A constructional approach to clefts(*).

Abstract

In this article I will argue that clefts differ in more than simply their information structure from their noncleft counterparts. Clefts are constructions in their own right, whose grammatical features convey specific representational semantics. I will focus on the two main coded relations in cleft constructions: the one expressed by the matrix clause and the anaphoric relation between the complement of the matrix clause and the relative clause. I will indicate in which ways these have not been described satisfactorily in the literature so far and will propose an alternative description. The different matrix clauses of clefts are specific subtypes of identifying, existential, and possessive clauses, which all impose a specific "quantificational" value on their complement: they express exhaustive specification of a set, enumeration, or cardinal quantification of instances. The relative clause takes this specifying, enumerating, or cardinally measured complement as antecedent and incorporates it in a value-variable relation. In this way, I will argue, the different types of clefts, viz. it clefts, there clefts, and possessive clefts, can be given an integrated semantic description.

Introduction

In this article I will revisit the grammar and semantics of the sentence types traditionally referred to as clefts. I will be concerned with it clefts, the neglected category of there clefts, and, marginally, have clefts.

(1) It's John who's causing us trouble.

(2) There's John who's causing us trouble.

(3) We have John who's causing us trouble.

In section 1, I will criticize the traditional approach to "clefts," which has tended to assume that they do not have any representational semantics independent of their noncleft counterparts. Focusing on Huddleston's (1984) articulation of this position,(1) I will point out in section 1 what seem to me the two main descriptive cop-outs in this approach, viz. the failure to give a positive characterization to the relative clause in clefts, and the claim that the matrix clauses of clefts cannot be described as identifying, existential -- or possessive -- clauses. Because these two main "component" structures have not been dealt with satisfactorily, the "composite" structure formed by the cleft has not been described adequately either.

The general approach to clefts advocated here is "constructional" within the long functional tradition represented by, amongst others, Haas (1954: 74), Halliday (1985: 32), Langacker (1987: 316; 1991: 5), and McGregor (1997: 39). According to Halliday's (1985: 32) succinct characterization a construction is a "configuration of functions." This functional approach is "top-down" in the sense that the functions are defined by the HIGHER-LEVEL unit in which they play a role, viz. the construction. It is also "semiotic" in the sense that both the composite structure and its component structures are viewed as SYMBOLIC form-meaning couplings, with the form "coding" the meaning.(2)

An exhaustive description of all the constructional features of clefts is far beyond the scope of one article. Therefore, I will focus on the two main coded relations in clefts, which, as indicated above, have not been described with sufficient precision so far.

In section 2, I will focus on the relative clause and its antecedent in clefts and point out fundamental differences with both restrictive relative clauses and nonrestrictive relative clauses. This will also lead to a revision of certain traditional descriptions of restrictive relative clauses.

In section 3, I will home in on the matrix clauses found in clefts and argue that they are specific subtypes of identifying, existential, and possessive constructions, which all impose a specific "quantificational" value on their complements. In particular, the distinction between "cardinal" and "enumerative" existentials as set out in Davidse (1999a) will be used to further distinguish two types of there cleft.

In section 4, I will bring these findings together into a description of the main constructional features of clefts, which will also lead to an integrated description of the semantic differences between the various types of cleft.

1. Traditional approaches to clefts

1.1. Recognition criteria of clefts

As pointed out by Huddleston (1984: 461), the most distinctive grammatical feature of the it cleft is the fact that the postverbal NP and the relative clause following it do NOT form one grammatical unit. In this respect the it cleft can be contrasted with an "ordinary" identifying clause whose complement contains a restrictive relative clause (henceforth RRC). Consider,

(4) Who was that on the phone? -- It was the boy who/that caused all the trouble.

(5) Who caused all the trouble? -- It was the boy (who/that) caused all the trouble.

Identifying clause (4) specifies that the person on the phone is the boy who caused all the trouble: in this clause the complement is formed by the whole NP with its RRC postmodifier. The it cleft (5) specifies that the person causing all the trouble was the boy. In (5) it is only the boy that functions as postverbal complement. The wh-/th clause is not part of this complement NP. This is clearly shown if we replace the boy by a proper name (Huddleston 1984: 460), as in

(6) It was Tom who caused all the trouble.

As is well known, a determinerless proper name cannot function as the head of a RRC postmodifier (Langacker 1991: 59).

The distinction between it clefts, on the one hand, and identifying clauses with complements postmodified by a RRC, on the other, is also signalled by the distinct intonation patterns that tend to be associated with them. Halliday (1967a: 237) has pointed out that it clefts are typically spoken on a compound fall-rise tone. This allows the speaker to place the marked or contrastive information focus on the postverbal NP, while also having a tonal nucleus on the final lexical item of the tone unit (Quirk et al. 1972: 1046).

(5') It was the boy who caused all the trouble.

In contrast, the identifying clause with complement postmodified by a RRC, (4), would normally be uttered with only one, tone-final, salient element.

(4') It was the boy who caused all the trouble.

As noted by Halliday (1985: 205), RRCs can be recognized intonationally by the fact that they do not constitute a separate tone unit.

Now, the same sorts of contrast as exist between identifying clauses with RRC and it clefts can be observed between existential clauses with RRC like example (7) on the one hand and constructions such as those illustrated by example (8) on the other. Consider:

(7) What can you see on the table? -- Well, there's one thing that has a funny shape.

(8) Could it be anything else? -- No, there's only one thing that's that shape.

The existential in (7) basically informs the hearer that there is "one" instance of a "funny-shaped thing" on the table. That has a funny shape is a RRC that modifies the head thing.(3) (Note that it is intonationally integrated into the matrix clause, which is uttered with one simple tone contour.)

In contrast, the there sentence in (8) asserts that "only one thing" exists that corresponds to the specific shape speaker and hearer have in mind. The complement of the existential predicator is just the NP one thing here. The th clause that is that shape is NOT a postmodifier of thing. This is signalled formally by the intonation associated with (8): the relative clause is not -- as it IS in (7) -- integrated into the tone contour with which the sentence is spoken. Rather, example (8) takes the compound "fall-rise" tone, which was also characteristic of the it cleft in (5).

(7') There's one thing that has a funny shape.

(8') There's only one thing that's that shape.

Moreover, the postverbal NP in (8) can be replaced by a proper name (as in [9]), which also shows that the th clause following it cannot be its postmodifier.

(9) There's only Humpty Dumpty that's that shape.

Clearly, sentences like (8) instantiate another construction than an ordinary existential with RRC. Because of the many similarities with it clefts, I will, following Halliday and Huddleston, refer to examples like (8) as there clefts. The descriptive justification of this categorization will be given later in this article. Meanwhile, it can be noted that the category of there clefts does not have common currency in the mainstream. To my knowledge, relatively few authors (Halliday 1967a; Huddleston 1984; Hannay 1985; Collins 1992; Davidse 1999b) have argued that there clefts should be recognized.

However, recognition of there clefts as a category parallel to it clefts puts some of the constructional features they share into sharper relief and also reveals that these features have not yet received an adequate explanation. The formal status and semantic value of the th/wh clause in it and there clefts is a case in point. This question has traditionally been skirted by describing sentences like (5)-(8) as "cleft" transforms of "noncleft" originals. In the next section, I will indicate a number of problems with such an approach.

1.2. The traditional "transformational" approach to clefts

Huddleston (1984) is an interesting example of the traditional approach to clefts, which, though not transformational in the strict sense, holds that "cleft" constructions can only be described "indirectly, in terms of the non-cleft counterpart in conjunction with the cleaving operation" (1984: 462). For examples (5) and (8), the noncleft counterparts are (5') and (8'):

(5') The boy caused all the trouble.

(8') Only one thing is that shape.

These single clauses, so the traditional explanation runs, are divided into two distinct parts assigned to different clauses, viz. a matrix clause, it was the boy, there's only one thing, and a "secondary" clause, (who/that) caused all the trouble; (that)'s that shape.

As to the matrix clauses of it and there clefts, Huddleston (1984: 462, 470) holds that both it + be and there + be are "fully grammaticalized features of the construction whose contribution to the meaning is not directly predictable from their use in other kinds of clause." The main semantic function he ascribes to clefts is a TEXTUAL one.(4) It clefts "highlight" an element, viz. the postverbal NP, and associate, due to the use of definite it, a component of uniqueness with this element (Huddleston 1984: 466-467); for example, in (6) it is Tom, and no one else, who caused the trouble. Informationally, the postverbal NP often provides "new" information, with the relative clause containing the "given" information (as in [5] and [6]). However, as pointed out by Huddleston (1984: 465), the information in the relative clause can also be new, in which case the highlighted element is often anaphoric or otherwise given, as in the following example about "The Burghers of Calais":

(10) Far from humbling himself before the king of England, the burgher was leaving the city to descend toward the camp. It is this that gives the group the feeling of march, of movement (Butler 1993: 204).

Huddleston (1984: 469) does not attribute a "highlighting" function to there clefts but holds that they recast the noncleft into a "thematic variant," which presents the description of the event in postverbal position (One man kept interrupting -- There was one man kept interrupting). In other words, despite the many structural similarities between it and there clefts, he does not recognize analogies between their semantic values. I will argue that this is unsatisfactory and will propose an alternative analysis that does bring out certain semantic parallels based on formal similarities.

Turning to the th/wh clause, then, Huddleston views it as displaying some, but not all, of the characteristics of a defining relative clause. First, antecedents of relative clauses in it clefts have a wider range than those of ordinary RRCs: they can be not only NPs with common noun heads, but also pronouns and proper names (Declerck 1988: 52), as well as prepositional phrases and embedded clauses (Quirk et al. 1972: 953).

(11) It was Tom/you who/that caused all the trouble.

(12) It was in September that/zero I first noticed it.

(13) It was because he was ill that/zero we decided to return (Quirk et al. 1972: 953).

Note that the complements in there clefts can be pronouns or proper names too.

(14) There's only Humpty Dumpty/him that's that shape.

Prepositional phrases and, marginally, clauses also seem possible as complements of there clefts.

(15) There's on the table that you may have left it.

(16) There's when you were away that it might have happened.

Further, in both it and there clefts, that and zero are more common than wh forms. With some types of antecedents, wh forms are even ungrammatical (Quirk et al. 1972: 953; Huddleston 1984: 393-396; 460-461); for example,

(17) It is in November that/*when you should prune the roses.

(18) It's only by train that/*how you will get in.

(19) There's on the platform that/*where you can wait.

Moreover, zero realization of the subject in the relative clause, which is restricted to informal registers in ordinary RRCs, occurs unproblematically in it and there clefts (Quirk et al. 1972: 959; Huddleston 1984: 460), as in (5') and (8').

(5') It was the boy caused all the trouble.

(8') There's only one thing is that shape.

Huddleston (1984: 462) concludes that the relative clause in it clefts is of a kind that is "sui generis, unique to this construction."

We thus see how this approach to clefts posits that they cannot be described as constructions in their own right, but only "indirectly" as transforms of their noncleft counterparts. The matrix clause, it is claimed, should not be described as an identifying or existential clause, despite its apparent similarities to these clause types. In fact, in the Huddleston approach, the matrix clause is denied any representational import. Similarly, no attempt is made to describe the special properties of the relative clause in clefts beyond the observation of fairly superficial differences with RRCs.

Opposed to this, I will argue that the two main component structures of clefts -- matrix and relative clause -- CAN be given positive functional characterizations, which, together, lead to a more precise insight into the semantics of cleft constructions.

2. The relative clause in cleft constructions

The first main point to be tackled in a constructional approach to clefts is the special status of its relative clause. The traditional observations about the relative clause in clefts stop short of the real issues, particularly THE EXTENT .AND THE STRUCTURAL STATUS OF THE ANTECEDENT. In this respect, I will show that Huddleston's analysis of the relative clause in clefts

1. does not go far enough in developing the differences with restrictive relative clauses;

2. does not account for the differences from nonrestrictive relative clauses;

3. does not recognize that this sort of relative clause is found in other constructions besides it clefts.

2.1. Differences between relative clauses in clefts and restrictive relative clauses

As we saw above, Huddleston makes the observation that relative clauses in it clefts allow for a wider range of antecedents than RRCs, viz. proper names, pronouns, adverbials, and clauses. In this section I will argue that there are more fundamental differences between RRCs and relative clauses in clefts. In fact, focusing on the contrast between RRCs and clefts will bring out a problem in the received approach to RRCs.

The traditional approach to RRCs tends to hedge on the question whether it is the whole NP or only the nominal head that functions as antecedent.(5) If mostly the assumption is withheld that RRCs have the whole NP as antecedent (e.g. Huddleston 1984: 394) this may be because the semantic function of the RRC is traditionally explained as giving information necessary to identify the referent of that antecedent. And "picking out a referent" presupposes a full NP with determiner. For instance, in (20) who stood in the corner is said to provide the necessary information to identify the girl in question.

(20) The girl who stood in the corner is Mary Smith (Quirk et al. 1972: 858).

Nonrestrictive relative clauses (henceforth NRRCs) are then said to merely add information that is not essential to identification, because the antecedent has already been uniquely identified. For instance, in (21) the antecedent is the proper name Mary Smith, which is as such uniquely identifiable.

(21) Mary Smith, who is in the corner, wants to meet you (Quirk et al. 1972: 858).

The traditional description is not very explicit either about the head--modifier relations in NPs with RRCs and mostly leaves it at noting that, for instance in (20), the determiner the is premodifier, girl head, and who stood in the doorway postmodifier, without specifying any internal bracketing within that dependency structure.

A further problem with the traditional explanation of RRCs versus NRRCs is that it applies only to definite NPs but does not clarify what sort of distinction is involved when the "antecedent NP" is indefinite, as in

(22) She was wearing a dress that I'd never seen before (Huddleston 1984: 398).

(23) She was wearing an exclusive dress, which I'd never seen before.

According to the received analysis the antecedent in (22) is a dress. Since this nominal is indefinite, the function of the RRC cannot be that of "aiding identification." In example (23) the antecedent (an exclusive dress) is also indefinite, so the NRRC cannot be said to add nonessential information to an already identified antecedent.

Against this traditional description, Langacker (1991: 430f.) explicitly says that the antecedent of the RRC is the nominal head minus determiners. Moreover, he specifies that the RRC modifies the nominal head only, and that this complex is grounded by the determiner. The internal dependency structure of a NP with RRC is thus, according to Langacker, that shown in Figure 1.

[Figure 1 ILLUSTRATION OMITTED]

Langacker's analysis of RRCs is part of an alternative description of the NP that correlates the various structural elements found in it with its four basic semantic functions, viz. type specification, instantiation, quantification, and grounding. NPs with common noun heads display the UNMARKED correlations between forms and basic functions. Since common nouns designate classes or types, such as dog, the common noun head expresses the "general type" of the entity designated.(6) Nondetermining modifiers express a subcategorization of this general type, such as golden dog, dog that is faithful. Head and nondetermining modifiers thus constitute the full "type specification." To tie the type specification to specific instances, it has to be modified by a determiner, such as a golden dog, the dog that is faithful. The grounding, or "identification," of the instances is realized by determiners such as articles, demonstratives, possessives, and relative quantifiers, such as a dog (one instance, not presumed known), your dogs (more than one instance, identified in terms of possession), either dog (one instance not presumed identifiable from a reference set of two). NPs consisting of proper names or pronouns present the MARKED option in that they do not have distinct structural elements to realize the four semantic functions of the NP. For instance, Mrs. Widmerpool implies a "type specification" (e.g. "female," "married") and designates a single identified instance.

Within this overall descriptive framework, Langacker analyzes the RRC as an element of the type specification: a RRC "restricts the head noun's type specification" (Langacker 1991: 432). The internal assembly of the type specification precedes the grounding of the instances by the determiner. Thus, in Langacker's description, the antecedent of the RRC is the nominal head designating the type (e.g. dress), not the full NP designating the instance (a/the dress).

In what follows I will argue that Langacker's analysis receives further support from an in-depth analysis of determiners and quantifiers in relative constructions. Huddleston (1984: 394-395) himself has made a convincing case that the relative pronoun does not have an ANAPHORIC relation with the full NP, but only with the head minus determiners. This is particularly clear when the determiner is a quantifier such as no, every, any, each, etc. Consider

(24) No one who knows her would consider her a threat (cf. Huddleston 1984: 394).

Here, the relative pronoun clearly does not point to no one, for this would entail that "no one knows her" and that is not part of the intended meaning of (24). Likewise,

(25) Every vehicle which they had tested had some defect (Huddleston 1984: 394).

does not entail that they had tested every vehicle. Huddleston (1984: 395) further points out quite rightly that even in the seemingly straightforward cases with definite article, as in

(26) I didn't like the guy who spoke first.

there is no pragmatic equivalence between who spoke first and the guy spoke first. In the latter expression, the description guy is sufficient to pick out the intended referent, whereas in (26) the whole description guy who spoke first is judged pragmatically necessary by the speaker for picking out the person in question.

With these observations, Huddleston seems to come close to the sort of position advocated by Langacker (1991), but rather than going the whole way, he ultimately reverts to the traditional position in which the antecedent of the RRC is considered to be the full NP. He then posits an alleged "lack of equivalence" between relative anaphor and antecedent, which goes against the logical assumption that the antecedent IS the meaning anaphorically pointed to.(7)

I subscribe to the internally coherent approach proposed by Langacker: the restrictive relative pronoun is an anaphor that requires the disentangling of "type specification" and "grounded, quantified instance" as grammatical categories. Who in (26) has got the "type" guy as antecedent. Once one recognizes that it is the "type description" that functions as antecedent in this anaphoric relation, there is no need to posit a -- counterintuitive -- mismatch between antecedent and anaphor.

Ultimately, Huddleston's problems with the delineation of nominal antecedents may be related to the traditional philosophical conception of "reference," which considers only the relation established between the full NP and the "instance" referred to (be it conceptualized as an instance in the real world or in the universe of discourse). In fact, the reference act is more complex: the speaker can bring the hearer into mental contact with the instance (Langacker 1991: 91) only if the hearer can also successfully process the "type description," which should be sufficiently informative in view of the information accumulated in the text up to that point. "Defining" modifiers such as RRCs play an important role in the reference act precisely by further specifying the type description.

It is clear enough that the determiners in (24)-(26) are not part of the antecedent, but to arrive at a coherent picture of determiners in NPs with RRC, we have to look into this matter more systematically. More specifically, we have to investigate how the main types of identifiers and quantifiers behave when the NP contains a RRC as postmodifier. Interestingly, it is by systematically comparing RRCs with clefts that light can be shed on this matter.

Let us begin with identifiers, which, as is well known, indicate whether the instances designated by the NP are "presumed known to the hearer" (definite reference) or "not presumed known to the hearer" (indefinite reference).

Let us first (re-)consider examples with definite identifier and attempt to bring a number of earlier observations together into a more coherent whole.

(27) At that meeting last night,

     a. I didn't like the man who spoke first.

     b. *I didn't like John who spoke first.

(28) When God asked why they were wearing clothes,

     a. It was the man who spoke first.

     b. It was Adam who spoke first.

In the it cleft, the complement slot can take NPs with definite determiner as well as proper nouns. By contrast definite NP the man in (27) cannot be replaced by a proper name: *I didn't like John who spoke first. This distributional fact is well known but little reflection has been devoted to why this is so. Following the lead of Langacker's analysis, I propose that a proper name cannot function as antecedent of a RRC because it does not symbolize the "type description" as a separate element that the relative pronoun can refer back to. This line of reasoning is confirmed by the fact that a RRC IS possible after determiner + proper noun in the special use illustrated by

(29) I've just met the John you used to go out with.

which could be used in a context containing at least two people with the name John. As pointed out by Langacker (1991: 59), John is used as a common noun in such examples: its lexical predication is not an individual but the category delineation "person with name John," and it requires a determiner to be tied to a specific instance. The RRC in (29) thus has the general type "person with name John" as its antecedent. In contrast, in clefts, NPs with definite determiner + common noun head Do alternate with proper names (example [28]). Clearly, the relative pronoun does not have an anaphoric relation to the "type" here but to the "grounded instance" designated by the full NP (the man, Adam).

How does the RRC function then in the referential act performed with a definite NP containing a RRC? As pointed out above, the more elaborate "type specification" construed by nominal head + RRC, such as man who spoke first in (27), is judged necessary by the speaker in that specific context to allow the hearer to pick out the instance in question. But in order to understand fully the contribution made by the RRC, we have to look more deeply into the mechanisms involved in definite identification. I (1999a) have proposed that, with NPs introduced by a definite determiner, NPs, the type description defines a reference mass(8) of "all the instances corresponding to that type in the discourse context." For instance, in both examples (30) and (31), the contextually relevant reference mass is "all exams that still need correcting." The definite article in (30) refers to all of those, while the possessive determiner in (31) refers to only some instances of that reference mass.

(30) Here are the exams that still need correcting.

(31) If you take care of your exams that still need correcting, I'll take care of mine.

NPs with definite article and singular count noun can be used without ambiguity if there is just one "contextually relevant instance" corresponding to that type, which also happens to be the one actually referred to, as is the case in (28). The context of the Garden of Eden evoked by this example contains only one instance of the type "man." Of course, many other contexts will contain more than one instance of the type "man." By adding a RRC to the general type, the "reference mass" may be narrowed down so that only one instance in the discourse context corresponds to it. For instance, in (27), there were presumably several "men" at the meeting, but by using the description man who spoke first, the set of potentially relevant instances is restricted to just one. Hence, the definite NP the man who spoke first can be used unambiguously to refer to the person in question. (This is the use that is so often cited as the paradigm case of RRC in textbooks.)

Let us now turn to RRCs with indefinite determiner, which contrast with relative clauses in there clefts.

(32) A tie is always an appropriate present for a man who is rather conservative.

(33) There is not a white man dances as well as black men.

In (32), the RRC contributes to the "type specification": speaker and hearer are thinking of an instance of the category "man who is rather conservative." By contrast, it is clear that the there cleft in (33) is not correctly paraphrased as "There's not an instance of the category `white man dances as well as black men.'" The point of this there cleft is that there is not a white man of whom it can be predicated that he "dances as well as black men." The zero relative pronoun clearly points back to the full NP not a white man so that the predication made in the relative clause is also -- indirectly -- made of this antecedent.

As for the RRC in (32), Langacker's point holds again that the RRC is frst assembled with the nominal head and only then is this elaborate type specification integrated with the indefinite article, which presents the instance as "not identifiable." "Not identifiable" can mean basically two things. The instance may be known to the speaker, but may be "being introduced" into the discourse for the hearer's benefit. In this case, the type of identification is "indefinite," but "specific." It is also possible that no specific "instance" corresponds to the indefinite NP for either speaker or hearer, in which case we have "nonspecific" indefinite reference as in (32). In either case, the "type specification" that contains the RRC has a purely CLASSIFICATIONAL function: it does not defne a reference mass. An indefinite NP merely designates "some" instantiation of a type, without indicating whether there are -- or aren't -- other instances of the type in the discourse context.

The next issue to be looked at is the role of quantifiers in RRCs and clefts. Following Milsark (1976) and Langacker (1991: chapter 2), I (1999) hold that the two main types of quantifying elements found in the NP are cardinal and relative quantifiers. Cardinal, or "absolute" (Langacker 1991: 82f.), quantifiers express the INTRINSIC magnitude of the designated mass without giving any information about how this amount "stacks up in relation to" (Langacker 1991: 82) all the instances of that type in the context. For instance, if you are fighting a mouse plague in your shed and you report that

(34) Three mice were found dead in the shed.

you give information about how many mice are dead, but not about how many mice are still alive in the shed. In contrast, a relatively quantified NP COMPARES the actually designated mass with the reference mass of all contextually relevant instances and indicates whether the former coincides with, or is a part of, the latter. For instance,

(35) Most mice in the shed are still alive.

conveys that "most" mice of the total mouse population in the shed are still alive.

How do cardinal quantifiers behave in RRCs and clefts?

(36) For one man who's just, I'll save the city.

(37) There's only one thing that is that shape.

We can note again that the cleft indirectly predicates "is that shape" of only one thing: the antecedent of that is the full complement NP only one thing. On the other hand, the RRC has only the noun man as antecedent and thus constructs the more elaborate type man who's just. Cardinal quantifiers such as one "man who's just," or five/some "men who are just" indicate the intrinsic size of the instantiation. Hence, the type specification containing the RRC does not contribute to the definition of a reference mass but has a purely "categorizing" function.

Finally, how do relative quantifiers in clefts behave in comparison with those associated with RRCs?

(38) All students who attended will receive a bonus point.

(39) It was all the passengers who had committed the murder.

In (38) the relative pronoun does not point back to all students; this clause does not state that "all students attended." Rather, it conveys that all instances of the category "students who attended" will receive a bonus point. Note that with a relatively quantified NP such as all students who attended, the type specification evokes a reference mass of all contextually relevant instances, over which all quantifies universally. By contrast, the relative clause in (39) is linked to all the passengers by the relative pronoun who: hence had committed the murder is indirectly predicated of all the passengers. In other words, the antecedent of the RRC excludes relative quantifiers, whereas that of the relative clause in clefts obviously includes them.

In conclusion, against the traditional assumption that the antecedent of RRCs is a full NP, we have seen that the antecedent of RRCs systematically excludes identifiers and quantifiers. In contrast, the relative clauses in clefts have the full complement NP as antecedent. We can also note that it is precisely the inclusion of identification and quantification in the antecedent of clefts that SANCTIONS the well-known systematic alternation with noncleft counterparts, as with (40) for (39):

(40) All the passengers had committed the murder.

In the cleft (39), the relative clause predicates information about a specific situation INDIRECTLY -- via the intermediary of the anaphoric relative relation -- of the antecedent NP. In the corresponding noncleft (40), the same clause is integrated DIRECTLY with that NP. Note however that the semantics of the cleft differ from those of the noncleft because the matrix clause frames the anaphoric relation between antecedent and relative clause in a specific way (see sections 3 and 4).

2.2. Differences between relative clauses in clefts and nonrestrictive relative clauses

In terms of their antecedent, nonrestrictive relative clauses differ, in turn, from the relative clauses in clefts. The antecedent of a NRRC can be a noun phrase, (41), an adjective, (42), an adverb or prepositional phrase, (43), a part of the matrix clause, (44), or the whole matrix clause, (45); for example,

(41) We'll ask Barry, who is a specialist on those matters.

(42) They're going to paint their house purple, which I would never paint my house.

(43) They talked about it yesterday/on Monday, when they vetted all the applications.

(44) The members of the board changed the order, which we have never done.

(45) They always looked down on him, which they should not have.

Purely in terms of a grammatical class analysis, the range of antecedents found in relative clauses in clefts might appear not to be very different from that in NRRCs. In it clefts, for instance, we also find, besides NPs, (46), adjectives, (47), adverbs and prepositional phrases, (48), and parts of, (49), as well as full clauses, (50).

(46) It's John who caused the trouble.

(47) It's blonde that she is.

(48) It's then/during the holidays that they met.

(49) It's holding hands that they were doing.

(50) It's while you were away that it might have happened.

However, there is a fundamental structural difference between the antecedents of NRRCs and those of relative clauses in clefts, as far as the "non-NP" fillers are concerned, that is, adjectives, adverbs, prepositional phrases, and clausal antecedents. The latter are rankshifted into the nominal complement slot of clefts, whereas the antecedents of NRRCs display the structural assembly congruent with their rank. The concept of "rankshift" has evolved within the Hallidayan tradition (Halliday 1981 [ 1965]; Huddleston 1984; Matthiessen and Thompson 1989; Hopper and Closs-Traugot 1993; McGregor 1997). McGregor's recent (1997: 127) definition brings out best both the "external" and the "internal" consequences of rankshift: rankshifting refers to the process whereby a unit of a given rank is "reclassified" as a unit of a different rank, "as a result of which it takes on certain grammatical and semantic properties inherent to that rank." In other words, not only does the rankshifted unit function in a foreign structural environment (such as a clause in a nominal slot), it is also internally reclassified.

Thus, when functioning at their own rank, adjectives and prepositional phrases express "relational" concepts (Langacker 1991: chapter 1) such as "quality" or "location." Likewise, the inherent designatum of a clause is a "situation," more specifically, the predicate designates a type of situation and the full clause a specific instance of a situation (Langacker 1991: 33). The antecedents of NRRCs that are not NPs all have the semantic value congruent with their rank, such as "attributable quality" in (42), temporal location in (43), type of situation in (44), and instance of situation in (45). However, the "non-NP" units functioning as complement in clefts are rankshifted into the nominal complement slot and are in this sense "nominalized" (Halliday 1985: 219). They function in a strictly "nominal" slot, such as that of the identifying complement in it clefts.(9) That is, they fulfil the role of a NP-type element of structure, which also requires them to discharge basic NP functions such as "identification" and "quantification." They fulfil these functions in a way similar to how proper names realize identification and quantification (cf. Langacker 1991: 148), viz. without separate symbolization by determiners. The rankshifted units are "cited" in their entirety like a proper name, and they designate single and uniquely identified entities, which may be individuals (e.g. then, they have always looked down on him) or generic entities (e.g. blonde, holding hands).

I propose that it is precisely the "rankshifted," nominalized, status of such antecedents in clefts that explains their resistance to relative adverbs such as when, where, etc.

(51) It's upstairs *where/that she keeps her records.

(52) It's in September *when/that you should plant them.

(53) It's before you get married *when/that you should travel.

If the complement in clefts is realized by a rankshifted element such as adverb, prepositional phrase, or clause, then its semantic profile is that of a -- reified -- entity. Hence, these "rankshifted" units are systematically referred to by relative pronoun that, and occasionally which, whose general categorial features are those of "inanimate/abstract entity." These antecedents cannot be referred to by relative adverbs such as when, where, and why, because the rankshift has superimposed NP features on them.

It should be noted that enumerative there clefts can also take rank-shifted prepositional phrases and clauses. Enumerative there clefts, which will be discussed in more detail in section 3, typically have definite NPs in complement position. Compare

(54) There's Tom and Dick that/who caused trouble.

(55) There's under the stove that still needs dusting.

(56) There's when you were away that it might have happened.

The mechanism of rankshifting found here is fully parallel to that in it clefts. Prepositional phrases and clauses are rankshifted into the nominal complement slot and are reclassified as definite NPs. Like proper names, they are "cited" in their entirety and given definite identification, but without separate coding of their definite status. As we shall see in the next section, the more "enumerative" feel associated with the complement stems from the distinct constructional semantics of the there clause.

Note also that the enumerating complement in there clefts accepts fewer types of rankshifted units than the identifying complement in it clefts. For instance, rankshifted adjectives, and prepositional phrases expressing receivers and beneficiaries, are possible in it clefts but seem virtually excluded in there clefts (K. Van den Eynde, personal communication).

(57) It's to John that you should give it.

(58) It's for John that you should do it.

(59) It's flightly that she is.

(60) ?There's to John that you should give it.

(61) ?There's for John that you should do it.

(62) *There's flighty that she is.

Acceptability judgments will vary here and one can always try to make the enumerative examples more acceptable, as in

(63) For whom am I doing all this? -- Well, there's for your mother and father that you could be doing it.

But there can be no doubt that it clefts allow for a greater variety of rankshifted units than there clefts. Perhaps this is due to the fact that "ordinary" identifying clauses readily accept rankshifted units as one of the terms in the equation they express, as in

(64) The best way is by train.

(65) The moment to do it is when the plants have shed their leaves.

"Ordinary" enumerative existentials do not, of course, predicate any equation and do not normally contain rankshifted units. They seem to need the "secondary" relation expressed by the antecedent and the relative clause in clefts to sanction the use of rankshifted units.

Let us, in conclusion to this section, summarize the main differences between NRRCs and relative clauses in clefts. The antecedents of NRRCs may belong to various grammatical classes such as NP, PrepP, PredP, and clause. There is no necessary match between the class of the antecedent and the function of the relative pronoun in the relative clause. Compare

(66) If I did fall off, which there is no chance of. :: *There is no chance of I did fall off.

(67) He had to do it by hand, which was time consuming. :: *Do it by hand was time consuming.

For instance, in (67) the antecedent is the predicate do it by hand, but the function of the relative pronoun is subject -- which cannot be realized by a bare infinitive (*Do it by hand was time consuming).

In contrast, the antecedents of the relative clauses in clefts are either NPs or reclassifications as NPs of other grammatical classes (adjectives, adverbs, prepositional phrases, infinitives, gerunds, clauses). In the latter case, we have rankshifted units that "quote" the form of the original class but superimpose the functions of a definitely identified NP on them. In clefts, there is a general match between the function of the relative anaphor and the original, quoted, class of the antecedent. In that form, the antecedent can generally simply replace the relative pronoun.(10) Compare

(68) It's his falling off that there is no chance of. :: There is no chance of his falling off.

(69) It's by hand that you should do it. :: You should do it by hand.

This is in fact the second factor, besides the inclusion of quantifiers and determiners in the antecedent, which enables the systematic alternation between cleft and noncleft counterpart. The rankshifted antecedent of the RC in clefts still displays the class that matches up with the function of the relative pronoun. Therefore, the relative clause can be reintegrated with its antecedent.

Thus, careful comparison of the relative clauses in clefts with RRCs and NRRCs has revealed the two most distinctive features of the antecedents of the former. First, the antecedent of the relative clause in clefts is the full NP, including the identifiers and quantifiers. Second, the antecedent of the relative clause in clefts is always a NP, either as such or as the result of a reclassification of non-nominal units. As we have seen, it is these two factors that motivate the well-known systematic alternation between cleft and noncleft. As a result, this famous alternation loses something of its mystique. Definitely, it provides no justification for regarding the cleft construction as a "transform" unanalyzable in its own right. The special characteristics of the antecedent in clefts CAN be given a positive description and provide part of the explanation of the specific semantics of the cleft construction.

2.3. The sort of relative clause found in it clefts is also found in other constructions

In this section, the point will be made (against Huddleston 1984: 462) that relative clauses of the sort found in it and there clefts are not restricted to these environments. They are also found in possessive clauses such as

(70) Who could be considered for this job? Well, we have/you've got Mick and Di that could be considered.

With possessive clauses we find a contrast that is very similar to that between identifying clause with RRC versus it cleft, and between existential clause with RRC versus there cleft. Consider

(71) I have one daughter who plays the guitar and two daughters who play the piano.

(72) So, you're all on your own then? -- No, I have a daughter/my daughter/Elizabeth who/that helps me.

Example (71) is a clause expressing possession. Its complement is formed by two coordinated NPs with postmodifying RRC: one daughter who plays the guitar and two daughters who play the piano. Who plays the guitar defines a subclass of "daughter," which contrasts with the subclass "daughters who play the piano" (Huddleston 1984: 400). In contrast, (72) turns out, on closer examination, to have the sort of relative clause whose antecedent is a full NP. Who/that helps me makes a predication about definite instances (a daughter, my daughter, Elizabeth), not about a type of which it delineates a finer subtype. Note that the antecedent of the relative clause in (72) can be a definite NP (e.g. Elizabeth, my daughter) or an indefinite NP (e.g. a daughter). With an indefinite complement, this type of example can be ambiguous between a "cleft" and an RRC reading:

(73) RRC: (What are your daughters like?) I have a daughter that helps me and two that don't.

(74) Cleft: (So, you're all on your own then?) No, I have a daughter that helps me.

The functional parallels between this type of construction and there clefts are obvious and have also been pointed out by Hannay (1985: 92), who has drawn attention to the existence of these "possessive" clefts. In the next section, the parallels between possessive and existential clefts will be described in more detail.

3. The matrix clauses in clefts

In the traditional approach to clefts, the "representational" semantics of the matrix clauses have generally been neglected or even denied. Thus, both there + be (Collins 1992: 432) and it + be have been claimed to be "fully grammaticalized features of the construction whose contribution to the meaning is not directly predictable from their use in other kinds of clauses" (Huddleston 1984: 462).

Against this position, I will make the case that the matrix clauses of it, there, and have/got clefts do have the semantic import of identifying, existential, and possessive clauses, respectively. I will argue that, if one does NOT bring in the distinct semantics of the matrix clauses, one cannot explain the distinct semantics of the corresponding clefts. Consider the contrast between examples (75) and (76):

(75) It's Jim who makes the coffee.

(76) There's Jim who makes the coffee.

As pointed out by Halliday (1967a: 238), (75) specifies that it is ONLY Jim who makes the coffee, while (76) enumerates Jim as one, "possibly among others," who makes the coffee. The very existence of the there cleft, besides the it cleft, shows that the latter's meaning of "exhaustive" identification is not simply due to the definite complement NP in (75). The complement NP in (76) is similarly definite, but (76) does not express "exhaustive" identification. I will argue below that the semantic contrast between (75) and (76) stems from the distinct semantics of the whole matrix clause.

In this section, I will show that the matrix clauses of clefts are very specific subtypes of identifying, existential, and possessive clauses, which all have a common denominator, viz. the fact that they impose a specific QUANTIFICATIONAL value on their COMPLEMENT. In subsection 3.1, I will discuss the subtypes of identifying, existential, and possessive clauses with specifying or enumerating complements that function as the matrices of clefts. In 3.2, I will look at the rather distinct subtype of the "cardinal" there cleft, in which the obligatory CARDINAL quantification of the complement in the matrix clause plays a central role.

3.1. Matrix clauses with specifying and enumerating complements

The identifying, existential, and possessive matrix clauses found in clefts correspond exactly to the clause types found in a context such as the following:

What's for supper?

(77) It's spaghetti.

(78) Well, there's spaghetti.

(79) Well, you've got/we have spaghetti.

In what follows I will briefly characterize the constructional semantics of each, focusing on the question of what sort of quantitative effect is associated with the complements of these constructions.

Example (77) is an identifying clause with subject it. Identifying constructions have been defined by Halliday (1967a: 224) as providing a definite value, the "identifier," for an element to be identified, the "identified." The identified can be likened to the unknown "x" in a mathematical equation and the identifier to its actual value in that equation. In (77) the identified is the general pronoun it and the identifier is spaghetti.

Example (78) is an enumerative existential (Lumsden 1988: 150f.; Davidse 1999a). This construction type enumerates instances -- in the limiting case just one instance -- of a contextually given type, such as "things that can be eaten for supper." It is generally accepted that (77) specifies exhaustively "what's for supper," while (78) merely lists -- potentially nonexhaustively "things available for supper."

Constructionally, this semantic contrast is triggered by the distinct subjects of the two clause types. Halliday and Hasan (1976: 101) have proposed that the general pronouns it and there stand in a systemic relation to each other: the first is definite and the second indefinite. Following their lead, I (1999a) have argued that definite it conveys quantificational exhaustiveness, while there designates an unspecific -- amount of the numerical scale. This proportionality is brought out nicely by the following alternation between clefts and pseudoclefts:

(80) It's Jules and Jim that got away with it.

(81) The ones that got away with it are Jules and Jim.

(82) There's Jules and Jim that got away with it.

(83) Ones that got away with it are Jules and Jim.

It points to ALL the instantiation in the relevant discourse context of the category in question. In this sense, its semantics involve comparison with a reference mass. It is the RELATIVE general pronoun, which quantifies exhaustively over the whole reference mass. Note that in the corresponding pseudocleft (81), the also encompasses all the instances in the discourse space of the general type expressed by ones that got away with it. On the other hand, there in (82) designates an unspecific amount of the numerical scale. In the indefinite pseudocleft (83), the corresponding form of quantification is expressed by the zero article, which conveys a cardinal form of quantification: it designates -- in an unspecific way -- the intrinsic size of the predicated mass ones that got away with it. There is no reference mass involved here. We can thus characterize there as the CARDINAL general pronoun.

Example (77) is an identifying clause with exhaustively quantifying it as subject. Therefore, the complement spaghetti is construed as a specificational element. "Specificational" is used here (as in Declerck 1988) in the strict sense of specifying the exact value corresponding to the unknown factor in the equation. Thus, spaghetti exhaustively specifies the entities corresponding to the contextual type "things to be eaten for supper." We can, therefore, characterize spaghetti as a specificational complement.

In contrast, the cardinally quantifying subject there in (78) points to its complement spaghetti as an enumerating complement. I (1999a) have proposed that this "enumerating" scheme is ultimately an "ordinal" scheme. In an enumerative there cleft such as

(84) First there was the Commonwealth and then also the European to go for (CB).

the ordinal scheme of the enumeration is made explicit by the adverbs first and then also. As argued in Davidse (1999a), the ordinal scheme found in enumerative existentials is an EXTENSION of the "cardinal" meaning of ordinary existentials. The extension from cardinality to ordinality is clearly illustrated in existentials that combine the cardinal and the enumerative aspect, as in

(84') There were two important competitions: first, the Commonwealth Games and, then, the European Championships.

The limiting case of enumeration is to enumerate just one instance. Thus, example (78) lists only spaghetti as an instance corresponding to the contextual type "things to be eaten for supper."

Lists are mostly felt to be "possibly incomplete" as in (85), but Rando and Napoli (1978) have rightly pointed out that, with falling intonation, lists can also be presented as "complete" as in (86).

(85) I don't have any friends. Oh, don't be silly! There's John and me and Susan and Peggy (Rando and Napoli 1978: 308).

(86) What's worth visiting here? There's the park, a very nice restaurant, and the library. That's all as far as I'm concerned (Rando and Napoli 1978: 300-301).

Ultimately, the difference between specificational identifying clauses and enumerative there clauses lies in the distinct quantificational mechanisms imposed on the complement. Like its subject it, the specificational complement involves RELATIVE quantification; it involves comparison with a reference set or mass, with which the actually predicated mass coincides completely. If in answer to the question Who murdered Caesar? I say, It was Brutus and the senate, I specify the full set of "murderers of Julius Caesar." In contrast, the enumerating complement is, like its subject there, concerned only with a part of the numerical scale: it merely pegs the enumerated instances to a numerical scale, as in There was Brutus and the senate. The default implication is that the list may not be exhaustive. It requires special intonation to indicate that the list is complete.

Finally, possessive clauses such as (79) have also been characterized as "enumerative" (Quirk et al. 1972: 961). The possessive verbs convey general meanings such as "have at one's disposal" or "have to consider." There is no constructional feature in them triggering a specificational meaning: the complement is not a specificational one, but one with enumerative features similar to those in enumerative there clauses, which can list items. For instance,

(87) Who could apply for that job? Well, we have Charles and Harry.

Depending on the intonation (low rise versus fall respectively), these lists may be interpreted as "possibly incomplete" or "complete."

3.2. Matrix clauses with cardinally quantified complements

In clefts such as There's only one thing that's that shape the matrix clause is an unmarked, not an enumerative (see 3.1), existential. In the mainstream, the unmarked existential has always been interpreted locatively: it is alleged to predicate an explicit or implicit location of the entity designated by the central NP. However, many decades ago Strawson (1959: 249) had already suggested that clauses with there is/there are should not be interpreted as predicting properties of entities, but as confirming or denying INSTANTIATION of the "nonparticulars" described by the central NP. I have given extensive linguistic argumentation elsewhere(11) for the same interpretative claim, while, however, also incorporating Milsark's cardinality restriction in it. As pointed out by Milsark (1976: 116f.), the focal NPs in unmarked existentials are subject to a "cardinality restriction": the quantifying elements found in the unmarked existential all express a form of cardinal quantification.

As we have seen, cardinal quantifiers, or absolute quantifiers as Langacker (1991: 82f.) calls them, designate the intrinsic magnitude of the designated mass, either by pegging it to a specific numerical scale, as in

(88) There were ten/nine/eight ... little Indians.

or by using a more "schematic" expression of cardinality, as in

(89) There were some little Indians.

In what way does the cardinal measure obligatorily expressed by the NP in these existentials contribute to its semantics? I have argued that it applies to the "instantiation" of the type -- Strawson's nonparticulars -- designated by the NP's "type specification." An existential such as There were ten little Indians cardinally measures the actual instantiation within that clause's spatiotemporal domain of the general category "little Indian." Or to put it very simply, it COUNTS the instances of "little Indians" to be found within the spatiotemporal coordinates set out by that clause. In this context, we can also return to the existential example (7) discussed at the beginning of this article and reproduced here as (90).

(90) There's only one thing that has a funny shape on the table.

Sentence (90) specifies that in the spatiotemporal domain "present time, on the table," only one instance of the category "thing that has a funny shape" can be found.

It is because of this central concern with the cardinal quantification of instances that I refer to the unmarked existential as the "cardinal existential."

4. A constructional approach to clefts

In this final section, the findings of sections 2 and 3 will be brought together and a general description of the semantics of clefts will be offered. The position assumed here is that so-called clefts can and should be analyzed as constructions in their own right. Their semantics are the result of the two basic relations coded by the construction. First, there is the relation expressed by the matrix clause, which, as we have seen, can be identifying, cardinal existential, enumerative existential, or possessive. The distinct semantics of the various matrix clauses are responsible for the differences in meaning between the various types of cleft, such as that between the identifying and the enumerating cleft discussed above (examples [75]-[76]). Second, there is the anaphoric relation between complement and relative clause, which is common to all clefts. As we have seen, the transformation-type approach has tended to confuse these two component structures, but in sections 2 and 3, I have attempted to put their precise grammatical outlines on the map more clearly.

What is striking about all the matrix clauses that occur in clefts is that they all construe "quantificational" information of some sort for their complement. As we have seen, we have a specificational complement in identifying clauses, an enumerating complement in enumerative existentials and possessives, and a cardinally quantified complement in cardinal existentials. I propose that it is these various quantificational meanings imposed on the complements that constitute the common denominator of the clause types that can function as the matrix of clefts.

What is more, these constructional quantificational features also bear on the complement NPs AS ANTECEDENTS. The antecedent in (80) is not just the coordinated NPs Jules and Jim, but the specificational complement, that is, the element of structure on which the identifying construction imposes exhaustive relative quantification. The meaning of the antecedent is "Jules and Jim as exhaustively specified set." Likewise, the antecedent in (82) is not just the coordinated NPs Jules and Jim. (As pointed out under 3, we cannot account for the difference in meaning between [80] and [82] if we bring only the grammatical class features of the complement NP, such as definite NP, into the picture.) The antecedent in (82) is the enumerating complement, which pegs instances to a numerical scale; its meaning is "Jules and Jim, as instances on an ordinal scale."

What, then, are the semantics of the antecedent-relative clause relation itself in clefts? Unlike the case with RRCs, they do NOT constitute a head-modifier relation, in which the "general type" is narrowed down to a more specific "type specification." Rather, as proposed by Declerck (1984, 1988) and Hannay (1985: 120), antecedent and relative clause in clefts construe a value-variable relation. The instances, as framed by the various quantificational mechanisms of the matrix clauses, constitute the specific VALUE. The entity involved in the situation designated by the relative clause constitutes the more general VARIABLE. Value and variable are used in a general sense here, which covers all the quantificational mechanisms found in the various cleft types.(12)

The constructional semantics of the various types of cleft can then be described as follows. It clefts have an identifying matrix, which equates universally quantifying it with the complement. This complement is thus an exhaustively specifying one. As such, it functions as antecedent of the relative clause. It specifies the total set of instances corresponding to the variable expressed by the relative clause. For instance, it cleft (91)

(91) The facts are: it's President Chirac who is going to carry the political can (CB).

specifies that the only politician who is going to be blamed is President Chirac.

In enumerative there clefts, the matrix is an enumerative existential with an "enumerative" complement. This complement lists instances, according to an implied ordinal scheme, as corresponding to the variable expressed by the relative clause. Often, we find the limiting case of enumeration, which is to list just one instance. Typically, the implication is that this list is potentially incomplete, as in (92), but intonation and added expressions such as only may also signal that the enumeration is complete, as in (93).

(92) I've really just got to fill them in on lexicographers' needs just because we've been doing a lot of it but there's other people that you think are doing kind of creative corpus lexicography. -- Well, there's McCarthy who's just building a new one (CB).

(93) Put something up? -- There's only the council can do that (LDC).

This sort of enumerative complement is also found in the more marginal "cleft" constructions with enumerative possessive matrix:

(94) I have my daughter who helps me (cf. Huddleston 1984: 399).

In (94), a general concept of possession is used to frame the list of values corresponding to the variable in the relative clause.

The matrix clause of cardinal there clefts is a cardinal existential. Hence, it is subject to the cardinality restriction: its complement always has absolute quantification. Therefore, it indicates the cardinal measure of the instance corresponding as value to the variable. As discussed above, the complement NP in clefts is typically focal. In cardinal there clefts, the information focus may be on the nominal head or on the cardinal quantifier of the value. In the former case, illustrated by one possible reading of (95), the most salient information in the cleft is the TYPE of thing or person ("boy") instantiating the variable. In the latter case, for instance (96), it is the CARDINALITY ("one"), with which the variable is attested.

(95) There's no boy in that school has more or better clothes than you (LDC).

(96) Look at the shape of it. There's only one thing that's that shape (LDC).

We should note here that possessive clefts with indefinite complement also impose cardinal quantification on the value being related to the variable, as in

(97) I have a woman comes in twice a week (Hannay 1985: 91).

I submit that these semantic characterizations do justice to the two main coded relations in clefts: the one expressed by the matrix clause and the one expressed by the anaphoric relation between antecedent and relative clause. Earlier descriptions have tended to reduce one to the other. For instance, in Halliday's (1967a) and Bolinger's (1972) analysis of it clefts, subject it and the relative clause are said to form one discontinuous constituent, which is analyzed as the identified. In this way, the it cleft is reduced to a simple identifying clause.(13)

The distinct semantics of the matrix clauses explain the semantic differences between the various types of cleft. In particular they show how the complement is set up as antecedent of the relative clause with different quantificational values. As we have seen, the distinct matrix clauses may express exhaustive specification of a set, enumeration, and cardinal quantification of instances. What the relative clause does is ADD a valuevariable dimension as part of the construction. The nature of the anaphoric relation created by the relative pronoun is the same in all types of cleft: it relates the antecedent as a value to the variable expressed by the relative clause. Thus, it becomes clear that clefts do not simply express specification, enumeration, or cardinal measuring of the instantiation -- they SPECIFY, ENUMERATE, OR MEASURE INSTANCES AS VALUES CORRESPONDING TO A VARIABLE.

As continually stressed throughout this article, these two relations expressed by matrix clause and relative anaphora are part of the constructional semantics. They form the grammatical substratum for the many "textual," or "pragmatic," patterns that have received so much attention in more recent studies of clefts (Geluyckens 1988; Collins 1991, 1992). As we saw in section 1.1, the relative clause in clefts may recap "given" information and encapsulate this in a "contextually given" variable (as in example [5]), or it may present "new" information in the variable (as in example [10]). Either informational schema is compatible with the grammaticosemantic description proposed here.

Conclusion

In this article, I have developed a grammatical analysis of clefts as constructions in their own right whose semantic properties can be correlated directly with their grammatical features.

In section 1, I have first recapitulated the incontrovertible formal recognition criteria of the cleft established in the literature, such as its intonation and the fact that its relative clause does not modify the nominal head. I have then identified the two main aspects of clefts that had not yet been dealt with satisfactorily, viz. the constructional value of the matrix clause and the nature of the anaphoric relation established by the relative pronoun.

In section 2, I have investigated the special status of the antecedent of the relative clause in the cleft. Unlike restrictive relative clauses, the antecedents in clefts include identifiers and quantifiers, and unlike nonrestrictive relative clauses, its "non-NP" antecedents are rankshifted and reclassified as NPs.

In section 3, I have set out descriptions of the identifying, existential, and possessive types of matrix clause found in clefts, with special reference to my "quantificational" interpretation of cardinal and enumerative existentials. I have pointed out that the complements of clefts, which already have "marked" properties as antecedent of the relative clause, are further foregrounded in clefts by the distinct quantificational values the matrix clauses impose on them, viz. as specificational, enumerative, or cardinally quantified, respectively.

In section 4, finally, I have shown that the two relations construed by clefts -- by the matrix clause and by relative anaphora -- can be integrated with each other, without reducing them to each other, as happened in earlier descriptions. Put very simply, clefts specify, enumerate, or measure instances as values corresponding to a variable.

In view of all this, I have to add as a postscript that the term "cleft" can be retained only as a conventional mnemonic label and should certainly be cleared of any implications that the construction is intrinsically "derived" and cannot be described directly.

Received 15 March 2000 Revised version received 26 September 2000

University of Leuven

Notes

(*) I wish to thank Karel Van den Eynde for his careful reading of a previous draft of this article. I owe a special debt of gratitude to him for his sharing of ideas, his sharp descriptive and methodological observations, and his general support. I also thank the anonymous referee of Linguistics for his generous comments. All the examples marked with "(LDC)" were extracted from the Leuven Drama Corpus and those marked with "(CB)" from the COBUILD corpus. The latter are reproduced here with the kind permission of Harper Collins. Correspondence address: University of Leuven, Department of Linguistics, Blijde Inkomststraat 21, B-3000 Leuven, Belgium. E-mail: [e-mail].

(1.) This article takes Huddleston's "transformational" approach to clefts as a starting point because his is a very clear articulation of this position. Moreover, it contains many interesting observations, as well as unresolved questions, which, as acknowledged in several places in this article, have served as a springboard for the alternative approach developed here.

(2.) It is precisely in its emphasis on the natural coding relation between form and meaning that this form of constructionism differs from that propagated by Goldberg (1995).

(3.) For a more detailed discussion of the semantics of existentials, see section 3.2 and Davidse (1999a).

(4.) This is often referred to, in the mainstream, as a "pragmatic" motivation, viz. one enabling specific discourse uses.

(5.) In fairness, it should be pointed out that there are also existing descriptions that unambiguously identify the nominal head as the antecedent of RRCs, such as McGregor (1997: 199-201).

(6.) Note that Langacker's (1991: chapter 1) concept of the "type specification" includes the semantic contribution made by the grammatical class of the head noun: a singular count noun such as bullet designates a "type of discrete entity," a mass noun such as buckshot designates a "type of homogeneous mass," while a plural count noun such as bullets designates a "type of heterogeneous mass."

(7.) Thus, in The, Dictionary of Phonetics and Linguistics, we find the antecedent defined as the "linguistic UNIT to which another unit in the SENTENCE refers (ANAPHORIC REFERENCE), typically a later unit" (Crystal 1991: 17).

(8.) The concept of the "reference mass" has been developed by Langacker (1991: 82f.), who introduces this concept as part of his definition of relative quantifiers (see below). However, unlike Milsark (1977), Langacker does not extend the notion of relative quanification to define determiners such as the definite article, and possessive and demonstrative determiners.

(9.) The constructional characteristics of the complements in the various types of cleft will be discussed in the next section.

(10.) Verbal complements in clefts contain the predicate, not the finite, component of the VP, which is quoted in either the infinitival or the -ing form. To integrate the antecedent with the relative clause, the substitute do has to be replaced by the specific lexical verb:

(iii) It's run/running that she did. :: She ran.

(iv) It's running that she was doing. :: She was running.

(11.) For extensive discussion of the semantics of cardinal and enumerative clauses, see Davidse (1999a).

(12.) Declerck (1988) in fact defines value and variable more narrowly as the exhaustive specification of the set of values corresponding to the variable expressed by the th/wh clause. In his approach, the concepts of values and variable are restricted to it clefts.

(13.) The analysis of, for instance, it ... who am to blame as one discontinuous constituent of It is I who am to blame also involves grammatical problems. As pointed out by Huddleston (1984:461), the first person marking of the finite verb cannot be reconciled with third person it.

References

Bolinger, Dwight (1972). A look at equations and cleft sentences. In Studies for Einar Haugen, Presented by Friends and Colleagues, Evelyn Scherabon Firchow et al. (eds.), 96-114. The Hague and Paris: Mouton.

--(1977). Meaning and Form. London and New York: Longman.

Butler, R. (1993). Rodin. The Shape of Genius. New Haven and London: Yale University Press.

Collins, Peter (1991). Cleft and Pseudo-Cleft Constructions in English. London: Routledge.

--(1992). Cleft existentials in English. Language Sciences 14, 419-434.

Crystal, David (1991). A Dictionary of Phonetics and Linguistics, 3rd ed. London: Longman.

Davidse, Kristin (1999a). The semantics of cardinal versus enumerative existential constructions. Cognitive Linguistics 10(3), 203-250.

--(1999b). Are there sentences that can be analyzed as there-clefts? In Thinking English Grammar. To Honour Xavier Dekeyser, Professor Emeritus, Guy Tops, Betty Devriendt, and Steven Geukens (eds.), 179-195. Leuven: Peeters.

Declerck, Renaat (1984). The pragmatics of it-clefts and WH-clefts. Lingua 64, 433-471.

--(1988). Studies on Copular Sentences, Clefts and Pseudo-Clefts. Leuven: Leuven University Press and Foris.

Geluyckens, Ronald (1988). Five types of clefting in English discourse. Linguistics 26, 823-841.

Goldberg, Adele (1995). Constructions. A Construction Grammar Approach to Argument Structure. Chicago: University of Chicago Press.

Gundel, Jeannette; Hedberg, Nancy; and Zacharski, Ron (1993). Cognitive status and the form of referring expressions in discourse. Language 69, 274-307.

Haas, William (1954). On defining linguistic units. Transactions of the Philological Society 1954, 54-84.

Halliday, Michael (1967a). Notes on transitivity and theme in English 2. Journal of Linguistics 3(2), 199-204.

--(1967b). Intonation and Grammar in British English. The Hague: Mouton.

--(1981 [1965]). Types of structure. In Readings in Systemic Linguistics, Michael Halliday and James Martin (eds.), 29-41. London: Batsford.

--(1985). Introduction to Functional Grammar. London: Longman.

--; and Hasan, Ruqaiya (1976). Cohesion in English. London: Longman.

Hannay, Michael (1985). English Existentials in Functional Grammar. Dordrecht: Foris.

Higgins, F. R. (1976). The Pseudo-Cleft Construction in English. Bloomington: Indiana University Linguistics Club.

Hopper, Paul; and Closs-Traugott, Elizabeth (1993). Grammaticalization. Cambridge: Cambridge University Press.

Huddleston, Rodney (1984). Introduction to the Grammar of English. Cambridge: Cambridge University Press.

Langacker, Ronald (1987). Foundations of Cognitive Grammar, vol. 1: Theoretical Foundations. Stanford: Stanford University Press.

--(1991). Foundations of Cognitive Grammar, vol. 2: Descriptive Applications. Stanford: Stanford University Press.

Lumden, Michael (1988). Existential Sentences. Their Structure and Meaning. London: Croom Helm.

Matthiessen, Christian; and Thompson, Sandra (1989). The structure of discourse and "subordination." In Clause Combining in Grammar and Discourse, John Haiman and Sandra Thompson (eds.), 275-330. Amsterdam and Philadelphia: Benjamins.

McGregor, William (1997). Semiotic Grammar. Oxford University Press.

Milsark, Gary (1976). Existential Sentences in English. Bloomington: Indiana University Linguistics Club.

--(1977). Toward an explanation of certain peculiarities of the existential construction in English. Linguistic Analysis 3, 1-29.

Quirk, Randolph; Greenbaum, Sidney; Leech, Geoffrey; and Svartvik, Jan (1972). A Grammar of Contemporary English. London: Longman.

Rando, Emily; and Napoli, Donna (1978). Definites in there-sentences. Language 54, 300-313.

Strawson, Peter (1959). Individuals: An Essay in Descriptive Metaphysics. London: Routledge.