ m >1 (iv)
[all y: y ⊆[Q x: Mx]] ∃ !m[ y:y ⊆ m=m] ∧ m >1 (v)
[all y: y ⊆[Q x: Mx]](Py) ≡ ∃ !m[ y:y ⊆ m∩ p = m] ∧ m >1(vi)
Line (iii) establishes the ensemble m and quantifies it; in effect, it translates
men. Line (iv) identifies y as a subensemble of m. Line (v) brings together lines
(ii), (iv) and (iii). Line (vi) simplifies on lines (vi) and (vii) of (15 ′ ). Countabili-
ty is determined in (16 ′ ) by the plural quantifier; like most universal quantifiers,
all ignores countability (exceptions are every and each). Hence line (vi) identifies
all subensembles of a discrete ensemble m and only by inference is this equiva-
lent to all members of a set M.
Every differs from all in two ways: (a) it identifies the NP as countable and
sentences like Take a pill every four hours suggest it quantifies over discrete
ensembles (sets) rather than individuals; and (b) it selects objects of roughly
identical quantity5 so that
[every x: Fx] ∀ x[Fx ≡ ∀ y,z[(y⊆ x ∧ z ⊆ x) → y ≈ z ]]
All and every are only interchangeable when these two conditions obtain. What
they have in common in [Qx:Fx](Gx) is that both imply f = f ∩ g ≠ 0. In
much of the discussion which follows, the common semantics of all and every
is more relevant than the difference between them.
(17) Every man in this room is a secret policeman.
(17 ′ ) [every y: y ⊆ [ ∅ Q x: Mx]](Py) (i)
[every y: y ⊆ [ ∅ Q x: Mx]]
∀ y,w,z[ y ≥ 1 ∧ (w ⊆y ∧ z ⊆ y) → w ≈ z ]] (ii)
y⊆ [ ∅ Q x: Mx] y⊆ m (iii)
[every y: y ⊆ [ ∅ Q x: Mx]] y ≥1 ∧
∀ y[y ⊆ m ≡ ∀ w,z[(w⊆ y ∧ z ⊆ y) → w ≈ z ]] (iv)
[every y: y ⊆ [ ∅ Q x: Mx]](Py) → y ⊆ m ∩ p (v)
[every y: y ⊆ [ ∅ Q x: Mx]](Py) ≡ m = m ∩ p ≥1 ∧
∀ y[y ⊆ m ≡ ∀ w,z[(w⊆ y ∧ z ⊆ y) → w ≈ z ]] (vi)
Line (iv) follows from lines (ii) and (iii). In lines (iv) and (vi), m is a constant.
Turning next to no:
(18) [no y: y ⊆ [ ∅ Q x: Fx]](Gy) (i)
[no y: y ⊆ [ ∅ Q x: Fx]] ¬ (y ⊆[ ∅ Q x: Fx]) (ii)
y⊆ [ ∅ Q x: Fx] x ≠0 ∧ y ⊆ x⊆ f, ∴ y ⊆f ∧ y ≠ 0 (iii)
[no y: y ⊆ [ ∅ Q x: Fx]] y =0 ∨ y f (iv)
[no y: y ⊆ [ ∅ Q x: Fx]](Gy) → y =0 ∨ y f ∩ g (v)
[no y: y ⊆ [ ∅ Q x: Fx]](Gy) ≡ f ∩ g =0 (vi)
Line (ii) says that the quantifier no of line (i), in effect, negates line (iii), i.e.
negates y ⊆ f ∧ y ≠ 0. The effect of this is spelled out as the conventional
implicature in line (iv): y is either null (there is nothing that is F) or, whether or
not y is null, it is no part of ensemble f. Line (v) follows logically: if there is
nothing that is F, then there is nothing that is both F and G. The latter is the
crucial disjunct, hence line (vi). Like all, no ignores countability.
12 KEITH ALLAN
Compare (18) with (19), not an/one F is G.

(19) [not z: z ⊆ [a(n)/one y: y ⊆ [∅ Q x: Fx]]](Gz) (i)
[not z: z ⊆ [a(n)/one y: y ⊆ [∅ Q x: Fx]]]
¬ (z⊆ [a(n)/one y: y ⊆ [ ∅ Q x: Fx]]) (ii)
[a(n)/one y: y ⊆ [ ∅ Q x: Fx]] y ≥1 ∧ y ⊆f (iii)
[not z: z ⊆ [a(n)/one y: y ⊆ [∅ Q x: Fx]]]
¬ (z⊆ ( y ≥1 and y ⊆f)), ∴ z 1 ∨ z f (iv)
[not z: z⊆[a(n)/one y: y⊆[∅ Q x: Fx]]](Gz) → z 1 ∨ z f∩g (v)
[not z: z⊆[a(n)/one y: y⊆[∅ Q x: Fx]]](Gz) ≡ f∩g 1 (vi)
[not z: z⊆[a(n)/one y: y⊆[∅ Q x: Fx]]](Gz) f∩g =0 (i.e.<1) (vii)
Lines (ii-iv) may be compared with those in (18). In line (iv) z 1 means that
the quantity of z is either not equal to 1 or it is less than one; z f means “there
are no z’s that are F”. In line (vii) the implicature is equivalent to the semantics
of no. Note that what is negated by not is the concatenated quantifier a(n)/one,
i.e. it has narrow scope. If this is the semantics for not one, why is Not one but
two Fs are G meaningful? This has the hallmark of radical presupposition
negation in its intonation and in the presence of the adversative but; what is
being denied is the implicature of One F is G: f ∩ g =1. Note that Not one but
two Fs are G only entails not one F is G in the dispreferred sense f ∩ g ≠1 —
the preferred sense being f ∩ g 1 i.e. f∩ g <1 cf. line (vii) in (19).
When all ranges over a numeral quantifier Qn, it limits the quantity from “at
least Qn” to “exactly Qn”, cf.
(20) All three men in this room are secret policemen.
(20 ′ ) [all z: z ⊆ [three y: y ⊆ [Q x: Mx]]](Pz) (i)
[all z: z ⊆ [three y: y ⊆ [Q x: Mx]]] z:z ⊆[three y:
y⊆ [Q x: Mx]]]=[three y: y⊆ [Q x: Mx]]] (ii)
[three y: y ⊆ [Q x: Mx]] y ≥3 ∧ y ⊆ m, ∴ m ≥3 (iii)
z ⊆ [three y: y ⊆ [Q x: Mx]] m ≥3 ∧ z ⊆ m (iv)
[all z: z ⊆ [three y: y ⊆ [Q x: Mx]]]
∃ !m[ z:z⊆ m=m] ∧ m =3 (v)
[all z: z ⊆ [three y: y ⊆ [Q x: Mx]]](Pz) ≡
∃ !m[ z:z⊆ m∩ p = m] ∧ m =3 (vi)
Line (v) recognizes that if all three men is the merge of all subensembles of three
men, there is no possibility that there are more than three; thus the relation ≥
“greater than or equal to” is reduced to the stronger relation = “(exactly) equal
to”. This is a function of all universal quantifiers, and correlates with their
structural location as leftmost quantifiers, thus c-commanding and controlling the
quantifiers within their scope.
At this point we introduce the notion of  as an entailment of
propositions containing quantifiers. One kind is NP monotone, also called ‘left
monotone’ (Chierchia and McConnell-Ginet 1990; Gamut 1991), ‘subject
monotone’ (Cann 1993); the other is Pred monotone, also called ‘right mono-
tone’. Each of these can be upward/increasing, symbolized ↑ , or downward/
decreasing, symbolized ↓ , and sometimes neither. Consider some examples to
see the mnemonic value of ↑ and ↓ . In (21) the second proposition is a valid
inference from the first; this is not so in (22).
(21) All linguists are Chomskyites All bald linguists are Chomskyites.
Herod killed all the newborn babies Herod killed all the blind
newborn babies
All women and children are at risk All children are at risk.
(22) All bald linguists are Chomskyites All linguists are Chomskyites.
Herod killed all the blind newborn babies Herod killed all the
newborn babies
All children are at risk All women and children are at risk.
Note that in (21–22) the denotation of ‘bald linguists’ is a subset of the denota-
tion of ‘linguists’, ‘blind newborn babies’ is a subset of ‘newborn babies’, and
‘children’ is a subset of ‘women and children’. In (21) the entailment goes from
the superset to the subset, i.e. downward, and (22) shows it cannot go in the
other direction. All is therefore NP ↓ monotonic (or simply, NP ↓ ).
Now consider
(23) All magpies sing All magpies sing sweetly.
(24) All magpies sing sweetly All magpies sing.
Those things that sing sweetly are a subset of the things that sing. In (24) the
entailment goes from subset to superset, so we say that all is Pred ↑ monotonic.
Every and each have the same monotonic properties as all, but no is NP ↓
and also Pred ↓, cf. (25).
14 KEITH ALLAN
(25) No linguists are elephant poachers No bearded linguists

are elephant poachers NP ↓
No bearded linguists are elephant poachers No linguists are ele-
phant poachers
No magpie snores No magpie snores loudly Pred ↓
No magpie snores loudly No magpie snores
Only radical negation (contrastive corrective negation) of no and each is possible,
e.g. Not no but three students failed the course! and Not each but only two students
got a test paper. However, no and each do have : when
Q1= ¬ Q2
then Q1 and Q2 are contradictories. The contradictory of no is at least some/one. Q¬
negation on no yields monotonicity like its contrary all/every, i.e. NP↓ Pred↑:
(26) No club member doesn’t smoke cigars No female club member
doesn’t smoke
Negating each negates distribution to every N to yield not every (or not all when
this is synonymous with not every); cf.
(27) ¬ [Each child] has an apple Not every child has an apple
¬ [Each participant] stood up in turn Not all participants stood up
in turn.
(28) [not all/every x: Fx](Gx) ≡ f ≠ f ∩ g ∨ f =0
[not all/every x: Fx](Gx) f∩g ≠0 ∧ f > f∩g
(Full details are omitted from (28) and many of the following examples.) The
semantics of [not all/every x: Fx](Gx) can be glossed “some F(s) is(are) G or,
disjunctively, no F(s) is(are) G” because
[no x: Fx](Gx) → [not all/every x: Fx](Gx)
The existential implicature in (28) is a function of the fact that, for nonexistence,
[no x: Fx](Gx) would be preferred under the cooperative principle because it is
a stronger, more accurate, statement of the facts. The rest of the implicature
accounts for the fact that not all/every is closer to most than lesser quantities.
The fact that
[not all/every x: Fx](Gx) f∩g ≠0
(cf. 28) is exactly why [some x: Fx](Gx) can be inferred from it. The quantifier
some is nonspecific as to quantity between all and nothing, but it has the
standard implicature “less/fewer than much/many” to which it is otherwise
semantically similar.
(29) [some x: Fx](Gx) (i)
[some x: Fx] f ≠0 ∧ x ⊆ f (ii)
[some x: Fx](Gx) ≡ f∩ g ≠0 (iii)
[some x: Fx](Gx) f∩ g < f − f∩ g ∧ f∩ g f − f∩ g (iv)
We leave aside how it is we know that some man is countable and some gold is
uncountable, see the discussion of (61–65) below. Line (ii) represents the existential
presupposition in the NP, which carries through to line (iii). f− f ∩ g in lines (iii)
and (iv) is “the quantity of F which is not G”; in countables “the number of Fs
that don’t G”.
There is a difference in monotonicity between not all/every and some: not
all/every are NP ↑ Pred ↓, whereas some is NP ↑ Pred ↑.
(30) Not all (the) boys drive carelessly Not all (the) boys drive
Some boys drove carelessly Some boys drove Pred ↑
¬ Q negation is not possible on some–presumably blocked by its contradictory, no.

The subformula f ≠0 in the semantics for some really means “Speaker
has reason to believe it is the case that the referent exists”; the truth of this can
be modified by various predications that hedge or otherwise vary the evidential
status, e.g. I think there is someone at the door; There might be some eggs in the
fridge; Possibly there are some eggs in the fridge; etc. This might be compared
with a partial semantics for any–often said to be the negative counterpart of some
because, e.g. I can see somebody there when negated becomes I can’t see
anybody there. Also when questioned: Can you see anybody there? and when
hypothesized: If anyone calls get their number. There is also Compare any two
lemons, … “no matter which two”. Thus when Speaker has reason to believe that
the referent does not exist or might not exist or might exist at some hypothetical
index, then — under certain syntactic conditions — any is used. It is this
inability to guarantee existentiality for referents that characterizes the use of any.
Quine (1960: 138ff) argued on the basis of examples comparable with (31) and
(32) that the universal quantifiers all and every are captured by ∀ with narrow
scope, and any by ∀ with wide scope.
16 KEITH ALLAN
(31) Ed didn’t kiss everyone. ¬ ∀ x (Ed kiss x)

(32) Ed didn’t kiss anyone. ∀ x ¬ (Ed kiss x)
Many scholars agree, though Horn (1972) has an alternative proposal using the
existential quantifier ∃ for any on the basis of its relationship with some. The
link between any and the universal seems to be intuitively correct for some
contexts, and the link with some in others. For instance, (33) is synonymous with
one interpretation of (34), i.e. when Speaker does not know that there is an
extension for ‘traitors’.
(33) Any traitors will be shot.
(34) All traitors will be shot.
However (34) with all can be used when the extension of ‘traitors’ is known,
whereas (33) with any cannot. Quine’s analysis is correct for negative contexts, i.e.
(35) ¬ ([any x: Fx](Gx)) → f = f∩ g =0
but for interrogative and hypothetical contexts the meaning is more like (36) —
the interrogative mood operator is ? and the optative-subjunctive operator is ¡.
(36) ?([any x: Fx](Gx))
→ f∩g ≠0
¡([any x: Fx](Gx))
Hypotheticals and expressions like Compare any two lemons, invoke a world, wh,
accessible from the current world, wi, where — loosely speaking — Speaker
imagines a possible world in which f ∩ g ≠0 (traitors are shot or two lemons
are compared, cf. Allan fc). The fact that Speaker cannot use any N in a positive
clause when the NP is either known to have extension or when it is known not
to, is what distinguishes any from all, every and some on the one hand, and no/
not one on the other. In the scope of a negative a(n) is, appropriately, NP ↓ Pred ↓
like no, but there are problems with NP monotonicity in negated conditionals
with if any. In fact from now on we ignore NP monotonicity because it is, by
and large, problematic and unsystematic (cf. Allan fc).
A question Is/Does any F G?  extension for f ∩ g, whereas the ques-
tions How much F Gs? or How many Fs G?  extension for f ∩ g. The
semantics of much and many are similar to the semantics for some, but the
standard implicatures are different. (37) refers to some quantity APPRO , a
nonstandard abbreviation for “judged appropriate” or “that might reasonably be
expected” for the reference set.
(37) [many/much x: Fx](Gx) ≡ f∩ g > APPRO

[many/much x: Fx](Gx) f ∩ g / f − f∩ g
The formula f ∩ g / f − f∩ g can be glossed “the quantity of F which is G

is not much-greater than the quantity of F which is not G”. Many and much are
Pred ↑ monotonic as are a little/few and some. Many/much are complements of
a few/little compare (37) with (38–39):
(38) [a few/little x: Fx](Gx) ≡ f ∩ g < f− f ∩ g ∧ f ∩ g / f − f∩ g
[a few/little x: Fx](Gx) f∩ g f− f ∩ g
There is something strange about a little/few: the monotonicity of their apparent

negations, not a little/few is Pred ↑ , identical to the positives.
(39) [not a little/few x: Fx](Gx) ≡
¬ ( f∩ g < f − f∩ g ∧ f∩ g / f − f∩ g )
[not a little/few x: Fx](Gx) f∩ g > APPRO
The semantics of the negative (39) is ambiguous; the upscale implicature is not
and is identical to the semantics of many/much in (37). Monotonicities are
upward in (40), and downward in the Q ¬ sentences of (41).
(40) Much of the hour (40 minutes, in fact) was taken up with a dispute
over procedure Not a little of the hour (40 minutes, in fact) was
taken up with a dispute over procedure
(41) Not a few women don’t smoke cigars Many women don’t smoke cigars
The situation with not much/many and a little/few appears to be the same,
compare (38) with (42).
(42) [not much/many x: Fx](Gx) ≡ ¬( f∩g > APPRO )
[not much/many x: Fx](Gx) f∩g < f−f∩g ∧ f∩g / f−f∩g
Again, the semantics of the negative is ambiguous, the implicature of the positive
in (38) is not. For a ¬ Q negated upscale quantifier like (42) the implicature is
downscale, cf. (39). But, the monotonicities of (38) and (42) do not coincide: not
much/many is Pred ↓ whereas a little/few is the reverse, Pred ↑. Hence, the
difference between the positive connotation of the upward entailing (43) and the
negative downward entailing (44):
(43) A few people turned up to the meeting (ten out of forty) Pred ↑
(44) Not many people turned up to the meeting (ten out of forty) Pred ↓
18 KEITH ALLAN
Few/little (Pred ↓ ) are the converses of most (Pred↑ ) as is obvious from the
comparison of of (45) and (46).
(45) [most x: Fx](Gx) ≡ f ∩ g f− f ∩ g
[most x: Fx](Gx) f∩ g / f − f∩ g
(46) [little/few x: Fx](Gx) ≡ f∩ g f− f ∩ g

[little/few x: Fx](Gx) f∩ g / f− f ∩ g
It has often been assumed that few/little are the converses of many/much. Against
this are their meanings, and the symmetry of the quantity scale (Fig. 2), which
pairs as converses most with few/little and many/much with a few/little. There is
also the fact that ¬ Q negation of most and few/little is not possible in the forms
*not most/few/little; whereas ¬ Q negation of many/much and a few/little is
normal. On the other hand, there is a good deal of overlap between most and
much/many upscale and between a few/little and few/little downscale; in fact most
can be regarded as constrained to the upper end of much/many which range way
downscale from most; conversely, few/little are restricted to the lower end of the
range quantified by a few/little. In consequence there is a converse relation
between most/much/many upscale and (a) few/little downscale as evidenced in
(47). The (Pred) monotonicity of the quantifiers is given.
(47) a. Most/many ↑ of my friends are professionals, but/?and a few ↑
have blue-collar jobs.
b. A few ↑ of my friends are professionals, but/??and most/many
↑ have blue-collar jobs.
c. Most/many ↑ of my friends are professionals and/*but few ↓
d. Few ↓ of my friends are professionals, and/*but most/many ↑
The differences in conjunction in (47) result from the fact that when partitioning the
ensemble of friends into nonoverlapping subensembles, a conjunction of quantifiers
of different monotonicity requires and (47c–d, 48), whereas identical Pred mono-
tonicity prefers but, though and is possible when going downscale, cf. (47a–b, 49).
(48) Not all ↓ of my friends are professionals and/*but a few ↑ have
blue-collar jobs.
Some ↑ of my friends are professionals and/*but not all ↓ have blue-
collar jobs.
All ↑ of my friends are professionals and/*but none ↓ have blue-

collar jobs.
(49) Not all ↓ my friends are professionals but/and few/none ↓ have blue-
collar jobs.
Not many ↓ of my friends are professionals but/and few/none ↓ have
blue-collar jobs.
Some ↑ of my friends are professionals but/??and many/most ↑ have
blue-collar jobs.
In (50) the subensembles of friends overlap.
(50) Some ↑ of my friends are professionals and/*but many ↑ are women.
Let’s now summarize the findings on monotonicity and see what relations they
reveal between quantifiers. NP monotonicity is a quagmire as can be seen from
the following:
NP ↑ some/a(n)/one/two, some/a(n)/one/two…not, not all/every, all/every…not, a few/

little, (not) a few/little…not, counterfactual if not any /if any…not.
Doubtful: ?many/much, ?not a few/little
NP ↓ no, not a(n)/one/two, all/every, no…not, any. Noncounterfactual if not any /if
any…not.
Doubtful: ?not many/much, ?few/little, ?few/little…not
NP nonmonotonic most…not, much/many…not, not a few/little…not
The conclusion must surely be that NP monotonicity is of no great interest to

systematic semantic inquiry. Before we abandon it entirely, there some interest-
ing points worth noting. First of all though, consider the traditional ‘square of
opposition’, cf. Aristotle On Interpretation, L. Apuleius (2nd c. 1968: 261ff),
Martianus Capella (5th c., 1969: 193ff), Boethius 6th c., 1894:319ff, 468ff,
775ff), Gamut (1991, II:238). The vertices are named from vowels in the Latin
Affffirmo ‘I affirm’ and nEgO ‘I deny’.
20 KEITH ALLAN
Affirmations Negations
Universals A E
A–E are contraries: ¬ (Q1 ∧ Q2)
A–O and I–E are contradictories: ¬ Q1=Q2
A–I and E–O are duals: Q1= ¬ Q2 ¬
I–O are subcontraries: ¬ ( ¬ Q1 ∧ ¬ Q2)
Particulars I O
E.g. A: [all x: Fx]F ≈ ∀ xF E: [no x: Fx]F ≈ ¬ ∃ xF
I: [a(n) x: Fx]F ≈ ∃ xF O: [not y: y ⊆ [all x: Fx]]F ≈ ¬ ∀ xF
Figure 3. The traditional square of opposition
If we eliminate doubtful cases of NP monotonicity, the two following boxes

show that the top line lists the contradictories of the bottom line, i.e. vertices
A–O and I–E from Figure 3:
A NP ↓ Pred ↑ all/every, no…not

O NP ↑ Pred ↓ not all/every, all/every…not, some/a(n)/one/two…not, a few/little
…not
I NP ↑ Pred ↑ some/a(n)/one/two, a few/little
E NP ↓ Pred ↓ no, not a(n)/one/two, ?any
We now turn Pred monotonicity, which is comparatively systematic.
Pred ↑ all/every, no…not, few/little…not, most, many/much, (not) a few/little,

some/a(n)/one/two, if not any, if any…not
Pred ↓ not all/every, all/every…not, no, any, most…not, not much/many, much/
many…not, not a few…not, few/little, not a(n)/one/two, some/a(n)/one/two…not
A comparison of the top and bottom lines here shows that Pred ↑ are all affirma-
tive (despite a couple of negative forms), cf. the A–I vertices of Figure 3; in
contrast, the Pred ↓ line is negative, corresponding to the E–O axis of the
traditional square of opposition. These vertices identify , something quite
different from dual number, and defined as follows:
If the dual of Q is Q*, then Q= ¬ Q*¬
If Q is ↑ (since we shall no longer be concerned with NP monotonicity, ↑=Pred ↑,

and ↓ =Pred ↓ from now on) then Q* is ↑ , if Q is ↓ then Q* is ↓ . Some
examples in which the underlined quantifiers are duals of one another:
(51) Not all cats are not sick Some/a(n)/one/two cat(s) is/are sick
cf. ¬ ∀ x¬ F ≡ ∃ xF
Not anyone [= ¬ someone] doesn’t know Harry Everyone knows
Harry cf. ¬ ∃ x¬ F ≡ ∀xF
It is not the case that not all cats are not sick No cat is sick
cf. ¬ ¬ ∀ x¬ F ≡ ¬ ∃ xF
It is not the case that no cat is not sick Not all cats are sick
cf. ¬ ¬ ∃x ¬ F ≡ ¬ ∀ xF
Not a few people don’t smoke A few people smoke
Summarizing, with monotonicity indicated on the right:
some/a(n)/one/two = ¬ all/every ¬ ↑
all/every = ¬ some/any/a(n)/one/two ¬ ↑
no = ¬ not all/every ¬ ↓
not all/every = ¬ no ¬ ↓
a few/little = ¬ a few/little ¬ ↑ (a few/little is its own dual)
NP conjunctions of different Pred monotonicity prefer but to and for the most part
in contrast to the facts concerning clause conjunction exemplified in (47–50).
(52) ?Every ↑ child and no ↓ adult died.
–Every child but no adult died.
(53) ?We tested all ↑ the cats and not all ↓ the dogs for rabies.
–We tested all the cats but not all the dogs for rabies.
(54) ?Not a few ↑ men and few ↓ women play rugby.
–Not a few men but few women play rugby
[Stylistically awkward]
(55) Many ↑ men and few ↓ women play rugby.
–Many men but few women play rugby.
(56) I’ll have some ↑ anchovies and not much ↓ lettuce, thanks.
–I’ll have some anchovies but not much lettuce, thanks.
(57) –A few ↑ olives and no ↓ garlic, thanks.
22 KEITH ALLAN
For no obvious reason (listed items? colloquial style?), (57) is completely acceptable.
And-conjunctions of NPs with identical Pred monotonicity are the only
possibility — but-conjunctions are ungrammatical.
(58) Every ↑ child and/*but a few ↑ women were vaccinated.
Some/many ↑ women and/*but all ↑ children were inoculated.
Not any ↓ women and/*but not all ↓ children escaped unharmed.
No ↓ women and/*but few ↓ men play Russian roulette.
Not many ↓ anchovies and/*but not much ↓ lettuce, thanks.
As well as being either countable or uncountable, every English NP is either
definite or indefinite. A  NP identifies a recognized or recognizable
ensemble of denotata uniquely. On the other hand, number marking is either
implicit or explicit in the fact that an  NP instructs Hearer to partition
some recognized or recognizable ensemble denoted by the head noun of the NP.
For example, supposing Sue wants a couple of trout from the fishmonger, she
might say (59).
(59) Give me two trout, please.
(59) instructs Hearer to select a discrete ensemble of two from the ensemble of
trout on the counter. Alternatively, indicating two trout, Sue can say (60).
(60) Give me those two trout, please.
In uttering (60), Sue identifies exactly the ensemble of trout that she wants. Note
that the difference between the indefinite and the definite is
the indefinite requires Hearer to create an ensemble x from an ensemble y
such that x ⊂ y;
the definite picks out the ensemble x for Hearer by equating it with ensem-
ble y such that x=y (which is what universals do), or just naming it, e.g. [h/
x]F where [h=Harry] and x is a variable in F.
The words ‘picks out’ link definites to demonstratives, and refer to the property
which makes Harry or the loo or the son of one of my students pick out the
contextually appropropriate referent at a in wi at time ti–which raises too many
issues to be discussed here.6 [the x: Fx] is semantically ∃!x[x ⊆ f → x=f] to be
read “there is exactly one ensemble x and if it is a subensemble of f, then x is
identical with f” which can be paraphrased by “there is exactly one ensemble f
at a in wi at ti”. This shows that the is a species of universal quantifier and the
identificational property is associated with the entire restricted quantifier.

Russell’s (1905) formula would translate (61) as (62), B={x:x is a lamb}
and S={x:x is sick}.
(61) The lamb is sick.
(62) ∃ x[Bx ∧ ∀y[By → y=x] ∧ Sx]
Reformalizing (62) into our metalanguage we get (63).
(63) [the y: y ⊆ [ ∅ Q x: Bx]](Sy) (i)
[the y: y ⊆ [ ∅ Q x: Bx]] ∃ !y[y ⊆[ ∅ Q x: Bx] →
y=[ ∅ Q x: Bx]] (ii)
[ ∅ Q x: Bx]] x ≠ 0 ∧ x ⊆ b, ∴ b ≠0 (iii)
y⊆ [ ∅ Q x: Bx] y⊆ x ⊆b, ∴ y ⊆b ∧ b ≠ 0 (iv)
[the y: y ⊆ [ ∅ Q x: Bx]] ∃ !y[y ⊆b → y=b] ∧ b ≠0 (v)
[the y: y ⊆ [ ∅ Q x: Bx]](Sy) → ∃ !y[y ⊆ b → y=b] ∧ y⊆ s,
∴ ∃ !y[y ⊆ b∩ s → y=b] (vi)
[the y: y ⊆ [∅ Q x: Bx]](Sy) → ∃ !y[y⊆ b ∩ s → y=b], ∴ b ∩ s=b (vii)
[the y: y⊆[∅ Q x: Bx]](Sy) → ∃!y[y⊆b∩s → y=b] ∧ b∩s ≠0 (viii)
Line (ii) particularizes on the generic translation of [the x: Fx] into ∃ !x[x ⊆ f →
x=f]. Line (iii) establishes ensemble b as non-null, in effect it translates lamb.
Line (iv) establishes y as a subensemble of b. Line (v) says that, quantified by
the, y is an improper subensemble of b, ∴ y=b; this is counterpart to the
subformula ∀ y[By → y=x] in (62) and shows the to be one kind of universal
quantifier. (v) translates the lamb into “there is exactly one ensemble y and it is
identical with b, which is non-null” which can be paraphrased by “there is
exactly one non-null ensemble of lamb at a in wi at ti”. (vi) brings in the clause
predicate. (vii) says that if (61) is true, then the identifiable ensemble of lambs
at a in wi at ti is identical with the ensemble of sick lambs at that location.
(viii) brings down the information that the ensemble of sick lambs is non-null.
Present in (62), but missing in (63) is the information that b∩ s =1. This
information is not directly available from the morpho-syntax of (61). The N0
form ‘lamb’ is compatible both with a countable singular interpretation and also
with an uncountable interpretation — compare (61) with (64).
(64) The lamb is delicious.
24 KEITH ALLAN
In (64) the uncountable NP ‘the lamb’ most probably refers to meat, and in (61)
the countable singular to an individual animal. The distinguishing factor for (61)
and (64) is, of course, the predication. A related matter is what we know about
about things, based on the encyclopedia entry for the noun (cf. Allan 1995); for
instance, we know that cats are not normally a source of meat but lambs are.
However, the question of whether the lamb referred to in (61) is or is not meat
is decided by the predicate. The strands in arguing that ‘the lamb’ in (61) is
countable are the following meaning postulates:
(65) ∀ x[lamb ′ (x) → animal ′ (x)] (i)
∀ x[animal ′ (x) → living ′ (x) ∨ food.source ′ (x)] (ii)
∀ y[lx[animal ′ (x) ∧ living ′ (x)](y) → y ≥ 1] (iii)
∀ x[be.sick ′ (x) → living ′ (x)] (iv)
∀ y[lx[lamb′ (x) ∧ be.sick ′ (x)](y) → y ≥ 1] (v)
(65) grossly oversimplifies, but provides grounds for replacing line (63viii) with
(63) [the y: y ⊆ [ ∅ Q x: Bx]](Sy) → ∃ !y[y ⊆ b∩ s → y=b], ∴ b ∩ s=b ∧
b ∩ s =1 (ix)
[the y: y=[ ∅ Q x: Bx]](Sy) ≡ ∃ !y[y ⊆ b∩ s → y=b] ∧ b ∩ s =1 (x)
In line (ix) the clause predicate is instrumental in determining reference to a sick
and therefore living animal; consequently the NP is countable. By default the N0
form in a countable NP indicates a single referent as shown by b∩ s =1; it is
“exactly one” and not “at least one” because the is a universal quantifier. Line
(x) says the lamb is sick is true at location a in world wi at time ti if and only
if there is exactly one ensemble of lambs that is identical with the ensemble of
sick lambs and it is an atomic ensemble: this is what is meant by saying the definite
makes unique reference. The is NP ↑ Pred ↑, like some, etc., and is its own dual.
Let’s turn to a more complex example, and one that involves a plural
definite:
(66) All (of) the men in this room are secret policemen.
(66 ′ ) [all z: z ⊆ [F the y: y ⊆ [Y Q x: Mx Y] F]](Pz) (i)
[all z: z ⊆ [F the y: y ⊆ [Y Q x: Mx Y] F]] z:z ⊆ F=F (ii)
[the y: y ⊆ [Y Q x: Mx Y]] ∃ !y[y ⊆Y → y=Y] (iii)
[Y Q x: Mx Y] x >1 ∧ x ⊆ m, ∴ m >1 (iv)
[F the y: y⊆ [Q x: Mx] F] ∃!y[y ⊆ x⊆ m → y=m] ∧ m >1,
∴ ∃ !y[y ⊆ m → y=m] ∧ m >1 (v)
[all z: z ⊆ [F the y: y ⊆ [Q x: Mx] F]] ∃ !m[∃ !y[ z:z ⊆ (y⊆ m →

y=m) = (y⊆ m → y=m)]] ∧ m >1 (vi)
[all z: z ⊆ [F the y: y ⊆ [Q x: Mx] F]] ∃ !m[∃ !y[ z:z ⊆ m=m ∧
(y ⊆ m → y=m)]] ∧ m >1 (vii)
[all z: z ⊆ [the y: y ⊆ [Q x: Mx]]](Pz) ≡ ∃!m[ ∃ !y[ z:z ⊆m ∩ p=m ∧
(y ⊆ m → y=m)]] ∧ m ∩ p >1 (viii)
Line (iv) identifies the semantics for Y or men. Line (v) substitutes Y in the
conventional implicature of line (iii) with its interpretation from line (iv) and in
doing so conflates lines (iii-v) of (63) to give the semantics for F or the men.
Line (vi) substitutes F in line (ii) with the interpretation given for it in line (v),
providing an interpretation for all the men. (vii) reduces both instances of (y ⊆ m
→ y=m) from line (vi) to m and restates the crucial identificational information
as a conjunct. (66) differs from All men in this room are secret policemen only in
the identificational property contributed by the.
(67) Three of the men in this room are secret policemen.
(67 ′ ) [three z: z ⊆ [the y: y ⊆ [Q x: Mx]]](Pz) (i)
[the y: y ⊆ [Q x: Mx]] ∃!y[y ⊆ m → y=m] ∧ m >1 (ii)
z ⊆ [the y: y ⊆ [Q x: Mx]]
∃ z[∃ !y[z⊆ (y ⊆m → y=m)] → z ⊆ m] (iii)
[three z: z ⊆ [the y: y ⊆ [Q x: Mx]]]
∃ z[∃ !y[z⊆ (y ⊆m → y=m)] ∧ z ⊆ m ≥ 3], ∴ m ≥ 3 (iv)
[three z: z ⊆ [the y: y⊆ [Q x: Mx]]] ∃ z[z ⊆ m ∧ z ⊆ m =3] (v)
[three z: z ⊆ [the y: y ⊆ [Q x: Mx]]](Pz) → ∃ z[z⊆ m ∧
z ⊆ m ∩ p ≥3], ∴ m∩ p ≥ 3 (vi)
[three z: z ⊆ [the y: y ⊆ [Q x: Mx]]](Pz) ≡
∃ z[∃ !y[z⊆ (y ⊆m → y=m) ∩ p ∧ z ⊆ m ∩ p ≥ 3]] (vii)
[three z: z ⊆ [the y: y ⊆ [Q x: Mx]]](Pz) m∩ p =3 (viii)
In the next example the, which is a universal quantifier, ranges over three, which
is not; whereas three men means “at least three men”, the three men and all three
men both mean “exactly three men” — the universal quantifiers create this
precision effect.
(68) The three men in this room are secret policemen.
(68 ′ ) [the z: z ⊆[F three y: y⊆ [Q x: Mx] F]](Pz) (i)
[the z: z ⊆[three y: y ⊆ [Q x: Mx]] ∃ !z[z ⊆F → z=F] (ii)
26 KEITH ALLAN
[three y: y ⊆ [Q x: Mx]] m ≥3 (iii)

[the z: z ⊆[three y: y ⊆ [Q x: Mx]]
∃ !z[z ⊆y ⊆ m → z=m] ∧ m =3 (iv)
[the z: z ⊆ [three y: y⊆ [Q x: Mx]]](Pz) ≡
∃ !z[z ⊆m ∩ p → z=m] ∧ m = m ∩ p =3 (v)
(69) All three of the men in this room are secret policemen.
(69 ′ ) [all w: w⊆ [three z: z ⊆ [the y: y ⊆[Q x: Mx]]]](Pw) (i)
[all w: w⊆ [F three z: z ⊆ [the y: y ⊆[Q x: Mx]] F]]
w:w ⊆ F=F (ii)
[three z: z ⊆ [the y: y⊆ [Q x: Mx]]]
∃ z[ ∃!y[z ⊆ (y ⊆ m → y=m)] ∧ z ⊆ m ≥3], ∴ m ≥ 3 (iii)
∃ !m[∃ z[ ∃!y[ w:w ⊆ (z ⊆(y ⊆ m → y=m))
= (z ⊆ (y ⊆ m → y=m))]]] ∧ m =3 (iv)
∃ !m[∃ !y[ w:w⊆ m=m ∧ (y ⊆ m → y=m)]] ∧ m =3 (v)
[all w: w⊆ [three z: z ⊆ [the y: y ⊆[Q x: Mx]]]](Pw) ≡
∃ !m[∃ !y[ w:w⊆ m=m ∩ p ∧ (y⊆ m → y=m)]] ∧
m = m ∩ p =3 (vi)
Line (iii) can be checked against (67 ′ ). Line (iv) substitutes the F of line (ii) with
its interpretation from line (iii), but because all is a universal quantifier it constrains
m ≥3 to m =3.
We have seen that the is a quantifier, and in particular a universal quantifier
such that in [the x: Fx](Gx), f=f ∩ g. The universal quantifiers differ from one
another as follows:
[the x: Fx](Gx) ∃ !x[x⊆ f ∩ g → x=f] ∧ f ∩ g ≠0
[all x: Fx](Gx) ∃ !f[ x:x⊆ f=f] ∧ f ≠0
[every x: Fx](Gx) f = f ∩ g ≥ 1 ∧ ∀x[x ⊆ f ≡ ∀ y,z[(y ⊆ x ∧z ⊆x) →
y ≈ z ]]
[no x: Fx](Gx) f ∩ g =0
[any x: Fx](Gx) f∩ g ≠0
Universal quantifiers always occur as the leftmost quantifiers within the structure of
simple NPs in English. Notably all the N and both the N aparently consist of two
NPs as revealed by the synonymous all of the N and both of the N. Other combina-
tions of quantifiers in which the second is universal must occur in the NP of NP

construction, cf. any of the N, every one of the Ns, two of the Ns, etc. Why is all
the men acceptable but not *the all men? Although the is a universal quantifier it
does not focus on quantity, its function is to mark the identifiability of the
referent, and quantification is incidental. Quantification is also incidental with the
other articles, a(n) and some, but not with all and other quantifiers: they focus on
quantity. Thus although the men means “all of the men referred to” referring to
the set as a whole, all the men focuses on the merge of the subsets.
Generalizing on the complex quantification we have examined: for any pair
of concatenated quantifiers Qi Qj within the structure of a NP, the relation is
that the leftmost quantifier, Qi, is a subensemble of Qj–which it c-commands; i.e.
[Qi y: y ⊆ [Qj x: F]]
We saw that a numeral quantifier is semantically interpreted “at least Q”, and has
a standard implicature “exactly Q”; however, when it concatenates to the right of
a universal quantifier, the numeral quantifier is interpreted “exactly Q”. E.g
(70) All five men are bald Exactly five [and not six] of the men at a
are bald
The five men are bald Exactly five [and not six] of the men at a
are bald
Any five men will do Exactly five men [and not six]
It may be this precision effect that constrains the concatenation of the universal
quantifiers with fuzzy quantifiers such as most, many, much, some, a few/little. If
the concatenation is grammatical at all, there needs to be a restrictive relative
clause for the concatenation to be good whereas this is not necessary with a
numeral quantifier.
(71) a. *All (of) some men were Russians.
b. All of some men I met were Russians.
(72) a. ?Themany sailors were all charming.
[OK when elliptical]
b. The many sailors I met were all charming.
The (b) examples paraphrase as follows:
I met some men; they were Russians.
I met many sailors; they were all charming.
28 KEITH ALLAN
(73b) is superlative, i.e. ‘most’ is not functioning like e.g. the quantifier five.
(73) a. *The most models were in New York.
b. The most models I ever saw was in New York
I saw more models in New York than anywhere else
What has been achieved in the foregoing analysis? A comprehensive, internally
consistent account of number, countability, (in)definiteness, and quantification in
English that links meaning with morphological form. The interesting question is
whether it applies to other languages. Some of the observations do: e.g. the facts
that quantifiers exhibit scope relations, are conservative, demonstrate Pred and
sometimes NP monotonicity, most quantifiers have contradictories. It is likely
that all languages have a counterpart to all and no and some; it would be
surprising if more than a handful of languages had an exact counterpart to any.
It is probable that all languages have some counterpart to the quantity scale,
much less probable that the scale is divided up in a way exactly parallel to
English. We used ensemble theory to capture the properties of both countable
and uncountable NPs.7 We identified a category of quantifiers, symbolized Q,
which includes articles and denumerators. The formula [Qx:Fx](Gx) with which
we began is equivalent to the labelled bracketing [S[NPQ N]i [Sei VP]] at the level
of LF in GB theory. When we embarked on a more complex compositional
semantics involving a series of concatenated quantifiers that include the ∅ Q
quantifier N0 and the morphological plural quantifier Q, there is a mismatch
with standard accounts of LF that needs to be resolved. Quantifiers determine a
measure function on ensembles and their overlaps. With some quantifiers it is
necessary to supplement the semantic facts about quantification with standard
implicatures, e.g. a singular countable is semantically consistent with a quantity
Q≥ 1 whereas it has the standard implicature Q=1. We identified the mono-
tonicity of most of the quantifiers discussed and described some relationships
between quantifiers that depend on their monotonicity. The compositionality of
quantification was focused upon throughout the discussion. We identified the
relation between concatenated quantifiers Qi Qj as [Qi y: y ⊆ [Qj x:F]]. There is
a left-quantifier condition (an effect of c-command) which has the leftmost
quantifier constraining and even contradicting the implications of other quanti-
fiers within its domain. It is significant that in spoken and written language the
leftmost quantifier is encountered first. We noted that universals are leftmost
within a simple NP. A limitation of this paper is that only one place predicates
have been considered, however it is easy enough in principle to extend the
analysis to n-place predicates, e.g. the formula for a two place clause predicate
H and any F and G is [Qx:Fx]([Qy:Gy](Hxy)); a three place predicate B would
have the form [Qix:Fx]([Qjy:Gy]([Qkz:Hz](Bxyz))). In the Pred3 formula, Qk is
in the scope both Qj and Qi, and Qj is in the scope of Qi.
Notes
* This essay is dedicated to Rodney Huddleston whose scholarship provides a standard against
which Australian linguists grappling with the grammar of English measure their efforts and too
often feel humbled. I’m sure this work would be vastly improved had I had the benefit of
Rodney’s always insightful comments. I was, however, fortunate to have comments on early
versions of part of this paper from audiences at colloquia given at Manchester University,
Monash University, and University College London. Particular thanks are due to Lloyd
Humberstone and Peter Kipka who corrected many errors; the remaining faults are, alas, not
theirs but all my own.
1. The use of … to delimit ensembles is not due to Bunt. Also, the term “parts” used
throughout is borrowed from mathematical logic and refers to both parts and pieces. “Parts” as
in spare parts for a car are often functional and/or motivated components of a structured
/configuration; “pieces” as in pieces of the car were littered all over the road are not. Cf. Cruse
(1986: 157ff).
2. Grice (1975) identified conventional implicature as a non-truth functional implication, but said
very little about it. Lyons (1995: 276) writes ‘all sorts of meaning are encoded — i.e. in Grice’s
terms, made conventional — in the grammatical and lexical (and phonological) structure of
particular languages.’ I define it: A B (B is a conventional implicature of A) when in all
possible worlds A implies B (but does not entail B) such that if B does not hold, then A does
not hold, either. We may conclude that an entailment would be a conventional implicature were
it not truth functional. See Allan fc for discussion.
3. Rather than being in different locations, the referents of two formally similar NPs may be time
differentiated, e.g. A cockatoo is sitting on that pole [at ti] exactly where a cockatoo sat yesterday
[at tj]. The referents of such NPs can be distinguished by a temporal index ti tj. (This spatio-
temporal differentiation of NPs correlates with the participant role of the NP in the clause, e.g.
in One cat spat at another we have an actor at one location distinct from an undergoer at
another.)
4. A category a c-commands category b iff the first branching node dominating a also dominates
b, and neither a nor b dominates the other. The scope of a node a is the set of nodes that a
c-commands at LF (cf. Reinhart 1976). c-command has perhaps been superceded by m-com-
mand, though it is of no consequence in this paper: A category a m-commands category b iff
the first maximal projection dominating a also dominates b, and neither a nor b dominates the
other.
30 KEITH ALLAN
5. With temporal expressions, recurrence at similar intervals seems to be the critical factor, cf.
Every few days, every now and then.
6. The identificational characteristic of the definite can be illustrated by the following: Where an
antecedent clause containing an indefinite NP is [Qindef x: Fx](Gx) for any restrictor F and
clause predicate G, the definite successor with a NP coreferential with the antecedent indefinite
is the subformula [the y: Fy → y=x](Hy) for any clause predicate H. More precisely, [Qindef x:
Fx](Gx ∧ [the y: Fy → y=x](Hy)) — see Allan fc.
7. One questionable outcome that space prevents us from discussing is that a proper name denotes
an ensemble.
References
Allan, Keith. 1980. “Nouns and Countability”. Language 56.541–67.

Allan, Keith. 1986. Linguistic Meaning (Vols 1 and 2). London: Routledge and
Kegan Paul.
Allan, Keith. 1995. “What Names Tell about the Lexicon and the Enyclopedia”.
Lexicology 1.
Allan, Keith. Fc. Natural Language Semantics: The Study of Meaning in Human
Languages. Oxford: Basil Blackwell.
Apuleius, L. 1968. “De philosophia rationali sive peri ermhneiaV”. Opera
Omnia II, ed. by G.F. Hildebrand, 261–78. Hildesheim: Georg Olms. [Said
to be wrongly attributed to Apuleius]
Barwise, Jon and Robin Cooper. 1981. “Generalized Quantifiers and Natural
Language”. Linguistics and Philosophy 4.159–219.
Boethius, Manlius S. 1894. Opera Omnia, ed. by J.-P. Migne. Patrologiae
Latinae Tomus LXIV. Paris: J.-P. Migne.
Bunt, Harry C. 1976. “The Formal Semantics of Mass Terms”. Papers from the
Third Scandinavian Conference on Linguistics ed. by F. Karlsson, 81–94.
Turku: Academy of Finland.
Bunt, Harry C. 1979. “Ensembles and the Formal Semantic Properties of Mass
Terms”. Mass Terms: Some Philosophical Problems ed. by Francis J.
Pelletier, 249–277. Dordrecht: D. Reidel.
Bunt, Harry C. 1985. Mass Terms and Model-Theoretic Semantics. Cambridge:
Cambridge University Press.
Cann, Ronnie. 1993. Formal Semantics. Cambridge: Cambridge University Press.
Capella, Martianus. Martianus Cappella, ed. by Adolfus Dick. (Bibliotheca

Scriptorum Graecorum et Romanorum Teubneriana). Stuttgart: B.G.
Teubner, 1969.
Chierchia, Gennaro and Sally McConnell-Ginet. 1990. Meaning and Grammar:
An Introduction to Semantics. Cambridge MA: MIT Press.
Cruse, D. Alan. 1986. Lexical Semantics. Cambridge: Cambridge University Press.
Gamut, L.T.F. 1991. Language, Logic and Meaning: Introduction to Logic (Vol.1),
Intensional Logic and Logical Grammar (Vol. 2). Chicago: Chicago University
Press.
Grice, H. Paul. 1975. “Logic and Conversation”. Syntax and Semantics 3: Speech
Acts, ed. by Peter Cole and Jerry L. Morgan, 41–58. New York: Academic
Press.
Horn, Laurence R. 1972. On the Semantic Properties of the Logical Operators in
English. Bloomington: Indiana University Linguistics Club.
Horn, Laurence R. 1989. A Natural History of Negation. Chicago: University of
Chicago Press.
Leśniewski, Stanisław. 1988. Collected Works, ed. by Jan Srzednicki et al. Dor-
drecht: Kluwer.
Levinson, Stephen. 1983. Pragmatics. Cambridge: Cambridge University Press.
Lyons, John. 1995. Linguistic Semantics: an Introduction. Cambridge: Cambridge
University Press.
May, Robert. 1985. Logical Form: Its Structure and Derivation. Cambridge MA:
MIT Press.
Moravscik, Julius M.E. 1973. “Mass Terms in English”. Approaches to Natural
Language ed. by Jaako Hintikka, Julius Moravscik and Patrick Suppes,
Dordrecht: Reidel.
Neale, Stephen. 1990. Descriptions. Cambridge MA: MIT Press.
Ojeda, Almerindo. 1993. Linguistic Individuals. CSLI Lecture Notes #31. Stanford:
Center for the Study of Language and Information.
Quine, Willard V.O. 1960. Word and Object. Cambridge MA: MIT Press.
Reinhart, Tanya. 1976. The Syntactic Domain of Anaphora. Doctoral Dissertation,
MIT.
Russell, Bertrand. 1905. “On Denoting”. Mind. Reprinted in Logic and Knowledge.
London: Allen and Unwin (1956).
Schein, Barry. 1993. Plurals and Events. Cambridge MA: MIT Press.
Language, linear precedence and parentheticals
Noel Burton-Roberts
1. Introduction
This chapter concerns the scope and nature of grammar, particularly syntax and
the geometric conception of it in which Immediate Dominance and Linear
Precedence play a central role. Parentheticals offer a point of entry, since they
exhibit features not easily handled within this conception.1 This has led some to
propose extensions to the grammar and others to suggest that parentheticals fall
outside the scope of grammar altogether, belonging to the domain of utterance
interpretation and conceptual (not linguistic) representation. However, the
proposals seem subject to an unacknowledged conflict, a conceptual unease. In
some, grammatical analyses are adopted which seem to acknowledge that
parentheticals fall outside the scope of grammar. Others, in pursuit of the latter
idea, nevertheless arrange for their grammatical representation.
It is this conflict/unease that concerns me. Haegemann (1988) pursues the
matter to the conclusion that there is no linguistic relation between parenthetical
and any “host” clause. I want to agree with this but — given prevailing concep-
tions of language and syntax — it is not as simple as it seems. The problems
have to do with the distinction between language and speech (what grammars
generate and what speakers produce) and, where the distinction is made, how the
relation between them is conceived. This impinges on geometric syntax, linear
precedence in particular.
In Section 2, I briefly review some relevant features of parentheticals,
concentrating on non-restrictive relative clauses. In Section 3, I review some
proposed analyses and discuss conceptual and technical issues. In Section 4, I
speculate that a representational approach (Burton-Roberts 1994, Burton-Roberts
& Carr 1996) to the distinction and relation between what grammars generate
and what speakers produce might clarify the issues.
34 NOEL BURTON-ROBERTS
2. The problem of non-restrictive relative clauses
Restrictive relative clauses (RRs) are subordinate clauses. They appear within,
are contained by, contribute to the structure and interpretation of, other clauses
and are subject to syntactic principles operating in the domain of linguistic
expressions in which they function as modifiers. This seems not to be the case
with non-restrictive relatives (NRRs). NRRs appear not to be constituents of the
expressions that (apparently) contain them. The evidence has been extensively
discussed by the authors mentioned.
RRs, but not NRRs, are within the scope of operators and expressions
outside the R-clause itself. For example, when RRs appear within the comple-
ments of a verb like say, they are included in what is reported to have been said.
Not so with the NRR.
(1) a. John said that the receivers who had done a good job should
be dismissed
b. John said that the receivers, who had done a good job, should
be dismissed.
(1a), but not (1b), raises the question of why John should say that the receivers
should be dismissed, when by his own admission they had done an excellent job.
By contrast, in (1b) John, and the clause of which John is subject, are “innocent of
the knowledge of” the NRR, which is the independent responsibility of (1b)’s utterer.
Fabb (1990:71) considers the evidence provided by fixed expressions such
as make headway, noting (though not in so many words) that their constituents
can be distributed on either side of the boundary of a RR but not of a NRR.
(2) a. The headway the students made last week was phenomenal.
b. *The headway, which the students made last week, was
phenomenal.
(3) a. The advantage they took of me last week was unbelievable.
b. *The advantage, which they took of me, was unbelievable.
I get the same result with certain other idioms:
(4) a. The cat which John let of the bag concerned your demotion.
b. *The cat, which John let of the bag, concerned your demotion.
Constituents of idioms are required to “co-occur” with their co-constituents to
maintain the integrity of the idiom as such. This is maintained by the RR but
LANGUAGE, LINEAR PRECEDENCE AND PARENTHETICALS 35
destroyed by the NRR. It is as if the NRR were not there. In some as yet
undefined sense, and despite appearances, the relevant expressions do not “co-
occur” in the (b) examples.
RRs, but not NRRs, come within the scope of main clause negation, which
thereby licences negative polarity items in RRs but not NRRs.
(5) a. None of the authors who had any imagination remained with
them.
b. *None of the authors, who had any imagination, remained with
them.
And pronouns in RRs, but not in NRRs, are bound by quantifiers in main clauses:
(6) a. She gave every boy(i) who/that cleaned his(i) teeth well a
new toothbrush.
b. *She gave every boy(i), who cleaned his(i) teeth well, a new
toothbrush.
Furthermore, as shown by Every boy(i) was given a toothbrush so that he(i) could
clean his(i) teeth everyday, the quantifier can bind a pronoun in a clause outside
the quantified NP. This suggests that the NRR is, not just outside the domain of
that NP, but outside the clausal domain as a whole.
Haegemann (1988) notes the following contrast between the adjunct clause
in (7a) and the parenthetical clause in (7b):
(7) a. John(i) always works better while his/*John’s(i) children are
asleep.
b. John(i) studies mathematics, while his/John’s(i) wife studies
physics.
Referential expressions such as John cannot be bound by (and hence co-indexed
with) c-commanding NPs in argument positions (Principle C of Binding Theory).
The coreferential interpretation of John(’s) in (7a) is thus ungrammatical. But in
(7b) this is fine, suggesting that the parenthetical clause is not c-commanded by
the first subject NP, and hence not “in” the clause at all. The same results seem
to obtain when RRs and NRRs are compared:
(8) a. John(i) gets on best with those firms who employ
him/*John(i) frequently.
b. John(i) gets on well with those firms, who employ
him/John(i) frequently.
This, incidentally, is consistent with the following contrast, with therefore:

(9) a. *John gets on best with those firms who therefore employ him
frequently.
b. John gets on well with those firms, who therefore employ
him frequently.
The contrast seems due to the fact that therefore establishes a (discourse)
connection between independent clauses, of the sort discussed by Blakemore
1987; it cannot be employed to connect two clauses one of which is a constituent
of the other.2
The (Principle C) explanation of the (8a/b) distinction would, as Fabb
(1990: 72) suggests, carry over to the distinction between (10a) and (10b):
(10) a. *The LAGB which organisation meets tomorrow is based here.
b. The LAGB, which organisation meets tomorrrow, is based here.
on the assumption that which organisation is a fully referential NP. While I
believe this contrast does show that NRRs are independent clauses, Fabb’s
explanation is problematic. He allows that which organisation is a fully referen-
tial NP on a par with John because it has lexical content (supplied by organisa-
tion). The difficulty with this is that, if we were to replace which organisation
in (10b) with just which, we would surely be replacing it with an expression
which was referential in the same sense and to the same degree. The pres-
ence/absence of “organisation” does not seem associated with any difference in
referential status. In an obvious sense, which organisation is not a fully referen-
tial NP on a par with John or Senate. Compare:
(11) Max knows Senate meets tomorrow.
(12) Max knows which organisation meets tomorrow.
The complement clause in (12) — but not in (11) — is interrogative. See (13) – (14):
(13) Senate meets tomorrow.
(14) Which organisation meets tomorrow?
(14), but not (13), must be construed as a question precisely because which
organisation is not a referring expression on a par with John/Senate; the use of
(14) is construed as requesting a referential expression in place of which
organisation.
Nevertheless, (10) and (14) do show this: NRRs (but not RRs) pattern with
interrogative clauses in permitting wh-NPs like which organisation. In doing so,
NRRs are shown to pattern with clauses that can function as independent clauses.
One of the more salient contrasts between RRs and NRRs is that, while the
relative pronoun can be omitted in the RR, it cannot in the NRR.
(15) a. The article I’m working on needs to be short.
b. *The article, I’m working on, needs to be short.
Fabb (1990: 72) suggests (following Chomsky 1986b: 84) that null operators
such as that in (15a) must be licenced by being “strongly bound” by a c-com-
manding antecedent. The assumption that the NRR clause in (15b) is an indepen-
dent clause, not c-commanded by the article, is thus consistent with the unaccept-
ability of (15b).
Notice I have followed Fabb in applying an asterisk to (15b) (his [76]). This
would suggest that (15b) is ungrammatical. However, the suggestion is not
consistent with his argument that NRRs are independent clauses. It is the NRR
in (15b), not (15b) itself, that is ungrammatical. (15b) could inherit the ungram-
maticality of the NRR only if the connection between its clauses fell within the
domain of the grammar, only if the NRR were a syntactic constituent of (15b).
But that is contrary to the argument. The point may seem pedantic but it will
assume importance as we proceed.
Assigning the asterisk to the NRR itself, rather than (15b), anyway better
reflects the fact that the ungrammaticality of I’m working on is independent of
the other clause “in” (15b), The article needs to be short, which is in itself
independently grammatical and (compared with that in (15a)) fully interpretable.3
The fact that NRRs (but not RRs) are associated only with other clauses that are
independently grammatical and interpretable explains why, when a relative clause
is added by another speaker, it must be construed as an NRR, never as an RR:
(16) A: My publications will include the article in Scientific American.
B: Which you’ve not even begun to write yet.4
Notice also that, having mentioned someone by means of a jerk of the head, a
speaker may continue with an NRR but never an RR:
(17) I asked … [jerk of the head towards John], who was no help at all.
NRRs, it might be argued, cannot be independent since they include an expres-

sion (a wh-expression) which must be construed as coferential with an expression
in another clause (17 aside). They are thus no more independent than RRs.
Against this, we need to distinguish between the dependence of clauses as
such and the dependence of utterances. Since B’s utterance in (16) is, in its
context, complete and coherent, it should be construed as the utterance of an
independent clause. However, in including a (wh) pronoun whose reference is
given in the (linguistic) context of utterance, it is dependent as an utterance. In
short, it is the dependent utterance of an independent clause. Compare B’s
utterance in (16) with that in (18).
(18) A: I think it amounts to twenty quid.
B: That seems right.
The suggestion is that NRRs involve the dependent utterance of an independent
clause in the same sense that B’s utterance in (18) does. That and the NRR wh-
expression have the same referential status (both pronominals). They have
implication of reference though the actual reference is only given in the context
of utterance. (In this respect the wh-expression in an NRR contrasts with that in
an RR. See below).
Their having the same status would explain the lack of restriction on the
category of expressions that can serve as antecedents with the NRR. As is well
known, while RRs only function as modifiers in NPs (with the wh-expression co-
indexed with the head N), the range of expressions that can serve as antecedents
for NRRs is much broader:
(19) He put it in the shed, which is the right place for it [That’s the
place for it] (PP)
(20) She’s rich and famous, which I’d like to be too [That’s all I want
to be] (AP)
(21) He bowled slowly, which is how you should. [That’s the way to
do it] (AdvP)
(22) Andrew proved the theorem, which no one else did. [I wish I’d
done that] (VP)
(23) The boss is away right now, which is convenient. [That’s conve-
nient] (S/CP)
As regards (23), were the NRR to be regarded as subordinate with respect to The
boss is away right now, the wh-expression would have to be regarded — paradox-
ically — as a constituent of its own antecedent. The fact that the wh-expression
in NRRs can take a (main) clause as antecedent is a function of the syntactic
independence of NRRs.
Attention has been drawn to some connection between NRRs and interrogat-
ives. One might speculate that the clauses we think of as specifically NRR
clauses are simply those wh-clauses in which the reference of the wh-expression
is, on the occasion of utterance, mutually manifest to speaker and hearer.
Otherwise, we think of them as interrogative. In (10b) the wh-expression is
straightforwardly, if not “fully”, referential in being coreferential with the LAGB.
An interrogative such as (14) defines a set of answers (Huddleston 1994); its use
demands an answer from within the set defined. The suggestion is that the wh-
expression in both NRRs and independent interrogatives demands either an
“antecedent” or an “answer”. Where the demand is satisfied in preceding
discourse the clause will be construed as an NRR. In default of a preceding
(co)referent, the hearer will recognize that he or she is being requested supply it.
In such terms, the difference between an NRR and an independent interrogative
is a matter of use and is the difference between (discourse-dependent) reference
and the request for a referent, and between a referent as such and an answer.5
The relevance of such connections between the wh-expression in NRRs and
pronominals, and between NRRs and interrogatives, is that they serve to
distinguish sharply between NRRs and RRs. The distinction, crucial here, is often
obscured. Many analyses (e.g. Emonds, Fabb) seem more concerned with
establishing connections between NRRs and RRs, with their analyses of RRs
influenced by what seems true only of NRRs. In claiming that “appositives have
no properties”, for example, Emonds is claiming that they have no syntactic
properties in addition to those possessed (a) by (restrictive) relatives and (b) by
parentheticals (for Emonds, formed by a Parenthetical Formation rule). The wh-
expression in the NRR is a pronominal and is treated as such by Emonds; so
Emonds also treats that in the RR as a pronominal, coreferential with another
expression. And, since the relation between the wh-expression in the RR and its
“antecedent” clearly is grammatically determined, that in the NRR must also be.
This is consistent with Emonds’ assumption that NRRs fall within the domain of
grammar, but it is contrary to the argument pursued by Haegemann (1988) and
Fabb (1990). Yet Fabb, like Emonds, argues that “there is no need for construc-
tion-specific stipulations which distinguish between [NRRs and RRs]”. For Fabb,
as for Emonds, “a relative pronoun must share a referential index with some
node outside the relative clause”, whether RR or NRR.
But Fabb encounters difficulties in sustaining this idea, finding himself
obliged to admit that “the sharing of referential indices is not in itself a guaran-
tee of coreference” (p. 61). And, despite insisting that “Both antecedent and
relative pronoun will carry a referential index by virtue of being [full NPs]” (p.
58), Fabb concedes that, in contrast to the wh-expression in an NRR – which is
indeed coreferential with a full XP, that in the RR never is. On the contrary, the
“antecedent” of the wh-expression in an RR is the head N, not a full NP. Fabb
resolves this by allowing that a noun inherits its “referential” index from the NP
of which it is the head. But this means that the wh-expression in an RR is
effectively co-indexed with the NP within which it functions as a modifier. This
makes it a constituent of (one of) its own “antecedents”. This is paradoxical, but
only if co-indexing is to be correlated with coreference. That these cannot be
correlated in the RR is anyway reflected in the fact that, in taking just the N as
the immediate antecedent, the RR makes a (necessary but not sufficient)
contribution, with that N, to the conditions of reference of a distinct, referring
NP expression.
Let us then disassociate the grammatical phenomenon of co-indexing from
coreference. As a relation between a speaker’s use of an expression and some
element in conceptual representation, reference (and hence co-reference) is
determined in the context of utterance, not by the grammar. Coreference is not
a relation between linguistic expressions as such, but co-indexing is. An RR
involves grammatical co-indexing, not coreference; an NRR involves coreference,
not co-indexing. The difference seems crucial to the contention that the relation
between NRRs and the clauses that contain expressions coreferential with their
wh-expressions is not a grammatical relation. If, as Fabb suggests, NRRs were
subject to any species of co-indexing that RRs are subject to, it would count
against that contention.
More generally, though, it also counts against that contention is that it calls
for a more uncompromising distinction between what grammars generate and
what speakers produce than anything usually contemplated. Those who claim
that, as utterance phenomena, NRRs fall beyond the domain of grammar need to
address these more general issues. I turn to this now.
3. Proposed analyses
Emonds does not consider the possibility that the relation between a NRR clause
and an antecedent-containing clause might lie outside the grammar. Why not?
Notice first that it has (as usual) just been described as a relation between
clauses. Clauses are grammatical objects. As a relation between clauses, it surely
should lie within the domain of grammar. We don’t escape this conclusion by
replacing talk of clauses with talk of utterances, for the relevant utterances are
universally regarded as utterances–of–clauses. And, as this is generally under-
stood, utterances are regarded as the instantiating tokens (or “realizations”) of
linguistic expressions (their types), instantiating expressions by instantiating their
grammatical properties. On these terms, the relation must be a matter of grammar.
Furthermore, the linear position of NRRs is constrained in much the way
one would expect in a conception of grammar in which linear precedence and
constraints thereon play a major part. Within such a conception, LP phenomena
are constrained only within ID domains. It should then follow unproblematically
that an (LP-constrained) NRR is a (dominated) grammatical constituent of some
other linguistic expression.
To accommodate the NRR, however, Emonds resorts to an idea that is not
unproblematic. He has the NRR dominated by an extra type of node: “E”. This,
rather than “S”, is now the Initial Symbol. The other clause is also dominated by
a (co-ordinated) “E” node. These two “E” nodes are in turn dominated by a third
“E” node. Material which in surface structure follows the NRR is moved across
it and attached to the topmost “E” node. The conceptual status of “E” is never
made clear. Although a grammatical proposal, this seems to concede that the
matter does in some sense lie beyond grammmar (previously defined by “S”). It
is as if Emonds wants to say both that one expression generated by the grammar
is in another and that it is not in it.
The proposal has been criticized (e.g. Mittwoch 1985, Haegemann 1988,
Fabb 1990) as confusing syntax with utterance phenomena. I want to agree.
However, given (a) the importance universally attached to Linear Precedence in
syntax, and (b) how the distinction/relation between sentence (grammar) and
utterance (speaker) is generally conceived, there is a problem here. In default of
addressing these general issues, the criticisms are at least too pat.
Safir (1986) proposes to extend the grammar with a further level of
grammatical representation. In addition to Logical Form (LF), Safir proposes to
handle the relation between the NRR and the antecedent(-containing) clause (co-
indexing etc, as in Fabb) at a level of LF-prime (LF’). This raises similar

questions, and has been dismissed on the same grounds by Haegemann. Fabb
identifies LF-prime with “discourse representation”.
In contrast, McCawley’s (1982) proposal demands no such conceptual
extension to the grammar. It explicitly embraces the geometric conception of
syntax universally assumed, rejects any extra components, or new species of
nodes. It consists in allowing for movement rules that don’t alter constituency.
The NRR in (10b), for example, remains immediately dominated by a root S
(which dominates both clauses in (10b)) even when moved to a linear position
within the other clause, which thereby is made discontinuous. Thus the NRR is
in the other clause on the LP axis, but not in it on the ID axis.
This is a solution to the problem of Emonds (1979). But it calls for output
trees with crossing branches. These are generally repudiated, but (as McCawley
points out) for no very good reason. The only objection to cross-branching is to
such branches in “deep structure” trees: base-rules are incapable of coherently
generating trees with crossing branches (Postal 1964). But McCawley is here
proposing cross-branches only in SS, not DS. Given the assumption that LP is
instantiated only in ID domains, and given either no distinction between clause
and utterance or the (instantiating) relation between them presented above, this
seems unobjectionable. I am not aware of any substantive principled criticism of
it. In fact, Haegemann (1988: 251) has it that “the parenthetical intervenes within
a clause” (my emphasis); this is precisely what McCawley is modelling as a
linguistic phenomenon of “discontinuous constituency”.
Only Espinal (1991) dismisses it on the grounds of its crossing branches.
Espinal, then, is objecting to S-structure crossing. Now, McCawley observes that
the objection to S-structure crossing is effectively an objection to any tree that
cannot be translated into a linear string of labelled bracketings (crossing trees are
not translatable so). In short, only surface trees equivalent to linear strings are
admitted. McCawley attributes this ban to an assumption among grammarians
that “strings are more basic than trees”. He offers no speculation as to the
grounds of this preference for strings over trees, so I shall. I suggest it is due to
a type/token view of the relation between what-is-generated-by-a-grammar
(linguistic expressions) and what-is-produced-by-speakers. If what is produced by
speakers instantiates, realizes, or exemplifies what grammars generate, and if that
is why speakers may be said to “utter linguistic expressions”, then what is
generated by a grammar must be utterable, produceable by a speaker. And,
undeniably, what a speaker produces/utters is — spatially or temporally — linear.
Linearity is paramount in Espinal’s own proposal. While Espinal objects to

crossing branches, she proposes to abandon a much more fundamental principle
of phrase structure, namely that LP phenomena are instantiated only in ID
domains. Her proposed “innovation in phrase structure theory” allows that a
structure can be a “constituent” of another without being dominated by any node
in the latter. The two structures intersect geometrically only at LP, defined by
Espinal as the “line”, or plane, on which all their respective terminal symbols lie.
However, as a matter of grammatical fact, there is no such “line”.6 The plane
that Espinal seems to have in mind is the temporal/spatial linearity of physical
phenomena produced by speakers/writers in their utterance behaviour.
It is indeed Espinal’s claim that the relevant phenomena (“disjunct constitu-
ents”) are utterance phenomena, “best analysed at a postsyntactic level of
representation”, interpreted only “at the moment of utterance processing”. What
Espinal appears to mean in writing of “independent syntactic constituents” is that
they are “constituents” of utterances, but not syntactic constituents and that is
why they are “independent”. But then, why should it call for any such innovation
in grammatical theory? Espinal appears to regard such innovations as necessary,
because she believes conventional grammar is not adequate to its task, which she
sees as the task of describing utterances. Espinal’s proposal is a vivid illustration
of the conflict alluded to earlier.
Though they adopt opposite positions on whether such phenomena fall
within the domain of the grammar (are strictly linguistic), only McCawley and
Haegemann seem free of unease. It is perhaps significant that McCawley never
alludes to any distinction between utterances and what grammars generate. For that
reason, and because it is the position I favour, I focus on Haegemann’s position.
Haegemann’s contention implies that, while grammars generate NRRs,
grammars do not generate any linguistic expression having an NRR as a
syntactic constituent.
But consider the “examples” cited above: (1b), (8b)–(10b), (19)–(23). Under
prevailing concepts of language and syntax, these are generally regarded as
examples-of-linguistic-expressions. But if that is what they really are, then they
are examples-of-linguistic-expressions-that-include/contain-NRRs. Then, contrary
to the contention, there are such expressions, and the grammar generates them.
Against this, the contention implies that these cannot be examples of linguistic
expressions: there are no such expressions. So we need to say what they are
examples of. To repeat, the bland insistence that they are examples, not of
linguistic expressions (LEs), but of utterances, is of no help as long as the
relation between LE and utterance is conceived of in terms of type-token

(realization, instantiation and, indeed, exemplification).
The contention thus calls for a different, more radical distinction between
language and (the products of) utterance behaviour. It might be thought that such
a distinction is drawn by Chomsky (e.g. 1986b: Chs.1–2) in his more recent
formulations of the competence-performance distinction: his identification of
“language” with I-language (grammar), “a state of mind/brain” radically distin-
guished from utterance behaviour and its products (“E-language”), which is
rigorously excluded from the scope of linguistic theory. I-language does not
consist even in behavioural abilities. It has “no objective existence apart from its
mental representation” (1972: 169). I believe the contention that parentheticals
can/should be excluded, as utterance phenomena, from the domain of grammar
(I-language) is sustainable only if just such a distinction can be sustained and
rendered plausible. I contend that such a distinction can be sustained only if
accompanied by a consistent account of the relation between them. For, however
radically (I-)expression and (E-)utterance are distinguished, they are related.
But I am not aware of any such consistent account of this relation. Chomsky’s
I/E distinction should, for example, imply a distinction between what-grammars-
generate and what-speakers-produce so absolute that, if (as he suggests) gram-
mars generate LEs, then LEs cannot be what speakers produce, and conversely.
Now this sets the question: if grammars do generate LEs, what is it that speakers
produce and what is its relation to the (I-)language? Insofar as these questions
have been addressed, they have received answers that seem more consistent with
the (type/token) view already outlined. Haegemann herself, in her excellent 1994
text, has it that “sentences of a language” are both generated by a grammar and
produced by speakers (5–6, see also 7). But in that case grammars generate what
speakers produce; E-language is what is actually generated by I-language.
It is difficult to reconcile this with Chomsky’s internalist dismissal of
“E-language” (or with excluding NRRs from the domain of grammar). But
Haegemann is not inaccurate here. Chomsky himself has it both that “we may
take the linguistic expressions of a given I-language to be the S[tructural]
D[escription]s generated by it” (e.g. 1992b: 211) and that “Jones has … capaci-
ties of mind that allow him to produce and interpret linguistic expressions” (e.g.
1992a: 121). Furthermore, Chomsky has it that E-language is not just “external”
but “externalized”, and I-language not just “internal” but “internalized”. This
suggests that language does have an external aspect. So does his (e.g. 1986b)
gloss of I/E as “intensional/extensional”, which suggests that the products of
speakers’ behaviour are linguistic extensions, having true intensional linguistic

descriptions. How does this differ from an account of what it is to “utter
linguistic expressions” in terms of type-token (physical production of tokens of
linguistic types)?7
The contention that there is no grammatical relation between NRR and any
host expression, and the wholly internalist conception of language on which it
depends, calls for a different conceptual analysis of what it is to “utter linguistic
expressions”. For consistency, it should be an account of the I/E relation that
allows us, with plausibility, to reject the assumption that NRR clauses, or any
(grammar generated) I-linguistic expression, can occur as, or be a constituent of,
an utterance (an E-behavioural product).
More generally, the required relation must allow us, with plausibility, to
deny that utterances as such have any linguistic property. This is anyway
indicated by a fundamental problem for a type-token account of the relation
between LE and utterance. The next three paragraphs deal with this.
Classical type-token ontology allows that types (e.g. LEs) are abstract
(lacking occasion/location), but insists that tokens are occurrent physical phenom-
ena (utterances). The problem is that we wish to say of complex linguistic
expressions as such (the types) that they are syntactically composed of other LEs.
LEs occur in the composition of LEs. In other words, complex LEs are constitut-
ed by one or more tokens of other LEs (such as the man in the hat: two tokens
of the definite article). However, classical tokens (but not types) are physical
phenomena. Now, constituent physical phenomena can only occur as constituents
of occurrent physical phenomena (which types are not). We cannot then say that
the (abstract) types themselves (the LEs) are composed of tokens of other LEs.
Only physically occurrent tokens of LEs could possibly be composed of tokens
of LEs. And, with utterances regarded as the tokens, this means that only
utterances are linguistic.
Even if coherent, this exclusively externalist conclusion is not what is
generally intended. We could achieve what is intended in linguistics adopting a
Principle of Type-Token Affinity and insisting on a distinction between types
that do have physical phenomena as tokens (for instance mercury, C-sharp, and
letters of the alphabet) and types that do not. Include linguistic types among the
latter. Then, linguistic tokens would be no less “abstract” than the types they are
tokens of. In that case, there would be no objection to the statement that LEs as
such are composed of (equally abstract) tokens of LEs.
This, I believe, reflects actual practice in linguistics, where “linguistic
occurrence” is understood quite distinctly from “physical occurrence” — the

latter, but not the former, being events. But it has the consequence that LEs are
instantiated (tokened) only as (abstract) constituents in the structure of other LEs,
never in or as physical utterance phenomena. On these terms, utterances could
not be the tokens of linguistic types. Nevertheless, LE and utterance are related,
and we need an account of the relation.
4. A representational approach
The relation proposed in Burton-Roberts (1994) is representation. To avoid

misunderstanding, call it M-representation. ‘M’ stands for Magritte. The allusion
to Magritte is intended as a reminder of his painting “Ceci n’est pas une pipe”,
itself a reminder that, in looking at the painting, we are looking not at a pipe but
at the representation of a pipe. The reminder is necessary because we (linguists)
are used to a distinct technical use of the term, which I shall distinguish as C-
representation. A C-representation is constitutive of what it represents. When, for
example, Chomsky writes of a grammar being “mentally represented”, the
grammar is not regarded as distinct from that representation. One might prefer
“mentally constituted” here. Consider also his C-knowledge (of language) and his
C-(structural) description (of LEs). The rationale of my appeal to “representa-
tion” is that a representation is not constitutive (and not a token, realization,
instantiation, or example) of what it is a representation of.
The proposal is that speaker/writer/signers produce wholly non-linguistic
external physical phenomena in aid of physically (M-)representing (internally
constituted) linguistic expressions generated by their grammar.
Representations come in two varieties, iconic and conventional. Iconic
representations (like Magritte’s of a pipe) represent in virtue of resembling what
they are representations of. Conventional representations (such as ‘p’, ‘∫’, ‘501’)
do not. It follows, both from the Affinity principle and from the exclusion of the
“E” from the domain of linguistic enquiry, that utterances are devoid of the
properties in terms of which LEs are wholly constituted, namely linguistic
properties. The representation must then be wholly conventional, not iconic.8
And, since the relevant phenomena (acoustic, graphic, manual) are in a relation
to the linguistic, the relation itself must be extra-linguistic. Hence the representa-
tional conventions must be non-linguistic conventions.
In Section 2, I cited linguistic expressions, identifying each by producing,
not the LE itself (LEs are not produceable by speaker-writers), but physical
phenomena in aid of representing it, in accordance with accepted (graphic)
conventions for physically representing that LE.9 No LE, nor any token of one,
is there on the page, only physical representations of LEs, consisting wholly of
linearly ordered tokens of physical types (letters). As speakers and linguists,
however, we are concerned with what is represented, not with the representation.
Inevitably, then, there is a tendency (compulsion, even) to respond to representations
in terms of what they represent. So, where what is represented is linguistic, we
respond to the representation as linguistic — and do talk of expressions as being
on the page. But the representational idea encourages us to understand this as we
(ungruesomely) understand, for example, “T. S. Eliot hangs beside Edith Sitwell”
in London’s National Portrait Gallery. And the sense to be attached to “utter a
linguistic expression” is exactly that in which we say of Monet that he “painted
his garden at Giverny”. These are legitimate locutions, but highly elliptical.
This representational idea is far removed from more usual views of the
relation between language and — for example — speech, between what grammars
generate and what speakers produce. Were the usual views entirely satisfactory, it
would, I concede, be unnecessary. But are they satisfactory?10 Our discipline is
rife with unease on these matters, as I believe the discussion of parentheticals
illustrates. I cannot pursue the ramifications of the idea here (see Burton-Roberts
1994, Burton-Roberts and Carr 1996), so I conclude with how it bears on NRRs
and with some speculations on the nature and status of linear precedence.
By this proposal, utterance phenomena, as conventional (M-) representations
of the linguistic, are not linguistic. This offers a view of the relation between the
linguistic and utterance phenomena (as non-linguistic) more clearly consistent
with Chomsky’s I/E distinction and the contention that parentheticals, as an
utterance phenomenon, fall outside the domain of grammar. We can now
differentiate between linguistic (I) and representational (E) senses of such terms
as “contain”, “(co-)occur”, and “appear”.
As usually understood, saying that tokens of LEs physically co-occur is
tantamount to saying that those LEs themselves co-occur. So, if a token of one
LE is contained by a token of another, that is tantamount to that LE being
contained by (instantiated within) the other LE. On this view, LEs (co-)occur if
and only if the physical phenomena that are their classical tokens (co-)occur.
Why this should be attractive is perhaps a matter of intellectual history. That it
is suspect is shown at least by the problems encountered above with NRRs. On
that view, (15b) as a whole would indeed be “ungrammatical”; and we could not
deny that the constituents of the relevant idioms do, as a matter of “linguistic”
fact, observably (physically) co-occur in (2b)–(4b)
By contrast, occurrence of a representation is entirely independent of what
is represented, especially where what is physically represented is not physical,
but a mentally constituted LE. Flying planes is dangerous is an ordered (physical)
co-occurrence of the conventional graphic representations of four words. Now the
represented words, the words themselves, do indeed (linguistically) co-occur in
the structure of a certain English clause. But there is no guarantee that the
representational co-occurrences just presented are (intended as) representational
of that clause (cf. Boarding flying planes is dangerous).
On this approach, it is not (the token of) an NRR that appears and is
contained within (the token of) another clause. It is the (non-linguistic, non-
clausal) representation of an NRR that physically occurs within the representation
of another clause. LEs do not “occur” in virtue of representations of them
occurring. It is thus consistent to say, of (2b)–(4b), that physical phenomena
there co-occur which, by the usual graphic conventions, are representational of
the words that constitute the relevant idioms, but the words themselves do not
co-occur in (2b)–(4b). If we think of ‘(2b)’ – ‘(4b)’ as labels of LEs, that is due
to our interest in what they are representations of. The parenthetical does not
“intervene within a clause”, nor is there any linguistic phenomenon of disconti-
nuity here (or anywhere). This is a spatio-temporal matter, occurring when
production of physical phenomena in aid of representing one LE intervenes
during the production of physical phenomena in aid of representing another
(independent) LE. The discontinuity is of representations, not of the linguistic
expressions represented.
In these terms, there is no call for “linearization”, of LEs or anything else
(Espinal 1991, Haegemann 1988). LEs do not figure in utterance. They don’t
need, and could not, be “linearized” in utterance — any more than any pipe
needed to be flattened onto Magritte’s canvas. Nor do the representations
themselves need to be “linearized”. They just are linear — as linear as Magritte’s
canvas is flat.
It is less clear though that, in serving as representations of LEs, the relevant
phenomena are serving as representations of anything linear. Quite the opposite,
I believe. In this connection, Gazdar et al’s (1986) ID/LP format, in which the
constraints on Immediate Dominance and those on Linear Precedence are strictly
separated, is relevant because it draws our attention to a clear distinction between
ID and LP. Both are geometric (positional) notions. But ID is used (by linguists)
in aid of physically (M-)representing something else, namely: all that we

understand by “syntactic structure” (constituency, the internal logic of syntactic
properties), bar what is known as “word order”. Given the compulsion alluded to
earlier, we may think of such aspects of syntax in geometric terms (trees), but
there is no necessity to suppose that they are, in themselves, geometric. The
linguistic relation of head-complement, for example, is not a geometric relation.11
LP is quite different. LP is inalienably positional. What does the LP axis in
syntactic trees relate to other than the linear positionings physically observable
in utterances? Having separated ID and LP, we should ask: what, apart from that,
are linear precedence constraints for?
On a type-token view, the answer appears self-evident. The sequentially
ordered physical phenomena are tokens of LEs. They are regarded as tokens of
sequentially ordered LEs. LP constraints are thus required to effect correct
orderings of LEs. This implies that LEs as such are sequentially ordered and
(since what pertains to LEs as such is a matter of grammar) that LP constraints
are grammatical constraints.
But consider the matter in representational terms. That the representations
are linearly ordered is physically manifest. However, they themselves are not
linguistic. Constraints on their ordering cannot, then, be linguistic constraints.
They are constraints on representational order. It is the physical representation of
an I-linguistic complement expression that is ordered with respect to the physical
representation of an I-linguistic head expression.
Seeking a rationale for LP constraints in such a framework, we might ask:
is this property of the physical representation (its linear order) itself representa-
tional? Is LP ever exploited in aid of the representational enterprise utterers
engage in? The answer is not inevitable; not all physical features of the represen-
tation are representational; a feature that is representational in one language may
not be in another. Tone, for example, is representational only in some languages.
Languages with “free word order” are, in this light, ones in which LP is not (or
to a lesser extent) exploited for representational purposes. But for any language
in which ordering of representations is highly constrained, it is reasonable to
suppose that LP is so exploited. In that case, the ordering in the physical
representation of a head-complement relation, for example, would contribute to
the representation of it as a head-complement relation.
LP constraints, then, are not linguistic though may be representational. In
English at least, linearity has been harnessed in aid of the conventional physical
representation of the (non-linear) syntactic structure of LEs. Considering the
notion of syntactic tree in this light might allow us to pinpoint the source of the
unease alluded to earlier. For, if the linearity exhibited by utterances contributes
to the representation of the syntax of LEs, then LP contributes to the representa-
tion of what is (by linguists) represented, at least in part, by means of Immediate
Dominance. In that case, trees, in including (as they must) both ID and LP,
include both a theoretical (M-)representation of I-linguistic structure (by means
of ID) and an (iconic) copy of the (LP) E-means of (M-)representing that
structure ordinarily used by speakers/writers. In such terms, trees embody a
fundamental confusion of the linguistic and what is merely (non-linguistically)
representational of it.
The contention that there is no grammatical relation between an NRR and
any host expression must itself imply that such trees embody some such confu-
sion. The linear position of NRRs is thought to be highly constrained. If syntax
were a matter of both LP and ID, it is reasonable to suppose that the linear
position of NRRs as such is grammatically constrained. Under the representation-
al idea, by contrast, it is not the linear position of any NRR that is constrained,
but the linear position of its physical representation. This is a matter of represen-
tational, not grammatical, fact.
The constraint seems to be that the representation of the NRR must be, if
not linearly adjacent to the representation of the antecedent of its wh-expression,
then close enough to that representation to guarantee that the antecedent can be
correctly identified. If this is correct, it indicates that, while English does exploit
linear precedence in aid of the representation of the structure of linguistic
expressions, not all constraints on linear precedence in English representations
have even a representational motivation. The constraint on the linear position of
the representation of NRRs seems instead to have a general cognitive (conceptu-
al) motivation.
Notes
1. Discussed by Emonds (1979), McCawley (1982), Safir (1986), Haegemann (1988), Fabb (1990),
Espinal (1991).
2. In fact this should be rephrased, in a way more consistent with the general argument. Is the idea
of “a connection between independent clauses” coherent? Clauses are syntactic objects; the only
kind of connections that can hold between such things are syntactic; hence connected clauses
cannot be independent. Instead, and more faithful to Blakemore, therefore establishes an
inferential connection, not between clauses, or even utterances, but between conceptual
representations.
3. More generally, in talking of clausal dependence, we are generally talking of mutual depen-
dence between clauses. Also, when I say “fully interpretable” I mean as fully interpretable as
any sentence as such could be.
4. The second person pronoun in (16)B and the first person pronoun in (16)A need to be
coreferential. See Mittwoch (1985) for a discussion of the general implications of this.
5. This speculation owes something to Wilson and Sperber’s (1988) approach to non-declaratives.
6. That is, if we seek an algorithm from PS rules to geometric objects such as trees, all distances
must be held constant, including those between mothers and daughters (in ID). Under this
constraint, it would be the merest accident if the terminal symbols were to form up in a plane
line.
7. The question remains when, as in Chomsky (1995: Introduction and Ch.1.), it is suggested that
I-language “issues instructions” to a productive (articulatory)-perceptual performance system.
8. Conventional, but not necessarily arbitrary. The number represented by ‘501’ is thereby
conventionally represented, but not arbitrarily, as witnessed by the fact that a change in any part
(digit) of that representation will change the number represented.
9. In an oral presentation, I would have produced phenomena in accordance with conventions
governing acoustic representations.
10. “Uttering is, presumably, a relatively unproblematic notion, though even here just what counts
as the uttering of the token of a sentence, or even what a sentence is, could well be debated at
length” (Cresswell 1985:41).
11. The geometry of my family tree would include a representation of the relation between my
father and myself, but this is clearly not a geometric relation.
References
Blakemore, D. 1987. Semantic Constraints on Relevance. Blackwells: Oxford.

Burton-Roberts, N. 1994. “Ambiguity, Sentence and Utterance: A Representa-
tional Approach”. Transactions of the Philological Society 92 (2). 179–212.
Burton-Roberts, N. and P. Carr. 1996. “On Speech and Natural Language”.
Newcastle & Durham Working Papers in Linguistics 4.13–40. To appear in
Language Sciences.
Chomsky, N. 1972. Language and Mind. Harcourt, Brace, Jovanovich: New York.
Chomsky, N. 1986a. Barriers. MIT Press: Cambridge, Mass.
Chomsky, N. 1986b. Knowledge of Language. Praeger: New York.
Chomsky, N. 1992a. “Language and Interpretation”. Inference, Explanation and
other Philosophical Frustrations ed. by J. Earman, 99–128. California:
University of California Press.
Chomsky, N. 1992b. “Explaining Language Use”. Philosophical Topics 20. 205–231.
Chomsky, N. 1995. The Minimalist Program. M.I.T. Press.

Cresswell, M. 1985. Structured Meanings. MIT Press: Cambridge, Mass.
Emonds, J. 1979. “Appositive Relatives Have No Properties”. Linguistic Inquiry
10. 211–43.
Espinal, M. 1991. “The Representation of Disjunct Constituents”. Language 67.
726–762.
Fabb, N. 1990. “The Difference between English Restrictive and Non-Restrictive
Relative Clauses”. Journal of Linguistics 26. 57–78.
Gazdar, G., E. Klein, G. Pullum, and I. Sag. 1985. Generalised Phrase Structure
Grammar. Blackwells: Oxford.
Haegemann, L. 1988. “Parenthetical Adverbials: The Radical Orphanage Ap-
proach”. Aspects of Modern English Linguistics ed. by S. Chiba, 232–254.
Kaitakushi: Tokyo.
Haegemann, L. 1994. Introduction to Government and Binding. Blackwells: Oxford.
Huddleston, R. 1994. “The Contrast between Interrogatives and Questions”.
Journal of Linguistics 30. 411–439.
McCawley, J. 1982. “Parentheticals and Discontinuous Constituent Structure”.
Linguistic Inquiry 13. 91–106.
Mittwoch, A. 1985. “Sentences, Utterance Boundaries, Personal Deixis and the
E-hypothesis”. Theoretical Linguistics 12. 137–152.
Postal, P. 1964. “Constituent Structure”. International Journal of American
Linguistics. 30.
Safir, K. 1986. “Relative Clauses in a Theory of Binding and Levels”. Linguistic
Inquiry 17. 663–689.
Wilson, D. and D. Sperber. 1988. “Mood and the Analysis of Non-Declarative
Sentences”. Human agency: language duty and value ed. by J. Dancy et al.,
77–101. Stanford UP: Stanford.
The English modifier well
Ray Cattell
1. Introduction
This paper will explore the semantics and the syntactic category of expressions
consisting of well + passive participle. Many linguists have commented on the
adjective-like properties of passive participles, and at least since Wasow (1977)
it has been generally accepted (well accepted, in fact) that “some passive
participles are adjectives, while others are verbs” (p. 338).
There are some expressions, such as well-heeled (“rich”), well-hung
(“sexually well-endowed”, used of a man) and well-grounded (“firmly founded”)
for which there are no related passives, so they are the examples which most
clearly must be called adjectives rather than verbs, if those are the only choices.
Despite the lack of corresponding passive sentences, these words have the
form of participles, so I will follow Wasow in using the term “passive participle”
neutrally as between verbs and what he calls adjectives. But I am not so sure that
the latter are really adjectives in the same sense as good, big, and the like, so I
am going to use the tentative class-name “predicatives” for them, so as not to
prejudge the question. This allows the possibility that they are not really
adjectives, though they have some properties in common with them.
Many people, including me, tend to use a hyphen between well and a
following participle when they feel that the latter is a predicative. However, I
will omit such hyphens throughout this paper so as not to prejudge the issue of
whether the expressions are predicative or not.
The behaviour of degree modifiers is cited by Wasow as one of the
indicators of the adjectival nature of many passive participles. Well is a degree
modifier, and much of its behaviour does tend to support the predicative nature
of the participles it precedes.
54 RAY CATTELL
One of the characteristics of adjectives to which Wasow pointed is the

ability to occur as a modifier in a noun phrase, and many passive participles
modified by well correspond to this description; for example, a well typed essay,
a well knitted jumper, as well as ones like those just discussed: a well heeled
couple, a well hung guy, a well grounded suspicion.
Wasow points out that adjectives can occur as complements of verbs like act,
become, look, remain, seem, sound; and again we find that many participles linked
with well have this characteristic: The essay looks well typed; The jumper seems
well knitted; They look well heeled; The suspicions seem well grounded, and so on.
However, a study of well reveals that it is still far from clear when a
passive participle is a verb and when it is a predicative, or whether the two are
neatly separable at all.
2. “Quality” well1
2.1 Well = “effectively”
Both the sentences in (1) seem like passives, at least at first glance.
(1) a. The ball was fielded well by the bowler.
b. The ball was well fielded by the bowler.
It is when well precedes the participle it modifies, as in (1b), that the question of
whether the sequence is a predicative arises. In other words, it is there that it
seems most relevant to ask whether the verb be is a copula or an auxiliary. When
well follows the participle, as in (1a), there is not much temptation to feel that
the sequence is a predicative, or that be is a copula. But we cannot leap to the
conclusion that when the modifier well precedes a passive participle it always
indicates that the latter is a predicative.
If we ask what the active sentences corresponding to the passives in (1) are,
we find that there is only (2a) and not (2b).
(2) a. The bowler fielded the ball well.
b. *The bowler well fielded the ball.
This would be explained if well fielded in (1b) were a predicative, and not
simply the passive of the verb accompanied by a modifier.
But an observer of the game where the incident referred to in (2a) occurred
THE ENGLISH MODIFIER WELL 55
might call out “Oh, well fielded!” and this seems to mean that the deed was well
done, rather than that the ball could be described as a well fielded ball. Another
characteristic of well fielded which seems to make it verb-like is that it has
application only to the period during which the person was engaged in the action
of fielding the ball.
If well fielded is an adjective, as Wasow implies, then it is surprising that it
doesn’t conjoin freely with other appropriate adjectives.
(3) *The ball was heavy and well fielded.
These facts don’t necessarily mean that well fielded is not an adjective, but they
do blur the distinction between adjective and verb.
Similar remarks apply to the sentences in (4).
(4) a. The ball was played well.
b. The ball was well played.
Again, if someone calls out from the grandstand, “Well played!”, the meaning seems
to be that it was well done, not that the ball could be described as a well played ball.
And again, the description well played in (4) seems to have application only
to the period during which the verbal action was carried out. To complete the
parallel between well fielded and well played, we don’t seem to be able to
conjoin well played freely with appropriate adjectives.
(5) *The ball was heavy and well played.
But a very interesting change occurs if we replace the ball by the shot, in (4b),
yielding (6).
(6) The shot was well played.
Suddenly, we find we can conjoin appropriate adjectives with the well expression.
(7) The shot was forceful and well played.
In the place of forceful, we could also have such adjectives as vigorous, aggres-
sive, and elegant.
The only difference between (4b) and (6) is that the shot refers to an action,
whereas the ball refers to a physical object. We can easily invent other such
examples, as long as we choose a subject that refers to an agentive action and
compare it with one that refers to a physical object.
56 RAY CATTELL
(8) a. The ball was well handled.

b. *The ball was white and well handled.
c. The pass was fast and well handled.
It may seem, then, that the crucial distinction is between nouns that refer to
actions, like shot and pass, and ones that refer to physical objects, like ball. But
that is not so.
(9) The rock was heavy and well aimed.
The subject of (9) refers to a physical object, a rock, but the well expression has
been successfully coupled with an adjective. The implication in (9) is that the
rock was being used to injure somebody or something. What is interesting is that
both heavy and well aimed are properties of the rock which would assist that
purpose. The purpose is not overtly stated in (9), but it is implied.
As a matter of fact, if we imagine that someone threw a hard cricket ball
with the intention of injuring someone, we could also have (10).
(10) The ball was hard and well aimed.
So what is wrong with (5) *The ball was heavy and well played is not that the
subject is a physical object, but that it is difficult to think of any purpose for
which both heavy and well played would be effective properties. Similar
comments apply to (8b).
We can now see why (7), The shot was forceful and well played, is com-
pletely acceptable. If a shot is forceful, it is likely to be more effective than if
it is not; and if it is well played, that is an added reason to expect it to be
effective. So predicatives which refer to a property which enhances the prospects
for success of the subject can be conjoined with adjectives which are likewise
success-enhancing for the subject.
It is not difficult to see what might constitute success for a shot (in cricket,
tennis, etc.) but what might constitute success for a ball, in the “physical object”
sense is not clear. And that accounts for the apparent deviancy of (5) *The ball
was heavy and well played.
However, this deviancy seems to be a pragmatic one rather than a linguistic
one. If we put (5) in the context of a cricket match, it is unsatisfactory, because
it is difficult to imagine how the weight of the ball could make much difference
to success or failure, all cricket balls being more or less the same weight. But
suppose someone invented a new game, called “Whammo”, in which the player
had to hit a ball through some sort of barrier, which resisted up to a certain
point, but gave way if the ball developed enough force. And suppose the game
allowed each player to bring his/her own set of Whammo balls, which could be
of different weights. Then (5) would seem to be a perfectly satisfactory sentence.
What was needed to render (5) satisfactory was the introduction of a set of
circumstances in which the properties “heavy” and “well played” would be
success-enhancing properties for the ball. Thus, the general principle that
adjective and predicative must both refer to properties that enhance the prospects
for success is confirmed.
There are some predicatives (and their verbs of origin) whose meaning is
such that success for the imagined agent would mean failure for someone else.
This is the case in (11).
(11) The sword was well parried (by the Black Knight).
The person doing the parrying can only succeed if the person wielding the sword
is not successful at that moment. This makes it more difficult, and perhaps
impossible, to find an adjective which will be predicated of the sword and yet
will suggest increased prospects of success for the parrier. In such a case, the
predicative seems less adjectival, and in fact, the noun phrase in (12) seems deviant.
(12) *A well parried sword.
Example (1b) above (The ball was well fielded by the bowler) seems to be another
such case. Given the rules of cricket, the person fielding can succeed only at the
expense of the batsman: they cannot both succeed at the same time. Again, we
find that the putative noun phrase in (13) is deviant.
(13) *A well fielded ball.
To sum up, then, the well expressions discussed in this section are partly verb-
like and partly adjective-like. Their verb-like properties include:
A. The description given by the well expression seems to have relevance
only to the period during which the (verbal) action was going on.
B. The expression can occur in the full range of tenses and aspects.
Although there are adjectives which can occur in progressive aspect (She
is being careful, He is being good, and so on), in such cases the adjective
is predicated of the subject with respect to his/her own actions. In The
ball is being well fielded, the predication is made to the subject, the ball,
but with respect to an action other than its own. This is, of course,
because the expression retains its passive orientation.
58 RAY CATTELL
The main evidence for the adjective-like nature of the predicatives is that they
can conjoin with adjectives. But, as we have seen, this is possible only under
certain conditions, namely, that both the adjective and the predicative refer to
properties that enhance the prospects of success for some project. That project is
usually (always?) not explicit but implied, as in The spear was sharp and well
aimed. The implied project here is to wound or kill some person or animal.
Rather different from the cases we have been considering so far are the
sentences in (14).
(14) a. The manuscript is well typed.
b. The garment is well knitted.
The predicatives in these cases are not restricted in reference to the period during
which the relevant (verbal) action was going on; in fact, they are more likely to
be used with “present” application after that action is finished. Thus, (14a) has
the interpretation “The manuscript has been well typed” and (14b) “The garment
has been well knitted”. The properties well typed and well knitted will continue
to be valid throughout the life of the manuscript/garment; unlike well aimed in
The spear was well aimed.
Predicatives like well typed and well knitted seem more adjectival than ones
like well played and well aimed, which in turn seem more adjectival than ones
like well fielded and well parried. Thus, a well typed manuscript and a well
knitted garment are excellent noun phrases; so are a well played shot and a well
aimed spear; however, a well fielded ball and a well parried sword are dubious.
There seems to be a graded scale of adjectivehood, then, amongst the
predicatives that have the form of passive participles.
2.2 Well = “favourably”
There is a rather different interpretation for well in the sentences in (15):

(15) a. The warders treated the prisoners well.
b. The public received Harry’s book well.
c. The critics reviewed the film well.
(15a) does not mean that the warders performed efficiently the task of treating
the prisoners, but that they treated the prisoners in a way that was favourable to
them. Likewise, (15b) does not mean that the public performed efficiently the
act of receiving Jane’s book, but that they received it favourably. And (15c) does
not necessarily mean that the critics did an efficient job of reviewing the film
(although there is one reading along those lines). It can mean that the critics
reviewed the film favourably. In each of these examples as described, then, the
interpretation of well seems to be “favourably” rather than “efficiently”. Why
should there be this difference?
When we look for the passive forms corresponding to the sentences in (15)
we again find there are two kinds of candidates. The forms in (16) are the verbal
passive forms, and the ones in (17) are the ones we have been calling predicatives.
(16) a. The prisoners were treated well by the warders.
b. Jane’s book was received well by the public.
c. The film was reviewed well by the critics.
(17) a. The prisoners were well treated by the warders.
b. Jane’s book was well received by the public.
c. The film was well reviewed by the critics.
In both (16) and (17), the by-phrases may be absent.
We can conjoin an ordinary adjective to the left of each of the expressions
in (17), whether the by-phrase is included or not.
(18) a. The prisoners were comfortable and well treated (by the warders).
b. Jane’s book was timely and well received (by the public).
c. The film was successful and well reviewed (by the critics).
What is interesting is that the adjectives which are conjoined with the predicat-
ives all represent properties that are “favourable” to the subject. So there is a
harmony between the adjective and the predicative in this regard, in each case.
Another interesting fact is that the by-phrases seem to be quite acceptable
as part of the predicative phrase, whereas it does not seem nearly as good if we
include the by-phrases in the following cases:
(19) a. ?*The shot was forceful and well played by the batsman.
b. ?*The spear was sharp and well aimed by the hunter.
c. ?*The manuscript was attractive and well typed by the
secretary.
If these judgements are correct, it means that the type of predicative in which
well has the meaning “effectively, efficiently” cannot take by-phrases as
complements, but the type in which well has the meaning “favourably” can. In
the former case, the by-phrases must be complements of the main verb, but in
60 RAY CATTELL
the latter case they must be complements of the predicative.

Seemingly related to this contrast in complementation is the fact that the
“active” sentences in (15) (where well means “favourably”) can be matched by
the close paraphrases in (20):
(20) a. The warders gave the prisoners good treatment.
b. The public gave Jane’s book a good reception.
c. The critics gave the film good reviews.
Give is a light verb,2 and the real predication in these sentences is (good)
treatment, (a) (good) reception and (good) reviews respectively, where these
phrases correspond to treat … well, receive … well and review … well respective-
ly. The passives of (15) shown in (16) correspond well in meaning to the
passives formed from (20); that is, the sentences in (21):
(21) a. The prisoners were given good treatment by the warders.
b. Jane’s book was given a good reception by the public.
c. The film was given good reviews by the critics.
But even more interesting, in the present context, is the fact that the sentences in
(17), which contain predicattive phrases, are closely matched by the following
sentences containing the “passive” light verb get.
(22) a. The prisoners got good treatment (from the warders).
b. Jane’s book got a good reception (from the public).
c. The film got good reviews (from the critics).
There seems to be the same relationship between passive give-sentences and get-
sentences as between the “ordinary verb” passive sentences in (16) on the one
hand and the sentences in (17) containing predicative phrases on the other. The
fact that the get-sentences have a passive kind of meaning, though they lack an
equivalent active form, is strikingly parallel to the way in which other predicative
phrases have a passive kind of meaning, though they have no corresponding
active forms.
There are no such relationships in the cases where well means “efficiently”
or “effectively”.
(23) a. Paul played the shot well.
b. *Paul gave the shot a good play.
c. The shot was well played.
d. *The shot got a good play.
(24) a. The hunter aimed the spear well.

b. *The hunter gave the spear a good aim.
c. The spear was well aimed.
d. *The spear got a good aim.
And so on. Only the examples where well means “favourably” have equivalent
light verb sentences with give and get. Presumably this relationship has a lot to
do with the lexical relationship between the relevant nouns and verbs. And
presumably the fact that we cannot have *The bowler gave the ball a good field,
or *The ball got a good field from the bowler is due to the fact that there is no
noun field with just this sense. Why that is so is another question, which this
paper will do nothing to resolve. But the same semantic factor that governs the
occurrence of paraphrasing get-sentences seems to influence the meaning of well
so that it is interpreted as “favourably”.
In the get-sentences of (22), it is clear that the from-phrases are comple-
ments of the predicational nouns. On the other hand, in the related passive give-
sentences, it is clear that the by-phrases are the complements of give.
If the from-phrases in (17) are similarly complements of the predicatives
which they follow, and not of get, it helps distinguish them from ordinary
passives, where it is clear that the by-phrases are the complements of the verbs.
There is an interesting symmetry about all of this.
3. “Quantity” well
All the cases we have considered so far have been ones in which the well-
phrases suggest notions of “quality”. But there is another group of well-expres-
sions in which it is “quantity”, rather than “quality”, that is suggested. Consider
the sentences in (25):
(25) a. The champion was well extended by the other swimmers.
b. The shirts were well advertised by the store.
c. The sisters were well separated by the other runners.
These are all examples of what is meant by “quantity” or “degree” well. Perhaps
(25b) can also be given a “quality” reading, but the more prominent reading is
a “quantity” one.
62 RAY CATTELL
Note that in all three sentences, the period during which the predicative has
relevance is equal to the period occupied by the implied verbal action; hence
they resemble the “quality” sentences like The ball was well fielded by the bowler
in this regard.
As in the case of that sentence, too, it is not easy to think of adjectives that
could be coordinated with the well-expressions. In the case of well advertised, it
does seem to be possible if we leave off the by-phrase, but with the by-phrase
it doesn’t seem as good.
(26) The shirts were cheap and well advertised (?by the store).
Cheap and well advertised combine successfully in the shorter version because
it is not difficult to imagine a common objective for which they would both
enhance the prospects of success; that is, the objective of selling a lot of shirts.
Significantly, it is also true that these three sentences cannot be paraphrased
by passive give-sentences.
(27) a. *The champion was given a good extension by the other
swimmers.
b. *The shirts were given a good advertising by the store.
c. *The sisters were given a good separation by other runners.
Nor can these expressions be paraphrased by get-sentences.
(28) a. *The champion got a good extension from the other
swimmers.
b. *The shirts got a good advertising from /by the store.
c. *The sisters got a good separation from the other runners.
As in the case of “quality” well-expressions, however, there are some “quantity”
well-expressions which can be paraphrased by give- and get-sentences. (For the
former, compare well treated, well received, well reviewed, discussed earlier.)
(29) a. The library is well used (by the students).
b. The executives were well indoctrinated (by the company).
c. The canteen is well patronised by the employees.
(30) a. The library is given good use (by the students).
b. The executives were given a good indoctrination (by the
company).
c. The canteen is given good patronage by the employees.
(31) a. The library gets good use (from the students).

b. The executives got a good indoctrination (from the company).
c. The canteen gets good patronage from the employees.
As in the earlier cases, the sentences in (29) can have adjectives conjoined to the
left of the predicatives, and this is true whether the by-phrases are present or not.
(32) a. The library is comfortable and well used (by the students).
b. The executives were loyal and well indoctrinated (by the
company).
c. The canteen was convenient and well patronised by the
employees.
And in each case, the adjective and the predicative both represent similar
properties, which will contribute to a common goal.
However, these cases must be distinguished from certain others where the
predicative sentences can also be paraphrased by give- and get-sentences.
(33) a. The boy was well thrashed by his father.
b. The device was well tested by the engineers.
c. The volunteers were well warned (about the dangers) by the
commander.
These resemble the sentences in (25) in that in each case the predicative seems
to have force only for the duration of the action, and in that it is difficult to
think of adjectives which will conjoin, either with or without the by-phrases.
However, the sentences in (33) can be paraphrased by give-sentences.
(34) a. The boy was given a good thrashing by his father.
b. The device was given a good testing by the engineers.
c. The volunteers were given a good warning (about the
dangers) by the commander.
Furthermore, there are get-sentences of the kind we have been associating with
the “predicatives”.
(35) a. The boy got a good thrashing from his father.
b. The device got a good testing from the engineers.
c. The volunteers got a good warning (about the dangers) from
the commander.
64 RAY CATTELL
There do not seem to be any similar cases involving the “quality” well-expres-
sions; only ones like these which are “quantity”, or “degree” ones. All the other
kinds of well-sentences that can be paraphrased by get-sentences can have
adjectives coupled with the predicative expressions. But these cannot.
(36) a. *The boy was guilty and well thrashed by his father.
b. *The device was ingenious and well-tested by the engineers.
c. *The volunteers were intelligent and well warned (about the
dangers) by the commander.
If the verb be is repeated before the predicative, of course, felicitous sentences
will result, but that is irrelevant to the point being made.
So what makes these so different? Notice that all the predicational nouns in
the give- and get-sentences in (34) and (35) end in -ing. In cases like thrashing
/ beating / spanking / smacking / tongue-bashing, and so on, this ending in itself
suggests a certain (high) degree of activity, even without the well. It is similar
with testing. Warning is different, however, and requires the addition of good to
suggest a high frequency or intensity of warning. There is obviously more to be
said in that direction, but for the moment I am mute.
In any case, well-expressions that are matched by the -ing type of nominals
in related give- or get-sentences cannot be conjoined with adjectives. The only
well-expressions that can be conjoined with adjectives are those whose related
nominals in give/get sentences are of other kinds. In fact, the two kinds of
nominals are similar in form to gerundive and derived nominals respectively, in
the sense of Chomsky (1970), although the structures being discussed here are
somewhat different. The appropriate conclusion seems to be that these predicat-
ives are at the other end of the spectrum from proper adjectives.
When we take into account all the different kinds of well-expressions
discussed in this paper, it seems clear that there is a continuum of “passive
participle” interpretations, running from more to less “adjectival”.3 The awkward
fact seems to be that at this point of the grammar, two categories flow together.
Notes
1. Bolinger (1972) gives a very insightful account of the semantics of well-expressions, but along
rather different lines from those followed here.
2. For further discussion of light verbs in English, see Jackendoff (1974), Cattell (1984),
Jayaseelan (1988) and references therein.
3. It is true that most characteristics described by adjectives can change over time. The description
Sue is thin is, of course, valid only while Sue is thin. (Shades of Tarski!) But it is not the
duration of an action referred to by a verb that determines the duration of her thinness. So what
we are talking about in the discussion about well fielded, well played, and so on is not simply
a general property of adjectives. The link with verbs means that these adjectives, if that is what
they are, are rather different from the common herd.
References
Bolinger, D. 1972. Degree Words. The Hague: Mouton.

Cattell, R. 1984. Syntax and Semantics, Vol. 17: Composite Predicates in English.
Sydney: Academic Press.
Chomsky, N. 1970. “Remarks on Nominalization”. Readings in English Transfor-
mational Grammar, ed. by R. Jacobs and P. Rosenbaum, 184–221.
Waltham, Mass.: Ginn and Company.
Jackendoff, R. 1974. “A Deep Structure Projection Rule”. Linguistic Inquiry 5.
481–505.
Jayaseelan, K. A. 1988. “Complex Predicates and q-Theory”. Syntax and Seman-
tics, Vol. 21: Thematic Relations, ed. by W. Wilkins, 91–111. San Diego:
Academic Press.
Wasow, T. 1977. “Transformations and the Lexicon”. Formal Syntax, ed. by
P.W. Culicover, T. Wasow and A. Akmajian, 327–360. New York: Aca-
demic Press.
The deictic-presentation construction in English*
Peter Collins
1. Introduction
In this paper I shall consider sentences of the type in (1) below.1

(1) a. # ^there’s his BOOK # [LL 2.6 96]
b. “No. Now dammit, I don’t want to go into any more explanations.
Here comes Jason. Keep this to yourself”. [Brown N04 30]
They will be regarded as instances of a construction which in English prototypic-
ally has the following configuration: Here/There + V + NP (where “V” repre-
sents a small set of verbs — be, come and go in the vast majority of cases). The
word “prototypically” is used here to allow for the inclusion of instances where
the NP precedes the verb (as it typically does when the NP is realized as a
nominative-case pronoun, as in Here we are!), and for cases where the NP is
followed by further material — what I shall call the “extension”.2 The term
“deictic” is intended to reflect the fact that here and there are, characteristically,
interpretable only relative to the context in which the sentence is uttered, while
“presentation” is intended to reflect the typical pragmatic function served by
instances of the construction of directing the addressee’s attention towards an
entity in the context of the situation and/or discourse.
Some languages have a fully grammaticalized deictic-presentation construc-
tion. Witness the grammatical specialization that has occurred in French, for
example, where deictic-presentation (introduced by voici or voilà) is structurally
distinct from existential predication (introduced by il y a). In English the deictic
presentation construction has evolved historically as a specialization of the more
general inversion construction involving the fronting of a complement (as in
More important are the moral objections) or of an adjunct (as in Over the hill
68 PETER COLLINS
appeared the cavalry), accompanied by postposing of the subject.3 As a result it

is sometimes difficult to determine whether a particular example falls under
deictic-presentation or the general construction. Consider:
(2) a. Spectators are gathered on the opposite cliff, cut off from
me by a chasm, and waiting for the chief event of the sports.
Here are townspeople and their visitors, with a few rustics from
the mountains inland. [LOB K12 64]
b. Much of this goes on in offices high up in Wall Street’s lofty
wind-swept towers.
There sit men who make moving averages of weekly volume,
monthly averages of price earnings ratios [Brown J39 460]
c. A little to the westward of this sounding lay an abyss of eigh-
teen hundred fathoms, beyond the reach of any trawl and the
haunt of alien species. Here swam the snake-like Cyclothones,
the rat-tailed Chimerae, [LOB N27 113]
These examples form a gradient. The further one moves from the prototype the
more likely it is that the following features will be present:
(i) The initial adverbial is likely to have an anaphoric rather than deictic role
and thus be replaceable by an earlier locative expression (on this cliff in
(2a), in these offices in (2b), and in that abyss in (2c)).
(ii) The verb will not be be, come or go, or be readily replaceable by one of
these, but rather a verb with more complex semantic features.
(iii) The verb will not be simple present (swam in (2c)).
(iv) There will be equivalence with a non-inverted alternant. (Each of the
examples in (2) satisfies this criterion: compare the non-equivalence
between Here comes Jason and Jason comes here in (1b).)
In the following two sections arguments are presented to support the claim that
a deictic-presentation construction has evolved in English, and the claim that it
is not as yet fully grammaticalized. Subsequent sections will pursue a detailed
examination of the various components of the construction (the initial deictic, the
verb, the post-verbal NP, and the extension).
THE DEICTIC-PRESENTATION CONSTRUCTION IN ENGLISH 69
2. A distinct construction
The position adopted in this paper is that there has been enough specialization in
English to warrant recognition of a deictic-presentation construction. One sign
that deictic-presentation is a separate construction is the semantic/pragmatic
difference between an example of the type in (3a) and its non-presentational
counterpart (provided in (3a′)):
(3) a. # the Dean has now emerged and is coming first out of the
sacRArium # and ^here come the CLERgy # [LL 10.6b 50]
a′. the clergy come here
These are not merely stylistic equivalents. In (3a) the verb has an instantaneous
interpretation (and would thus be incompatible with an adjunct such as some-
times), and the clergy has a specific interpretation; in (3a′) the verb has a generic
or iterative interpretation (and accordingly would be incompatible with an adjunct
such as now in most contexts), and the clergy has a non-specific interpretation.
In some cases of deictic presentation a non-presentational counterpart is not
available at all, because the post-verbal NP could only be interpreted specifically
and the verb only non-iteratively. Examples follow (compare *The announcement
comes here and *A cock goes there):
(4) a. # ^here comes the anNOUNCEment # [LL 10.3 96]
b. # – – er in ^fact there goes a COCK # ^LOOK #
[LL 10.8a 44]
The non-availability of an acceptable non-presentational alternant in cases of this
type is a strong indication that the deictic presentation construction has developed
special features in English such that it cannot be merely subsumed under a
general inversion analysis. Such non-availability is clearly associated with
idiomatic uses of the construction, as in the following:
(5) a. “Never!” Andrea declared passionately. “Never! He has
robbed me — ”
“And here is your chance to make him pay back!” Madam
interrupted. [LOB L08 18]
b. “Don’t start making a fuss all over again Charlotte, there’s a
good girl” Esmond said. [LOB P24 16]
c. Gaffer lifted the half-empty whisky bottle from the sink-
board and sloshed a liberal quantity into a tumbler. “Here’s
70 PETER COLLINS
to your information being correct, Cuddy, because if it isn’t …

” [LOB L10 130]
d. “Have a look first, Wull. You’ll need to work fast. The
water’s cold.”
“I ken fine it’s cold. I can feel it. Here goes,” and Willie
eased himself overside into the sea. [LOB N24 190]
There are several further properties of the deictic-presentation construction which
suggest that it has attained the status of a distinct construction. The first is that
here and there have assumed certain subject properties, and the post-verbal NP
correspondingly lost some. The pressure to interpret here/there as the subject is
particularly strong when the post-verbal NP is realized as an accusative pronoun,
as in There’s me (see further Section 4 below). A similar point could be made
with respect to the existential there construction. Notice, however, that while the
two constructions share a superficial grammatical similarity, they are prosodically
quite different: on an existential reading There’s me (as, perhaps, a response to
a question such as “Who’s planning to come to the theatre?”) will have stress on
the personal pronoun, while on a deictic reading there will be stressed.
A further distinctive property of the deictic-presentation construction is that
a nominative pronoun may appear as the post-verbal NP, as in There am I,
covered in mud, … (for which there is no non-inverted counterpart (?I am there,
covered in mud, …). This is not possible with general inversion (*In the corner
am I). Furthermore, with deictic-presentation there is alternation between
instances with a nominative pronoun in pre- and post-verbal position, but not
with general inversion (In the corner I am would not qualify as inversion, but
rather simply as fronting).
Finally, another property which indicates the specialization of the deictic-
presentation construction is the restriction of the set of verbs that may be
selected to a small subset of those that are used in “locative inversion” (see
further Section 6 below).
3. A semi-grammaticalized construction
Two pieces of evidence are advanced here for the claim that the deictic presenta-
tion construction is as yet only partly grammaticalized: its grammatical inflexibil-
ity and its grammatical indeterminacy. The inflexibility of the construction is
evident in its inability to enter into a wide range of syntactic processes. There is
no interrogative clause type:
(6) a. *Is there my pen?
b. *Does here come the bus?
Negation is not possible:
(7) a. *There is not my pen
b. *Here doesn’t come the bus
Subordination, though not impossible, is restricted. It is possible to subordinate
a prototypical deictic-presentational as a content clause in certain cases where it
clearly encodes new information (as in (8a)), but not otherwise (compare the
unacceptability of (8b)):
(8) a. I’m glad to say that here comes the bus
b. *I know that here comes the bus
Secondly, we may note the indeterminacy of the structure. The deictic element
bears some resemblance to a pronominal deictic subject, as can be seen from the
possibility of substituting this for here in an example such as (9):
(9) Less committed critics, however, were faced with two unpleasant
alternatives. Here was undoubtedly a bad opera; [LOB G44 187]
In some such cases the construction is not sharply distinct from identifying
constructions with subject-complement switch:
(10) a. Here is the best short explanation of the Cold War that has
been written. [Brown J36 1640]
b. You’ll need your Weider Power Stands for this fine exercise
and here’s the way it’s done: [Brown E01 1630]
Although they are in some ways similar to a pronominal, the deictic here and
there of the deictic presentation construction have not been “subjectivalized” as
much as existential there, which may invert with the operator in interrogative
clauses as in Is there anyone home?, may govern subject-verb agreement as in
There’s two biscuits left, and may occur before an ellipsis-site as in Tom believes
there’s still a chance of winning, but there isn’t. The subject properties of here and
there in the deictic-presentation construction will be explored in the next section.
72 PETER COLLINS
4. Subject properties of deictic here and there
Deictic here and there have some subject properties, albeit rather ambivalent,
relating to subject-verb agreement. When the post-verbal NP is plural, number
agreement is typically determined by that NP (more often with here than there
— the low figures with there perhaps resulting from the possibility of miscon-
strual with existential there). Examples follow:
(11) a. # and ^here COME # the ^Life GUARDS # ^trotting DOWN
# to^wards the ABbey # [LL 10.6b 601–4]
b. Aside from the Ruger carbine, a number of hunting rifles
have been introduced for the first time. Here are the brands
(in alphabetical order) and the new models. [Brown E10 780]
With other types of example number agreement is determined by the deictic item
rather than the post-verbal NP, as in:
(12) a. # ^here IS # ^Captain Mark PHILLips # and his ^best MAN
# [LL 10.6b 718–9]
b. “There goes our grub an’ ammunition”! [Brown N03 1550]
The fact that in (12a) it is here, rather than Captain Mark Phillips and his best
man that determines number agreement is confirmed by the fact that in basic
clauses a coordination of NPs takes a plural verb (as in Captain Mark Phillips
and his best man are here). A similar argument is applicable to (12b).
Predictably, the substitution of ’s for are is more common in informal than
formal registers in the corpus, as in:
(13) # ^there’s the two teachers that come UP WITH them # [LL 4.7 98]
Corpus examples with a personal pronoun in post-verbal position are rare.
Because of their typically ‘given’ informational status personal pronouns, like
proforms in general, are normally disfavoured in clause final position. The first
example in (14) below has the informationally unmarked NP + V ordering, the
second the marked V + NP ordering (with the focus on is in the former, on he
in the latter):
(14) # … er ^here’s er a frusTRATed # ap^parently unhappy WIFE #
con{FIDing} IN him # ^there she IS # with this bungalow on the
outskirts of HOUNslow # – – – er ^there is HE # with his er
^{SHOOTing} VAN # ^doing ROUNDS # er involving regular
journeys to DITton # [LL 12.3 322–3]
When the displaced subject is a personal pronoun in accusative case, agreement

is invariably determined by the deictic element. Compare the following invented
examples (to be understood as occurring in a context in which speaker and
addressee are, say, looking at a set of photographs):
(15) a. There’s me/us/?them
b. *There am me/are us/are them
The very possibility of accusative forms here provides further evidence for
treating the deictic element as the subject (the accusative form of the pronouns
suggests strongly that they do not function as subject, leaving no other candidate
for subjecthood than the deictic element). By contrast, when the pronoun is
nominative, agreement is determined not by the deictic element but by the
pronoun (as in There am I/I am, digging away furiously, when this car pulls up).
It is notable that lack of agreement with the post-verbal NP is more likely
with contracted ’s than uncontracted is (so that There’s me, second from the left
is more likely than There is me, second from the left), presumably because the
former has more of the character of a clitic, and clitics do not govern agreement.
5. Deictic reference of here and there
In general terms here and there contrast as “proximal” (close to speaker) and
“distal” (distant from speaker) deictics. Compare:
(16) a. Do you think I ought to be holding a gun in one hand and a
pair of handcuffs in the other? Here is my warrant. You will
verify, I hope, that it is perfectly in order.’ [LOB N10 182]
b. When they were ready to leave, Benson and Ramey walked
back around the rear of the trailer.
“There’s a body you won’t mind looking at”, Benson said and
they stopped. She had driven up with her husband in a convert-
ible with Eastern license plates. [Brown N22 480]
In (16a) the speaker is holding the warrant, but in (16b) the woman is observed
in the convertible that has just arrived.
A feature of the sporting commentary genre, in which instances of the
deictic-presentation construction are very common, is the neutralization of the
distal-proximate distinction. In the following small selection from horse racing
commentaries, here and there are in virtually free variation:
74 PETER COLLINS
(17) a. # and ^here’s Polar WAY — ^number THREE #

[LL 10.4c 620]
b. # and ^here’s number {THREE} Pretty CAGE #
[LL 10.4d 1018]
c. # now there’s er ARCHie’s rider #
[LL 10.4c 624]
d. # – – – and ^there’s What NOW # ^number SIX #
[LL 10.4d 991]
There may be used for situations in which an entity moves away from the
speaker (and is thus compatible with go). In (18) the man referred to is disap-
pearing from the scene (notice the occurence of went in the final clause):
(18) “From now on I am going to walk a little among the hills every
day, extending the length of my walks slightly each day till I can
get myself back in condition among these wonderful mountains.”
“Now there’s a man I’d tie to, if he ever give me the chance,” the
constable told himself happily as Goddard went into the hotel.
[LOB N03 27]
By contrast here is used for situations in which an entity moves towards the
speaker (and is thus compatible with come). In (19) Jason is understood to be
approaching the speaker:
(19) “No. Now dammit, I don’t want to go into any more explanations.
Here comes Jason. Keep this to yourself”. [Brown N04 30]
A metaphorical extension of the there goes versus here comes distinction occurs
with “discourse deixis” (or “text anaphora”). Here may be used to anticipate a
proposition that is yet-to-be-presented, whereas there is used with anaphoric
reference to an earlier proposition. Thus Here’s an interesting suggestion would
most likely be used to refer to a suggestion not-yet-made, but There’s an
interesting suggestion to one already made. (Note that a comparable distinction
occurs with this and that: This is what you should do anticipates advice, whereas
That’s what you should do recapitulates.) In the following corpus examples here
has forwards-anaphoric reference, there backwards-anaphoric reference:
(20) a. And here’s an idea to help avoid any ill-effects. Next time you
have pork make a point of having stewed fruit in the meal.
[LOB F33 33]
b. But farther North? One has to book at least six weeks in

advance. Put on extra flights? There’s another idea for the
waste-paper basket.
Yours faithfully,
Mark Murray Threipland. [LOB B09 117]
Writers and speakers sometimes exploit the proximal-distal distinction associated
with basic uses of deictic here and there for rhetorical purposes. For example, in
(21) the American writer indicates his concern for the distant region in question,
and his empathy for its people, by selecting here:
(21) But, just as we drew on Europe for assistance in our earlier years,
so now do these new and emerging nations that do have this faith
and determination deserve help. Respecting their need, one of the
major focal points of our concern is the South-Asian region. Here,
in two nations alone, are almost five hundred million people, all
working, and working hard, to raise their standards, and in doing
so, to make of themselves a strong bulwark against the spread of an
ideology that would destroy liberty. [Brown G35 730]
6. The verb
As mentioned above, the verb is prototypically be, normally the simple present
form is (which in speech is usually contracted to ’s).4 Next in order of frequency
are come and go (whose typical correspondence with here and there respectively
has already been discussed).
Next in frequency are verbs of “spatial configuration” (to borrow Levin and
Rappaport’s (1994) term), those which add components of positional information
to a basic existential meaning, and which are potentially substitutable by the
appropriate form of be:
(22) a. At the distant edge of the river, I caught a glimpse of roofs
and chimneys, and the quick glitter of glass that marked the
hot-houses in the old walled garden that had belonged to the
Hall. There, too, lay the stables, and the house called West
Lodge, [LOB L09 65]
b. There, in the centre of the “Ring”, stands a magnificent statue
of Jan Hus, the Bohemian reformer. [LOB G66 12]
76 PETER COLLINS
Occasionally verbs with more specific semantic features are found, but they tend
to produce a literary or archaic effect:
(23) Hudson’s first error of the fourth voyage occurred only a few
miles down the Thames. There at the river’s edge waited one Henry
Greene, whom Hudson listed as a “clerk”. [Brown F16 920]
The strong tendency for verbs other than be, come and go to be realized as past
forms (as in (22a) and (23)) indicates that the sentences containing them are non-
prototypical.
The verbs permitted in the deictic-presentation construction appear to be a
subset of the several hundred that Levin and Rappaport (1994) list as occurring
in “locative inversion”. Most of these would be barely possible in deictic-
presentation, and if used would produce a sentence that was at best peripheral to
the construction. Compare:
(24) a. In this factory worked a large number of migrant women
a′. ?There worked a large number of migrant women
(25) a. Through the tunnel rumbled the Southern Aurora
a′. ?Here rumbled the Southern Aurora
7. The post-verbal NP
The post-verbal NP must normally have a specific interpretation, so that for an

example such as (26), where the NP contains a non-specific quantifier, an
existential interpretation is possible, but not a deictic interpretation:
(26) There are many/few species of sheep
The NP cannot contain no as determiner (a particular case of the restriction on
the deictic presentation construction to positive polarity). The following could
only carry an existential interpretation:
(27) There are no sheep
Counterexamples occasionally occur, but in tokens that are at best marginal. For
example in (28) here is clearly anaphoric rather than contextually deictic, and the
sentence is pragmatically close to an existential (compare Here there is little
scope for original thought):
(28) Some of the first attempts at “composition” will be the retelling of

stories read or heard. Here is little scope for original thought, and
the severe discipline of the sequence of detail has to be observed.
[LOB E24 139]
8. The extension
The term “extension” seems to be more appropriate for the post-verbal material
in the deictic-presentation construction than the existential construction. Compare:
(29) a. There’s someone working after hours
b. There’s John working after hours
In the case of (29a) (on its more salient existential reading) only one proposition
is expressed: it is predicated of “someone” that they are working after hours. In
the case of (29b) two propositions are expressed: it is predicated of John that
“He is there”, and that “He is working after hours”. In the deictic-presentation
construction the extension presents ancillary information, and it is not surprising
that in the vast majority of instances it is not present (often separated from what
precedes it by a comma in writing or a pause in speech), whereas in the existen-
tial there construction it presents integral information, and is rarely absent.
One piece of evidence that the extension represents a separate proposition
from the deictic proposition, as noted by Lakoff (1987: 502), derives from the
possibility of inserting different types of temporal adjunct into the primary
clause and the extension, as in:
(30) There’s John again, always working after hours
Here again indicates a repeated state-of-affairs, while always indicates one that
is continuing: they could not be applied simultaneously to the same situation
(*There’s John again always; *John is again always working after hours).
A variety of types of phrase and non-finite clause may function as the
extension. The following examples are all taken from the corpus and are unlikely
to exhaust all the possible types.
(i) Prepositional phrase (expressing, for example, location as in (31a), accom-
paniment as in (31b), and attribution as in (31c)):
78 PETER COLLINS
(31) a. # ^there’s the man round the CORner # [LL 2.7 1329]
b. # so ^here are the Snowdons and the Gloucesters and the
Kents and the OGilvys # — with ^several of their CHILDren
# [LL 10.6b 485–6]
c. # ^here COMES # the ^special TRAIN. {^WITH #}# the
^decorated HEADboard # — the ^four saloon COACHes # .
^and . the royal saLOON # [LL 10.7b 567–71]
(ii) Adjectival phrase:
(32) “Here are two old men, mad at each other and too proud to pick up
the phone”, said a House Democratic leader. [Brown A37 1530]
(iii) Non-finite clause (infinitival as in (33a), present-participial as in (33b), and
past-participial as in (33c)):
(33) a. Here are some key areas to examine to make sure your
pricing strategy will be on target: [Brown E28 200]
b. Here, in two nations alone, are almost five hundred million
people, all working, and working hard, to raise their standards
[Brown G35 730]
c. # ^here is this station PLATform # — ^BEAUTifully
decorated # ^as ALways {for these ^state ocCASions #}#
[LL 10.7b 347]
(iv) Absolute with-construction:
(34) a. Here and there on work table or pedestal stood a shape with a
sheet or a tarpaulin draped over it. [Brown L11 1640]
b. One day you will go to the door and there will be a little
envelope with a publisher’s name on it; [LOB K25 168]
In general the types of element that can function as the extension are those
which could function as complement to be if the extension were expanded to full
finite clausal status. This generalization appears not to work with absolute
constructions, which in most cases lend themselves more readily to expansion
with have than be. However the differences between absolute constructions and
the other types would be reduced under an analysis in which all extensions were
analyzed as elliptical clauses in which the post-verbal NP functions as the
ellipsed subject (in the case of (34a) and (34b), The shape had a sheet or a
tarpaulin draped over it and The little envelope will have a publisher’s name on it
respectively). Some evidence supporting an ellipsis analysis is available in (32)

from the presence of each other in the AdjP extension. As Huddleston (1988:100)
notes, each other is best analyzed as a single (“reciprocal”) pronoun which
occurs under the same structural conditions as reflexive pronouns in their basic
use — that is, in the straightforward cases, in the same clause as its antecedent.
In (32), on the assumption that the extension is separate from and independent
of the primary clause (a “peripheral dependent”) each other could only meet this
structural condition if mad at each other were regarded as an elliptical version of
(the) two old men are mad at each other.
9. Conclusion
We have argued that English has a deictic-presentation construction, albeit one

whose grammatical inflexibility and indeterminacy suggest that it is not fully
grammaticalized. In prototypical instances deictic here or there is not replaceable
by an alternative locative expression, the verb is a simple present form of be,
come or go, and there is no non-inverted alternant.
Notes
* This chapter stems from the Cambridge English Grammar project supported by the Australian
Research Council. I wish to acknowledge that Rodney Huddleston provided me with most of
the initial ideas on deictic presentation that are developed here. More generally, I wish to thank
Rodney personally for his immense intellectual generosity over the years. Thanks are also due
to Peter Peterson, David Lee, Ray Cattell and Pam Peters for helpful discussion of the issues
discussed here. I am solely responsible for the final product.
1. This study is based on an exhaustive search of three standard corpora of contemporary English,
the Lancaster-Oslo/Bergen Corpus (“LOB”), Brown University Corpus (“Brown”), and the
London-Lund Corpus (“LL”). The location of each example is provided in square brackets as
follows: corpus, text, line/tone unit number(s). Transcription symbols used for LL are: # (end
of tone unit), ^ (onset), —. (pause markers), CAPS (syllable carrying nuclear tone), {}
(subordinate tone unit).
2. The term “extension” is borrowed from Hannay (1991) who uses it to refer to the comparable
section of the existential there construction.
3. This construction is commonly referred to as “locative inversion”, but Huddleston suggests
(personal communication) that this term is too narrow, since non-locative elements may be
inverted, and that a more appropriate term would be “S-P-X inversion” (where “X” applies to
80 PETER COLLINS
both complements and adjuncts as in the examples above).

4. In the corpus is was contracted to ’s on 72% of occasions in speech, but only 35% in writing.
References
Hannay, M. 1991. English Existentials in Functional Grammar. Dordrecht: Foris.

Huddleston, Rodney. 1988. English Grammar: An Outline. Cambridge: Cam-
bridge University Press.
Lakoff, George. 1987. Women, Fire and Dangerous Things: What Categories
Reveal about the Mind. Chicago: Chicago University Press.
Levin, B. and Rappaport, M. 1994. Unaccusativity. Cambridge: Cambridge
University Press.
Relative clauses
Structure and typology on the periphery of standard

English
Bernard Comrie
1. Relative clauses in standard English
Most accounts of relative clauses in standard English agree that, on one relevant
dimension, two types can be recognized, the first introduced by a wh-word, the
second introduced by that or zero. The analysis of the structure of the first type
is also reasonably uncontroversial, at least to a certain level of detail and theory-
specificity. The wh-word has clear nominal properties, for instance in that it can
be preceded by a preposition, as in (1), and has inflectional forms, as in (2) and,
in certain styles, (3).
(1) the stick with which the man chased the dog
(2) the man whose dog ran away
(3) the man whom I saw
Moreover, the wh-word occurs at the beginning of the relative clause, more
accurately as part of the first major constituent of the relative clause, but must be
related to a syntactic position (or “gap”) that would more often than not occupy
a different position in the linear order of clause constituents, a striking feature
in a language with word order as tightly constrained syntactically as English.
Thus, the instrument wh-expression of (1) would normally, and the direct object
wh-expression of (3) almost invariably, occur postverbally, if they were ordinary
noun phrases, as in (4) and (5).
82 BERNARD COMRIE
(4) The man chased the dog with the stick.

(5) I saw the man.
Different theoretical approaches may differ in how the clause-initial wh-position
is to be related to the non-initial position (the “gap”) that such ordinary noun
phrases with the same syntactic-semantic function would occupy, for instance in
whether or not they would envisage a process of movement, but there is a
consensus analysis that such a relation must be established in some way.
When one turns to relative clauses introduced by that or zero, this consensus
disappears, with at least two radically different analyses being proposed in the
recent and current literature. In what follows, I will use primarily illustrations
with that, although in general the same points will apply to the relative clause
introduced by zero.
Quirk et al. (1985: 366, 1066, 1244) argue that the relevant occurrences of
that are relative pronouns, and by extension, the same is true of the relevant
occurrences of zero. Thus, in discussing (6), they say “the relative pronoun is
subject”, while in discussing (7) they say: “The relative pronoun is to be
distinguished from the subordinator that, which does not operate as an element
in the subordinate clause”.
(6) The news that appeared in the papers this morning was well received.
(7) The style that we are examining in this exhibition was an unusual one.
By contrast, Huddleston (1984: 397–398), while acknowledging that the analysis
of that as a relative pronoun is widespread, especially in traditional grammar,
argues rather that that introducing a relative clause should be identified with the
subordinator that of sentences like (8).
(8) He said that he was ill.
Arguments in favour of this analysis are that that introducing relative clauses
behaves more like the subordinating conjunction that than like clear cases of
relative pronouns such as who and which. Thus that, unlike which, cannot be
preceded by a preposition, as in (9), and is usually omissible, unlike the general
run of English pronouns and like the subordinate conjunction that, as in (10),
with which (11) should be compared.
(9) *the stick with that the man chased the dog
(10) the man (that) I saw
RELATIVE CLAUSES 83
(11) He said (that) he was ill.

There are, of course, instances where that is not omissible in relative clause, as
in (12), though these are paralleled by uncontroversial examples of non-omis-
sibility of the subordinator that, and Huddleston follows the line of argument that
these are to be regarded as grammaticalizations of perceptual constraints. If one
accepts this analysis, one argument for regarding these as grammaticalized rather
than non-grammaticalized perceptual constraints is that there is some dialect
variation, with some non-standard dialects accepting the starred variant of (12).1
(12) I didn’t like the guy that/*Ø spoke first.
Note that these two arguments essentially say that that introducing a relative
clause lacks the basic properties of other pronouns, and of noun phrases more
generally. This is taken a stage further by Lehmann (1984: 107), who argues that
that also lacks the inflectional morphology that would be expected if it were a
relative pronoun, though this strikes me as at best a weak addition to this set of
arguments. More specifically, Lehmann says that if that were a pronoun, one
would expect its plural to be those, following the morphology of the demonstra-
tive pronoun. However, there are languages that have relative pronouns that have
distinct morphology, distinct even from nearly homophonous pronouns of other
types. For example, in Latin the relative pronoun differs from the interrogative
pronoun in the nominative (neuter: also accusative) singular, although the forms
are otherwise identical (relative masculine qui, feminine quae, neuter quod;
interrogative masculine-feminine quis, neuter quid, with the distinction in the
neuter being absolute). One would hardly want to argue that you is not a
personal pronoun because it lacks the number distinction obligatory for all other
English personal pronouns. Moreover, the relative pronoun which is equally
lacking in inflectional morphology; the absence of a genitive *which’s is
particularly striking, given that ’s can be attached to such noun phrases as the
man who came early to give the man who came early’s dog. And if one were to
argue that whose serves as a suppletive genitive of which, as in (13), it is unclear
what would stop one from saying that whose serves equally as a suppletive
genitive of that. At best, the lack of morphology of that seems a small addition
to the battery of its absent nominal properties.
(13) the house whose roof was blown off
84 BERNARD COMRIE
It should be noted that analyzing that introducing relative clauses as a subordina-

tor rather than as a relative pronoun does not obliterate the distinction between
relative clauses (as in 14) and nominal (content) clauses (as in 15):
(14) the fact that I discovered
(15) the fact that I ran away
Even under this analysis, it remains the case that (14) contains a “gap”, a
position (direct object of discover) that would normally have to be occupied by
an overt noun phrase in English syntax, whereas (15) contains no gap.
In a recent addition to the literature on relative clauses in standard English,
Haiman (1990) argues that the that which introduces relative clauses is some-
times a subordinator and sometimes a pronoun, more specifically that it is a
subordinator when relativization is on a non-subject, as in (16) and (17), but a
relative pronoun when relativization is on a subject, as in (18):
(16) the dog (that) you fed
(17) the one (that) I compared you to
(18) the dog that bit you
On this analysis, omissibility of that in relative clauses has a straightforward
account: the subordinator can be omitted, but not the relative pronoun. Actually,
Haiman’s analysis is slightly more complex, and interesting, than this, since his
proposed generalization (19) (Haiman 1990: 91) also covers so-called that-trace
effects (the unusual paradigm of examples in 20 and 21, where the conjunction
that must be omitted if there is relativization on the subject of a subordinate clause).
(19) If the missing constituent of the relative or complement clause is
the subject, that is interpreted as the subject. If the missing
constituent plays any other role, that is interpreted as a
complementizer [=subordinator — BC].
(20) the man that I think (that) you saw
(21) the man that I think (*that) saw you
In (20), the missing constituent or “gap” is a non-subject, so that, in fact both
occurrences of that, are interpreted as subordinators. In (21), the gap is a subject,
so both occurrences of that are interpreted as subjects, which leads to too many
subjects for the number of verbs available, since each occurrence of that plus I
would be a subject, giving three subjects but only two verbs (think, saw).
RELATIVE CLAUSES 85
Haiman’s analysis will, no doubt, raise some eyebrows, in particular the

analysis of the second that in (21) as a relative pronoun.2 It does, however,
provide an elegant explanation for so-called that-trace effects, a real stumbling
block for a principled analysis of constraints on relative clause formation in
English. And the crucial point for the presentation in Section 2 is the possibility
that English that introducing relative clauses might have a schizophrenic nature,
being sometimes a subordinator and sometimes a relative pronoun.
2. Relative clauses on the periphery of standard English
Data from the periphery of standard English provide even more striking evidence
for the schizophrenic nature of English that introducing relative clauses. My use
of the notion of the “periphery” of standard English no doubts merits a much
more extensive and profound discussion than I can give here, or indeed than I
have worked out to date. Roughly, what I mean, as far as syntax is concerned,
are constructions that (i) occur systematically in the speech of (at least some)
speakers of English whose English would probably be considered to be standard,
in its syntax and morphology (but not necessarily pronunciation), by other
speakers of standard English, and (ii) might well not be considered standard, on
explicit reflection, by speakers of standard English, including those who use the
constructions systematically; in particular, they would be rejected in writing. In
a sense, these are constructions hovering around the edge of the standard
language and waiting to be accepted into the standard, although there is, of
course, no guarantee that they ever will be. The notion is clearly at least
somewhat subjective — witness, for instance, the exclusion of examples like the
asterisked version of (12) — and no doubt in part its justification is that it gives
some interesting results in relation to the discussion of Section 1, in particular
the question of whether that introducing relative clauses is a relative pronoun or
a subordinator. The subjectivity can be made more honest: perhaps these are
constructions that I think I might say myself, or that I wouldn’t immediately
react to as non-standard if someone else were to say them.
2.1 Genitive that’s
One of the pieces of evidence noted in Section 1 in favour of the analysis of that
as a subordinator in relative clauses is its lack of any inflectional morphology.
86 BERNARD COMRIE
As noted there, this is a weak argument, but if its inverse were to hold, this
would provide a strong argument in favour of analysing that as a relative
pronoun. In other words, if that had inflectional morphology parallel to that of
clear cases of pronouns, we would be almost forced to analyse it as a pronoun,
since English subordinators do not have inflectional morphology. The case that
arises is the possessive form. In the standard language, that lacks a possessive
form, indeed the only inflectional possessive form allowed to introduce relative
clauses in the standard language is whose, as in (22); in addition, the preposition-
al phrase of which can be used with inanimate nouns, as in (23):
(22) Do you remember the man whose house got burnt down?
(23) the house the roof of which was blown off
On the periphery of the standard language, however, the possessive form that’s
is attested. In his contribution to Milroy and Milroy (1993) — a useful source for
non-standard and peripherally standard English syntax — Harris (1993: 150–151)
notes (24) for Irish English.
(24) Remember the man that’s house got burnt down?
While I suspect that I do not use this construction myself, I do not find it clearly
unacceptable in the colloquial speech of standard English speakers, and some (by
no means all!) other English speakers I have asked have a similar reaction. In
this respect, this construction differs markedly from nearly all of the other
constructions that Harris cites for Irish English.
One might object that ’s can be added to a range of elements in English, as
in the example the man who came early’s dog, cited in Section 1, where it is
attached to an adverb. But this relates to phonological attachment. As an enclitic,
’s must attach phonologically as part of the same phonological word as the
material that immediately precedes it. But syntactically, ’s can only be attached
to a noun phrase, in the case of the example just cited the man who came early.
In (24), ’s attaches not only phonologically, but also syntactically to that, which
must thus be analyzed, in varieties that contain this possibility, as a noun phrase.
One might also contemplate an alternative analysis of (24), namely that the
’s is not the possessive marker but rather the possessive pronoun his, reduced
and cliticized, since the construction of (25) is widely attested on the periphery
of standard English. (See further Section 2.2.)
(25) Remember the man that his house got burnt down?
RELATIVE CLAUSES 87
While destressing and cliticization of his may have played some part in the
historical development of the construction (24), and indeed more generally in the
development of the ’s possessive in the history of English, this is clearly not how
it is analyzed synchronically in this construction, since ’s can be used even
where a different possessive pronoun would be required, as in (26):
(26) Remember the woman that’s house got burnt down?
(27) Remember the woman that her/*his house got burnt down?
Examples from the periphery of the standard thus provide evidence for an
extension of the range of pronominal that in relative clauses, in comparison with
the situation as analyzed by Haiman: it encodes not only subject relatives, but
also possessor relatives. The absence of that preceded by a preposition, even in
varieties that have (24), would still be accounted for by saying that when an
object (of a verb or preposition) is relativized, that functions as a subordinator.
2.2 Heads that aren’t taken up in the relative clause
In all the examples of relative clauses considered so far, the same notional noun
phrase has played a role in both the main and the relative clause. In example (7)
for instance, the style functions as subject of the main clause and as direct object
of the subordinate clause. In (22) the man functions as direct object of the main
clause and as possessor of the subject in the relative clause. Indeed, for many
linguists this is part of the definition of relative clause. However, if that intro-
ducing English relative clauses is, at least in some instances, to be analyzed as
a subordinator rather than as a relative pronoun, then this leaves open at least the
logical possibility that one might have a construction that would otherwise be
identical to a relative clause, but where the head noun would not be taken up in
the “relative clause”. Note that none of the examples considered so far is of that
type (with the possible exception of (15) — not usually considered a kind of
relative clause — and (25)–(27), for which see further below). While an example
like (14), especially with omission of the that, has no overt nominal element that
refers back to the head noun, the fact that the English verb discover requires a
direct object means that there must be a covert element of this kind; that I
discovered cannot in isolation be a well-formed subordinate clause introduced by
a subordinator (cf. *he said that I discovered).
However, on the periphery of the standard language there are attested
examples where a head noun in what otherwise looks to be a relative clause is
88 BERNARD COMRIE
not taken up again in the relative clause itself. A number of examples are cited
by Matsumoto (1997), including (28), from an American speaker.
(28) You come to a group that you have to eat certain foods.
The interpretation of (28) is clearly like that of a relative clause rather than like
a complement structure such as (15): the subordinate clause tells us what kind of
group is at issue. Similar examples are noted for Scottish English by Miller
(1993: 112) in his contribution to Milroy and Milroy (1993), such as (29).
(29) I haven’t been to a party yet that I haven’t got home the same night.
Miller treats (29) and such examples as instances of omission of the preposition,
but while one can rather easily reconstruct from in (29), with some examples it
takes a fair amount of imagination to come up with a plausible preposition,
especially if the subordinator that is replaced by where, as in (30) (Matsumoto
1988: 174), in which case it is impossible to insert a preposition.
(30) a cake where you don’t gain weight
Indeed, the construction ends up similar to the pseudo-English such that-construction
beloved of logicians (a group such that you have to eat certain foods; a party such
that I haven’t got home the same night; a cake such that you don’t gain weight).
In varieties of English that contain the construction of (28)–(30), clearly the
that that introduces the relative clause cannot be analyzed as a relative pronoun,
since it plays no syntactic or semantic role — subject, object, adjunct, possessor
of any of these, and the like — within the relative clause. The periphery of
standard English seems, then, to have extended the possibilities found in the
standard language for having relative clauses introduced by the plain subordinator
that. Such relative clauses must still restrict the reference of the head noun, but
not necessarily in a way that would be expressible in terms of any readily
reconstructable grammatical relation.
As noted above, examples like (14), though paralleling those just discussed
in being introduced by a subordinator rather than by a relative pronoun, differ
from the latter in that they have a grammatically readily reconstructable missing
element, in the case of (14) the direct object. If the construction of (28)–(30),
with a complete clause following the subordinator, were to be extended to
sentences like (14) as well, what would the resultant sentences look like? Well,
in order to result in a grammatical structure of English, they would need to
contain an overt noun phrase in the place of the missing element. And the
RELATIVE CLAUSES 89
material presented in Milroy and Milroy (1993) shows widespread occurrence of

this type, as in examples (31)–(32) from Miller’s account of Scottish English
(Miller 1993: 111–112).
(31) the spikes that you stick in the ground and throw rings over them
(32) the girl that her eighteenth birthday was on that day
Examples from the periphery are particularly frequent where otherwise con-
straints on relativization would be violated, as in (31) (Coordinate Construction
Constraint), and also (33) (relativizing the subject of a subordinate clause with an
overt complementizer).
(33) the road that I don’t know where it leads
Examples like (32) seem to reflect the relative unnaturalness of the possessive
form whose in many varieties of informal speech, even for standard speakers.
Traditionally, in particular within generative and typological studies, such
examples as (31)–(33) have been considered to illustrate a distinct pronoun-
retention type of relative clause. For varieties of English that lack the construc-
tion type of (28)–(30), and indeed for other languages that have the construction
of (31)–(33) but lack that of (28)–(30), this is probably the correct analysis. But
for varieties that include the type (28)–(30), an alternative analysis suggests itself
which avoids having to specify a distinct set of pronoun-retention relative
clauses: rather, such varieties simply allow a relative clause introduced by that
followed by a complete clause; this allows both that the following clause might
refer back to the head noun (as in (31)–(33)) and that it might not (as in (28)–(30)).
A further possible re-analysis arises if we take seriously the fact that the
relative clause of example (30) is introduced not by that but rather by who, with
similar examples found with “pronoun retention”, as in (34) (from Miller’s
Scottish English material) and (35):
(34) an address which I hadn’t stayed there for several years
(35) the road which I don’t know where it leads
What these seem to suggest is that the relevant varieties of English have at least
the possibility of reinterpreting even the traditionally clearly pronominal elements
who and which as non-pronominal relative clause introducers, such that they
function much as that does in introducing relative clauses, with the difference
that who and which do not parallel that in being able to introduce other kinds of
subordinate clauses.3 Of course, to the extent that such varieties also allow
90 BERNARD COMRIE
examples like (1)–(3), (6)–(7), (12)–(14), (16)–(18), (20)–(21), or some subset of

these, they still have the traditional, unequivocally standard English relative clause
construction with a gap. But on the periphery, other possibilities also exist.
3. Conclusion
The survey of standard English relative clauses in Section 1 took us from the
traditional analysis of that introducing relative clauses as a relative pronoun
through its analysis as a subordinator to a seemingly more complex analysis in
which it can be either, depending on its grammatical function. Consideration of
examples from the periphery of standard English in Section 2 suggests that the
range of both types of construction is being extended. Of course, not all varieties
of English allow all of the possibilities considered in Section 2, and to the extent
that varieties of English differ in this way their grammars (in the broadest sense)
must differ accordingly. But at least I hope to have shown that even as exten-
sively investigated a topic as relative clauses in as extensively investigated a
language as English can still provide empirical and analytical surprises.
Notes
1. A third point noted by Huddleston is that relative clauses introduced by wh-expressions have
non-finite (infinitival) versions, whereas those introduced by that do not, as in (i)–(ii);
uncontroversial subordinate clauses introduced by that must likewise be finite.
(i) He was looking for a box in which to store her letters.
(ii) *He was looking for a box that to store her letters in.
But note that (iii), the version of (i) in which the preposition is stranded as in (ii), seems as bad
as (ii).
(iii) *He was looking for a box which to store her letters in.
The version of (ii)–(iii) in which zero appears in place of that or which is grammatical, as in
(iv), though it is perhaps not obvious that this should be analyzed as a relative clause.
(iv) He was looking for a box to store her letters in.
2. It should be noted, however, that there are languages like Irish that have an overt relative
marker rather than a subordinator in the position of the second that of (21). Compare (i) and (ii)
(Ó Siadhail 1989: 316), where REL glosses the relative marker, which aspirates the initial
segment of a following verb (whence chuirfeadh); the subordinator go lenites the initial segment
of a following verb (whence gcuirfeadh).
(i) Dúirt sé go gcuirfeadh an leabhar ar mo chosa mé.
said he that would-set the book on my feet me
‘He said that the book would set me on my feet.’
RELATIVE CLAUSES 91
(ii) an leabhar a dúirt sé a chuirfeadh ar mo chosa mé

the book REL said he REL would-set on my feet me
‘the book that he said would set me on my feet’
(Note that in Irish this repetition of the relative marker also occurs when relativization is on the
direct object of such a clause.)
3. A number of languages have non-pronominal elements that serve solely to introduce relative
clauses; for example, yang in Malay-Indonesian.
References
Haiman, J. 1990. “Schizophrenic Complementizers”. Studies in Typology and

Diachrony: Papers Presented to Joseph H. Greenberg on his 75th Birthday.
Typological Studies in Language 20, ed. by W. Croft, S. Kemmer and K.
Denning, 79–94. Amsterdam: John Benjamins.
Harris, J. (1993). “The Grammar of Irish English”. Milroy and Milroy, 139–186.
Huddleston, R. 1984. Introduction to the Grammar of English. Cambridge
Textbooks in Linguistics. Cambridge: Cambridge University Press.
Lehmann, C. 1984. Der Relativsatz: Typologie seiner Strukturen; Theorie seiner
Funktionen; Kompendium seiner Grammatik. Language Universals Series 3.
Tübingen: Gunter Narr.
Matsumoto, Y. 1988. “Semantics and Pragmatics of Noun-modifying Construc-
tions in Japanese”. Berkeley Linguistic Society 14. 166–175.
Matsumoto, Y. 1997. Noun-Modifying Constructions in Japanese: A Frame
Semantic Approach. Amsterdam: John Benjamins.
Miller, J. (1993). “The Grammar of Scottish English”. Milroy and Milroy, 99–138.
Milroy, J and Milroy, L. 1993. Real English: The Grammar of English Dialects in
the British Isles. Real Language Series. London: Longman.
Ó Siadhail, M. 1989. Modern Irish: Grammatical Structure and Dialectal Varia-
tion. (Cambridge Studies in Linguistics, Supplementary Volume.) Cam-
bridge: Cambridge University Press.
Quirk, R., S. Greenbaum, G. Leech and J. Svartvik. 1985. A Comprehensive
Grammar of the English Language. London: Longman.
Post nominal modifiers in the English noun phrase
Peter H. Fries
1. Introduction
All linguists agree that the English noun phrase is a very complex construction.
If we focus on those modifiers which follow the head function, most linguists
(typical analyses include Huddleston 1984, Quirk et al. 1985, and Radford 1988)
identify three different functions. Huddleston is typical when he distinguishes
between what he calls complement, modifier and peripheral dependent. This
paper will focus exclusively on the distinction between complement and modifier
and the arguments which have been adduced to support that distinction. Huddle-
ston and Radford use the term “complement” for one of these functions in order
to highlight the similarity between these constituents of the noun phrase and
constituents of the clause which form part of the nuclear structure of the clause
and which typically express arguments of the verb. Thus, similar to complements
of the clause, complements in the noun phrase “generally correspond to argu-
ments of a semantic predicate expressed in the head noun” (Huddleston 1984:
262). One consequence of that feature is that one can subclassify nouns on the
basis of whether they may or may not accept modification by a complement, and
if they may accept a complement, which type of complement is possible. Such
subclassification is not possible with modifiers. Since Huddleston’s position is
similar to that of Radford, and Radford expresses his arguments for the distinc-
tion very overtly, I will organize this paper around Radford’s presentation, but I
will periodically point out places where Huddleston differs from Radford.
94 PETER H. FRIES
2. Arguments for distinguishing complement from modifier.
Radford’s (1988) term for the other post-nominal function under focus here is
“adjunct”. Radford’s adjuncts are equivalent to what Huddleston calls modifiers,
and are roughly equivalent to what more traditional linguists often call restrictive
modifiers. (Since in this section I will be either directly quoting Radford, or
closely paraphrasing what he says, I will use his term “adjunct”. However, since
his use of that term for this function is somewhat unusual, I will use the subscript
“R” (as in adjunctR) to make it clear that the term is being used in his sense.)
Radford’s examples of complements include those in (1):
(1) (= Radford example (90), p. 193)
a. your reply to my letter
b. the attack on the Prime Minister
c. the loss of the ship
d. her disgust at his behavior
e. his disillusionment with linguistics
As examples of adjunctR Radford cites those in (2):
(2) (= Radford example (91), p. 193)
a. the book on the table
b. the advertisement on the television
c. the fight after the match
d. his resignation because of the scandal
e. a cup with a broken handle
Radford distinguishes adjunctsR and complements on the basis of the seven
features described below.
A. Complements will always be “closer” to their head noun than adjunctsR.
Compare:
a student of Physics with long hair (complement + adjunctR)
*a student with long hair of Physics (adjunctR + complement)
B. A noun phrase may contain at most one complement, while a noun phrase
may contain more than one adjunctR. Compare:
*the student of Physics of Chemistry (complement + complement)
the student with long hair in the corner (adjunctR + adjunctR)
POST NOMINAL MODIFIERS IN THE ENGLISH NOUN PHRASE 95
C. Complements may not follow substitute one, while adjunctsR may follow
substitute one. Compare:
a student of Physics ⇔ *one of Physics (complement)
a student with long hair ⇔ one with long hair (adjunctR)
D. AdjunctsR attribute a separate semantic property while complements do not.
Rather the head and complement together attribute a single semantic property.
The difference is illustrated in the ambiguity of a student of high moral principles
(“the person studies high moral principles” versus “the person has high moral
principles”).
E. It is impossible to coordinate complements with adjunctsR:
*the student of Physics and with long hair
F. AdjunctsR may be extraposed to the end of the sentence more freely than
complements. Compare:
A student came to see me yesterday with long hair.
*A student came to see me yesterday of Physics.
G. Complements are more freely preposed than adjunctsR.
What branch of Physics are you a student of?
*What kind of hair are you a student with?
All the examples provided by Radford in his discussion involve prepositional
phrases, but he comments that the complement/adjunctR distinction is valid for
clauses as well as for prepositional phrases, thus implying that clausal comple-
ments have the same features that his prepositional complements have.
In fact, not all the arguments advanced by Radford hold. While what he
says about adjunctsR is generally accurate, there are more serious problems with
what he says about complements. As a result, at this point let me focus on
examples of complements and some of the features of complements which
Radford says are diagnostic.
3. Arguments for distinguishing complement from modifier: revisited
This section will repeat Radford’s points but they will be discussed in terms of
some of my data.1
96 PETER H. FRIES
A. The relative ordering of complements and adjunctsR: Complements will

always be “closer” to their head noun than adjunctsR.
In fact, while this statement describes a general trend — complements are
usually closer to their head than adjunctsR when both occur within the same noun
phrase — it is not always true. In (3a) and (3b) the phrases in your September
issue and in Hong Kong are adjunctsR, but precede the complements (an of-
phrase in 3a, and a complement clause in 3b).
(3) a. [near the close of] his review in your September issue of The
Western Intellectual Tradition, by J. Bronowski and Bruce
Mazlish, [C. P. Snow comments … ]
b. There were suggestions in Hong Kong that the issue divides
along cultural lines
Radford ignores two basic issues which must be considered in the ordering of
modifiers which follow the head of the noun phrase. The first issue concerns
ambiguity. If one were to postpone the prepositional phrase in Hong Kong in (3b)
to a position after the complement clause, it would most likely be read as a part
of the complement clause, not as a separate modifier of suggestion. Placing the
prepositional modifier before the complement avoids that ambiguity. A second
issue to be considered in ordering modifiers which follow the head concerns the
relative information value expressed in the complement and the adjunct. If the
information expressed in the complement is more important to the development
of the text, then that information will more than likely be expressed last. In each
case, in the context, the complements expressed in examples (3a) and (3b)
convey information that is more important than the information conveyed by the
adjunctsR. For example, the nature of what is being responded to by the author
of (3a) relates more closely to the fact that the stimulus was a review of a
particular book than that the review was found in a particular issue of a journal.
Similarly, in (3b), in the context of the particular example the specific nature of
the suggestions was more important than that they were being made in Hong Kong.
B. A noun phrase may contain at most one complement, while a noun phrase
may contain more than one adjunctR.
Radford’s wording is far too strong. If complements are considered to be
those phrases and clauses which “depend on the selection of an appropriate noun
as head” (Huddleston 1984: 264), then many noun phrases will be seen to
contain more than one complement. Examples include the ones listed in (4a–f).
The first complement has been underlined and the second has been italicized to
emphasize the fact that there are two complements in each example.
(4) a. the duty of each district to raise a leiðangr …
b. the reluctance of some to accept his own unsupported testimony
c. a very moving appeal to all Christians to do all they can
d. the division of European history into national histories
e. the distribution of income within a country
f. The nearly simultaneous discoveries of non-Euclidean geome-
tries by Gauss, Lobachevski and Bolyai
It is clear that examples (4d–f) constitute nominalizations of processes, and that
the of-phrase expresses the actor or goal of the underlying process, while the
second prepositional phrase expresses a circumstance or actor. The particular
preposition used in the noun phrase is the same as the one that would be used,
if the process had been expressed in clausal form. If this is true there would be
a limitation on how many complements can occur in a single noun phrase, but
this limit would be greater than one. Huddleston (1984: 260f) takes an approach
which is similar to the one advocated here. He gives examples which illustrate
noun phrases containing more than one complement, and then says that “The
number of complements in any one noun phrase is strictly limited.” (261) Clearly
this position is more reasonable than Radford’s, since it makes clear that the
limitation is to a number greater than one.
C. Complements may not follow substitute one, while adjunctR may follow
substitute one.
This feature holds in general. However there are many exceptions to both
parts of this statement. The examples in (5) illustrate that at least several
adjunctsR do not post-modify one.2
(5) a. the way to do it
*the one to do it
b. a way to do it
*a one to do it
c. the other solo ballet dancers of the evening
*the other ones of the evening
d. his wife of a year
*his one of a year
98 PETER H. FRIES
e. the strategic arms race in its present form

*the one in its present form
f. the first minute after entering the chamber
?the first one after entering the chamber
Similarly the examples in (6) demonstrate that many (though not all) comple-
ments may post-modify one.
(6) a. a loss of a ship
one of a ship
b. the attack on the Prime minister
the one on the Prime Minister
c. a suit for damages
one for damages
d. their belief in the right of the strongest
??their one in the right of the strongest
e. an orientation toward the future
one toward the future
f. a rap on the wall
one on the wall
g. the fact that he had already arrived
??the one that he had already arrived
h. the wish to o end
?the one to o end
i. the idea that we could sell it
?the one that we could sell it.
j. the philosophy that adults have to do dull jobs
the one that adults have to do dull jobs
D. AdjunctsR attribute a separate semantic property while complements do not.
Rather the head and complement together attribute a single semantic property.
The difference is illustrated in the ambiguity of a student of high moral principles.
Here Radford seems to be presenting the semantic ambiguity as an argument to
support his point. However, while this example is indeed ambiguous in the way
Radford claims, it is dangerous to accept this ambiguity as an argument for two
different functions. The problem is that the word of has so many meanings that
one must decide which of several possible meanings are relevant to positing a
difference in grammatical function. For example, a phrase such as the develop-

ment of the Impressionists might be understood as any of (7a–d):
(7) a. the Impressionists developed (something)
b. (something) developed the Impressionists.
c. the Impressionists developed
d. the development is characterized by being associated with the
Impressionists.
Clearly the list can be continued for some time. (For example, alienable posses-
sion has not been illustrated.) My point here is that ambiguity of itself does not
argue for a difference in grammatical functions. On the other hand, the proposed
difference in function can be used as a tool to “account” for an ambiguity which
is discovered.
In a more positive vein, the basic semantic point which Radford makes
describes the general trend. Those items which are analyzed as adjunctsR
typically express separate properties or qualities of the referent, while those items
which are analyzed as complements express a more integral part of the reference.
In an example such as (8), of the hunters is not a property of the shooting but
rather an integral part of the reference of the entire noun phrase.3
(8) the shooting of the hunters.
Related to this point is the notion of cooccurrence restrictions. As Radford points
out “In the case of a PP Complement, there are severe restrictions on the choice
of P heading the PP.” (192). Again, this is true, but it raises an interesting issue.
Radford describes the coccurrence restrictions which hold between adjunctsR and
their heads in the following terms: “the type of PP which functions as adjunct
can be used to modify any type of head noun (subject to semantic and pragmatic
restrictions)” (192).4 While Radford is clearly focusing on the lack of limitation
on the coccurrence potential of adjunctsR, his wording clearly admits that it is not
true that anything can cooccur with anything. The problem, then, is to locate the
boundary where semantics and pragmatics leave off and grammar begins.5 If one
adopts, as some grammarians do, the view that there is no major distinction of
type between semantic/pragmatic cooccurrence restrictions and grammatical
cooccurrence restrictions, then the differences in cooccurrence restrictions
becomes merely a matter of the strength of these restrictions and how they are
best described. One cannot use the difference in type of restriction as an
argument for a difference in grammatical function.
100 PETER H. FRIES
While Radford’s account of the semantic difference between adjunctsR and

complements holds in general, there are a few examples which do not seem to
fit the pattern of either type of post-modifier, the complements or the adjunctsR.
These examples do not describe a property nor are they part of the reference of
the noun phrase. Rather they seem to concern the applicability of the term to the
referent. The examples in (9) illustrate these post-modifiers.
(9) a. [They had] no windows to speak of
b. the dynamics of the arms race as a whole
In (9a) the infinitive clause hedges the reliability of using the terms no windows
to indicate the referent. In this way, it is similar in interpretation to They had
almost no windows. One could very well imagine a sequence of sentences such as
(10), in which the hedged nature of the claim is taken up in the second sentence.
(10) They had no windows to speak of. Well, they did have a few sort
of holes near the ceiling that let in some light and air.
By contrast, one could not imagine a sequence in which the second sentence
takes up the notion of speaking as in:
(11) They had no windows to speak of. In fact, they didn’t have much
to talk about at all.
Looking at the issue more formally, notice that when an infinitive clause acts as
a typical adjunctR in a noun phrase, the subject of the infinitive clause is either
actually expressed (typically as the object of the preposition for) as in (12a), or
is interpreted as identical to some other constituent of the sentence (for example,
the head of the noun phrase which includes the infinitive clause, or the subject
of the sentence which includes the infinitive clause) as in (12b) and (12c). In
other cases, the subject of the infinitive is simply left implicit and it is left to the
listener/reader to infer the subject. However, even in these cases it is possible to
supply some subject.
(12) a. Allow time for us to travel there.
b. He had no children to succeed him on the throne.
c. We have nothing to hope for.
d. Allow time to travel there.
In (12d) it is left to the listener/reader to figure out who is doing the travelling.
However, it is possible to make the subject explicit. By contrast, in (9a) it is
impossible to make the subject explicit.6 Notice that all of the examples in (13a
– d) change the meaning of the original and seem very odd. Even (13c), which
uses a radically different construction, seems less than successful.
(13) a. ?They had no windows for us to speak of
b. ?They had no windows for them to speak of
c. ?They had no windows for one/anyone to speak of
d. ?They had no windows that one could speak of
E. Cannot coordinate complements with adjunctsR.
Statement E is true — complements cannot be coordinated with adjunctsR.
However, again, we must be very careful about the conclusions we draw from
that fact. The problem is that many apparently similar post-modifiers may not be
coordinated with each other. Coordination requires that the coordinated items be
seen as similar. As a result, only certain complements may be coordinated with
each other. Similarly only certain adjunctsR may be coordinated with each other.
This is true even when the coordinated items express similar meanings.
Radford would consider the post-nominal modifiers in (14a–c) to be
complements. Notice that volume and consistency have much the same relation
to the head change, regardless of the preposition which introduces them. That is,
in both cases, the post-nominal modifier describes what changes (the volume
changes, the consistency changes). However, these two post-modifiers may be
coordinated only if they are introduced by the same preposition.
(14) a. a change in volume and in consistency
b. a change of volume and of consistency
c. *a change in volume and of consistency
The issue is much worse if the complements have different semantic relations to
the head as in (15).
(15) *the division of European history and into national histories
Just as the coordination of complements is a complex issue, the coordination of
adjunctsR is also complex, as in (16a) and (16b):
(16) a. a book for children
b. a book of great importance
Here the prepositional phrases for children and of great importance would both
be considered to be adjunctsR. However, it would be most unlikely that these two
102 PETER H. FRIES
prepositional phrases would be coordinated with one another in a phrase such as

(17a) or (17b):
(17) a. *a book for children and of great importance
b. *a book of great importance and for children
As one can see, while coordinatability strongly implies7 similarity in grammatical
function, a similarity in grammatical function does not guarantee coordinatability.
When two items are not coordinatable, the analyst needs to decide whether that
impossibility arises from a difference in grammatical function, or from some
other source. To summarize the point here, the lack of coordination cannot be
used as a strong argument to support a difference in function. The possibility
that two constructions may be coordinated can usually be used to establish that
two constructions function in similar ways in an example or set of examples.
F. AdjunctsR may be extraposed to the end of the sentence more freely than
complements.
Since the wording is hedged, this statement is falsifiable only with data
concerning relative frequencies. Extraposition of post-nominal modifiers of noun
phrases is in general rare, so it is difficult to get solid statistics on it without
automatic tagging and parsing. In general an element can be extraposed to the
end of its sentence only when the extraposed element is sufficiently important to
the purposes of the text and the text segment in which it occurs to warrant the
prominence it receives as a result of being postponed to the end. Examples
(18a–c) demonstrate that complements may be extraposed provided they contain
sufficiently important information.8
(18) a. Work is well underway on sites such as Hopetoun Quays next
to Birchgrove Public School, which will soon go into its
second stage of selling 111 units.
[The Balmain and Rozelle Village Voice July 2nd Birthday
edition, p. 1]
b. Balmain Police have lodged an application with the Leichardt
Council for the creation of an alcohol-free zone in the
vicinity of Gladstone Park.
[The Balmain and Rozelle Village Voice July 2nd Birthday
edition, p. 2]
c. Only a token start was made in attacking the tax reappraisal
question
In all three examples, the extraposed post-modifiers provide information which

is critical to the point of the sentence. This is most obvious in (18b), where it is
clear that the nature of the application is an important point in this sentence.
Just as extraposing complements to the end of the sentence requires that
they be worthy of informational prominence, extraposing adjunctsR to the end of
the sentence also requires that the adjunctsR be worthy of informational promi-
nence. Examples (19a–c) show that adjunctsR may be extraposed when they
contain important information.
(19) a. Every day decisions are being made by local officials in our
communities that could drastically affect the quality of our
lives.
b. Afterwards a layer of plastic was applied the thickness of a
nickel.
c. After the victory, Athens’ position was assured as a leader in
the Greek world.
In the case of (19a), Fries (1995: 240–244) discusses ways in which the mean-
ings expressed in the italicized relative clause convey information that is crucial
to the functioning of the sentence in that text.
Since information prominence is at issue both in the case of extraposing
adjunctsR and of extraposing complements, it is easily possible to construct
examples which involve such extraposition with either type of function which do
not satisfy. All that is necessary is to place unimportant or obvious information
in the postposed element at the end of the sentence. The examples in (20)
illustrate unsuccessful extraposition of adjunctsR, while the examples in (21)
illustrate unsuccessful extraposition of complements.
(20) a. ?The other solo dancers were quite excellent of the evening
b. ?The lake was very pretty which we saw.
(21) a. ?The shooting was quite surprising of the hunters.
b. ?His description didn’t make sense of how he did it.
The examples in both (20) and (21) seem unlikely at best.
If there is a difference in the relative proportion of adjunctsR which are
extraposed vs. complements which are extraposed, then one would guess that the
difference arises ultimately from the claim made in Point D above concerning
the different semantic interpretations expressed by adjunctsR and by comple-
ments. If, as is implied in Point D, complements are seen to be more integrally
104 PETER H. FRIES
related to the reference of the noun phrase as a whole (rather than simply
expressing a property of the referent) then it would seem less likely that the
complement would be worthy of receiving the separate emphasis which is
entailed by extraposition.
G. Complements are more freely preposed than adjunctsR.
I have no examples of preposed complements (or adjunctsR) which are
comparable to the ones Radford proposes, and am unable to construct truly
convincing examples. However, one type of construction does allow preposing
of the complement, this is the of construction with collective nouns, partitive
nouns and quantitative nouns.9 Consider (22):
(22) a. Of Dorian Grey I haven’t seen a picture, [but I have seen a
photo of Oscar Wilde.]
b. Of these excellent fish a small school went by soon after we
started swimming.
c. Of this poison one small taste will kill you very effectively.
d. Of the people we invited only sixty showed up.
e. Of the seventy sets that we ordered none arrived in good
condition.
f. Of the seventy sets which arrived many were damaged.
The example in (22a) is cited by Matthiessen (1995): (22b) and (22c) are further
constructed examples. Similar pre-positionings are accepted by certain quantif-
iers, as in (22d) and (22e).
The examples in (22d–f) may be compared with those in (23a–c), which
have a more normal order.
(23) a. Only sixty of the people we invited [showed up].
b. None of the seventy sets that we ordered [arrived in good
condition].
c. Many of the seventy sets which arrived were damaged.
4. Discussion
In summary, the evidence concerning whether or not to differentiate between

complement and adjunctR is far from conclusive. Certainly there are differences
between the various post-nominal modifiers. Indeed, it is clear that these
differences tend to cluster around the distinction which most linguists make
between adjunctsR and complements. However, the next question to ask is why
that should be. To take two of the formal features which Radford uses as
important distinguishing tests for his analysis, why should complements tend to
be placed nearer to the head than adjunctsR, and why should complements be
somewhat more difficult to postpone to the end of the sentence? Assigning these
examples to two different grammatical functions implies that the difference
arises because of the difference in grammatical function. In this view, positing
a difference in grammatical function is used to “explain” or “account for” a
cluster of features including differences in meaning, differences in selectional
restrictions and (more relevant here) differences in ordering potential. That is the
approach which Radford, Huddleston, and Quirk et al. have taken. But another
answer is possible — that the two sets of examples regularly differ in their
informational status. The “cause” of the difference in ordering potential lies in
their informational status not in their different grammatical functions. Such an
answer predicts that if an example of either type of construction is found which
contains information which is essential to the function of the text or the text
segment, it is likely to be placed last within the noun phrase or even postponed
to the end of the clause. Since in this view the differences in ordering principles
relate to informational status, using a difference in grammatical function to
account for the difference in ordering potential would be to mistake a correlation
with a causal relation. Of course, in this second approach, features such as
selectional restrictions will have to be dealt with using some other theoretical
construct. So, the ultimate issue dividing these two approaches is whether or not
all these different features (meaning, selectional restrictions, ordering potential
etc.) are most efficiently “explained” by a single (grammatical) factor, or
whether it is more efficient, in the long run, to posit several different (grammat-
ical) factors, all of which are seen to operate in the noun phrase (as well as
elsewhere in the language), with each factor used to account for different
features of the potential (ordering, selectional restrictions, etc.) of particular
constructions.
Another of Radford’s arguments, the difference in meaning between
modifiers of the two types of functions, again is not convincing to someone who
does not agree with the analysis beforehand. The reason is, as was discussed
under Point D, that linguists cannot posit separate functions for each difference
in meaning, even when those differences in meaning are accompanied by differ-
ences in formal potential. The examples in (7) illustrate some of the issues that
might arise should one attempt to do so. Let me simply add one further example
106 PETER H. FRIES
of two modifiers which would be assigned to adjunctR but which express quite
different relations to the head of the noun phrase.
(24) a. the other solo ballet dancers of the evening
b. the road in the northern part of the county
The two examples differ grammatically in that (24b) may be paraphrased by a
relative clause, while (24a) cannot. In spite of this semantic and grammatical
difference, they are typically assigned to the same post-nominal function.
A number of linguists, in particular ones who advocate a Systemic approach
to grammar, do not distinguish between complements and adjunctsR. Thus,
Downing and Locke (1992), Halliday (1985, 1994), Lock (1996), and Matthies-
sen (1995) all posit only one function where Huddleston, Quirk et al., Radford
and others posit two. Since linguists who do not distinguish between comple-
ments and adjunctsR do not argue for their position, it is impossible to say what
evidence was considered when they made their decisions.10 However, one can
assume that it resembled the sorts of points that have been made in this paper.
But whichever approach is taken to the description of the post-nominal
modifiers in the noun phrase, it is incumbent on all grammarians to be sure that
all examples of the phenomena being described can be accounted for using the
tools of the descriptive system being used. Of course, one cannot simply look at
individual examples, but it is necessary that any description fit into an integrated
theory. Let me review a few places that seem to be problems for description.
Since meaning is of central interest to most linguists, I will focus on examples
which illustrate a range of semantic relations. Each semantic difference can be
correlated with some difference in grammatical property, thus these differences
cannot merely be discarded as being “purely” semantic.
Radford and others point out two11 major types of meanings which are
typically expressed by post-nominal modifiers. These meanings are described as
(a) expressing a quality of the referent, and (b) forming part of the reference
itself. Two examples, (9a) and (9b), have already been mentioned in which the
post-nominal modifier seems to fit into neither category (a) nor category (b), but
rather to describe how the wording of the noun phrase is to be understood.
A second set of examples also seems not to fit the basic two-category
distinction in important ways. Examples are given in (25):
(25) a. my brother Charles
b. the number four
c. the two words do
The first thing to notice about these examples is that while they are “appositives”
in that the second term describes the first, they are not non-restrictive apposit-
ives.12 That is, Charles, four and do all provide essential information within their
constructions. Notice also that these words are grammatically essential in that
they cannot be omitted (my brother is not merely a short version of my brother
Charles, the number is not simply a short version of the number four) Similarly,
the phrases cannot be reversed13 (*four the number and *do the two words are not
the same as the number four and the two words do14). Charles, four and do do not
seem to be separate qualities of the referent, but part of the reference. However,
they are not part of the reference in the way that Radford’s complements are.
A third set of examples which do not seem to fit into the adjunctR/complement
distinction is illustrated by the examples in (26):
(26) a. He’s a terror of a child
b. That brute of a dog bit me
c. They’ve got a dream of a garden
Downing and Locke (1992: 468) point out that these examples have an unusual
relation between head and post-nominal modifier in that the post-nominal
modifier indicates the referent and the head of the noun phrase indicates a
quality. One can paraphrase these examples with (26’), but not with (26”).
(26′) a. That child is a terror
b. That dog is a brute
c. Their garden is a dream
(26″) a. That terror is a child
b. That brute is a dog.
c. Their dream is a garden.
Downing and Locke go so far as to analyze terror of, brute of, and dream of as
prenominal modifiers in their noun phrases. Thus the heads of these noun phrases
are child, dog and garden. If one does not wish to go to such a radical analysis,
one is left with a problem when it comes to describing the semantic relation
between head and postnominal modifier.
Examples (25) and (26) illustrate the most dramatic exceptions to the
traditional descriptions of the post-nominal modifiers. There are, of course, less
dramatic examples of indeterminacies. The examples in (27) illustrate some of
the range of post-modifiers in the noun phrase. Many of these examples are
108 PETER H. FRIES
repetitions of examples which have already been cited. However, several have
not been mentioned.
(27) a. a layer the thickness of a nickel
b. breathing organs such as gills
c. an error of .01 degrees Centigrade
d. a region of some importance
e. a day like this
f. a terror of a child
g. Network Light
h. the number four
i. the road back
j. those children of theirs
k. two main problems for a mammal
l. the division of European history into national histories
m. the arms race as a whole
Of course the examples in (27) are chosen to include both clear adjunctsR and
clear complements as well as some that do not belong clearly in either category.
Clearly, even without going into the more fine-grained distinctions in meaning
cited in (7) above more than two semantic relations may be expressed between
the head and the postnominal modifier in the noun phrase. These examples
challenge all linguists regardless of the approach they use. They challenge those
who make a distinction between adjunct and modifier to justify the division into
just two grammatical functions. They challenge linguists who do not distinguish
between adjunct and modifier to account both for the similarities and the
differences between these examples.
Notes
1. The data on which these comments are based consist of a core corpus of about 1,000 examples
gathered from ten pages of the January 1961 issue of Scientific American, supplemented by a
large number of additional examples gathered more casually in the years since then.
2. Definiteness is relevant to the acceptability of the examples in (6). For example (6g) is
extremely unlikely to occur as an indefinite noun phrase (?a fact that he had already arrived).
As a result, I have tried to illustrate both definite and indefinite noun phrases, and to preserve
the definiteness of the paraphrases which involve substitution with one.
3. This is true regardless of whether one interprets (8) as involving a “subjective genitive” (“The
hunters shot something”), or an “objective genitive” (“Someone/something shot the hunters”).

4. Emphasis in the original.
5. For example, there is a restriction that back only post-modifies nouns that involve some sort of
trajectory. Thus we can have examples such as (a–c) but not (d–e).
a. the road back
b. the run back
c. the trip back
d. the picture back
e. the telephone back
Is this restriction best regarded as a grammatical restriction or a semantic restriction? Note that
at least in the case of run the restriction is related to selectional restrictions inherent in the verb
to run. Similarly, one of the readers of an earlier version of this article objected to treating at
once in the example I do not speak of emancipation at once, but of a decision at once to
emancipate gradually as a complement within the noun phrase. Clearly the presence of at once
depends on the presence of an action nominalization as the head. (We cannot say *the boy at
once, or *the fact at once.) Thus, that modifier ‘subcategorizes’ the head in some sense.
However, it should be noted that when we examine the clause which is related to this form,
(e.g. We decided at once to emancipate gradually), the adverbial at once is not a complement of
the clause.
6. One might legitimately take the approach that to speak of is a set phrase, an idiom, and should
not be treated as a true infinitive clause. One argument for treating this phrase as an idiom is
the difficulty of using any other verb, or even any other form of the verb in this construction.
However, if we treat this phrase as an unanalyzable idiom, we still need to deal with the fact
that it appears as an adjunctR in the noun phrase.
7. It should be noted that while most coordinated items do fill similar functions, that strong
tendency is not an absolute requirement for a grammatical sequence to result.
8. Indeed, Radford (448 ff.) provides several examples of extraposed complements, but says
nothing of their possible discourse status.
9. Downing and Locke (1992) discuss this construction explicitly and decide that the preposed of-
phrase is not part of the noun phrase but a kind of adjunct of the clause.
10. In fact, they should not be required to provide evidence for not making a distinction. To require
them to do so is to require them to argue a negative case. Rather they simply need to show that
the evidence which has been used to support making a distinction between the two putative
grammatical functions is not persuasive, and/or can be handled in some other manner.
11. The number two arises, of course, because we have omitted peripheral dependents (that is, non-
restrictive modifiers) from consideration throughout this chapter.
12. Some traditional grammarians describe examples such as these as “close appositives”, thereby
distinguishing them from what one might term “regular” appositives.
13. This property distinguishes close appositives from “regular” appositives.
14. There is an interesting relation between these phrases and classifiers such as the four number
and the do word.
110 PETER H. FRIES
References
Bolinger, D. 1967. “Adjectives in English: Attribution and Predication”. Lingua

18. 1–34.
Downing, A. and P. Locke. 1992. A University Course in English Grammar. New
York: Prentice Hall International Language Teaching.
Halliday, M.A.K. 1985. An Introduction to Functional Grammar, 1st edn.
London: Edward Arnold.
Halliday, M.A.K. 1994. An Introduction to Functional Grammar, 2nd edn.
London: Edward Arnold.
Huddleston, R. 1984. Introduction to the Grammar of English. Cambridge:
Kress, G. ed. 1976. Halliday: System and Function in Language. London: Oxford
University Press.
Lock, G. 1996. Functional English Grammar: An Introduction for Second Lan-
guage Teachers. Cambridge: Cambridge University Press.
Matthiessen, C. 1995. Lexico-Grammatical Cartography. Tokyo: International
Language Sciences Publishers.
Pocheptsov, G.G. 1968. “Indispensable Attributive Adjuncts in English”. Lingua
20. 1–14.
Prakasam, V. 1996. “ ‘Ngp of Ngp’ Constructions: A Functional-structural
Study”. Meaning and Form: Systemic Functional Interpretations. Meaning
and Choice in Language: Studies for Michael Halliday ed. by M. Berry, C.
Butler, R. Fawcett and G. Huang, 567–583. Norwood NJ: Ablex.
Radford, A. 1988. Transformational Grammar: A First Course. New York:
Elliptical clauses in spoken and written English
Sidney Greenbaum Gerald Nelson
Quirk et al. (1985: 883) define ellipsis as grammatical omission, which is then
contrasted with other kinds of omission in language. These are said to include
phonological loss, as in ‘cos for because; morphological clipping, as in flu for
influenza, though Quirk et al. prefer to subsume clipping under phonological loss;
and semantic implication, as in Frankly, it is too late, where frankly can be
expanded in various ways, such as I am speaking frankly when I say.
The distinction between grammatical omission and semantic implication is
not always clear, since wherever there is grammatical omission there is also
semantic implication. One criterion that favours ellipsis as the analysis over
semantic implication is that for ellipsis the omitted words must be uniquely
recoverable (compare Quirk et al. 1985: 884); that criterion rules out frankly as
elliptical in the example given above because it is not possible to say precisely
what has been omitted. A further criterion that favours an elliptical analysis is
that what has been omitted is normally obligatory in the particular structure. This
requirement, however, is problematic, since it depends on what a grammar
defines as obligatory (compare Quirk et al. 1985: 885).
In this paper we focus on elliptical clauses. We therefore exclude ellipsis in
phrases, such as the italicized phrases in (1) and (2), which might be analyzed as
containing ellipsis:
(1) Archaeological and philological evidence in fact confirms that
early Rome was the product of a union of Latin and Sabine
communes, [ … ] [W2A-001–9]
(2) They are also enormously proud of the skill and courage of their
armed forces. [W2E-002–10]
112 SIDNEY GREENBAUM AND GERALD NELSON
In (1) the reconstructed full forms would be Archaeological evidence and

philological evidence and Latin communes and Sabine communes and in (2) the
skill of their armed forces and the courage of their armed forces.
If (1) had contained Archaeological evidence and philological evidence we
would not have considered the first noun phrase as the separate subject of the
elliptical clause “Archaeological evidence in fact confirms …”. Rather, we would
have regarded the two phrases as coordinated and functioning as subject of the
rest of the sentence. The same applies to verbs or verb phrases that share a
complement, as in (3):
(3) Developing first as an expression of society and an administrative
tool with the advent of the phonetic script and later printing press
it has changed and homogenised societies by a process of literacy
and literary coercion. [W1A-012–47]
(4) The greatest enemy of the Arab world is not and never has been
the United States. [W2E-001–72]
We also exclude phrases that might be analyzed as fragments of clauses because
the precise words cannot be recovered, as in this extract from a conversation:
(5) b: Are you going to go to that?
a: Which?
b: The first two Palmer and Firth Yes
a: Oh definitely [S1A-005–61 ff.]
Partly for the same reason we do not regard nonfinite and verbless clauses as
elliptical for corresponding finite clauses, but we also exclude them because they
are full clauses in their own right. Hence, the -ed participle clause in (6) and the
verbless clause in (7) are not considered elliptical:
(6) The last components to be fitted should be the hookswitch as this
tends to make the board rather clumsy if fitted too early in
construction [W2B-032–66]
(7) He stumbled downstairs, conscious of heavy footsteps behind him.
[W2F-001–23]
A different reason motivates us to exclude (8) and (9). They do not involve
ellipsis of repeated subjects and auxiliaries because their insertion would change
the meaning:
ELLIPTICAL CLAUSES IN SPOKEN AND WRITTEN ENGLISH 113
(8) I’ll try and fit it into some time during the week [ … ] [S1A-
059–53]
(9) When dropping a spanner on their foot it would be uh oh heck or
or dash it oh bother I’ve gone and dropped the on my f foot [S1B-
042–24 ff]
The intransitive substitute verb do (compare Quirk et al. 1985: 875) does not
permit ellipsis, since the insertion of the substituted items would result in an
ungrammatical construction:
(10) Thank goodness I didn’t say anything awful because I could’ve
done [S1A-091–299]
On the other hand, the operator do often involves ellipsis:
(11) b: No I don’t ever share cooking no
a: Do other people? [S1A-059–147f.]
(12) b: Did you take the camera with you?
a: Yeah I did but I didn’t take any photographs [S1A-036–157f.]
However, in some contexts if we insert the omitted items we convey an emphasis
that is not present in the original and thereby change the meaning. The addition
of draw in (13) is an instance in point:
(13) And Mr Deputy Speaker I would like to start by congratulating
my honourable friend on drawing the number he did in the ballot
[S1B-051–3]
Clauses functioning as backchannels (compare Greenbaum 1996: 6.1) may
contain verbs that are intransitive though otherwise they are transitive. The
absence of complements in backchannels such as you see, you know, I mean, I
know is perfectly normal. We therefore do not regard I know in (14) as elliptical:
(14) a: I don’t know how I’ll cope with anybody else’s
b: Yeah I know [S1A-005–75 ff.]
Similarly, we do not analyze constructions such as those in (15) – (17) as
involving the ellipsis of a subordinate clause (compare Quirk et al. 1985: 908, 909):
(15) I cooked Rosie dinner Did I tell you? [S1A-048–262 f.]
(16) Actually I knew I had seen it and I couldn’t think where [S1A-
046–17]
(17) Or you could put one either side in actual fact if you want to [S1B-
071–135]
We regard such constructions as grammatically complete.
The decisions that we have enumerated have in effect confined our
investigation of elliptical clauses to clauses that are finite, though as we shall
later see the verb phrase may itself be ellipted.
For this investigation we selected a sub-corpus of spoken and written texts
drawn from the British component of ICE, the International Corpus of English
(see Greenbaum and Nelson 1996a). In our previous studies of clauses in British
English (Greenbaum and Nelson 1995a, 1995b, 1996b, 1996c) we had used a
selection of 42 texts representing several varieties of speech and writing, a sub-
corpus that we had called the Leverhulme Corpus. For the present investigation
we added a further 40 texts so as to include samples from all the text categories
in ICE-GB, for which purpose we provided two texts from each category that
were not in the Leverhulme Corpus. A text contains about 2,000 words; the
number of words in the present sub-corpus of 82 texts comes to 176,968, of
which 115,107 are from spoken texts and 61,861 from written texts.
The composition of the sub-corpus is shown in Figure 1, together with the
number of texts in each category. The total number of words in each of the
major categories is given in brackets. References to citations from the sub-corpus
consist of three sets of digits, for example S1A-046–17. The first set is the
category, in this instance a conversation, the initial S indicating that the text is
from speech in contrast to the initial W used for written texts. The second set is
for the identity number of this text (046). The third set identifies the number of
the text unit (17) within the text 046. A list of the sources of the texts appears
in the Appendix to Greenbaum (1996).
Figure 1. The composition of the sub-corpus

SPEECH (54) [115,107 words] WRITING (28) [61,861 words]
Conversations (22) – S1A [46,060] Non-printed (10)
face-to-face conversations (20) student essays (4) – W1A [8,351]
telephone conversations (2) letters (6) – W1B [13,184]
Public dialogues (15) – S1B [32,316] Printed (18)
class lessons (2) academic writing (4) – W2A [9,187]
broadcast discussions (5) non-academic writing (4) – W2B [9,402]
broadcast interviews (2) press reports (2) – W2C [4,428]
parliamentary debates (2) instructional writing (4) – W2D [8,507]
legal cross-examinations (2) administrative writing (2)
business transactions (2) skills and hobbies (2)
Unscripted monologues (11) – S2A [24,041] press editorials (2) – W2E [4,264]
spontaneous commentaries (2) fiction (2) – W2F [4,538]
unscripted speeches (5) Total number of texts = 82
demonstrations (2) Total number of words = 176,968
legal presentations (2)
Scripted monologues (6) – S2B [12,690]
broadcast news (2)
broadcast talks (2)
non-broadcast speeches (2)
For this research we were interested in the type of clausal ellipsis and its location
within the clause, and we annotated these features manually in the corpus. We
also decided to find out whether the ellipsis type and location correlated with
differences between our text categories.
In reporting the results of our investigation we make an initial distinction
between two major types of elliptical clauses, depending on whether the ellipsis
is or is not conditioned by the coordination of the elliptical clause with another
clause. We were motivated to make this as our initial distinction because a
preliminary inspection of the results highlighted this as differentiating important-
ly between our text categories. Ellipsis that is not conditioned by coordination we
termed independent ellipsis, in contrast to coordination ellipsis.
Independent ellipsis is illustrated in (18), coordination ellipsis in (19):
(18) That little plant that grows doesn’t matter what the soil conditions
are whether it’s very acidic or chalk or whatever [S1A-036–209]
(19) And uh so we unpacked our stuff and trooped in [S1A-036–140]
Coordination ellipsis may be asyndetic, as in (20):
(20) Carling calls for the ball inside his own half gives it to Andrew
[S2A-002–6]
In all three instances the subject is ellipted. We know that the ellipted subject is
it in (18) because we know that this type of construction requires anticipatory it
as subject of the main verb. In (19) and (20), on the other hand, the recoverabili-
ty of the subject depends on the presence of the subject in the preceding
coordinated clause. In (19) the recoverable subject we is identical with the first
subject, whereas in (20) the recoverable subject he is the pro-form for Carling.
Altogether, there were almost a thousand instances of clause ellipsis in our
sub-corpus. Table 1 contrasts the spoken and written components with respect to
whether the ellipsis is independent ellipsis or coordination ellipsis. Since the
number of texts varies between the categories and there are also some differenc-
es in the number of words in a text, Table 1 and subsequent tables give in
parentheses the number of elliptical clauses per 2,000 words.
Table 1. Independent and coordination ellipsis in speech and writing.

Category Independent ellipsis Coordination ellipsis Total
speech 356 (6.2) 278 0(4.8) 634 (11.0)
writing 039 (1.3) 310 (10.0) 349 (11.3)
Total 395 (4.5) 588 0(6.6) 983 (11.1)
Table 1 exhibits striking contrasts between the two components in our corpus in
the type of ellipsis they favour. Coordination ellipsis occurs more than twice as
frequently in writing, a result that is in harmony with what Meyer found in his
study of coordination ellipsis in American English (Meyer 1995: 251). Citing
Tannen (1989: 47–53), Meyer argues that repetition is more important in speech
because of the transitory nature of speech. Repetition helps the listener to
understand what is being said by making the discourse less dense. Full forms, which
involve repetition, tend therefore to be preferred over ellipted forms in speech.
Even more striking is the difference between the two components in their use
of the other type of ellipsis. Independent ellipsis emerges as characteristic of speech.
At first sight, the preference for independent ellipsis in speech seems to
subvert the explanation that repetition is helpful to the listener and therefore there
should be less ellipsis of this type too in speech. However, unlike coordination
ellipsis, independent ellipsis does not necessarily involve anaphoric reference.
The full forms that correspond to independent ellipsis do not necessarily repeat
previous words. On the other hand, speech provides ample scope for deictic
reference, since speech is generally located in situations where participants are
in visual and aural contact. Thus, in (21), where the context is a game of Scrabble,
it is obvious what the grammatical subjects are even though there is no repetition:
(21) a: Got an E?
b: No Could have
The intonation of A’s utterance in (21) would indicate that it is a question, and
the interpretation of the missing subject and auxiliary is clearly Have you. In B’s
utterance, the question-response pair provides the clue that the omitted subject is
I and the omitted direct object is an E. This example also shows that our use of
“independent“ in “independent ellipsis” must not be interpreted literally. The
interpretation of B’s utterance is in fact dependent on the interpretation of A’s
utterance. By independent ellipsis we therefore mean ellipsis that is independent
of the influence of coordination and we acknowledge that other contextual factors
may be at work.
Like coordination ellipsis, independent ellipsis is motivated by a desire for
economy of expression. That has to be balanced in speech by a need to be easily
comprehensible. Independent ellipsis is used where the balance can be shifted —
or needs to be shifted — towards economy. Speech can sometimes afford to be
more economical because, if utterances puzzle the listeners, they can ask for
clarification and elaboration. Such an assumption suggests that there might be
differences between dialogues and monologues, since only dialogues can
generally provide the opportunity for listeners to intervene in this way. Table 2
displays the results for the categories in the spoken component.
Table 2. Independent and coordination ellipsis in the spoken categories.

conversations 236 (10.2) 063 (2.7) 299 (13.0)
public dialogues 049 0(3.0) 061 (3.8) 110 0(6.8)
unscripted monologues 067 0(5.6) 113 (9.4) 180 (15.0)
scripted monologues 004 0(0.6) 041 (6.5) 045 0(7.1)
The relative frequency of independent ellipsis in conversation meets our expecta-

tion. It is satisfyingly high, well above the frequency for the spoken component
as a whole (compare Table 1). The low frequency in scripted monologues (whose
source is in writing) was also expected, since most of the material (broadcast
news and broadcast talks) does not lend itself to deictic reference, nor is there an
opportunity for the dependencies found in interaction between participants in a
dialogue. In accordance with our hypothesis, the public dialogues exhibited a
higher proportion of independent ellipsis than we observed for the written
component, though the frequency was somewhat less than for coordination
ellipsis. However, the results for unscripted monologues were surprising, since
the frequency of independent ellipsis in that category (5.6 per 2,000 words) was
distinctly higher than for the public dialogues (3.0). We searched for an explana-
tion by examining the varieties of texts within the unscripted monologues.
Our investigations showed conclusively that the anomalous results for the
unscripted monologues were induced by the category of spontaneous commentar-
ies. The two texts in this category were radio sports commentaries. Table 3
contrasts these with the set of nine other texts in the category of unscripted
monologues, comprising unscripted speeches (5), demonstrations (2), and legal
presentations (2).
Table 3. Independent and coordination ellipsis in unscripted monologues.
Commentaries 62 (29.1) 46 (21.6) 108 (50.7)
Others 05 0(0.5) 67 0(6.8) 072 0(7.3)
The sports commentaries are replete with deictic references, since they report and
evaluate the movements of the players as they occur. For games such as soccer
and rugby, which are the topics of our two commentary texts, the commentators
are forced to rely on ellipsis to keep in time with the swift succession of the
players’ acts. Here are typical examples of the use of independent ellipsis by
commentators:
(22) England have the possession
John Barnes just approaching the centre circle
Leaves it for Geoff Thomas who plays it square to Derigo on the
far left
Derigo plays it back to Thomas just on the half-way line
Now just clips it forward
Smith meets it
Gets it down dangerously [S2A-001–93ff.]
(23) The scrum fifty metres in from the far touch-line the Irish right
[S1A-002–177]
(24) So Francis at the front here [S2A-022–211]
In (22) the auxiliary is is ellipted in the first elliptical construction and the subject in
the other constructions. In (23) and (24) it is the main verb is that is ellipted.
Our analysis of the text varieties in the spoken component has revealed that
it is in the dialogues that we find most instances of independent ellipsis. Com-
mentaries are exceptional in their frequent use of independent ellipsis. It is used
because of pressure of time, the requirement to keep the speech in time with the
acts that are being described or evaluated.
We now turn to the categories in the written component. These are dis-
played in Table 4.
Table 4. Independent and coordination ellipsis in the written categories.

essays 0 29 0(6.9) 29 0(6.9)
letters 15 (2.3) 72 (10.9) 87 (13.8)
academic writing 0 27 0(5.9) 27 0(5.9)
non-academic writing 0 55 (11.7) 55 (11.7)
press reports 3 (1.4) 15 0(6.8) 18 0(8.1)
instructional writing 2 (0.5) 62 (14.6) 64 (15.0)
editorials 4 (1.9) 11 0(5.1) 15 0(7.0)
fiction 15 (6.6) 39 (17.2) 54 (23.8)
Within the written component, coordination ellipsis is most frequent in fiction

and in instructional writing. There is some evidence that written narratives favour
coordination in linking descriptions of events (compare Beaman 1984: 57), and
therefore there is greater scope for coordination ellipsis, as exemplified in (25):
(25) Brett swallowed, cleared his throat and forced out the words.
[W2F-001–54]
In this respect it is significant that 34 of the 39 instances of coordination ellipsis
appear in the narrative portions and only five in the dialogue. As to instructional
writing, instructions sometimes present more than one condition or situation, a
circumstance conducive to coordination ellipsis, since otherwise an identical
subject and perhaps also an identical auxiliary would be repeated. An extreme
example, offering six instances of coordination ellipsis within one long sentence,
is displayed in (26):
(26) If you do not get an extra amount for a wife or husband you may
be able to get one for someone looking after a child if: you get
Child Benefit for the child, or can be treated as getting it, and the
person is not earning more than the extra amount for an adult
dependant, and is living with you, or is living elsewhere but
maintained by you, or is living elsewhere but working for you,
provided the cost to you of employing him or her is not less than
the extra amount for an adult dependant. [W2D-002–9]
Three of the written categories in Table 4 contain no independent ellipsis. The
highest frequencies for independent ellipsis are found in fiction and in letters. All
the instances in fiction but one occur in representations of conversation, and
should therefore be classed in this use with the conversations in the spoken
component. Here is an example:
(27) “I told Dad what I thought last night,” she said quietly. “You must
have heard.”
“Had my headphones on — didn’t miss much.” [W2F-001–170.]
There are six texts of letters, but seven of the fifteen instances of independent
ellipsis are indebted to one letter-writer. We therefore ascribe the frequency for
letters not to a feature of this category but to the idiosyncratic usage of a
particular writer, who offers sentences such as Miss you and Can’t write
anymore. However, it must be said that personal letters, which make up four of
the composite texts in the category, readily allow elliptical sentences of this type.
Table 5. Locations for independent ellipsis in the spoken categories.

Category -S -S,-aux -V, -C -C Other
conversations 068 (2.9) 036 (1.7) 075 (3.3) 38 (1.6) 09 (0.8)
public dialogues 005 (0.3) 004 (0.2) 024 (1.5) 13 (0.8) 03 (0.2)
unscripted monologues 028 (2.3) 014 (1.2) 005 (0.4) 0 20 (1.7)
scripted monologues 001 (0.2) 0 003 (0.5) 0 0
Total 102 (1.5) 054 (0.9) 107 (1.9) 51 (0.9) 42 (0.7)
Table 6. Locations for coordination ellipsis in the spoken categories.

Category -S -S, -aux Other
conversations 039 (1.7) 20 (0.9) 04 (0.2)
public dialogues 038 (2.3) 20 (1.2) 03 (0.2)
unscripted monologues 081 (6.7) 24 (2.0) 08 (0.7)
scripted monologues 029 (4.6) 11 (1.7) 01 (0.2)
Total 187 (3.2) 75 (1.3) 16 (0.3)
We now focus on the spoken component to examine the major locations for
ellipsis. These are exhibited in Table 5 for independent ellipsis and in Table 6
for coordination ellipsis.
There are four major locations for ellipsis in the spoken category, though
only two constitute major categories for coordination ellipsis. Two of the
locations are at the beginning of the clause and two at the end. Those at the
beginning have ellipsis of the subject (-S) or ellipsis of the subject plus the
auxiliary (-S, -aux). Those at the end have ellipsis of the main verb and any
complement of the verb (-V, -C) or ellipsis of the complement alone (-C).
Examples of the four locations appear below:
-S
(28) Haven’t wasted much cash [S1A-006–193]
-S, -aux
(29) Be leaving about half five or something I think [S1A-006–306]
-V, -C
(30) a: […] you’ve seen Teenage Mutant Ninja Turtles haven’t you
b: I have [S1A-006–156f.]
-C
(31) a: Didn’t there used to be deer in Richmond Park
b: Yeah there still are [S1A-006–225ff.]
Table 5 shows that all four locations for ellipsis occur only in the dialogues.
Ellipsis of the complement (-C) does not occur at all in the monologues, and this
suggests that it may be dependent on interaction between speakers, as in (31),
where the complement which is ellipted is recoverable from the question posed
by another speaker. Looking at the table as a whole, the -V,-C location is the
dominant type, followed by -S. However, there is a clear difference — on the
surface at least — between dialogues and monologues. In dialogues, the predom-
inance of -V,-C is largely due to the public dialogues, which display very little
subject ellipsis (-S). In conversations there is little difference between these two
types, as there is much more subject ellipsis than in the public dialogues. For
monologues, Table 5 shows the reverse pattern: this category displays much
more subject ellipsis than any other type, especially among the unscripted texts.
However, we found once again that this result was due to the spontaneous
commentaries: all 28 instances of subject ellipsis in this category occurred in the
two sports commentaries. The idiosyncracies of this type of monologue have
again produced anomalous results.
We now turn to the corresponding results for the written categories in our
corpus. Table 7 shows the locations for independent ellipsis, and it is noticeable
that in this mode, all four locations are available only in fiction, where there is
a fairly even distribution between initial and end ellipsis. Although the figures
are small, Table 7 shows that the -S and -V,-C locations are again the dominant
types. The figure for -V,-C, however, is largely due to the fiction category. In
Table 5 we saw that this location is characteristic of dialogue, and the evidence
from fiction supports this: 4 of the 5 instances in fiction occur in the dialogue
portions of novels.
Table 7. Locations for independent ellipsis in the written categories.

Category -S -S,-aux -V,-C -C Other
essays 0 0 0 0 0
letters 10 (1.5) 5 (0.8) 0 0 0
academic writing 0 0 0 0 0
non-academic writing 0 0 0 0 0
press reports 0 0 0 0 3 (1.4)
instructional writing 0 0 2 (0.5) 0 0
editorials 0 0 2 (0.9) 1 (0.5) 1 (0.5)
fiction 5 (2.2) 1 (0.4) 5 (2.2) 3 (1.3) 1 (0.4)
Total 15 (0.5) 6 (0.2) 9 (0.3) 4 (0.1) 5 (0.2)
Table 8. Locations for coordination ellipsis in the written categories.

Category -S -S,-aux Other Total
essays 25 0(6.0) 4 (0.9) 0 29 0(6.9)
letters 51 0(7.7) 20 (3.0) 1 (0.1) 72 (10.9)
academic writing 21 0(4.6) 6 (1.3) 0 27 0(5.9)
non-academic writing 44 0(9.3) 5 (1.1) 6 (1.3) 55 (11.7)
press reports 5 0(2.2) 4 (1.8) 6 (2.7) 15 0(6.7)
instructional writing 36 0(8.5) 18 (4.2) 8 (1.9) 62 (14.6)
editorials 6 0(2.8) 3 (1.4) 2 (0.9) 11 0(5.1)
fiction 34 (15.0) 5 (2.2) 0 39 (17.2)
Total 222 0(7.2) 65 (2.1) 23 (0.7) 310 (10.0)
The locations for coordination ellipsis in speech and writing are shown in
Table 6 and Table 8 respectively. With this type of ellipsis there are only two
main locations, -S and -S,-aux.
-S
(32) The Romans themselves saw in this practice a major factor in their
rise to world power and traced it back to the legendary origins of
their city. [W2A-001–8]
-S,-aux
(33) Get there early enough and you can queue up and get a tour round
the White House [S2B-021–5]
Both -S and -S,-aux occur in initial position in the second coordinated clause, so
they may be considered identical in terms of location. If we combine the two, we
find that in speech as a whole, this initial position is favoured in 94% of cases
of coordination ellipsis, while in writing the corresponding figure is 92.6%.
Meyer (1995: 247–8) found that the initial position is favoured in coordination
ellipsis generally, and suggested that this position has a high potential for ellipsis
because it is likely to contain old information which has low information value. As
such it does not need to be repeated, and may be ellipted without any loss of clarity.
In both speech and writing, ellipsis of the subject alone is the dominant
type, and there is little variation within the two modes. Subject ellipsis may be
considered the norm when clauses with the same subject are coordinated, as in
(32), where repetition of the subject would be considered redundant. When the
subjects are different, ellipsis of the subject is not available.

We have shown that there is a considerable difference between speech and
writing in the type of clausal ellipsis they favour. Independent ellipsis is charac-
teristic of speech, and especially of conversations, while coordination ellipsis is
favoured in writing. Dialogues are characterized by independent ellipsis of the
verb and its complement, and we have speculated that this type of ellipsis
depends largely upon interaction between speakers, the ellipted items being
recoverable from previous speaker turns. The locations for coordination ellipsis
are much less dependent upon mode or text category. We found that ellipsis of
clause initial elements — the subject or the subject plus the auxiliary — is
favoured throughout the corpus, regardless of mode or category. However, our
results for unscripted monologues, and so for speech generally, have been
heavily influenced by the sports commentaries. These contain an idiosyncratic
type of discourse with a great deal of both types of ellipsis, and probably deserve
a separate study.
References
Beaman, K. 1984. “Coordination and Subordination Revisited: Syntactic Com-

plexity in Spoken and Written Discourse”. Coherence in Spoken and Written
Discourse ed. by D. Tannen, 45–80. Norwood, N.J.: Ablex.
Greenbaum, S. 1996. The Oxford English Grammar. Oxford: Oxford University
Press.
Greenbaum, S. and G. Nelson. 1995a. “Clause Relationships in Spoken and
Written English.” Functions of Language 2. 1–21.
Greenbaum, S. and G. Nelson. 1995b. “Nuclear and Peripheral Clauses in
Speech and Writing”. Studies in Anglistics ed. by G. Melchers and B.
Warren, 181–190. Stockholm: Almqvist and Wiksell.
Greenbaum, S. and G. Nelson. 1996a. “The International Corpus of English
(ICE) Project”. World Englishes 15. 5–17.
Greenbaum, S. and G. Nelson. 1996b. “Complement Clauses in English”. Using
Corpora for Language Research ed. by J. Thomas and M. Short, 76–91.
London: Longman.
Greenbaum, S. and G. Nelson. 1996c. “Positions of Adverbial Clauses in British
English”. World Englishes 15. 71–83.
Meyer, C.F. 1995. “Coordination Ellipsis in Spoken and Written American

English”. Language Sciences 17. 241–269.
Quirk, R., S. Greenbaum, G. Leech, and J. Svartvik. 1985. A Comprehensive
On the nature of ?I believe Jack to arrive tomorrow*
Hisashi Higuchi
There is an apparently peculiar restriction of some sort on NP-to-V… type

complements of verbs such as believe. Hudson (1971:209) mentions the restric-
tion, which is exemplified in (1a):
(1) a. *I didn’t believe John to die.
b. I didn’t believe John to be dying.
c. I didn’t believe John to have died so long before.
Postal (1974:25, footnote 25) notes the same point, contrasting the sentence pair
in (2a–b) with that in (3a–b) below, offering no explanation:
(2) a. I believe that Bill will win tomorrow.
b. *I believe Bill to win tomorrow.
(3) a. I expect that Bill will win tomorrow.
b. I expect Bill to win tomorrow.
Despite the unacceptability of (1a), all of the following are quite acceptable:
(4) a. I didn’t believe John to be going to die.
b. I didn’t believe John to be about to die.
c. I didn’t believe John to be likely to die.
Hudson (1971:209), speaking of (1a) and (1b) above, says: “The restriction
seems to be that in these constructions the referent of the noun-clause must in
general be a ‘state’ rather than an ‘event’”. For (1c) he adds: “if the ‘event’ is
earlier in time than the time of ‘believing’ (or whatever type of cognition may be
involved), then the restriction is waived”.
Mair (1990) takes the view that this restriction has crucial bearings on the
distinction between “monotransitive” and “complex transitive” patterns of
128 HISASHI HIGUCHI
complementation in his descriptive framework. While admitting that “complex

transitive” patterns show strong affinity with “monotransitive” ones, Mair
(1990:175) says one important difference between the two constructions is that
in “complex transitive” patterns “the infinitive generally has to be stative or
perfective”, where the term “perfective” seems to refer to a construction involv-
ing the perfect have.
This issue has been discussed in the literature often in terms of stativity,
possibly because the notion of stativity could be extended to cover some aspects
of the perfective meaning, inasmuch as there is a sense in saying that having
done something is a kind of state rather than an event. Borkin (1984:61ff)
speaks of stativity as one of the semantic characteristics of the kind of “raising”
constructions she deals with, that is, NP-to-V… complements of believe-type
verbs. Similarly, Menzel (1975:105) states: “ … verbs of belief allow this rule
[Raising to Object] only with stative verbs”. Stockwell et al. (1973:570f) propose
a condition to the effect that a complement of the relevant sort cannot be derived
unless the verb in the complement is marked as “stative”. Kilby (1984:154) also
says: “The restriction on the verbs [of believe-type verb complements of the
relevant sort] … seems to be largely that they should be stative”, giving the
following contrast:
(5) a. I believe her to hate knitting.
b. ??I believe her to hit John.
He then adds that “a generic interpretation of an active verb is quite possible”,
giving the following example:
(6) I believe her to beat her children.
While the act of beating is certainly not a state, we could probably extend the
notion of stativity to cover generic or habitual uses of verbs such as beat,
regarding “beating children regularly” as a kind of state characterised by the
habitual behaviour of beating children.
Thus, the characterisation of the restriction in terms of stativity has some
intuitive appeal. But it is not altogether clear exactly what it means for a
predicate to be stative, and the restriction doesn’t seem to be as straightforward
as it may first appear to be. It is not so easy to distinguish systematically
between predicates such as die that are not allowed in the construction, and such
predicates as be tall, live, be dying, be going to die, be about to die and the like,
as well as verbs such as beat with their habitual readings. Moreover, different
ON THE NATURE OF ?I BELIEVE JACK TO ARRIVE TOMORROW 129
matrix verbs may impose different constraints on their complements: according

to Borkin (1984:58), both I know Sam to be competent in everything he does and
I believe Sam to be competent in everything he does are fine, but *I know Sam to
be ready to leave now is not as good as I believe Sam to be ready to leave now.
In view of this, it is not surprising that while there have been attempts to
formalise this restriction in some way or another, none of them has turned out to
be satisfactory. Watts (1983:67), who takes *John believes Mary to come late to
be “ungrammatical”, says “it may even be the case that only be with a predicate
NP, AP or PP is allowed” in the relevant construction. But obviously this is a
very unsatisfactory generalisation which doesn’t cover (1b-c) and (6).
Stockwell et al. (1973:570f) attempt to account for this restriction in such a
way that the rules of grammar will not generate unacceptable sentences such
as (5b). This wouldn’t account for the acceptability of (6), however, unless beat
is marked as “stative” for its generic interpretation — which would make such
markings in the lexicon vacuous.
May (1987) proposes a similar account: he formulates a condition requiring
that an infinitival complement of the verb believe with a verb such as murder
have positive aspectual marking, according to which ?Linda believes Gary to
murder David can be shown to be ungrammatical, but Linda believes Gary to
have murdered David and Linda believes Gary to be murdering David are allowed.
But it is not clear from his discussion whether his formalisation would allow
sentences such as (6), where beat, an “event” verb like murder, appears with a
generic interpretation.
Thus the constraint does not seem to be statable strictly in grammatical
terms, and attempts to formalise it so as to incorporate it into the description of
the grammar seem almost bound to fail.
The constraint may have something to do with the fact that infinitive clauses
cannot include a modal. Dixon (1991:223) says: “A TO clause cannot include a
Modal and there is no means of coding the information shown by the Modal in
I know that Mary may/must/should be clever into a Judgement TO construction
[NP-to-V… complementation pattern of believe-type verbs]”. Accordingly the
corresponding finite clause counterparts of NP-to-V… complements of believe-
type verbs will be clauses with no modal, and non-modal finite clauses are more
often associated with stative (in an extended sense, including habitual) meaning
than specific future action meaning. The oddity of ?I believe him to die will be a
consequence of the unnatural ?I believe that he dies.
With verbs such as intend, the interpretation of the complement is more
130 HISASHI HIGUCHI
dependent on the meaning of the matrix verb, and this is a source of modality
meaning characteristically associated with complements of this kind. The
information that is not in the complements can be readily interpreted from the
meaning of the matrix verb.
With verbs such as believe, on the other hand, the meaning of a complement
is more independent of that of the matrix clause, with the complement constitut-
ing a proposition of its own. Information regarding time reference can be
encoded, as in I believe the mission to have arrived. When there is no such
marking, NP-to-V… complements will have “stative” or “habitual” interpreta-
tions: I believe him to go to school corresponds to I believe that he goes to school.
It may also be noted that the addition of specific future time reference does not
help: ?I believe him to go to school tomorrow cannot be used for I believe that he
will go to school tomorrow.
The so-called “futurate”, where a non-modal finite clause does express a
specific future situation, could have a NP-to-V… counterpart: compare The sun
sets at 6:30 this evening and They know the sun to set at 6:30 this evening. This,
however, may well be less frequent, because the “futurate” expresses some
objectively determined event or state of affairs that is to be realised (such as the
setting of the sun), which does not fit the subjective meaning generally associat-
ed with the NP-to-V… complements of believe-type verbs. Other non-stative uses
of non-modal finite clauses fail to be complements of believe-type verbs, most
probably because their meanings are not compatible with the verbs. The “instan-
taneous present” as in Now I put this rabbit in this box, for example, cannot be an
object of belief or knowledge.
In this connection the following contrast (from Palmer 1987: 196) may be
mentioned:
(7) a. *I believe Mary to arrive tomorrow.
b. ?Mary is believed to arrive tomorrow.
The subjective meaning associated with (7a) is removed to a certain extent in
(7b), due to the absence of the NP designating the one who believes. This, I
think, makes the “futurate” reading more likely in (7b) than in (7a). In general,
the removal of the subjective meaning associated with the construction by this
means seems to have the effect of improving acceptability. Wierzbicka (1988: 52)
notes that Mary is believed to drive well is better than I believe Mary to drive well,
although the complement predicate here has a habitual reading.
There are a few other environments where the constraint is somehow
ON THE NATURE OF ?I BELIEVE JACK TO ARRIVE TOMORROW 131
weakened. Although the acceptability judgements of sentences in such cases can

be rather subtle, they might be worth noting here.
When the perfect have is involved in the matrix verb, the restriction is
somehow weakened, as we can see from the following examples (from Bolinger
1974:73; 89–90) with know:
(8) a. *I know them to run for years.
b. I’ve known them to run for years.
c. I know them to have run for years.
(8b) is judged as acceptable although the NP-to-V… sequence is identical to that
in (8a), which is not acceptable. Bolinger says that (8b) means that they have
run: “if I’ve known them to run for years, they must have run for years.” The
acceptability of (8b), then, parallels that of (8c). He relates this to such cases as
negation (I think she is not there versus I don’t think she is there) or I’ll hope to
see you there where, he says, “the future of see attaches to hope”. I do not know
whether such a relation can be maintained.
Another environment in which the constraint is weakened is the so-called
participial construction:
(9) a. *He guessed roses to appeal to her more.
b. Guessing roses to appeal to her more, he sent her a dozen of
the best.
Of these examples involving guess, again from Bolinger (1974: 74), (9b) is more
acceptable than (9a), although the verb takes the same NP-to-V… complement
in both examples. The reason for this is not clear, but it at least indicates that the
matter doesn’t have much to do with the configuration of words in a NP-to-V…
complementation pattern but rather is related to other factors.
The agentless passive construction may improve otherwise unacceptable
sentences, too. May (1987: 32) judges (10c) to be more acceptable than (10b):
(10) a. ?Hamletbelieved Claudius to murder his father.
b. ?Hamletbelieved his father to be murdered by Claudius.
c. Hamlet believed his father to be murdered.
The reason for this seems to be that to be murdered in (10c) is interpreted
resultatively, probably because there is no explicit agent phrase, with its meaning
parallel to “to have been murdered”. Dixon (1991: 223) discusses the sentence I
believed John to be beaten, in connection with the fact that the complement
132 HISASHI HIGUCHI
predicate in this kind of construction is most often be. As (10b) shows, however,
the presence of be is not the sole factor, but it is rather the absence of the NP
designating the agent of the action described in the complement predicate that
seems responsible for the kind of resultative interpretation which makes such
sentences more acceptable.
Note
* This paper is based on part of Higuchi (1993).
References
Bolinger, D. 1974. “Concept and Percept: Two Infinitive Constructions and their
Vicissitudes”. World Papers in Phonetics: Festschrift for Dr Onishi’s Kiju,
65–91. Tokyo: Phonetic Society of Japan.
Borkin, A. 1984. Problems in Form and Function. New Jersey: Ablex.
Dixon, R.M.W. 1991. A New Approach to English Grammar, on Semantic
Principles. Oxford: Clarendon Press.
Higuchi, H.Q. 1993. A Linguistic Study of NP-to-V… Complements in English.
MA Thesis, The University of Queensland.
Hudson, R. 1971. English Complex Sentences: An Introduction to Systemic
Grammar. Amsterdam: North Holland.
Kilby, D. 1984. Descriptive Syntax and the English Verb. Kent: Croom Helm.
Mair, C. 1990. Infinitival Complement Clauses in English: A Study of Syntax in
Discourse. Cambridge: Cambridge University Press.
May, T. 1987. “Verbs of Result in the Complements of Raising Constructions”.
Australian Journal of Linguistics 7. 25–42.
Menzel, P. 1975. Semantics and Syntax in Complementation. The Hague: Mouton.
Palmer, F.R. 1987. The English Verb. (2nd ed.) London: Longman.
Postal, P.M. 1974. On Raising. Cambridge, Mass.: MIT Press.
Stockwell, R.P., P. Schachter, and B.H. Partee. 1973. The Major Syntactic
Structures of English. New York: Holt, Rinehart and Winston.
Watts, R.J. 1983. “On Infinitival Complement Clauses”. Studia Anglica Posna-
niensia 16. 45–69.
Wierzbicka, A. 1988. The Semantics of Grammar. Amsterdam/Philadelphia: John
Benjamins.
Intransitive prepositions
Are they viable?
David Lee
1. Introduction
The aim of this paper is to challenge the assumption made in generative gram-
mar that native speakers access syntactic rules based on projections from lexical
categories. This challenge arises out of a practical problem: how to classify a set
of words in English illustrated by such commonplace examples as aboard,
abroad, away, downstairs, here, home, now, somewhere, then, there, when,
where. I will argue that these words are problematic with respect to the distinc-
tion between adverbs and prepositions and, more significantly, that the nature of
this problem calls into question widespread assumptions about the relationship
between word classes and phrase markers.
The issues addressed here arise directly out of many hours of discussion in
workshops for the Cambridge Grammar of English. The difficult question of
where to draw the boundary between prepositions and adverbs is one that
exercised us over several sessions and for which a number of creative solutions
were canvassed at various stages. The particular format adopted here (argument,
counterargument, response) is chosen in order to attempt to capture something of
the flavour of those debates. Much of the material comes directly out of the
workshop discussions and therefore owes a great deal to those who took part:
Ray Cattell, Peter Collins, Pam Peters, Peter Peterson, Geoff Pullum — but most
of all to Rodney Huddleston, who was always the dominant figure. However, I
suspect that none of these would wish to identify themselves too closely with the
particular “spin” that I have imposed on the material here — in particular the
134 DAVID LEE
claim that the best framework for conceptualizing this problem is an emergence
theory of category structure.
Current models of generative grammar assume that a language contains a
small number of discrete lexical categories from which syntactic rules project
phrase markers, the categories in question being defined by clusters of morpho-
syntactic properties. For such rules to be viable, each word needs to be unambig-
uously assigned in the lexicon to the relevant word class. There is no problem in
reconciling this requirement with the widely recognized fact that word class
membership is characterized by fuzziness, since peripheral members of the noun
category (for example) can be unambiguously marked as nouns, even though
they may not be as ‘nouny’ as central members. However, in this model we
would not expect to find a set of words where no clear category assignment is
possible, since this would make the projection rules unworkable.
One way of dealing with such a problem would be to assume that native
speakers have been more successful than linguists in their identification of the
relevant categories. This would be rather surprising, given that word classifica-
tion is such a basic issue. Alternatively, one might surmise that traditional ways
of conceptualizing word classes and their role in the grammar is seriously
flawed. This would require an alternative way of dealing with the undeniable fact
that morphosyntactic properties do indeed manifest strong clustering patterns in
many cases.
2. The preposition analysis
Traditionally, words like aboard, abroad, away, downstairs, here, home, now,
somewhere, then, there, when, where and so on are classified as adverbs.
However, it has been known for some time that there are strong arguments for
analysing them as (intransitive) prepositions — cf. Emonds’s (1972: 550–3)
treatment of afterwards, apart, away, back, downstairs, now, overhead, together
and Jackendoff’s (1973: 350) treatment of home, here, there. This claim involves
a fundamental reanalysis of the preposition category as a whole. I will argue here
that the question of whether these words (which I will refer to as the “X-words”)
are prepositions or adverbs is in fact an unanswerable question, since it is based
on an assumption that generative theory has taken over from traditional grammar
— namely, that there is a finite set of discrete word classes, such that all words
in the language are unambiguously assigned to one class or another (or to more
INTRANSITIVE PREPOSITIONS 135
than one class in the case of words like catch, fast and walk). Instead, I will
argue that these words are problematic because the clustering patterns that
differentiate adverbs and prepositions (as this class would be characterized under
the proposed reanalysis) are not sufficiently coherent to define a clear boundary
between the two categories.
The basis for classifying the X-words as adverbs is a relatively slender one.
They are certainly not prototypical adverbs since they lack typical adverbial
properties such as -ly morphology and the ability to take degree adverbs as
dependents. The one property that they do share with indisputable adverbs is that
they can function as heads of phrases which fill the adjunct function at clause-level:
(1) He found the money immediately
(2) He found the money there1
The strongest reason for classifying these words as adverbs is in fact a negative
one. Traditionally, adverbs are a residual category, encompassing a very disparate
set of words that cannot plausibly be assigned to any other category.
The arguments for taking the X-words to be prepositions were originally
advanced by Emonds (1972) and Jackendoff (1973) and have recently been
reviewed by Burton-Roberts (1992). The arguments are based on the observation
that these words share a significant number of properties with prepositions or PPs:
(a) Unlike (prototypical) adverbs these words can function as verb complements
and the verbs in question are precisely those that take PP complements.2
(b) They can post-modify nouns.3
(c) They cannot premodify adjectives or other adverbs.
(d) They can function as complements of prepositions.
(e) They can take PPs as complements.
(f) Most of them can take right and straight as pre-modifiers.
(g) Many of them occur in the Locative Inversion construction (e.g. Into the
room ran John!).4
(h) A few (particularly away) occur in the “expletive” construction exemplified
by the sentence Into the river with the traitor!
These observations suggest that the X-words should be analysed as “intransitive”
prepositions; in other words, the category of prepositions should be extended to
include words that do not require a complement. On this analysis, the traditional
argument for taking the X-words to be adverbs (namely that they can function as
136 DAVID LEE
heads of phrases filling the adjunct function in the clause) collapses, since, if
some prepositions are intransitive, then this property no longer distinguishes
adverbs from prepositions.
There are a number of general considerations which make this proposal an
attractive one. In the traditional framework prepositions are anomalous with
respect to the other major word classes, in that they are the only words that take
an obligatory dependent. NPs, VPs, AdjPs and AdvPs may all consist of a head
only, but a PP may not. The notion of “intransitive preposition” removes this
anomaly. Moreover, the analysis solves a problem involving understood comple-
ments. In sentences like (3), (4) and (5), for example, it is not obvious in the
traditional framework whether to treat outside, before, behind as prepositions
with understood complements or as adverbs.
(3) John was in the house but I stayed outside
(4) Sue left at three but Jo left before
(5) I’ll stand in front of Ed and you stand behind
This point also impinges on “particles”. For example, it is arguable that in has an
understood complement in (6) but less plausible to make the same claim about
up in (7):
(6) I let the cat in
(7) I picked the cat up
In fact the main point of the arguments advanced by Emonds (1972) and
Jackendoff (1973) was to show that the behaviour of “particles” in general
provides a particularly strong case for extending the class of prepositions to
include intransitive members (see also Dixon 1982).
The proposal is also attractive from a semantic point of view in that all the
X-words express locative or temporal meanings — meanings that are typically
expressed by PPs rather than by AdvPs. In many cases, they can be paraphrased
by PPs. In this sense the relationship between intransitive prepositions and PPs
can be seen as similar to that between pronouns and NPs. The circumstances in
which the pronoun she can be substituted for the NP Mary, for example, are very
similar to those in which there can be substituted for the PP in the garden (that
is, when the context makes the fuller specification unnecessary). Moreover, just
as pronouns are more restricted than nouns in terms of the range of dependents they
allow, so the X-words are more restricted than transitive prepositions in this respect.
The proposed analysis has consequences for the class of words that contains
items such as although, because, whereas, while, whilst etc., traditionally
analysed as “subordinating conjunctions”. The main difference between these
and prepositions is that prepositions typically take nominal complements (and
certain types of clause complement in some cases), whereas a subordinating
conjunction is typically followed by a finite declarative clause.5 However, if we
regard this clause as a complement of the “conjunction”, there is a good case for
assimilating these words to the preposition class too, since it seems unsatisfactory
to base a major word class difference on a relatively minor difference of
complement type.6 Thus, a structure such as although he knew the answer could
be analysed as a PP, with the preposition although as head and the clause he
knew the answer as complement. In this respect prepositions would resemble
verbs in that they would subcategorize for such features as +[ __NP], +[__finite
declarative clause], +[ ——] with a significant number of items appearing in
different sub-classes.
3. Problems with the preposition analysis
Since the proposal for intransitive prepositions has such far-reaching implica-
tions, it clearly deserves close consideration. However, I will now argue that the
case for the analysis is less compelling than appears at first sight. For ease of
presentation, I take each of the arguments indicated above in turn, with the
exception of (h) which relates to a very small set of the X-words. First I give a
supporting argument or elaboration where appropriate. Then I outline some
objections in the form of a counterargument. Finally I give a response to the
counterargument where appropriate.
3.1 Argument from verb complements
Unlike (prototypical) adverbs the X-words can function as verb complements, the
verbs in question being precisely those that take PP complements.
This argument relates to verbs expressing locative or directional meanings: be,
come, go, lie, put, sit, stand, take and so on. The proposal, then, is to assign a
similar structural analysis to He went to the door, He went there and He went in.
138 DAVID LEE
3.1.1 Counterargument
In fact the relevant constraint has nothing to do with prepositions or PPs but with
locative meanings, which can be expressed by PPs, particles or adverbs (that is,
the X-words). In support of this counterargument, note that not all PPs can occur
in this position. (Those headed by prepositions such as except, since, with,
without and many others are disallowed). As further support, note that indisput-
able adverbs such as word-initially, word-medially and word-finally can occur as
the complement of verbs like go and put (e.g. You can’t put that morpheme word-
finally). Note too that put does not necessarily require a PP in second comple-
ment position — fused relatives can also occur there, as in He puts it where he
always puts it.7 In this example where cannot be analyzed as a preposition, since
its putative “complement” (he always puts it) is not a complete grammatical unit.8
3.1.2 Response
It is certainly true that not all PPs can occur in this position but this may simply
mean that the relevant constraint involves an intersection of syntactic and
semantic factors. The elements occurring in this position are required to be both
prepositional (PPs or intransitive prepositions) and locative. It is true that adverbs
such as word-initially etc. can occur after go and put. However, these patterns
are marginal — go does not have its normal meaning in an example like This
morpheme goes word-initially and with put the example is of doubtful acceptabil-
ity — so that the “cost” of dealing with these as exceptions is far outweighed by
the benefits of treating the X-words as prepositions. Furthermore, other proto-
typical adverbs expressing locative meanings such as locally, regionally,
nationally do not readily occur as the complements of the verbs in question,
which supports the view that the relevant constraint involves the intersection of
semantic and grammatical factors. As for the structure involving the fused
relative, it is difficult to decide whether this is a true counterexample without an
independently motivated analysis of the structure.9
3.2 Argument from post-modification of nouns
The X-words can post-modify nouns.10

This involves assigning the same grammatical structure to the man in the corner,
the man there and the batsmen out. In this case there are almost no restrictions on
PPs in this function. Virtually all prepositions can occur here (except those like
although which take finite clauses).
The post-head modifier position is not restricted to PPs. In particular some
indisputable adverbs can occur there:
(8) People locally/ regionally/ nationally have been protesting
(9) The news lately/ recently has been most disturbing
Any claim that these adverbs are clausal adjuncts is undermined by the following
cleft patterns, where the adverb is fronted along with the remainder of the NP:
(10) It was people locally who were protesting
(11) It is the news lately that has been most disturbing
3.2.2 Response
The analysis of the adverbs in (8) and (9) as post-head modifiers is clearly
problematic — as the rather marginal status of (10) and (11) shows. On the other
hand, there is no question that there and over there are post-head modifiers in
examples like the man there, the man over there. The absence of any semantic
constraint on prepositionals here is a particularly strong argument.
3.3 Argument from premodification of adjectives and adverbs
The X-words cannot premodify adjectives or adverbs.
This argument is based on the assumption that all adverbs can premodify
adjectives or other adverbs. However, this is true only of a subset of adverbs, i.e.
those expressing degree-type meanings. Many prototypical adverbs, including
words like carefully and slowly (not to mention locative adverbs like locally,
regionally) do not have this property. Even adverbs that do not express a degree-
type meaning in other contexts acquire such a meaning in this construction. For
example, surprisingly does not express the notion of degree in Surprisingly he
was late but does so in He was surprisingly late. In other words, the character of
adverbs that premodify adjectives or other adverbs must be such as to allow a
degree-type meaning to emerge. The X-words do not have this character.
Moreover, even if it were true that all prototypical adverbs had the relevant
property, whereas the X-words did not, the argument would not be fully convinc-
ing. It might simply mean that the X-words are not prototypical adverbs.
140 DAVID LEE
3.4 Argument from prepositional complements
The X-words can function as complements of prepositions.11

This argument assigns the same structural analysis to from under the bed on the
one hand and from there, from here, from home and so on on the other.
This argument assumes that adverbs cannot function as complements of preposi-
tions. In fact, this is untrue, as the following examples show:
(9) until recently, until later
(10) [I’ll keep it] for later
A further problem is that it may be the case that there, here and home are able
to function as complement of from and some other prepositions not because they
are prepositional but because they are nominal (a property which shows up in the
fact that they can also fill the subject function in certain constructions.) This is
confirmed by the fact that there, here, home show a greater collocational
potential than the relevant set of PPs in this construction. For example, they all
collocate with near, whereas under the bed does not. Similarly home collocates
with at, whereas no PPs do so.
Examples like up there, down here, over there cannot be invoked to bolster
this argument, since these are not examples of head-complement structures. In
He was standing over there for example, there is head, with over as dependent.
3.5 Argument from head of PP
The X-words take PPs as complements.

This argument assigns the same analysis to from under the bed, on the one hand,
and there in Britain, here in Australia, home to Mum, somewhere near Chipping
Sodbury and so on on the other.
In an example like here in Australia it is not clear that the PP in Australia is a
dependent of here. The strongest candidate for such an analysis is home to Mum,
since He went home to Mum seems structurally and semantically analogous to He
went home rather than to He went to Mum. On the other hand, there are no
grounds for saying that He lives here in Australia is closer to He lives here than
it is to He lives in Australia.
Moreover, even in the case of home to Mum the relation between home and
to Mum is unlike that between a head and a dependent in a number of ways. For
example, home can be fronted:
(14) Home went John to Mum
whereas this is not possible with prep-PP structures (*From he came under the
bed). In general, these constructions (home to Mum, there in England, here in
Australia) are more plausibly analyzed as appositional structures than as head-
complement structures.
As for somewhere near Chipping Sodbury, the somewhere seems to be a
dependent rather than a head, given the fact that the sentence is analogous to He
lives near Chipping Sodbury rather than to He lives somewhere. This structure
should therefore be analyzed as parallel to examples such as vaguely near
Chipping Sodbury or really near Chipping Sodbury; that is, as a PP with some-
where functioning as pre-head modifier and therefore analyzable as an adverb.
Note finally that PPs do in fact occur in some cases as the complements of
adverbs (for example Independently of these considerations).12
3.6 Argument from right and straight
The X-words take right and straight as pre-modifiers.
Again this is not a property of all prepositions, only those expressing locative
and directional meanings, so the relevant constraint may be semantic rather than
syntactic. Admittedly, words like locally, regionally, nationally do not allow
right or straight as premodifiers, even though they express locative meanings.
But this may be because they also contain elements of meaning relating to
manner, which are incompatible with right and straight. As far as straight is
concerned, the constraints on its distribution are rather tight. It does not co-occur
with all locative expressions (*It is straight inside the house), only with those
where there is some notion of path. Even in the case of right it could be argued
that there is a covert notion of path in the sense that right inside the house, for
example, designates the endpoint of a path that leads into the house. The non-
occurrence of right and straight on adverbs like locally could then be explained
in terms of the absence of the notion of path in their semantics.
142 DAVID LEE
3.6.2 Response
Again the relevant constraint may involve an interaction between semantic and
syntactic properties. The fact that locally, regionally, nationally etc. are not
prepositions is a more plausible explanation for the fact that they do not take
right and straight as premodifiers than the explanation offered above.
3.7 Argument from Locative Inversion
The X-words occur in the Locative Inversion construction (e.g. Into the room ran
John!).
The name of this construction again suggests that the relevant factors may be
semantic rather than syntactic. Moreover, although most examples of Locative
Inversion can be thought of as pragmatically-oriented reorderings of the corre-
sponding declarative structure (Into the room rushed John! ← John rushed into
the room), it is not plausible to think of examples such as There goes the bus!,
Here comes John!, Away went Mary! in these terms. These have acquired the
status of independent construction types.
Note too that certain adverbs can also trigger Locative Inversion:
(15) Next came John13
3.7.2 Response
Again there may be an interaction here between semantic and syntactic features.
Notice that it is not sufficient for an expression to express a locative or direc-
tional meaning for it to occur in this construction. Thus, whereas (16) is well-
formed, (17) is not:
(13) Under the bush was a rare flower
(14) *Locally was a lot of trouble
As for next it may be that this should be included in the X-class. It is obviously
not a prototypical adverb (it does not allow an adverbial pre-modifier, for example).
4. Descriptive options
The above arguments present us with at least three descriptive options. The first
option (option A) can be outlined as follows:
– The defining property of prepositions is that they take an obligatory nominal
complement.
– The X-words are adverbs.
– “Particles” are a sub-class of adverbs.
– Although etc. are “subordinators” and although he left early is a subordinate
clause.
This is close to the traditional analysis. There are a number of objections to it.
It fails to capture the fact that the X-words have many prepositional properties.
It treats the adverb category as one with minimal internal cohesion. It fails to
capture the fact that there is a major lexical overlap between particles and
prepositions and it does not explain why particles often have understood
complements. Moreover, if although is a subordinator like that (as in that he left
early), there is no way of accounting for the fact that clauses like although he left
early have a quite different distributional potential from subordinate clauses such
as that he left early. In particular, they function as adjuncts but not as subjects or
objects, whereas that he left early has precisely the converse properties. More-
over, it fails to account for the fact that words such as as, after, before, until are
both prepositions and subordinators.
One way of improving option A is to analyze although, because, while and
so on as prepositions — that is, to allow that prepositions may take either
nominal complements or finite clause complements or both. This solves the
problem with as, after, before, until as well as explaining the fact that structures
like although he left early have a different distributional range from true
subordinate clauses such as that he left early. That is, although he left early is
now analyzed as a PP containing a finite clause complement rather than as a
finite clause in its own right.
A further improvement on option A would be to analyze particles as
deprepositional adverbs. (This possibility was canvassed by Rodney Huddleston
at one stage in the CGE workshops.) This would account for the extensive
lexical overlap with prepositions and would also help to explain the fact that they
have acquired specialized meanings in many cases (as in many of the phrasal
verbs: give up, give in, take up, find out and so on). I will call this modified
analysis Option B.
144 DAVID LEE
Option B does not solve the problem with the X-words. In order to address
this problem, we have to move to the prepositional analysis, where the preposi-
tion class is extended significantly to include intransitive prepositions. The
objections to this move are outlined in the counterarguments given in the body
of this paper. They consist mainly in the fact that this analysis does not capture
the fact that the X-words also have an affinity with at least some members of
the adverb class — specifically the locative adverbs. Note too that all these
difficulties would remain if a fourth option were canvassed — namely that a
special category be established for the X-words (temporal-locatives, say). The
real problem is that these words hover uncertainly between adverbs and preposi-
tions. This in fact suggests yet another option — a supercategory to cover both.
It is difficult to imagine however that there would be widespread support for this
option, given the fact that the words located at one extreme of such a continuum
have nothing in common with those at the other.14
This discussion has now led us to a stalemate. Given the framework within
which this discussion is located, there is no fully satisfactory way of solving this
descriptive problem. Certainly there would be few linguists who would endorse
option A, but it is difficult to imagine any consensus emerging between options
B and C, precisely because the correlational patterns are not coherent enough to
provide a solution. But if this is so, the obvious question is: how is it that native
speakers are able to process sentences containing the X-class words, if sentence
structures are projections from categories unambigously marked in the lexicon?
5. An emergence theory
The solution to this problem seems to lie in an emergence theory of category

structure of the kind described by Hopper (1987), Langacker (1990: 261–88). It
is beyond the scope of this paper to explore in detail how such a theory might
deal with the specific question of the X-class words. Nevertheless the outlines of
such an approach should be sketched.
An emergence theory shares with traditional models of category structure
the notion that categories are constructed on the basis of shared properties. To
this extent there is no significant difference between the way in which the
category of English nouns, for example, is conceptualized in each approach. As
noted above, nouns are characterized by an extensive set of shared morpho-
syntactic properties which converge on a specific set of English words and which
therefore give that category strong internal cohesion.

However, the two approaches diverge with respect to the question of how
far we can expect this kind of cohesion to apply to all linguistic categories
(focusing here specifically on lexical categories). Emergence theory predicts that
the process of schematization (Langacker 1990: 103–4) that applies in category
construction will operate at different levels of generalization. The category of
nouns in English is highly schematic, since it ranges over a large set of proper-
ties. There is no expectation in emergence theory, however, that all schematiza-
tion will operate to this degree of abstraction. In some cases, any schemas that
are constructed on specific groups of lexical items may remain close to the
lexical level and range over only a narrow range of properties. In this sense there
is no necessary expectation that all the words in a given language will necessari-
ly slot unambiguously into a restricted set of lexical categories. Moreover, we
might expect a good deal of variation between extremely tight correlations across
properties in certain areas and very loose correlations across others. Tight
correlations across an extensive feature set will facilitate a high level of schema-
tization. Loose correlations will confine schematization to much lower levels.
This latter situation is precisely what we find in the preposition/adverb area.
The traditional category of preposition is defined by a relatively restricted set of
properties (the ability to take a nominal complement being the most salient one).
The same is true of adverbs, their most salient property being their morphologi-
cal character. As soon as we attempt to move beyond these properties, the
correlational patterns cut across the two categories in a confusing manner making
higher level schematizations problematic.
One further characteristic of this model needs to be mentioned here. Distinct
schemas from different domains can be applied simultaneously to the same data
set, without necessarily privileging one particular type of schema. (By contrast,
the projection of syntactic structures from lexical categories in generative
grammar privileges syntactic schemas.) For example, from sentences like (15)
and (16) (and many similar sentences) it is possible to construct several distinct
schemas in parallel: [Verb + PP] on the category dimension (assuming a narrow
characterization of PP along the traditional lines), [Predicate + Complement] on
the functional dimension and [Process + Locative] on the semantic dimension.
(15) He went to the city
(16) She lives in the country
146 DAVID LEE
The [Predicate + Complement] and [Process + Locative] schemas can also be

constructed on sentences like (17) and (18) but the [Verb + PP] schema cannot.
(17) He went there
(18) He lives here
Thus we can capture some of the similarities between the X-words and preposi-
tions on the one hand and adverbs on the other without claiming that these words
are in fact prepositions or adverbs.
This is clearly only the beginning of a solution to the problems outlined in
this paper. The full content of the schemas that constitute prepositions, adverbs,
locatives needs to be elaborated together with an analysis of the relationships
between them, taking account of the arguments given here. No doubt a whole
range of schemas of various levels of abstraction will be needed to handle the
very complex cross-cutting patterns observed here. The cost of this model — by
comparison with the generative model — is that it assumes that the degree of
schema construction by speakers of a language is vastly more intricate and
complex than is envisaged in a model that operates with a small number of
schemas in the form of lexical categories. However, this level of complexity
seems to be required by the data. There is no reason for making the a priori
assumption that linguistic data are necessarily amenable to minimalist theories.
Notes
1. Even this observation does not apply to all X-words — for example away and home.
2. This argument applies primarily to the locative items in the set of X-class words. Thus,
although where is a complement in Where did he go?, when is not a complement in When did
he go? (though it is arguably one in When did he live?).
3. Again this argument applies to only a subset of the X-class words (but to a different subset
from that picked out by the first argument). It applies to there, here, somewhere, aboard,
abroad, now, then but hardly to where, when and home (*people where, *somebody when,
*children home are not well-formed NPs).
4. Burton-Roberts (1992: 169) calls this construction “Exclamatory Fronting”.
5. Huddleston (1984: 340) notes that it is insufficient to simply identify “clause” as the relevant
subcategorisation feature, since words traditionally analysed as prepositions take clause
complements of certain types. For example, about and of take interrogative clauses as
complements (He was worrying about who he should trust, He raised the question of why it had
been concealed) and the prepositions on and to can take non-finite clauses as complements (He
insisted on leaving early, He objected to being forced to vote).

6. As Huddleston notes, nobody would argue that a word class boundary should be established
within the verb class based on different complementation patterns — clause vs PP, for
example.
7. I owe this example to Rodney Huddleston.
8. Rodney Huddleston, personal communication.
9. For example, if we analyse He put it where he always puts it as a fusion of He put it at x and He
always puts it at x, then it is not a counteraxample.
10. Again this argument picks out only a subset of the X-class words (though a rather different
subset from those affected by the first argument). It applies to there, here, somewhere, aboard,
abroad, home, when, now, then but hardly to where, when, home (?People where/ when/ home
have agreed … )
11. Note, however, that this is not a position in which particles occur naturally.
12. Rodney Huddleston, personal communication.
13. As far as now is concerned, it seems more plausible to analyse an example like Now comes the
difficult part as being structurally closer to (15) than to an example like Into the room marched
John.
14. However, this option might well be canvassed by Lakoff, given his notion of radial categories
(Lakoff 1987), which predicts precisely this situation.
References
Burton-Roberts, N. 1992. “Prepositions, Adverbs and Adverbials”. Language

Usage and Description ed. by I. Tieken-Boon van Ostade and J. Frankis,
159–172. Amsterdam: Rodopi.
Dixon, R.M.W. 1982. “The Grammar of English Phrasal Verbs”. Australian
Journal of Linguistics 2. 1–41.
Emonds, J. 1972. “Evidence that Indirect Object Movement is a Structure-
Preserving Rule”. Foundations of Language 8. 546–61.
Hopper, P. 1987. “Emergent Grammar”. Proceedings of the Thirteenth Annual
Meeting of the Berkeley Linguistics Society.
Jackendoff, R. 1973. “The Base Rules for Prepositional Phrases”. A Festschrift
for Morris Halle ed. by S.R. Anderson and P. Kiparsky, 345–356. New
York: Holt, Rinehart and Winston.
Lakoff, G. 1987. Women, Fine and Dangerous Things: What Categories Reveal
About The Mind. Chicago and London: University of Chicago Press.
Langacker, R.W. 1990. Concept, Image and Symbol. Berlin and New York:
Mouton de Gruyter.
148 DAVID LEE
Sentences, clauses, statements and propositions
John Lyons
1. Introduction
The topics with which I am concerned in this article are topics that I have dealt
with in various publications over a number of years (including, Lyons 1968,
1969, 1977a, 1977b, 1991b, 1995a). In the present context, I will be formulating
what I have to say somewhat differently, and I will be making one or two points
that I have not made in print before.1 My general purpose is to bring together
(and at times to update) what is otherwise dispersed throughout several books
and articles and, in doing so, to clarify or modify points made in previous
publications.2 Although much of the article deals with grammatical structure, my
primary concern throughout will be the semantic (and/or pragmatic) justification
for postulating the linguistic and metalinguistic entities whose descriptive value
for English, though not necessarily for all languages, I wish to establish. I will
take for granted throughout an understanding of my terminological distinction
between forms (whether as types or tokens) and expressions (to which the type-
token distinction does not apply, on the one hand, and between system-units
(system-sentences, system-clauses, system-phrases, and so on) and text-units
(text-sentences, text-clauses, text-phrases, and so on), on the other.3,4
2. Sentences and clauses
The validity of the distinction between sentences and clauses, as far as English
(and many other languages) is concerned, is generally accepted by descriptive
grammarians: the arguments are set forth briefly, but cogently, in Huddleston
(1984: 18–21; see also Matthews 1981: 29–38). There are two points that can be
150 JOHN LYONS
usefully made here. One has to do with the generality of the distinction; another
with the question whether one of the two, the sentence or the clause, is grammat-
ically and/or semantically less basic than the other.
Let us begin, however, by establishing the distinction as it is traditionally
drawn. What we commonly refer to as “(Western) traditional grammar” is, of
course, far less uniform than many authors (including Lyons 1968) would seem
to imply. But the following two propositions are commonly included in (or are
implied by) what is said about sentences and clauses in those versions of
traditional grammar that do in fact draw the distinction:
(1) The sentence is the maximal unit of grammatical analysis.
(2) Clauses are composed [typically] of a subject and a predicate.
Taken together, (1) and (2) will serve adequately to focus our attention on the
two points that are important for the present purpose.4
The first point is that neither unit is defined here in terms of the other. The
situation, in this respect, is comparable with that of words and morphemes as
defined by Bloomfield (1933). If morphemes and words are by definition
minimal forms and minimal free forms, respectively, it is logically possible for
a language to have both kinds of units and for them to coincide: that is, for all
morphemes to be words and, conversely, for all words to be morphemes. Indeed,
this is exactly what we would find in a maximally isolating (or analytic)
language. This theoretical ideal may not be actualized in any known natural
language, but it is generally accepted that there are languages, generally classi-
fied typologically as isolating, notably Vietnamese and Classical Chinese, which
approximate to it.
As with morphemes and words, so with clauses and sentences: it could well
turn out that in particular languages, characterized as they are in (1) and (2), they
coincide. Indeed, it has been argued recently, on the basis of data from English,
Russian and German: (a) that “the clause is [the basic unit of syntactic analysis
in that it is] the essential locus of both dependency relations and distributional
properties” and (b) that “[t]here is very little evidence to support either text-
sentences or system-sentences in spontaneous spoken language” (Miller 1995)5.
If clauses are the maximal units of grammatical analysis in a particular language,
then in that language, in terms of (1) and (2), all clauses are sentences, and converse-
ly. Miller, it will be noted, explicitly distinguishes, as I do, between sentences as
units of text (or discourse) and sentences as units of the language-system.
It is of course clausal hypotaxis, and more especially the embedding of
SENTENCES, CLAUSES, STATEMENTS AND PROPOSITIONS 151
subordinate clauses as subjects or objects (rather than more peripherally as

adjuncts) in superordinate clauses, that provides grammarians with the principal
reason for establishing a distinction between simple and complex sentences and,
therefore, between sentences and clauses. Many languages, however, do not seem
to have embedded clauses of this kind, and therefore, arguably, do not have
complex sentences. As for compound sentences, formed by the conjoining or
paratactic co-ordination of clauses, these have always been suspect as a syntactic-
ally justifiable subcategory of what may be referred to more generally as
composite (that is, non-simple) sentences. For many languages, including
English, the principal criterion used by grammarians for deciding whether a
stretch of (spoken) text consists of a single compound sentence or a sequence of
simple sentences is that of intonation and rhythm. And it is at least arguable that
intonation and rhythm are not part of the structure of system-sentences, but, when
relevant, like punctuation for written text, of text-sentences (and text-phrases, text-
clauses, and so forth).6 It is interesting to observe at this juncture — and I will
come back to this — that, as far as relating grammatical structure to proposition-
al structure is concerned, conjoining is the sole means of forming composite
propositions in standard first-order propositional logic (Lyons 1995a: 157–158).
Let us turn now to the question whether sentences are more basic than
clauses or conversely. Another proposition that is commonly included in what is
said about clauses in the English school-grammatical tradition can be extracted
from the following definition:
(3) “A clause is a sentence which is part of a larger sentence.”
(Nesfield 1939: 3)
This definition is not of course in conflict with (2). Indeed, it is commonly
combined with the equivalent of (2) in many traditional accounts. But, unlike (2),
which in itself is neutral in this respect, it definitely makes sentences more basic
than clauses. Like most definitions of traditional grammar, (3) needs to be
adjusted in various ways for consistency and generality; but it conveys the
essence of one traditional way of relating sentences and clauses (and, of course,
in doing so, of making clauses theoretically and descriptively redundant).
If we drop “larger” in (3) and allow (as any logician will) that X can be
part of Y without Y being larger than X (or having any residue that is not part
of X), we can talk, as one does traditionally, of one-clause sentences as well as
of multi-clause sentences. We can, in short, draw the traditional distinction
between simple and non-simple (or composite) sentences; and we can go on to
152 JOHN LYONS
divide composite sentences (in terms of the distinction between coordination and
subordination) into compound and complex sentences. This is in effect what was
done in the earliest versions of Chomskyan transformational-generative grammar,
in which the content of (2), in contrast with that of (3), was at best formalized
only incidentally and derivatively. It is the approach that I myself followed in
Lyons (1968).
Despite what I said or implied in such works as Lyons (1968, 1977a), I now
take the view that clauses, rather than sentences, are the basic units of syntax
(and especially of government and dependency) and that (if one retains the
traditional concept of the sentence) sentences are best defined in terms of
clauses, rather than conversely (Lyons 1995b: 235–237).7 I should emphasize,
however, that nothing of real consequence follows from my failure to give to
clauses the status of primary syntactic units that I should have given them in my
earlier writings. As I have made clear, most recently in Lyons (1995a), much
(though not all) of what I have said about the meaning of sentences can be
reformulated, where appropriate, with reference to clauses.8
I will not go further into the distinction between sentences and clauses. As
I have already mentioned, clauses correspond more closely to simple propositions
than do simple — that is, non-composite — sentences and may also have more
psycholinguistic validity (in both competence and performance). All that needs
to be said now is that, from a more general point of view and more especially
for semantic purposes, one needs to generalize that part of the definition implied
in (2) which restricts the clause to languages in which the subject-predicate
distinction is grammatically justifiable and to reformulate it in terms of predica-
tors of variable valency (prototypically verbs) and their arguments. But this is
straightforward enough and nowadays would, I think, be readily accepted by
most grammarians.
3. Sentences (and clauses) and statements
As units of the language-system, sentences — and more particularly simple

sentences — may fall into a number of grammatically distinct subclasses, the
utterance of each of which can be associated with a set of one or more character-
istic illocutionary functions. For example, one subclass of (simple) sentences in
a particular language may be associated with questions and will, for that reason,
be given the label “interrogative”; another may be associated with statements —
that is, with saying that something is or is not so — and will be given the label
“declarative”; and so on.
The terms “question”, “statement”, and the like, are all being used, it will
be noted, to refer to functionally defined subclasses of utterances; “interrogative”,
“declarative”, etc., are being used, in contrast, to refer to grammatically defined
subclasses of sentences (or, as we shall see presently, clauses).9 Statements, then,
are but one subclass of utterances — constative utterances — for which particu-
lar natural languages may or may not provide a grammatically distinct subclass
of sentences.
The term “statement” is subject to exactly the same by now well known
process/product ambiguity as “utterance” (Lyons 1995a: 18, 35). It is statements
as the products of the activity of utterance with which we are primarily con-
cerned in the present context; that is, with utterances as the products of perfor-
mance (to employ the now widely used Chomskyan terminology). In all that
follows, therefore, “statement” will be used exclusively in its product sense; that
is, to refer to a functionally definable subclass of utterance-inscriptions. The term
“sentence” in contrast, will be used exclusively to refer to units of the language-
system. Statements, thus defined, are entities that belong to ontologically
different categories from sentences and clauses and, therefore, cannot in
principle be defined as a subclass of the latter. Regrettably, this fact is obscured
in the school-grammatical tradition by its failure to distinguish sentences from
utterances and its consequent use of “statement”, “question”, “command”, and so
on, for subclasses of sentences.
Statements need not, and in my view should not, be assumed to be logically
or ontologically more primitive than — or to have any other kind of privileged
status with respect to — other kinds of illocutionary acts, such as questions,
commands, promises, exhortations, proposals, and the like. Whether Austin
(1962) or his followers were on the right track in their elaboration of the details
of what has come to be known (unfortunately) as speech-act theory is, for present
purposes, irrelevant.10 Austin was unquestionably correct in his criticism of what he
called the descriptive fallacy: the view that describing the world, or states of
affairs, by making statements is the sole or the primary function of language.
Associated with the descriptive fallacy, but logically independent of it, is the
common assumption, that declarative sentences (and clauses) are necessarily, and
in all languages, more basic than non-declarative sentences (and clauses).
Obviously, the term “basic” has to be explicated before we can accept or reject
this assumption; and there are several ways in which it can be explicated within
154 JOHN LYONS
this or that theory of grammatical structure. But it is often pre-theoretically clear

in the description of particular languages that, given two syntagms X and Y, X
is more appropriately derivable from Y than Y is from X; that is to say that,
independently of any other sense that might be assigned to “basic” and regardless
of the way in which this sense of “basic” is explicated in any particular theory
of grammar, X is therefore descriptively more basic than Y. It may be readily
conceded that in English, for example, declarative sentences are descriptively
more basic than corresponding interrogative sentences (though not necessarily
more basic than all other subclasses of non-declarative sentences).11 But this is
a matter of empirical discovery. Moreover, the fact that in English (as in many,
but not all, natural languages) declarative sentences are descriptively more basic
than interrogative sentences is, from the viewpoint of general linguistic theory,
contingent, rather than necessary. It is easy enough to construct pragmatically
and sociolinguistically plausible versions of Quasi-English in which declarative
sentences are derivable from interrogative sentences.12 It is commonly assumed,
and sometimes asserted by philosophers of languages, that a language in which
one cannot make statements, in contrast with a language in which one cannot ask
questions or issue commands and requests, is inconceivable. De imaginatione non
est disputandum! I for one have no difficulty in imagining indefinitely many
pragmatically plausible, admittedly non-natural (but certainly not unnatural),
languages which do not have declarative sentences and which, moreover, do not
provide the means for making (except indirectly and by implicature) straightfor-
ward statements. There are, after all, many natural languages that have neither
declarative nor interrogative sentences, but sentences which can be uttered either
to make statements or to ask questions according to the context in which they are
uttered and/or the prosodic and paralinguistic modulation that is associated with
them in (spoken) utterance. Obvious examples, as far as yes-no questions and
corresponding statements are concerned, are Spanish, Italian and Modern Greek
(Lyons 1995a: 186).
I must now draw the distinction between what I referred to above, loosely,
as straightforward statements and other kinds of statements. By a straightforward
statement, in the context of the present discussion, I mean what logicians
frequently refer to as a categorical assertion (“assertion” being synonymous with
“statement” in the usage of certain logicians), but which may be called, more
revealingly for the purpose of linguistic semantics, epistemically, evidentially and
attitudinally neutral (or unqualified) statements. Such statements are utterances
with which speakers (more precisely illocutionary agents operating in whatever
medium) say that something is or is not so, without concomitantly (in the
utterance itself) qualifying their commitment to the truth of what they are saying
or revealing the grounds, or evidence, for their (epistemic) commitment and
without expressing (once again, in the utterance itself) any emotional attitude
(surprise, regret, pleasure, and the like) towards whatever it is that they have said
to be or not to be so.
Epistemically, evidentially and attitudinally neutral statements (which I will
refer to henceforth collectively as neutral statements) are of relatively rare
occurrence in everyday colloquial discourse, even in languages (such as English)
which do provide their users with the means of making them. Even if there is no
grammatical or lexical indication of the speaker’s epistemic commitment (or
warrant) or attitude, there will usually be some such prosodic or paralinguistic
indication. This fact is well known to linguists and has been discussed and
exemplified for English in several readily accessible, non-technical, accounts of
intonation, rhythm, tempo, pitch range, laryngeal and labial adjustments, and so
on. It requires neither emphasis nor illustration here. But English does at least
allow its users to make neutral statements in speech (with what is referred to as
neutral, or unmarked, intonation); and in non-colloquial written texts they are of
course quite common.
There are, however, many natural languages in which it is impossible to
make neutral statements either in speech or writing. Of particular interest are
languages in which every declarative sentence must be in one of several (seman-
tically non-equivalent) non-indicative moods expressing a particular kind of
epistemic commitment or warrant (Palmer 1986). The fact that such phenomena
as epistemic commitment are grammaticalized in a wide range of the world’s
languages makes it essential for the linguist to give non-neutral, as well as
neutral, statements their rightful place in semantics. As we shall see, it also
reinforces the need to distinguish statements, not only from sentences, but also
from propositions.
Not only is the difference between sentences (or clauses) and statements
frequently ignored, so too is the terminological distinction between “declarative”
and “’indicative”, especially by philosophers and logicians, who commonly use
“indicative” with the sense that linguists writing in English normally assign,
nowadays, to “declarative”. A declarative sentence, as we have seen, is one that
belongs to a grammatically definable subclass whose members are used, charac-
teristically (but not necessarily in the case of each member on all — or indeed
any — occasions of its use) to make (neutral or non-neutral) statements.
156 JOHN LYONS
The term “indicative”, in contrast with “declarative”, applies only secondari-

ly and derivatively to sentences (and clauses). Like “subjunctive”, “imperative”,
“optative”, and similar terms inherited from traditional grammar, it applies
primarily to one of a set of grammatical moods which may or may not be
manifest in particular languages: an indicative sentence (or clause) is a sentence
(or clause) in the indicative mood. As we have noted already, in many languages
(though not in English) there are non-indicative declarative sentences. Converse-
ly, in many languages (including English) there are non-declarative (for example,
interrogative) indicative sentences. Mood, which is by no means universal in the
languages of the world, is the category which results from the grammaticalization
of (subjective) modality; and the indicative (in those languages that have such a
mood) is by definition the epistemically neutral such mood (Lyons 1995a:
320–335).13 In complex sentences (in languages with mood that have such a
subcategory of system-sentences), each clause will have its own mood (with
more or less strict rules of compatibility between the mood of the main clause
and that of the subordinate clause or clauses); and it will be the mood of the
main clause that signals, of itself or together with other lexical or grammatical
features, the characteristic illocutionary function of particular sentences.
Throughout this section I have talked mainly, except in the preceding
paragraph, about sentences, and more especially simple sentences, rather than
about clauses. As we saw in the previous section, for languages without compos-
ite sentences (that is, languages in the description of which linguists do not see
the need to postulate the existence of non-simple, complex or compound,
sentences) it makes little difference whether one associates such grammatical
properties as declarativity and interrogativity with clauses or with sentences. As
for languages that also have either complex or compound sentences, or both, it
may well be the case that (as for mood and perhaps also for tense) they should
be associated primarily with clauses and only secondarily with sentences (and
then perhaps only for complex sentences).14 To say this is, of course, to concede
that, like propositional structure, they are (universally?) clausal, rather than
sentential, properties. For the present, I would prefer to leave the question open.
4. Propositions
We have now distinguished statements, not only from sentences, but also from
that subclass of sentences (and/or clauses) to which the label “declarative” is
assigned (in languages that have such a grammatically definable subclass) by

virtue of the characteristic association of typical members with the (illocutionary)
activity of making statements; that is, of saying that something is or is not so.
Let us now turn to propositions.
The status of propositions has been as problematical in twentieth-century
philosophy as has been that of sentences in twentieth-century linguistics. Both
disciplines were strongly influenced in the 1930s by empiricism of one kind or
another. Several schools of modern linguistics (and most notably that of Bloom-
field and his followers in the United States) avoided using the term “sentence”
altogether or, alternatively, defined sentences as a subclass of utterances.
Similarly, many twentieth-century philosophers of language of an empiricist bent
of mind have anathematized the term “proposition” because of its (to their mind)
undesirable metaphysical or psychological connotations; and (when writing in
English) they have preferred to employ either “sentence” or “statement”.15 Not
surprisingly, it is not always clear (to a non-philosopher at least) when an
apparent difference of opinion expressed by two philosophers of language about
the ontological or theoretical status of propositions is genuine and substantive or
is merely a matter of terminological preference.
Much of the heat has now gone out of past controversies. But the status of
propositions in relation to sentences and statements is still both terminologically
and substantively confused. It is not part of my purpose to attempt to contribute
to the clarification of any of the purely philosophical issues. In this context, I
will simply adopt a recognizable, more or less standard (if not wholly uncontro-
versial), view of the matter and exploit it for the more restricted purposes of
linguistic semantics. Whatever might be the situation in logical semantics and in
the analysis of non-natural, formal, languages, in my view there is no doubt that
in the semantic analysis of (so-called) natural languages the postulation of
propositions, in addition to both sentences (and/or clauses) and statements, is
descriptively justifiable.
Some or all of the following properties are commonly ascribed to proposi-
tions by such philosophers as postulate their existence (or descriptive or explana-
tory utility as theoretical constructs):
(i) they are either true or false: that is, they are the bearers of truth-values;
(ii) they can be the objects of so-called propositional attitudes: that is, they can
be known, believed, doubted or entertained;
(iii) they can be the objects of illocutionary acts: that is, they can be asserted (or
stated), denied, queried, and so on;
158 JOHN LYONS
(iv) they are (or can be) held constant under translation from one language
(whether natural or non-natural) to another.
Not all of these properties are ascribed equally to propositions by all philoso-
phers, and each of them merits a word or two of comment.
The first, the property of being the bearer of a truth-value, is unquestionably
the most essential of the four that I have listed here. There are however various
ways in which the truth and falsity of propositions can be explicated and
formalized. I will take the standard, classical, view: that propositions are bivalent
(that is, two-valued) and eternally constant with respect to their truth-value. That
is to say, I will assume: (a) that every proposition is either true or false, there
being no other truth-value (such as “indeterminate”); and (b) that a proposition
cannot be true at one time and false at another. It is very important, in this
connection, to draw a clear distinction between the truth-value of a proposition
(its being, eternally, true or false) and the truth-conditions of the sentence (or
sentences) that may be uttered to express it. It is also very important not to
confuse epistemological with (what might be referred to as) ontological indeter-
minacy. The fact that we may never be able, even in principle, to discover
whether a given proposition is true or false does not mean that its truth-value is
intrinsically indeterminate.16
The second and third of the four properties generally ascribed to proposi-
tions that I have listed above are closely connected. This is reflected in the fact
that in many languages the same constructions are used for both kinds of
propositions: (tensed) that-clauses in English, (untensed) accusative-and-infinitive
clauses in Latin, and so on. This does not mean, however, that propositions as
the objects of epistemic attitudes (and other kinds of mental states and activity)
can be identified ontologically with propositions, without difficulty, as the
objects of assertion, denial and other kinds of illocutionary activity. For various
reasons, there seem to be logical and semantic differences between them; and
there are languages in which the constructions used for referring to the one class
differ, in a semantically relevant way, from the constructions used for referring
to the other.
I will not go into this question in the present article. Let us simply note two
points: (i) that (in some languages at least) there are similar problems that arise,
in both cases, with respect to the interpretation of what I will call ego-referring
propositions; that is, propositions which refer to the epistemic (judgemental, and
the like) agent, on the one hand, or the illocutionary agent, on the other; (ii) the
grammatical structure of English (and a number of other languages) is such that
(using English) we can (generally) treat corresponding mental and illocutionary

propositions as identical. Some of the problems I am alluding to under (i) —
notably the problems of various kinds of referential opacity and de re/de dicto
ambiguity (one or two of which I will discuss briefly below) — are by now
familiar enough to linguists, as well as to philosophers. As to (ii), it is supported
grammatically by the fact that in English one can construct such (apparently
well-formed) sentences as:
(4) The Count of Paris thinks and says that he is the rightful heir to
the throne of France.
(5) The Count of Paris thinks that he is the rightful heir to the throne
of France and says so.
Rather more questionable as system-sentences of English are:
(6) The Count of Paris thinks /says that he is the rightful heir to the
throne of France and it is true.
And even:
(7) The Count of Paris (thinks) /says that he is the rightful heir to the
throne of France because it is true.
Or:
(8) The Count of Paris thinks /says — and it is true — that he is the
rightful heir to the throne of France.
The it in (8), in the second of the two conjoined co-ordinate clauses (in particular
utterances of this sentence; that is, in particular utterance-tokens) can be inter-
preted as being co-referential with the that-clause of the first of the two con-
joined clauses. They are questionable, of course, because the undoubted accept-
ability of the corresponding text-sentences, and their production and interpretabil-
ity, can be handled in terms of the (pragmatic) text-structure, or discourse-
structure, process of what I have elsewhere referred to as “contextualization”
(Lyons 1977a: 588 ff). This is in the present context irrelevant, except in so far
as it supports the more general point made earlier, that the class of composite
sentences (that is, system-sentences) in English (and many other typologically
comparable languages) may be smaller than most grammarians take it to be. The
acceptability (in terms of textual cohesion) and apparent comprehensibility of the
corresponding text-sentences is sufficient to support (though it does not of
160 JOHN LYONS
course prove) the assumption that propositions as the abstract and eternal bearers
of truth-values are identifiable (at least by the users of languages such as
English) with propositions as the objects of mental states and illocutionary acts.
So far, so good. Let us now couple with the assumption that we have just
made the further assumption that two different persons can both know (believe,
doubt, and so on) and assert (deny, query, and so on) the same proposition. Once
again, this assumption is supported by the possibility of constructing (in English)
such (apparently well-formed) sentences as (9) and of assigning to the conjoined
complex propositions that are expressed, on particular occasions of their utterance,
what seem to be logically coherent and empirically verifiable interpretations.
(9) John thinks/says that the Count of Paris is the rightful heir to the
throne of France and so does Peter.
But what then of such sentences as (10), where the subordinate clause in the first
conjunct is construed (on particular occasions of utterance) as ego-referring (with
respect to the referent of the verb of mental state or illocutionary activity)?
(10) John thinks/says that he is the rightful heir to the throne of France
and so does Peter.
Here the first conjunct does not seem, at first sight, to be (relevantly) ambiguous.
As soon as we set about assigning an interpretation to the second conjunct,
however, we see that it can be understood in two different ways; that is, as
expressing either of the following two semantically (truth-functionally) non-
equivalent complex propositions:
(11) “Peteri thinks (says) that Johnj is the rightful heir to the throne of
France.”
(12) “Peteri thinks (says) that hei is the rightful heir to the throne of
France.”
And yet the rules that we generally apply in the interpretation of the English do
so construction (and seem to be applying here) require us to say that the so is
anaphoric and refers (implicitly) to the same proposition as is referred to by the
antecedent that-clause. This is one of the linguistically (as well as philosophical-
ly) interesting problems, or antinomies, that arise in respect of ego-referring
propositions. It is one that has been much discussed in the literature both by
linguists and by logicians.
Many logicians have dealt with it in terms of the notion of referential
opacity and/or an updated version of the scholastic distinction between exten-

sional (de re) and intensional (de dicto) reference. This is a distinction which has
also been applied to the resolution of a wide range of other more or less
comparable phenomena. There is as yet no consensus about the validity of the
distinction itself (as it has been formalized in this or that system of modern
intensional logic) or about the identity or uniformity of the phenomena that it has
been held to cover. I have referred to it here, not because I wish to declare any
particular view in relation to its philosophical validity, but because the phenome-
na with respect to which it is currently invoked, both by philosophers and by
linguists, supply especially cogent evidence for the distinction between sentences
(or clauses) and propositions. They also make it clear that the notion of proposi-
tional identity is not always as straightforward as it might appear to be at first sight.
Comparable problems, and others, arise in respect of the fourth property
commonly (though not of course universally) ascribed to propositions: constancy
under translation from one language to another. It seems to me that we not only
can, but must, accept that there are certain propositions which can be expressed
in one (natural or non-natural) language X, but which cannot be expressed in
some other language Y that differs significantly in its lexical or grammatical
structure from X. I will return to this point in the final section.
That propositions cannot be identified either with statements (as we have
defined them) or with the sentences of natural languages is now obvious enough
and requires no further argument. Propositions considered as the objects of
illocutionary acts are components of such acts; and it also seems reasonable to
say that they are components of the meaning of the utterances that are the
products of such acts, including not only statements, but also questions, com-
mands, requests, promises, and so on. As for the relation between propositions
and sentences: that this is not one of identity follows immediately from the fact
that the same sentence can be used to express different propositions and that
some, though not all, propositions are expressible in different natural languages.
Are propositions to be identified, then, if not with sentences or clauses, with the
meanings of sentences or clauses, which, in so far as they are interlingually
translatable (and intralingually paraphrasable), are held constant under translation
(and paraphrased — salva veritate — within a single language)? This is one of
the questions to be addressed (and answered briskly in the negative) in the
following section.
162 JOHN LYONS
5. Sentence-meaning and propositional content
Since the publication of Frege’s seminal paper on what is conventionally

translated into English as “Sense and reference” (1892), there has developed an
important tradition in logical semantics within which propositions are identified
not with sentences, but with their meaning, their (Fregean) sense (“Sinn”), or
following Carnap (1942, 1947), their intension.17 Within this tradition, whether
it is associated with Montague’s (1974) explication of the intension of a sentence
as its extension in all possible worlds or with some other extensionalist or non-
extensionalist version of formal semantics, it is customary nowadays to say that
sentences (or clauses) express propositions.
This way of talking is perhaps acceptable enough in itself, and it has the
merit of brevity and simplicity. In this respect it is comparable with saying that
certain expressions of natural languages are referring expressions, rather than that
they can be used, by locutionary agents, to refer to what locutionary agents pick
out as referents (on particular occasions of utterance). That particular sentences
(or clauses) of natural languages do not express particular propositions (indepen-
dently of their context of utterance) follows immediately from the fact that, on
any standard definition of the sentence (or the clause) as a unit of the language-
system: (i) on the one hand, the same sentence (or clause) can generally be
uttered to express, and in so doing to assert, deny, query, and so on) indefinitely
many different propositions, some of which may be true and others false; and
(ii) on the other, the same proposition can be expressed by different sentences
either (a) of the same language or (b) of different languages. Hence the impor-
tance of drawing the now common distinction between the truth-value of a
proposition and the truth-conditions of a sentence (Lyons 1995a: 146–149).
It would be quite possible, in principle, to relativize the meaning of a
sentence to its context of utterance (or its so-called point of reference); and some
logicians have followed this approach. But there is no reason for linguists to
follow them when it is possible to handle the phenomena instead by invoking the
distinction between sentence-meaning and utterance-meaning. There is nothing to
be said in favour of adopting an eccentric and untraditional definition of
sentences, whose only purpose, it would seem, is to make it possible to say that
the meaning of a sentence is the proposition that it expresses.
What then is the relation between propositions and sentences? This is best
explained, I think, by introducing the term “propositional content” and by
defining it in a particular way. (The term is common enough, but its relation to
“proposition” is not always made clear.) The propositional content of a sentence,

we shall say, is that part of its meaning which is truth-conditional; that is,
sentences will have the same propositional content if and only if they have the
same truth-conditions (Lyons 1995a: 147).
This has the effect, it will be noted, of making the propositional content of
a sentence comparable, from the viewpoint of formal logic, with a propositional
function; that is, with an expression which can be converted into one of a set of
propositions by fixing the value of its unbound or open (restricted or unrestrict-
ed) variables.18 Similarly, the context-independent propositional content of a
sentence is converted, as it were, into one of a set of propositions, in the act of
utterance, by fixing the reference of its potentially-referring (or referential)
expressions (and of such features of its grammatical structure as, most notably,
tense) in the context of utterance and making of them referring (that is, actually-
referring) expressions.
That propositional content is a proper part, but not the whole, of sentence-
meaning would be denied by many formal semanticists. For some of these, all
meaning (properly so called) is by metatheoretical definition truth-functional, and
all sentence-meaning is, in consequence and a fortiori, truth-conditional. For
others, so-called pragmatic meaning (which may or may not be identified with
utterance-meaning) is not necessarily truth-conditional, though sentence-meaning
is. In my view, there are no convincing metatheoretical or methodological
reasons for linguists to adopt either these two positions; and, as I have argued
elsewhere, there is a good deal of pretheoretical and (relatively) theory-neutral
empirical evidence which should lead us to reject both (Lyons 1995a: 153–199).
I will not repeat these arguments here. I will simply note that they involve a
consideration of the way in which different natural languages grammaticalize or
lexicalize what ordinary, relatively naive (that is, metatheoretically-unprejudiced)
users of the languages in question take to be differences of meaning: these
involve such phenomena as thematic meaning (or information structure),
coordination and subordination, various kinds of negation, sentence-type (and
clause-type), mood and tense. This list is by no means exhaustive, but it is
certainly comprehensive enough to give anyone who wishes to identify sentence-
meaning with (truth-functionally explicable) propositional content pause for thought.
164 JOHN LYONS
6. Propositional content, subjectivity and propositionalization
In conclusion, I want to emphasize the importance, for linguists, of not assuming

that propositional content is universal throughout the languages of the world.
The thesis of universal and unrestricted intertranslatability or expressibility
— the thesis that whatever can be expressed in any one arbitrarily chosen
language can be expressed in all languages — is demonstrably false as far as
natural languages are concerned. The fact that it has been defended, or assumed
to be defensible, by several eminent theoreticians (despite the existence of readily
accessible empirical evidence which falsifies it) is hard to explain. Indefinitely
many everyday examples can be produced from the vocabularies of any two
natural languages (or any two dialects of the same language) to refute it. In the
present context, however, I shall be concerned, not primarily with differences of
lexical meaning, but with differences of expressive power and semantic structure
that have to do with the propositionalization or non-propositionalization, in
different languages, of the subjectivity of utterance. This is only partly, and only
in some languages, a matter of lexical (rather than grammatical) meaning. By
“propositionalization” is meant “making propositional what is basically (or,
considered from a diachronic point of view, was in an earlier state of a particular
language) non-propositional”. What is meant by “the subjectivity of utterance”
(more precisely, “locutionary and illocutionary subjectivity”) I have dealt with in
some detail elsewhere (Lyons 1984, 1994, 1995a). For present purposes, it will
suffice to say that the subjectivity of utterance is speakers’ (or writers’) expres-
sion of themselves (that is, their will, desires, expectations, attitudes, beliefs, and
so on) in the utterances (that is, utterance-inscriptions) that they produce. It is a
very restricted (and simplified) treatment of subjectivity that will be given here.
I shall be defending two related theses: (i) that languages differ as to
whether they permit or facilitate the propositionalization of the subjectivity of
utterance; and (ii) that propositionalization alters, and may increase, expressive
power. These two theses will be illustrated from the domains of deontic modali-
ty. The kind of expressive power which comes from propositionalization is
plausibly subsumed under the general notion of abstraction which has been seen
as one of the products, historically, of literacy and of the development, in certain
cultures, of what I have described elsewhere (in the non-pejorative, anthropologi-
cal, sense of “myth”) as the myth of literal meaning (Lyons 1991b).
Deontic modality has been much discussed by philosophers in recent years
and brought within the scope of modern formal logic, which has greatly extended
the application of the classical (aletheutic) modalities of necessity and possibility.

Whatever is obligatory is deontically necessary; whatever is permitted is
deontically possible; and so on. This much, I trust, will be intuitively clear; and
it has, of course, become quite common now for linguists, as well as philoso-
phers, to talk in such terms (Huddleston 1984: 160–176; Palmer 1986; Lyons
1995a: 329ff). The connection between deontic logic and the logic of commands
(sometimes referred to as imperative logic) will also be intuitively clear (though
it is perhaps not so easily formalized in terms of standard classical logic: if X,
with the authority to do so and in the appropriate circumstances, commands Y to
perform some action a, X thereby imposes upon Y the obligation to do a (X
makes it deontically necessary for Y to do a). X may then be said to have
created, or brought into existence, the obligation to do a. But the person describ-
ing what has happened cannot, of course, make this additional statement unless
the language of description provides the means for moving to the higher level of
abstraction in which obligations, prohibitions, rights, and so forth, can be
hypostatized or objectified (that is, referred to as entities). And it is the differ-
ence between languages that do and languages that do not provide their users
with the means for making such statements that is of primary concern to us in
the present connection.
The expression of deontic modality in English, as in many natural languag-
es, is not simply a matter of grammar and vocabulary. It is distributed over the
whole language-system, including, as far as spoken language is concerned, the
intonational sub-system. For example, the declarative sentence “I am telling you
to stop smoking” can be used, without change of grammatical or lexical struc-
ture, to perform a variety of illocutionary acts, including those of making a
statement and issuing a command. The illocutionary force of the resultant
utterance, its status as a command or a statement, may well be evident from the
context, but it may also be signalled, in speech, by its intonation-pattern.
Now, it so happens that Standard English allows its users to make neutral
statements in which there is no indication in utterance-tokens (that are the
products of acts of utterance) of an utterer’s attitude towards what is said or his
or her grounds, or evidence, or authority for saying it. For example, there is
nothing subjective (in the relevant sense) in the grammatical or lexical structure
of the sentence:
(14) Smoking is forbidden (here).
166 JOHN LYONS
In uttering it the speaker (or writer) expresses — and typically, if making a

statement, will be asserting — nothing more and nothing less than the proposi-
tion that smoking is forbidden (at the time and in the place in which the
statement is made). In speech, of course, as was noted earlier, there will
commonly be some indication of what we may now refer to as the subjectivity
of utterance in the (non-neutral, or marked) intonation-contour and stress-pattern
that is, as it were, superimposed upon the utterance-token. But let us accept, for
the sake of the argument, that even in speech there can be, in English, subjec-
tively neutral statements. As was mentioned above, it is in any event a matter of
legitimate dispute among linguists whether stress and intonation, neutral or non-
neutral, count as part of sentence-structure, but the general point being made here
is unaffected by any metatheoretical decision that we might take on this issue.
The proposition that smoking is forbidden does not create the prohibition
against smoking. Nor does its assertion as such, though its assertion may of
course, in the appropriate circumstances, have the effect of bringing the prohibi-
tion into existence (an effect which some authors would account for in terms of
the notion of so-called indirect speech-acts). The statement made by uttering (14)
represents the prohibition as existing, objectively, prior to and independently of
the utterance of (14) and can be evaluated for its factual truth or falsity.
Let us now compare (14), which in terms of its grammatical structure is
both declarative and indicative with (15) and (16):
(15) Do not smoke (here)
(16) You must not/may not (cannot) smoke (here).
Of these, (15) differs semantically from (14) in that, being in the imperative
mood, it would be used, characteristically, not to assert the prior existence of the
prohibition, but (provided that the person uttering the sentence has the requisite
deontic authority — legal, moral, or whatever) to create the prohibition. In such
cases, as generally in languages that have the grammatical category of mood, the
imperative mood grammaticalizes (that is, encodes grammatically, rather than
lexically) that part of the subjectivity of utterance which consists in speakers’
expression of their deontic authority (or warrant): their authority (acknowledged
or assumed) to prohibit, or prevent, the addressee, or addressees, from doing
what might otherwise be held to be permitted (that is, deontically possible).
As for (16), this is more problematical, in that it is by no means clear
whether, literally interpreted, it is more like (14) or (15). Most truth-conditional
semanticists would no doubt say that it resembles the former, but the question is
empirically undecidable. What is clear, however, as a matter of (relatively)

theory-neutral fact is that it can be used, in particular contexts, either subjectively
(to create a prohibition) or objectively (to assert that the proposition exists).
Now, the fact that we can construct sentences like (14) in English depends,
first of all, on the fact that English, by virtue of its grammatical structure and
vocabulary, enables its users to refer to acts and activities as second-order
(abstract) entities by nominalizing the verb that denotes them. In this case,
nominalization (the formation of a noun or noun-phrase) involves the use of what
is traditionally called the gerund (identical in form with the present participle
ending in -ing). There are many other kinds of nominalization, grammatical and
lexical, in English; and associated with them there are several important semantic
differences, which tend to be obscured in traditional discussions of their meaning
by the use of the catch-all term “abstract”: it is important not to confuse the
different kinds of abstraction (Lyons 1989, 1995c: 244ff).
The fact that we can propositionalize and objectify the modality of prohibi-
tion depends also on the fact that English provides its users with a set of
lexemes, including adjectives such as “forbidden”, “wrong”, “immoral”, “illegal”,
and so on, which can be employed to ascribe second-order properties to such
second-order entities as acts and activities. In other words, English allows us to
form sentences like “Smoking is forbidden”, which are not only grammatically
comparable with Tarski’s (1944) now famous (but by no means unproblematical)
“Snow is white”, but can also be interpreted as being semantically and logically
comparable, due allowance being made for the ontological differences between
concrete and abstract (second-order) entities and properties. The increase in
expressive power that is provided by this kind of abstraction, it may be observed,
is exactly the same as the increase in expressive power that is achieved by the
move from the first-order to the second-order predicate calculus in modern
formal logic. It is one aspect of what Quine (1960) has called semantic ascent
(Lyons 1989).
But not all natural languages provide their users with the lexical and
grammatical resources to propositionalize and thus to objectify deontic (and other
kinds of) modality in this way. And languages also differ with respect to the
distinctions that they draw in their deontic vocabulary between what is a matter
of religion, law, custom, ethics, morality and so on. At one extreme we might
envisage a language with, let us say, imperative constructions like “Do not
smoke” (or “Honour thy father and thy mother”) and first-order declarative
sentences like “Snow is white” (or “The table is round”), but no abstract
168 JOHN LYONS
vocabulary (of the relevant kind) and no means of nominalization. In such a first-
order language there would be, ex hypothesi, no way of forming sentences like
“Smoking is forbidden”, “It is wrong to smoke”, or even, let us assume, “You
must not/may not (cannot) smoke”. At another extreme, we find languages such
as English, furnished with the kind of grammatical and lexical devices for
propositionalizing deontic modality that I have mentioned and for differentiating
linguistically between what have developed historically, in certain cultures, as
different kinds of deontic modality.
As I observed at the beginning of this section, many philosophers of
language, logicians and even linguists currently subscribe to the thesis of the
universal intertranslatability of natural languages, which implies (amongst others)
the proposition that all natural languages have the same expressive power. In
doing so, they are rejecting (amongst others) the thesis that there have been
stages in the historical development of certain (so-called) natural languages, such
that some languages can be said to be more advanced (richer, more expressive,
more powerful, and so forth) than others. I would certainly not wish to defend
the view that some natural languages are in all respects more primitive, less
advanced, than others in the form in which this view was commonly formulated
and defended (on empirically spurious grounds) in the last century (and even,
notoriously, by Jespersen as late as 1922). At the same time, I do not believe that
it is in principle impossible to evaluate natural languages in terms of their
expressive power. Indeed, there is an immediately applicable measure of
expressive power which comes from the study of formal languages. This tells us
that if one language, X, properly includes within itself another language, Y, then
X is (in this respect) richer, or more powerful, than Y. For example, the
extended propositional calculus which contains modal operators of necessity or
possibility is more powerful than the simple propositional calculus; and a higher-
order logical language, such as the second-order predicate calculus, is more
powerful than a lower-order logical language, such as the first-order predicate
calculus. It follows similarly that, if two natural languages, X and Y, differ from
one another only in that, whereas X has the grammatical and lexical resources
for the propositionalization of deontic modality, Y does not, X is (in this
respect), more powerful, or richer than, Y. And if being able to objectify deontic
modality is rightly regarded as a mark of cultural progress, a language which
facilitates this ability may also be regarded, in this respect at least, as being more
advanced, more progressive, than a language which does not.
However that may be, regardless of whether it is rightly regarded as a mark
of cultural progress or not, the objectification and codification of law (and

morality) was certainly facilitated historically, first of all, by the development in
particular cultures of a concept of literal meaning (or, rather, of a variety of
concepts of literal meaning) and, second, by the development in particular
languages of the necessary devices for the propositionalization of deontic
modality: that is, for the construction of declarative sentences, such as “Smoking
is forbidden” or “It is right to honour one’s parents”, which can be used, in
contrast with imperative sentences, such as “Do not smoke” or “Honour your
parents”, to assert the objective existence of obligations and prohibitions. And
both of these developments would seem to have been associated, historically,
with societal literacy (Lyons 1991b).
We must, of course, be careful not to assume, as swiftly as many nine-
teenth-century thinkers may have done, that such differences of expressive
power in different languages as I have mentioned necessarily reflect different
stages of mental development in their users. To the best of my knowledge, there
is no evidence that this is so, just as there is no evidence of any intrinsic
connection between race and language, or between race and intellectual or artistic
ability. On the other hand, we must be equally careful not to accept uncritically
a version of linguistic egalitarianism according to which all languages (and all
dialects) are intertranslatable and equally appropriate for all purposes.
So far, in what I have said about propositionalization and reference to
abstract (hypostatized) entities, I have failed to make explicit and to exploit the
distinction that I currently draw between second-order extensional and second-
order intensional expressions (Lyons 1989, 1995: 325–326).19 These two
different kinds of reference are seldom, if ever, properly distinguished in either
traditional or modern discussions of what for convenience (and without commit-
ment to any particular approach to grammatical analysis) may be called nominal-
ization and complementation. There is, however, a crucial semantic difference
between them. Extensional reference is reference to real-world, physical, entities:
to such first-order entities as persons, animals and things or to such second-order
entities as (hypostatized or, more precisely, reified) situations and processes or
activities in which first-order entities are involved. Intensional reference, in the
sense in which I am using the term, is reference to a particular class of non-
physical entities: to such first-order non-physical entities as those which some
philosophers call individual concepts (which do not concern us in the present
context) or to such second-order non-physical entities as propositions. The
traditional term “abstract noun” does not of course distinguish between these two
170 JOHN LYONS
kinds of abstractness: physical and non-physical. Limitations of space prevent me

from developing this topic in detail. In the present context it must suffice for me
to make just two points.
The first, obvious enough perhaps in the light of what has been said above,
but none the less worth emphasizing, is that the expressive power of a language,
such as English, that provides its users with the grammatical and lexical resourc-
es for making intensional (as well as extensional) second-order reference is
greater (ceteris paribus) than that of a language that does not.20 There is a further
distinction to be drawn between languages that permit the construction of
expressions that can be used as the complements of verbs of saying, knowing,
believing, and so forth — verbs of illocutionary activity and verbs of mental
state or activity — and those that also provide their users with the means of
ascribing to second-order intensional entities such second-order intensional
properties as truth and falsity. Apart from anything else, it is only in the latter
subclass of higher-order languages, whether formalized or not, that one can do
theoretical semantics (and write such articles as this one!).21
The second point to be made is that the issues I have raised in the final
section of this article, relating to propositionalization and subjectivity, on the one
hand, and, on the other, to differences of expressive power which derive from
differences of grammatical structure, are currently very difficult to investigate
empirically, because existing reference grammars of particular languages (and
even otherwise theoretically sophisticated descriptions of such topics as complex
sentences and complementation) do not operate with a rich enough set of
semantic and ontological distinctions and therefore do not provide the necessary
information in a form that makes it accessible to those who do not themselves have
the requisite competence in a sufficiently wide range of languages. It is to be hoped
that this situation will, in time, be remedied (Dixon 1995, Lyons 1995b: 243–246).
Notes
1. I am pleased that this article should be appearing, in its present form, in a volume which
honours Rodney Huddleston for his work in linguistics and more particularly for his application
of general linguistic theory to the description of English. Huddleston is one of the relatively few
authors who defines his metalanguage with care and, equally carefully, relates his English-
specific descriptive metalanguage to the theoretical metalanguage of general linguistics. He is
also one of the few linguists writing on the grammatical structure of English who operates with
much the same set of theoretical distinctions (and, for the most part, the same terminology) as
I have done myself. There would seem to be, however, one or two differences of viewpoint,
and I will make it my business, in the present context, to comment specifically upon these.
2. A fuller version of several sections of this article, with differences of focus and emphasis and
in a broader context, will be published in Lyons (forthcoming).
3. In the version of this article submitted for publication, these terminological distinctions, which
are absolutely critical for a proper understanding of the argument, were reinforced by the use
of a set of typographical conventions first adopted in Lyons (1997a): (i) italics for forms,
including utterance-inscriptions and units or stretches of text; (ii) single quotation-marks for
lexemes, and other expressions, including system-sentences; (iii) double quotation-marks for the
meaning (including propositional content) of an expression and also for propositions (compare
Lyons 1995a: xvii). Regrettably, these typographical conventions have not been maintained in
the published version. In particular, the distinction between single and double quotes has
disappeared. I have made compensatory adjustments, ad hoc, in the later sections, to mark the
difference between sentences and propositions. I trust that this is sufficient to determine major
inconsistencies.
4. The type-token distinction is not relevant to system-systems because (a) this distinction, in my
view, is best restricted to that particular class of physical entities members of subclasses of
which can be described as (semiotic) inscriptions in a particular medium, phonic, graphic, and
the like, and (b) system-sentences, in contrast with text-sentences and other utterance-inscrip-
tions, are not physical entities. Obviously, it is possible to broaden the definition of “type” and
“token” (as many authors do nowadays) to the point that they become more or less synonymous
with “class” and “member”. But this, in my view, has the effect of depriving the more specific
terms of their usefulness.
5. Huddleston’s characterization of sentences and clauses is in this respect eminently traditional
(1984: 18–21).
6. Independently of the generality and validity of his main thesis, Miller usefully reminds us of the
powerful influence that has been, and still is, exerted on linguistic theory by the normative and
literary prejudices of traditional grammar. He also makes it clear that theoretically-minded
linguists should once again take more seriously than many of them have done since the
pendulum-swing from relativism to universalism in the 1960s the possibility that such concepts
as word and sentence (as they are commonly defined) are relevant in the description of some,
but not all, languages and may also be, to some degree, medium-dependent and style-dependent.
7. In his interesting comparative study of complementation in English, Fijian and Dyirbal, Dixon
(1995) notes, among much else that is relevant to what I am discussing in this and other
sections of this paper, that compound sentences in Dyirbal can be defined only in terms of their
intonational contour. This is perhaps tantamount to saying that Dyirbal has no compound text-
sentences, but that it may well be justifiable to talk about compound sentences as (spoken) text-
sentences as the products of utterance.
8. Some of the reasons for taking this view — including the difficulty of establishing compound
(in contrast with complex) sentences and the fact that syntactic dependencies can run over “a
sequence of what would normally be regarded as separate sentences” — were actually given in
Lyons (1968: 170ff). I did not, however, draw from this evidence the conclusions that,
arguably, I should have done.
172 JOHN LYONS
9. Huddleston very kindly read Lyons (1995a) for me in draft, and it was he, more than anyone
else at the time, who was responsible for getting me to see that this was so.
10. Generally speaking, the terms that I use here are the same as those used by Huddleston in his
work on English (especially Huddleston 1984). More recently, he has chosen to use “question”
for the semantic classification of sentences, rather than for the (pragmatic) classification of
utterances (Huddleston 1995). I am not sure that his reasons for doing so would have the same
force within a descriptive framework which exploits the distinction between system-sentences
and text-sentence as I should wish to do. But then, even when one does draw this distinction,
there are many different points at which linguists will disagree, reasonably, as to whether
undoubted differences of meaning between tokens of one utterance-type and tokens of another
should be attributed to the language-system (and therefore, in the usage of many linguists these
days classified as semantic). As will be clear from the several sections of this article taken
together, there is a whole complex of interrelated metatheoretical issues, such as the status of
compound sentences and of intonational features in relation to system-sentences, which bear on
this issue. I do of course approve of the general thrust of Huddleston’s argument and welcome
his subtle discussion of a range of well-chosen examples. I also recognize that a majority of
linguists would probably accept as well-formed system-sentences of English many word-strings
whose generability within the language-system I might be inclined to query.
11. The major parts of Austin’s theory (and the only parts that have been adopted, and developed
and modified, by linguists) are, of course, medium-neutral and applicable to (the products of)
both speech and writing.
12. It might be argued that in many languages imperative and subjunctive sentences (that is,
sentences in which the main verb is in the imperative or subjunctive) are no less basic,
pretheoretically, than sentences in which the main verb is in the indicative.
13. By Quasi-English I mean a (possibly infinite) set of language-systems which are, in general,
typologically identical with Standard English (itself, of course, a many-member set of
intersecting language-systems) except for a more or less limited number of deviations from it
on selected parameters. Some reviewers of Lyons (1977a) have objected to my use of artificial
(but realistic) Quasi-English for expository theoretical purposes. In doing so, I think they may
have failed to understand both my purpose in invoking the concept of Quasi-English and the
use I made of it and also the difference between natural, non-natural and unnatural languages.
Standard English, as described by linguists, is of course a non-natural language (Lyons, 1991a:
46–72). In the current, necessarily very tentative, state of empirically based language typology,
so-called implicational (or Greenbergian) universals which might be thought to exclude as
impossible the kinds of parametric variation that I postulate, here and elsewhere, as theoretically
possible should not be invoked quite so confidently as they are by many linguists these days
(Croft 1990: 44ff).
14. A good case can, of course, be made for dealing with the modal auxiliaries of Modern English
in terms of the category of mood (Huddleston, 1984: 164ff). Whether something is grammatical
or lexical is not always a matter of yes or no, but rather of degree. The modal auxiliaries of
Modern (Standard) English are certainly more fully grammaticalized than the semantically
comparable modal auxiliaries of German, French, Italian, and so on.
15. This is the view that Huddleston (1984) takes as far as English is concerned. His arguments
may very well apply more generally.

16. The situation is further complicated by the fact that, independently of any motivated preferences
that scholars writing in English might have for the use of “proposition”, “sentence” and
”statement”, many of the seminal works of logical semantics were originally written in
languages (notably German or Polish) without exactly equivalent terms and by scholars who
were brought up in a different school-grammatical tradition. The now standard English-
language versions (which are often more widely cited than the originals) are sometimes
misleading because at times they import into the text, whether intentionally or not, a particular
(possibly anachronistic) view which may not be (and perhaps for historical reasons could not
have been) the view of the author on the issues being discussed in this section of my article.
The misleading (and anachronistic) standard translation of Frege’s (1892) “Bedeutung” with the
English word “reference” is comparable. But this is a much larger topic, wich cannot be dealt
with here.
17. What I have just said about the eternal constancy (and determinacy) of truth-values would
require some adjustment if we were to adopt the point of view adopted by some of the
proponents of possible-worlds semantics (Montague 1974). But once this adjustment is made,
there is (I believe) no radical incompatibility between the notion of truth-relativized-to-a-
possible-world and the eternal constancy of truth-values (that is, the so-called God’s-eye view
of truth “sub specie aeternitatis”). It is, to say the least, easier to operate for present purposes
with the classical notion of absolute truth (and falsity).
18. The term “intension” is used in a variety of related, but different, senses (in the non-technical
sense of “sense”) in philosophical and logical semantics. My own use derives from, but is not
strictly to be identified with, Carnap’s.
19. This process of fixing the values of a propositional function can, in turn, be compared with that
of fixing the value of a simple algebraic expression or function, by replacing the numerical
variables with numerical constants: for example, by determining the value of z as 5 in z = x +
y by replacing x with 2 and y with 3.
20. In earlier work I operated with what was, or terminologically might have appeared to be, a one-
dimensional, three-way, distinction (Lyons 1977a: 441ff). But the distinction between the two
kinds of abstract reference, extensional and intensional, was drawn within that and propositions,
in contrast with real-word properties and situations, were identified as intensional entities.
21. It is a matter for empirical determination whether there are any (so-called) natural languages in
which it is possible to construct second-order intensional, but not second-order extensionsal,
referring expressions. It is easy enough to construct a version of Quasi-English which has this
property.
22. What is in itself, grammatically and lexically, a first-order language without complement
clauses can, of course, be used to construct utterances (by parataxis) in which a demonstrative
is understood to refer to the proposition expressed in the clause with which the demonstrative
is anaphorically associated (Lyons 1978). An example might be the English text-sentence John
said this: it is raining (in which the semicolon symbolizes parataxis). It is of course arguable
that the clausal indirect-discourse constructions with originally demonstrative complementizers
(such as English that) are diachronically explicable along these lines.
174 JOHN LYONS
References
Austin, J. L. 1962. How To Do Things with Words. Oxford: Clarendon Press.

Bloomfield, L. 1933. Language. New York: Holt, Rinehart and Winston.
Carnap, R. 1942. Introduction to Semantics. Cambridge, Mass. : MIT Press.
Carnap, R. 1947. Meaning and Necessity. Chicago: Chicago University Press.
Croft, W. 1990. Typology and Universals. Cambridge: Cambridge University Press.
Dixon, R. M. W. 1995. “Complement clauses and complementation strategies”.
F. Palmer 1995: 175–220.
Frege, G. 1892. “Über Sinn und Bedeutung”. Zeitschrift für Philosophie und
philsophische Kritik 100. 25–50. English translation: “On sense and refer-
ence”. Translations from the Philosophical Writings of Gottlob Frege ed. by
P.Geach &M. Black, 56–78. Oxford: Blackwell, 1960.
Huddleston, R. 1984. Introduction to the Grammar of English Cambridge: Cam-
bridge University Press.
Huddleston, R. 1988. English Grammar: An Outline. Cambridge: Cambridge
University Press.
Huddleston, R. 1995. “The Contrast between Interrogatives and Questions”.
Journal of Linguistics 30.411–39.
Jespersen, O. 1922. Language: Its Nature, Origin and Development. London:
Allen and Unwin.
Lyons, J. 1968. Introduction to Theoretical Linguistics. London and New York:
Lyons, J. 1969. “Intervention in ‘Formal logic and Natural Languages’”. Founda-
tions of Language 5. 269.
Lyons, J. 1977a. Semantics. 2 vols. London and New York: Cambridge University
Press.
Lyons, J. 1977b. “Statements, Questions and Commands”. Linguistic Structures
Processing ed. by A. Zampolli, 255–280. Amsterdam, New York and
Oxford: North Holland.
Lyons, J. 1978. “Deixis and Anaphora”. The Development of Conversation and
Discourse ed. by T. Myers, 88–103. Edinburgh: Edinburgh Press. (To be
reprinted in Lyons forthcoming.)
Lyons, J. 1984. “La subjectivité dans le langage et dans les langues”. E. Ben-
veniste Aujourd’hui, vol. 1., ed. by G. Serbat, 131–139. Paris: Société pour
l’information grammaticale.
Lyons, J. 1989. “Semantic Ascent: a Neglected Aspect of Syntactic Typology”.

In Essays on Grammatical Theory and Universal Grammar ed. by D. G.
Arnold et al., 153–186. London: Oxford University Press.
Lyons, J. 1991a. Natural Language and Universal Grammar: Essays in Linguistic
Theory, Vol. 1. Cambridge: Cambridge University Press.
Lyons, J. 1991b. “Linguistics and Law: the Legacy of Sir Henry Maine”. The
Victorian Achievement of Henry Maine ed. by A. Diamond. Cambridge:
Lyons, J. 1994. “Subjectivity and Subjecthood”. Subjectivity and Subjecthood:
The Status of the Subject in Linguistic Theory ed. by M. Yaguello, 9–17.
Paris: Ophrys.
Lyons, J. 1995a. Linguistic Semantics: An Introduction. Cambridge: Cambridge
University Press.
Lyons, J. 1995b. “Grammar and Meaning”. F. Palmer 1995:221– 249.
Lyons, J. 1995c. “Performance and Competence and Related Notions”. Perfor-
mance and Competence in Second Language Acquisition ed. by G. Brown, K.
Malmkjaer and J. Williams. Cambridge: Cambridge University Press.
Lyons, J. Forthcoming. Semantics, Subjectivity and Localism: Essays in Linguistic
Theory. Cambridge: Cambridge University Press.
Matthews, P. H. 1981. Syntax. Cambridge: Cambridge University Press.
Miller, J. 1995. “Does Spoken Language have Sentences?”. Palmer 1995: 116–135.
Montague, R. 1974. Formal Philosophy: Selected Papers ed. by R. H. Thomason.
New Haven: Conn.: Yale University Press.
Nesfield, J. C. 1939. Manual of English Grammar and Composition. London:
Macmillan.
Palmer, F. R. 1986. Mood and Modality. Cambridge: Cambridge University Press.
Palmer, F. R. (ed.) 1995. Grammar and Meaning: Essays in Honour of Sir John
Lyons. Cambridge: Cambridge University Press.
Quine, W. V. (1960). Word and Object. Cambridge, Mass.: MIT Press.
Tarski, A. 1944. “The Semantical Conception of Truth”. Philosophy and Pheno-
menological Research 4. 341–375.
Some interactions between tense and negation in
English
James McCawley
When an English sentence is negated in the normal manner, with -n’t attached to
the tensed auxiliary verb (to an appropriate form of do if there is otherwise no
auxiliary verb), inversion in a corresponding interrogative sentence normally is
of the whole V-n’t combination; inversion leaving the negative element behind
often results in lowered acceptability:
(1) a. I can’t help you.
a′. Can’t you help me?
a″. ??Can you not help me?
b. John didn’t bring the money.
b′. Didn’t John bring the money?
b″. ?Did John not bring the money?
However, in sentences such as (2a) and (2b), inversion of V-n’t results in
substantially lowered acceptability, while inversion just of the auxiliary verb,
stranding the not, yields a perfectly normal sentence:
(2) a. He actually can’t get it through his head that we want to help him.
a′. *Can’t you actually get it through your head that we want to
help you?
a″. Can you actually not get it through your head that we want to
help you?
b. They still haven’t answered our letter.
b′. *Haven’t they still answered our letter?
b″. Have they still not answered our letter?
178 JAMES MCCAWLEY
c. He really isn’t feeling well.

c′. *Isn’t he really feeling well?
c″. Is he really not feeling well?
It is evidently the preverbal adverb that is responsible for this deviation from the
normal pattern of inversion, since corresponding sentences without the adverb
display the pattern in (1):
(3) a. He can’t get it through his head that we want to help him.
a′. Can’t you get it through your head that we want to help you?
a″. ??Can you not get it through your head that we want to help you?
b. They haven’t answered our letter.
b′. Haven’t they answered our letter?
b″. ?Have they not answered our letter?
c. He isn’t feeling well.
c′. Isn’t he feeling well?
c″. ?Is he not feeling well?
A motivation for the deviation from normal Inversion in (2) emerges from a
consideration of scope relations. In (2a) and (2b), the adverb has the negation
and the auxiliary verb in its scope (what is actually the case is that he can’t get
it through his head that we want to help him; what still is the case is that they
haven’t answered our letter; what really is the case is that he isn’t feeling well).
Inversion moves the tensed auxiliary verb to a position higher in the structure
than everything else in the clause, thus to a position in which, if scope relations
corresponded exactly to structural relations in surface structure, everything else
should be in the scope of that auxiliary verb. Scope relations, of course, don’t
correspond in anything like such a uniform manner to surface structural relations,
but avoidance of discrepancies between surface structure and logical scope can
at least motivate deviations from otherwise general syntactic rules. English polar
(“yes/no”) interrogatives require inversion, and thus if (2a–c) are to have polar
interrogative counterparts, the tensed auxiliary verb must appear higher in the
structure than the adverb and thus deviate from the pristine match between scope
and surface constituency. If the negative element is stranded, at least that element
remains in a surface position that matches its scope relation to actually or still.
By contrast, when the adverb is absent, the motivation for exempting -n’t from
Inversion is also absent: Inversion alters only the structural relation of the
auxiliary verb and the negative element to something (the personal pronoun
INTERACTIONS BETWEEN TENSE AND NEGATION 179
subject) to which they have no significant scope relation.

What if the subject were something that could stand in a significant scope
relation to the auxiliary and/or a negative, say, a quantified NP? We find again
Inversion affecting only the auxiliary verb and not the negative element:
(4) a. Someone hasn’t replied to the invitation.
a′. Has someone not replied to the invitation?
a″. Hasn’t someone replied to the invitation?
(≠ interrogative of (4a))
a″′. Hasn’t anyone replied to the invitation?
(≠ interrogative of (4a))
b. Most of the students couldn’t solve problem #3.
b′. Could most of the students not solve problem #3?
b″. Couldn’t most of the students solve problem #3?
(≠ interrogative of (4b))
The only plausible interpretations of (4a) and (4b) are ones in which quantifier
has the auxiliary and the negation in its scope (that is, (4a) says that “x hasn’t
replied to the invitation” is true of someone, and (4b) that “x couldn’t solve
problem #3” is true of most of the students). The forms that would result if the
negation were included in the Inversion have either an interpretation in terms of
fake negation (“Someone has replied to the invitation, right?”, “Most of the
students could solve problem #3, right?”) or one with different scope relations
((4a″′) is an interrogative counterpart of No one has replied to the invitation).
Again, Inversion in which only the auxiliary verb and not the negation partici-
pates serves as a compromise between the requirement that there be Inversion
and the goal of avoiding mismatches between surface constituent structure and
logical scope relations.
I turn now to the possibilities for combining negation and a present perfect.
English allows both sentences in which a present perfect is negated (5a) and
sentences in which a present perfect is applied to a negated verb phrase (5b):
(5) a. John hasn’t received any encouragement.
b. John has [not returned my calls] many times.
In (5a), one says that (in the relevant past interval that stretches up to the
present) there is no event of John receiving some encouragement; in (5b), one
says that there are many past events of John not returning my calls. Let us try
combining a sentence modifier such as probably with (5a):
180 JAMES MCCAWLEY
(6) a. John has probably not received any encouragement.

a′. ??Johnhasn’t probably received any encouragement.
Note that while has is higher than not in the surface form of (6a), which suggests
that in logical form the negation is embedded in the present perfect, as in (5b),
(6a) in fact only receives an interpretation in which the present perfect is in the
scope of the not. If the negation is put in the position that is the usual one for an
interpretation in which the tense and auxiliary verb are in the scope of the
negation, in other words, if it is made a contracted adjunct to the auxiliary verb,
probably sounds quite odd in the position after the tensed auxiliary verb (6a′). I
suggest that here too the normal preference for a contracted negation is overrid-
den by the need to avoid a surface structure that appears to put into the scope of
the negation material that is actually outside its scope, here, probably. The
surface structure of (6a) of course does not reflect the scope relations correctly
either, as was pointed out above; however, the syntax/semantics mismatch in
which has incorrectly appears to be outside the scope of not seems to be
admitted more readily than the one in which probably incorrectly appears to be
within the scope of not.
Consider now (7a), which allows the three interpretations (7b–b″):
(7) a. I didn’t spend $20.
b. $20 isn’t the amount that I spent.
b′. I spent less than $20.
b″. There is $20 that I didn’t spend (i.e. I had $20 left).
The polar interrogative counterparts of (7a) differ with regard to which of these
meanings they can relate to:
(8) a. Didn’t you spend $20?
a′. Did you not spend $20?
The most obvious interpretation of (8a) is one with a fake negation (“You spent
$20, right?”); it can also marginally be given interpretations corresponding to
(7b) and (7b′), but not one corresponding to (7b″). By contrast, (8a′), in which
the negative element does not take part in the inversion, can be interpreted as
corresponding to any of the three senses of (7a). In a corresponding wh-question,
inclusion of not in the inversion is virtually excluded:
(9) a. *How much money didn’t you spend? (Kuno and Takami 1992:297)
a′. How much money did you not spend?
The one obvious interpretation of (9a′) is a counterpart to (7b″); more specifical-

ly, it refers to a specific past occasion, presupposes that there is some amount of
money that you did not spend on that occasion, and asks what that amount is.
Kuno and Takami argue that (9a) is unacceptable because there is no context that
would motivate a wh-question with a meaning corresponding to (7b) or (7b′); for
example there would always be an open-ended range of amounts x such that
what you spent wasn’t x. The only interpretation of (7a) under which a corre-
sponding wh-interrogative could satisfy the pragmatic conditions for asking a wh-
question is (7b″) (an interpretation that Kuno and Takami did not take into
account), and that interpretation only allows a wh-interrogative form in which
inversion leaves the negative element stranded.
I suggest that (9a′) and (8a′) reflect a sort of underlying structure that is
available in the scheme of syntactic analysis developed in McCawley (1988) but
which did not figure in the analyses presented there. Specifically, I proposed that
tenses and auxiliary verbs are deep structure predicate elements that take
sentential complements, and that the Ss that are posited in that proposal behave
like deep structure Ss in all respects, in particular that they can be the hosts of
S-modifying adjuncts, can be the locus of conjoining or of negation, and can be
the scope of quantifiers. These assumptions allow not only for deep structures in
which a tensed S is negated but also for deep structures in which the comple-
ment of a tense is negated:
(10) a. S b. S
not S NP 0′
NP 0′ S 0
Pres
S 0 not S
Pres
Only structures such as (10a) actually were made use of in the analyses of
McCawley (1988), and the availability of structures like (10b) was a minor
embarrassment. What I have said above, however, suggests that (8a’) and (9a’)
may indeed call for structures such as (10b), in which the tense is outside the
scope of the negation, and that a negation that is below the tense does not
participate in inversion.1 Under the assumption that tenses provide the time
182 JAMES MCCAWLEY
reference of their complements, a deep structure of the form (10b) fits the
meaning (7b″), as contrasted with (10a), which would fit the meanings (7b) and
(7b′): while (7b) and (7b′) deny that on the occasion in question the money John
spent equalled or reached $20, (7b″) says that on that occasion there was $20
that he didn’t spend. Further evidence that the not is below the tense in the deep
structure of the (7b″) sense of (8a) is provided by the unacceptability of V′-
deletion when didn’t spend $20 has that sense. That is, (11a) (and likewise
(11a′)) is acceptable as a statement that neither John nor Mary made an expendi-
ture of $20, but not as a statement that both of them had $20 remaining after
paying for their purchases:
(11) a. John didn’t spend $20, and Mary didn’t Ø either.
a′. John didn’t spend $20, nor did Mary Ø.
Another pair of sentences whose difference in acceptability can also plausibly be
accounted for in terms of a semantic demand that a negation be in the scope of
a tense is (12a) and (12a′):
(12) a. ??Whatyear didn’t Citizen Kane win the Oscar?
a′. What year did Citizen Kane not win the Oscar?
Since Citizen Kane never won the Oscar (for best picture), every year is, strictly
speaking, a year in which it didn’t win the Oscar. However, there is only one
year in which it was eligible to win the Oscar, and (12a′) asks what that year
was. While the semantic motivation for a choice between a (10a) and a (10b)
deep structure is less clear than in the case of (8) and (9), it is at least plausible
to take a logical structure matching (10a) as corresponding to the silly question
(12a) that would ask for the year in which it was not the case that Citizen Kane
won the Oscar, and one matching (10b) as corresponding to the sensible question
(12a′), with an interpretation presupposing that there is some specific year having
the property of Citizen Kane not winning the Oscar that year.
I will close this note by alluding to a direction in which I hope to expand
the treatment of tense and negation sketched above, though one whose execution
will have to depend on the treatment of some semantic differences whose
existence is relatively clear but whose precise nature is not, for example, the
differences in the following pairs of sentences:
(13) a. Smith is still alive. (Indeed, he’s 85 and in surprisingly good
health.)
a′. Smith still is alive. (I told you idiots that I wanted him killed.)
b. Everyone is still being treated rottenly. (Universal
mistreatment of the people has been going on for centuries.)
b′. Everyone still is being treated rottenly. (These people have
had it bad all their lives.)
In McCawley (1988), I presented an analysis in which sentences in which a
S-modifying adverb precedes or follows a tensed auxiliary verb differed with
regard to whether in deep structure the adverb modified the whole tensed S or
the complement of the tense (thus, whether it was in the position of not in (10a)
or in (10b)). S-modifiers generally allow optional conversion into V′-modifiers;2
both word orders involve the conversion of still into a V′-modifier, and “Attrac-
tion to Tense”, the cyclic transformation that raises an auxiliary verb to where
the tense is, skips over any intervening modifiers, thus yielding word orders such
as … is still …
The problem that sentences as in (13) pose is that of how to reconcile this
account of the syntax of the two word orders with an adequate account of the
semantics of the corresponding sentences. Examples in which the adverb is still
are particularly problematic, since it is not even clear that a semantic structure
corresponding to the deep structure posited for … still is … is even semantically
coherent. Still refers to two times: the reference time t at which still(p) is
evaluated, and the time t′ of a presupposed prior instance of a given state proposition
p; still(p) says that p is true at all points on the time interval [t′, t]. The reference
time for an occurrence of still can be in the past, the present, or the future:
(14) a. When I arrived, John was typing, and when I left, he was still
typing.
b. When I arrived, John was typing, and he’s still typing now.
c. John is typing right now, and I bet that when we leave, he’ll
still be typing.
According to what I have said about the position of adverbs in word order, these
sentences should require deep structures in which still is in the complement of
the tense. But that conclusion conflicts with the most obvious inference to draw
from the dependence of the interpretation of still on a reference time, namely that
in semantic structures still should be in the complement of the tense whose
interpretation provides its reference time. The only obvious possibilities for
modification in the analysis sketched here that would leave it relatively intact
184 JAMES MCCAWLEY
while accommodating these observations are either (i) to allow (14a-c) to have
deep structures in which still is below the tense, by positing some mechanism
that will reverse the order of still and the auxiliary verb (perhaps a rule allowing
a tensed auxiliary verb to be cliticized to the subject even when it isn’t adjacent
to it, thus reversing the order of the auxiliary and a preceding adverb), or (ii) to
deviate from strict compositionality of interpretation by providing an interpreta-
tion for still even when it is above the corresponding tense. Even so, pairs of
sentences as in (13) differ in meaning by more than would appear to correspond
to a difference in whether still is above or below the tense in logical structure.
Perhaps the differences can be made to follow from the interaction of the tense
and the adverb with some third item, such as the quantified subject in (13b) and
(13b’), which has to be respectively inside and outside the scope of still, but I
have not yet achieved a satisfactory account of how the forms of these sentences
correlate with their meanings. The correlation is not a simple parallelism between
surface constituent structure and logical structure, since, if anything, the relation
between the surface forms of (13b) and (13b’) and their logical structures is anti-
iconic: still has higher scope in the sentence in which it appears lower in the
surface constituent structure.
Notes
1. While the negative element does not participate in Inversion, it does combine in a contracted
form in non-inverted sentences such as (7a). The proposal made here thus requires that there be
two derivations for contracted combinations of auxiliary verb and not, one in which the not is
above the tense in deep structure and is combined with it by the syntactic transformation of
Negative-placement and one in which the not is below the tense in deep structure and is
combined with the tensed auxiliary verb not syntactically but only morphologically.
2. One of the few S-modifiers in English that does not allow conversion into a V′-modifier is yet:
Bill has been quite nasty to me, and yet I can’t help liking him.
*Bill has been quite nasty to me, and I yet can’t help liking him.
Languages differ with regard to the optionality of this transformation; in English, it is almost
always optional (not is one of the few S-modifiers that are obligatorily converted into V’-
modifiers), whereas in Chinese, a large proportion of underlying S-modifiers can appear in
surface structure only in V′-modifier positions.
References
Kuno, S. and K. Takami. 1992. “Negation and Extraction”. Papers from the 28th
Regional Meeting, Vol. 1. Chicago: Chicago Linguistic Society, 297–317.
McCawley, J. D. 1988. The Syntactic Phenomena of English. Chicago: University
of Chicago Press.
The English accusative-and-infinitive construction
A categorial analysis
John Payne
Introduction
In this paper, my intention is to provide a new solution to the syntactic problems

raised by English accusative-and-infinitive constructions such as (1):
(1) a. Joan believes him to be a genius.
b. Joan proved him to be guilty
The proposed solution is a variant of the “wrap” analysis originally proposed by
Bach (1979) and the theoretical framework in which the solution is presented is
a version of categorial grammar using wrap, infix, and gap operators. The paper
concentrates on syntactic issues: a formal semantic description of the construc-
tion is not developed, but would follow quite naturally from the categories and
operations proposed.
In Section 1, the initial motivation for the “wrap” type of analysis is
provided through an outline comparison of three families of solutions to the
syntactic problems raised by the accusative-and-infinitive construction. Section 2
then presents the basic categories and operations of the categorial grammar
which will be used. The new analysis, which involves treating objects as infixed
functors, is described in Section 3. The consequences of the new analysis for
accusative-and-infinitive constructions involving Heavy NP Shift, coordination
and extraction are then developed in detail in Sections 4, 5 and 6. Section 7
discusses the treatment of passive versions of the construction. The paper
concludes with a summary of the categorial operations which are needed, and a
note on the nature of the categorial system being developed.
188 JOHN PAYNE
1. Background
There are essentially three families of solution to the accusative-and-infinitive

construction in the literature:
A. Clausal complement analyses

The clausal complement analyses are analyses in which believe and prove take
non-finite clausal complements him to be a genius and him to be guilty, and the
noun phrase him is treated as the subject of the non-finite clause. They can be
represented in a schematic way as follows:
(2) S
NP VP
Joan V S
believes NP VP[INF]
proved
him to be a genius
to be guilty
A representation of this form automatically accounts for the predication relation-

ship between the accusative noun phrase him and the infinitivals to be a genius
and to be guilty, which is taken to be the standard predication relationship
between any subject and predicate. Joan believes him to be a genius is the direct
non-finite equivalent to Joan believes he is a genius. It is then necessary to
provide additional mechanisms which handle the object-like properties of him.
These properties minimally include the accusative case of the pronoun and its
forced adjacency to the higher verb believe or prove. One standard account of
these properties is the “exceptional case-marking” rule of the theory of Govern-
ment and Binding (Chomsky 1981), which simply allows a verb like believe or
prove to assign accusative case to any noun phrase to its immediate right,
whether subject of the subordinate clause in Joan believes him to be a genius or
straightforward direct object in Joan believes him. However, the exceptional case-
marking solution is in the end untenable: it does not account for further object-
like properties of the pronoun based on anaphor-binding and scope relations.
These properties were first pointed out by Postal (1974) and subsequently
THE ENGLISH ACCUSATIVE-AND-INFINITIVE CONSTRUCTION 189
developed by Bresnan (1976, 1982) and Lasnik and Saito (1991). For example,
a standard negative object licenses any in an adverbial at the same level (exam-
ples from Lasnik and Saito 1991:329):
(3) The DA accused none of the defendants during any of the trials
The same behaviour can be observed with the accusative noun phrase in the
accusative-and-infinitive construction:
(4) The DA proved none of the defendants to be guilty during any of
the trials
The subject of a finite complement does not, however, license any:
(5) *The DA proved that none of the defendants were guilty during any
of the trials
Such contrasts, which seem reasonably secure regardless of the exact levels of
acceptability, appear to demonstrate that the accusative noun phrase occupies a
scope position analogous to that of a standard direct object, rather than the
position of subject in a subordinate clause. Accordingly, following the original
transformational approach of Rosenbaum (1967), further mechanisms have to be
proposed within the single complement solution which involve the raising of the
accusative noun phrase from subject to object position, either overtly as in the
original approach, or covertly as in the Minimalist Program (Chomsky 1995).
The possibility of inserting an adverbial with matrix scope in between the
accusative noun phrase and the infinitival, as in (6), suggests, however, that raising
in the single complement analysis must be an overt rather than a covert process:
(6) Joan believes him fervently to be a genius.
In an attempt to avoid the conclusion that overt raising is necessary, Chomsky
(1995:333) speculates that examples like (6) involve extraposition of the
infinitival from a structure like Joan believes [him to be a genius] fervently. They
would therefore be analogous to examples like (7), which derives from Joan felt
[an obligation to leave immediately] last night:
(7) Joan felt an obligation last night to leave immediately.
The extraposition argument against overt raising does not naturally extend,
however, to examples like (8), noted by Postal (1974:413) and Seuren (1985:73–75),
in which the accusative noun phrase precedes the particle of a matrix phrasal verb
like figure out:
190 JOHN PAYNE
(8) Joan figured him out to be a genius

The putative pre-extraposition source for (8) would have to be the ungrammatical *Joan
figured him to be a genius out, with extraposition in this case an obligatory rule.
B. Separate complement analyses

Separate complement analyses are analyses in which believe and prove take
separate accusative noun phrase and infinitival complements, rather than a single
clausal complement. For discussion along these lines, see Huddleston (1988:194–5).
Separate complement analyses can be represented schematically as follows:
(9) a. S
NP VP
Joan VP VP[INF]
V NP
believes him to be a genius

proved to be guilty
(9) b. S
NP VP
Joan
V NP VP[INF]

proved to be guilty
Such solutions immediately present the accusative noun phrase in a syntactic

position plausibly identifiable with the position of a standard direct object (sister
of the matrix verb), and the case property of the accusative noun phrase therefore
needs no special treatment. The binary-branching version (i) is rarely proposed;
it can be found for example as an analysis of the parallel construction Joan
believes him a genius in Kang (1995). One problem it immediately creates arises
from the fact that the pronoun him is put in a position where it is structurally
lower than the infinitival. It is then not a straightforward matter to account for
the fact that the object can bind a reflexive pronoun in the infinitival, as in (10):
(10) Joan believes himi to have perjured himselfi
In the ternary-branching proposal (ii) this problem seems to be resolved: the
object commands the infinitival. Indeed, when the ternary-branching proposal is
compared with the clausal complement analysis, it can be seen that there is no
need for any raising process to put the accusative noun phrase in a position in
which its anaphor-binding and scope properties can be accounted for. Instead,
however, a separate statement has to be made concerning the predication
relationship between the accusative noun phrase and the infinitival, which are no
longer a syntactic constituent. One way of doing this is as in Lexical-Functional
Grammar (Bresnan 1982), namely to assume separate levels of constituent
structure and functional structure, and to assign the accusative noun phrase the
subject relationship with respect to the infinitival predicate at the level of
functional structure. Another, as in Generalized Phrase Structure Grammar
(Gazdar, Klein, Pullum and Sag 1985), is to assume a meaning postulate that
requires the semantic translation directly derived from combining believe and
prove separately with their noun phrase and infinitival complements to be
equivalent to one in which believe and prove take a clausal complement, i.e.
believe′(to be a genius′)(him′) ≡ believe′(to be a genius (him′)). These proposals
in their different ways represent the inverse of the raising proposals in the
clausal complement analysis. More radically, it is possible to assume, following
Chierchia (1984) and Langacker (1995) in widely differing frameworks, that the
semantic structure of the accusative-and-infinitive construction is fully isomor-
phic to the proposed syntactic structure, treating the accusative noun phrase as a
genuine argument of believe and the infinitive to be a genius as a property. One
difficulty with this more radical kind of proposal, however, is that dummy
subjects like existential there have to be invested with some independent, and not
simply vacuous, interpretation in examples like (11):
(11) Joan believes there to be a party tonight
C. Complex predicate analyses
Neither the clausal complement nor the separate complement solutions permit the
discontinuous strings believe to be a genius in Joan believes him to be a genius or
192 JOHN PAYNE
proved to be guilty in Joan proved him to be guilty to act as constituents. That

such a constituent structure might be required is demonstrated by sentences like
(12) and (13) involving antecedent-contained deletion (Jacobson 1992:156–162):
(12) Joan believes everyone that Fred does
(13) Joan believes everyone that Fred does to be a genius
In (12), the appropriate interpretation of the pro-verb does is the transitive
predicate believes, i.e. (12) is interpreted as Joan believes everyone that Fred
believes. Analogously, the pro-verb in (13) must be interpreted as the complex
transitive predicate believes to be a genius, which is simply achieved if believes
to be a genius is a constituent. In the wrap kind of analysis, first proposed by
Bach (1979, 1980) and subsequently developed by Dowty (1982), Hoeksema
(1984), Pollard (1984) and Jacobson (1983, 1987, 1992) among others, verbs like
believe and prove are first combined with their infinitival complements to form
a complex transitive verb phrase. The complex transitive verb phrase is subse-
quently combined with its direct object by an operation which “wraps” the
complex transitive verb phrase around the direct object, creating a discontinuous
constituent. Schematically, this might be represented as follows, with heavy lines
representing the two parts of the “wrapped” category:
(14) S
NP VP
Joan VP
V NP VP[INF]

proved to be guilty
Variants of the complex predicate idea which do not directly involve a wrap
operation combine believes to be a genius first with the object to the left, giving
him believes to be a genius, then move the matrix verb to the left of the object
(Jacobson 1987), or combine believes to be a genius first with the object to the
right, giving believes to be a genius him, then subsequently “flattening” the
structure to give the correct word order with the object after the verb (Keenan 1987).
In addition to solving the problem of antecedent contained deletion in

examples such as (13), the wrap idea elegantly accounts for many aspects of the
accusative-and-infinitive construction. In particular:
(a) The accusative noun phrase is the syntactic object of a transitive verb
phrase, and hence in the appropriate case without the necessity for special
rules such as exceptional case marking, or overt or covert raising.
(b) The accusative noun phrase is structurally higher than the infinitival without
the operation of any raising rule, accounting immediately for the anaphor-
binding and scope relations between them.
(c) The adjacency of the object to the matrix verb is a direct consequence of
the wrap operation. Adverbs such as fervently in (6) or the phrasal verb
particle out in (8) will straightforwardly follow the object.
(d) Combining the complex transitive verb phrase directly with the object to the
right, rather than wrapping it around the object, results in “Heavy NP Shift”
examples such as Joan believes to be a genius that professor who gave the
lecture on gravitation. The existence of Heavy NP Shift can then be
interpreted as a block against wrapping a lighter constituent around a
heavier one.
In many respects, the wrap type of analysis is more like the separate complement
type of analysis than the clausal complement type of analysis. It straight forward-
ly accounts for the object-like properties of the accusative noun phrase, and treats
the accusative noun phrase and the infinitival as separate complements of the
matrix verb. Equally, a meaning postulate is required if it is desired to handle the
semantics of the construction in such a way that the accusative object is inter-
preted as a subject of the infinitival. The wrap type of analysis, however, also
accounts for the discontinuous constituency of the matrix verb and the infinitival,
realised in the antecedent contained deletion and Heavy NP Shift constructions,
and provides a neat statement of the forced adjacency of the accusative noun
phrase to the matrix verb.
2. Basic Categorial Grammar
As is customary in Categorial Grammar, we begin with a set of basic categories.

For the first sections of the present article, those in (15) will suffice:
194 JOHN PAYNE
(15) NP[CASE, PERSON, NUMBER]

S[TENSE]
PP[PREP]
VP[VFORM]
The symbols CASE, PERSON, NUMBER, TENSE, VFORM range over finite
sets of features as follows:
CASE: SUB, OBJ
PERSON: 1, 2, 3
NUMBER: SG, PL
TENSE: PAST, PRES
PREP: TO
VFORM: INF, BSE, PSP
The finite set of basic categories is then used to generate an infinite set of
derived categories, using rule (16):
(16) If A and B are categories (basic or derived), then A/B and A\B
are categories.
For example, the following are all possible categories:
(17) NP[SUB, 3, SG]
S[PRES]\NP[SUB, 3, SG]
(S[PRES]\NP[SUB, 3, SG])/NP[OBJ, 3, SG]
VP[INF]/NP[OBJ, 3, SG]
PP[TO]/NP[OBJ, 1, SG]
The most straightforward and fundamental way in which categories combine is
by the rules of functional application (18):
(18) X/Y Y ⇒ X (Forward Functional Application)
Y X\Y ⇒ X (Backward Functional Application)
In these rules, the categories X/Y and X\Y are functor categories, the category
Y is an argument category, and category X is a result category. In the case of
forward functional application, an expression belonging to the functor category
X/Y combines with an expression belonging to the argument category Y to its
right, and the result is an expression belonging to category X. Similarly, in the
case of backward functional application, an expression belonging to the functor
category X\Y combines with an expression belonging to the argument category
Y to its left, and the result is equally an expression belonging to category X. The
application of the functional application rules can be represented simply in tree
diagram form:
(19) Forward Functional Application Backward Functional Application
X X
X/Y Y Y X\Y
functor argument argument functor
Given these rules, the simplest way of representing a transitive sentence like
Joan believes me would be as in (20):
(20) S[PRES]
NP[SUB,3 SG]
, S[PRES]\NP[SUB,3 SG]
,
(S[PRES]\NP[SUB,3 SG]
, )/NP[OBJ,1 SG]
, NP[OBJ,1 SG]
,
Joan believes me
In this representation, believes combines by forward functional application with

its object me, the result being a category which combines by backward functional
application with the subject Joan to give a sentence. The representation in (20)
neatly captures the idea that a transitive verb combines first with its object, then
with its subject. Nevertheless, this simple analysis is open to improvement.
First, and most simply, it is clear that English transitive verbs do not agree
with their objects. The inclusion of object person and number features in the
category of believe is therefore otiose. Using the idea of unification (Shieber
1986), we can simply omit these features and allow the verb to belong to the
following category: (S[PRES]\NP[SUB, 3, SG])/NP[OBJ]. The functional application
rules require not identity, but compatibility of features.
196 JOHN PAYNE
Secondly, it is useful to treat subjects as belonging not to the basic category

NP[SUB], but rather to the type-raised category S/(S\NP[SUB]). Given a string
consisting of the argument Y and the functor X\Y, type-raising converts the
argument Y into a functor X/(X\Y) which takes the original functor X\Y as
argument. Its effect is to reverse the functor-argument structure (without
however altering the word-order). Similarly, given a string consisting of the
functor X/Y and the argument Y, type-raising converts the argument Y into the
functor X\(X/Y). In the case of a string consisting of a subject and verb-phrase,
instead of the subject being the argument of the verb-phrase, the effect is to
make the subject into the category S/(S\NP[SUB]) which takes the finite verb-
phrase of category S\NP[SUB] as an argument. The representation of the transitive
sentence Joan believes me then becomes (21):
(21) S[PRES]
S[X]/(S[X]\NP[SUB,3 SG]
, ) S[PRES]\NP[SUB,3 SG]
,
(S[PRES]\NP[SUB,3 SG]
, )/NP[OBJ] NP[OBJ,1 SG]
,
Joan believes me
The representation S[X]/(S[X]\NP[SUB, 3, SG]) for the type-raised subject, where X

ranges over feature values, ensures that the tense feature PRES of the verb
phrase is carried up to become a feature of the sentence.
As pointed out by Steedman (1985) and Dowty (1988), the type-raised
category of the subject, in conjunction with rules of functional composition,
allows for a neat treatment of right-node-raising constructions such as (22):
(22) Joan believes, but Hilary distrusts, my story
In functional composition, two functors combine to form a new complex functor:
(23) X/Y Y/Z ⇒ X/Z (Forward Functional Composition)
Y\Z X\Y ⇒ X\Z (Backward Functional Composition)
In tree diagram format, we might show the results of these combination rules as
follows, using dotted lines rather than the unbroken lines of functional application:
(24) Forward Functional Composition Backward Functional Composition
X/Z X\Z
X/Y Y/Z Y\Z X\Y
functor functor functor functor
In the case of example (22), the type-raised subject Joan of category

S/(S\NP[SUB, 3, SG]) combines with the transitive verb believes of category
(S[PRES]\NP[SUB, 3, SG])/NP[OBJ].to form a new functor Joan believes of category
S[PRES]/NP[OBJ]. Similarly, the subject Hilary combines with the transitive verb
mistrust to form a new functor Hilary mistrusts. These two new functors, Joan
believes and Hilary mistrusts, being of the same category, can be conjoined by
the conjunction but. The result, likewise a functor of category S[PRES]/NP[OBJ],
then combines by simple functional application with the object my story:
(25) S[PRES]
S[PRES]/NP[OBJ] NP[OBJ,3 SG]

,
S[PRES]/NP[OBJ] but S[PRES]/NP[OBJ] my story
S[X]/(S[X]\NP[SUB,3 SG]
, ) (S[PRES]\NP[SUB,3 SG]
, )/NP[OBJ] S[X]/(S[X]\NP[SUB,3 SG]
, ) (S[PRES]\NP[SUB,3 SG]
, )/NP[OBJ]
Joan believes Hilary mistrusts
Tense features of S and person/number features of NP will henceforth be omitted

for simplicity.
3. Wrap and Infix
In previous categorial grammar analyses of the English accusative-and-infinitive

construction which involve a wrap operation, it is assumed that the verb com-
bines first with the infinitival by forward functional application, then the
resulting complex verb-phrase functor wraps round the object, placing the object
adjacent to and to the right of the verb. We will symbolise “right-wrap” by /W,
198 JOHN PAYNE
to distinguish it from standard forward functional application. A full definition of

right-wrap could be made either to place the argument linearly immediately
following the first word or immediately following the lexical head of the functor
(the relevant head being the head which takes the object as argument). Empirical-
ly, a definition using the notion of lexical head seems preferable, since we do not
want a verb-phrase like fervently believes to be a genius to combine with its
object by placing it to the immediate right of the adverb fervently rather than to
the immediate right of the verb believes. There may also be subtle theoretical
advantages to choosing the notion of lexical head rather than first word (see
Jacobson 1992 for discussion), though there are complications with coordination
(see Section 5). In particular, when two transitive verbs are coordinated, the wrap
operation must place the object following the rightmost of the coordinated heads,
giving for example believed and assumed him to be a genius rather than *believed
him and assumed to be a genius. For present purposes, we will assume that this
problem can be resolved, and a definition of wrap based on the notion of head
will be used.
Let concatenation be represented by •, and let a1 •a2,.•.am • h • b1 •b2,.•.bn
(m, n≥0) be a string of lexical items which belong to category X/WY with head
h, then right-wrap can be stated as follows:
(26) X/WY Y ⇒ X, where X = a1 •a2,.•.am • h • Y • b1 •b2,.•.bn
(Right Wrap)
The right wrap operation has obvious affinities to the basic operation of
functional application: the functor-argument structure is identical, but the linear
order of the elements of the resulting category in right-wrap involves a discontinuity.
Using right-wrap, the verb believe in the accusative-and-infinitive construc-
tion then belongs to the category ((S\NP[SUB])/WNP[OBJ])/VP[INF]. This yields the
following representation of the sentence Joan believes him to be a genius, with
bold lines as before indicating the wrapped functor:
(27) S
S/(S\NP[SUB]) S\NP[SUB]
(S\NP[SUB])/WNP[OBJ]
Joan
((S\NP[SUB])/WNP[OBJ])/VP[INF] NP[OBJ] VP[INF]
The representation in (27) has all the basic advantages of the wrap analysis
outlined in Section (1c). In particular, the object properties of the pronoun him
(adjacency to the verb, accusative case, scope over the infinitival) emerge straightfor-
wardly, as does the constituency of the discontinous string believe to be a genius.
Nevertheless, in this paper a variation on the standard categorial analysis
with right-wrap will be explored. It involves the combination of the basic
wrapping idea with the proposal of Dowty (1988) that objects in English, like
subjects, should be type-raised to functor status. In this case, instead of the
transitive verb-phrase functor wrapping itself around its object argument, the
object as functor will infix itself into its transitive verb-phrase argument. Infixing
operations, symbolized in their most general form by ↓, are discussed in
Moortgat (1988). The operation required for English objects will be infixation of
the object to the immediate right of the lexical head of the verb phrase. Moortgat
suggests a specialised symbol < for a similar type of infixation based on the
notion of first word rather than lexical head, but since there appears to be no other
type of infixation in English, we will use the distinctive downward-pointing arrow.
As before, let concatenation be represented by •, and let a1 •a2,.•.am • h • b1
•b2,.•.bn (m, n≥0) be a string headed by h of lexical items which belong to
category Y, then right-infixation can be defined as follows:
(28) Y X↓Y ⇒ X, where X = a1 •a2,.•.am • h •X↓Y• b1
•b2,.•.bn (m, n≥0) (Right Infixation)
The category of finite transitive verb-phrase in English will then be
(S\NP[SUB])/NP[OBJ], and the type-raised category of the objects of such verb-
phrases will be (S\NP[SUB])↓((S\NP[SUB])/NP[OBJ]).
200 JOHN PAYNE
The representation of Joan believes him to be a genius is then as in (26),

with the infixed functor this time being marked by a heavy line:
(29) S
(S\NP[SUBJ])/NP[OBJ]
Joan
((S\NP[SUB])/NP[OBJ])/VP[INF] (S\NP[SUB])¯ ((S\NP[SUB])/NP[OBJ]) VP[INF]
The basic conceptual advantage of the revised representation in (29), apart from
the parallel treatment of subjects and direct objects as functors, is the fact that
the most striking syntactic property of direct objects in English, namely their
forced adjacency to the verb, is represented in the category of direct objects
themselves, rather than in the category of verbs. In other words, it is being
claimed that the adjacency property is essentially a property of the noun phrase
rather than of the verb. The consequences of this shift of perspective will be
explored in the following sections.
4. Heavy NP Shift
The standard wrap analysis of direct objects, from Bach (1979) to Jacobson
(1992), assumes that transitive verb phrases belong to two categories: there is a
redundancy rule to the effect that any verb-phrase in category
(S\NP[SUB])/WNP[OBJ] also belongs in category (S\NP[SUB])/NP[OBJ]. However, the
infixation idea allows us to have a single category (S\NP[SUB])/NP[OBJ] for
transitive verb phrases and to represent “heaviness” where it conceptually
belongs, namely in the category of the object. To be precise, all objects of finite
transitive verbs without exception belong to the “infix” category
(S\NP[SUB])↓((S\NP[SUB])/NP[OBJ]), but “heavy” objects also belong to the
category (S\NP[SUB])\((S\NP[SUB])/NP[OBJ]) and can combine with the verb phrase
by straightforward backward functional application. The pair of sentences (30a)
and (30b), respectively with and without the shift of a heavy NP, can then be
represented as in (31):
(30) a. Joan believes to be a genius that professor who gave a lecture
on gravitation
b. Joan believes that professor who gave a lecture on gravitation
to be a genius
(31) a. S
(S\NP[SUB])/NP[OBJ]
Joan
((S\NP[SUB])/NP[OBJ])/VP[INF] VP[INF] (S\NP[SUB])\((S\NP[SUB])/NP[OBJ])
believes to be a genius that professor who gave a

lecture on gravitation
(26) b. S
(S\NP[SUB])/NP[OBJ]
Joan
((S\NP[SUB])/NP[OBJ])/VP[INF] (S\NP[SUB])¯ ((S\NP[SUB])/NP[OBJ]) VP[INF]
believes that professor who gave a to be a genius

The definition of “heaviness” in Heavy NP Shift is complex, and whether the

shifted or non-shifted variant is preferable seems essentially to be a stylistic
judgement. However, NPs such as unstressed pronouns can only belong to the
unshifted category. For example, the dummy pronouns it and there, which are
always unstressed, cannot undergo Heavy NP Shift:
(32) a. Joan believes it to be raining
b. *Joan believes to be raining it
(33) a. Joan believes there to be no solution
b. *Joan believes to be no solution there
202 JOHN PAYNE
This seems to be a grammatical rather than a stylistic constraint, so such

pronouns must be grammatically excluded from the “heavy” object category
(S\NP[SUB])\((S\NP[SUB])/NP[OBJ]. This is straightforward in a framework in which
the category of the object rather than the category of the verb reflects whether
Heavy NP Shift takes place.
As in the standard wrap analysis, the contrast between NP objects and sentential
complements with respect to adverbial position also follows straightforwardly:
(34) a. Joan believes the professor fervently to be a genius
b. *Joan believes fervently the professor to be a genius
c. Joan believes fervently that the professor is a genius
Assume that the adverbial fervently belongs to the category (((S\NP[SUB])/X)/Y)\
(((S\NP[SUB])/X)/Y), where X and Y are variables representing any category
(including zero). This allows the adverbial to combine by backward functional
application with any finite verb phrase of up to three arguments, and will suffice
for the examples in the text. Then in (34a) the object NP the professor is
straightforwardly infixed into the verb phrase believes fervently to be a genius:
(35) S
Joan
(S\NP[SUB])/NP[OBJ]
((S\NP[SUB])/NP[OBJ])/VP[INF]
((S\NP[SUB])/NP[OBJ])/VP[INF] (S\NP[SUB])¯ ((S\NP[SUB])/NP[OBJ]) (((S\NP[SUB])/X)/Y)\(((S\NP[SUB])/X/Y) VP[INF]
believes the professor fervently to be a genius
Infixation after the string believes fervently is excluded by definition, since this
string consists of two words. Hence (34b) is ungrammatical.
By contrast, sentential complements do not involve an infixing category:
they simply belong to the category S, and, assuming that believes also belongs
to category (S\NP[SUB])/S, the grammaticality of (34c) follows:
(36) S
Joan ((S\NP[SUB])/S
((S\NP[SUB])/S) ((S\NP[SUB])/X)/Y)\((S\NP[SUB])/X/Y) S
believes fervently that the professor

is a genius
The only example in this set to permit Heavy NP Shift is (37), in which the
object follows the entire transitive verb phrase:
(37) Joan believes fervently to be a genius that professor who gave a
This is the shifted analogue of (34a), and has the following analysis:
(38) S
Joan
(S\NP[SUB])/NP[OBJ]
((S\NP[SUB])/NP[OBJ])/VP[INF]
((S\NP[SUB])/NP[OBJ])VP[INF] ((S\NP[SUB])/X)/Y)\((S\NP[SUB])/X/Y) VP[INF] (S\NP[SUB])\((S\NP[SUB])/NP[OBJ])
believes fervently to be a genius that professor who gave

a lecture on gravitation
Given that ordinary transitive verb-phrases of category (S\NP[SUB])/NP[OBJ] permit

both infixed and heavy objects of category (S\NP[SUB])↓((S\NP[SUB])/NP[OBJ]) and
(S\NP[SUB])\((S\NP[SUB])/NP[OBJ]) respectively, the original right wrap operation
can be used restrictively for those transitive verb-phrases whose objects do not
permit Heavy NP Shift. It has been well-known since Postal (1974:138) that
double object verbs do not permit Heavy NP Shift of the dative (recipient/
beneficiary) object:
204 JOHN PAYNE
(40) a. Joan gave the professor the manuscript

b. *Joan gave the manuscript the professor who gave a lecture on
gravitation
This restriction can now plausibly be represented as an idiosyncratic property of
this class of verb. Double object verbs will be analysed as belonging to the
category ((S\NP[SUB])/WNP[OBJ])/NP[OBJ]), forcing a wrap around the dative object.
In order for this to work with type-raised categories for objects, the dative object
must be given the matching category (S\NP[SUB])↓((S\NP[SUB])/WNP[OBJ]). Such
a category is indeed proposed by Jacobson (1992:146) as the general result of
(syntactically) type-raising the objects of transitive verb phrases in a wrap
analysis. In the analysis proposed here, though, only the dative object has this
category: the second (theme) object retains a basic infix category ((S\NP[SUB])/
WNP[OBJ])↓(((S\NP[SUB])/WNP[OBJ])/NP[OBJ]), and neither object category is
produced by a freely applicable syntactic type-raising operation.
Given these categories, (40a) can be represented as follows:
(41) S
S/(S\NP[SUBJ]) S\NP[SUB]
Joan
((S\NP[SUB])/WNP[OBJ]/NP[OBJ] (S\NP[SUB])¯ ((S\NP[SUB]/WNP[OBJ]) ((S\NP[SUB])/WNP[OBJ])¯ (((S\NP[SUB])/WNP[OBJ]/NP[OBJ])
gave the professor the manuscript
These categories neatly prevent the dative object from undergoing Heavy NP
Shift, since a heavy NP of of category (S\NP[SUB])\((S\NP[SUB])/NP[OBJ]) must
combine backwards by functional application with a verb-phrase of category
((S\NP[SUB])/NP[OBJ]) rather than the wrap category ((S\NP[SUB])/WNP[OBJ]) .
Some simplification is possible in the representation of these categories
given a unification schema. Non-dative objects have the general form X↓(X/NP),
where X represents any category. Dative objects have the somewhat similar form
X↓(X/WNP), while heavy objects have the form X\(X/NP). Subjects have the
form X/(X\NP), and are thereby like dative NPs automatically excluded from the
process of Heavy NP Shift.
It should be noted that it now follows from the categories given for the
double object construction that Heavy NP Shift of a dative NP to the right of an
adverbial will be impossible under any circumstances. Thus, even when the
theme NP itself is heavy and there is therefore no infelicity caused by a heavy
dative NP preceding a light theme NP, nevertheless no movement of the dative
NP is possible over an adverbial:
(42) *Joan handed reluctantly the professor who gave a lecture on
gravitation the manuscript she had been working on for five years
As pointed out by Postal (1974:138), this phenomenon is directly inconsistent
with any formulation which simply makes reference to the heaviness of the
dative NP. In the present framework, dative NPs are forced by the infix opera-
tion to be directly adjacent to the verb, and cannot be combined in a position
following the adverbial. Equally, there is no possibility of Heavy NP Shift, since
the wrap operator introducing dative objects in the category of double object
verbs is incompatible with the category of heavy objects.
5. Coordination
The advantages of categorial grammars in the analysis of coordinate structures

are well-known: in particular, various kinds of coordination involving apparent
non-constituents receive natural treatments using functional composition, one
particular case being the right-node raising construction exemplified in (22)
above. Some authors have however been pessimistic about the possibility of
incorporating wrap and similar non-linear operations into such a framework (see
Steedman 1996:46 for discussion and references). The immediate problem raised
by an infix operation in the analysis of coordinate structures has already been
mentioned in Section 3: if infixation is defined to place the object immediately
after the relevant verbal head, then in coordinate constructions involving two or
more verbs, the rightmost verb must be designated as the relevant head. Such a
designation permits (43a), and correctly blocks (43b):
(43) a. Joan believes, but does not like the professor
b. *Joan believes the professor, but does not like
Equally, it is not possible for a string such as believes, but does not like to have
no head designated at all, otherwise (43a) will be underivable. For the purposes
of this paper, we will simply assume that an appropriate mechanism for designat-
ing heads can be provided (perhaps along the lines of Jacobson 1992, where
206 JOHN PAYNE
individual categorial combination rules contain information about which element

in a pair of combined categories is head). It is interesting that the various
coordination possibilities of the English accusative-and-infinitive construction
then emerge from the infix analysis in a relatively straightforward manner.
Consider first a coordinate structure such as that in (44):
(44) Joan believes and assumes the professor to be a genius
This involves coordination of believes and assumes, both of category
((S\NP[SUB])/NP[OBJ])/VP[INF]. For concreteness, we will assume that coordinators
such as and and but belong to the polymorphic category (X\X)/X, where X is
any conjoinable type (Oehrle 1991). Then the representation of the string believes
and assumes (with assumes as the designated head) is (45), and (44) has the
overall representation in (46):
(45) ((S\NP[SUB])/NP[OBJ])/VP[INF]
((S\NP[SUB])/NP[OBJ])/VP[INF] (((S\NP[SUB])/NP[OBJ])/VP[INF])\(((S\NP[SUB])/NP[OBJ])/VP[INF])
believes (X\X)/X ((S\NP[SUB])/NP[OBJ])/VP[INF]
and assumes
(46) S
X/(X\NP[SUB]) S\NP[SUB]
(S\NP[SUB])/NP[OBJ]
Joan
((S\NP[SUB])/NP[OBJ])/VP[INF] X¯ (X/NP[OBJ]) VP[INF]
believes and assumes the professor to be a genius
A construction such as (47) straightforwardly involves coordination of the strings

believes to be a genius and also believes to be mad, both of category (S\NP[SUB])/
NP[OBJ], with a Heavy NP in the right peripheral position:
(47) S
(S\NP[SUB])/NP[OBJ]
Joan X\(X/NP[OBJ])
believes to be a genius that professor who gave

but also believes to be mad a lecture on gravitation
By contrast, an attempt at right-node raising in a construction such as (48) is

predictably awkward:
(48) ?Joan believes, and Fred assumes, the professor to be a genius
In the infix analysis, the string the professor to be a genius is not a constituent,
and neither can Joan combine with believes (nor Fred with assumes) by standard
first-order functional composition. Sentence (48) is derivable, however, with the
addition of a second-order composition rule of the following type:
(49) X/Y (Y/Z)/W ⇒ (X/Z)/W
(Second-order Forward Functional Composition)
This rule would permit Joan to combine with believes as follows:
(50) (S/NP[OBJ])/VP[INF]
X/(X\NP[SUB]) ((S\NP[SUB])/NP[OBJ])/VP[INF]
Joan believes
The strings Joan believes and Fred assumes both then belong to category
(S/NP[OBJ])/VP[INF], and can be conjoined into the string Joan believes and Fred
assumes of the same category, with designated head assumes. It is then possible
to derive (51):
208 JOHN PAYNE
(51) S
S/NP[OBJ]
(S/NP[OBJ])/VP[INF]
X¯ (X/NP[OBJ]) VP[INF]
Joan believes and Fred assumes the professor to be a genius
We assume the awkwardness of (48), therefore, to be explicable by the need for

second-order, rather than first-order composition.
A natural extension of first-order composition to infix categories will further
allow the combination of the verb and object into a single constituent. This will
be represented by a dotted line for the verb and a heavy line for the infixed object:
(52) (S\NP[SUB])/VP[INF]
((S\NP[SUB])/NP[OBJ])/VP[INF] X¯ (X/NP[OBJ])
believes the professor
As before, let concatenation be represented by •, and let a1 •a2,.•.am • h • b1

•b2,.•.bn (m, n≥0) be a string headed by h of lexical items which belong to
category Y/Z, then the necessary rule, infixing functional composition, can be
defined as follows:
(53) Y/Z X↓Y ⇒ X/Z, where X/Z ⇒ a1 •a2,.•.am • h •X↓Y• b1 •b2,.•.bn
(m, n≥0) (Infixing Composition)
Just like the basic infixation rule, this rule will place the object to the right of the
verb, but the use of composition delays combination with the infinitival VP to
the right to a subsequent operation. It should be noted that infix composition
does not lead to permutation of the object and the VP, and is therefore not
susceptible to the same objections as crossing combination rules (such as Y/Z
X\Y ⇒ X/Z), which lead to a collapse into free order grammars.
Given infix composition, it is then a simple matter to derive coordinate
structures such as those in (54):
(54) Joan believes the professor, and also believes the dean, to be a genius
Assuming that also believes the dean can be constructed as (S\NP[SUB])/VP[INF] in
a similar way to believes the professor, (54) has the structure in (55):
(55) S
(S\NP[SUB])/VP[INF] VP[INF]
Joan believes the professor, to be a genius

and also believes the dean
Similarly, by using forward functional composition to combine Joan with

believes the professor into S/VP[INF], we can also straighforwardly derive (56):
(56) Joan believes the professor, and Mary believes the dean, to be a
genius
In fact, the only type of coordinate structure which is not now immediately
derivable is illustrated by the so-called “non-constituent coordination” or
“argument cluster coordination” in (57), where it appears that the object and VP
complements of the verb must somehow form a constituent (see Steedman
1996:45–46 for an interesting discussion):
(57) Joan believes [the professor to be a genius], and [the dean to be mad]
One way in principle of creating such constituents, following the proposal
initially made by Dowty (1988), is to type-raise all arguments, including the
infinitival VP. It might then be possible to combine the object and infinitival VP
by functional composition. It is doubtful, however, that such an analysis can be
extended to cases in which one of the functors is an infix: an infixing object is
a functor looking for a head verb to infix to the right of, and it is dubious that
a rule could exist which allowed such a functor to combine with a constituent
which does not include that head. The infix analysis of objects therefore seems
to force the use of an alternative solution to argument cluster coordination such
as that pioneered by Oehrle (1988) and Wood (1988), a solution involving the
use of the product or “dot” operator.
210 JOHN PAYNE
Product categories are introduced by the following rule:

(58) If A and B are categories, then A•B is a category
In effect, this allows the concatenation of any two categories A and B to be a
category A•B. In (57), the string the professor to be a genius would thus belong
to category (X↓(X/NP[OBJ]))•VP[INF]. Coordination involving this category then
might allow (57) to be derived, with some complexity. One peculiarity of the
construction (a peculiarity significantly shared by gapping examples) is the fact
that the first product constituent in the coordination appears to be syntactically
integrated into the string to its left, forming a complete clause, while the second
(and subsequent) product categories are stranded concatenations of otherwise
non-combinable categories. This can be seen in an example like (59), where the
object of the first conjunct the professor must be infixed next to the verb:
(59) Joan believes [the professor] fervently [to be a genius], and [the
dean • to be mad]
Compare this with a gapping example such as (60), where the product constituent
may consist of a subject and object, separated by the verb in the first conjunct:
(60) [Joan] believes [the professor] and [Mary • the dean]
The complexities which the gapping construction is known to involve are
discussed in Oehrle (1987). Examples such as (59) suggest that the argument
cluster coordination construction may be comparable in complexity, and it is
therefore an interesting property of the system outlined here that special treat-
ment involving more than composition rules is necessary.
6. Extraction
For extraction constructions, we will use a “gap-forming” operator of the type

initially proposed by Moortgat (1988). As pointed out by Oehrle (1991), gap-
forming operators have clear affinities to the SLASH category of Generalised
Phrase-Structure Grammar (Gazdar, Klein, Pullum and Sag 1985). The operator,
which will be symbolised by ↑, is introduced by a rule of the following type:
(61) X op Y ⇒ X↑Z (Gap Introduction)
Here op (= operator) must include at least \, /, and for many speakers /W. Gap
categories derived from ↓ have no use in English, but no harm appears to be
done if ↓ is also included in the scope of the rule.
A constituent of category X↑Z is a constituent of category X with a
constituent of category Z extracted from it. The gap operator is not eliminated by
an application rule of the type X↑Z Z ⇒ X. The difficulties associated with
such a rule are pointed out by Moortgat (1988): information about the linear
position of Z in X is lost when ↑ is introduced. However, a constituent of
category X↑Z can be eliminated as an unanalysed whole by functional applica-
tion using a category of the form Z/(X↑Z). The relative and interrogative
pronouns involved in extraction constructions will belong to categories of this form.
As a simple example, let us take Q (= S[INT]) to be the category of interrog-
ative clauses, and Q/(S↑NP[OBJ]) to be the category of interrogative object who
(the corresponding category for subject who will be Q/(S↑NP[SUB])). Then in
object extraction, the derivation might proceed as follows:
(62) S
X/(X\NP[SUB]) S\NP[SUBJ]
(S\NP[SUB])/Q Q
Q/(S NP[OBJ]) S NP[OBJ]

gap introduction
S/NP[OBJ]
X/(X\NP[SUB]) S\NP[SUB]/NP[OBJ]
I wonder who Joan believes
Joan combines with believes by functional composition to give S/NP[OBJ], and

the object gap is then introduced, producing the category S↑NP[OBJ]. The object
interrogative pronoun is then able to combine with this gap category by function-
al application, yielding an interrogative clause of category Q.
This analysis obviously extends straight forwardly to the extraction of the
NP object in the accusative-and-infinitive construction, since believes to be a
genius in Joan believes the professor to be a genius is of the same category as
believes in Joan believes the professor:
212 JOHN PAYNE
(63) S
X/(X\NP[SUB]) S\NP[SUBJ]
(S\NP[SUB])/Q Q

gap introduction
S/NP[OBJ]
X/(X\NP[SUB]) (S\NP[SUB])/NP[OBJ]
I wonder who Joan ((S\NP[SUB])/NP[OBJ])/VP[INF] VP[INF]
believes to be a genius
A number of accusative-and-infinitive verbs (examples in my speech are say and

think) seem to permit extraction even though they do not permit the basic
accusative-and-infinitive construction(except perhaps with heavy stress on the
object, see Postal 1974:305):
(64) a. *Joan said the professor to be responsible
b. I wonder who Joan said to be responsible
A natural assumption here is that these verbs are lexically specified as belonging
to the gap category ((S\NP[SUB])↑NP[OBJ])/VP[INF]. This implies the existence of
composition rules for the gap operator analogous to the standard forward and
backward composition rules (see Oehrle 1991 for a similar assumption):
(65) (X/Y) (Y↑Z) ⇒ X↑Z (Forward Gap Composition)
(Y↑Z) (X\Y) ⇒ X↑Z (Backward Gap Composition)
With forward gap composition, (64b) can then be derived as follows:
(66) S
(S\NP[SUB])/Q Q
X/(X\NP[SUB]) (S\NP[SUB]) NP[OBJ]
((S\NP[SUB]) NP[OBJ])/VP[INF] VP[INF]

I wonder who Joan
said to be responsible
Should perhaps all gaps initiate in the lexicon, rather than through free syntactic
application of the gap introduction rule? This would mean that believe belonged
to both gap and non-gap categories in the lexicon, while say belonged only to the
gap category. We will not resolve this issue here, noting only that the lexical
version of gap introduction requires also a gap counterpart of the infix composition
rule (see (70) below). For the purposes of this paper, the syntactic version of gap
introduction will continue to be used, together with the assumption that the introduc-
tion of gaps in the lexicon is a marked property of a very limited set of verbs.
It is interesting to observe that, in conjunction with the infix analysis of
objects, the proposed rules give a straightforward account of so-called “non-
peripheral extraction” in examples like (67), avoiding the necessity for second-
order composition rules as in the analysis of Steedman (1985, 1996):
(67) I wonder who Joan gave the manuscript
Because of the infixing nature of objects, the recipient argument of gave in (67)
is no longer peripheral, and can be easily extracted (as long as gap-introduction
is allowed to apply to right-wrap, an operation which may be barred for some
speakers):
214 JOHN PAYNE
(68) S
(S\NP[SUB])/Q Q
X/(X\NP[SUB]) (S\NP[SUB]) NP[OBJ] gap introduction

I wonder who Joan
((S\NP[SUB])/WNP[OBJ])/NP[OBJ] X ¯ (X/NP[OBJ])
gave the manuscript
The most internal argument is instead the theme argument, and this can be
extracted by using infix composition to combine the double-object verb with the
recipient argument:
(69) S
(S\NP[SUB])/Q Q
X/(X\NP[SUB]) (S\NP[SUB]) NP[OBJ] gap introduction

(S\NP[SUB])/NP[OBJ]
I wonder what Joan
((S\NP[SUB]/W)NP[OBJ])/NP[OBJ] X ¯ (X/WNP[OBJ])
gave the professor
Precisely the same device can be used to extract an object from within the
infinitival VP of the accusative-and-infinitive construction, as in (70):
(70) I wonder what Joan believes the professor to like
(71) S
(S\NP[SUB])/Q Q

gap introduction
S/NP[OBJ]
X/(X\NP[SUB]) (S\NP[SUB])/NP[OBJ]
I wonder what Joan

((S\NP[SUB])/NP[OBJ])/NP[OBJ]
((S\NP[SUB])/NP[OBJ])/VP[INF] VP[INF]/NP[OBJ]
X¯ (X/NP[OBJ]) VP[INF]/VP[BSE] VP[BSE]/NP[OBJ]
believes the professor to like
It should be noted that using a lexical version of gap introduction in this case
would require a (quite reasonable) rule like (71) combining infixing and gap
composition. As before, let concatenation be represented by •, and let a1 •a2,.•.am
• h • b1 •b2,.•.bn (m, n≥0) be a string headed by h of lexical items which belong
to category Y↑Z, then the necessary rule can be defined as follows:
(72) Y↑Z X↓Y ⇒ X↑Z, where X↑Z = a1 •a2,.•.am • h •X↓Y• b1
•b2,.•.bn (m, n≥0) (Infix Gap Composition)
The present set of rules predict that extraction of any element from within a
subject or a heavy NP is excluded:
(73) a. *I wonder what subject [X/(X\NP) the professor of] will become
dean
b. *I wonder what subject Joan believes to be a genius [X\(X/NP)
the professor of]
Constructions of this type would require rules of the following form, allowing
gaps in forward or backward-looking functors to be carried forward (rules of this
nature are called “transparency rules” in Oehrle 1991):
(74) (X/Y)↑Z Y ⇒ X↑Z
Y (X\Y)↑Z ⇒ X↑Z
216 JOHN PAYNE
The non-existence of rules of this kind simultaneously accounts for the ban
against extractions from adjuncts.
If such rules are indeed barred in English, we can see a further advantage
in maintaining the VP argument in the accusative-and-infinitive construction as
a non-type-raised category: if it were a backward-looking functor, the possibility
of extraction from the VP would not exist. On the other hand, extraction from
within an object is somewhat more acceptable (though not without its difficul-
ties: early accounts of extraction constraints proposed that no extraction at all
from within NP should be possible). An example of extraction from within the
object NP of the accusative-and-infinitive construction is given in (75):
(75) ?Iwonder what subject Joan believes the professor of to be a
genius
To the extent that this extraction is possible, it requires a transparency rule for
objects. Objects are of course here analysed as infixes, and therefore they are not
automatically susceptible to the constraint against transparency for forward and
backward looking functors. What would be required to derive (74) is a transpar-
ency rule for infixes:
(76) Y (X↓Y)↑Z ⇒ X↑Z (Infix Transparency)
Taking the professor of to be of category (X↓(X/NP[OBJ])↑NP[OBJ], and given
infix transparency (which we assume, like second-order composition, to have a
somewhat marked status in English), it is possible to derive (75) as follows:
(77) S
(S\NP[SUB])/Q Q
Q/(S NP[OBJ]) S NP[OBJ] (by gap transparancy)
X/(X\NP[SUB]) (S\NP[SUB]) NP[OBJ] (by infix

transparancy)
I wonder what subject Joan
(S\NP[SUB])/NP[OBJ])
((S\NP[SUB])/NP[OBJ])/VP[INF] VP[INF]
believes (X¯ (X/NP[OBJ] ) NP[OBJ])
the professor to be a genius
To conclude this discussion of extraction, we note that all the mechanisms are
now in place to handle a verb like assure (see Kayne 1983:111, Steedman
1996:60), which take an NP object (the person assured) in addition to the NP of
the accusative-and-infinitive construction, but which like say force extraction:
(78) a. a professor who I assure you to be a genius
b. *I assure you the professor to be a genius
The category of assure is (((S\NP[SUB])/WNP[OBJ])↑NP[OBJ])/VP[INF]. This category
forces the NP representing the person assured to act like the recipient NP in a
double-object construction, blocking for example any possibility of Heavy NP Shift:
(79) *a professor who I assure to be a genius all you students who have
gathered for the lecture
The gap category of the accusative-and-infinitive object of course forces extrac-
tion of that object.
Interestingly, the existence of verbs like assure forces us to assume the
availability of the infix gap composition rule posited for the analysis of examples
like (70), under the assumption that gaps arise in the lexicon. We must therefore
definitively add this rule to the set of rules required for English, deriving (78a)
as follows:
218 JOHN PAYNE
(79) b. X/(X\NP[SUB])
(X/(X\NP[SUB]))/N N
N N\N
(N\N)/(S NP[OBJ]) S NP[OBJ] (by gap composition)
X/(X\NP[SUB]) (S\NP[SUB]) NP[OBJ] (by gap composition)
a professor who I
((S\NP[SUBJ])/WNP[OBJ]) NP[OBJ]
(((S\NP[SUB])/WNP[OBJ]) NP[OBJ]/VP[INF] X¯ (X/NP[OBJ]) VP[INF]
assure you to be a genius
Here N\N is taken as the category for relative clauses, and (N\N)/(S↑NP[OBJ])
therefore as the category of the object relative pronoun.
7. Passive
A natural categorial syntax of the passivized form of the accusative-and-infinitive

construction would take the passive verb to belong to the category VP[PSP]/VP[INF],
and the finite auxiliary correspondingly to the category (S\NP[SUB])/VP[PSP]. The
sentence The professor is believed to be a genius then has the analysis in (80):
(80) S
the professor (S\NP[SUB])/VP[PSP] VP[PSP]
is VP[PSP]/VP[INF] VP[INF]
believed to be a genius
A further natural assumption would then be to assume that the object status of
the accusative NP in the accusative-and-infinitive construction, i.e. its position in
the category of the verb as the next-to-last argument to be combined with (the
subject being the last), is the crucial factor which permits passivization to take
place. This, in effect, is the proposal of Bach, the originator of the wrap
analysis: “all and only transitive verb phrases occur in the passive” (Bach
1979:521). Just as believe belongs to category (S\NP[SUB])/NP[OBJ] of transitive
verb phrase in Joan believes the professor, so does believe to be a genius in Joan
believes the professor to be a genius. One way of instantiating this idea is to
assume a syntactic rule which creates passive verb phrases from active verb
phrases (for discussion of this proposal see Keenan 1980). Introducing such a
rule is however clearly out of line with the present lexically-based categorial
framework. A lexical approach would have to assume a lexical redundancy rule
of the form in (81), the two parts of which could of course be generalized:
(81) Category:
(i) ((S\NP[SUB])/NP[OBJ] ⇔ VP[PSP]
(ii) ((S\NP[SUB])/NP[OBJ])/VP[INF] ⇔ VP[PSP]/VP[INF]
Form: Base Base+ed
This rule would derive passive participle believed in the correct combinatory
categories directly from active believe (ignoring verb-agreement). Unfortunately,
however, despite its seductiveness, the claim implied in (81) will have to be
abandoned. The passivizability of the accusative-and-infinitive construction does
not seem to follow from the object status of the accusative NP. We consider
below the classes of verbs for which the proposed equivalence fails:
Class 1: order, allow, permit, enable, authorize
Active verbs in this class in principle permit two interpretations, one (the
“control” interpretation) in which the accusative NP bears a direct semantic
relationship with the verb, and a second (the “raising” interpretation) in which
the accusative NP is interpreted solely as subject of the infinitival VP. It is this
second interpreation which parallels the interpretation of the accusative NP with
verbs like believe. For example, with the verb order, one can either order a
person to do something, or one can order that a situation comes about, no matter
who actually carries out the order. A sentence like (82) actually has both
interpretations. Although the control interpretation is the most salient, it is
possible to imagine a scenario in which the general orders the situation rather
than the soldiers:
(82) The general ordered the soldiers to blow up the bridge at dawn
220 JOHN PAYNE
On the other hand, the raising interpretation is forced when the accusative NP is
a dummy or inanimate item:
(83) a. The general ordered there to be an attack at dawn
b. The general ordered the bridge to be blown up at dawn
Other verbs which even more predominantly have control interpretations may,
given the right context, appear in the same kind of construction, for example
oblige, force, forbid:
(84) a. The law obliges there to be an election every five years
b. The PM’s resignation will force there to be an election within
a month
c. The headmaster forbids there to be any more disturbances in
the classroom
Interestingly, the passivizability of these constructions seems to depend on the
interpretation. In the control interpretation, passivization is straightforwardly
possible. (82) has the passive (85):
(85) The soldiers were ordered to blow up the bridge at dawn
On the other hand, passives of (83) and (84) are at least very awkward and in
some cases impossible:
(86) a. ??There was ordered by the general to be an attack at dawn
b. ??The bridge was ordered by the general to be blown up at dawn
(87) a. *There is obliged by the law to be an election every five years
b. *There will be forced by the PM’s resignation to be an
election within a month
c. *There is forbidden by the headmaster to be any more distur-
bance in the classroom
For this class of verbs, then, it is not the case that ((S\NP[SUB])/NP[OBJ])/VP[INF]
⇒ VP[PSP]/VP[INF].
Class 2: cause
The verb cause straightforwardly accepts a raising interpretation, and perhaps
only accepts this interpretation, yet passivization is impossible:
(85) a. The accident caused there to be considerable delays on the

motorway
b. *There were caused by the accident to be considerable delays
on the motorway
As with the verbs in Class 1, it is not the case therefore that ((S\NP[SUB])/
NP[OBJ])/VP[INF] ⇒ VP[PSP]/VP[INF].
Class 3: want, hate, like, love, wish, desire, prefer, can’t bear
The verbs in Class 3 have long been recognized as a class which resist passiviza-
tion (see Postal 1974:179):
(89) a. Everyone wants Joan to be president
b. *Joan is wanted by everyone to be president
They are therefore prima facie equivalents to the verbs in Classes 1 and 2 as
counterexamples to the claim that ((S\NP[SUB])/NP[OBJ])/VP[INF] ⇒ VP[PSP]/VP[INF].
One argument which has been used in an attempt to avoid this conclusion
is the fact that this class of verbs permits a complement clause beginning with
the complementizer for. In British English, this is typically restricted to comple-
ment clauses which are not adjacent to the governing verb:
(90) a. I want very much [for Joan to be president]
b. *I want [for Joan to be president]
Nevertheless, the claim is made that Joan in (89a) is really the subject of the
infinitival clause, with the complementizer obligatorily omitted immediately
following the verb. An early statement of this position can be found in Bach
(1979:524). An immediate difficulty, however, is the fact that the construction
in (89a), though not the construction in (90a), otherwise mirrors the standard
accusative-and-infinitive construction in all important respects. In particular, it is
possible for the accusative NP to be infixed to the left of an adverb which
modifies the matrix verb:
(91) I want Joan desperately to be the new president
What is more, there is a class of verbs (including intend, expect, and require)
which permits the for-complement, but also permits passivization (though
significantly not when the complementizer is present):
222 JOHN PAYNE
(92) a. I expect Joan to be elected president

b. Joan is expected to be elected president
c. I expect confidently for Joan to be elected president
d. *Joan is expected confidently for to be elected president
We prefer therefore to treat the verbs in Class 3 as genuine equivalents to those
in Classes 1 and 2.
Class 4: say, think
The verbs say and think were analysed in Section 5 as belonging to the gap
category ((S\NP[SUBJ])↑NP[OBJ])/VP[INF] on the basis that they only permit an
extracted object:
(93) a. *Everyone says the professor to be a genius
b. the professor who everyone says to be a genius
These verbs do however permit passives:
(94) The professor is said to be a genius
They therefore represent counterexamples in the opposite direction to the equivalence
in (81). We do not have that VP[PSP]/VP[INF] ⇒ ((S\NP[SUB])/NP[OBJ])/VP[INF].
Class 5: rumour, repute
These two verbs only occur in the passive:
(95) a. *Everyone reputes the professor to be a genius
b. *the professor who everyone reputes to be a genius
c. The professor is reputed to be a genius
Like the verbs in Class 4, they are therefore counterexamples, though for a
different reason, to the claim that VP[PSP]/VP[INF] ⇒ ((S\NP[SUB])/NP[OBJ])/VP[INF].
Given the collapse of the equivalence in (81), and therefore the total generality
of the lexical redundancy rule relating active and passive in the accusative-and-
infinitive construction, are we therefore reduced to a separate listing of those
verb forms which belong to the active category ((S\NP[SUB])/NP[OBJ])/VP[INF] and
those which belong to the passive category VP[PSP]/VP[INF]? This is probably in
the end necessary. However, one generalization which seems to be very strong
is the following:
(96) VP[PSP]/VP[INF] ⇒ VP[PSP]/C
In this implication, C is the category of complementizer-introduced clauses, i.e.

clauses introduced by the complementizer that. In other words, given the
existence of a passive version of the accusative-and-infinitive construction, we
can nearly always predict the existence of a construction with the same passive
verb taking a complement clause:
(97) a. The professor is believed to be a genius
b. It is believed that the professor is a genius
(98) a. The professor is rumoured to be a genius
b. It is rumoured that the professor is a genius
(99) a. The professor is said to be a genius
b. It is said that the professor is a genius
A clear exception to this generalization is the verb hear, where we do appear to
have a passive with an infinitival VP, but not one with a that-clause:
(100) a. The professor was heard to enter the lecture room
b. *It was heard that the professor entered the lecture room
However, verbs of perception have their own peculiarities: for example, hear in
the active takes a VP[BSE] rather than a VP[INF]:
(101) a. Joan heard the professor enter the room
b. *Joan heard the professor to enter the room
The overall conclusion that can be drawn is that the object status of an NP as a
syntactic argument of a verb does not guarantee its passivizability. If an NP is a
syntactic argument and at the same time bears a direct semantic relation to the
verb, passivizability is almost guaranteed. However, we have argued that the
object NP in the accusative-and-infinitive construction has an identical syntactic
category and status, in the categorial system used here, to any object which does
bear a direct semantic relation to the verb. The passivizability of such objects
does not seem to be clearly related to their object status. Rather, the verbs which
permit a passive construction with VP[INF] are (almost) a subset of those which
permit a passive construction with a complementizer-introduced finite clause.
One subtle final piece of evidence in favour of this conclusion is the fact
that many accusative-and-infinitive verbs distinctly prefer their VP[INF] to denote
a property rather than an activity:
224 JOHN PAYNE
(102) a. I acknowledge Mary to be the best driver

b. ?I acknowledge Mary to drive well
This restriction does not however emerge in the passive:
(103) a. Mary is acknowledged to be the best driver
b. Mary is acknowledged to drive well
As predicted, the passive with a finite complement clause parallels (103) rather
than (102):
(104) a. It is acknowledged that Mary is the best driver
b. It is acknowledged that Mary drives well
8. Conclusion
In this paper, we have treated objects syntactically as infixed functors. As

infixes, objects are constrained to occur adjacent to their head verb. As functors,
they have interesting syntactic properties which play a significant role in the
account of Heavy NP Shift, coordination and extraction. The categorial system
which is required for this to work requires the addition of the “exotic” operators
↓, /W and ↑ to the standard / and \ of basic categorial systems. In addition, the
product operator • is required for argument cluster coordination and gapping.
Rather than derive the properties of these operators axiomatically, as in the
Lambek calculus, we have concentrated on a direct statement of the rules which
are required in a syntactic description of English. In this sense, the system
developed is more similar to the Combinatory Categorial Grammar developed by
Steedman (1996). Steedman however chooses to eschew the use of wrap-like
operations, introducing some corresponding complexity in other areas of the
system (for example the use of second-order composition and feature-restricted
crossing composition rules).
The rules which are used here can be summarized as follows (omitting
details of the linear order resulting from rules involving ↓):
Functional Application: X/Y Y ⇒ X
Y X\Y ⇒ X
Functional Composition: X/Y Y/Z ⇒ X/Z
Y\Z X\Y ⇒ X\Z
Infixation: Y X↓Y ⇒ X
Infix Composition: Y/Z X↓Y ⇒ X/Z
Gap Introduction: X/Y ⇒ X↑Z
X\Y ⇒ X↑Z
X/WY ⇒ X↑Z (for many
speakers)
Gap Composition: (X/Y) (Y↑Z) ⇒ X↑Z
(Y↑Z) (X\Y) ⇒ X↑Z
Infix Gap Composition: Y↑Z X↓Y ⇒ X↑Z
In addition, the following rules have marginal status, allowing marked coordina-
tions and extractions:
Second Order Composition: X/Y (Y/Z)/W ⇒ (X/Z)/W
Infix Transparency: Y (X↓Y)↑Z ⇒ X↑Z
References
Bach, E. 1979. “Control in Montague Grammar.” Linguistic Inquiry 10. 515–31.

Bach, E. 1980. “In Defense of Passive.” Linguistics and Philosophy 3. 297–341.
Bresnan, J. 1976. “Non-arguments for Raising.” Linguistic Inquiry 7. 485–501.
Bresnan, J. (ed). 1982. The Mental Representation of Grammatical Relations.
Cambridge, Massachusetts: MIT Press.
Chierchia, G. 1984. Topics in the Syntax of and Semantics of Infinitives and
Gerunds. PhD Dissertation. University of Massachusetts, Amherst.
Chomsky, N. 1981. Lectures on Government and Binding. Dordrecht: Foris
Publications.
Chomsky, N. 1995. The Minimalist Program. Cambridge, Massachusetts: MIT Press.
Dowty, D. 1982. “Grammatical Relations and Montague Grammar.” The Nature
of Syntactic Representation, ed. by P. Jacobson and G.K. Pullum, 79–130.
Dordrecht: Reidel.
Dowty, D. 1988. “Type-raising, Functional Composition, and Non-constituent
Coordination.” Categorial Grammars and Natural Language Structures, ed.
by R.T. Oehrle, E. Bach and D. Wheeler, 153–98. Dordrecht: Reidel.
226 JOHN PAYNE
Gazdar, G., E. Klein, G.K. Pullum and I.A. Sag. 1985. Generalized Phrase Struc-
ture Grammar. Oxford: Blackwell.
Hoeksema, J. 1984. Categorial Morphology. PhD Dissertation, Rijksuniversiteit,
Groningen. (Outstanding dissertations in linguistics series, 1985. New York:
Garland Inc.)
Jacobson, P. 1983. Phrase Structure and Grammatical Relations. Colloquium talk,
LSA Annual Meeting, Minneapolis.
Jacobson, P. 1987. “Phrase Structure, Grammatical Relations and Discontinous
Constituents.” Syntax and Semantics 20: Discontinuous Constituency, ed. by
G. Huck and A. Ojeda, 27–69. New York: Academic Press.
Jacobson, P. 1992. “Flexible Categorial Grammars: Questions and Prospects.”
Formal Grammar: Theory and Implementation, ed. by R. Levine. Vancouver
Studies in Cognitive Science, Volume 2, 129–67. New York, Oxford:
Oxford University Press.
Kang, B-M. 1995. “On the Treatment of Complex Predicates in Categorial
Grammar.” Linguistics and Philosophy 18. 61–81.
Kayne, R. 1983. Connectedness and Binary Branching. Dordrecht: Foris Publications.
Keenan, E.L. 1980. “Passive is Phrasal (not Sentential or Lexical).” Lexical
Grammar, ed. by T. Hoekstra, H. van der Hulst and M. Moortgat, 214–49.
Dordrecht: Foris Publications. Reprinted in E.L. Keenan 1987. Universal
Grammar: 15 Essays. London, Sydney, Wolfeboro: Croom Helm.
Keenan, E.L. 1987. “Parametric Variation in Universal Grammar.” Universal
Grammar: 15 Essays, ed. by E.L. Keenan, 245–315. London, Sydney,
Wolfeboro: Croom Helm.
Langacker, 1995. “Raising and Transparency.” Language 71. 1–62.
Lasnik, H. and Saito, M. 1991. “On the Subject of Infinitives.” Chicago Linguis-
tic Society 27. Part 1: The General Session, ed. by L.M. Dobrin, L. Nichols
and R.M. Rodriguez, 324–43. Chicago: Chicago Linguistic Society.
Moortgat, M. 1988. Categorial Investigations. Dordrecht: Foris Publications.
Oehrle, R.T. 1987. “Boolean Properties in the Analysis of Gapping.” Syntax and
Semantics 20: Discontinuous Constituency, ed. by G. Huck and A. Ojeda,
203–40. New York: Academic Press.
Oehrle, R.T. 1988. “Multidimensional Compositional Functions as a Basis for
Grammatical Analysis.” Categorial Grammars and Natural Language Struc-
tures, ed. by R.T. Oehrle, E. Bach and D. Wheeler, 349–90. Dordrecht: Reidel.
Oehrle, R.T. 1991. “Categorial Frameworks, Coordination, and Extraction.”

Proceedings of the 9th West Coast Conference on Formal Linguistics.
Stanford: Center for the Study of Language and Information, 411–25.
Pollard, C. 1984. Generalized Phrase Structure Grammars, Head Grammars and
Natural Languages. Ph.D. Dissertation, Stanford University.
Postal, P. 1974. On Raising: One Rule of English Grammar and its Theoretical
Implications. Cambridge, Massachusetts: MIT Press.
Rosenbaum, P.S. 1967. The Grammar of English Predicate Complement Construc-
tions. Cambridge, Massachusetts: MIT Press.
Seuren, P. 1985. Discourse Semantics. Oxford: Blackwell.
Shieber, S. 1986. An Introduction to Unification-Based Approaches to Grammar.
Stanford: Center for the Study of Language and Information.
Steedman, M. 1985. “Dependency and Coordination in the Grammar of Dutch
and English.” Language 61, 523–68.
Steedman, M. 1996. Surface Structure and Interpretation. Cambridge, Massachu-
setts: MIT Press.
Wood, M.M. 1988. A Categorial Syntax for Coordinate Constructions. PhD
Dissertations: University of London.
On the boundaries of syntax
Non-syntagmatic relations*
Peter Peterson
1. Syntagmatic and non-syntagmatic relations
Syntactic units within a sentence may be related to each other in one of two
ways: syntagmatically, or non-syntagmatically.
Syntagmatic relations involve the linking of two or more elements to form
a single grammatical construction, giving the familiar hierarchical constituency
relationships which can be expressed in the form of branching tree diagrams.
These syntagmatic relations are of two types: parataxis (“side-by-side arrange-
ment”) and hypotaxis (“underneath arrangement”). With parataxis (eg. coordina-
tion), each element is of equal status in the construction (they are on the same
“grammatical level”), and no element is subordinated to another. Such construc-
tions are logically symmetrical but not necessarily semantically reversible; “X
and Y” implies “Y and X”, but the two are not usually synonymous. With
hypotaxis (for example, subordination and complementation), the elements are
hierarchically arranged; one element is the “head” of the construction and the
other elements are dependent upon that head. Here, primacy is determined by
hierarchical structure rather than by linear order, and therefore in some cases the
construction may be semantically reversible (for example, subordinate
clause/main clause). Syntactic inequality is reflected in semantic inequality: the
head is free, whereas its dependents are constrained. Thus a subordinate clause
typically does not have independent illocutionary force, may lack independent
tense choice, and may provide presupposed rather than new information.
Non-syntagmatic relations, on the other hand, involve a loose linking of two
230 PETER PETERSON
or more items in a linear sequence which does not constitute a single grammati-
cal construction. The units do not form any larger syntactic unit, and are related
only by linear adjacency, not by hierarchical construction. Because such relation-
ships do not form constituents, they are not directly representable either in tree
diagram form or in terms of labelled brackets. The items are separate grammati-
cal units which are syntactically independent of each other. The sequence of
items has discourse unity, usually signalled by tone concord; but there is no
superordinate syntactic unit.
Items in a non-syntagmatic relationship may be sequentially ordered, one
peripheral to the other (sometimes called “juxtaposition”); or one item may be
interpolated “inside” the other (so-called “parentheticals”). This distinction
between juxtaposition (end-to-end sequencing) and parentheticals (“interrupting”
one unit with another), however, is to a large extent a matter of superficial
ordering. Many “loosely linked” items are able to appear either at the periphery
of or within their “host” element, without any apparent change in semantic
function. This is illustrated by the equivalence of the “juxtaposed” and “paren-
thetical” examples in the sets of sentences in (1):
(1) a. Kim, I’ll be grateful to you for the rest of my life.
I’ll be grateful to you, Kim, for the rest of my life.
I’ll be grateful to you for the rest of my life, Kim.
b. It was Mary who came yesterday, wasn’t it?
It was Mary, wasn’t it, who came yesterday?
c. John will be coming, and probably Sally as well.
John — and probably Sally as well — will be coming.
Because of this typical lack of distinctness, it will be useful to have a cover term
for all non-syntagmatic relationships. I will in general use the term “juxtaposi-
tion” for this purpose, but will on occasion refer to the more specific sub-types
of “peripherals” and “parentheticals.”1
In the following section I illustrate the notion of juxtaposition with some
straightforward examples. The justification for treating these as non-syntagmatic
will be presented in Section 3, and the proposed analysis will be outlined in
Section 4. In Section 5, I discuss constraints on possible sites for the “insertion”
of parentheticals within the “host”, and seek to resolve the apparent paradox
inherent in the concept of “constraints” on non-syntagmatic relationships. The
final section of the paper justifies the extension of the concept of non-syntag-
matic analysis to peripheral clauses, right node raising, and apposition.
ON THE BOUNDARIES OF SYNTAX 231
2. Examples of non-syntagmatic relationships
Examples of juxtaposition will involve a word, phrase or clause appended to or

inserted into a structure to which it is not syntactically dependent, functioning as
an aside or providing material peripheral to the main message. The juxtaposed
item is typically marked off from its “host” clause, in writing by parentheses,
dashes or commas, and in speech by discernible pauses.
Following Espinal (1991), we can classify examples of juxtaposition
according to whether the juxtaposed item is a “self-contained” unit or has some
apparent “link” with the “host” clause. The former category includes vocatives,
interjections, “loose adjuncts”2 and parenthetical (full) clauses. The “linkage”
involved in the latter category may be a coordinator, a pronominal with ana-
phoric reference into the “host”, or ellipsis with the gap “filled” by the “host”.
The theoretical status of the link will be discussed briefly in Section 4 (and see
Burton-Roberts, this volume, for extended discussion).
2.1 Vocatives and interjections
Vocatives and interjections are clear examples of non-syntagmatic relationships.

As Onions (1971: 17) states, “Vocatives and interjections, since they form no
part of the subject or predicate of the sentences or clauses with which they are
connected, are outside the formal structure of the sentence.”
Vocatives (or “address terms”) may occur in isolation, but are commonly
juxtaposed to a host sentence, either at the periphery (initial or final), or inter-
posed, as shown by the examples in (1a) and (2):
(2) a. You people over there, please sit down.
b. It’s good to see you again, old friend.
c. It’s time, Ted, that you and I had a chat.
“Interjection” is a cover term for items such as yes, no, well, oh, hey, here which
can stand alone as responses or exclamations, or introduce sentences serving
these functions. Included under this heading are what we might call “pseudo-address
terms”, which have the form of an address term (for example a proper name) but do
not refer to an actual or intended addressee: boy, God, my saints, Jeez, and so on.
When attached to a host sentence, interjections are typically in initial position:
(3) a. Hey, what are you doing?
b. Here, I know how to fix that.
232 PETER PETERSON
However, they may introduce the second of two coordinate clauses since such
clauses each have independent illocutionary force:
(4) It was a great party, but boy, was it noisy!
As an extension of this, interjections may occur after coordinators used as
discourse connectors:3
(5) a. But hey, why should we care?
b. And yes, I am going to sue.
2.2 Parenthetical clauses
Another illustration of non-syntagmatic relationships is provided by parenthetical

clauses, as illustrated in (6):
(6) a. John Smith — at least I think that’s his name — is asking to
see you.
b. John Smith — is that his real name? — is asking to see you.
c. John Smith — he’s persistent, isn’t he? — is asking to see you.
d. John Smith — boy! is he persistent — is asking to see you.
These clearly function as interpolations, with the marked intonation or punctua-
tion of such asides. As the above examples show, there are no constraints on the
clause type of the parenthetical; it may take the form of a declarative, an
interrogative or an imperative. It may also be introduced by a coordinator, as in (7):
(7) a. John Smith — and he’s a very important client – is asking to
see you.
b. Even Amanda, and she’s my best friend, won’t believe me.
c. The Hawks will win, or at least so I’ve been told, by at least
10 points.
The parenthetical clause may be “elliptical” (Quirk et al. 1985). The “gap” is
typically the complement of the verb; the antecedent may be supplied by a
parallel complement in the “host” clause (8), or more typically by the whole of
the “host” clause (9):
(8) a. Amanda is, or at least she used to be, my best friend.
b. Amanda is, and there’s no doubt in my mind that she always
will be, my best friend.
(9) a. It will stop raining, I expect, before Sunday.

b. John Smith, would you believe, is asking to see you.
c. The Hawks will win, says John, by at least 10 points.
The sentences in (8) are examples of the so-called “right node raising construc-
tion”; they are discussed in more detail in Section 5, where I justify their
analysis as parentheticals.
The examples in (9) illustrate a very common form of interpolation,
typically containing a verb of saying or thinking such as I think, I’m fairly sure,
he said, it must be admitted, would you believe. When they occur as the main verb
in a matrix clause, these verbs require a finite clause as complement. In this
parenthetical use, the complement is “supplied” by the “host” clause. There is
thus an anaphoric linkage between the interpolated clause and its “host”, created
by the ellipsis.
The inverted pattern of subject and verb shown in (9c) is possible only
under restricted circumstances, usually when reporting direct speech or thoughts
with a third person subject; so we may have said John, thought Charles, admitted
Mary, but hardly *expected I or *hope you.
A distinctive property of the elliptical clauses in (9) is that they do not
accept an introductory coordinator, as evidenced by:
(10) a. *It will stop raining, and I expect, before Sunday.
b. *John Smith, or would you believe, is asking to see you.
c. *The Hawks will win, but says John, by at least 10 runs.
d. *The Hawks will win, or at least I’ve been told, by at least 10
points.
Compare in particular (10d) with the minimally different but acceptable (7c).
Quirk et al. (1985) extend the notion of “elliptical clause” to include
examples such as (11) (examples from Quirk et al.):
(11) a. The train arrived — on time for a change.
b. He is playful — even mischievous.
c. I only met him once — at your place.
I take these to be “loose adjuncts” (Peterson 1996), a classification which would
also include “disjuncts” such as those in Footnote 2. (For evidence for the non-
syntagmatic analysis of disjuncts see Espinal 1991.4)
234 PETER PETERSON
3. Properties of juxtaposition
In this section I provide justification for treating the structures discussed in

Section 2 as non-syntagmatic; that is, I will present evidence that the juxtaposed
item does not form a syntactic unit with its “host”.
3.1 VP-ellipsis
McCawley (1988: 763) gives a very simple proof that vocatives are not constituents
of their “host” sentence. Consider the process of VP-ellipsis, as illustrated in (12):
(12) a. A: Didn’t you claim, Pat, that the exam was unfair?
B: No, I didn’t.
b. I didn’t claim that the exam was unfair.
c. I didn’t claim, Pat, that the exam was unfair.
VP-ellipsis elides a VP which is identical to a preceding VP. Pat cannot be part
of that preceding VP because it is not understood to be part of the elided
material; the response in (12a) must be interpreted as (12b), not the (incoherent)
(12c). In general, as McCawley notes, any linguistic phenomenon that depends
on constituent identity will behave as if the vocative is not there at all. This
argument extends quite directly to other non-syntagmatic examples as well. So,
for instance, in an extension of (9c) as in (13):
(13) The Hawks will win, says John, by at least 10 points, and Bill
says so too.
so must be interpreted as The Hawks will win by at least 10 points not as The
Hawks will win, says John, by at least 10 points.
3.2 Any
McCawley (1988: 710) also provides evidence that parenthetical vocatives are
not c-commanded by items in the “host” clause. Thus the indefinite any in the
parenthetical vocative in (14) (McCawley’s example 23a) is not licensed, since
it is not within the scope of the negative in the main clause:
(14) I didn’t predict, *(any of) you bastards, that we would win.
McCawley presents this as evidence in favour of his “crossed branch” analysis,

which gives (15) as the (simplified) derived structure for (14):
(15) S
I didn’t predict that we would win

any of you bastards
However, further investigation shows that the lack of a command relationship

works both ways. It is not only the case that the parenthetical is not c-command-
ed; it also does not c-command into the “host” clause. So, for example, the
negative of don’t think licenses any in (16a) but not in (16b):
(16) a. I don’t think anyone will solve this problem.
b. *Anyone, I don’t think, will solve this problem.
And consider the following examples:
(17) a. I hope the rain stops before Sunday.
b. *The rain stops, I hope, before Sunday.
c. I hope the rain will stop before Sunday.
d. The rain will stop, I hope, before Sunday.
e. *The rain stops before Sunday.
The unacceptability of (17b) is caused by the fact that the verb hope is no longer
governing the present tense form stops. In other words, (17b) is unacceptable
because and in the same way that (17e) is unacceptable. Simple present tense in
English can only be used for future reference when the future event is “programmed”
or otherwise highly predictable, a criterion that is not met in these examples.
Consider now the analysis of parentheticals proposed in McCawley (1982)
in which the structure corresponding to (17d), shown in outline in (18), would be
derived from a structure underlying (17c):
236 PETER PETERSON
(18) S
the rain will stop before Sunday

I hope
McCawley’s analysis maintains the command relationship between hope and will
stop in (18) that would obtain between those items in (17c). However, as we
have just seen, this cannot in fact be the case. There is no such relationship in
the parenthetical examples, and therefore the superordinate S’ posited in (18)
must be rejected. More generally, we must reject any account that incorporates
the parenthetical item as a constituent of the “host” clause.
3.3 Further evidence
The non-equivalence of the examples in (9) (repeated below) and the non-
interpolated constructions in (19) provides further evidence against any account
that treats parentheticals as “derived” from a more basic “higher clause” structure:
(9) a. It will stop raining, I expect, before Sunday.
b. John Smith, would you believe, is asking to see you.
(19) a. I expect (that) it will stop raining before Sunday.
b. Would you believe (that) John Smith is asking to see you.
(9b), for instance, is more closely equivalent to (20), which gives the appropriate
illocutionary force, than to (19b), which is interrogative rather than declarative:
(20) Would you believe, John Smith is asking to see you.
4. Proposed analysis for juxtaposition
The facts outlined in Section 3, combined with the lists of properties presented
by Fabb (1990), Espinal (1991) and Burton-Roberts (1998), provide strong
support for the analysis of juxtaposition proposed in this paper, in which all
examples of juxtaposition, whether parenthetical or peripheral, are treated as non-
syntagmatic. That is, the juxtaposed item is not a constituent of any larger
“containing” or “host” item. There is no structural relationship between the

juxtaposed item and any other element in the stretch of discourse (and therefore
there can be no relevant tree diagram representation). This analysis thus follows
and extends the tradition of Emonds (1979) and McCawley (1988), who treat
non-restrictive relative clauses in the way just outlined. Burton-Roberts (1975,
1998), Haegemann (1988), Espinal (1991), and Peterson (1992), although
differing in some key details, all propose essentially non-syntagmatic analyzes
for certain subsets of parentheticals. Espinal, for instance, discusses the relation-
ship between “disjunct” and “host” as involving linear precedence without immediate
dominance. In Peterson (1992), I propose an analysis of apposition (see Section
5.3 below) based on the use of separate trees which are nevertheless linked, but
not syntactically. Following this proposal, the structural analysis for juxtaposition
is represented diagrammatically as in (21), the (skeletal) structure for (9b):
(21) a. [S[NP John Smith] [VP is asking to see you]]
[Swould you believe]
b. S
NP VP
John Smith is asking to see you

S
would you believe
The dotted line in (21) is intended to show that, although the apposed element is
not a constituent of S, it does have a semantic “bond” with its “host” constituent.
Burton-Roberts (1998) argues convincingly that “contextually interpretable”
relationships such as pronoun-antecedent and gap-antecedent relationships which
hold between parenthetical and “host” involve not (syntactic) coindexing but
(semantic/pragmatic) coreference.
5. Permitted positions of parentheticals
Juxtaposed elements are typically “moveable”; that is, they have a number of
available alternative points at which they can attach to the “host”.
238 PETER PETERSON
(22) a. Clause-initial:
Hey, what are you doing?
Would you believe, John Smith is asking to see you.
It must be admitted, John could win any fight in Australia.
b. Medially (subject to constraints in Section 5.1 below):
John Smith, would you believe, is asking to see you.
John, so I’ve been told, could win any fight in Australia.
c. Clause-final:
John Smith is asking to see you, would you believe.
John could win any fight in Australia, so I’ve been told.
d. Inside coordinate structures:
John and, I think, Mary will play the next round.
John — he’s really clever — and Mary — boy, is she intelli-
gent! — seem made for each other.
His old and (it must be admitted) dilapidated hat caused great
dismay.
5.1 Constraints
Despite the lack of any formal constituency relationship between the juxtaposed
item and its “host” sentence, there are constraints on what positions within the
“host” can serve as a “niche” for the “insertion” of a parenthetical, as shown in
the following examples:
(23) a. We’re glad, Kim, that you could come.
b. We, Kim, are glad that you could come.
c. *We’re, Kim, glad that you could come.
(24) a. Don’t leave me here, you guys.
b. *Don’t, you guys, leave me here. (McCawley 1988)
(25) a. He shouldn’t have washed up the dishes, in my opinion.
b. *He shouldn’t have washed, in my opinion, up the dishes.
c. *He shouldn’t have washed up, in my opinion, the dishes.
(26) a. Mary believes, it must be admitted, that George is a hero.
b. ??Marybelieves, it must be admitted, George to be a hero.
The first constraint, first noted in Emonds (1973), requires that what follows the
parenthetical within the domain of the host must be a constituent of the host.
Emonds captures this restriction through the details of the structural description
of his “Parenthetical Formation” transformational rule. Since Emonds’s account
treats the parenthetical as a constituent of the “host” clause both before and after
it has been “moved”, his formulation of the constraint is not appropriate here.
However, an equivalent effect can be achieved in terms of a “surface structure”
constraint, as in (27):
(27) Constraint I
Let “host” be the maximal constituent “containing” the
parenthetical:
i.e. [a X — P — Y ] where P = parenthetical; a = Cmax
Then Y = constituent.
This constraint accounts for the unacceptability of (25b) and (26b). It may also
help to account for the difference in interpretation between:
(28) a. John could, it must be admitted, not win the fight.
b. John could not, it must be admitted, win the fight.
where (28a) admits the possibility of John not winning, whereas (28b) admits the
impossibility of John winning, a difference which is directly captured if not win
the fight is taken as a single constituent in (28a) but not (28b). (For further
discussion of similar examples, see McCawley (this volume).)
Among other conditions on the positioning of parentheticals, we can note
the following:
(29) Constraint II
A parenthetical cannot (usually) intervene between a verb and its
object.
This constraint rules out examples such as (25c) and (30)5:
(30) a. *John likes, it must be admitted, icecream.
b. *John couldn’t win, in my opinion, the fight in Australia6
Examples containing “heavy NP” objects, however, seem to be acceptable:
(31) a. John likes, it must be admitted, icecream with peanut butter
topping.
b. John couldn’t win, in my opinion, any fight in Australia.
240 PETER PETERSON
which suggests that Constraint II is a matter of “end-weight” preferences; that is,

there seems to be a general preference to avoid inserting parentheticals immedi-
ately before a lightweight (non-focus) constituent. This is supported by example
(32) (from Emonds 1973):
(32) They will sooner or later injure, I predict, themselves.
Emonds lists this example as acceptable; but this judgement depends crucially on
the associated stress pattern. If the stress falls (naturally) on the verb injure, the
sentence is unacceptable; it becomes acceptable only if there is focal stress on
themselves.
5.2 Nature of constraints
If parentheticals are a production phenomenon, then the constraints proposed in

Section 5.1 must be seen purely as conditions on performance outputs (compare
Espinal’s 1991 notion of “processing constraints”). It is not a coherent position
to say that there are syntactic constraints on non-syntagmatic relationships.
Therefore we cannot adopt an analysis of juxtaposition as non-syntagmatic and
at the same time apply McCawley’s (1988) term “external syntax” to parenthet-
icals. If parentheticals are not constituents, then, as Burton-Roberts (1998) points
out, “the grammar doesn’t generate sequences including them”. Parentheticals are
“outside the domain of the grammar”. However, this does not prevent us from
investigating the syntax of the host and the syntax of the parenthetical item
independently of each other. And it may well be the case that the syntax of the
host clause is such that there are “weak spots” — “possible interpolation sites”
— which allow more readily than others for interpolation. But the interpolation
itself remains a production event, not a grammatically generated phenomenon.
6. Widening the net
In the preceding sections, I hope to have established the validity of the notion
that certain elements bear a non-syntagmatic relationship to the “host” clause to
which they are linearly linked. The question then arises as to how widely spread
this phenomenon may be. In this section, I examine the properties of peripheral
clauses, “right node raising”, and apposition, and show that these also may be
given a coherent account by utilizing the concept of non-syntagmatic analysis.
6.1 Juxtaposed clauses
Juxtaposed clauses are clearly non-syntagmatic when parenthetical (Section 2.2

above). When peripheral, as in the following examples, their status is somewhat
less clear.
(33) a. Ned is poor; nevertheless he is happy.
b. Kim loves onions; Pat prefers garlic.
c. The politician cheated on his tax returns; therefore he had to
resign.
d. Joe’s had an accident; he’s in hospital.
e. Joe’s in hospital; he’s had an accident.
In some such examples, it is possible to insert a coordinator without a significant
change in meaning, as shown in (34):
(34) a. Ned is poor but nevertheless he is happy.
b. Kim loves onions but (and) Pat prefers garlic.
c. The politician cheated on his tax returns and therefore he had
to resign.
d. Joe’s had an accident and he’s in hospital.
e. Joe’s in hospital because she’s had an accident.
For this reason, it is tempting to analyze the examples in (33) as “asyndetic
coordination” (that is, coordination without overt marking). However, the fact
that the choice of appropriate coordinator is not always clear (34b), together with
the fact that some examples would more readily accept a subordinator such as
because in (34e), indicates that the relationship between juxtaposed clauses is
typically much looser than between overtly (syntactically) linked clauses. More
crucially, these examples do not display the syntactic unity typical of coordinate
constructions. The examples in (35) and (36) show important contrasts between
coordination, which provides a syntactic unit, and juxtaposition, which does not:
(35) a. Everyone thinks that [Ned is poor but nevertheless he is hap-
py], but they are wrong.
b. *Everyone thinks that [Ned is poor; nevertheless he is happy],
but they are wrong.
(36) a. The politician [whoi [ei cheated on his tax returns] and [ei
therefore had to resign]] was reelected.
242 PETER PETERSON
b. *The politician [whoi [ei cheated on his tax returns], [ei

therefore had to resign]] was reelected.
The examples in (35) show that coordination, but not juxtaposition, can serve as
complement S, while (36) shows that coordination, but not juxtaposition, allows
“across-the-board” extraction. Sentences such as those in (33), then, are best
treated simply as examples of juxtaposition, and hence as further examples of
non-syntagmatic relations.
6.2 Right Node Raising
The set of interpolated coordinate clauses in Section 2.2 included examples

which have elsewhere been labelled “Right Node Raising” (such as (8a) and
(8b)). These are shown to be interpolations by the marked parenthetical nature of
their intonation contours:
(37) a. It seemed likely to me, though it seemed unlikely to everyone
else, that he would be impeached.
b. Kim sells, and Pat knows someone who repairs, washing
machines.
c. Amanda is, or at least she used to be, my best friend. [=8a]
d. Amanda is, and there’s no doubt in my mind that she always
will be, my best friend. [=8b]
It is instructive to compare examples of “Right Node Raising” with straightfor-
ward cases of coordination. In simple cases of conjoined clauses, certain
grammatical properties “distribute” across all conjuncts. For example, in (38):
(38) John dedicated and gave the book to Bill.
John is the subject of both dedicated and gave; the book is the object of both
verbs; and Bill is the indirect object of both verbs. That is, each argument
“distributes” to each conjoined verb. This and similar facts could be explained by
regarding the conjoined structure as the result of combining two full sentences
(S1 + S2). However, this presents a problem, which comes to light when we
compare the sentences in (39):
(39) a. John bought and sold a house yesterday.
b. John bought a house yesterday and John sold a house yesterday.
A natural interpretation of (39b) could be that John was involved with more than
one house, a reading not possible for (39a). The same house is necessarily
involved in both actions in (38a), but not necessarily in (39b). In general, the full
coordination of S1 + S2 has more possible meanings than the “reduced” form.
“Right Node Raising” differs from coordination precisely in that it is not a
“distribution” phenomenon. Consider, for example, the following:
(40) a. John bought, and Bill sold, a house yesterday.
b. John promised Mary and persuaded Sue to leave.
c. John promised Mary, and Bill persuaded Sue, to leave.
d. *John promised and persuaded Sue to leave.
Like (39a), and unlike (39b), example (40a) does not impose an identity require-
ment on the object of the two predicates buy and sell. And in (40b), unlike its
coordination counterpart (40d), promise and persuade do not impose conflicting
requirements on the controller of the infinitival phrase to leave. Thus the “Right
Node Raising” examples show the properties not of coordinate constructions but
of the full unreduced conjoined sentences. These properties follow from an analysis
that takes the italicised sequences in (40a–c) as parentheticals, with the special
property that the missing argument of the verb is supplied by the “host” clause.
6.3 Apposition
Apposition is here also analyzed as a special case of juxtaposition, special in that the
juxtaposed items are referentially equivalent. In its simplest form, apposition is the
placing side by side of two or more items of similar grammatical form, as in (41):
(41) a. Paul Jones, my favourite movie star, is coming to Australia.
b. I wish to speak to the company commander, Captain Manners.
c. She got killed, run over, by one of those heavy lorries.
d. Five poems, five real masterpieces, appear in this book.
Apposition typically involves the juxtaposition of syntactically equivalent
categories. “Syntactically equivalent” is to be interpreted here as “serving an
equivalent grammatical function”. Apposition is like coordination in that
categorial identity is not essential, as long as functional equivalence is preserved.
So, in (42a) we have a PP under the bed in apposition to an NP my favourite
hiding place, which is quite acceptable since the PP in question is capable of
serving as subject. Similarly, in (42b) the apposed PP without help has as “host”
a functionally equivalent adverb alone.
244 PETER PETERSON
(42) a. My favourite hiding place, under the bed, has been ruled out
of bounds.
b. I couldn’t have done it alone, without help.
In fact, as the examples in (43) show, even this functional equivalence require-
ment is not always strictly observed:
(43) a. The reason that he gave, that John is sick, failed to convince me.
b. His claim, that he didn’t see the car, was unconvincing.
c. Jones, at one time a law student, wrote several best sellers.
d. Amanda, no longer my best friend, voted against me.
In these examples, unlike those in (41), the apposed element could not stand in
place of its host.
Notice also that in some cases, rather than appearing adjacent to its host, the
apposed element may be extraposed to a position later in the host clause:
(44) a. An unusual present was given to him, a book on ethics.
b. An argument broke out after the party, a dispute that finally
put an end to their friendship.
c. The Smith family walked in together — John, Mary, Kim and
Pat.
d. Five poems appear at the end of the book, five real
masterpieces.
This option is not always available, as the following examples show:
(45) a. *My favourite hiding place has been ruled out of bounds,
under the bed. (cf. 42a)
b. *Jones wrote several best sellers, at one time a law student.
(cf. 43c)
c. *Amanda won’t believe me, no longer my best friend. (cf. 43d)
This suggests that, at least for non-prototypical examples of apposition, there is
a requirement of “strict adposition” (McCawley 1988: 448), a requirement that
also applies to non-restrictive relative clauses (Emonds 1979).
The key semantic characteristic of apposition is that the juxtaposed items
are referentially equivalent. In (41a) the NP Paul Jones refers to the same entity
as the NP my favourite movie star; in (42a) my favourite hiding place and under
the bed are also co-referential. This referential equivalence distinguishes apposi-
tion quite clearly from coordination which typically involves referentially distinct
items. It also leads to the possibility in most cases of a copular paraphrase,

semantically non-distinct from a non-restrictive relative clause — which has led
some grammarians to suggest a derivational relationship, treating appositional
constructions as reduced non-restrictive relative clauses. However, although I would
claim (along with Haegemann 1988, Fabb 1990, Burton-Roberts 1998) that non-
restrictive relative clauses, like apposition, are non-syntagmatic, I do not analyze non-
restrictive relative clauses as the source for apposition. It has been well established
(see, for instance, Burton-Roberts 1975) that there are examples of apposition
which are difficult to paraphrase sensibly in the form of a non-restrictive relative
clause, a point that is illustrated straightforwardly by the following:
(46) a. The next speaker was a well-known novelist, Kim Brown.
b. An unusual present was given to him, a book on ethics.
c. Five poems appear at the end of the book, five real
masterpieces.
d. The Smith family walked in together — John, Mary, Kim and
Pat.
e. Your relations, John in particular, are not welcome here.
Non-restrictive relative clauses cannot therefore be regarded as a viable source
for apposition, although they have in common their parenthetical character. This
does not prevent us from pointing out the obvious semantic parallels between (at
least some) non-restrictive relative clauses and (some) examples of apposition,
particularly those with an attributive semantic function as in:
(47) a. Paul, a shy person at heart, appeared to be overly confident.
b. The next speaker was Kim Brown, a well-known novelist.
(cf. 46a)
Following the analysis of juxtaposition presented in Section 4 above, apposition
is represented diagrammatically as shown in (48), corresponding to (41a):
246 PETER PETERSON
(48) S
NP VP
Paul Jones is coming to Australia
NP
my favourite movie star
It is important to note that this non-syntagmatic analysis of apposition takes only

the first element to be a constituent of S. The apposed element provides semantic
elaboration of that constituent, without however forming a syntactic unit with it.
Thus I am explicitly rejecting the assumption, made for example by Matthews
(1981), that apposition is a construction, and I am following Bloomfield (1933)
in taking apposition as a special case of what he calls “parataxis”, (what I have here
called “juxtaposition”) in which forms are united only by their intonational unity.
The claim that apposition is an example of juxtaposition is a claim that it is
neither paratactic (under the definition of parataxis established in Section 1) nor
hypotactic. That this is indeed the case can be seen by comparing its properties
with those of coordinate and subordinate constructions.
The clearest difference between apposition and coordination is seen by
looking at agreement properties. In coordinate constructions, agreement is
determined by the set properties of the construction as a whole. In apposition, on
the other hand, only the first element determines agreement (where applicable),
as shown in (49):
(49) a. My favourite food, pork chops, is (*are) not on the menu.
b. Car accidents, the major cause of deaths in Australia, are
(*is) increasing.
c. The most common complaint, too many mosquitos, is (*are)
being remedied.
This agreement pattern follows directly from the analysis in (48). Since only the
first element is a constituent of the tree, there is no superordinate constituent which
is able to accept the transfer of agreement properties from the apposed element.
There is also a clear syntactic distinction between apposition and subordination.
One point of difference is that, at least in core cases of apposition, the apposed
element is functionally equivalent to its host; that is, it is capable of serving the
same function. This is a somewhat weak characteristic of apposition, in that it
does not apply to all the cases we would wish to include under this heading, but
where it does apply it clearly differentiates apposition from subordination. A
more central difference is that apposed items are not semantically reversible,
whereas subordinate clauses are freely ordered with their main clause; compare
(50) with (51):
(50) a. When he comes to Australia, Paul Jones will get a hero’s
welcome.
b. Paul Jones will get a hero’s welcome when he comes to Aus-
tralia.
(51) a. Paul Jones, my favourite movie star, is coming to Australia.
b. My favourite movie star, Paul Jones, is coming to Australia.
In (51a), the apposed phrase, my favourite movie star, gives a relevant character-
ization of the person named in the host NP (a “designation” relationship: Quirk
et al. 1985); it is paraphrasable with the explicit apposition marker that is, or
with a non-restrictive relative clause construction who is my favourite movie star.
(51b), on the other hand, is an example of “appellation”, where the apposed
phrase names the person referred to in the host NP and can be paraphrased using
the apposition marker namely. The non-equivalence of reversed sequences is
predicted by the juxtaposition analysis of apposition, since the two elements of
the apposition have non-equivalent grammatical positions: only the first element
is actually a constituent of S; the apposed element is in loose association with
that “head” element, linked semantically and intonationally (by tone concord) but
not syntactically.
Perhaps the clearest syntactic evidence that an apposed element does not
form a single constituent with its host is provided by sentences such as (52)
(based on McCawley 1988):
(52) a. John sold Mary, his best friend, a lemon, and Max did too.
b. Tom owns a Stradivarius violin, once the property of Heifetz,
and Jane has one too.
The elliptical clause in (52a) is to be interpreted as Max sold Mary a lemon, not
Max sold Mary, who was his best friend, a lemon. There is no implication in (52a)
that Mary is (or was) Max’s best friend. Thus the VP which is serving as the
248 PETER PETERSON
antecedent for the ellipsis in the coordinated clause does not include the apposed
NP. Similarly, in (52b), the violin owned by Jane is not necessarily the former
property of Heifetz; again, the apposed element (in this case a clause) is not a
part of the antecedent clause.
There is thus both semantic and syntactic evidence to show that appositional
elements should be analyzed as juxtaposed to their host, and do not combine
with that host to form a single syntactic unit. Apposition, then, may be taken as
a further instance of a class of constructions where traditional concepts of
hierarchical structured constituency do not apply — the class of non-syntagmatic
relationships.
Notes
* Some of this material has been presented at workshops of the “Cambridge Grammar of English”
research group. I am indebted to the members of that group for their comments, and exonerate
them from any responsibility for the claims made herein.
1. The term “parataxis” is used by some grammarians for what I here call “juxtaposition”.
Halliday (1985), Quirk et al. (1985) and others use the term “parataxis” (as I do here) for non-
hypotactic grammatical constructions, but Quirk et al. (1985) extend its use to include
“juxtaposition” as well. It is important to keep these relationships quite distinct. The relationship
labelled here as “parataxis” is a type of grammatical construction, forming a syntactic unit;
“juxtaposed” items on the other hand do not constitute a syntactic unit at all. I use the term
“parataxis” for the construction type for three reasons: the morpheme -taxis implies construc-
tion; there is an appropriate term (“juxtaposition”) already available for the non-syntagmatic
relation; there is no other useful term for the syntagmatic relation.
2. “Loose adjuncts” (Peterson 1996) subsumes the “disjuncts” of Quirk et al. (1985) as well as
those of Espinal (1991). It therefore includes examples such as the following:
The Hawks will win, in my opinion, by at least 10 points
The Hawks will win, supposedly, by at least 10 points
3. Note that examples such as (4) and (5) provide evidence for an analysis of coordinate clauses
along the lines of (i) rather than (ii):
(i) [S [COORD [S NP VP]]]
(ii) [S [COORD NP VP]]
since only under (i) can we maintain the generalization that interjections occur in clause-initial
position.
4. The title of Espinal’s article, “disjunct constituents”, is in fact self-contradictory, since the
essence of her argument is that disjuncts are not “constituents” at all.
5. For reasons that are not yet clear to me, this restriction does not apply to examples of “Right
Node Raising” (Section 6.1 below).
6. This example becomes more acceptable if in Australia is construed as restrictively modifying

the fight, in which case it parallels the “heavy NP” structure in (31b).
References
Bloomfield, L. 1933. Language. London: George Allen and Unwin.

Burton-Roberts, N. 1975. “Nominal Apposition.” Foundations of Language 13.
391–419.
Burton-Roberts, N. 1993. “Apposition.” The Encyclopedia of Language and
Linguistics. Oxford: Pergamon Press.
Burton-Roberts, N. 1998. “Language, Linear Precedence, and Parentheticals.”
(this volume)
Emonds, J. 1973. “Parenthetical Clauses.” You Take the High Node and I’ll Take
the Low Node ed. by C. Corum, T. Smith-Stark and A. Weiser. Chicago:
Chicago Linguistic Society.
Emonds, J. 1979. “Appositive Relatives Have No Properties.” Linguistic Inquiry
10. 211–242.
Erteschik-Shir, N. 1987. “Right Node Raising.” MIT Working Papers in Linguis-
tics 105–117.
Espinal, M.T. 1991. “The Representation of Disjunct Constituents.” Language
67.726–762.
Fabb, N. 1990. “The Difference between English Restrictive and Non-Restrictive
Relative Clauses.” Journal of Linguistics 26.57–78.
Haegeman, L. 1988. “Parenthetical Adverbials: the Radical Orphanage Ap-
proach.” Aspects of Modern English Linguistics ed. by S. Chiba, 232–54.
Kaitakushi: Tokyo. [Cited in Espinal 1991.]
Halliday, M.A.K. 1985. An Introduction to Functional Grammar. London: Edward
Arnold.
Hannay, M. and E. Keizer. 1992. “Non-Restrictive Apposition in English.” Paper
presented at ICFG5, Antwerp.
Huddleston, R. 1984. An Introduction to the Grammar of English. Cambridge:
Huddleston, R. 1988. English Grammar: an Outline. Cambridge: Cambridge
University Press.
Koboyashi, R. 1972. “Problems in Generating Coordinate Sentences with
Parenthetical Elements.” Studies in English Linguistics 1. 71–91.
250 PETER PETERSON
McCawley, J.D. 1982. “Parentheticals and Discontinuous Constituent Structure.”

Linguistic Inquiry 13. 91–106.
McCawley, J.D. 1988. The syntactic phenomena of English. Vols. 1 and 2. Chica-
go: University of Chicago Press
Matthews, P. 1981. Syntax. Cambridge: Cambridge University Press.
Meyer, C. F. 1987. “Apposition in English.” Journal of English Linguistics 20.
101–121.
Onions, C.T. 1971. Modern English Syntax. London: Routledge and Kegan Paul.
Peterson, P. 1992. “A Non-Syntagmatic Analysis of Apposition and Related
Structures.” Paper presented at the Linguistic Association of Great Britain.
Peterson, P. 1996. “Tight and Loose Adjuncts.” Paper presented at the Australian
Linguistic Society Conference.
Prince, E. 1988. “Discourse Analysis: A Part of the Study of Linguistic Compe-
tence.” Linguistics: the Cambridge Survey, Vol II ed. by F. Newmewyer,
164–182. Cambridge: Cambridge University Press.
Quirk, R., S. Greenbaum, G. Leech, and J. Svartvik. 1985. A Comprehensive
Sampson, G. 1975. “The Single Mother Condition.” Journal of Linguistics 11. 1–12.
Schachter, P and S. Mordechay. 1983. “A Phrase-Structure Account of ‘Non-
Constituent’ Conjunction.” WCCFL 2. 260–274.
Sells, P. 1985. “Restrictive and Non-Restrictive Modification.” Report No. CSLI-
85–28. Stanford: CSLI
Gerund participles and head-complement inflection
conditions*
Geoffrey Pullum Arnold Zwicky
Traditional grammars contributed enormously to our understanding of the facts

of English, yet there are facts about the language that some or most of them
have missed, facts that were not discovered or described until the generative
grammar period. The constraints on long-distance grammatical dependencies
seem very largely to be in this category. So do inflection constraints of the sort
examined in Ross (1972a). Ross began by noting this grammaticality contrast:
(1) a. It continues to rain.
b. It continues raining.
c. It is continuing to rain.
d. *It is continuing raining.
Many and perhaps most speakers share these judgments. With verbs such as
keep, Ross noted, it is very strong: an example like *It is keeping raining quite
hard here is very clearly ungrammatical for essentially everyone.
Ross argued for the existence of a surface structure constraint in English
grammar denying grammaticality to certain sentences in which two -ing-inflected
verbs are adjacent. He called it the “Doubl-ing” constraint. His formulation was
complex; he stated it with a global codicil about the clauses having to have been
subjacent (one immediately embedded in the other) in deep structure, and adds
a paragraph, based on observations of George Lakoff’s, suggesting that the right
formulation might be transderivational. There were various later attempts to
remove any global or transderivational reference from Ross’s treatment (Emonds
1973 was one, and Pullum 1974 offered a friendly amendment). But sixteen
years after Ross’s article, Milsark (1988: 625) proposed that the right answer was
as simple as this:
252 GEOFFREY PULLUM AND ARNOLD ZWICKY
(2) The Doubl-ing Filter (Milsark’s formulation)

At PF, mark as ill-formed any sentence containing contiguous -
ing-affixed words.
By “PF”, Milsark means the most superficial level of description the grammar provides,
the one at which phonetic interpretation of surface structure is made explicit.
Milsark’s formulation exhibits a familiar feature of transformationalist
papers of the last twenty years. The strategy is to present a highly over-general
description of some phenomenon and then try to show that appropriate principles
of Universal Grammar (UG) can rein in the excess generality. The program aims
at discovering interesting and powerful principles of UG. However, it seems to
us most unlikely that reliable principles of UG will arise out of unreliable
description at the parochial (that is, non-universal) level. Milsark’s level of
generality here cannot be made compatible with the facts, and the ways in which
it fails do not seem to be of a sort that we could expect universal grammar to
rectify. In this paper we will pursue the question of just how and why it fails,
arriving ultimately at a satisfyingly accurate formulation that is at the same time
remarkably conservative and traditional in what it says and what theoretical
concepts it relies on. Though first noted (it seems) in generative grammatical
research, Doubl-ing phenomena are best described in a way thoroughly compati-
ble with the assumptions of traditional grammar.
Milsark does have a proposal for covering the well-known contrasts between
superficially similar structures such as the pair in (3):
(3) a. Terry was enjoying reading aloud.
b. *Terry was starting reading aloud.
His explanation of this difference is that the surface structures (at the PF level,
where the Case Filter applies) are as follows:
(4) a. Terry was enjoying [NP[+ACC] [NP[+GEN] e] [[N reading] aloud]].
b. Terry was starting [IP [NP e] [V reading] aloud].
Cased PRO (here shown as [NP[+GEN] e]) is assumed, under the hypothesis of
Jaeggli (1980), to be visible to filters, and hence it renders the two -ing-forms in
(4a) noncontiguous, so the filter does not exclude (3a). But Milsark needs a lot
more than this to save his version of the constraint. He explicitly adopts (p.614)
the very strong assumption that there is only a single -ing suffix in English, and
his statement in (2) mentions no categories, so he is claiming that any pair of
words of any category that happen to end in the -ing suffix and happen to be
GERUND PARTICIPLES AND HEAD-COMPLEMENT INFLECTION 253
adjacent in a surface string with no Cased empty category intervening will yield
a violation of the filter. The erroneous predictions this makes go way beyond
what other principles can cope with.
Given Milsark’s assumption that all instances of -ing are to be identified, it
is highly relevant that the list of syntactically or semantically distinct contexts in
which a suffix of the shape -ing shows up in English is quite formidable. In (5)
we list eight derivational processes that give rise to words in -ing (see Marchand
1969: 302–5), and in (6) we list twenty-five distinguishable syntactic construc-
tions that call for an -ing-inflected verb.
(5) a. Prepositions and subordinating conjunctions (according (to),
during, concerning, considering).
b. Deverbal adjectives (charming, fascinating, disgusting).
c. Deverbal nouns of several sorts (locational: mountings, hous-
ings; material: coatings, coverings; concrete result: writings,
buildings; activity: drinking, smoking; event: meetings, shoot-
ings).
d. Denominal nouns of material (planking, sheeting, towelling).
e. Deadjectival nouns naming fruit varieties (golding, greening,
sweeting).
f. Deverbal first elements of participle-noun compounds (eating
apple, carrying case, dancing girl, wading bird).
g. Deverbal second elements of noun-participle compounds
(spear-fishing, bicycle-riding).
h. Deverbal second elements of compound adjectives of various
sorts (heart-breaking, night-flying, all-embracing, self-destroy-
ing, self-loading, easy-going, good-looking).
(6) a. Adverbial VPs with go and come (go fishing, come dancing).
b. Adverbial VPs with get (get going, get cracking).
c. Progressives (am watching, is singing; also prospective
present am going to).
d. Complements to intransitive verbs of temporal aspect (keep
looking, start running).
e. Post-particle complements to intransitive verbs of temporal
aspect (keep on looking, went on running).
f. Post-object complements to transitive perception or causation

verbs (saw Mommy kissing Santa Claus, leave them wanting
more).
g. From + VP complements of intransitive abstention verbs
(can’t keep from crying, please refrain from reciting poetry).
h. From + VP complements of transitive prevention verbs (keep
us from leaving, stop you from reciting poetry).
i. Close adverbial VP adjuncts (I arrived wearing only a T-shirt).
j. Loose adverbial VP adjuncts (Wearing only a towel, I rushed
into the street).
k. Subject-predicate absolute clauses (There being no objection,
the meeting was adjourned).
l. Preposition-subject-predicate absolute clauses (With you being
in L.A. half the time, it’s difficult to hold things together).
m. Preposition + VP loose adverbial adjuncts (by lurking in a
culvert).
n. Prenominal modifiers (a sleeping dog, screaming children).
o. Postnominal “reduced relative” modifiers (anyone wearing a
tie, two birds chirping merrily).
p. Obligatorily subjectless circumstantial complements (We had
trouble keeping the engine running).
q. Extraposed nominal gerund phrases (It’s no use your looking
at me with those cow eyes).
r. The What’s X doing Y construction (What are you doing
reading my mail?).
s. Nominal gerunds with no determiner (Starting the car is never
easy).
t. Nominal gerunds with genitive determiner (my doing this,
your having broken it).
u. Nominal gerunds with lexical determiner (I don’t approve of
this running away whenever things get tough).
v. Gerunds with accusative subject (I hate them doing that).
w. Action nominals (the continuous burning of the Amazon jungle).
x. Covertly passive complements (This coat needs cleaning by a
professional).
y. Exclamatives (Little Kim running a business!).
From the above it follows that there are literally hundreds of syntactically or
semantically distinct contexts in which it could fortuitously turn out that two
word forms with the -ing suffix might be contiguous in a surface string. Given
the taxonomy of thirty three -ings above, the upper bound on the number of
distinct possibilities for two -ing-forms to fall adjacent in a string is 332 − 33 =
1,056. Nearly all the possibilities that can be exemplified fail to engender Doubl-
ing violations. A few are listed in (7), with the sequence of -ing-forms in
boldface. All of them violate Milsark’s filter, but in fact none of them have the
characteristic ungrammaticality of Doubl-ing violations.
(7) a. Preposition with deverbal noun:
You should report any pain experienced during walking or
other normal activities.
b. Conjoined action nominals:
There was a lot of pushing, shoving, and elbowing of ribs.
c. Progressive aspect and adverbial -ing (Silva 1975):
They are all going fishing.
d. Attributive adjective before noun:
It was a truly amazing building.
e. Noun before predicative adjective:
The novel design made the building amazing.
f. Action nominal before gerund participle complement:
We could hear the screaming coming out of the air vents.
g. Denominal material noun before exclamatory -ing:
Just imagine: aluminum siding selling for a dollar a foot!
Milsark’s formulation thus fails, literally, a thousand times over. But what, at
root, is the problem with his account? It is distinguished by its attempt to
maintain both that (i) no special syntactic condition is needed to draw the
distinction illustrated in (3), and that (ii) no reference to the grammatical
categories of the -ing-suffixed words need be made. By adhering to both of
these positions, Milsark casts the filter in a form that looks essentially morpho-
logical (in that it mentions particular suffixal subparts of words) or phonological
(in that it mentions a specific phonological shape). It contains nothing at all that
is framed in the theoretical vocabulary of syntax. This is what dooms it to
failure. The Doubl-ing constraint is in fact a syntactic condition, not a morpho-
logical or phonological one. Its presence in the grammar may perhaps owe
something to a phonesthetic dispreference for jingling sequences of similar-
sounding endings (see Bolinger 1979 for a claim that nothing more than this —
the same stylistic preference that would disfavour Was his the token taken? — is
involved in the Doubl-ing constraint), but it has clearly been grammaticized in a
very specific way, as contrasts like those in (3) and (7) show. Synchronically it
must be treated as an ordinary syntactic restriction in the grammar of English,
making no reference to morphology or phonology.
We claim that -ing is not unitary in English. We separate the verbal
inflectional suffix -ing from the derivational noun-forming, adjective-forming,
and/or preposition-forming suffix (or suffixes) that happen to share its shape.
That is, we take the traditional view that instances of a suffix that are attribut-
able to word formation operations (with a concomitant semantic value) are to be
distinguished from instances of a similarly shaped suffix that are attributable to
the inflectional realization of morphosyntactic categories (with no semantic
consequences).
Interestingly, in the case of the -ing suffix it is possible to offer some
independent empirical evidence from colloquial usage to support our decision. It
comes from studies of variation in pronunciation in British English. There is a
morphophonemic alternation between -ing and -in’ in the colloquial speech of
most varieties of English. This alternation very frequently treats inflectional -ing
differently from derivational -ing or other instances of the phonological sequence
/I]/. It affects inflectional -ing significantly more (in terms of percentage of
tokens showing the -in’ form) than other instances of the same phonological
sequence. (Speakers who find no difference in their dialect are of course neutral
rather than disconfirming for the claim made here.) We do not need to rely on
intuition to support this claim, because Houston (1991) reports some relevant
statistical results from an analysis of a sample of British working class speakers
interviewed by William Labov in the 1970s, and comes up with clear evidence
of the distinction under discussion.
In Houston’s sample, unstressed ing syllables in monomorphemic forms
were pronounced with a velar nasal 63% of the time, derivational -ing suffixes
had the velar nasal 23% of the time, and inflectional -ing suffixes had the velar
nasal only 13% of the time. Her statistical analysis (see Table 18.7 of probability
coefficients for the velar variant of -ing broken down by inflectional, deriva-
tional, and monomorphemic contexts, p. 250) shows a figure of p = 0.26 for the
property of being an instance of inflectional -ing suffix, meaning that this
property disfavors the occurrence of the velar variant fairly strongly, and p =
0.79 for the property of being in a monomorphemic word, meaning that a
monomorphemic context strongly favors preservation of the velar nasal. The

figure for the property of being a derivational -ing is p = 0.43, meaning that the
influence of this factor is relatively weak and does not significantly affect the
variation in pronunciation of the suffix (0.43 is not very far removed from the
0.50 that is found as the probability coefficient of a factor that does not affect
the likelihood of a variable’s realization).
It should be made clear that Houston argues against regarding the distinction
between derivation and inflection as a sufficient basis for explaining all the
variation; she maintains that the best fit is with a more ramified model, and
corresponds very closely to the “nouniness” gradation argued for by Ross
(1972b, 1973). There are some questions about some of her finer classifications
of the data and about her choice of statistical tests, and the full implications of
her data are not clear to us; but it does seem clear enough to us that the statistical
patterns of distribution of the two morphophonemic alternants in speech indicate that
English working class speakers distinguish derivational -ing from inflectional -ing;
and Houston cites studies making the same claim for American English.
While we distinguish derivational from inflectional -ing, we nonetheless
agree with Milsark on uniting all the inflectional cases of -ing. In particular, we
would defend the position long taken by Rodney Huddleston that no separate
gerund and progressive forms of the verb should be morphologically distin-
guished. There seems no more reason to distinguish the gerund -ing form from
the progressive -ing form than to posit a different -ing suffix for each of the 25
constructions listed in (6) above. There is a compelling argument from parsimony
for analyzing them as the same word form being called upon in different
syntactic constructions.
As it happens, we can again show a factually-based argument for the
analysis we advocate. It is based on English compound verbs like those in (8).
(8) spear-fishing, kite-flying, bicycle-riding, truck-driving, hand-
holding, basket-weaving, beer-drinking, name-taking
Kiparsky (1974) noted that the constructions in which such compound verbs can
occur are all and only the constructions in which the verb is inflected with -ing.
Examples like those in (9) are all ungrammatical, yet any of the inflectional -ing
contexts in (6) will permit these compounds, as illustrated by the representative
four examples in (10).
(9) a. *They usually spear-fish near the rocks.

b. *We kite-flew all this morning.
c. *Chris wants to truck-drive.
d. *I have seldom bicycle-ridden.
(10) a. Let’s go spear-fishing near the rocks.
b. We were kite-flying all this morning.
c. Chris wants to carry on truck-driving.
d. I have seldom enjoyed bicycle-riding.
Clearly, rather than postulate a number of homophonous -ing suffixes that just
happen to be the only ones permissible in this construction, it is preferable to
postulate that the construction is limited to a single grammatical category.
Houston’s data may also be interpretable as supporting this analytical
decision. Her classification of subtypes of inflectional -ing is not as fine as the
one suggested by (6), but as far as we can tell, she found broadly similar
phonological variation behaviour in all the various inflectional -ing constructions
she distinguishes: progressive (I’m workin’ at a caterer’s), quasi-progressive (I
started gettin’ pains), VP complement (I don’t mind watchin’ rugby), periphrastic
future (things are goin’ to happen), appositive participle (We’ve been to Jersey,
drivin’ all over), gerundive adjunct modifier (a waitin’ list), participial adjunct
modifier (the plain workin’ man), ACC-ING (’im gettin’ battered), and gerundive
nominal (havin’ a fall). This could be taken as an argument for analyzing them
all as instances of a single grammatical (morphosyntactic) category, rather than
treating them as an array of distinct suffixes that coincidentally have identical
morphophonemics and highly similar variation profiles.
We will follow Huddleston in using the term “gerund participle” for the
grammatical category of which the -ing inflection is the realization. The gerund
participle is the most regular and reliable form of the verb paradigm in English;
every verb that has non-finite forms has a gerund participle, and no gerund participles
are irregular, not even those of highly irregular verbs like be or go or have.
This result, together with the fact that no instance of noun-forming or adjective-
forming -ing is ever implicated in a Doubl-ing violation, permits a dramatic improve-
ment in the accuracy of Milsark’s formulation of the Doubl -ing filter. All that is
necessary is to replace the phonological or morphological reference to -ing-
affixed words by the morphosyntactic reference to gerund participles:
(11) The Doubl-ing Filter (first revision)

Mark as ill-formed any sentence containing contiguous gerund
participles.
(We omit Milsark’s qualifier “At PF” because, whatever the merits of Chom-
sky’s suggestion that there is a “PF” level combining surface syntactic and
phonetic properties, the level at which filters apply would surely be a matter of
universal grammar, not something to be mentioned in a particular constraint in
English, so there was no reason for Milsark to mention it in his statement. As we
shall see, it will not be needed in our statement either.)
However, it is still the case that there are numerous ways in which gerund
participles can fall together in a surface string, and only a limited set of them
occasion violations of the Doubl-ing filter. We now have to consider Milsark’s
suggestion that the exemptions from the filter are simply the structures in which
in which Case is assigned to the constituent containing the second -ing form, so
that a Cased PRO linearly intervenes between the two and renders them noncon-
tiguous in the representation against which the filter is checked. In fact, this
suggestion does not cover the full range of cases. Examples of sequences of
contiguous non-Case-marked gerund participle verbs that do not invoke the
Doubl-ing filter include the following (some of these are based on examples
found in Ross (1972a, 73) and Pullum (1974, 114)).
(12) a. Conjoined premodifiers:
Crawling, flying, and jumping insects present three different
kinds of problem.
b. Stacked premodifiers:
These herbivorous flying insects are not as much of a
nuisance as the biting flying insects we get in the summer.
c. Progressives in apposition:
I was sitting thinking about my troubles.
d. Aspectual complement before premodifier:
Waldo keeps molesting sleeping gorillas.
e. Progressive before progressive:
I heard a man who was dying describing his feelings.
f. What’s X doing Y construction where Y = VP:
What are you doing reading my mail?
g. Get going idiom:
This is a new company; we’re only just getting going.”
This list could be greatly lengthened, but this heterogeneous sample will suffice here.
The crucial syntactic distinction that must be drawn by (or must somehow
constrain the applicability of) the Doubl-ing filter is, in traditional terms, the
distinction between (clausal) direct objects and postverbal complements that are
not objects. This might be reconstructed in terms of nominal versus verbal
clause-like constituents, as suggested in Milsark (1972), following Emonds
(1970), or it might be reduced to the distinction between Case-marked and non-
Case-marked immediately postverbal constituents. The choice is not at all crucial
here, but for compatibility with Milsark’s (1988) current assumptions, let us
assume the latter.
Case-marked complements of verbs occur as passive subjects and tough-
movement subjects and are required to be immediately adjacent to their govern-
ing verbs, while non-Case-marked complements lack these properties. This is
illustrated in (13), (14), and (15) with the verb try, which on the reading “test or
experience” takes a Case-marked object and on the reading “attempt” takes a
non-Case-marked infinitival complement.
(13) a. Everyone here has tried unflavoured oat bran.
b. Unflavoured oat bran has been tried by everyone here.
c. Everyone here has tried to cut down on fat.
d. *To cut down on fat has been tried by everyone here.
(14) a. To try unflavoured oat bran is difficult for some people.
b. Unflavoured oat bran is difficult for some people to try.
c. To try to cut down on fat is difficult for some people.
d. *To cut down on fat is difficult for some people to try.
(15) a. *I have tried many times unflavoured oat bran.
b. I have tried many times to cut down on fat.
We suggest that the correct generalization about the Doubl-ing filter is that it
applies to sequences of a matrix verb immediately followed by the verb of (what
transformationalists would call) a non-Case-marked complement. This is strongly
suggested by Milsark in his text (pp. 624–631), but not reflected in his formula-
tion of the filter. But there is a more traditional way to put this. The Doubl-ing
restriction affects only cases where the constituent following the first verb does
in fact have the grammatical function of complement rather than that of direct object.
Revising Milsark’s formulation of the filter to mention the crucial syntactic
distinction produces the purely syntactic formulation in (16).
(16) The Doubl-ing Filter (second revision)

Mark as ill-formed any sentence containing a sequence XY where
X and Y are both gerund participles and Y is the head of a phrase
bearing the complement relation to X.
Milsark (1988: 620 n.10) notes a problem for his own attempt to embrace this
generalization. If all clauses have full CP structures (that is, are abstractly full
clauses introduced by subordinators), as is widely assumed in generative
grammar, then even if he assumes that the verb moves to Infl to amalgamate
with the gerund participle-defining features and acquire the -ing suffix, it will
not be in the head position of the complement CP. The configuration the filter
has to rule out is not a local one, but must span the boundaries of a CP and an
IP. The items whose adjacency the filter proscribes are in fact neither adjacent
(there are various empty categories in between) nor subjacent (the second is
contained within more than one maximal project that does not contain the first).
In theories that assume direct VP complementation of verbs, this does not have
to be the case. Assuming for the sake of concreteness that the morphosyntactic
feature [Ger] (gerund participle) is present on phrase nodes as well as lexical
heads, and that a non-Case-marked (non-object) complement VP is represented
as a VP sister to V under VP, the Doubl-ing constraint can be stated as in (17).
(17) The Doubl-ing Filter (third revision)
The following type of local tree is not permitted:
VP[Ger]
V VP[Ger]
This mentions just a single local tree. This is highly significant. What it
means is that the condition can be represented as a constraint on a constituency
principle for local tree configurations directly, instead of as a non-local filter that
scans the whole surface structure of a sentence. Pullum and Zwicky (1991)
suggest that the Doubl-ing Constraint is best located as a language-specific
codicil to the English instantiation of a universal constituency principle defining
VPs. The generalization is independent of the properties of any particular lexical
head, but sensitive to the difference between direct objects and non-object VP
complements.
There is a very interesting difference between the informal statement in

(16) and the claim made by saying that local trees matching the template (17) are
not admissible. While (16) mentions contiguity between a gerund participle head
and the gerund participle head of its complement, (17) does not. A single local
tree can include the head verb of the matrix VP or the head verb of the comple-
ment VP, but cannot include both, because they are not in a mother-daughter or
sister-sister relation. Hence, although there is contiguity in (17) between the
matrix head and the complement VP, this is not enough to prevent material from
intervening between the two heads; it is only enough to prevent material in the
matrix VP domain from intervening. That is, while a constituent in the position
marked X in a tree like (18a) would prevent the tree from meeting the template
in (17), a constituent in the X position in (18b), in the same linear position,
would not be relevant, and the constraint would still apply.
(18) a. VP b. VP
V X VP V VP
X V ...
Remarkably, there is once again empirical evidence that our analysis of the Doubl-
ing constraint is correct. We are predicting that intervening material renders the
constraint inapplicable only if it belongs to the matrix VP. Although there is
always some alleviation of the constraint when material intervenes between the
two -ing verb forms, simply because the unpleasant phonetic jingle effect is
attenuated, speakers who have the Doubl-ing constraint will find that there is
nonetheless a very clear contrast of grammaticality between the following pairs:
(19) a. [VP Keeping right on [VP drinking]] would be most unwise.
b. *[VP Keeping [VP secretly drinking]] would be most unwise.
(20) a. I hope you won’t be [VP continuing throughout next week [VP
going over the same material]].
b. *I hope you won’t be [VP continuing [VP suddenly jumping out
and scaring people]]
(21) a. [VP Beginning straight away [VP being less hostile]] would be
a good first step.
b. *[VP Beginning [VP not always being so hostile]] would be a

good first step.
Thus formulating the constraint in terms of a single local tree makes additional
confirmed predictions about previous unnoticed facts, and brings out an addition-
al way in which Milsark’s formulation is incorrect: it is not contiguity between
the two -ing-marked verbs that is required, but adjacency between the first verb
and its complement VP.
Occurrence of an overt NP between the two verbs very clearly eliminates
the effect of the Doubl-ing constraint, of course, as seen in contrasts like this:
(22) a. We have difficulty keeping the pump running.
b. *The pump has difficulty keeping running.
This follows under the present formulation if the pump in (22a) is in the matrix
clause; that is, if the structure of the keep VP is (23a), a raising-to-object
structure rather than a small-clause structure like (23b).
(23) a. [VP keeping [NP the pump] [VP running …]]
b. [VP keeping [SC[NP the pump] [VP running…]]]
The latter structure fails to bring (22a) under the scope of the generalization
illustrated in (19)–(21), namely that constituents embedded in the complement of
the matrix verb do not have an adjacency-interrupting effect. And Milsark’s
assumption that any Case-marked NP (overt or empty) between two -ing-marked
verbs will always nullify the Doubl-ing constraint seems to be incorrect: if wh-traces
are Case-marked NPs, they should block the constraint; but in fact (24a) seems no
better than (22b), or (24b), where a heavy intervening NP is shifted rightward:
(24) a. *Which pump did you have difficulty keeping t running?
b. *The campus police are stopping t drinking all students who
get bad grades.
The optimal hypothesis would appear to be that (22a) has the structure (23a)
rather than (23b), and empty categories are never visible for purposes of
assessing adjacency in constraints of this type. Thus we have an argument
against small clauses (at least with the type of matrix verb considered here), and
an argument in favor of the conclusion of Halpern (1991), where a phonological
argument against empty category visibility is given.
A brief discussion by Williams (1983: 302–6) deserves attention at this
point, since it presents a V – VP analysis similar to what we are advocating, but
also attempts a collapsing of the Doubl-ing constraint with another grammatical

fact about English, with a view to capturing a wider generalization. Williams’
statement of the Doubl-ing constraint is as follows (p.303, example 62):
(25) *V+PrP VP+PrP
But he generalizes this from sequences where both constituents are gerund
participles to cover also sequences where neither is a gerund participle, to deal
with facts like these:
(26) a. Everybody saw John leave.
b. *John was seen leave by everybody.
But this collapsing is made possible by what is no more than a trick of feature
encoding. Having assumed in (25) a feature ±PrP, Williams generalizes his
constraint through the device of alpha variables over feature values, yielding the
following (p.303, example 63):
(27) *VaPrP VPaPrP
But there is no independent reason for classifying English verb forms in terms of
a binary PrP feature, which is crucial here. In fact, since the number of distinct
forms that need to be recognized for English verbs for reasons not having to do
with tense or agreement is five, the most straightforward way of distinguishing
them is to use a single feature VFORM with the possible values Fin (finite), Irr
(irrealis), Base, PstP (past participle), and Ger (gerund participle). We know of
no argument for analyzing verbs in a way that involves a binary feature whose
“+” value denotes the gerund participle form and whose “–” value denotes the
disjunction of Finite, Irrealis, Base, and Past Participle.
Furthermore, even granting Williams’ various assumptions about how his
constraint applies (for example, that it will not apply to verbs followed by
subjectless nominal gerunds because they are NPs and it will not “look into”
these (1983: 305), and that *John was seen leave will be structurally differentiat-
ed from John was expected to leave in some way), the generalization seems
spurious. Contrasts like those in (28) are surely a matter of subcategorization, not
inflectional marking incompatibility:
(28) a. (i) The police saw John leave.
(ii) John was seen to leave by the police.
b. (i) The police watched John leave.
(ii) *John was watched to leave by the police.
c. (i) The interview committee had the candidate leave.

(ii) *The candidate was had to leave by the interview
committee.
d. (i) *The interview committee forced the candidate leave.
(ii) The candidate was forced to leave by the interview com-
mittee.
And Williams’ proposal that a V has to differ from its VP sister in respect of
whether or not it is a gerund participle is incompatible with the analysis we
would assume for all of the following (assuming, as in Gazdar et al. 1985, that
to is a defective non-finite auxiliary verb):
(29) a. would have been nice (V[VFORM:Base] VP[VFORM:PstP])
b. would have to be nice (V[VFORM:Base] VP[VFORM:Base])
c. would be nice (V[VFORM:Finite] VP[VFORM:Base])
d. to be nice (V[VFORM:Base] VP[VFORM:Base])
e. was eaten by a bear (V[VFORM:Finite] VP[VFORM:PstP])
f. has eaten a bear (V[VFORM:Finite] VP[VFORM:PstP])
g. was believed to be a waste
(V[VFORM:PstP] VP[VFORM:Base])
Consider (29a), for example. It is almost universally accepted that perfect have
is a verb taking a subjectless complement, at least when a modal (in the Aux or
Infl position) precedes it, yet Williams’ constraint would not allow this analysis.
Williams’ constraint is also incompatible with the well-supported claim that
in the intransitive quasi-serial verb construction Go get your shoes the second
verb is in a bare VP complement of the first; his constraint would block verb
phrases such as the following, with the V – VP analysis of Pullum (1990) :
(30) a. go be a fool if you want to
(V[VFORM:Base] VP[VFORM:Base])
b. usually go get the paper
(V[VFORM:Finite] VP[VFORM:Finite])
c. has come put things right
(V[VFORM:PstP] VP[VFORM:PstP])
Finally, Williams’ proposed constraint has an unacknowledged problem with the
key fact that he wants to collapse with Doubl-ing, namely the puzzling and long-
known fact about the naked infinitives that they resist matrix passivization.
Williams himself notes that (what are standardly taken to be) NP-movement
traces must be invisible to the constraint. This is because the structure of (26b)
would be (31).
(31) *Johni was seen ti leave by everybody.
The question arises of whether (what are standardly taken to be) wh-movement
traces are invisible to the constraint. But here the two halves of the constraint
show different behaviors. The non-gerund participle half of Williams’ purported
generalized constraint seems definitely to be voided when a wh extraction site
intervenes between V and VP:
(32) a. This chairi I arranged to have ti made for me.
b. I can’t trust a man whomi I have seen ti cheat his friends.
c. I disapprove of this tendency, whichi I have watched ti spread
among journalists.
d. It was as if one might at any moment see ti emerge the hand
of God.
But the Doubl-ing constraint is insensitive to wh traces, so that (33b–d) are just
as bad as (33a).
(33) a. *Don’t try keeping laughing for two hours.
b. *This is the mani that we’ve been keeping ti laughing for two
hours.
c. *I can’t trust a man whomi I have been watching ti cheating
his friends.
d. *It was as if one might at any moment be seeing ti emerging
the hand of God.
We conclude that Williams’ extra degree of generalization is spurious, and (26)
must have an explanation that is not bound up with the Doubl-ing constraint.
What we are led to, therefore, is a statement of the Doubl-ing Constraint that
says exactly what (17) said above. Williams is right about the support that the
Doubl-ing constraint offers for nonclausal VP complements, but not about the
feature trick with which he attempts to assimilate the constraint to other phenomena.
It is particularly interesting that there is evidence of a very similar kind
from a different construction in which the adjacency of two verbs has been
wrongly taken to be crucial. The go get construction mentioned above has
generally been assumed to involve two verbs that not only have to be devoid of
visible inflection but also have to be strictly adjacent. Thus, for example, the
attempt in Perlmutter (1971) to account for the absence of inflection by means

of a surface filter is explicitly supported by evidence that no syntactic material
can intervene between the two verbs. That is, Perlmutter claims that (34a) is
ungrammatical for the same reason as (34b) and (34c).
(34) a. *He goes get(s) his shoes whenever I tell him to.
b. *Please go inside get your shoes.
c. *Now go as fast as you can get your father.
In (34a), the morpheme -es intervenes between go and get; in (34b) it is the
word inside; and in (34c) it is the phrase as fast as you can. Although virtually
everyone agrees that the examples in (34) are bad, there is wide agreement
among American speakers on contrasts like the following.
(35) a. *I want you to go upstairs [VP rewrite this on a clean sheet of
paper].
b. I want you to go [VP neatly rewrite this on a clean sheet of
paper].
(36) a. *Why don’t you go outside [VP put the lizard back on its rock]?
b. Why don’t you go [VP carefully put the lizard back on its rock]?
(37) a. *Go over [VP glance at her left ear] and tell me what you notice.
b. Go [VP casually glance at her left ear] and tell me what you
notice.
After these data were presented in a paper by Pullum at the 1992 annual meeting
of the Linguistic Society of America, James McCawley presented them to a
number of speakers (students in classes at the University of Chicago) for
judgment, and recorded the results. He reports strong support for the grammatic-
ality judgments we affirmed (see McCawley 1996). There are some speakers
(Zwicky is one) who find even the (b) examples rather bad, but even those
speakers find the (b) cases perceptibly better than the (a) cases.
The configuration involved in the go get construction is exactly the one
involved in the Doubl-ing constraint, namely this:
(38) VP [F1:v1]
V VP [F2:v2]
In the case of the Doubl-ing constraint, the configuration in (38) is forbidden if

[F1:v1] = [F2:v2]= [VFORM:Ger]. In the case of go get, the configuration in (38)
is again involved and is under the restriction that neither [F1:v1] nor [F2:v2] may
determine any morphological operation that modifies the stem shape of the head
V; in other words, the V has to be realized in the Base form or — what suffices
for most speakers — in a form phonologically identical to the Base form (see
Pullum 1990 for further details). It seems likely that both constraints are best
expressed as conditions on constituency statements; that is, generalizations
governing which phrasal constituents may be combined with which heads to
define which types of larger constituents.
To put things this way relates the intrusion asymmetries above to a very
general fact about how syntax works: the reason that right modifiers of the
matrix verb affect the Doubl-ing constraint and the go get construction but left
modifiers of the complement verb do not has to do with the fact that true syntax
is never about putting words together with other words, but always about putting
phrases together with heads.
There is a further advantage of our reformulation of the Doubl-ing con-
straint: it eliminates a violation of the Principle of Phonology-Free Syntax, which
we have argued for in a number of other publications (see, for example, Pullum
and Zwicky 1988 for a conceptual introduction). This principle forbids reference
to phonological information in syntactic statements. It is well supported by
research on a large number of languages. If there can be PF filters on the output
of the syntax that can mention the phonological or morphological form of words,
it is not clear that phonological conditions on syntactic structure can be correctly
excluded (imaginary languages with ridiculous rules like “in a passive clause the
subject must not begin with a bilabial nasal” might be describable if such filters
are permitted). An appropriately restrictive theory of Universal Grammar will
exclude phonological and morphological references from the formulation of
syntactic conditions in principle, and thus require that restrictions like the Doubl-
ing constraint be stated in purely syntactic terms. This we achieve under our
treatment. The consistency of our analysis with the principle increases our
confidence in both (as do the four reanalyses of apparent counterexamples in
French that we discuss in Miller, Pullum and Zwicky 1997, where again we
propose that simple and plausible traditional descriptive analyses turn out to be
mututally supportive with the phonology-free syntax).
The syntactic statement we have given here, though vastly more adequate in
descriptive terms than Milsark’s account, is still extremely simple, and can be
expressed without difficulty in entirely traditional terms:
(39) It is not acceptable in most varieties of modern English for a com-
plement (as opposed to an object) marked with gerund participle
inflection to be adjacent to its matrix-clause verb when that verb is
likewise in the gerund participle form.
There are puzzles that remain, and of course one of them is how a learner might
learn something like this. No utterance ever heard would be able to provide
evidence for it; instead, it would have to be learned by perceiving an absence
from the set of utterances encountered. But in this, our descriptive statement is
no different from Milsark’s. We are inclined to think that the answer might be
that infants learning languages do so through slow and conservative generaliza-
tion, never assuming that some configuration like (38) is permissible with a
given instantiation of [F1:v1] and [F2:v2] until a confirming example has been
heard. Determining whether this is too conservative a standpoint to permit
learning to occur must await further research on language acquisition.
Note
* It is a pleasure to be able to offer this study of a small point in English grammar as a tribute
to Rodney Huddleston, a grammarian whose work on English grammar (small points and large)
we have admired greatly for more than two decades. We thank a number of colleagues who
contributed correspondence, comments, conversation, or useful references, particularly James
McCawley, Bonnie McElhinny, Louise McNally and (we do have some friends whose names
begin with something other than Mc-) John Rickford. Assistance and support was received from
the staff at the Center for Advanced Study in the Behavioral Sciences, where some of the
research for this paper was done, and were supported in part by NSF grant number BNS
87–00864 and a sabbatical leave from the University of California, Santa Cruz (Pullum) and a
Distinguished Research Professor award from The Ohio State University (Zwicky).
References
Bolinger, D. 1979. “The Jingle Theory of Double -ing”. Function and Context in
Linguistic Analysis: A Festschrift for William Haas ed. by D.J. Allerton, E.
Carney and D. Holdcroft, 41–56. Cambridge: Cambridge University Press.
Emonds, J.E. 1970. Root and Structure-preserving Transformations. Doctoral

dissertation, MIT; distributed by Indiana University Linguistics Club,
Bloomington, Indiana.
Emonds, J.E. 1973. “Alternatives to Global Constraints”. Glossa 7. 39–62.
Halpern, A. 1991. “Phonological Evidence for the Surface Nonexistence of
Empty Categories”. The Proceedings of the Ninth West Coast Conference on
Formal Linguistics, 233–247. Stanford Linguistics Association, Stanford,
California.
Houston, A. 1991. “A Grammatical Continuum for (ING)”. Dialects of English:
Studies in Grammatical Variation ed. by P. Trudgill and L. Chambers,
241–257. Longman: London.
Jaeggli, O. 1980. “Remarks on To Contraction”. Linguistic Inquiry 11. 239–245.
Kiparsky, P. 1974. “Remarks on Analogical Change”. Historical Linguis-
tics,Volume 1 ed. by J.M. Anderson and C. Jones, 257–75. Amsterdam:
North-Holland.
Marchand, H. 1969. The Categories and Types of Present-Day English Word-
Formation: A Diachronic Approach (2nd ed.) Munich: C. H. Beck.
McCawley, J.D. 1997. “Acceptability Judgments in the Teaching and Doing of
Syntax”. Papers from the 32nd Regional Meeting of the Chicago Linguistic
Society, part 2, 119–132.
Miller, P., G.K. Pullum and A.M. Zwicky. 1997. “The Principle of Phonology-
Free Syntax: Four Apparent Counterexamples in French”. Journal of
Linguistics 33.67–90.
Milsark, G.L. 1972. “Re: Doubl-ing”. Linguistic Inquiry 3. 542–9.
Milsark, G.L. 1988. “Singl-ing”. Linguistic Inquiry 19. 611–34.
Perlmutter, D.M. 1971. Deep and Surface Structure Constraints in Syntax. New
York: Holt, Rinehart and Winston.
Pullum, G.K. 1974. “Restating Doubl-ing”. Glossa 8. 109–120.
Pullum, G.K. 1990. “Constraints on Intransitive Quasi-serial Verb Constructions
in Colloquial English”. Working Papers in Linguistics (The Ohio State
University) 39. 218–239.
Pullum, G.K. 1991. “Nominal Gerunds as Noun Phrases with Verb Phrase
Heads”. Linguistics 29. 763–799.
Pullum, G.K. and A.M. Zwicky. 1988. “The Syntax-Phonology Interface”.
Linguistics: The Cambridge Survey, Volume 1 ed. by F.J. Newmeyer,
255–80. Cambridge: Cambridge University Press.
Pullum, G.K. and A.M. Zwicky. 1991. “Condition Duplication, Paradigm Homo-
nymy, and Transconstructional Constraints”. Proceedings of the Seventeenth
Annual Meeting of the Berkeley Linguistics Society, 252–266.
Ross, J.R. 1972a. “Doubl-ing”. Linguistic Inquiry 3. 61–86.
Ross, J.R. 1972b. “The Category Squish: Endstation Hauptwort”. Papers from the
Eighth Regional Meeting ed. by P.M. Peranteau, J.N. Levi and G.C. Phares,
316–28. Chicago Linguistic Society, Chicago, Illinois.
Ross, J.R. 1973. “Nouniness”. Three Dimensions of Linguistic Theory ed. by O.
Fujimura, 137–258. Tokyo:TEC Company.
Silva, C. 1975. “Adverbial -ing”. Linguistic Inquiry 6. 346–50.
Williams, E. 1983. “Against Small Clauses”. Linguistic Inquiry 14. 287–308 .
Zwicky, A.M. 1990. “Syntactic Words and Morphological Words, Simple and
Composite”. Yearbook of Morphology 3, 201–216.
Isolated if-clauses in Australian English*
Lesley Stirling
1. Introduction
This paper will address the usage and grammatical status of isolated if-clauses
such as (1) and (2).
(1) Okay if you’d like to get dressed now.
(2) If I’d somewhere to go, some friend’s room.
(1) was produced by the doctor in the examination phase of a general practice
consultation. It belongs to a class of formally identical utterances which have the
characteristic force of a polite request. (2) is from dialogue represented in a
novel, and expresses a wish on the part of the speaker. In both cases, a clause
which is formally subordinate occurs without the expected accompanying main
clause. Although well-established in the linguistic community, usages such as (1)
are typically overlooked in traditional grammars of English, while those such as
(2) receive minimal and sporadic attention. The question for the grammarian is
what their grammatical status is: are they strictly ungrammatical, elliptical, or do
they represent an independent minor sentence type? A number of writers have
pointed to a phenomenon whereby subordinate clauses come to be reanalysed as
independent clauses, and the examples in question appear to represent English
examples of such a process in action.
Two corpora of Australian English were surveyed to collect examples of
these types of isolated if-clause. The communicative function and distribution of
the collected examples will be considered, focussing in particular on the ill-
described directive type, and then their grammatical status will be discussed.
274 LESLEY STIRLING
2. The syntax of non-isolated if-clauses
If is used in two types of subordinate construction, illustrated by (3) and (4) below.
(3) I wonder if the play is still on.
(4) a. If the weather is fine, (then) the play will be on.
b. The play will be on, if the weather is fine.
(3) illustrates the use of if to introduce a subordinate interrogative complement
clause. In this context if is an alternative to whether. (4a) and (4b) exemplify its
use in the “conditional” construction, where the clause introduced by if is
traditionally referred to as the antecedent or protasis, and the main clause as the
consequent or apodosis. The main clause normally follows the if-clause (compare
Ford and Thompson 1986), and in this position it is optionally introduced by
then. Much has been written about the semantics and pragmatics of conditional
constructions (for instance Traugott et al. 1986), in particular with respect to a
range of functions which this construction may have that do not fit the logical
conditional; these issues will not be of concern here.
While the traditional analysis of the conditional construction treats if as a
subordinating conjunction introducing an adverbial clause, Huddleston (1984:
340f., 390–1; 1988: 33, 124, 154) argues for an alternative analysis on which if
is a preposition taking a subordinate clause as its complement. Here the tradition-
al terminology will be used, owing to its monopoly in both traditional grammars
and the literature on conditionals. Under a PP analysis, the observations made
about usage still stand, and an analysis in terms of grammaticalization would still
be possible.
3. A survey of isolated if-clauses in Australian English
The two corpora surveyed were (a) a corpus of transcribed general practice
consultations and (b) the Macquarie Dictionary Corpus (Ozcorp).1
Observation, and previous work by Ford and Thompson (1986), had
suggested that conversation was the most fruitful place to look for examples of
isolated if-clauses. The General Practice Corpus consists entirely of conversation-
al dialogue. The consultations were recorded in Queensland in 1980 with the
cooperation of the RACGP Family Medicine Program, as part of a much larger
study in which over 300 consultations from 17 doctors were collected. For the
ISOLATED IF-CLAUSES IN AUSTRALIAN ENGLISH 275
current survey a random selection of 40 consultations, involving 11 doctors, was

used. Consultations averaged 12 minutes in length and the transcribed text
consisted of approximately 42,000 words.
The Macquarie Dictionary Corpus consists of approximately 20 million
words from approximately 250 texts, predominantly in Australian English. These
represent a range of genres, but with most of the data coming from written
language. For this study a subset of genres was chosen, to represent as “conver-
sation-like” language as possible, with the caveat that there are clear limitations
in taking the representation of dialogue in plays, and of dialogue and third person
thought in novels, as an indication of actual usage. The genres selected were:
Oral Group: The transcription of a public hearing into Aboriginal
Deaths in Custody, 1989; approximately 90,937 words.
Drama: Printed texts of two plays; approximately 28,053 words.
Private: Letters and email; approximately 1,775 words.
Newspaper: Including newspaper database material; approximately
3,759,328 words.
Fiction: Novels; approximately 10,158,816 words, of which one
novel was surveyed.
Both corpora were searched for instances of if, and all examples of if-clauses
occurring without a main clause were then identified. As anticipated, most
examples of isolated if-clauses from the two corpora fell into one of the two
categories identified in the Introduction, although no examples of optatives like
(2) occurred in the General Practice Corpus, and there were only 9 examples of
directives like (1) in the Macquarie Dictionary Corpus. A systematic quantitative
survey was possible for the General Practice Corpus, and Table 1 shows the 221
if-clauses which occurred classified according to syntactic type.
276 LESLEY STIRLING
Table 1. If-clauses in the General Practice Corpus

Type of If-clause: Number: Percentage:
Complete subordinate interrogative constructions 035 15.8
Complete conditional constructions 141 63.8
Incomplete 013 05.9
Contextually supplied consequent 006 02.7
Directive isolated if-clauses 019 08.6
Other 007 03.2
Totals 221 100
In addition to the directive and optative examples, two types of isolated if-clause
occurred in dialogue across all text types considered in the corpora: these are
classified as “Incomplete” and “Contextually supplied consequent” in Table 1.
These will be excluded from consideration here, since they seem uncontro-
versially analysable as incomplete or fragmentary instances of normal conditional
constructions. They are briefly described in sections 3.1 and 3.2 before more
extensive consideration is given to the directive and optative kinds. A further
small set of examples, categorized as “Other” in Table 1, consisted of utterances
difficult to classify due to inaudible stretches, and relatively infrequent uses of
if which will not be considered here, for instance as if constructions.
3.1 Incomplete utterances
The utterance of a conditional construction may be truncated or “tail off” and

the protasis may consequently be missing. Matthews (1981: 40–2) distinguishes
“incomplete utterances” from “incomplete sentences”; in the former the speaker
stops not because he/she expects the hearer to supply the missing words, but for
circumstantial reasons such as speaker reformulation of the utterance in progress,
or interruption by the other discourse participant. The resulting fragments have
no standing of their own and are cued with cut off, not completion intonation.
In the written texts (in particular the plays) examined here, punctuation signals
this category by indicating incompleteness. For Matthews, such utterances are “of
no concern to syntax, except as a source of confusion in our data” (p. 41). (5)
and (6) are representative examples.2
(5) P: We thought if we saved in the bank and try, and, it’s very
difficult to know what to do. [GP 1:9]
(6) PEARL: [disdainfully] I was only tellin’ you how the whole thing
looked to me. If a person can’t pass an opinion …
OLIVE: You pass too many damned opinions, that’s yur trouble.
[SUM Act 2 Scene 1]
It is however sometimes difficult to tell whether an isolated if-clause which
finishes a turn is intended to be complete or falls into this category. In the
quantitative study reported in Table 1 a conservative approach was taken in that
examples were classified as (directive) isolated if-clauses only if they were
clearly intended to be complete.
3.2 Elliptical if-constructions with contextually supplied main clauses
The consequent to an ellipsis if-clause is sometimes clearly supplied in the

linguistic context. The most common cases are those where the if-clause
constitutes a response and in some cases a qualification to an utterance previous-
ly made by a different speaker. (7) and (8) are representative examples (under-
lining indicates the relevant part of the previous utterance). Examples (9) and
(10) illustrate a further subtype, where the conditional is jointly constructed by
the two participants.
(7) D: It’s actually tender to touch then?
P: Only if you push it, push on it or ummm it’s …
[GP A76:3]
(8) SIMMONDS: […] There’s a good life here for you in the force if
you know how to organise yourself.
ROSS: It’s a pretty good life is it?
SIMMONDS: If you know how to organise yourself and get your
priorities straight.
[REM Act 1]
278 LESLEY STIRLING
(9) MR CLARKE: […] It all depends on the extent of the work

that’s got to be done, and if we say, “All right,
move your family in there — ”
MS CONNELLY: It might end up $20,000 worth of damages,
mightn’t it?
[TRANS p. 251]
(10) MR CLARKE: […] But if you want to buy a motor car or
something —
MS N. MOORE: You save up.
MR CLARKE: Yes, save up, but the thing is […]
[TRANS p 253]
In determining whether relevant material is contextually supplied there is a
continuum of cases, and again a relatively strict approach has been taken here.
This subtype is taken to include only those utterances for which there is a clearly
identifiable clause which provides the consequent, and which is not analysable as
part of the same syntactic sentence as the if-clause.
3.3 Isolated if-clause directives
This type was exemplified in (1) and further representative examples are given
in (11)–(19). As indicated in the Introduction, in these examples the speaker uses
the isolated if-clause with the illocutionary force of a directive, most frequently
a request (but in some cases perhaps more appropriately categorized as a
suggestion, such as in example 16).
(11) D: Deep breaths … If you’d like to move your head a little. Thank
you. [GP 5:14]
(12) D: Ohh, … well I’d be most surprised … ahhh, if you’d come
back in a month’s time. [End of consultation]
[GP A76:3]
(13) D: […] If you finish off your Microgynon pack for this month, and
then just when you finish that one start on the Neogynon in the
same spot, go around this month.
[GP A74:3]
(14) D: […] Perhaps if you could just pop back in a week and let me
check it again to make sure it’s not shot up
M: Uh huh
D: too much but certainly on today’s reading it’s normal.
[GP A74:6]
(15) D: […] I, I, they’re not on the NHS any more, um so Gavascon,
so if you just ask the chemist to sell you a box of Gavascon.
P: Gavascon. [GP 3:19]
(16) D: Yes, yeah. Well if you can get someone to massage those
muscles for you, and just the hot shower, or a hot water
bottle, at the end of the day, and that’s really about all you
can do for it. […] [GP A76:8]
(17) FIONA: (to the men) Perhaps if you could take the double bed
from our bedroom.
(ROSS and THE REMOVALIST move offstage.)
[REM Act 2]
(18) THE COMMISSIONER: […] Whose job is it to start something
off here. Is it the police or the Aboriginal Legal Service or the
country?
MR PETTIT: If I could answer that, Mr. Commissioner. My job in
1988 — and it took most of that year — was to consult with com-
munities all around Victoria regarding the concept of the
community justice panels. […] [TRANS p. 11]
(19) MS HIGGINS: It’s also available at the local office.
MR CLARKE: Yes. If you went to the local office here and just
said, “Look, what sort of schemes have you got going? Give us a
look at them.” [TRANS p. 268]
The subject of the if-clause is normally second person. In cases such as (18),
where a first person subject occurs, the utterance has the force of a request for
permission — in fact, most of the few examples from the Macquarie Dictionary
Corpus were of this type. The directives naturally refer to a future action, and so
it is not surprising that the simple past tense only rarely occurs (just once in the
examples found in this study); instead one finds either the present tense form of
the main verb, the present tense form of the modal can, or the past tense form of a
modal (could, ’d, might); with the latter type making up over half the cases found.
280 LESLEY STIRLING
There has been little discussion of this usage for English. For example,
although Quirk et al. (1985: 11.38, 11.41) include the isolated if-optatives in a
category of “irregular sentences”, the directive type is not mentioned. Directives
are occasionally noted in corpus-based studies, but without extensive discussion:
this is the case in Mulholland and Stirling (Ms: 33–4, 39–40, 42–3) and Ander-
son et al. (1991: 356–7). Ford and Thompson (1986: 365) identify a function for
conditionals of “polite directives” but do not explicitly distinguish the isolated
type, saying merely that “in many cases a consequent clause is very difficult to
isolate”. Evans (Ms: 5, 10) also notes their existence. Similar uses occur in many
other languages. Evans reports that parallel request uses of if-clauses occur in
French, Basque, Dutch, Spoken Mon, and Japanese, and Buscha (1976) includes
the translation equivalents in an extensive study of isolated subordinate clause
types in German. Brown and Levinson (1987: 153f.) note that in Tzeltal, the
normal word for “if”, me, has come to be used also as a particle which softens
commands, turning them into polite suggestions, and which turns assertions into
gentle commands.
Nor have these examples been mentioned in the literature on speech acts
and politeness (for instance Brown and Levinson 1987), although polite uses of
complete conditional constructions have been discussed. There are two types of
example, of which the first was not found in the General Practice Corpus and the
second will be discussed below:
1. The proposed action is coded in the main clause, with the if-clause provid-
ing a hedging precondition, often expressing respect or deference or
suspending felicity conditions on the illocutionary act being performed; for
example if that would be agreeable to you, if you can, if I may ask you
(compare Levinson 1983: 266, Ford and Thompson 1986: 368, Brown and
Levinson 1987: 162f.).
2. The proposed action is coded in the if-clause with the main clause express-
ing some (desirable) consequence of compliance or an evaluation of the
situation arising as a result of compliance.
The directive use of isolated if-clauses in English does however fit the general
pattern exhibited by indirect directives. Brown and Levinson (1987: 227) identify
ellipsis as one of the main mechanisms of indirection and Sadock and Zwicky
(1985: 193) note that “Numerous languages use some typically subordinate
clause form, a free-standing infinitive or subjunctive, for example, as a circumlo-
cution for the imperative”. They and others working on “indirect speech acts”
also note the tendency for such acts to involve overt expression of various
felicity conditions on the performance of speech acts, which include the desire
for the act by the speaker, and the ability of the addressee to carry it out. As we
shall see, in the examples of interest here, if we could reconstruct omitted
material, it would have to do with either the speaker’s wish for the act (the sincerity
condition on requests: that S wants H to do A) or the hearer’s ability/willingness to
perform it (a preparatory condition on requests, that S believes H can do A).
Brown and Levinson (1987: 172f.) note one class of indirect directives as
characterized by avoidance of coercing the hearer’s response, which may be done
by explicitly giving H the option of not doing A (they cite as an example use of
past tense forms of modals such as could, would, might to express doubt about
the likely eventuality of the act).
The isolated if-directives seem to be of this kind. Their communicative
meaning seems to be something like: “I want you to do A, but I don’t necessari-
ly believe that you will do A”; that is, use of the if-clause construction allows the
communication of the possibility of not-A. This is based on the assumption that
at least part of the difference in meaning between an assertion and an if-clause
is the difference between “I believe that p” and “I believe that there is some
chance that p”. Thus, the if-clause allows the speaker to express that he/she is
not assuming the performance of the act requested of the hearer; the hearer has
an option.
The social meaning of choice of an isolated if-clause rather than a more
direct form is less easy to determine. Davison (1975: 149) notes that “the extra
factor in indirect speech acts is often described as politeness [ … ], but polite-
ness is hard to define narrowly enough to be of use”. Where mentioned in the
literature these uses have been described as polite directives, and they cooccur
with other markers of tentativeness, such as the adverbs just and perhaps, and
formulations such as if you can think about A or if you’d like to do A. However,
isolated if-directives potentially span a range of degrees of politeness, cf. the
(unattested) example If you’d leave the dog alone! (with appropriate intonation).3
Furthermore, impressionistically, there are individual differences in speaker style
in frequency of use of the isolated if-directives, an observation which awaits
empirical investigation. The best evidence we have at present as to their social
meaning comes from their distribution in the corpora.
All the examples from the general practice consultations were uttered by
doctors (who produced the vast majority of directives overall; 91.5% of them
according to Mulholland and Stirling Ms: 11). Interestingly, the isolated if-
282 LESLEY STIRLING
directives and you-imperatives were the only forms of directives which patients
did not use (pp. 40, 42f.). All the examples were found in two phases of the
consultation: in the concluding phase of detailing of treatment and, much less
commonly, in the examination phase of the consultation. In this they pattern like
directives generally which are concentrated in these two phases of the consulta-
tion. Examples from the examination phase, “internal directives”, express
requests by the doctor for the patient to move parts of their body. Examples from
the treatment phase, “external directives”, involve specification of regimen or
further treatment involving acts the doctor recommends the patient to perform
outside the context of the consultation; for these acts the patient’s compliance is
delayed and will be unsupervised by the doctor. These examples often relate to
acts which have already been mentioned in the consultation and function as a
final reinforcement of the importance of carrying them out.
In both cases the directive-issuing situation is conducive to indirectness. In
the examination phase, directives involving patient body movement in compli-
ance with the examination require relatively intimate acts of the patient and are
dependent on the patient’s cooperation to achieve. In the treatment specification
phase, a major concern of the doctor’s is patient compliance with directives the
outcome of which the doctor will not be in a position to oversee, and the use of
the isolated if-clause appears to give the patient responsibility for making the
decision to carry through the act. In some cases use of this form for the directive
reflects real potential constraints on the patient’s ability to fulfil the request, as
for instance with example (16), if you can get someone to massage those muscles
for you. These directives also occur at the close of the interaction, a point at
which the on-going relationship between the participants may be at issue.
In an attempt to examine the circumstances of use of directive isolated if-
clauses more systematically, a corpus of Scottish English dialogues was investi-
gated. This corpus provided the opportunity to look at frequency distribution of
such clauses in the context of a methodological design which controls for certain
aspects of the interactive relationship between the speakers. Despite the differ-
ence in dialect, the results are suggestive. The HCRC Map Task Dialogue
Database was collected by the Human Communications Research Centre (see
Anderson et al. 1991, who note the presence of isolated if-clause directives in
passing (p. 356–7)). The corpus consists of 128 dialogues obtained from 64
talkers, all undergraduate students from the University of Glasgow, all but 3
Scottish and with a mean age of 20. Participants worked in pairs, each with a
map in front of them that the other could not see. One participant had a route
drawn in and was required by the task to instruct the other in drawing the correct
route onto their own map. The maps were mostly identical, but differed in
certain features. Systematic manipulation of the design variables resulted in two
crosscutting subject conditions of “familiarity” and “eye-contact”, with half the
dialogues between speakers who knew each other well and half between speakers
who had never met, and similarly half in which the speakers had eye contact and
half in which they did not.
Analysis of frequency distribution data gathered by the author shows that
the isolated if-clause directives occurred surprisingly frequently in this corpus. In
a total of 21,251 turns over the 128 dialogues (comprising 151,455 words), 578
instances of if occurred, of which 267 were the isolated directive type. Thus,
46% of all if-clauses which occurred in the data were of this type, compared with
8.6% of the if-clauses in the General Practice Corpus. The greater frequency of
examples is most likely due to the nature of the map task, which must involve
a higher proportion of directive utterances overall than the general practice
consultations (for these, Mulholland and Stirling Ms: 11 report that approximate-
ly 10.3% of speech exchanges included one or more directives). However,
differences in dialect and in size of the two corpora may also be important.
Moreover, usage of isolated if-directives in the HCRC corpus correlates with
familiarity of speakers, with directive if-clauses occurring more frequently in
dialogues where the speakers were unfamiliar with one another. The average over
the 64 familiar speaker dialogues of the number of directive if-clauses per 100
turns gave a mean normalized frequency for familiar speakers of 0.87 directive
if-clauses per 100 turns. This compared to a mean normalized frequency for
unfamiliar speakers of 2.04 directive if-clauses per 100 turns. After normalizing
for dialogue length, a two-way ANOVA for the factors of familiarity and eye-
contact in the use of directive if-clauses showed a highly significant effect of
speaker unfamiliarity on frequency of use of directive if-clauses (F=7.32, df
1,124, p < 0.01, hÐ2=0.055). There was no effect for eye-contact and no interac-
tion between the two factors, and the number of non-isolated if-clauses did not
differ significantly across the conditions. The frequency of directive if-clauses
was normalized by dialogue size because dialogues differed substantially in
length (as measured by number of speaker turns), and length of dialogue
correlated significantly with factors such as familiarity of speakers and whether
they had eye contact.
It seems, then, that both the nature of the act requested and the relative
social distance between the dialogue participants may contribute to the likelihood
284 LESLEY STIRLING
of an isolated if-clause directive being used. It is worth noting, however, that the
eta squared value of 0.055 for the map task data demonstrates that the factor of
familiarity accounts for only 5.5% of the variance of normalized isolated if-
clause frequencies over this corpus of dialogues, confirming that a full explana-
tion for their use must be more complex.
As noted above, if-clauses complete with main clause may also be used to
issue directives. Examples from the General Practice corpus include:
(20) D: […] I wonder if you’d pass a little urine test for me. […]
[GP A73: 1]
(21) P: I was wondering if I could have something, for my ears and my
throat. [GP A76:5]
(22) D: Right, well, if you can manage to keep on the four a day I
think that’s preferable. [GP A73:2]
(23) D: […] OK, if you just pop your arm up there I’ll check your
blood pressure. […] [GP A74:6]
(24) D: We’ll just check your weight again too, if you’d like to pop
over on the scales. [GP A74:2]
These include both syntactic types of if-clause: the subordinate interrogative type
in (20) and (21) and the conditional type in (22)–(24). In the conditional cases
the if-clause expresses the requested act and the main clause specifies a desirable
consequence of compliance (23, 24) or an evaluation (22). Classification of these
constructions as directives can be difficult, since in all cases a substantive
conditional meaning is conveyed also, although in some the directive meaning
clearly has prominence — for instance (24), with the if-clause following the main
clause and the use of the expression if you’d like to. The interrogative examples
also lie upon a continuum, from examples such as (20) which have relatively
more directive force, to those such as (21), which may express a greater degree
of sincere wonderment or doubt.
If we consider all types of if-clause in the General Practice corpus according
to whether they were used to issue directives, the pattern in Table 2 emerges. 39
uses of if-clauses, or 17.6% of the total number of if-clauses in the corpus, had
this function. Just under half of these (48.7%) were the isolated type. Note that
as indicated above, the conditional category counts only those complete condi-
tionals in which the if-clause encodes the requested act; in a further 19 examples,
the main clause expressed a directive for which the if-clause specified a substan-
tive precondition on performance (for example if it’s not settling in a week come
back or give us a ring). These are not included here, since the degree of direct-
ness or politeness of the directive is independent of the presence of the if-clause.
Table 2. Directive if-clauses in the General Practice Corpus
Type of If-clause: Number:
Complete subordinate interrogative constructions 05
Complete conditional constructions 13
Directive isolated if-clauses 19
Incomplete 02
TOTAL 39
These figures can be compared to Ford and Thompson’s (1986) findings from a
quantitative study of a spoken American English database consisting of lectures,
presentations and task-based conversations. They found that the “polite directive”
type of example (not distinguishing those with consequent clauses from those
without) made up 9% of initial if-clauses in their conversational corpus, substan-
tially fewer than the 17.6% noted here (p. 365).4
Ford and Thompson found that the polite directive type of if-clause did not
occur at all in a written language database (consisting of three non-fiction books)
that they considered, and the current study found isolated directive examples only
in actual conversation or in represented conversation in plays and novels.
However, the isolated directive type may also occur in more conversation-like
written language, as an example from Corbett (cited in Evans Ms: 5) indicates.
The following is taken from a circular from a milkman about Christmas deliveries:
(25) If you would kindly indicate in the boxes below your requirements
and then hand the completed form back to your Roundsman by no
later than the 16th December 1995.
3.4 Isolated if-clause optatives
Isolated if-optatives are used to express a wish by the speaker. They may include
the intensifying adverb only. They may have exclamative force. Examples with
and without only will be discussed in turn.
No examples with only were found in the General Practice Corpus. Five
examples were found in the Macquarie Dictionary Corpus. All of these were in
286 LESLEY STIRLING
the novel, all were in the “free indirect style” of representing the thought of a
third person protagonist, and all were exclamative. Some examples are:
(26) If she could only go to the bottom of the dike now, with the men, and
spend the night with them, thigh-deep in the sweet water, catching
fish, saying nothing, looking out to sea! [FLA p.62]
(27) If only Kitty had not done everything without her! [FLA p.96]
(28) If only Miss Hawkins would get a job … [FLA p.195]
A second group of 11 examples occurred without the adverb only. Representative
examples are given below. Once again, these were found only in the Macquarie
Dictionary Corpus, but they had a wider distribution within it across genres, with
examples from fiction, drama and the transcription of the public hearing.
(29) OLIVE: It was all true, everythin’ I told her was true, an’ an’ she
didn’t see any of it.
ROO: Hon, don’t cry now, you couldn’t help it.
OLIVE: B-but if she could have seen just a little bit, so she’d
know. [SUM Act 3 Scene 1]
(30) He smiled shyly. “Oh-ho! That’s too much to ask. Otherwise, it’s
clerking in the public service, or teaching, is that worth it? If I’d
somewhere to go, some friend’s room.” [FLA p.205]
(31) MR D. NICHOLLS: […] But that’s where it’s failing, you know.
If there was someone there to say, “Well, we’ve got your kiddy here,
look he’s on the streets.” [TRANS p. 149]
(32) MR SHANNON: […] I believe if we can start with the children as
they walk, as they’re big enough to understand what goes on, if we
can get them out of the environment where sometimes unfortunately
there’s a drinking problem at home or there’s a matrimonial
problem at home and the kids are brought up in that environment,
and it’s not fair to the kids. […] [TRANS p. 89]
For both kinds, two subtypes can be distinguished depending on whether the
event in question is situated in non-past or past time. Thus, (27) and (29) are
counterfactual in that the time is past at which the event described could have
taken place. Either a past tense modal or perfect aspect is used to indicate this.
The other examples, regardless of whether the verb form is in the past or present
tense, describe events which could potentially still take place. All examples with
only from this corpus have past tense verbs; examples without only may have
past or present tense.
Parallel complete examples including a consequent clause may have a
similar communicative function in that they express a wish, as well as providing
the motivation for it. These require the presence of the intensifying adverb only
to have a clearly optative meaning. See for instance the constructed example (33).
(33) If he would only make up with his parents, he would be happier.
While the examples with and without only share a general optative function,
there are some differences in meaning and use between them. For instance, at
least some of those without only could not be rephrased to include it without
significantly changing their meaning. Examples (31) and (32) from the public
hearing are of this kind: rephrasing to include only adds the here unwanted
implication that the speaker (and possibly the hearer) are not in a position to do
anything to bring about the desired act. The if-clauses in these examples in
contrast are used to propose desirable courses of action presumably with the
hope in this context that something will be done to bring them about.
Thus, in noncounterfactual examples, “if only p” seems to convey some-
thing like the following:
1. p is not the case
2. speaker desires p to be the case
3. speaker believes it is unlikely that p will come about
4. speaker believes that neither speaker nor hearer can bring about p
It is clauses (3) and (4) which are evidently not conveyed by optative if-clauses
lacking the intensifier, and it is these which lead to the sense of frustration
conveyed by the if only examples.5
Optative isolated if-clauses have been discussed by Quirk et al (1985: 11.38,
11.41) who include them in the category of “irregular sentences”. Quirk et al.
focus on the exclamative quality of these and other sentences in this subcategory,
“the omission of the matrix clause [ … ] being mimetic of speechless amaze-
ment” (p. 841). They are also distinguished in the Oxford English Dictionary,
where examples of “exclamation” uses of two kinds are given dating back to c.
1000; one is the “wish or determination” kind discussed here (with and without
only) and the other a (non-optative) expression of surprise or indignation of a
kind which seems no longer to occur very frequently (If ever I heard the like of
that!; The wretch! If he has not smashed the window!). Akatsuka (1986: 336f.)
also briefly discusses similar examples to those considered above, with respect
288 LESLEY STIRLING
to the question of how we can figure out what the omitted consequent would be.
Parallel uses can be found in other languages, for instance Greenberg (1986:
259ff.) describes an independent or main clause use of the protasis form in
Ancient Greek called the “indicative of wish”.
4. The grammatical status of isolated if-clauses
The question to be considered here is to what extent the isolated if-clauses

exemplified above should be considered to be functioning as independent or main
clauses, and therefore as worthy of attention by a grammar as comprising distinct
sentence types. That is, to what extent are the uses exemplified conventionalized?
The alternative would be to treat them as one of many examples of ellipsis
which occur in speech, to be handled by the discourse or pragmatic component
of the language description.
Certainly these utterances have broadly similar functions to some of those
of complete conditional and interrogative if-clauses: both the latter may be used
to issue directives, and the optative use is paralleled in complete conditional
clauses. It is notoriously difficult to tell when a particular indirect illocutionary
form has become conventionalized to a sufficient extent to deserve the status of
a distinct sentence type (compare Sadock and Zwicky 1985: 192–3). I shall argue
that both types of isolated if-clause are well along the path of conventionaliza-
tion, the directive as well as the optative, which is effectively treated as a minor
sentence type in many descriptions already.
To some extent this question hangs on the difficult issue of the definition
of “sentence”, which depends on what counts as “completeness” (see for instance
the discussion in Matthews 1981, ch. 2). As noted in the discussion of the
isolated optative if-clauses, Quirk et al. (1985: 11.38, 11.41) use these to
exemplify one of three ways they identify in which sentences can be “irregular”,
that is, by being marked as subordinate. However, a second way in which
sentences can be irregular according to Quirk et al. is for them to be fragmen-
tary, with normally obligatory constituents recoverable from linguistic context, or
jointly constructed by the two participants in a dialogue; the examples discussed
earlier where the main clause was contextually supplied clearly fall into this
category. This makes their category of “irregular sentences” unhelpfully diverse.
It is relevant to distinguish between three cases: (1) incomplete utterances,
(2) elliptical sentences (where the ellipsis is governed by rule and the context
supplies the missing element), and (3) minor sentence types, which are in some
way unusual in clausal structure, such as by virtue of apparently containing a
marker of subordination. I would argue that the isolated directive if-clauses (and
indeed the isolated optative if-clauses) should be seen as (3) minor sentence
types rather than as “fragmentary” sentences of type (1) or (2), although their
similarity to complete if-constructions with similar functions indicates their origin
and explains their meaning.
First, the isolated if-clauses are prosodically complete, with terminal intonation
contour. In this they differ from the incomplete utterances described earlier.
Second, there are problems with analyzing these utterances as elliptical.
Definitions of ellipsis vary widely. Many, including Matthews (1981: 39) and
Quirk et al. (1985: 12.32–12.38), take as criterial whether the ellipted material is
verbatim recoverable from linguistic context. Matthews adds that the speaker
should be able to be made to complete the sentence, if necessary. Quirk et al.
assume a continuum of ellipsis, with the following criteria ranked in order of
least to most strict (and what they call “standard” ellipsis meeting all but 5.):
1. ellipted words are precisely recoverable
2. elliptical construction is grammatically “defective”
3. insertion of missing words results in a grammatical sentence of same meaning
4. missing words are textually recoverable
5. missing words are present in the text in exactly the same form
In the case of the isolated if-clauses considered here, postulated ellipted material
cannot be readily supplied from linguistic context, at least with any degree of
specificity. In each case it is clear what the range of meaning of the omitted
material may be. However, there is no way of determining it within more narrow
limits. In particular, there is no way to determine whether the directive isolated
if-clauses should be seen as elliptical upon a complete conditional or a complete
interrogative construction, given that both of these can have directive functions
(lack of a paraphrase with whether is unimportant given that there are other
structural differences between if and whether clauses; compare Huddleston 1988:
155). Evans (Ms: 5,10) comments that in Basque the two constructions are
formally distinct and both are available for use as isolated requests, and notes
that in English informants give both expansions for the isolated clauses when
asked. Apart from this difficulty, in either case, the meaning we would have to
assume to be supplied is of the most general kind. For the conditional case, it
amounts to an assurance that the act performed would have positive consequenc-
290 LESLEY STIRLING
es, most generally that it would fulfill the speaker’s wish for the act. For the
interrogative case, it amounts to an expression of doubt on the speaker’s part that
the act will be performed by the hearer, whether owing to inability or unwilling-
ness. In the case of the general practice consultations, the overall purpose of the
consultation is that the patient should reach a better health outcome, and in cases
where the isolated directive is treatment oriented, this is often the assumed
positive result, sometimes by way of more specific, condition-related outcomes.
Similarly, in many cases the broader context may indicate why the act described
in an optative isolated if-clause is desirable. Thus, it is understood in the
examples from the public hearing into Aboriginal Deaths in Custody that if the
desired acts were performed, children would do better at school or the problems
of drinking would be alleviated and the overall purpose of the inquiry — that
there would be fewer deaths in custody — would be furthered. However, this
kind of contextual support for utterances is a general phenomenon in discourse,
and fails to meet the criterion of recoverability required to define the utterances
as elliptical.
Furthermore, the meaning of the isolated if-clauses in themselves is contextually
unambiguous, and “shortcircuits inference” in the way that conventionalized uses
typically do (compare Brown and Levinson 1987: 290 n. 35).
Third, in general the uttering of a subordinate clause does not constitute an
illocutionary act (compare Huddleston 1984: 354), but here the supply of the
omitted component is not required for the utterance to constitute an illocutionary
act. Focussing on the directive examples, these utterances have the effective
illocutionary force of an indirect request, and as is usual with indirect speech
acts, they have distributional characteristics associated with their effective force.
For instance, they are likely to attract a response appropriate to a directive
speech act (for example explicit acquiescence or commitment to undertake the
act by the hearer). They readily cooccur with reason clauses which motivate the
intended illocutionary act, as in (34). This is unlikely with directive complete
conditionals, since in such cases the main clause often provides the reason.
Finally, they may belong to a chain of formulations of the same directive,
involving more direct commands as well as the indirect if-clause directive. This
kind of reformulation is particularly common in the HCRC Map Task Corpus,
where many examples like (35) occur. In contrast, the complete conditional
polite requests seem to retain as part of their meaning the assertion of a causal
relation between the two propositions mentioned. The isolated cases have shed
this element of assertion.
(34) D: So it’s actually half a teaspoonful of both of them um three

times a day and if you can pop her back Monday or Tuesday
just to let me know how she is going, OK? […] [GP A74:4]
(35) G: Yeah, if you go down … Do you have a great rock?
F: Nope.
G: Along from … Right okay, go down to the top of Indian
country.
Finally, at least the directive isolated if-clauses readily function as independent
clauses in complex and compound sentences. For instance, in example (36)
(observed at a university committee meeting) the adverbial clause introduced by
since provides a reason for the act proposed in the if-clause. If-clauses also
frequently coordinate with other clauses, including reasons for the requested act,
as in example (37). Example (38) shows that they may even cooccur with
hedging if-clauses of the kind described earlier.
(36) If I can explain what’s driving it, since I was at that meeting.
(37) D: […] Right if you just push up your sleeve and I’ll check your
blood pressure. […] [GP A76:8]
(38) D: There is a script for some Achromycin and uh and if you
would get back on the Actidel if you would.
P: Okay, mm.
As indicated in the Introduction, it has been noted that subordinate clauses may
be used as main clauses and that a process of grammaticalization may ensue in
which they become reanalyzed as conventional main clauses. Evans (Ms: 1) has
named this type of grammaticalization “insubordination”: “the conventionalized
independent (main-clause) use of a formally subordinate clause”. He gives a
typological survey of ‘insubordination’ and argues that it is widespread among
the world’s languages and has been a major historical source for a range of
grammatical categories. He identifies three functions of insubordinate clauses
crosslinguistically: (1) interpersonal coercion, especially commands; (2) modal
framing (recruitment of new tense/mood categories through a shift from relative
tense/mood (between clauses) to absolute tense/mood); and (3) marking of
various discourse contexts, such as negation or contrastive statements high in
presuppositionality.
Isolated if-clauses provide the most obvious exemplars of this phenomenon
in English, and the directive examples clearly fit into Evans’ first functional
292 LESLEY STIRLING
category (although it is less clear where the optative examples would fit). On this
analysis of them, they would lie somewhere on the scale of historical develop-
ment from an elliptical complex construction to an independent main clause.
Evans (p. 3) summarises the process of historical development as involving four
stages, listed below, and the examples discussed here would seem to lie some-
where around stages 3 and 4.
STAGE 1 Subordinate construction with overt main clause
STAGE 2 Ellipsis of main clause
STAGE 3 Conventionalization of ellipsis with certain syntactically permitted
reconstructions becoming excluded by convention
STAGE 4 Conventionalized main clause use of formally subordinate clause:
grammaticalization of construction with specific meaning of its
own; may not be possible to restore any ellipsed meaning
5. Conclusion
I have argued that isolated if-clauses functioning as directives should, like those
functioning as optatives, be analyzed as minor sentence types worthy of descrip-
tion within the grammar of English. In many respects they behave like indepen-
dent clauses, and crosslinguistic evidence supports the hypothesis that they may
be in the process of conventionalization as main clause usages.
There is a clear similarity in meaning between the directive and the optative
types. Both express the desire of the speaker in a context where the likelihood of
the act to be carried out is at issue. However they differ both formally and in
illocutionary force.
Directive isolated if-clauses occur almost exclusively in conversation,
although their frequency of occurence seemingly varies widely, with notable
differences between the general practice and map task dialogues. These clauses
fit the pattern of indirect directives, in that they explicitly leave open the
possibility of the hearer not performing the requested act. While they are
generally taken to be more ‘polite’ usages, their social meaning is probably better
characterized by examining their circumstances of use: their distribution in the
General Practice Corpus suggests that the nature of the act requested may be a
factor in determining their use, while the analysis of the map task data shows
that familiarity between the speakers is important.
If-clauses would appear to be the most obvious examples of “insubordina-

tion” in English — the reanalysis of an isolated subordinate clause as an
independent construction. Further work could usefully consider whether other
examples also occur, and whether certain types of subordinate clause are more
likely to give rise to such reanalysis (for instance, adjuncts rather than comple-
ments, since they are less tightly integrated into the structure of the clause).
Notes
* Thanks to Graham Barrington, Nick Evans and Roger Wales for helpful discussion of the
content of this paper.
1. I am grateful to Joan Mulholland for the use of the General Practice Corpus, which was
collected as part of her project “The language of doctor-patient communications: the develop-
ment of a methodology and a study in analysis”, funded by the Australian Research Grants
Commission. The directive isolated if-clause utterance type was first noted, for a larger and
overlapping subset of this corpus, by Mulholland and Stirling (Ms). For permission to use the
Macquarie Dictionary Corpus I am grateful to Sue Butler; many thanks are also due to James
Lambert and Mark Newbrook for helping me access it.
2. Henceforth, examples from the corpora are coded for their source. Thus “GP 1:9” indicates
general practice consultation 9 from doctor 1. Examples from the Macquarie Dictionary Corpus
are coded with the abbreviations: “FLA” for Christina Stead’s novel For Love Alone (1944),
“REM” for David Williamson’s play The Removalists (1972), “SUM” for Ray Lawler’s play
Summer of the Seventeenth Doll (1957), and “TRANS” for the transcription of the public hearing
into Aboriginal Deaths in Custody (1989). “[ … ]” indicates that material has been omitted
from the turn quoted. Isolated if-clauses are in italics.
3. Further, a contrasting use of isolated if-clauses, which is not attested in the corpora, is to issue
threats. These are also indirect directives: that the hearer not perform some act. An example
might be If you hit him … , spoken with appropriate intonation, where the consequent is left
unspecified, perhaps as a device to make the threat more frightening: “something bad will
happen”, the specific details to be filled in by the hearer’s imagination. These kinds of
examples seem to be less complete than the types discussed here, for instance they do not have
terminal intonation.
4. No breakdown was given for final conditional clauses, but it seems that this type was not
represented among them.
5. Such clauses are sometimes used in situations where the assumption that the speaker believes
the hearer will not be able to bring about p is cancelled. Compare If only you’d stop treating me
like a child!; If only I had a bicycle (by child to parent before Christmas). Such examples are
used in context to convey a very indirect request, and part of the indirectness is due to the
normal assumption that the speaker believes the hearer not to be able to bring p about.
294 LESLEY STIRLING
References
Akatsuka, N. 1986. “Conditionals Are Discourse-Based”. Traugott et al., 333–351.

Anderson, A., M. Bader, E. Bard, E. Boyle, G. Doherty, S. Garrod, S. Isard, J.
Kowtko, J. McAllister, J. Miller, C. Sotillo, H. Thompson, and R. Weinert.
1991. “The HCRC Map Task Corpus”. Language and Speech 34. 351–366.
Brown, P. and S. Levinson. 1987. Politeness: Some Universals in Language
Usage. Cambridge: Cambridge University Press.
Buscha, A. 1976. “Isolierte Nebensätze im Dialogischen Text”. Deutsch als
Fremdsprache 13. 274–279.
Davison, A. 1975. “Indirect Speech Acts and What to Do with Them”. Syntax
and Semantics, Vol. 3: Speech Acts ed. by P. Cole and J.L. Morgan,
143–185. New York: Academic Press, 143–185.
Evans, N. Ms. “Insubordination and Its Uses”. Department of Linguistics and
Applied Linguistics, University of Melbourne.
Ford, C.E. and S.A. Thompson. 1986. “Conditionals in Discourse: A Text-Based
Study from English”. Traugott et al., 353–372.
Greenberg, J. 1986. “The Realis-Irrealis Continuum in the Classical Greek
Conditional”. Traugott et al., 247–264.
Huddleston, R. 1988. English Grammar: An Outline. Cambridge: Cambridge
University Press.
Levinson, S. 1983. Pragmatics. Cambridge: Cambridge University Press.
Matthews, P. 1981. Syntax. Cambridge: Cambridge University Press.
Mulholland, J. and L. Stirling. Ms. “Directives in Doctor-Patient Discourse; an
Account of Speech Exchanges”. English Department, University of Queensland.
Sadock, J. and A. Zwicky. 1985. “Speech Act Distinctions in Syntax”. Language
Typology and Syntactic Description, Vol. 1: Clause Structure ed. by A.
Zwicky, 155–196. Cambridge: Cambridge University Press, 155–196.
Traugott, E.C., A. ter Meulen, J.S. Reilly and C.A. Ferguson eds. 1986. On
Conditionals. Cambridge: Cambridge University Press.
Functional and structural
The practicalities of clause knowledge in language

education
Lynn Wales
1. Introduction
In this chapter, the role of structural knowledge in language processing is

reviewed, with particular reference to the clause. This is followed by discussion
of the importance of the clause to language educationalists, especially to those
who favour the functionalist perspective proposed by Halliday. The focus then
turns to classroom implications, showing the need for knowledge of clause
structure on the part of teachers and its lack of availability in classroom texts.
The influence of Systemic Functional Grammar (SFG) in some government
materials for teachers is discussed and the presence of other models is shown.
Finally, the fusing of functional and structural approaches in language education
is illustrated by instances from discourse studies, second language teaching and
the reading process. This process, as well as implications of the earlier discus-
sions, demonstrates the need for a sound model of the English clause that clearly
indicates grammatical relations and phrase structure. Tribute is then paid to
descriptive linguists who have developed this much-needed tool.
2. Parsing: automatic in receptive processes
Fodor (1983, dedication) quotes the observation by Garrett that inspired his book,
“What you have to remember about parsing is that basically it’s a reflex.” This
296 LYNN WALES
refers to the fact that practised native-speaker listeners parse the incoming speech
stream subconsciously and automatically.
The same automaticity of word recognition and syntactic parsing is seen to
be needed by fluent readers. Research on the reading process points to the need
to keep the reading rate at a minimum of 200 words a minute, if comprehension
is not to be affected adversely, a point made in Eskey and Grabe (1988). Given
the array of subskills involved in such a complex process as reading comprehen-
sion, the way to achieve a good reading rate is to have automaticized low-level
linguistic decoding skills since, if word recognition and syntactic parsing are
automatic reflexes, the reader is free to put in cognitive effort on higher-level
skills such as inferencing and interpreting, without slowing down the reading rate
and thus demanding too much of the memory.
Word recognition studies consistently provide robust evidence that the best
single discriminator between good and poor readers is good readers’ automatic
word recognition skills (Stanovich and Stanovich 1995). This means rapid,
accurate and effortless word recognition. While there is debate on whether
recognition of syntactic structures is similarly automaticized in fluent readers, it
is a logical conclusion from the research on speech perception and the reading
process that fluent reading also requires automaticized syntactic parsing, for if
cognitive effort were required for parsing it would have the same adverse effect
upon reading rate, and thus upon comprehension, as poor word recognition skills.
Perfetti (1990) argues that there is an initial automatic parsing of syntax in
reading, the results of which provide for the establishment of semantic represen-
tation and contextual interpretation.
In this cognitive model of the reading process, the syntactic knowledge used
for automatic parsing is said to be encapsulated, that is, not part of world
knowledge available for general cognitive operations, but merely available to the
language processing module, where low-level decoding processes take place.
Similarly, the linguistic decoding processes, including parsing, are said to be
impenetrable, that is, taking place without interference from knowledge in long-
term memory, or executive control processes. In short, it is an advantage for
fluent receptive language skills that syntactic knowledge should not be part of
world knowledge and that parsing should not require cognitive effort.
FUNCTIONAL AND STRUCTURAL 297
3. Writing: the need for conscious linguistic knowledge
It is commonly noted that in their early years of writing children’s language is

much the same syntactically and lexically as their spoken language. However,
Kroll (1981) observes that, after consolidating the technicalities of writing,
children typically go on to differentiate between their written and spoken
language, so that eventually they use certain structures in written work that they
either use hardly at all in speech or at least use less frequently. Thus as children
develop their writing over the school years they become more aware of formal
features peculiar to written language (see Perera’s 1990 study of children aged
from 8 to 12).
Writing and reading differ with respect to the encapsulation of linguistic
knowledge and processes mentioned above. So argue Perfetti and McCutchen
(1987), discussing schooled language competence in which the acquisition of
literacy plays the major part. While the same kinds of linguistic knowledge are
needed for both reading and writing, the writing process seems to require, as
more mature texts are called for, that linguistic knowledge be under the con-
scious control of the writer. Writing is a cyclic process in which writers generate
sentences using many sources of information, including their world knowledge
and their linguistic knowledge. In addition, writing is slower as a process, and
the text remains permanent, thus being available as a source of information for
further text generation or for monitoring and editing.
Bereiter and Scardamalia (1993) discuss two models of the writing process.
The simpler version, known as the knowledge-telling model, describes what they
call the ordinary ability to put one’s thoughts and knowledge into writing. The
authors observe that this model reflects the early years of children’s writing. It
is characterized by little in the way of goal-setting, planning or problem-solving
(except at a local level), little awareness of the need to consider the reader’s state
of knowledge and the tendency to avoid text revision or to confine it to cosmetic
language changes. Coherence is achieved more from overall adherence to topic
than intersententially, the authors note, and from adherence to basic discourse
structures, although even in secondary school years significant numbers of
students fall short of achieving argumentation in exposition and plot development
in narrative, as observed for example in work produced for the National Assess-
ment of Educational Progress in the USA.
The authors go on to observe (judging, for example, from texts produced
and from think-aloud protocol evidence) that this simpler model is not merely a
298 LYNN WALES
reflection of the early writing of children but may also stand as a model of the
approach of many adults, including some undergraduates. We may note that,
though in the authors’ terms this is the simpler model of the writing process, it
nevertheless requires knowledge of formal distinctions between spoken and
written forms, as mentioned above, if it is to be acceptable to a maturer reader-
ship. In fact the authors point out that older students with more elaborate
knowledge of discourse forms can produce more richly structured compositions
within this model, especially if they monitor and edit their work for appropriate-
ness to topic, genre, clarity, interest and the like.
Bereiter and Scardamalia go on to discuss a more demanding form of the
writing process, described as the knowledge-transforming model. This model,
which entails the simpler model as a subprocess, more appropriately reflects the
work of professional writers, or of people at advanced levels in any intellectual
discipline. It is seen as a problem-solving model, in which there is an interaction
between text-processing and knowledge-processing. Writing in this case is not
merely recalling what one knows and writing it down, but also discovering
through the writing process what it is that one wants to say. The problem-solving
lies in two areas: working out knowledge and beliefs on the one hand (content),
and working out appropriate modes of expression on the other (rhetoric).
Decisions in one area affect those in the other; hence the intensely interactive
and demanding nature of this type of composition process. (Those who engage
in this kind of writing can testify to the cognitive effort required.)
4. Organizing linguistic knowledge
Karmiloff-Smith (1986:175) argues that

the human organism (both linguistic and cognitive) incorporates a drive to
have control not only over the external environment (the input stimuli) but
also, and importantly, over its own internal representations and finally over the
intricate relationship between the two. (p. 175)
For language, this means that the learner moves from data-intake to internal data-
organization, in order to gain control over accessing and manipulating linguistic
knowledge for different language processes. One could imagine, for example, the
setting up of a category NOUN after a number of examples had been processed
from the input, thus facilitating access to the class NOUN in the lexicon for
receptive or productive tasks. Once established, this category could be the file for
new examples, and would not need adjustment to accommodate new data from
the input. On the other hand, the subsequent recognition of a pluralizing process,
for example in English, might require some restructuring of the file to link this
rule to file entries. Morphophonemic rules would cover most of the data in an
economical way, but exceptions would also have to be entered. Such restructuring
operations are argued to be recurring processes, whenever new learning occurs
that requires an adjustment in the internal organization to facilitate control.
5. Analysis and control for metalinguistic tasks
Bialystok and Ryan (1985) see two parameters in the organization of linguistic
knowledge, analysis and control, the need for either of which varies according to
the linguistic task involved. Analysis means the extent to which the speaker’s
linguistic knowledge is analyzed or categorized, and control refers to relative
command of procedures such as accessing and manipulating such knowledge. For
everyday conversation, as we have noted above, linguistic knowledge does not
need to be highly analyzed or controlled, since the decoding skills are automat-
icized. For these authors, reading and writing are seen to require higher levels of
analysis and control, and metalinguistic skills to require even higher levels of
both. Thus if learners are asked to identify an example of a word class with its
grammatical function in a sentence, and then to substitute another example of the
identified function and word class ( the sort of activity that might be needed in
some types of second-language grammar exercise), the authors argue that the
first part of the task places high demands on analysed knowledge, and the second
demands high levels of control.
Bialystok and Ryan’s model has been influential in second language
learning studies (see, for example, Birdsong 1989, Ellis 1990). If it is correct, it
can be seen that, for more mature text production within either of Bereiter and
Scardamalia’s writing-process models, but especially in the knowledge-transform-
ing model, metalinguistic skills akin to the one just described are regularly
needed for conscious monitoring and editing of written text forms.
Taking the research and arguments of these scholars cumulatively, it is reason-
able to conclude that linguistic knowledge which serves best in encapsulated
conditions for receptive language tasks, such as listening and reading, should
increasingly also be available for use by executive thought-control processes for the
purpose of writing, and particularly writing of a more demanding intellectual nature.
300 LYNN WALES
In short, for skilled writing, rhetorical, syntactic and lexical knowledge has to become
part of world knowledge, available for conscious manipulation by the writer.
6. Significance of the clause in parsing
Working on the assumption that parsing was a necessary part of the process of
speech comprehension, psycholinguists in the sixties and seventies asked
themselves what units of syntactic structure listeners were actually using when
they parsed. In a series of studies employing click methodology (see discussion
in Fodor, Bever and Garrett 1974 ), or probe latency techniques (for example
Caplan 1972), it was shown that there was an effect of clause boundaries upon
subjects’ accuracy in either locating click sounds or detecting probe words during
sentence processing. This effect, which was shown by tape-splicing techniques
not to be caused by any acoustic properties in the signal, can be taken as
evidence that listeners do analyse utterances into clauses as part of the process
of comprehension. The clausal constituents of sentences, in other words, are a
psychological reality for English native speakers. We can therefore say that the
clause structure of sentences is part of the subconscious syntactic knowledge
used for automatic parsing of incoming text in fluent listening and reading.
7. Reading more formal text
We have noted that the clause is a subconscious, but nonetheless real psychologi-
cal unit in the parsing of speech, and have assumed that this is also the case in
the reading process. To the extent that reading material stays close in form to
spoken language this does not present too great a challenge for young readers.
One may note, for example, that much of current popular children’s literature is
characterized by long stretches of dialogue and short paragraphs of description,
in contrast to earlier works which tended to have longer sections of sustained
prose. However, more formal written text, such as literary or expository prose
encountered in the later years of school, differs more markedly from speech, one
important difference being the use of syntactic structures that are not common
in everyday language.
Perera (1984, 1990) summarizes work in this area showing a number of
structures which have been found to present difficulty to school-age readers at
various levels. They include clauses which will not work using the accustomed
“Subject-Verb-Object = Actor-Action-Patient” parsing strategy, such as passive
verb clauses or those with fronting of non-subject clause constituents, such as
objects or adjuncts. Such clauses are more frequent in written English because of
the absence of the prosodic features which allow speakers to emphasize clause
constituents without change of order. Writers learn to adjust the default order of
clause constituents in order to preserve thematic continuity, draw attention, show
emphasis, give thematic prominence and so forth. Hence the exploitation of first
and last position in the clause and of passive structures, and also the use of cleft
and existential there clause types.
Another characteristic of written language which Perera (1984) stresses is its
decreased redundancy, compared with the high degree of repetition that is
tolerated in speech. The permanent and sustained nature of written text may
account for this difference. Thus there is a higher frequency in written English
of such structures as non-finite subordinate clauses, verbless subordinate clauses,
ellipsis and nominalizations, all of which decrease redundancy (but of course also
have the effect of compressing the text, and placing a heavier load upon the
parser because of the diminished number of linguistic cues).
It is reasonable to suppose that parsing skills for reading texts containing
such unfamiliar clause structures are not automaticized at first and only become
so with practice. For some students, however, the difficulties encountered may
be an obstacle to achieving such practice on their own, and some assistance from
teachers may be needed in reading more difficult texts.
8. The clause and language education
It was noted above that rhetorical, syntactic and lexical knowledge needs to be
accessible as part of general knowledge, to be used in producing more advanced
written texts. It is therefore particularly in the area of writing instruction that
teachers may well be expected to be able to discuss a range of linguistic forms.
The question arises as to how important knowledge of clause structure per se is
for the writing process in its various levels of complexity. Such knowledge must
include the various surface forms that clauses can take in sustained written text
(some of which were noted above), as well as the essential grammatical relations
that make up the basic clause types.
The 1994 Australian National Curriculum for English called for students to
302 LYNN WALES
have a knowledge of linguistic structures and features, but in relation to their use
in texts. This reflects the growing interest in text that has characterized the last
two decades of language education. In Australia this has been particularly
promoted by the genre school whose work is based upon Halliday’s model of
SFG. Similarly in the UK, the Language in the National Curriculum (LINC)
materials developed under the direction of Ronald Carter (1990a) contain
statements that the linguistic approach taken in the work is much influenced by
Halliday’s functional theories of language.
Halliday’s emphasis upon language in text and in social context is also
frequently cited as the inspiration for much work in discourse analysis, a field
which is claimed as a second source of inspiration in the LINC materials.
Discourse analysis has now also become a major area of research in the field of
Teaching English as a Second Language (TESOL). In general, study of spoken
discourse became important as a result of the development in communicative
approaches to language teaching, but more particularly studies of both written
and spoken discourse have proliferated with the development of specialist
teaching areas. English for Specific Purposes (ESP) includes such fields as
English for Science and Technology, English for Academic Purposes and English
for Business Purposes. The growing interest in discourse among TESOL
practicioners is reflected by the title of McCarthy’s (1991) work Discourse
Analysis for Language Teachers, a work in which, as in those referred to above,
Halliday’s proposals are frequently cited.
Halliday (for example 1985a,1985b) emphasises the importance of the clause.
By and large, therefore, the chapters that follow take as their domain the
traditional realm of syntax, the terrain from the sentence to the word. Gram-
matically, that is where the action is; and within that, the fundamental unit of
organization is the clause. (1985a:xxi)
The clause is the gateway from the semantics to the grammar. (1985b:66)
It is not surprising therefore to find that the clause is the subject of much
investigation in discourse analysis, whether it be analysis of cohesive devices
linking clauses in meaningful ways (Halliday and Hasan 1976), of the ordering
of clause elements in information packaging (for example Halliday 1985b), or of
logical relations between clauses and selection of particular clause types for
particular discourse purposes (e.g. Winter 1977, Hoey 1983).
One often discussed area of interest in discourse analysis, which is relevant
to the clause, is the difference that can be observed between speech and writing.
Halliday himself uses the clause as the focal point of his proposals concerning
the respective complexity of spoken and written language. Spoken language is
characterized by lexical sparsity (with grammatical intricacy), while written
language is characterized by high lexical density, “measured in the number (and
informational load) of lexical items per clause … ” (1985b:75) The clause, then,
is seen as focal in those areas of language education that are influenced by
communicative and discoursal perspectives.
9. Grammar teaching
Gathering together these several points from the fields of cognitive psychology,
psycholinguistics and language education, it is reasonable to conclude that
teachers need to enhance understanding of clause structure on students’ part to
assist them to read more demanding written texts with less familiar clause types.
In addition, students need to become metalinguistically aware of clause structure
and be able to exploit this knowledge during the writing process, as they
themselves are asked to produce more advanced texts. Some students, perhaps
many, will need assistance from the teacher to develop a working knowledge of
the possibilities in clause and sentence structure for their written texts. Does this
mean a return to grammar teaching?
Researchers concerned about students’ knowledge of language structure
usually share the misgivings expressed by many teachers about ushering in a
return to decontextualized grammar teaching, so when classroom approaches are
considered there is an effort to indicate methods that integrate teaching about
language with other curriculum activities, endorsing such practices as: teachers
reading aloud a range of narrative and expository texts to familiarize students
with the more challenging written structures (Perera 1984); discovery learning
through exploiting knowledge that students already have (Hudson 1992);
“exploring grammar in relation to extended, preferably complete spoken or
written texts.” (Carter 1990:119).
However, these and other researchers still see the need for teachers to be
able to discuss linguistic points even if only incidentally, as “new” structures are
encountered, or as a less-used structure appears to be useful for particular
expression in students’ writing. It follows that, while teachers may not engage
proactively in grammar instruction, they need a sound grasp of the structure and
functions of English language themselves, if they are to raise students’ awareness
304 LYNN WALES
on these topics. As far as the clause is concerned, what teachers therefore need
is a good clear description of clause structure, of the different clause types and
of their functions, both prototypical and in pragmatic variation. Given that
linguistics is not commonly part of the curriculum in teacher education programs,
and that relatively few teachers have been trained in linguistics in their under-
graduate programs, it is worth examining what resources are commonly available
to teachers who are seeking the grammatical knowledge they need.
10. Instruction about the clause in secondary school coursebooks
Beginning teachers may well rely on the information about grammar provided in
the school’s chosen textbook. This is all the more likely after a long period
(some 30 years, according to Perera 1994) in which little attention has been paid
in schools to the structure of English. Huddleston (1989) examined the grammar
sections in a wide range of secondary English coursebooks used in Australia. He
found that the majority of texts relied heavily upon traditional grammar, for
example offering definitions that confounded semantic and morphosyntactic
categories of description. Wales examined a smaller range, mostly of Year 8
texts, that had been published later than those surveyed by Huddleston (Laughren
and Wales 1996). Most of the texts examined made no mention at all of clauses
or elements of clause structure.
One text gave a much fuller account of language structure than the rest.
This is Text 1 in the Appendix where extracts from three coursebooks are
presented, showing some terms used and definitions or descriptions offered that
are relevant to the issue of clause and sentence structure. For the present
discussion, the relevant terms used in Text 1 are summarized as follows:
sentence, statement, command, question, subject, noun, pronoun, personal
pronoun, object, verb, tenses, continuous actions, perfect tense, simple and
compound verbs, finite and infinite verbs, helping verbs, noun, adjective,
adjectival phrase, possessive adjective, adverb, adverbial phrase, conjunction,
clause, simple and compound sentences, helping clauses and complex sentences,
and adjectival clauses.
This text was unusual within the survey sample in that it indicated the
existence of different kinds of simple sentence, and discussed compound and
complex sentences, indicating different types of subordinate clause. It also
mentioned “object”, in addition to “subject”. (See sample extracts in the Appendix.)
In a recent survey to discover what textbooks were being used across the
curriculum in secondary schools in Brisbane it emerged from data provided by
48 schools that one English text was more popular than the others. Extracts from
this text make up Text 2 in the Appendix. Grammatical terms used in this
textbook were: noun, word (base word), phrase, clause, sentence, paragraph,
verb, tense, adjective, adverb, participle. It will be immediately apparent that this
text, which is much more representative than Text 1 of the general extent of
information available in the coursebooks surveyed, covers far less ground. The range
of the terms listed above (which are treated very briefly) reflects the total extent
of discussion of grammatical structure in the book, which is 288 pages in length.
Finally, Text 3 in the Appendix has some relevant extracts from a text
published for Year 8 students in response to the demands of the 1994 Australian
National Curriculum for English. The following terms are used in this book :
subject, predicate, pronoun (personal, possessive, relative), preposition, sentence,
subject, verb, finite verb, infinitive, participle, auxiliary (helping) verb, tense,
subjunctive mood, indicative, connectives, phrase, main clause, subordinate
clause, sentence patterns. In terms of coverage this text comes somewhere
between the first two. It is akin to Text 1 in that its coverage of aspects such as
verb structure, and complex sentences is much more detailed than that of Text 2
but is less detailed than Text 1, for example in its discussion of clause types.
Features that the three coursebooks have in common are as follows:
1. They all remain within a traditional grammar framework, confusing seman-
tic and structural categories in their descriptions. In this respect the situation
seems to have changed very little from that observed earlier by Huddleston.
2. They all suffer from confusion on various issues. Text 1, for example,
appears to equate non-finite verb forms with infinitives only, Text 2 is very
misleading on participles, and Text 3 falls short in its list of English
auxiliary verbs.
3. None of them contain a clear and reliable description of clause structure; for
example, apart from the traditional grammar definitions of subject and verb
in all texts, and the mention (without explanation) of the terms “object” and
“predicate” (Texts 1&3 respectively), there is no indication of the grammati-
cal relations in the clause.
4. They all show confusion on the definition of a phrase and, not surprisingly
perhaps, contain no description of English phrase structure, or of how
phrases relate to clauses on the one hand and to words on the other.
306 LYNN WALES
The point of examining these texts is to illustrate clearly that the assistance they
offer teachers for the task of raising learners’ consciousness about clause
structure is not only sparse, but, sadly, also unreliable. (It must be emphasized
that the problems noted here are common to the majority of secondary English
course books reviewed in Huddleston (1989) and also to those in the study
reported in Laughren and Wales (1996).
11. Second language study
The sparsity and unreliability of information about clause structure in English

secondary coursebooks presents a particular problem when one considers that
many students are learning another language. At one end of the scale the
problem can be that the second language is presented communicatively, so that
students are required to focus attention on meaning and intuit the structures
themselves (much in the manner of first language acquisition). A lack of
awareness of how English works may well inhibit their ability to conceptualize
the grammar of the second language. Conversely, increased awareness of the first
language can enhance proficiency in the second (Lowe and Wales 1996).
At the other end of the scale the student may well receive grammatical
information about L2 from the teacher or the text book, particularly where forms
or structures in the L2 differ from those of English. Case forms and verb
placement in German are examples where a knowledge of clause structure is
involved, as is interrogative formation in Japanese or Mandarin. Comparison with
clause structure in English is an inevitable part of instruction at this point, and it
is common, accordingly, to hear second language teachers in schools complain
that they are the only ones teaching English grammar, or tertiary students
reporting that the first time they really learnt any English grammar in school was
when they studied another language. Given, however, the difficulty of access to
reliable information about language structure faced by English teachers, one can
understand why this situation persists.
Through a survey of Year 8 Modern Language texts (French, German,
Indonesian, Japanese and Mandarin) in a Brisbane school, undertaken to see what
the possibilities were of teaching grammar cross-linguistically in the English
curriculum, it emerged that it was possible to link the different languages in
grammatical discussion, if clause structure (that is, grammatical relations and
phrase structure) were used as a major unifying factor. It is also possible, of
course, to use functional perspectives, such as how questions are asked in

different languages, or how definiteness is expressed, and this perspective can be
incorporated. The point being made here is that such discussion needs to include
reference to forms, and clause structure proves to be a particularly useful
framework for describing similarities and differences across languages (see
Wales 1996).
12. Government guidelines and the functional model
Government publications are another possible source of assistance for teachers.

They may be commissioned, as in the case of the LINC materials, which were
originally commissioned by the UK government, or, as in the case of the
Queensland English syllabus guidelines (1994a, 1994b), they may be produced
by curriculum sections in government Departments of Education.
Such publications can differ considerably in their approach to assistance
with grammatical knowledge. The Australian National Curriculum for English,
for example, has no specific section to discuss what may be meant by linguistic
structures and features. The LINC materials, on the other hand offer a compre-
hensive glossary of grammatical terms, and the Queensland English syllabus
guidelines offer a short appendix on “key elements of grammar” (1994b). The
latter two documents are explicitly functional in their linguistic perspective.
Halliday (1985a: xix) describes a functional grammar as one that is pushed
in the direction of the semantics. It is of course the ultimate goal of all linguists
to show how structure and meaning relate in language. Structural models start
with forms and attempt to show how meanings are mapped on to them. Propo-
nents of functional models must of course also start with forms (what else is
there to work with?) in order to ascertain the kinds of meanings that are ex-
pressed in particular languages and/or in human language generally, but the
resultant analysis proceeds from meanings and seeks to show how they are
mapped on to forms.
This meaning-to-form direction has been a very attractive linguistic model
for many language educationalists, as noted above, since it seems more relevant
to their work. “Halliday has always placed meaning at the very centre of theories
of language and LINC supports that position” (Carter 1990a:6). The Queensland
syllabus is similarly based on the Hallidayan model, taking a functional view of
language that sees language users as making choices within the various systems
308 LYNN WALES
of language (1994(a): vii).

Given the human capacity for syncretism, polysemy or ambiguity of
language forms (in such structures as Visiting relatives can be dangerous), a
model which works in the meaning-to-form direction may become very complex,
and at times difficult to follow. Halliday himself comments that his theory
contains “a wealth of apparatus” (1985a: xix). Elsewhere he states that the clause
is impossible to define, and that there is no one right way of describing it
(1985b: 67). Certainly, the summarizing description that follows indicates the
considerable complexity of the apparatus:
The clause is a functional unit with a triple construction of meaning: it func-
tions simultaneously (1) as the representation of the phenomena of experience,
as these are interpreted by the members of the culture; (2) as the expression of
speech function, through the categories of mood … .; and (3) as the bearer of
the message, which is organised in the form of theme plus exposition. To each
of these functions corresponds a structural configuration (1) in terms of a
process (action, event, behaviour, mental process, verbal process, existence, or
relation) together with participants in the process and circumstances attendant
on it (‘Medium’, Agent, Beneficiary, Time, Cause, etc.): (2) in terms of an
element embodying an arguable proposition (Subject plus Finite) and residual
elements (Predicator, Complement, and Adjunct): (3) in terms of a thematic
element, given prominence as what the message is about, and a residual
element (summarised as the ‘Rheme’). In addition, (4) the clause provides a
reference point for the information structure in spoken discourse, closely
related to (3) — there is systematic interplay between the Theme-Rheme
organisation of the clause and the Given-New organisation of the information
unit (realised as a tone group) … . The ‘systems’ (sets of options that embody
the choices in meaning) that are expressed through the various functional
configurations (1), (2), and (3) above are, respectively, those of TRANSITIVITY,
MOOD and THEME … . A clause, then, can be defined as the locus of
choices in transitivity, mood and theme. (1985b: 67–68)
The above is, of course, a summary and information is particularly difficult to
assimilate in this form. The model is explained in detail, however, in Halliday
(1985a). It may be noted that the definition pertains to function and discourse.
The formal means by which these features are realised (for example verb group,
nominal group) have still to come into the picture.
13. A mixture of models
When one examines guidelines for English teachers that are avowedly based
upon Hallidayan proposals (such as the above for clauses), one finds a consider-
able depletion in the amount of the apparatus that is utilised. Under the heading
Clause Structure the Queensland guidelines contain a brief discussion of
Processes, Participants and Circumstances (no reference to Medium, Agent,
Beneficiary, Cause), a brief discussion of independent clause types under the
heading Mood and Modality (with a small paragraph on probability and obliga-
tion), and one paragraph each on Theme, Active and Passive voice, Nominali-
zation, and Cohesion (including reference and conjunction).
There is no mention of the terms Predicator, Adjunct, Rheme or Transitivi-
ty. The terms “subject”, “finite”, “object” and “complement” are used, but only
once (or twice) en passant without explanation; for example:
Declarative clauses…usually have the structure, subject (noun phrase) + verb
phrase + optional object or complement (noun phrase). (e.g., The cockatoo
screeched. Canberra is the capital of Australia.)… In yes-no and alternative
questions, the subject and the first part of the verb phrase, the finite, are
inverted (e.g. Did you hear it? Would you prefer tea or coffee?). (1994b: 52–3)
The key concepts that relate to sentence structure are held to include: sentence,
main clause, subordinate clause, simple sentence, complex sentence, compound
sentence, noun clause, adverb clause, adjective/relative clause, embedded clause
(p54). It is interesting to observe here that apart from the inclusion of embedded
clause (the latter term familiar from models of generative grammar) this list (and
indeed much of the description that accompanies it in the text) bears a consider-
able resemblance to the content of the more traditional descriptions in the
English coursebooks 1 and 3 discussed earlier. The model, in other words, is not
distinctively that of SFG.
Key concepts relating to phrase structure are stated to be: group of words,
head word, noun phrase, verb phrase, adjective phrase, adverb phrase, preposition
phrase. (p54). It can immediately be seen that this description is influenced by
modern linguistics in a way that the English coursebooks were not. The phrases
are distinguished as the types that make up clause constituents, and the notion of
“phrase head” is introduced. A similar advance is seen in the following section
on word classes where, in addition to the terms already used in the course books,
there are such terms as “content words”, “grammatical words”, “determiners”,
“inflect”, and “modify” (55–6). Again, the terms used and much in the descrip-
310 LYNN WALES
tions offered bears an observable resemblance to work on English language

structure by descriptivists such as Quirk or Huddleston. The model at this point
is again not distinctively Hallidayan.
A similar point can be made about the LINC glossary. While in the overall
introduction to the LINC materials (p. 2) the proposals of Halliday are declared
to be the inspirational model for the approach taken, when it comes to a linguis-
tic model for the glossary (p. 331) the authors acknowledge a debt to the work
of Quirk et al. (1985) and to three works of Crystal who, of course, used the
Quirkian model. The influence of this descriptive structuralist model is apparent
throughout the glossary; one might cite in particular the sections on phrase
structure, and on sentence and clause. These occur alongside sections on topics
arising from functionalist models (for example discourse analysis). Thus what
results is an approach which in fact synthesizes functional and structural models.
(Perhaps in line with this synthesizing approach adopted by language educa-
tionalists, a more recent text on SFG by Eggins (1994) glosses the Hallidayan
term “nominal group” as “noun phrase”.)
It may be that the apparatus proposed in SFG is considered too complex for
teachers who have no linguistic training. Or it may be in some cases that the
functional insights offered by SFG are considered more successful than the
proposals concerning structure (Huddleston 1996 points to some problems with
the latter). Whatever the reason, we can see that, although these materials for
teachers claim SFG as their informing model, they use other models as well and,
significantly, they use structural models.
14. Functional and structural
In this section three instances are discussed of areas of language education in

which we can see both formal and functional perspectives being incorporated and
integrated. The first is in the field of spoken discourse, the second in language
teaching and the third in the teaching of reading.
Several researchers have drawn attention in recent years to important
differences between speech and written language which need to receive greater
focus in linguistic descriptions, (for instance Miller 1994, McCarthy and Carter
1995). Carter and McCarthy (1995) observe from corpus data on spoken
language that there seem to be available pre- and post- clause slots that can be
used for topicalizing or reinforcing functions. Sentence examples are:
a. Well, Sharon, where I’m living, a friend of mine, she’s got her
railcard and …
b. This friend of mine, her son was in hospital and he’d had a
serious accident …
c It’s very nice that road up through Skipton to the Dales.
Carter and McCarthy argue, accordingly, that the grammar of the extended clause
for conversational language ought to have the following pattern sequence:
STRUCTURAL POTENTIAL OF THE CLAUSE IN CONVERSATION

Sequence pre-clause clause post-clause
Function TOPIC CORE TAIL
Constituents S-tp/O-tp/RI S/V/O/C/A TAG/S-tl/O-tl
Forms NP/NFC NP/VP/ADVP NP/VP
(S-tp and O-tp = subject or object in the topic slot: S-tl and O-tl = subject or
object in the tail slot: RI = related item: and TAG = any of the types of English
sentence-tags. S/V/O/C/A represent the core clause constituents (subject, verb,
object, complement, and adjunct), which may be re-ordered for a variety of
thematic and focusing purposes. NP = noun phrase: NFC = non-finite clause: VP
= verb phrase: ADVP = adverbial phrase.) (extract from p. 152).
Clearly this is an interesting area with regard to discussion of clause
structure in general. With regard to the issues in this chapter, however, there are
two relevant points. The first is that, with some small adjustments, the linguistic
model used by Carter and McCarthy for the core of the clause could well be that
of Quirk et al. (1985). The second and related point is that this work constitutes
a good example of the benefits of linking the perspectives of function and
discourse to a clear structural model. It is obviously useful and appropriate to use
authentic spoken data, and to observe patterns therein that do not occur so much,
if at all in some cases, in written language. It is equally important to be able to
demonstrate the point with a model of clause structure that clearly distinguishes
discourse functions, clause relations (called constituents here) and phrase
structure (forms)
The eighties saw a flourishing of communicative approaches to language
teaching. The development of such approaches in the UK is attributed to the
work of applied linguists such as Candlin and Widdowson who drew their
inspiration from, among others, the functional linguistic emphasis upon language
312 LYNN WALES
as a communicative social tool (Richards and Rogers 1986 )

In the nineties there has been a preoccupation among many proponents of
Communicative Language Teaching (CLT) with the challenge of how to
incorporate input enhancement, or focus-upon-form, into second language
programs. This has come about because of the realization that a focus on
communication and meaning alone in second language programs seems to
encourage fossilization (the habitual production of entrenched non-native-like
errors) in learners, as evidenced, for example, in the results of the Canadian
French immersion programs (Hammerly 1991). The concern is to preserve the
perceived benefits of CLT (fluent, confident communication in a range of
settings), while assisting learners also to notice the forms of the language (see
for example Lightbown and Spada 1990).
As with school English programs for native speakers, there is a deep
concern not to return to former practices, such as decontextualized grammar
exercises, and so a range of methods are being sought that will help learners to
focus on form without impairing the development of communicative skills.
Nevertheless teachers themselves are being expected to know more about
grammatical structure than during the heyday of CLT. In any case, foreign-
language teachers, whose time allotment is insufficient to allow students to learn
inductively, are thereby constrained to give explicit instruction on the forms of
the language. Also, ESL teachers who have to teach literacy skills must try to
help students to gain an automaticized knowledge of English language structure
for reading, and a conscious knowledge of the same for writing.
With regard to models of the reading process, it is worth noting the dispute
of many years which in some ways mirrors the situation in language teaching.
Among the various models of the reading process, the one which has had the
greatest influence upon educationalists is that of Goodman (see for example
1967, 1976). In Goodman’s model the assumption is that good readers engage in
a psycholinguistic guessing game in which they use their knowledge of context
(world knowledge, syntactic and semantic knowledge) to help them recognize
words in the text, thus not needing to rely so much on the graphic display. Poor
readers have problems because they do not use the linguistic and semantic
context to assist them with word recognition. This model has given rise to the
whole-language approach to the teaching of reading, which encourages whole
word recognition (including guessing from context) rather than learning to make
phoneme-grapheme correspondences, and reading strategies such as guessing
(predicting, anticipating) what the words will be.
However, a large number of studies have continued to show that backward

readers do use context, that they use it to work out the meanings of particular
words and phrases, and that they do so probably more than the average reader.
Thus they know that it can help them work out word- and phrase- meanings. To
sum up the results of word recognition research in first language reading, good
readers are not good simply because they are better predictors or make better use
of context. “It is simply not true that good readers take decoding lightly: they
fixate almost every content word”. Contrary to Goodman’s hypothesis, “what is
really wrong with poor readers is that they recognize isolated words inaccurately
and far too slowly, and compensate for their lack of decoding skills with context-
dependent guessing or hypothesis testing… Good readers with their superior decoding
skills can decode letters and words rapidly in a bottom-up fashion and therefore do
not normally need to resort to guessing strategies”. (Van Dijk and Kintsch 1983).
Stanovich (1991) explains the paradox that highly skilled readers appear
better able to exploit context to enhance their comprehension of a passage than
do less skilled readers. On the other hand highly skilled readers appear to make
less use of context to aid word recognition. Not only do skilled readers not need
to rely on context for normal word recognition, it is faster and more efficient for
them to avoid relying on context. Thus it is important to teach phoneme-graph-
eme correspondences, to help students to automaticize word recognition skills.
The model of reading advocated by researchers like Stanovich is called the
interactive model. It views reading as an interactive process between low-level
decoding skills and higher-level skills of inferencing, interpreting etc. This
proposal encompasses both the knowledge of language forms for decoding and
the use of context (including the rest of the text) for interpreting the text on-line
and globally. Grabe (1991) shows the relevance of the interactive model for the
teaching of second language reading.
In these three instances we see researchers concerned with different areas
of language education bringing together functional and formal aspects of
language and language processing. The last two instances in particular indicate
the importance of formal linguistic knowledge (however it may be acquired) in
language learning and processing.
15. Contribution of research on structure
It is important that assistance given to teachers concerning language structure be

as clearly expressed and as reliable as possible. Where the description is not
314 LYNN WALES
clear confusions can easily arise. For example, in the Queensland guidelines
(1994b:54) we are told that a noun phrase
is a group of words which may realise a Participant. It comprises a noun as
head word (underlined) and, optionally, dependent words such as articles and
adjectives (e.g., the new motor-car, sixteen candles, a once-in-a lifetime
opportunity, these very old clothes, some recent novels with lurid covers). Noun
phrases can consist of a single word (e.g., Digger), a pronoun (eg., we) or of
a noun clause (e.g., That rain had been forecast helped us make to our minds).
To make a few comments: we do not seem to be told what a noun phrase does
when it does not realise a Participant; articles are not always optional (hence the
problem with the single word NP example, Digger); the structure of the English
noun phrase is not really dealt with (e.g. pre- and post-modification); noun
clauses are not noun phrases. How is it possible for us to make these points?
Because research on the structure of English has provided the necessary informa-
tion and analysis to enable us to discern where and how descriptions fall short.
What is needed here, as the confusion over noun phrases and clauses shows,
is a model of clause structure that distinguishes clearly between grammatical
relations (subject, object and the like.) on the one hand, and constituent structure
(phrase, clause) on the other. We then do not need to say that a noun phrase can
consist of a noun clause, but we can say that both can play the part of subject.
It is important that teachers have a good grasp of the grammar, because they are
the ones who will be clarifying points for students. The value of a reliable
structural model that is accessible to teachers becomes obvious.
For practical purposes also, it is important to have a clear and reliable
model of clause- and phrase- structure. For example, assuming the validity of
Halliday’s proposals on lexical density, if teachers want to calculate the lexical
density of a text (in order, for example to ascertain the relative complexity of
potential teaching materials) what they need for this task is a sound knowledge
of word classes and clause boundaries, not what clauses mean. That is to say, if
one is to divide the number of lexical items by the number of clauses in order to
assign mean lexical density to a text, it follows that one must have a reliable way
of defining both lexical items and clauses, in other words a clear grasp of clause
and phrase structure. (Greenbaum and Nelson (1995) discuss structural factors
involved in undertaking measurements of complexity.)
16. Summary and conclusion
We have seen that the clause is a psychological reality in speech perception, and
that automaticized parsing of clause structures must also be part of decoding
skills in fluent reading. For more formal texts this means that students must
acquire knowledge of some ‘new’ structures, if they are to automaticize them.
Moreover, metalinguistic awareness of clause structure is needed for monitoring
and editing of text in the writing process, the more so as writers engage in
composition of the knowledge-transforming type.
In order to help students with both these challenges, and also for their own
professional competence in gauging suitability (in terms of levels of complexity)
of written materials for particular student groups, English teachers need a sound
knowledge of clause structure and of its use in discourse. So also do second and
foreign language teachers. Even within CLT circles there is now a greater
acknowledgement of the need for some means of getting learners to focus on
language forms. The teaching of L2 literacy skills in particular calls for a good
knowledge of clause structure and its variations.
For all these reasons it is important that there be clear and accessible
models of how the structure of English works, how it relates to meaning and is
used in discourse, and how it relates to the way other languages work. Tribute is
rightly paid by language educationalists to Halliday’s functionalist perspective
which has done so much to promote interest in communication and discourse.
The world of language education should likewise acknowledge its debt to
the careful and perceptive work on English language structure, carried out over
many years, by scholars of the calibre of Quirk and Huddleston. This chapter has
been written with the intention of demonstrating some of the many reasons why
a lucid and comprehensive account of clause structure is needed in language
education. We can be grateful that such accounts have been provided, and that
language education can therefore benefit from the development of this most
useful, practical tool.
Appendix: Selected Extracts on Language Description from Three Secondary English coursebooks
TEXT 1
Sentence: A sentence can be a statement, a question, a command or an explanation . . . A sentence
must have a subject and a verb.
Statements: Most sentences are statements, they must be complete and must make sense.
Commands: are a special type of sentence in which the subject is not written but is understood. The
316 LYNN WALES
subject of command sentences is really you.

Subject: The subject of the sentence is the main person or thing being spoken about. There should be
agreement between subject and verb.
Finite and infinite verbs: A finite verb is a complete verb: it has a subject and a tense. Infinite verbs
don’t have a subject and have no tense: they are easily recognized because they usually have the
word to in front of them.
Adjective: tells more about a noun or pronoun. It is sometimes called a describing word: it describes
the person, animal, object or behaviour, or says what kind it is.
Adjectival phrase: A phrase is a group of words that has a meaning but does not have a finite verb.
An adjectival phrase is a phrase that does the same work as an adjective. It tells you more about a
noun or pronoun. (Examples follow with claimed adjectival phrases in italics. They include: “The guy
with the high-heeled boots jumped the fence.” “Having moved quickly to the scene, she grabbed the
fighting girls.” Further examples include: “splattered in the air”, “carefully curled and set”, “having
been stuck into his backside”)
Conjunction: is a joining word: it joins words, groups of words or sentences together. (Examples are:
and, but, (either) or, (neither) nor.
Clause: A clause is like a sentence within a sentence. It is a statement that contains a subject and a
finite (complete) verb and is part of a larger sentence.
Simple and Compound sentences: Simple sentences contain a subject, verb and often an object . . . .
A simple sentence may contain a number of phrases but has only one finite verb. (Examples include:
“You would laugh at anything.: You is the subject, laugh the verb and anything is the object.” “We
laughed at John Cleese pretending to be nuts.: Laughed is a finite verb, at John Cleese and pretending
to be nuts are phrases.”) Compound sentences contain at least two clauses: each clause contains a
finite verb. The clauses are joined by a conjunction.
Helping clauses and complex sentences: Complex sentences are made up of a main clause and one or
more helping clauses. These helping clauses usually do the work of adjectives or adverbs.
Adjectival clauses: describe nouns (usually begin with relative pronouns).
Adverbial clauses: usually tell more about verbs: begin with conjunctions (conjunctive adverbs)
(Examples given)
TEXT 2
The Label — The Noun: the words we use to name things (person, place, thing, event, idea, feeling)
(proper nouns and common nouns distinguished).
Phrases: Phrases are two or three words spoken together. They are short: e.g. more drink. A phrase
can be even one word: e.g. drink. (Examples of phrases are: warm and cosy, nice smells, end of the
street. Except in poetry or informal talk phrases not used by themselves, because do not tell whole
story.)
Clauses: (most have three main parts): 1. The action: All clauses have a verb. The verb tells us what
is happening. 2. The doer: In most clauses there is someone or something doing the action. 3. The
where, how or when: In many clauses the action happens at a certain place or in a certain way or at
a certain time. E.g.
Linda (doer) struggled (action) with the question silently (how):
The pilot (doer) battled (action) against the storm over Perth (where).
Sentences: do not need to be long: can be just one clause. E.g. Adnil faded and vanished.
The Action Maker — The Verb: The aim of the verb is to add action and meaning to what we say and
write . . . . The verb is the central action of any sentence. A verb can be one or more words. E.g ‘I
yelled’ and ‘I will yell’. Verbs can also change the time the action takes place : ‘The crowd
scattered’: Past tense. ‘The crowd is scattering’: Present tense. ‘The crowd will scatter’: Future tense.
Action Words — Participles: (Some examples are given first.) The author has made good use of
action verbs and words we call participles to create this atmosphere. The words dodged and ran are
verbs, whereas the other action words — twisting, turning, dodging, swerving — are not verbs but
participles. Usually participles are attached to a verb but sometimes they are used independently of
the verb, as they are in the example above. They are effective in adding action to the story.
TEXT 3
Subject and predicate: Only the verb is distinguished in the predicate.
A relative pronoun: is a word which helps you link or relate ideas about a noun . . . . The five
relative pronouns are who, whom, whose, which and that. Relative pronouns are always placed in a
sentence as near as possible to the noun they are related to. This is to prevent possible confusion of
meaning or ambiguity
Sentences: are extraordinarily difficult to define! However, three characteristics or ‘rules’ are worth
bearing in mind: 1. A sentence must have a finite verb (Some sentences contain only the finite verb,
such as the sentence Listen! ). 2. A sentence must have a subject (a person or thing doing the action
described by the verb). 3. A sentence must contain a complete idea and therefore make complete
sense. ( It is the finite verb that is most crucial in doing this, because it explains what is happening.)
Verb: The most basic rule of English grammar is that a sentence must have a verb . . . . The most
important (verbs) for you to know are the infinitive, the finite, the participle and the auxiliary
verbs.The infinitive form can be preceded by the word to and is the form used in dictionaries (e.g.
see, walk, be, accept, get).
The finite: the key verb form used in sentences. It changes slightly depending on number and tense.
(Illustrative tables follow using the verbs be and walk.)
Participles: cannot be used on their own: they must be accompanied by a special kind of finite verb
called an auxiliary verb . . . to make complete sense in a sentence. Participles end in either -ing, -(e)d
(e.g.walked, heard ) or -(e)n (e.g. chosen, grown.) (Examples follow.)
Auxiliary or ‘helping’ verbs: are used with participles to create sentences. They are also finite verbs.
The auxiliary verbs are may, can, will, be and have.
Subject: We need to know . . . . who was performing the action described by the verb. This is the
subject of the sentence. . . . To find the subject ask yourself who or what is performing the action
described by the verb. (Remember that in the case of the verb to be, it‘s not so much action as
presence that is being described.)
Phrase: A phrase is a group of words without a finite verb: (Examples: A hearty meal, Having missed
the bus, For ever) Phrases do not make complete sense on their own, because there is no finite verb
to say what is actually happening. Therefore phrases are not sentences.
Clause: A main clause is a group of words containing a finite verb. (Examples: It was a hearty meal.
Having missed the bus I was late for school. I will love you for ever.) As you can see a main clause
can be regarded as a sentence because it contains a whole idea. A subordinate clause is a group of
words which may contain a finite verb but doesn’t make complete sense on its own because it
318 LYNN WALES
depends on (or is subordinate to) a main clause. (Several examples follow, among which are: While
I waited for the next bus (subordinate clause), I witnessed a car accident (main clause). Waiting for the
next bus (present participle, no subject), I witnessed a car accident (main clause). Having finally
rigged the boat (auxiliary and participle, no subject), we were able to set sail (main clause)) . . . . A
main clause and its subordinate clause can usually go in any order, by the way.
References
Australian Government. 1994. A Statement on English for Australian Schools.

Carlton, Victoria: Curriculum Corporation.
Bereiter, C. and M. Scardamalia. 1993. “Composing and Writing”. Teaching
Literacy: Balancing Perspectives ed. by R. Beard, 155–175. London: Hodder
and Stoughton.
Bialystok, E. and E.B. Ryan. 1985. “A Metacognitive Framework for the
Development of First and Second Language Skills”. Metacognition, Cogni-
tion, and Human Performance ed. by D.L. Forrest-Pressley, G.E. MacKin-
non, and T.G. Waller, 207–252. Orlando: Academic Press.
Birdsong, D. 1989. Metalinguistic Performance and Interlinguistic Competence.
New York: Springer-Verlag.
Caplan, D. 1972. “Clause Boundaries and Recognition Latencies for Words in
Sentences”. Perception and Psychophysics 12.73-76.
Carter, R. ed. 1990a. Language in the National Curriculum: Materials for Profes-
sional Development. Dept. of English, University of Nottingham.
Carter, R. ed. 1990b. Knowledge about Language: and the Curriculum. London:
Hodder and Stoughton.
Carter, R. and M. McCarthy. 1995. “Grammar and the Spoken Language”.
Applied Linguistics 16.141–158.
Eggins, S. 1994. An Introduction to Systemic Functional Grammar. London:
Pinter.
Ellis, R. 1990. Instructed Second Language Acquisition. Oxford: Basil Blackwell.
Eskey, D. and W. Grabe. 1988. “Interactive Models for Second Language
Reading: Perspectives on Instruction”. Interactive Approaches to Second
Language Reading ed. by P. Carrell, J. Devine and D. Eskey, 223–238. New
York: Cambridge University Press.
Fodor, J.A. 1983. The Modularity of Mind. Cambridge, MA: Massachusetts
Institute of Technology Press.
Fodor, J.A., T.G. Bever and M.F. Garrett. 1974. The Psychology of Language: An
Introduction to Psycholinguistics and Generative Grammar. New York:
McGraw-Hill.
Goodman, K.S. 1967. “Reading: a Psycholinguistic Guessing Game”. Journal of
the Reading Specialist (May).126–135.
Goodman, K.S. 1976. “Behind the Eye: What Happens in Reading”. Theoretical
Models and Processes of Reading (2nd. edition) ed. by H. Singer and R.B.
Ruddell, 259–272. Newark, Del: International Reading Association.
Grabe, W. 1991. “Current Developments in Second Language Reading Re-
search”. TESOL Quarterly 25.375–406.
Greenbaum S. and G. Nelson. 1995. “Clause Relationships in Spoken and
Written English”. Functions of Language 2.1–21.
Halliday, M.A.K. 1985a. An Introduction to Functional Grammar. London:
Edward Arnold.
Halliday, M.A.K. 1985b. Spoken and Written Language. Victoria, Australia:
Deakin University Press.
Halliday, M.A.K. and R. Hasan. 1976. Cohesion in English. London: Longman.
Hammerly, H. 1991. Fluency and Accuracy: Toward Balance in Language Teach-
ing. Clevedon: Multilingual Matters.
Hoey, M.P. 1983. On the Surface of Discourse. London: Allen and Unwin.
Huddleston, R.D. 1989. “English Grammar in School Textbooks: Towards a
Consistent Linguistic Alternative”. Occasional Paper No. 11, Applied
Linguistics Association of Australia.
Huddleston, R.D. 1996. “What is an Appropriate Model of Grammar for Lan-
guage Teachers?” Who’s Afraid of Teaching Grammar? ed. by C. Arbones
Sola, J. Rolin-Ianziti and R. Sussex, 59–70. Conference Proceedings, June
1995: University of Queensland,.
Hudson, R. 1992. Teaching Grammar: A guide to the National Curriculum.
Oxford: Blackwell.
Karmiloff-Smith, A. 1986. “Stage/Structure versus Phase/Process in Modelling
Linguistic and Cognitive Development”. Stage and Structure: Reopening the
Debate ed. by I. Levin, 164–190. Norwood, NJ: Ablex.
Kroll, B.M. 1981. “Developmental Relationships between Speaking and Writ-
ing”. Exploring Speaking-Writing Relationships: Connections and Contrasts
ed. by B.M. Kroll and R.J. Vann, 32–54. Urbana, Illinois: National Council
of Teachers of English.
320 LYNN WALES
Laughren, M. and M.L. Wales. 1996. “A Cross-linguistic Approach to Teaching

Grammar: the Role of Linguistics”. Who’s Afraid of Teaching Grammar? ed.
by C. Arbones Sola, J. Rolin-Ianziti and R. Sussex, 71–81. Conference
Proceedings, June 1995: University of Queensland.
Lightbown, P.M. and N. Spada. 1990. “Focus on Form and Corrective Feedback
in Communicative Language Teaching: Effects on Second Language
Learning”. Studies in Second Language Acquisition 12.429–448.
Lowe, M.A. and M.L. Wales. 1996. “Language Awareness in L1 and its Impact
upon Progress in L2”. Who’s Afraid of Teaching Grammar? ed. by C.
Arbones Sola, J. Rolin-Ianziti and R. Sussex, 121–128. Conference Proceed-
ings, June 1995: University of Queensland.
McCarthy, M. 1991. Discourse Analysis for Language Teachers. Cambridge:
McCarthy, M. and R. Carter. 1995. “Spoken Grammar: What is it and How Can
We Teach it?” English Language Teaching Journal 49.207–218.
Miller J.E. 1994. “Speech and Writing”. The Encyclopedia of Language and
Linguistics ed. by R. E. Asher and J.M.Y. Simpson, 4301–4306. Oxford:
Pergamon.
Perera, K. 1984. Children’s Writing and Reading: analysing Classroom Language.
Oxford: Basil Blackwell.
Perera, K. 1990. “Grammatical Differentiation between Speech and Writing in
Children aged 8 to 12”. Knowledge about Language: and the Curriculum ed.
by R. Carter, 216–233. London: Hodder and Stoughton.
Perera, K. 1994. “Linguists, Teachers and Politicians: Current Controversies over
English Teaching in Britain”. Linguistic Society of America Conference
Jan.6th, Boston, MA.
Perfetti, C.A. 1990. “The Co-operative Language Processors: Semantic Influenc-
es in an Autonomous Syntax”. Comprehension Processes in Reading ed. by
D. Balota, G. Flores d’Arcais and K. Rayner, 205–230. Hillsdale, NJ: Law-
rence Erlbaum.
Perfetti, C.A. and D. McCutchen. 1987. “Schooled Language Competence:
Linguistic Abilities in Reading and Writing”. Advances in Applied Psycho-
linguistics 2: Reading, Writing and Language Learning ed. by S. Rosenberg,
105–141. New York: Cambridge University Press.
Queensland Government. 1994. English in Years 1 to 10: Queensland Syllabus
Materials. (a) “English Syllabus for Years 1 to 10”. (b) “A Guide to
Teaching in English”. Brisbane: Department of Education, Queensland.
Quirk, R.J., S. Greenbaum, G. Leech and J. Svartvik. 1985. A Comprehensive

Richards, J.C. and T.S. Rodgers. 1986. Approaches and Methods in Language
Teaching. Cambridge: Cambridge University Press.
Stanovich, K.E. 1991. “Word Recognition: Changing Perspectives”. Handbook of
Reading Research (Vol.2) ed. by R.Barr, M.L.Kamil, P. Mosenthal and P.D.
Pearson, 418–452. New York: Longman.
Stanovich, K.E. and P.J. Stanovich. 1995. “How Research Might Inform the
Debate about Early Reading Acquisition”. Journal of Research in Reading
18.87–105.
Van Dijk, T.A. and W. Kintsch. 1983. Strategies of Discourse Comprehension.
New York: Academic Press.
Wales, M.L. 1996. “Developing a Cross-linguistic Grammar Programme”.
Journal of Language Awareness 5.196–218.
Winter, E.O. 1977. “A Clause-relational Approach to English Texts: a Study of
some Predictive Lexical Items in Written Discourse”. Instructional Science
6.1–92.
Subject Index
A British English, 256

absolute with-construction, 78–9 Brown University Corpus, 79
accusative (case), 73, 189 by-phrase, 59
accusative and infinitive, 187–227
adjacency, 230
adjective, 53 C
adjunct, 93–110 Categorial Grammar, 187, 193–7
coordination with complement, 95, c-command, 28, 35
101–2 Chinese, 184
extraposition of adjunct, 95, 102–4 clause, 149–52
loose adjunct, 231, 233, 248 clausal complement, 188
prepositioning of adjunct, 95, 104 clitic, 86
relative ordering, 94, 96 co-indexing, 35, 40
restrictions on occurrence, 94, 96–7 collective noun, 104
restrictions on use of substitute one, command, 235, 236
95, 97–8 communicative language teaching, 312
semantic interpretation, 95, 98–101 competence vs performance, 44
adverb, 133, 178–9, 183–4 complement, 59–60, 93–110
agreement, 246 coordination with adjuncts, 95,
ambiguity, 98–9, 106–8 101–2
Ancient Greek, 288 extraposition of complement, 95,
apposition, 230, 237, 240, 243–8 102–4
appositive, 107 prepositioning of complement, 95,
Australian English, 273 104
relative ordering of complement,
94, 96
B restrictions on occurrence, 94, 96–7
backchannel, 113 restrictions on use of substitute one,
Basque, 289 95, 97–8
be, 75 semantic interpretation, 95, 98–101
binding, 35, 36 complex predicate, 191
324 SUBJECT INDEX
complex sentence (see sentence) Emergence Theory, 134, 144–6

composite sentence (see sentence) enclitic, 86
compound sentence (see sentence) end-weight, 240
conditional construction, 274 Ensemble Theory, 2
conservativity (of quantifiers), 5 entailment
constraint, 230, 238–40 downward entailment, 17
contextualization, 159 upward entailment, 17
contraction, 177–8, 184 exceptional case marking, 188
cooccurrence, 35, 47–8, 99 existential, 76
coordination, 102, 229, 205–10, 241–6 explanation (in grammar), 104–8
coreference, 40 extraction, 210–18
count(ability), 1
crossing branches, 42–3
F
free indirect style, 286
D from-phrase, 61
declarative, 153, 155 fused relative, 138
definite(ness), 1, 22 fuzzy quantifier (see quantifier)
deictic-presentation, 67–80
deixis, 73
dependency, 152 G
descriptive fallacy, 153 gap, 81, 82, 84, 90
dialect, 83 generalized quantifier (see quantifier)
discontinuous constituency, 42 Generalized Phrase Structure Gram-
discourse, 295, 302 mar, 191, 210
do generation vs production, 40, 42, 44,
operator do, 113 47
substitute do, 113 gerund participle, 251–71
double-ing constraint, 251–71 given information, 72
dual, 20 go get construction, 266
government, 152
Government and Binding Theory, 188
E grammar, teaching of, 303–4
ellipsis 111–25, 248, 277–8, 288, 289 grammaticalization, 67, 83, 155, 156,
coordination ellipsis, 115–24 291
independent ellipsis, 115–24
ellipsis of complement, 121
ellipsis of subject, 121 H
ellipsis of VP, 234 HCRC Map Task Dialogue Database,
elliptical clause, 111–25 282, 290
SUBJECT INDEX 325
head, 87, 88, 89 Left Quantifier Control Condition, 8

Heavy NP Shift, 187, 193, 200–5 Leverhulme Corpus, 114
hypotaxis, 150, 229 lexical density, 303
Lexical Functional Gramar, 191
light verb, 60
I linear precedence, 33–51
idiom, 34–5 linearization, 48
if, 274 locative inversion, 70, 135, 142
incomplete utterance, 276 logic
indefinite(ness), 1, 22 deontic logic, 165
indeterminacy, 71, 158 formal logic, 164
indirect directive, 280 logical form, 41
indirect speech act, 280 London-Lund Corpus, 79
infix, 197–200 long distance grammatical dependen-
inflection, 251–71 cy, 251
inflectional morphology, 83, 85
information status, 96, 102–4, 105
insubordination, 291 M
Interactive Model, 313 Macquarie Dictionary Corpus, 274
interjection, 231 Malay-Indonesian, 91
International Corpus of English, 114 m-command, 29
interpolation, 232, 233, 240 meaning postulate, 24
interrogative, 36–7, 39, 152, 177–82 measure function, 4
intransitive preposition (see preposi- Minimalist Program, 189
tion) modal, 129
inversion, 67, 177–81 modifier, 53
Irish, 91 monotonicity, 13
Irish English, 86 morpheme, 150
isolated if-clause, 273–94 movement, 82
J N
juxtaposition, 230, 231, 234–7, 240–3, negation, 35, 177–85
245–8 fake negation, 179–80
negative polarity item 35
nominalization, 167
L non-restrictive relative clause (see
Lancaster-Oslo/Bergen Corpus, 79 relative clause)
language education, 295, 301–3 non-syntagmatic relation, 229–50
Latin, 83 noun, 56
326 SUBJECT INDEX
nouniness, 257 Q
noun phrase, 57 quantification, 1
number, 1 quantifier, 1–31
number agreement, 72 fuzzy quantifier, 3, 27
generalized quantifier, 3
restricted quantifier, 4
O universal quantifier, 26
of-construction 104 quantitative noun, 104
only, 285
operator, 187
optative, 285 R
raising, 128
raising to object 128
P reading, 295, 313
parataxis, 229, 246, 248 reference, 36, 38, 39
parenthesis, 33–51, 230–3, 235–40, unique reference, 24
241 referential index, 40
parsing, 300 referring expression, 36
participial (construction), 131 relative clause, 81–91, 244, 247
particle, 136, 143 non-restrictive relative clause,
partitive noun, 104 34–50
passive, 131, 218–24 restrictive relative clause, 34–40
passive participle, 53 relative marker, 90
politeness, 280 relative pronoun, 37, 40, 82, 84, 85,
possessive, 86–7, 89 88
postnominal modifier, 100–1 relativization, 89
predicative, 53 representation, 46–50
preposition restrictive relative clause (see relative
intransitive preposition, 133–47 clause)
stranding of preposition, 90 right node raising, 230, 233, 240,
present perfect, 179–80 242–3, 248
processing (of language), 295
projection rule, 134
pronominal, 38–9 S
proposition, 156–61 scalar implicative, 5
propositional content, 162–70 schematization, 145
propositionalization, 164–70 scope, 178–84
psycholinguistics, 312 scope relation, 28
psychological reality, 300 Scottish English, 88–9
second language, 295, 306–7
SUBJECT INDEX 327
semantics, 1–31 time reference, 130

compositional semantics, 1 Traditional Grammar, 150, 156
sentence, 149–75 Transformational Generative Grammar,
complex sentence, 151 152
composite sentence, 152 truth-conditional semantics, 166
compound sentence, 151 type/token, 41–7, 149
simple sentence, 151
system sentence, 156
text sentence, 159 U
simple sentence (see sentence) universal grammar, 252, 268
Speech Act Theory, 153 universal quantifier (see quantifier)
square of opposition, 20 utterance, 38, 41, 43–5
Standard English, 85 utterance token, 165
standard implicature, 5
statement, 152–6
stativity, 128–30 V
subcategorization, 264 verb, 53
subject-verb agreement, 72 verbless clause, 112
subjectivity, 164–70 vocative, 231, 234
subordinate clause, 34, 151, 273 VP-ellipsis (see ellipsis)
subordinating conjunction, 137
subordinator, 82, 84, 85, 87–8, 91
syntagmatic relation, 229 W
system sentence (see sentence) well, 53–65
Systemic Functional Grammar, 295, where, 88
307–8 which, 83–89
who, 89
whose, 83, 86, 89
T wh-word, 81
teacher education, 304 word, 150
tense, 181–4 word class, 133
text sentence (see sentence) word classification, 134
that, 82–8, 89, 90 word order, 48, 81
that-relative clause, 82 wrap, 197–200
that-trace effect, 84–5 writing, 297, 300
In the STUDIES IN LANGUAGE COMPANION SERIES (SLCS) the following volumes
have been published thus far or are scheduled for publication:
1. ABRAHAM, Werner (ed.): Valence, Semantic Case, and Grammatical Relations.
Workshop studies prepared for the 12th Conference of Linguistics, Vienna, August 29th
to September 3rd, 1977. Amsterdam, 1978.
2. ANWAR, Mohamed Sami: BE and Equational Sentences in Egyptian Colloquial Ara-
bic. Amsterdam, 1979.
3. MALKIEL, Yakov: From Particular to General Linguistics. Selected Essays 1965-
1978. With an introd. by the author + indices. Amsterdam, 1983.
4. LLOYD, Albert L.: Anatomy of the Verb: The Gothic Verb as a Model for a Unified
Theory of Aspect, Actional Types, and Verbal Velocity. Amsterdam, 1979.
5. HAIMAN, John: Hua: A Papuan Language of the Eastern Highlands of New Guinea.
Amsterdam, 1980.
6. VAGO, Robert (ed.): Issues in Vowel Harmony. Proceedings of the CUNY Linguistics
Conference on Vowel Harmony (May 14, 1977). Amsterdam, 1980.
7. PARRET, H., J. VERSCHUEREN, M. SBISÀ (eds): Possibilities and Limitations of
Pragmatics. Proceedings of the Conference on Pragmatics, Urbino, July 8-14, 1979.
Amsterdam, 1981.
8. BARTH, E.M. & J.L. MARTENS (eds): Argumentation: Approaches to Theory Forma-
tion. Containing the Contributions to the Groningen Conference on the Theory of
Argumentation, Groningen, October 1978. Amsterdam, 1982.
9. LANG, Ewald: The Semantics of Coordination. Amsterdam, 1984.(English transl. by
John Pheby from the German orig. edition “Semantik der koordinativen Verknüpfung”,
Berlin, 1977.)
10. DRESSLER, Wolfgang U., Willi MAYERTHALER, Oswald PANAGL & Wolfgang
U. WURZEL: Leitmotifs in Natural Morphology. Amsterdam, 1987.
11. PANHUIS, Dirk G.J.: The Communicative Perspective in the Sentence: A Study of Latin
Word Order. Amsterdam, 1982.
12. PINKSTER, Harm (ed.): Latin Linguistics and Linguistic Theory. Proceedings of the 1st
Intern. Coll. on Latin Linguistics, Amsterdam, April 1981. Amsterdam, 1983.
13. REESINK, G.: Structures and their Functions in Usan. Amsterdam, 1987.
14. BENSON, Morton, Evelyn BENSON & Robert ILSON: Lexicographic Description of
English. Amsterdam, 1986.
15. JUSTICE, David: The Semantics of Form in Arabic, in the mirror of European lan-
guages. Amsterdam, 1987.
16. CONTE, M.E., J.S. PETÖFI, and E. SÖZER (eds): Text and Discourse Connectedness.
Amsterdam/Philadelphia, 1989.
17. CALBOLI, Gualtiero (ed.): Subordination and other Topics in Latin. Proceedings of the
Third Colloquium on Latin Linguistics, Bologna, 1-5 April 1985. Amsterdam/Philadel-
phia, 1989.
18. WIERZBICKA, Anna: The Semantics of Grammar. Amsterdam/Philadelphia, 1988.
19. BLUST, Robert A.: Austronesian Root Theory. An Essay on the Limits of Morphology.
20. VERHAAR, John W.M. (ed.): Melanesian Pidgin and Tok Pisin. Proceedings of the
First International Conference on Pidgins and Creoles on Melanesia. Amsterdam/
Philadelphia, 1990.
21. COLEMAN, Robert (ed.): New Studies in Latin Linguistics. Proceedings of the 4th
International Colloquium on Latin Linguistics, Cambridge, April 1987. Amsterdam/
Philadelphia, 1991.
22. McGREGOR, William: A Functional Grammar of Gooniyandi. Amsterdam/Philadel-
phia, 1990.
23. COMRIE, Bernard and Maria POLINSKY (eds): Causatives and Transitivity. Amster-
dam/Philadelphia, 1993.
24. BHAT, D.N.S. The Adjectival Category. Criteria for differentiation and identification.
25. GODDARD, Cliff and Anna WIERZBICKA (eds): Semantics and Lexical Universals.
Theory and empirical findings. Amsterdam/Philadelphia, 1994.
26. LIMA, Susan D., Roberta L. CORRIGAN and Gregory K. IVERSON (eds): The Reality
of Linguistic Rules. Amsterdam/Philadelphia, 1994.
27. ABRAHAM, Werner, T. GIVÓN and Sandra A. THOMPSON (eds): Discourse Gram-
mar and Typology. Amsterdam/Philadelphia, 1995.
28. HERMAN, József: Linguistic Studies on Latin: Selected papers from the 6th interna-
tional colloquium on Latin linguistics, Budapest, 2-27 March, 1991. Amsterdam/Phila-
delphia, 1994.
29. ENGBERG-PEDERSEN, Elisabeth et al. (eds): Content, Expression and Structure.
Studies in Danish functional grammar. Amsterdam/Philadelphia, 1996.
30. HUFFMAN, Alan: The Categories of Grammar. French lui and le. Amsterdam/Phila-
delphia, 1997.
31. WANNER, Leo (ed.): Lexical Functions in Lexicography and Natural Language
Processing. Amsterdam/Philadelphia, 1996.
32. FRAJZYNGIER, Zygmunt: Grammaticalization of the Complex Sentence. A case study
in Chadic. Amsterdam/Philadelphia, 1996.
33. VELAZQUEZ-CASTILLO, Maura: The Grammar of Possession. Inalienability, incor-
poration and possessor ascension in Guaraní. Amsterdam/Philadelphia, 1996.
34. HATAV, Galia: The Semantics of Aspect and Modality. Evidence from English and
Biblical Hebrew. Amsterdam/Philadelphia, 1997.
35. MATSUMOTO, Yoshiko: Noun-Modifying Constructions in Japanese. A frame seman-
tic approach. Amsterdam/Philadelphia, 1997.
36. KAMIO, Akio (ed.): Directions in Functional Linguistics. Amsterdam/Philadelphia,
1997.
37. HARVEY, Mark and Nicholas REID (eds): Nominal Classification in Aboriginal
Australia. Amsterdam/Philadelphia, 1997.
38. HACKING, Jane F.: Coding the Hypothetical. A Comparative Typology of Conditionals
in Russian and Macedonian. Amsterdam/Philadelphia, 1998.
39. WANNER, Leo (ed.): Recent Trends in Meaning-Text Theory. Amsterdam/Philadel-
phia, 1997.
40. BIRNER, Betty and Gregory WARD: Information Status and Noncanonical Word
Order in English. Amsterdam/Philadelphia, 1998.
41. DARNELL, Michael, Edith MORAVSCIK, Michael NOONAN, Frederick
NEWMEYER and Kathleen WHEATLY (eds): Functionalism and Formalism in Lin-
guistics. Volume I: General papers. Amsterdam/Philadelphia, 1999.
42. DARNELL, Michael, Edith MORAVSCIK, Michael NOONAN, Frederick
NEWMEYER and Kathleen WHEATLY (eds): Functionalism and Formalism in Lin-
guistics. Volume II: Case studies. Amsterdam/Philadelphia, 1999.
43. OLBERTZ, Hella, Kees HENGEVELD and Jesús Sánchez GARCÍA (eds): The Struc-
ture of the Lexicon in Functional Grammar. Amsterdam/Philadelphia, 1998.
44. HANNAY, Mike and A. Machtelt BOLKESTEIN (eds): Functional Grammar and
Verbal Interaction. 1998.
45. COLLINS, Peter and David LEE (eds): The Clause in English. In honour of Rodney
Huddleston. 1999.
46. YAMAMOTO, Mutsumi: Animacy and Reference. A cognitive approach to corpus
linguistics. 1999.
47. BRINTON, Laurel J. and Minoji AKIMOTO (eds): ollocational and Idiomatic Aspects
of Composite Predicates in the History of English. 1999.
48. MANNEY, Linda Joyce: Middle Voice in Modern Greek. Meaning and function of an
inflectional category. 2000.
49. BHAT, D.N.S.: The Prominence of Tense, Aspect and Mood. 1999.
50. ABRAHAM, Werner and Leonid KULIKOV (eds): Transitivity, Causativity, and TAM.
In honour of Vladimir Nedjalkov. 1999.
51. ZIEGELER, Debra: Hypothetical Modality. Grammaticalisation in an L2 dialect. 2000.
52. TORRES CACOULLOS, Rena: Grammaticization, Synchronic Variation, and Lan-
guage Contact.A study of Spanish progressive -ndo constructions. 2000.
53. FISCHER, Olga, Anette ROSENBACH and Dieter STEIN (eds.): Pathways of Change.
Grammaticalization in English. 2000.
54. DAHL, Östen and Maria KOPTJEVSKAJA TAMM (eds.): Circum-Baltic Languages.
Volume 1: Past and Present. n.y.p.
55. DAHL, Östen and Maria KOPTJEVSKAJA TAMM (eds.): Circum-Baltic Languages.
Volume 2: Grammar and Typology. n.y.p.
56. FAARLUND, Jan Terje (ed.): Grammatical Relations in Change. 2001.
57. MEL'C UK, Igor: Communicative Organization in Natural Language. The semantic-
communicative structure of sentences. n.y.p.

(Prof. Peter Collins, David Lee (Eds.) ) The Clause

Uploaded by

Copyright:

Available Formats

You might also like

(Prof. Peter Collins, David Lee (Eds.) ) The Clause

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

(Prof. Peter Collins, David Lee (Eds.) ) The Clause

Uploaded by

Copyright:

Available Formats

THE CLAUSE IN ENGLISH

STUDIES IN LANGUAGE COMPANION SERIES (SLCS)

Peter Collins and David Lee (eds)

The Clause in English

JOHN BENJAMINS PUBLISHING COMPANY

of American National Standard for Information Sciences — Permanence of

Library of Congress Cataloging-in-Publication Data

Sentences, clauses, statements and propositions 149

KEITH ALLAN BERNARD COMRIE

JOHN LYONS GEOFFREY PULLUM

Peter Collins David Lee

doubt that he taught a significant number of outstanding students. Of course the

Although he is undoubtedly one of the leading international researchers in

The arrangement of papers presented us with a formidable challenge as

DATE OF BIRTH: 4 April 1937.

Academic qualifications and awards

“Some Observations on Tense and Deixis in English”. Language 45 (1969),

My purpose in this essay is to use just one metalanguage to provide a compre-

We will consider only essential characteristics of ensemble theory, details are

x∩ y is the  of ensembles x and y and is defined in terms of suben-

(6) [a(n) x: Cx](Sx) (i)

(6) [a(n) x: Cx](Sx) ≡ c ∩ s ≥1 (v)

many several ≥ a few few a(n) / one

the superficial zero quantifier is semantically consistent with both interpretations,

(13 ′ ) [a(n)/one y: y ⊆ [ ∅ Q x: Fx]](Gy) (i)

[all y: y ⊆[ ∅ Q x: Gx]] ∃!g[ y:y ⊆ g=g] ∧ g ≠0 (v)

Compare (18) with (19), not an/one F is G.

(25) No linguists are elephant poachers No bearded linguists

¬ Q negation is not possible on some–presumably blocked by its contradictory, no.

(31) Ed didn’t kiss everyone. ¬ ∀ x (Ed kiss x)

(37) [many/much x: Fx](Gx) ≡ f∩ g > APPRO

The formula f ∩ g / f − f∩ g can be glossed “the quantity of F which is G

There is something strange about a little/few: the monotonicity of their apparent

(46) [little/few x: Fx](Gx) ≡ f∩ g f− f ∩ g

All ↑ of my friends are professionals and/*but none ↓ have blue-

NP ↑ some/a(n)/one/two, some/a(n)/one/two…not, not all/every, all/every…not, a few/

The conclusion must surely be that NP monotonicity is of no great interest to

Figure 3. The traditional square of opposition

If we eliminate doubtful cases of NP monotonicity, the two following boxes

A NP ↓ Pred ↑ all/every, no…not

We now turn Pred monotonicity, which is comparatively systematic.

Pred ↑ all/every, no…not, few/little…not, most, many/much, (not) a few/little,

If Q is ↑ (since we shall no longer be concerned with NP monotonicity, ↑=Pred ↑,

identificational property is associated with the entire restricted quantifier.

[all z: z ⊆ [F the y: y ⊆ [Q x: Mx] F]] ∃ !m[∃ !y[ z:z ⊆ (y⊆ m →

[three y: y ⊆ [Q x: Mx]] m ≥3 (iii)

tions of quantifiers in which the second is universal must occur in the NP of NP

Allan, Keith. 1980. “Nouns and Countability”. Language 56.541–67.

Capella, Martianus. Martianus Cappella, ed. by Adolfus Dick. (Bibliotheca

2. The problem of non-restrictive relative clauses

This, incidentally, is consistent with the following contrast, with therefore:

NRRs, it might be argued, cannot be independent since they include an expres-

indexing etc, as in Fabb) at a level of LF-prime (LF’). This raises similar

Linearity is paramount in Espinal’s own proposal. While Espinal objects to

relation between LE and utterance is conceived of in terms of type-token

speakers’ behaviour are linguistic extensions, having true intensional linguistic

occurrence” is understood quite distinctly from “physical occurrence” — the

The relation proposed in Burton-Roberts (1994) is representation. To avoid