<?xml version="1.0" standalone="yes"?>
<Paper uid="C92-1045">
  <Title>Aspect-Switching and Subordination: the Role of/t-Clefts in Discourse*</Title>
  <Section position="2" start_page="0" end_page="0" type="metho">
    <SectionTitle>
2 Buccleuch Place Edinburgh Scotland
</SectionTitle>
    <Paragraph position="0"> To our knowledge, Prince \[1978:902\] was the first to observe that cleft constructions serve a SUBOR-DINATING function in discourse. She observed that for examples like (1) the information conveyed is 'background material ...subordinate in importance to what follows':  (1) It is through the writings of Basil Bernsteiu that many social scientists have become aware of  the scientific potential of sociolinguistics ... Yet their very popularity has often deformed Bernsteiu's arguments; ...he has been made to say that lower class children are linguistically 'deprived' ...In fact, Bernstein's views are much more complex than that. First ...</Paragraph>
    <Paragraph position="1"> She notes in particular that the subordination relation involved is often (although not always) one of cause emd effect, where the clefted proposition is often intended to be interpreted as the cause. She gives the following example: (48a) Here ... were the ideas which Hitler was later to use.., lfis originality lay in his being the only politician of the Right to apply them to the Germ~m scene after tile First World War. It was then that the Nazi movement, alone among the nationalist and conservative pa*ties, gained a great mass following and, having achieved this, won over the support of the Army, the President of the Republic, and.., big business--three 'longestablished institutions' of great power. The lessons learned in Vienna proved very usefnl indeed. null Prince \[1978:902\] explains the effect of the cleft in her (48a) as follows: *.. If the third sentence of (48a) read Then, the Nazi movement . .., it would tend to suggest a separate event, and we would lose tile notion that it was all It's doing a notion conveyed very strongly by tim i/~cleft's subordinating effect, and underlined (though still not asserted) by the last sentence.</Paragraph>
    <Paragraph position="2"> \[Prince 1978:902\] Prince's suggestion, then, is that clefts can serve as suitable vehicles for deliveriug information that is baekgrouudcd to the main flow of the discourse, or that is contingently related to it, by cause-and-effect.</Paragraph>
  </Section>
  <Section position="3" start_page="0" end_page="0" type="metho">
    <SectionTitle>
ACIES DE C(IL1NG-92, Nhgll'.'S, 23-28 AO~t 1992 2 8 1 PROC. OF COLING-92, NAI, rrES, AUG. 23-28, 1992
</SectionTitle>
    <Paragraph position="0"/>
    <Section position="1" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
Temporal Regression
</SectionTitle>
      <Paragraph position="0"> In addition to the cause-and-effect relations noted by Prince, our own data reveals a further 'backgrounding' function: the use of clefts for temporal subordination. In (2), for example, an //,cleft is being used to introduce background information elaborating on the nature of a protagonist in the discourse (Mr. Butler). This is done by describing an eventuality that he was involved in at some previous time: (2) 1. Mr. Butler, the Home Secretary, decided to meet the challenge of the 'Ban-the-Bomb' demonstrators head-on.</Paragraph>
      <Paragraph position="1">  2. Police leave was cancelled 3. and secret plans were prepared.</Paragraph>
      <Paragraph position="2"> 4. It was Mr. Butler who authoriscd action which ended in 3Y, members of the Committee of 100 being imprisoned.</Paragraph>
      <Paragraph position="3"> 5. The Committee's president and his wife were  each jailed for a week.</Paragraph>
      <Paragraph position="4"> The effect of the cleft is to cause the 'background' information about the anthorisation of action to be interpreted as as occurring prior to the events introduced in lines 1-3--the decision, the cancellation of leave, and the preparation of secret plans.</Paragraph>
      <Paragraph position="5"> If we look at a de-clefted version of the same di~ course, we can see that the temporally subordinating effect of the cleft is removed, creating a rather different effect. The result, (3), has the 'authorisation of action' described in the de-clef ted sentence occnrring in simple temporal progression from the 'cancellation of police leave'--in other words, after the events introduced in lines 1-3:  (3) 1. Mr. Butler, the Home Secretary, decided to meet the challenge of the 'Ban-the-Bomb' demonstrators head-on.</Paragraph>
      <Paragraph position="6"> 2. Police leave was cancelled 3. and secret plans were prepared.</Paragraph>
      <Paragraph position="7"> 4. Mr. Butler authorised action which ended in 3~ members of the Committee of lO0 being imprisoned. null 5. The Committee's president and his wife were  each jailed for a week.</Paragraph>
    </Section>
    <Section position="2" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
The Known Fact Effect
</SectionTitle>
      <Paragraph position="0"> Examples such as (1), (48a) and (2) share a property that has been characterised as the KNOWN FACT EF-FECT. Prince states: Their function, or at least one of their functions, isTO MARK A PIECE OF INFORMATION AS FACT~ known to some people although not yet known to the intended hearer. Thus they are frequent in historical narrative, or wherever the speaker wishes to indicate that s/he does not wish to take personal responsibility for the truth or originality of the statement being made.</Paragraph>
      <Paragraph position="1"> \[Prince 1978:899-900\] The cleft can introduce 'new' information to the discourse, while at the same time signalling that the information is to be treated as if it had been there all along. A significant feature, then, is that the  information must be regarded as not open to conversational negotiation. Delin \[1991\] proposed that a speaker who uses an it-cleft that conveys new information in the complement is indicating that the information they are communicating did not originate with the speaker, and that they are therefore not to be held responsible for its truth value.</Paragraph>
      <Paragraph position="2"> This Known Fact account has an intuitive appeal; yet it does not constitute a meehanisable explanation of the role of the cleft in discourse. One possible avenue to such an account would be to exploit Polanyi and Scha's \[1988\] Linguistic Discourse Model. By adding an appropriate rule to the grammar for discourse constituent units (DCDS), we could represent the cleft as introducing a DCO to be attached as subordinate to the current node, deriving a local discourse parse tree such as that in Figure 1.</Paragraph>
      <Paragraph position="3"> Hence, we can represent the proposition conveyed by the cleft sentence as subordinate to the existing discourse structure. But mere representation does not make obvious how the syntactic or semantic features of the cleft are supposed to drive the assignment of discourse structure. Nor is it obvious that such a subordination structure supports the Known Fact Effect. There are plenty of other subordination structures in Polanyi and Scha's framework that don't indicate that a Known Fact reading should be associated with the subordinate elements.</Paragraph>
      <Paragraph position="4"> The Aspectual Effect of the Cleft What we want at this point is an account which can recruit the syntactic and semantic features of the cleft, to explain the background and regress data that has been observed, feed into the discourse parse process, and explain the Known Fact Effect. The basic proposal we explore here is that it is the aspectual effect of the cleft that provides the required explanation.</Paragraph>
    </Section>
    <Section position="3" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
Aspect ual Class
</SectionTitle>
      <Paragraph position="0"> Following Vendler \[1967\], much consideration has been given to the &amp;quot;aspectual types&amp;quot; of utterances of English sentences (cf. Hinrichs \[1986\]; Dowty \[1986\]; Moens and Steedman \[19871). An utterance denotes an eventuality of some type', the aspectual type will determine the relation to other eventualities mentioned in a discourse. Vendter's inventory includes ACTIVITIES~ ACCOMPLISHMENTS, ACHEIVE-MENTS and STATES, Bach \[1986\] takes the space of eventualities to include STATES and NON-STATES; in turn, states consist of DYNAMIC and STATIC states, Acr~ DE COLING-92, NANTES, 23-28 AnenT 1992 2 8 2 PROC. OF COLING-92, NANTES. AUG. 23-28, 1992 while non-states consist of PROCESSES and EVENTS.</Paragraph>
      <Paragraph position="1"> Events are then either PROTRACTED or MOMENTA-NEOUS; momentaneous events are either HAPPEN-INGS or CULMINATIONS.</Paragraph>
      <Paragraph position="2"> For current purposes, the relevant distinction is that between states and non-states; in particular, between states and events. From Bach \[1986:6\], paradigmatic cases of verb phrases exlfibiting this distinction include the following: States: sit, he drunk, own x, love x Events: build z, walk to Boston, notice, reach the top The aspectual class of an utterance is typically determined by the aspectual class of the lexical verb, by other elements within the verb phrase, by temporal adverbials with which the verb phrase co-occurs, and by the noun phrase itself. Linguistic context will also influence aspectual class assignment. For example, a verb normally taken to denote a process, such as (4), can form part of a verb phrase deuoting a protracted event, as ill (5); and in combination with certain noun phrases, tile same verb phrase cau form part of a sentence (6) de~oting a habitual state:  (4) ran (5) ran to the station (6) trains ran to the station  We can now frame the basic proposal we wish to discuss: it-cleft sentences are stative; tbe presence of the copular in the cleft head ensures this. We can thus view a cleft as a function taking either nonstate-expressions or state-expressions as input, and returning state-expressions as output. (Ta) and (8a) denote an event and a state respectively; but both  (Tb) and (8b) denote states.</Paragraph>
      <Paragraph position="3"> (7) a. Victoria found the body.</Paragraph>
      <Paragraph position="4"> b. It was Victoria who found the hody.</Paragraph>
      <Paragraph position="5"> (8) a. Victoria knew the killer's identity.</Paragraph>
      <Paragraph position="6"> b. It was Victoria who knew the killer's identity.</Paragraph>
    </Section>
    <Section position="4" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
Temporal Overlap
</SectionTitle>
      <Paragraph position="0"> Consider now theories which attempt to derive the temporal structure of discourse from the syntactic structures of a sequence of input sentences. In the framework of discourse representation theory, work by Partee \[1984\], Kamp and Rohrer \[1983\] and Hinrichs \[1986\] has indicated that it is possible to exploit Reichenbach's \[1947\] notions of speech-time, eventtime, and reference-time to drive a process which will add temporal constraints to a discourse representation structure (DRS). 1 In particular, in past tense narrative, simple event-expressions are taken to locate an event at an event-time corresponding to the existing reference-time, and, in addition, to update the reference-time to a point 'just after' that reference-time. This new time wil! constitute the reference-time for the loca-tion of the next input expression. By contrast, state-expressions firstly locate the state as overlapping the 1We do not wish to maintain that a reference-time based account is the best that can be provided. It is, however, a convenient representational tool.</Paragraph>
      <Paragraph position="1"> existing reference-time; and secondly do not update that reference time. Hence, the next input expression (denoting event or state) will be evaluated with respect to the same reference time again. In this way, DRS construction can encode the relative temporal locations of the various eventualities. In general, one can say that simple event-expressions 'move a narrative along', while simple state-expressions leave it where it is. More complex expressions, containing temporal adverbials and perfective or progressive aspect, require some complication in the DRSconstruction rules. Take an example like (9): (9) John met Mary in town. She had broken her leg, but looked well in spite of it.</Paragraph>
      <Paragraph position="2"> The use of the past perfect can be taken to either introduce a flashback sequence, with a set of 'secondary reference points' (as in Kamp and Rohrer \[1983:260\]), or else to turn an event expression into all expression denoting tile consequent state of an earlier occurrence of the contained event (adapting the somewhat different analysis in Moens and Steedmall \[1987:4\]). Assuming the DRT account of states in general, we would say here that the consequent state (of Mary having a broken leg) overlapped with the existing reference time (associated with the event of John meeting Mary); tile earlier occurrence of an event (of Mary breaking her leg) being inferrable from the perfective description of the leg-breakage.</Paragraph>
      <Paragraph position="3"> The DRT notion of temporal overlap is a permissive relation; in a case like (10), we can follow a pair of event-expressions with various state-expressions, all  of which DRT would say denote states which overlap the event already introduced.</Paragraph>
      <Paragraph position="4"> (10) Someone stole Victoria's car on Friday; they wrecked it.</Paragraph>
      <Paragraph position="5"> a. She was very attached to it.</Paragraph>
      <Paragraph position="6"> b. She was very annoyed.</Paragraph>
      <Paragraph position="7"> c. It was unlocked.</Paragraph>
      <Paragraph position="8">  In fact, we would want to say that Victoria was attached to the car before (and perhaps not after) it was wrecked; that sbe was annoyed after (and probably not before) it was wrecked; and that its being unlocked fully overlapped the stealing and wrecking.</Paragraph>
      <Paragraph position="9"> Arguably, we can view the states in (10a-c) as providing respectively some background, a result and an explanation for the events ill (10).</Paragraph>
      <Paragraph position="10"> One approach to representing states is to represent them via intervals of time, bounded by (artefactual) begin-events and end-events. Such an approach is adopted, for example, in Kowalski and Sergot's \[1986\] Event Calculus. In discourse, of course, it is not always possible to find explicit reference to anch beginnings and endings. Whilst not axlvocating such a reductivc approach to states here, we note that in some cases, such as the resultant state in (10b) or the perfect state in (9), the event which initiated that state may be explicitly mentioned. In other cases, such as the background in (10a) and tile explanation in (10c), the event which lead to the state may be only implicit. 2 ~Capturing these differences in a DRT-based theory of ACFES DE COLING-92, NANTES, 23-28 Aor~' 1992 2 8 3 PROC. OF COLING-92, NANTES, AUG. 23-28, 1992</Paragraph>
    </Section>
    <Section position="5" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
Explaining the Data
</SectionTitle>
      <Paragraph position="0"> Now, consider the use of clefts as state-denoting expressions. We would suggest that, in this respect, they be treated like the others we have considered.</Paragraph>
      <Paragraph position="1"> We can say that clefts will denote states which:  1. Overlap with the existing reference time 2. Do not update that reference time 3. Have been initiated by some event, which may be  either explicit or implicit.</Paragraph>
      <Paragraph position="2"> These facts arise directly from the aspectual type of the cleft; in turn, they directly account both for Prince's observations, and our own. Recall examples (1), (48a) and (2). In the first ease, the information about Basil Bernstein's influence is presented via a cleft. Hence, it is presented as a state, overlapping with any previously established time. There is no update to the reference time; hence the information that follows it temporally overlaps with it as well. What event brought about the influential status of Bernstein's writings is not specified. Thus, Bernstein's influence is indeed, as Prince suggests, background to what follows; this is a ease of background, like (19a).</Paragraph>
      <Paragraph position="3"> In Prince's (48a), the timing of the Nazi movement's gathering of mass support is presented via a cleft identifying it as the time of llitler's application of various ideas. Hence, the information about the timing is presented via a state--that of having gathered mass support. In this case, on the DRT analysis, the state overlaps with a reference-time 'just after' IIitler's application of the ideas. Again, the state does not itself update the reference time for the next sentence, so what follows overlaps with the state. What event brought about the state of mass support is clear from the context: it is in fact Hitler's application of the ideas, mentioned in the previous sentence. Thus, this wmdd be a ease of result, like (10b); Prince's suggestion of a causal relation is entirely compatible with this.</Paragraph>
      <Paragraph position="4"> Finally, Mr Butler's authorisation of various actions is presented via a cleft (example (2)). Hence, we have a state of Mr Butler~of having authorised action--and this state overlaps the reference-time established by the previous sentence. The state does not update the reference-time, and so tile subsequent sentence overlaps with this state. Here, the event which brought about the state of Mr Butler fit clearly his authorisation of action. It must have initiated the state, so it lies before tile current referenee time; but we cannot totally order it with respect to the reference-times from the previous sentences of the discourse. This explains why there is a feeling of &amp;quot;temporal regression' and the associated removal from the main time-line; further world knowledge would be required to find the actual relative Incatree of Mr Butler's action.</Paragraph>
      <Paragraph position="5"> The reason de-elefting seenm to disrupt the meaning of the discourse lies in the fact that it can, as here, convert a state-expression back into an eventexpression. This then gives the impression that discourse would, of course, require additional theoretical apparatus; cf. Lascarides and Asher \[1991\].</Paragraph>
      <Paragraph position="6"> the speaker-writer is introduciug a new event into the discourse and updating it in the relevant ways; whereas in the clefted versions, any events introduced by the state itself are either implicit, or identifiable in the previous context. Safe de-clefting must therefore involve the preservation of the stative aspect of the relevant cleft sentence; replacement with a perfect de-clefted sentence should normally suffice.</Paragraph>
      <Paragraph position="7"> Note that where the de-clefted sentence is already stative, de-clefting should not disrupt the coherence of the narrative so severely.</Paragraph>
      <Paragraph position="8"> Explaining the Known Fact Effect We have indicated that the discourse subordination effect of clefts can be traced to their aspectual class. This suggests that we can correlate the syntactic construction with a semantic feature, and that this feature could therefore be recruited by a discourse parsing mechanism, such as the Linguistic Discourse Model proposed by Seha and Polanyi \[1988\].</Paragraph>
      <Paragraph position="9"> As we noted earlier, Prince \[1978\] proposed that what the various clefts had in common was that they marked a piece of information as fact, known to some people, but not necessarily to the hearer. By indicating that they do not accept responsibility for the truth of the statement, the speaker at once denies that they are the 'informational origin', and makes it clear that the validity of the statement is nonnegotiable. null We would like to suggest that the aspectual effect of the cleft can explain the Known Fact Effect, in the following way. On one interpretation of Grice's \[1975:46\] Maxim of Quality, we can say that a considerate speaker's explicit commitments must bc supported by adequate evidence, but that their implicit implicatures need not be, so long as the speaker does not actually believe the implicatures to be false. In the discourses we have discussed, each piece of information the speaker wishes to convey can be transmitted via either an event-expression or a state-expression. When the speaker uses an event expression, they are explicitly introducing a new referential element to the discourse: an event. Let us say that speakers are 'responsible' for explicit introductions only. Now, when a speaker uses a state expression, they do two things: they explicitly introduce a state to the discourse, and they also implicitly refer to two further events; the beginning and ending of that state. But the speaker is not responsible for those events, because they have chosen to use a construction which leaves the events merely inferrable, or loeatable in tile previous discourse context.</Paragraph>
      <Paragraph position="10"> Lascarides and Oberlander \[1992\] suggest that if there is no 'explicit' indication of where a state starts--via tile mention of causes or the use of temporal adverbials--then the start of the state is assumed to be irrelevaut. Here we may gloss 'irrelevant' as: unknown, unknowable or simply to be taken for granted. Thus, conversely, if the speaker deems the start of tile state to be irrelevant to the discourse in this sense, then they can use a simple state-expregsion. This makes a cleft a natural choice for a speaker who wishes simply to assert that an eventuality is current at the reference4ime, without ACRES DE COLING-92. NANTES. 23-28 ^Oral&amp;quot; 1992 2 8 4 PROC. OF COLING-92, NANTES. AUG. 23-28, 1992 indicating anything further about it. So clefts can deliver information which might otherwise have been stated earlier without disrupting the flow of the discourse (of. Polanyi's \[1986:85-87\] 'true starts'); and they call also deliver information without generating responsibility for an initiating event whose loca-tion may be unknown, unknowable or simply to be taken for granted. The former type might be assimilated to what Prince \[1978\] has termed STIU,;SSED-FOCUS ibclefts, and the latter to her INFOItMAT,VE-</Paragraph>
    </Section>
  </Section>
  <Section position="4" start_page="0" end_page="0" type="metho">
    <SectionTitle>
PRBSU PPOSITION clefts.
</SectionTitle>
    <Paragraph position="0"/>
    <Section position="1" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
Clefts and Contrast
</SectionTitle>
      <Paragraph position="0"> The observation that the cleft initiates a subordinate discourse segment also provides us with a potential explanation for a further set of data, namely those clefts which play a eontrastive role in discourse. 3 Contrast (cf. Lyons \[1977\], Werth \[1984\] for a discussion) can be described as relationship of opposition or comparison between two (or more) discourse elements that operates on the basis of soine predicate: For example, in tile following case a contrast holds between the cleft head element the angel and a preceding element, Boaz, with respect to the predicate use this form of greeting: (11) To this the reply is given that from the verse dealing with Boaz there is no proof of divine atlproval, only that Boaz used this form of greeting.</Paragraph>
      <Paragraph position="1"> But in tile second verse it is the angel that uses this form of greeting and hence there is evidence of divine approval.</Paragraph>
      <Paragraph position="2"> It is important to note that the contrastive relationship t,as two distinct components: the two (or more) coutrastive elements themselves, and the semantic content relating those elements, thereby allowing the contrast to take place. In (11), for example, the relating semantic content is easy to find, since it is explicitly stated twice ill a way that allows the commonality between the contrast-supporting predicates to be retrieved inunediately (used this form of greeting ... uses this form of greetiug). In other cases, however, the relating semantic content is not so simple: understanding the contrast between doubling the selling space to 700 square feet and the newfixturcs and fittings in (12), for example, requires a contrastive relation to be constructed out of the non-identical content of the predicates be the greatest e~:pcnse and be costly: (12) Doubling the selling space to 700 square feet was not to be tile greatest expense. It was the new fixtures and fittings ~o fill ~his space that would be costly.</Paragraph>
      <Paragraph position="3"> A similar effect can be observed in (13), in which a contrast takes place between tile cleft head element the lady who obliges and the antecedent a aice 3As noted above, while we would hesitate to make a complete assimilation between the two classes, contrastive clefts seem to fall into the class that P.rince \[1978\] terms STRESSED-FOOUS it-clefts.</Paragraph>
      <Paragraph position="4"> 4The notion that relations of contrast and other kinds of coherence are supported hy inferrable SEAL,S (cf.</Paragraph>
      <Paragraph position="5"> tlitschberg \[1985\], Ward \[1985\]) wmdd also he a useful one for this analysis.</Paragraph>
      <Paragraph position="6"> old-fashioned housemaid in tile following advice to visitors to grand homes: (13) Quite a few of you have asked about tipping, and these days problems can arise. A nice old-fashioned housemaid, labelled by cap and apron, is easy enough; when you leave you will give her your little present as a thankyou for looking after you. It is the 'lady who obliges' thai can confound you; on that point, tile simplest way is to quietly consult your hostess.</Paragraph>
      <Paragraph position="7"> The contrast here operates on the basis of reader perceiving the relationship between the two predicates easy enough and can confound yon. The inferrable predicate for the contrast is therefore something like ease of tipping, and the actual predicates that appear serve to range the two elements--the housemaid and the lady who oblige~-at opposite ends of a scale of ease and difficulty: (14) Quite a few of you have asked about tipping, and these days problems can arise. A nice old-fashioned housemaid, labelled by cap and apron, is easy enough; when yon leave you will give her your little prczent as a thankyou for looking after you. The &amp;quot;lady who obliges' can confound you; on that point, the simplest way is to quietly consult your hoste.'m.</Paragraph>
      <Paragraph position="8"> We would suggest that it is in these more difficult cases, where the contrast-supporting semantic relation is less obvious, or where the contrastive antecedent is less accessible in some other way (for example, in terms of its embeddedness within the structure of the discourse) that the cleft comes into its own. Evidence for this comes from the fact that de-clefting in the simpler cases such as (15) does not cause loss of coherence: (15) To this the reply is given that from the verse dealing with lloaz there is no proof of divine approval, only that Boaz used this form of greeting.</Paragraph>
      <Paragraph position="9"> But in tile second verse the angel uses this form of greeting and hence there is evidence of divine approval.</Paragraph>
      <Paragraph position="10"> In other eases, however, de-clefting has more disruptive effects. While contrast is successfully established by the cleft ill (12), tile de-cleft version, shown in (16), is much less acceptable: (16) ?Doubling tile selling space to 700 square feet was not to be the greatest expense. The uewfiztu,es and fittings to \]ill this space would be costly. What is happening in the de-clefted cases ill order to disrupt the retrieval of the relationship along which the contrast takes place? In our discussion of subordination abow~., we observed that de-clefting gives the impression that the speaker-writer is introducing a new event into the discourse, while in the clefted versions, any events introduced by the state itself are either implicit, or identifiable in the previous context. In the same way, in the contrast cases, the loss of the cleft causes the content of the de-cleft to be interpreted as a new and distinct thematte development. In this way, the de-clefted information fails to identify with information already in AcrEs DE COLING-92, NANTES, 23-28/tOOT 1992 2 8 5 PROC. OF COLING-92, NANTES. AUG. 23-28. 1992 the previous context. Because of this, the identification of the contrastive antecedent, and the semantic information linking it to the current proposition, are not retrieved. In situations where this relationship is not made clear by means other than the cleft (and it can be effected by intonation, or through the availability of an obvious and immediately-preceding antecedent--we do not suggest that clefts are unique in their contrasting function) the reader's default will be to introduce a new eventuality into the discourse, probably (in the absence of other signals) as a co-ordination in the discourse structure. In this way, the information upon which the contrast depends-that the proposition is to be seen as an elaboration on existing content--is not preserved.</Paragraph>
      <Paragraph position="11"> Comments and Conclusions In this paper, we have tried to show that various apparently unrelated aspects of it-cleft function-subordination, the Known Fact Effect, and the facts surrounding contrast--can be explained in terms of the fact that i~-clefts perform a 'stativizing' function. It is as well at this point, however, to sound a cautionary note. We have not yet examined in full thoee cases where de-elefting leaves a state-expression. The prediction is that these cases should not seem as bad as when de-clefting reveals an event-expression, but we have not yet tested the prediction.</Paragraph>
      <Paragraph position="12"> A related issue which ought to be amenable to a pragmatic explanation is the unacceptability of it as clefted constituent3 R-clefts (and wh-clefts, for that matter) cannot take it as clefted constituent: (17) *It is it that John has decided he wants.</Paragraph>
      <Paragraph position="13"> The alternative forms with this and that, however, are acceptable: (18) It is this/that that John has decided he wants.</Paragraph>
      <Paragraph position="14"> The restriction on if cannot be accounted for by a simple restriction on pronouns in cleft head position, as personal pronouns can appear, s An obvious, but incorrect, explanation would be that it is the unstressed variant of that (cf. for example Declerck \[1988:14\], following Kuroda \[1968\]), and so cannot appear in the cleft's 'stressed' position. However, we know that it-clefts regularly appear with no stress on the head constituent (cf. Delin \[1989\] for an analysis); it also appears that it is in any case stressable, as the following (attested) data shows: (19) S: Judy, is there any more soap? J: If yon look in the basket there's that purple one S: I thought you were drying some out on the SBall (p.c.) has pointed out that it is not acceptable as complement of a copular sentence whose subject is it in any case: e.g. *it's it vs. that's it.</Paragraph>
      <Paragraph position="15"> SBall \[1991\] and p.c., in her study of the development of the it-cleft from Old English to Late Modern English, finds no occurrences of it in focus position either in the modern-day it-cleft or in any of its ancestors. This is in spite of the fact that the paradigm of personal pronouns in focus position can be considered complete around the 15th century, with objective case pronouns (e.g. it was me) appearing in the 16th (Ball \[1991:274\]).</Paragraph>
      <Paragraph position="16"> window. What happened to IT? J: That's it S: Oh, so it is A second plausible explanation may be that it, unique among the pronouns, has no cuntrastive reading (el. Werth \[1984:134\]). A contrastive function, however, does not appear to he obligatory for ilclefts anyway, as Declerck \[1988\] among others points out. /t-clefts are frequently found with old, non-C/ontrastive, anaphoric information in the clefted coustituent: null (20) A: Joe Wright you mean B: Yes yes A: I thought it was old Joe Wright who'd walked in at FIRST and Prince's \[1978:898\] written example: (21) It was also during these centuries that a vast internal migration.., from the south northwards took place, a process no less momentous than the Amhara expansion southwards during the last part of the 19th century A possible explanation for it-lessness in it-clefts may be found in work by Linde \[1979\], who relates the alternation of it and thai to the 'in focus' status of the referent in relation to the structure of the discourse. In her study of subjects' descriptions of their apartments, she notes that it is preferred for 'reference within the discourse node in focus', which Linde takes to be a continuation of a segment of discourse describing the same room in the apartment. That is used for reference within the discourse node in focus only when there is some contrast to a preceding node; most of the time, that is used for transitions between nodes.That is, when a room is being described, a second room may be described as leading off lhat. That therefore tends to mark progressions from one node to the next. Can we therefore expect the it-lessness of it-clefts to relate to their position in the discourse structure? That is, do it-clefts appear only in these node-transition situations, and not in the positions of same-node reference? We would expect that an exploration of the lack of it in clefts along these lines might be fruitful.</Paragraph>
    </Section>
  </Section>
class="xml-element"></Paper>