A Multi-Level Account of Cleft Constructions 
in Discourse 
J. L. Delin 
Centre for Cognitive Science 
2 Buccleuch Place 
Edinburgh 
United Kingdom 
EH8 9LW 
email: judy@uk.ac.ed.cogsci 
Phone: +44 31 668 4515 
Introduction 
This pal~;r presents an analysis and synthesis of the 
factors relevant to the decision to use a cleft construc- 
tion in discourse. The model described, based on a 
corpus of 587 naturally-occurring cleft constructions 
in written and spoken discourse, consists of two 
stages. The first stage concerus the decision to use a 
cleft construction rather than a non-cleft; the second 
describes the factors relevant in deciding between 
three types of cleft: it-clefts, such as that in example 
(1), wh-clefts, such as (2), m~d reverse wh-clefts, such 
as (3): 
(1) It was the mango that Martin ate. 
(2) What Martin ate was the mango. 
(3) The mango was what Martin ate. 
Below, we examine some previous explanations of 
the functions of clefts in discourse, paying particular 
attention to the prevalent view that clefts of all kinds 
are 'focu,;ing' constructions. Analysis of the corpus 
data reveals that this view does not stand up to close 
scrutiny: a more complex model of cleft function is 
required, incorporating factors at more than one 
level of description, if the unique function of clefts as 
an option for speakers and writers is to be properly 
understood. "Iqae paper shows how factors at the 
levels of syntax, semantics, and pragmatics interact, 
and sketches how these factors might be synthesised 
in a decision procedure for generating cleft construc- 
tions in connected discourse. The work presented 
here appears in greater detail as a doctoral disserta- 
tion (Delin \[1989\]). 
As a terminological note, in what follows, the irrune- 
diate complement of the cleft copula be (the mango, 
in each of the cleft examples above) will be termed 
throughout the cleft HEAD. The clausal constituent 
(that Martin ate, or what Martin ate) in each case will 
be termed the cleft COMPLEMENT. The class of cleft 
constructions in general will often be referred to 
simply a,; CLEFTS; the use of other terms will be 
made clear as and when they arise. 
Previous Explanations of Cleft Function 
In the literature on cleft constructions to date, three 
explanations of the function of clefts have gained 
currency. These are as follows: 
Clefts provide a syntactic indication of 
the position of particular statuses of in- 
formation, such as 'Given" or "New'; 
Clefts provide unique options for the 
presentation of information in a particu- 
lar serial order; and 
Clefts are 'focusing' constructions, indi- 
cating the position of information that is 
'salieng in the discourse 
A detailed analysis of 150 of each of the cleft types, 
taken from the Survey of English Usage spoken data 
(Svartvik and Quirk \[1980\]), however, reveals that 
significant classes of counterexamples are available 
for all three explanations: the data is more varied 
than is often thought. 
First of all, it can be demonstrated that clefts do not 
reliably indicate 'Given' and 'New' information in 
particular syntactic positions in the sentence. That is, 
clefts neither contain Given information in the com- 
plement and New in the head (which we can term the 
STRONG version of this hypothesis, adopted by 
Hornby \[1972\], Haviland and Clark \[1974\],Clark and 
Clark \[1977\] inter alia), nor do they divide Given from 
New information in other syntactic ways (as the 
WEAK version of the hypothesis suggests, adopted by 
Prince \[1978\], Geluykens \[1984\], Quirk et al \[1985\], 
and Cruttenden \[1986\] among others). 
Although definitions of 'Given' and 'New' informa- 
tion are not always easy to apply formally, it is 
possible to obtain a formal approximation to Halli- 
day's \[1967\] intended notion by applying the follow- 
ing criteria: 
On the basis of a formal representation, 
New information is taken to be any argu- 
ment, predicate, or eventuality (state, 
1 83 
event or process, following Bach \[1986\]) 
that appears explicitly in the representa ~ 
tion of the sentence under analysis but 
not in that of the context; and 
The occurrence of a prosodic nucleus is 
taken to be corroborative evidence of the 
presence of New information. 
The data showed, however, that neither the Strong 
nor the Weak hypothesis were upheld. Counterex- 
amples to the strong version were clefts that con- 
tained one or more New elements in the comple- 
ment: 90% of/t-clefts, 92% of wh-clefts, and 100% of 
reverse wh-clefts examined had this information ar- 
rangement. The prevalence of such examples was 
confirmed by the fact that 41 it-clefts, 40 wh-clefts, 
and all of the reverse wh-clefts had one or more 
nuclear accents in the part of the sentence commonly 
thought to convey "Given' information. The Weak 
version of the hypothesis requires that Given infor- 
mation is segregated syntactically from New infor- 
mation in some sense. Counterexamples to this ver- 
sion, however, were a large group of clefts that con- 
tained some New information in both head and com- 
plernent: 60% of it-clefts, 92% of wh-clefts, and 14% of 
reverse wh-clefts. Again, as corroborative evidence, 
prosodic nuclei appeared in both head and comple- 
ment of 25 it-clefts, 40 wh-clefts, and 4 reverse wh- 
clefts. 
The hypothesis that clefts provide unique options for 
the presentation of information in a particular serial 
order has been suggested by Chafe \[1976\], Werth 
\[1984\] and Geluykens \[1984\] among others. How- 
ever, in the corpus examined it was frequently the 
case that a cleft would be used in a context where the 
corresponding declarative, assuming that such a 
sentence were syntactically possible, would have 
given an identical serial ordering. For example, the 
wh-cleft in (4a) does not offer significantly different 
ordering options from the declarative in (4b), yet the 
cleft was chosen (capital letters indicate the constitu- 
ent upon which a nuclear accent appears): 
(4) But I have no reason to believe he teaches 
linguistics 
(a) ...what he teaches is modern LANGUAGES 
(b) ...he teaches modern LANGUAGES 
In addition, there were sentence types other than de- 
claratives, such as some instances of topicalisation, 
that allowed an information ordering the same as 
that of a cleft to be achieved. This means that serial or- 
dering options, while not irrelevant as a considera- 
tion iin clefting (as we shall see below), do not suffice 
as an explanation of clefts as a group of construc- 
tions, since the ordering options provided are not 
unique to clefts. 
Finally, there exists in the literature a common as- 
sumption that clefts serve to 'focus' an element 
appearing in a particular syntactic position, on some 
definition of the term. Two particular definitions of 
focus can be taken as as examples: the computational 
notions of focus applied to cleft constn~ctions by 
Sidner \[1979\] and Reichman \[1981, 198!5\] in their 
work on the resolution of pronominal anaphora, and 
the sentence-accent related notions of focus sug- 
gested by Chafe \[ 1976\], Creider \[1979\] and Quirk et aI 
\[1985\] as explanations of clefts. Taking the computa- 
tional notions first, Sidner \[1979:60\] and Reichman 
\[1985:75\] predict that some particular syntactic posi- 
tions in cleft constructions are more likely than oth- 
ers to denote the antecedent of a subsequent pro- 
nominal anaphor. Sidner claims that the cleft is among 
a group of sentence types that 'make the recognition 
of focus easy, since \[they\]...have the purpose of sin- 
gling out one element from others', assuming that 
the 'singling out' function is performed by the head of 
the cleft; while Reichman builds into her algorithm 
the principle that the cleft subject is assigned a "high 
focus' rating. 
The corpus as a whole yielded 92 examples of ana- 
phors that were co-referential with an antecedent 
somewhere in a cleft construction. These antece- 
dents, however, did not appear in the predicted po- 
sitions more frequently than they did elsewhere. The 
examples relevant to Sidner's prediction showed 
that it-cleft heads provided antecedents on 43% of 
the total occasions, with wh-clefts more successful at 
78% and reverse wh-clefts much less so at 11%. Reich- 
man's prediction was less successful still: cleft sub- 
jects acted as antecedents of subsequent pronouns in 
43% of/t-clefts, 22% of wh-clefts, and 11% of reverse 
wh-clefts. While it is not in question that Sidner's and 
Reichman's algorithms do resolve anaphora, it is not 
clear that the assumption that a particular syntactic 
position is more likely to denote a 'focus' is likely to 
help to resolve anaphors encountered in naturally- 
occurring data. 
Turning to the accent-related definitions of focus, it 
can be demonstrated that the appearance of accents 
does not correlate reliably with the particular syntac- 
tic positions predicted to bear 'focal' information. 
For example, Chafe and Creider suggest that the 
head constituent of any cleft is focal, focus being 
marked by a nuclear accent. In my corpus, however, 
only 10% of it-clefts, 8% of wh-clefts, and no reverse 
wh-clefts had a sole nucleus in this position. For wh- 
clefts and it-clefts, a nucleus in both head and com- 
plement was frequent (cf. examples (5) and (6) be- 
low), while in reverse wh-clefts the popular position 
for nuclei was in the complement alone, as in (7): 
(5) A: But have you got a kettle 
B: Well what I would USE is one of those 
little solid FUEL jobs 
(6) A: Now where did I hear that from 
B: Probaby me was it-it was the day AFTER 
when I RANG 
(7) We're big enough to stand on our own 
two feet now and this is what Vincent 
said NO about 
2 84 
Quirketal \[1985\] suggest that clefts will contain one 
accentually-marked focus in the head constituent, 
and one in the complement. Although this is closer to 
the patterns shown in (5) and (6) above, there are still 
significant exceptions. Only 50% of it-clefts and 8% of 
reverse wh--clefts showed the required accent pat- 
tern, 'with wh-clefts more successful at 80%. 
It seems therefore that accent and s3n~tax do not cor- 
relate in the way that would be required to support 
the view that cleft constructions indicate the position 
of an accentuaUy-marked focal element. The assump- 
tion that such a correlation can be found can perhaps 
be accounted for in terms of a past over-reliance on 
decontextualised examples, which tend to be read 
aloud with the default accentual pattern of a nucleus 
in head position and none elsewhere. However, when 
looking at clefts in discourse, the assumption that 
there is a straightforward correlation between ac-- 
cent, syntax, and the infm3nation status 'focus' is far 
too simple to explain the facts--the large literature on 
accent placement in context (cf. Gussenhoven \[1983\] 
inter alia) bears witness to this. It is clear that, if a 
focusing explanantion is to be found at all for clefts, 
an alternative formulation of the notion is required. 
As a general observation, one of the most marked 
shortcomings of the putative explanations of cleft 
constvoctions examined above is their tendency to 
consider only one level of description--usually that 
of pragcnatics---in accounting for discourse function. 
It is plausible to suppose that, while clefts may well 
be differentiated from other sentence types at the 
level of pragmatics, their use in discourse is likely to 
be influenced additionally by other factors~for ex- 
ample, the well-understood semantic fact that clefts 
are presupposition-inducing Syntactic structures. In 
what follows, an alternative view will be presented 
of how a range of factors at the levels of syntax, 
semantics, and pragmatics combine to carve out a 
niche for clefts in the range of sentential syntactic 
structures available to speakers at any given point in 
a discourse. The account of clefts will be formulated 
in two stages: factors that differentiate clefts from 
non-cleft constructions, and factors that may plausi- 
bly serve to determine the choice between the three 
cleft types. 
Diff~arentiating Clefts and Non-Clefts 
The hmction of clefts as a class of construction can be 
explained in terms of three factors, some of which 
have been observed before in relation to clefts, but 
not integrated in a complete picture of their function. 
First, clefts of all kinds serve to indicate by means of 
their syntax the site of two statuses of propositional 
content: PRESUPPOSITION and ASSERTION. The view of 
presupposition adopted here is a seman tic one, where 
presupposition is syntactically induced by particular 
structures. The presupposition induced by the cleft 
contains an existentially-quantified variable, whose 
value is supplied by the asserted proposition. The 
variable-containing presupposition is given as ex-- 
pression P in the analysis of the cleft in example (8) 
below, while the assertion is ~ven as expression A. 
The notation is a linearised form of that used in 
Discourse Representation Theory (cf. Kamp \[1981\]), 
with the addition of eventuality indices as argu- 
ments standing for the events, states and processes 
described by the cleft: 
(8) It was the mango that Martin ate. 
P: \[el, x, m\] ate(el, x, m) 
A: Is1, x, y\] be(s1, x, y) and mango(y) 
As suggested by van der Sandt \[1988\] the presuppo- 
sition serves the function in discourse of communi- 
cating to the hearer that a propositional ANTECEDENT 
needs to be found or constructed for the presup- 
posed proposition. Note that this view says nothing 
about the actual state of the discourse context: it does 
not suggest that the presupposed information has to 
be available to the hearer already at the time of 
encountering the presupposition. It is therefore per- 
fectly coherent to suggest thai: New information ap° 
pears in the presupposed proposition: indeed, as 
seen above in the discussion of the information struco 
ture of clefts, this is commonly the case. The appear° 
ante of New information in presupposed form has 
been observed by Kartunnen \[1974:191\] and Stalnaker 
\[1974\] in relation to presuppositional constructions 
in general. 
The secondfactor involved in differentiating clefts 
and non-cleftsis that clefts convey the fact that the in- 
stantiation of the variable in the presupposition is 
UN1QUE or MAXIMAL with respect to the context. In 
the case of (8), for example, there must be a unique set 
of objects currently in the discourse context such that 
Martin ate them. This is not a new observation (cf. for 
example Atlas and Levinson \[1981\]), but independ~ 
ent mechanisms have always been advanced to 
explain it. In fact, the requirement that the element 
denoted by the cleft head must be the totality of 
objects currently in the discourse context able to 
satisfy the variable is identical to a condition on the 
appropriate use of definite referring expressions-. 
also presupIx~sitional constructions, as noted by 
Hawkins \[1978\]. For example, the phrase the bishops 
must not be used as a referring expressions if there 
exists in the current context any element that satisfies 
the predicate bishop but is not intended to be in the set 
of referents specified by the phrase. The same is true 
of the cleft ca~: there must be no elements in the 
current context that satisfy the predicateate themango 
but that are outside the (singleton) set specified by 
Martin. 
The third relevant factor in determining the function 
of clefts is a pragmatic one, namely that the instanta- 
tion of the variable in the presupposition must be 
NOVEL. As van der Sandt \[1988\]has pointed out, 
examples of the following kind are unacceptable in 
both clefts and declaratives: 
(9) //John won and it was John who won. 
However, (10) is acceptable as a cleft: 
3 85 
(10) I knew someone won, and I saw John a 
moment ago, but I didn't know it was 
John who won. 
The distinction between (9) and (10) is that, while the 
participants in the event described by the cleft in each 
case, and the nature of that event, may already be 
available in the discourse, only in the second case is 
the connection between the winning event and the 
winner a novel one. This connection, which can in 
fact be characterised as that between the variable in 
the presupposition and its instantiation, needs to be 
informative for the cleft to be acceptable. An impor- 
tant part of the function of clefts in general appears to 
be the highlighting of this connection. 
Having outlined the three factors determining the 
function of clefts as a class of construction, we can 
turn to ways of choosing between the three cleft 
types. 
Choosing Between Clefts 
The factors contributing to the differentiation in be- 
tween the three cleft types are argued here to be as 
follows: 
The three cleft types provide different 
options for the serial ordering of infor- 
mation; 
Different constraints exist on the rela- 
I:ionships that can exist between the pre- 
supposed information borne by each cleft, 
and the discourse context; 
The clefts present different options for 
information structure in general; and 
Syntactic differences between the clefts 
serve to constrain how information can 
be divided up between presupposition 
and assertion in the three cleft types. 
While none of these factors serve in isolation to 
explain the function of the cleft types, in combina- 
tion, and in addition to the functions claimed above 
to be general to clefts, they are powerful determi- 
nants of syntactic choice. 
First of all, the three types offer different options for 
the serial ordering of information. One serial order- 
ing constraint appears to be related to the way texts 
are processed psychologically: as Garnham et al \[ 1982\] 
have shown, speakers may enhance ease of process- 
ing for the hearer by optimising REFERENTIAL 
CONTINUITY in the discourses they produce. For ex- 
ample, the/t-cleft in (11) below may be preferred to 
the wh-cleft in (12), because in (11) the order of in 
which important referents of the discourse are speci- 
fied is the discontinuous rabbit-birthday-rabbit-birth- 
day, while in (12) it is the continuous rabbit-rabbit- 
birthday-birthday: 
(11) Kevin wants a rabbit, because it was a 
rabbit I got for my birthday. My birthday 
is in January. 
(12) Kevin wants a rabbit, because what I got 
for mybirthday was a rabbit. My birthday 
is in January. 
Psychological research (cf. Clark and Haviland \[1977\]) 
suggests, in addition, that speakers may also prefer 
not to postpone the presupposed part of the message. 
This is because the presupposed part of the utterance 
may be the important part with respect to the embed- 
ding of information in memory. In some case,,~, choice 
of a particular type of cleft would allow the bulk of 
information in the cleft head to postpone the presup- 
position unduly, as might be the case in/t-clefts and 
reverse wh-clefts. This leads to a prefereno~ for the 
cleft in which the presupposed information is pre- 
sented firstmthe wh-cleft--in cases where the non- 
presupposed information is lengthy. This could ac o 
count for the preferability of the wh-cleft in (13) over 
the /t-cleft in (14) for presenting the following con- 
tent: 
(13) What I want to find is some formal thing I 
can use to follow the complex definitions 
of these feature structures. 
(14) It's some formal thing I can use to follow 
the complex definitions of these feature 
structures that I want to find. 
As a second determinant of the choice between clefts, 
it is possible to show on the basis of an analysis of the 
relationship between the presupposed proposition 
and the context that the kinds of information that can 
appear as the presupposition of each of the three 
types of cleft are distinct. As noted above, it is already 
well known that presuppositions defined in a fairly 
conventional sense need not consist of information 
that is currently available to the hearer. In fact, it is 
clear that for clefts at least the status of the presup- 
posed information with respect to context is more or 
less independent of whether it is presented as a 
presupposition. 
The corpus reveals that clefts of all three types have 
different profiles with respect to the types of infor- 
mation that it is possible to convey as presupposi- 
tion. For example, wh-clefts cannot be use(| to pre- 
suppose information that is New in context, while it- 
clefts can. This produces a strong preference for the 
/t-cleft in (15b) over the wh-cleft in (15a) below, for 
example: 
(15) A: Sue, this is Marta. 
(a) B: #Hi! What I stayed in when you were 
in the States was your fiat. 
(b) B: Hi! It was your fiat I stayed in when 
you were in the States. 
86 4 
Other distinctions of this nature exist between the 
three cleft types (cf. Delin \[1989:178ff for a full expo- 
sition). For example, inferrable relations of some 
kinds are unavailable to both/t-cleft and reverse wh- 
cleft presuppositions, while Old information-that 
is, information that is currently available in context- 
-appears very rarely in reverse wh-cleft presupposi- 
tions. These tendencies, although not a suitable basis 
for rules about the information structure of clefts, 
could form a.valid basis for a set of preference weight- 
ings to be used in the decision between the cleft types 
in a given context. 
In addition to the differences in the information 
status of presupposed information in the three cleft 
types, the analysis of the corpus showed a more 
general pattern of distinctions in information struc- 
ture. Firstly, as Prince \[1978\] has observed, there 
appear to be two distinctive arrangements of infor- 
mation preferred for it-clefts: the most common has 
New information in the complement, and Old or 
Inferrable in the head; less common was a second 
pattern, with New information borne by the head 
constituent and Old or inferrable information in the 
complement. These arrangements are comparable to 
Prince's informative presupposition and stressed-focus 
it-clefts respectively. The most common pattern for 
wh-cleft,; had Old or more usually inferrable infor- 
mation in the complement and New in the head, 
while the pattern in all the reverse wh-clefts was an 
anaphoric (Old) head and New or inferrable infor- 
marion borne in the complement. As with the prefer- 
ences for presupposition types noted above, these 
factors could form the basis for weightings to be used 
in syntactic choice. 
The fourth factor distinguishing the clefts from one 
another is the different range of constituent types 
that can appear as cleft head in each case, serving to 
constrain in different ways what can be conveyed by 
the cleft. The full range of constraints is presented in 
Delin \[1989:55ff\]; to demonstrate briefly here, only 
wh-clefts are comfortable with bare infinitives as 
head: 
(16) What Martin wanted to do was eat the 
mango. 
(17) *It was eat the mango that Martin wanted 
to do. 
(18) *Eat the mango was what Martin wanted 
to do. 
Similarly, only/t-clefts can take NP + not until as 
head: 
(19) It was not until three that she left. 
(20) *Not until three was when she left. 
(21) *When she left was not until three. 
Reverse wh-clefts are the only cleft type able to take 
pronominal it and which as head: 
(22) This camera--and it is what I need-is 
expensive. 
(23) This camera, which is what I need, is 
expensive. 
(24) *This camera--and what I need is it.-- is 
expensive. 
(25) *This camera, what I need is which, is 
expensive. 
(26) *This camera--and it is it that I need--is 
expensive. 
(27) *This camera, it is which that I need, is ex- 
pensive. 
Syntactic factors exist, therefore, that would serve to 
condition the choice between the three types of cleft. 
Conclusion 
Using the syntactic, semantic, and pragmatic factors 
discussed above it is possible to suggest in outline a 
two-stage decision procedure for the generation of 
chfft constructions. The first stage is concerned with 
determining whether to use a cleft. Taking a given 
proposition to convey as content, the generator has 
first to determine the following: 
• Is a variable-containing presupposition 
required? 
Does the residue of con tent left over from 
presupposing constitute a suitable in- 
stantiation of the variable in the presup- 
position? 
• Is that instantiation the maximal one 
possible in the context? 
• Is the instantiation novel? 
The second phase deals with the choice between the 
three cleft types. On a 'generate-and-test' approach 
to sentence generation, the possible reali.~ations of 
the arrangements of semantic content arrived at on 
the basis of the first stage of the process can be 
examined for the following features: 
• Which are syntactically acceptable? 
• Which provides the optimum serial or- 
dering of information? 
Which places information of the appro- 
priate type in the presupposed portion of 
the sentence? 
® Which has appropriate overall informa- 
5 87 
tion structure? 
It is predicted that the result of the above procedure 
may be no clefts at all, or more than one. Either result 
is acceptable, since it is clear that, just as it is not 
always appropriate to produce a cleft, neither is it 
always the case that only one cleft is possible. 
It is dear that the above suggestions are only a small 
step towards an implementable algorithm. How- 
ever, in the course of this paper I hope to have shown 
two things in particular: the importance of corpus- 
based descriptive work to underpin decision proce- 
dures for syntactic choice, and the importance of 
avoiding the assumption that syntactic choice is a 
uniquely pragmatic, or even a merely" stylistic', phe- 
nomenon. 

References 

Atlas, J. D. and Levinson, S. C. \[1981\]/t-clefts, infor- 
mativeness, and logical form. Chapter 1 ,.'n Cole, P. 
(ed.) Radical Pragmatics, ppl-59. New York: Aca- 
demic Press. 

Bach, E. \[1986\] The algebra of events. Linguistics and 
Philosophy, 9, 5-16. 

Chafe, W. \[1976\] Givenness, contrastiveness, defin- 
iteness, subjects, topics, and points of view. in Li, C. 
(ecl.), Subject and Topic, pp25-56. New York: Aca- 
demic Press. 

Clark, H. and Clark, E. \[1977\] Psychology and Lan- 
guage: An Introduction to Psycholinguistics. New York: 
Harcourt Brace Jovanovich. 

Clark, H. H., and Haviland, S. E. \[1977\] Comprehen- 
sion and the Given-New Contract. In Freedle, R. O. 
(ed.) Discourse Production and Comprehension, Vol- 
ume 1, ppl-40. Norwood, New Jersey: Ablex. 

Creider, C. \[1979\] On the explanation of transforma- 
tions. In Givon, T. (ed.) Syntaxand Semantics. Volume 
12: Discourse and Syntax, p3-22. New York, Academic 
Press. 

Cruttenden, A. \[1986\] Intonation. Cambridge: Cam- 
bridge University Press. 

Delin, J. L. \[1989\]. Cleft Constructions in Discourse. 
PhD dissertation, Centre for Cognitive Science, Uni- 
versity of Edinburgh. 

Garnham, A., Oakhill, J., and Johnson-Laird, P. 
\[1982\] Referential Continuity and the Coherence of 
Discourse. Cognition, 11, 29-46. 

Geluykens, R. \[1984\] Focus Phenomena in English: 
an Empirical Investigation into Cleft and Pseudo- 
Cleft Sentences. Technical Report No. 36, Departe- 
ment Germaanse, Universitaire Instelling Antwer- 
pen, Antwerp, 1984. 

Gussenhoven, C. \[1983\] Focus, Mode, and the 
Nucleus. Journal of Linguistics, 19, 377-417. 

Halliday, M. A. K. \[1967\] Notes on transitivity and 
theme in English: Part 2. \]ournal of Linguistics, 3, 199-244. 

Haviland, S. E., and Clark, H. I-L \[1974\] What's new? 
Acquiring new information as a process in compre- 
hension. Journal of Verbal Learning and Verbal Behaviour, 
13, 512-521. 

Hawkins, J. A. \[1978\] Definiteness and Indefiniteness. 
London: Croom Helm. 

Hornby, P. A. \[1972\] The psychological subject and 
predicate. Cognitive Psychology, 3, 632-642+ 

Kamp, H. \[1981\] A theory of truth and semantic rep- 
resentation. In Groenendijk, J., Janssen, T., and Stok- 
hof, M. (eds.) Formal Methods in the Study of Language, 
Volume 136, pp277-322. Amsterdam: Mathematical 
Cantre Tracts. 

Karttunen, L. \[1974\] Presuppositions and Linguistic 
Context. Theoretical Linguistics, 1, pp55-69. 

Prince, E. \[1978\] A comparison of WH-clefts and it- 
clefts in discourse. Language, 54, 883-906. 

Quirk, R., Greenbaum, S., Leech, G., and Svartvik, 
J. \[1985\] A Comprehensive Grammar of the English 
Language. London: Longman. 

Reichman, R. \[1981\] Plain Speaking: A Theory and 
Grammar of Spontaneous Discourse. PhD thesis. 
Published as BBN Report No. 4681. 

Reichman, R. \[1986\] Getting Computers to Talk Like 
You and Me.. Cambridge, Mass.: MIT Press. 

van der Sandt, R. A. \[19881 Context and Presupposi- 
tion. London: Croom Helm. 

Sidner, C. L. \[1979\] Towards a Computational The- 
ory of Definite Anaphora Comprehension in English 
Discourse. Technical Report No. 537, MIT Artificial 
Intelligence Laboratory, June, 1979. 

Stalnaker, R. \[1974\] Pragmatic Presuppositions. In 
Munitz, M. K. and Unger, P. K. (Eds.) Semantics and 
Philosophy, pp197-213. New York: New York Univer- 
sity Press. 
