Professional Documents
Culture Documents
(Prof. Peter Collins, David Lee (Eds.) ) The Clause
(Prof. Peter Collins, David Lee (Eds.) ) The Clause
(Prof. Peter Collins, David Lee (Eds.) ) The Clause
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
Edited by
PETER COLLINS
University of NSW
DAVID LEE
University of Queensland
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
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
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
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
“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
‘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
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 .
THE SEMANTICS OF ENGLISH QUANTIFIERS 5
(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 SEMANTICS OF ENGLISH QUANTIFIERS 7
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″ ).
THE SEMANTICS OF ENGLISH QUANTIFIERS 9
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
“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
(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
THE SEMANTICS OF ENGLISH QUANTIFIERS 15
(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 ↑
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
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 ↓
have blue-collar jobs.
d. Few ↓ of my friends are professionals, and/*but most/many ↑
have blue-collar jobs.
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.
THE SEMANTICS OF ENGLISH QUANTIFIERS 19
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
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*¬
THE SEMANTICS OF ENGLISH QUANTIFIERS 21
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
THE SEMANTICS OF ENGLISH QUANTIFIERS 23
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)
THE SEMANTICS OF ENGLISH QUANTIFIERS 25
(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
THE SEMANTICS OF ENGLISH QUANTIFIERS 29
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
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
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.
36 NOEL BURTON-ROBERTS
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.
38 NOEL BURTON-ROBERTS
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,
40 NOEL BURTON-ROBERTS
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.
LANGUAGE, LINEAR PRECEDENCE AND PARENTHETICALS 41
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-
42 NOEL BURTON-ROBERTS
4. A representational approach
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
48 NOEL BURTON-ROBERTS
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)
LANGUAGE, LINEAR PRECEDENCE AND PARENTHETICALS 49
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.
LANGUAGE, LINEAR PRECEDENCE AND PARENTHETICALS 51
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
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
2. “Quality” well1
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
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.
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
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.
THE ENGLISH MODIFIER WELL 63
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.
THE ENGLISH MODIFIER WELL 65
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
Peter Collins
1. Introduction
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
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
THE DEICTIC-PRESENTATION CONSTRUCTION IN ENGLISH 71
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
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]
THE DEICTIC-PRESENTATION CONSTRUCTION IN ENGLISH 73
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
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
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
THE DEICTIC-PRESENTATION CONSTRUCTION IN ENGLISH 79
9. Conclusion
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
References
Bernard Comrie
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
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.
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.
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
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
References
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
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.
This section will repeat Radford’s points but they will be discussed in terms of
some of my data.1
96 PETER H. FRIES
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
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
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
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
POST NOMINAL MODIFIERS IN THE ENGLISH NOUN PHRASE 107
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
POST NOMINAL MODIFIERS IN THE ENGLISH NOUN PHRASE 109
References
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
(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.]
114 SIDNEY GREENBAUM AND GERALD NELSON
(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).
ELLIPTICAL CLAUSES IN SPOKEN AND WRITTEN ENGLISH 115
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):
116 SIDNEY GREENBAUM AND GERALD NELSON
(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 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
ELLIPTICAL CLAUSES IN SPOKEN AND WRITTEN ENGLISH 117
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.
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.
Category Independent ellipsis Coordination ellipsis Total
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.]
ELLIPTICAL CLAUSES IN SPOKEN AND WRITTEN ENGLISH 119
(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.
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.
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-
122 SIDNEY GREENBAUM AND GERALD NELSON
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.
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
124 SIDNEY GREENBAUM AND GERALD NELSON
References
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
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
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
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.
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.
INTRANSITIVE PREPOSITIONS 137
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.
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.
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
3.2.1 Counterargument
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.
3.3.1 Counterargument
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
3.4.1 Counterargument
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.
3.5.1 Counterargument
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
INTRANSITIVE PREPOSITIONS 141
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
3.6.1 Counterargument
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.
The X-words occur in the Locative Inversion construction (e.g. Into the room ran
John!).
3.7.1 Counterargument
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).
INTRANSITIVE PREPOSITIONS 143
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
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
INTRANSITIVE PREPOSITIONS 147
References
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
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
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.
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
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
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
SENTENCES, CLAUSES, STATEMENTS AND PROPOSITIONS 157
(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
SENTENCES, CLAUSES, STATEMENTS AND PROPOSITIONS 159
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
SENTENCES, CLAUSES, STATEMENTS AND PROPOSITIONS 161
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
SENTENCES, CLAUSES, STATEMENTS AND PROPOSITIONS 169
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
SENTENCES, CLAUSES, STATEMENTS AND PROPOSITIONS 171
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
SENTENCES, CLAUSES, STATEMENTS AND PROPOSITIONS 173
References
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
(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.)
INTERACTIONS BETWEEN TENSE AND NEGATION 183
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.
INTERACTIONS BETWEEN TENSE AND NEGATION 185
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
1. Background
NP VP
Joan V S
believes NP VP[INF]
proved
him to be a genius
to be guilty
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
(9) a. S
NP VP
Joan VP VP[INF]
V NP
(9) b. S
NP VP
Joan
V NP VP[INF]
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
(14) S
NP VP
Joan VP
V NP VP[INF]
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).
THE ENGLISH ACCUSATIVE-AND-INFINITIVE CONSTRUCTION 193
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
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
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
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
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]
(27) S
S/(S\NP[SUB]) S\NP[SUB]
(S\NP[SUB])/WNP[OBJ]
Joan
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
S/(S\NP[SUB]) S\NP[SUB]
(S\NP[SUBJ])/NP[OBJ]
Joan
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)
THE ENGLISH ACCUSATIVE-AND-INFINITIVE CONSTRUCTION 201
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/(S\NP[SUB]) S\NP[SUB]
(S\NP[SUB])/NP[OBJ]
Joan
(26) b. S
S/(S\NP[SUB]) S\NP[SUB]
(S\NP[SUB])/NP[OBJ]
Joan
S/(S\NP[SUB]) S\NP[SUB]
Joan
(S\NP[SUB])/NP[OBJ]
((S\NP[SUB])/NP[OBJ])/VP[INF]
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:
THE ENGLISH ACCUSATIVE-AND-INFINITIVE CONSTRUCTION 203
(36) S
S/(S\NP[SUB]) S\NP[SUB]
Joan ((S\NP[SUB])/S
((S\NP[SUB])/S) ((S\NP[SUB])/X)/Y)\((S\NP[SUB])/X/Y) S
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
lecture on gravitation
This is the shifted analogue of (34a), and has the following analysis:
(38) S
S/(S\NP[SUB]) S\NP[SUB]
Joan
(S\NP[SUB])/NP[OBJ]
((S\NP[SUB])/NP[OBJ])/VP[INF]
S/(S\NP[SUBJ]) S\NP[SUB]
Joan
(S\NP[SUB])/WNP[OBJ]
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
THE ENGLISH ACCUSATIVE-AND-INFINITIVE CONSTRUCTION 205
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
((S\NP[SUB])/NP[OBJ])/VP[INF] (((S\NP[SUB])/NP[OBJ])/VP[INF])\(((S\NP[SUB])/NP[OBJ])/VP[INF])
and assumes
(46) S
X/(X\NP[SUB]) S\NP[SUB]
(S\NP[SUB])/NP[OBJ]
Joan
(47) S
S/(S\NP[SUB]) S\NP[SUB]
(S\NP[SUB])/NP[OBJ]
Joan X\(X/NP[OBJ])
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]
((S\NP[SUB])/NP[OBJ])/VP[INF] X¯ (X/NP[OBJ])
(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
X/(X\NP[SUB]) S\NP[SUB]
(S\NP[SUB])/VP[INF] VP[INF]
6. Extraction
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
X/(X\NP[SUB]) S\NP[SUB]/NP[OBJ]
(63) S
X/(X\NP[SUB]) S\NP[SUBJ]
(S\NP[SUB])/Q Q
X/(X\NP[SUB]) (S\NP[SUB])/NP[OBJ]
believes to be a genius
(66) S
X/(X\NP[SUB]) S\NP[SUB]
(S\NP[SUB])/Q Q
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
X/(X\NP[SUB]) S\NP[SUB]
(S\NP[SUB])/Q Q
((S\NP[SUB])/WNP[OBJ])/NP[OBJ] X ¯ (X/NP[OBJ])
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
X/(X\NP[SUB]) S\NP[SUB]
(S\NP[SUB])/Q Q
((S\NP[SUB]/W)NP[OBJ])/NP[OBJ] X ¯ (X/WNP[OBJ])
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
THE ENGLISH ACCUSATIVE-AND-INFINITIVE CONSTRUCTION 215
(71) S
X/(X\NP[SUB]) S\NP[SUB]
(S\NP[SUB])/Q Q
X/(X\NP[SUB]) (S\NP[SUB])/NP[OBJ]
((S\NP[SUB])/NP[OBJ])/VP[INF] VP[INF]/NP[OBJ]
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:
THE ENGLISH ACCUSATIVE-AND-INFINITIVE CONSTRUCTION 217
(77) S
X/(X\NP[SUB]) S\NP[SUB]
(S\NP[SUB])/Q Q
((S\NP[SUB])/NP[OBJ])/VP[INF] VP[INF]
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
a professor who I
((S\NP[SUBJ])/WNP[OBJ]) NP[OBJ]
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
X/(X\NP[SUB]) S\NP[SUB]
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 ENGLISH ACCUSATIVE-AND-INFINITIVE CONSTRUCTION 219
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:
THE ENGLISH ACCUSATIVE-AND-INFINITIVE CONSTRUCTION 221
8. Conclusion
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
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.)
Huddleston, R. 1988. Introduction to the Grammar of English. Cambridge:
Cambridge University Press.
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.
THE ENGLISH ACCUSATIVE-AND-INFINITIVE CONSTRUCTION 227
Non-syntagmatic relations*
Peter Peterson
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
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.
3. Properties of juxtaposition
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.
ON THE BOUNDARIES OF SYNTAX 235
(18) S
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.
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.
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
ON THE BOUNDARIES OF SYNTAX 237
NP VP
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.
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.
ON THE BOUNDARIES OF SYNTAX 239
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
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.
ON THE BOUNDARIES OF SYNTAX 241
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
ON THE BOUNDARIES OF SYNTAX 245
(48) S
NP VP
NP
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).
ON THE BOUNDARIES OF SYNTAX 249
References
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).
254 GEOFFREY PULLUM AND ARNOLD ZWICKY
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-
256 GEOFFREY PULLUM AND ARNOLD ZWICKY
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
GERUND PARTICIPLES AND HEAD-COMPLEMENT INFLECTION 257
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).
GERUND PARTICIPLES AND HEAD-COMPLEMENT INFLECTION 261
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.
262 GEOFFREY PULLUM AND ARNOLD ZWICKY
(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.
GERUND PARTICIPLES AND HEAD-COMPLEMENT INFLECTION 263
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
GERUND PARTICIPLES AND HEAD-COMPLEMENT INFLECTION 267
(38) VP [F1:v1]
V VP [F2:v2]
268 GEOFFREY PULLUM AND ARNOLD ZWICKY
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.
270 GEOFFREY PULLUM AND ARNOLD ZWICKY
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
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.
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
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.
(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.
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]
ISOLATED IF-CLAUSES IN AUSTRALIAN ENGLISH 279
(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”
ISOLATED IF-CLAUSES IN AUSTRALIAN ENGLISH 281
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
ISOLATED IF-CLAUSES IN AUSTRALIAN ENGLISH 283
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-
ISOLATED IF-CLAUSES IN AUSTRALIAN ENGLISH 285
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.
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
ISOLATED IF-CLAUSES IN AUSTRALIAN ENGLISH 287
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”.
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.
ISOLATED IF-CLAUSES IN AUSTRALIAN ENGLISH 291
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.
ISOLATED IF-CLAUSES IN AUSTRALIAN ENGLISH 293
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
Lynn Wales
1. Introduction
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
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.)
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.
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.
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.
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
FUNCTIONAL AND STRUCTURAL 301
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.
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.
FUNCTIONAL AND STRUCTURAL 303
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.
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.)
FUNCTIONAL AND STRUCTURAL 305
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).
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
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:
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.)
FUNCTIONAL AND STRUCTURAL 315
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
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.
FUNCTIONAL AND STRUCTURAL 317
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
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
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