Anaphora: lexico-textual structure, or means for utterance integration within a discourse? A critique of the functional-grammar account. (1).

Cornish, Francis

Abstract

This article is a critical examination of Dik's (1997b: chapter 10) account of discourse anaphora, within the framework of the theory of functional grammar (but it highlights features of anaphora theory that hold more generally). I show first that Dik's definitions of the phenomenon involve two contradictory conceptions of this discourse procedure (the anaphor refers to a mental representation of its referent within a mental model of the ongoing discourse yet at the same time needs first to connect up with a segment of cotext--its linguistic antecedent); second, that Dik's account of the relationship between given (pronominal) anaphor types and the entity order of their potential referents is both too rigid and too narrow; and third, that his description of the underlying structure of anaphors, which includes both the referential index of their actual referent/antecedent and a variable specifying the latter's entity order, does not allow for the necessary flexibility and dynamic character of anaphor use and interpretation. A discursively more realistic account of discourse anaphora needs to specify the necessary interaction between "bottom-up" factors of these kinds, on the one hand, and "top-down" relationships involving the wider discourse context, on the other. This is what I briefly outline at the end of the article.

1. Introduction

In the most recent two-volume presentation of the standard theory of functional grammar, Dik (1997b) devotes a whole chapter to anaphora (both sentence and discourse anaphora) (chapter 10), and a further chapter to an outline of an approach to discourse (chapter 18). Anaphora figures prominently throughout the current development of functional grammar (henceforth FG) as a whole--for example, in its account of relative clauses, and in its definitions of the variables marking each of the different layers within underlying clause structure: indeed, the possibility of distinctive anaphoric reference to each of these layers provides important evidence in favor of their independence. The opening up of the originally clause-bound model to discourse that is now being undertaken is developing in parallel with a concern for anaphora. (2) This is only natural, since anaphora and deixis at the discourse level are an important means of coordinating the attention of the discourse participants, as well as of converting text into discourse--and vice versa from the speaker's point of view, of course.

This is why I think now is a good time to examine the current, still programmatic, FG approach to anaphora, as set forth in Dik (1997b: chapter 10), and to see to what extent it can be situated within what is now known about discourse and anaphora generally. I will discuss three basic issues raised by Dik's account, which may be seen to be relevant generally, whatever the approach to anaphora and reference in discourse.

These are as follows:

1. the very fundamental question of the nature of the anaphoric relation, in other words, whether it is necessarily mediated by the prior mention (that is, use) of an antecedent expression in the cotext or not;

2. the relation between the form of an anaphor and the potential entity order(s) of its referent; and

3. the formalism used by Dik to represent anaphors in underlying clause structure, and the (necessarily) dynamic relationship between anaphors, their host predication, and the utterance containing a relevant antecedent trigger.

We shall see that these three issues are closely linked. In what follows, I devote a section--respectively, sections 2, 3, and 4--to each of these basic questions. Finally, section 5 is an attempt to spell out the steps involved in resolving an anaphor, integrating the language-system-based aspects of the procedure and the discourse-contextual ones.

2. The current FG view of anaphora

At the outset, it is necessary to confront a central issue regarding the relationship between the speaker's and the addressee's perspectives, concerning the establishment of both reference and anaphora. Now, as pointed out by both Co Vet at the conference where an earlier version of this paper was presented (see note 1 for details) and an anonymous Linguistics reader on the basis of the version submitted to this journal, the speaker is in control of his or her reference, hence the marking of coreference via identical indices on anaphoric expressions and their antecedents in underlying clause structures is justified--since the speaker can be expected to know in advance even of a given subsequent reference, that a subsequent anaphoric expression is coreferential or not with his/her earlier antecedent reference. Both individuals invoke as justification for this view the fact that Dik (1997a) conceives of the derivation of well-formed expressions in a given object language in terms of the PRODUCTION of such expressions from the speaker's point of view.

However, a careful examination of the source text shows that Dik hedges slightly in making such a statement (cf. 1997a: 14, where he observes: "... FG as presented in this work MORE, CLOSELY APPROXIMATES a production model than an interpretation model" (emphasis added). And further on (1997a: 57), he qualifies this statement: "Although the model of FG is presented in a quasi-productive mode, I do not wish to suggest that this order of presentation necessarily simulates the various steps that a speaker takes in producing linguistic expressions." So the mode of presentation of an FG clause derivation, although quasi-productive, is not intended to model the actual process of producing given expressions in context--that is, to correspond to a performance model, in generative-grammar terminology. Hence in my view, it would not be appropriate to indicate in two given underlying clause structures the intended coreference relation between two referring expressions, this being a function of the speaker's USE in context of the expressions involved, and not of the context-independent lexicostructural properties of these expressions. In any case, regarding the establishment of reference, Dik (1997a: 127) specifically makes the point that referring is a joint, collaborative affair, involving the active cooperation of speaker and addressee alike (see also in this regard his model of verbal interaction: 1997a: 8, Figure 1). Under this conception, reference cannot be said to have succeeded in a given utterance context unless and until the addressee shows by his/her subsequent behavior (linguistic or otherwise) that s/he has correctly picked out the referent intended by the speaker. In the absence of such mutual agreement, the ensuing communication will be seriously interrupted. I will argue in the present article that the FG clause-derivation system should not be viewed exclusively in terms of clause PRODUCTION, but rather in a more user-independent way, so as to be compatible with both the speaker's and the addressee's perspective. In practice, in any event, it is not the case that all aspects of the grammar are oriented in terms of the speaker's production: Dik's account of selection restrictions (1997a: section 4.4), for example, clearly takes the addressee's perspective in specifically allowing for the possibility of creative, metaphorical readings (by the addressee, then, clearly) whenever these restrictions are violated. This language-system/language-use distinction will be of particular relevance in the discussion of coindexing and anaphor type in section 4 below.

Dik (1997b: 215, chapter 10) rightly distinguishes between "the underlying anaphorical [sic] relation" and "its formal expression." He then attempts to define well-formedness conditions for anaphora in functional terms. Dik gives what I would argue are two different definitions of anaphora, cited below as definitions I and II, respectively.

Anaphora: Dik's (1997b: 215) definition I

I speak of anaphora as occurring when an element of underlying clause structure refers to an entity which has already been established, directly or indirectly, in the preceding discourse (discourse anaphora) or is being established in the same clause (sentence anaphora).

Anaphora: Dik's (1997b: 216) definition II

... The expression with which the entity in question [i.e. the referent] has been or is being established in the discourse is the antecedent of the anaphorical [sic] element. (...) The relation between anaphoric element and antecedent will be called an anaphorical relation.

I view these definitions as involving two distinct conceptions of the anaphoric relation involved here: in the first (definition I), the anaphor refers to an ENTITY--or more accurately, to the mental representation of an entity as evoked via the preceding discourse--that is already established in the addressee's discourse model: this is what would most commonly be called nowadays a discourse referent; but in the second conception (definition II), the relata involved are the antecedent (a co-occurring linguistic expression) and the anaphor.

In Dik's principle (ii) (1997b: 215), he states that, "All anaphors have an antecedent in the discourse. The antecedent itself is not used anaphorically, but it serves to establish some entity in the discourse."

Before we proceed any further, I think we need a definition of the notions of text and discourse. This is given under (1) below.

(1) Text vs. discourse

 a. Text denotes the connected sequence of verbal signs and
 nonverbal signals in terms of which discourse is coconstructed
 by the participants in the act of communication.

 b. Discourse denotes the hierarchically structured, situated
 sequence of utterance, (3) indexical, and illocutionary acts
 carried out in pursuance of some communicative goal, as
 integrated within a given context.

 c. The context is subject to an ongoing process of construction
 and revision as the discourse unfolds. (These are my
 definitions; see Cornish [1999: subsection 2.3] for further
 development and illustration of this distinction, and its
 importance for the study of discourse anaphora.)

The text, then, is the tangible, perceptible record of at least one utterance act, whether realized in terms of a verbal, linguistic trace or of a nonverbal trace--which may be gestural, sensory-perceptual, or prosodic. The discourse partners make use of this record, in conjunction with their invocation of a relevant context, in order to create discourse. The distinction is close to that drawn recently within an FG context by van den Berg (1998) between the concepts of utterance and message (structure), and in a text worlds context by Werth (1999: 46, 302). (4) So when Dik says that "all anaphors have an antecedent in the discourse," he no doubt means text. His definition I, on the other hand, is framed within the context of discourse, according to my definition (1).

Under Dik's conception of anaphora, it is necessary for there to be a previous mention in the surrounding cotext of the entity referred to by an anaphor, and it is this mention that constitutes the antecedent--that is, a linguistic expression used in the cotext of some anaphor.

But prior mention is in reality neither a necessary nor a sufficient condition for the existence of anaphora--as indeed Dik himself suggests in formulating his first conception of anaphora (definition I above), when he writes that "an element of underlying clause structure refers to an entity which has already been established, directly or INDIRECTLY, in the previous discourse" (emphasis added). If we take discourse seriously, as Dik claims to do--particularly in his chapter 18 devoted to this topic--then it is clear that naturally occurring discourse (5) is full of instances where there exists a mutually assumed discourse referent that is retrieved by a given anaphor, without that referent having been explicitly introduced into the discourse by a prior mention.

Let us look at the examples under (2) below. In order to be felicitous (i.e. coherent), these instances of implicit or indirect anaphora all require that some kind of connection be available for the addressee to be able to retrieve the correct, intended referent.

(2) a. [Context: a neighbor's father has been in hospital for a week
 already]
 Anne to her neighbor, seeing her looking haggard:
 How is he?

(2a) trades on the existence within their episodic memories of certain specific prior knowledge shared by the interactants. We may assume that the subject of the neighbor's father's health has been intensely discussed over the past days and so is constantly at the forefront of the neighbors' consciousness. The trigger of that discourse domain is Anne's seeing her neighbor looking haggard and immediately realizing the reason for it.

(2) b. The high street bank on the corner has been broken into twice
 this month. But they only took the small change
 [they = `the bank-robbers']

In (2b), it is the stereotypical knowledge frame evoked via the initial sentence (a bank break-in), making available a slot for the perpetrators of the act, that motivates the indirect anaphoric reference: the default reference realized via the "indefinite"-type pronoun they. The speaker cannot assume that the addressee has any particular individuals in mind at the point of use: but their nonspecific existence is nonetheless assumed.

(2) c. [Context: Mary and Barbara are discussing Sarah, who Mary
 knows well but who Barbara has only met once]
 Mary: She's always so good-humoured, you know ... but HE's
 a bore ...
 [HE = `Sarah's husband/partner']

In (2c), there is obviously an abductive, or backward-looking inference needed to instantiate (i.e. accommodate) a referent not hitherto presupposed to exist. The reference of HE here is deictic (or anaphorico-deictic), not strictly anaphoric: a contrastive pitch accent on this pronoun is required in order for the reference to have any chance of succeeding. Accented pronouns are deictic, in that there is an extra "pointing" element to them, namely the presence of the pitch accent, which also has an effect on the phonetic value of the vowel within the pronoun. Use of the unaccented variant, where the vowel would be a lax, close front vowel phonetically, would be totally incapable of picking out the referent intended here, which is only available in long-term memory through the existence of the "frame" corresponding to the notion of "a couple": it is via this stereotypical frame knowledge, triggered by the use of contrastive HE, that a representation of Sarah's partner is introduced into the addressee's discourse model (presumably, such a representation would already be present within the speaker's). In addition, the focus structure applied to the predicative content of the two clauses clearly indicates a contrast between the personalities of the two individuals concerned, both pronouns occurring in parallel subject position in their respective clauses. Under the conception assumed here, anaphora serves to maintain the preexisting attention focus on one or more referents, as well as the situation in which these referents are involved: while deixis is a means of changing the attention focus, by exploiting various key features of the deictic context (speaker, addressee, place and time of utterance).

(2) d. [Context: a young goat wanders through the open front door]
 A to B, observing the event in fascination:
 What do you think it's looking for?

In (2d), the key factor in the exophoric use of the pronoun is the copresence of the referent in the situational context, and the fact that the discourse partners are focusing on this unexpected event, which is immediately attracting their attention.

(2) e. [Context: Woman returning from country walk where she had
 intended to pick blackberries. Local retired man sitting on log
 observing her walk back with empty box]
 Man: So you didn't find any [PHI], then?

Finally, (2e) trades upon a forward-looking inference based on the man's observation of the scene, and on his knowledge of the common plans of action as well as motivations that people in society typically have.

Apart from (2c) where the reference is deictic (or anaphorico-deictic), all these instances are cases of anaphora, where the anaphor is intended to pick up a representation of a referent assumed by the speaker to be accessible and salient to the addressee at the point in the discourse where that anaphor occurs. But in no case is there a prior explicit MENTION of the intended referent. Note that in all types of anaphora, there is a semantic/pragmatic connection of one kind or other between what I call the antecedent trigger (6) and a relevant salient discourse representation. In discourse anaphora, the antecedent trigger is an utterance token, a gesture or percept that evokes or boosts a given mental representation of a state of affairs (SoA) into speaker's and addressee's current discourse models.

One other very important factor in the operation of anaphora is the nature of the anaphoric predication: (7) that is, what is predicated of the referent of the anaphor--which may still be unascertainable at the point when it is uttered--acts as a pointer toward a referent of a certain type; in other words, it places a semantico-pragmatic constraint on its potential values. A pair of examples presented in Wilson (1992) makes the point very clearly:

(3) a. Sean Penn attacked a photographer. The man was quite
 badly hurt.
 b. Sean Penn attacked a photographer. The man must be
 deranged.

Here, each anaphoric predication most naturally continues the perspective involving a different discourse referent mutually available to the speech partners at the point where the definite subject NP of the second sentence occurs--a term that is semantically appropriate for retrieving either of these referents. Note, though, that if we replace these two occurrences of the man by the pronoun he, the first but not the second becomes slightly unnatural. This is due to the preference for subjects, but not nonsubjects, to be construed as topics, as well as to the restriction of unaccented third person pronouns to referents that are highly discourse-active. However, (4) below shows that the anaphoric predication does not always perform an orienting function with regard to the choice of a suitable referent for the anaphor: (8)

(4) [Mary.sub.i] told [Jane.sub.j] that [she.sub.i/j] would have to see a doctor.

In all these examples, the anaphoric predication continues one aspect of the mental scene evoked via the context. So it seems, on the basis of an observation of real discourse in specific contexts of utterance, as opposed to careful, invented one-or two-sentence examples where there is an explicit prior mention of the intended referent, that prior mention is but one way in which a given referent may enter the addressee's discourse model, justifying its later reaccessing via an appropriate anaphor. Prior mention is of course normatively required in careful written genres of discourse, and linguists studying anaphora have in the past tended to base their analyses on invented minitexts that have a definitely "written" feel about them. But are we justified in extrapolating from normative, written prose to the entire range of discourse in characterizing the anaphoric relation? In my view, we are not, and the observation of a wide range of discourse genres (9) seems to suggest a rather different model of anaphora from the traditional one whereby an anaphor--typically a pronoun of some kind--has first to be paired with a cooccurring antecedent expression to enable its full sense and reference values to be instantiated.

There is, then, a tension, it seems to me, between the two conceptions of anaphora (reference via an antecedent expression available in the cotext, and reference in terms of a discourse representation of an entity of some kind) that Dik simultaneously entertains. And Dik states (1997b: 220) that "the form of the anaphoric element varies according as the ANTECEDENT ENTITY is a first-order entity, a second-order SoA, or a propositional content" (emphasis added). Here, it seems, the two conceptions of anaphora distinguished earlier are conflated into a single weave, with the concept of antecedent entity. Dik is guilty of a similar confusion as well as of contravening his own principle (i) (1997b: 217), namely that "the anaphor does not refer to the antecedent," in saying (1997b: 228), "In all these cases, the anaphorical element refers to an antecedent which is a full speech act." Anaphors can of course refer to speech acts, but a speech act is not an antecedent (i.e. an expression, in Dik's sense). Dik's traditional conception of the antecedent further loses its force when he notes the possibility of an anaphor referring to a subtopic inferrable from the entity introduced by the antecedent, as in the examples of associative anaphora he presents (1997b: chapter 10, [2a] and [2b]):

(5) a. On a bench in a park he saw an elderly couple. The man ...
 The woman ...

 b. John bought a book, but after reading the first few pages he
 threw it away.

Here, the notion of antecedent trigger would seem more appropriate to characterize the role of the NPs an elderly couple and the book: that is, as introducing into a discourse a number of entities that may later be retrieved via an appropriate anaphor.

The fact that each separate subtype of anaphor bears a distinctive bundle of semantic and referential properties means that they are not simply the passive recipient of a sense and reference assigned derivatively by a lexically fuller cooccurring expression (the antecedent); this is specifically highlighted by Dik in his discussion of the various types of anaphoric expression and their discourse-pragmatic roles (cf. 1997b: subsection 10.3.1, "Pragmatic factors"). Instead, they are actively responsible for the accessing of a referent or a denotation, the result of which has certain consequences both for the current interpretation of the wider discourse segment in which antecedent trigger and anaphoric predication cooccur, and for the subsequent direction of the discourse to follow.

The notion of antecedent expression is also perhaps motivated by the convenience, in evidence since the advent of TG, of annotating linguistic segments in an example for same or different indices, thereby marking coreference vs. noncoreference. But as pointed out by Lambrecht (1994: 47-49) in connection with the similar annotating of linguistic segments in terms of their topic or focus functions, this is not always either feasible or perspicuous.

3. Anaphor type and entity order

Dik's major concern within the body of the chapter, though, is with the entity-order types of referent that an anaphor may have--thus with the first conception of anaphora mentioned earlier. Indeed, he is concerned here not with the LINGUISTIC identity of segments--antecedent segment in the cotext and the segment corresponding to the anaphor--but with the various forms of the anaphors concerned in relation to the ontological (and discourse) status of the referents of those anaphors.

Two parameters determine the use of a particular anaphor type in a discourse. They are (a) the functioning of the anaphor with respect to the DISCOURSE STATUS OF ITS INTENDED REFERENT, assumed already to be present in that discourse--the formation of different chains of anaphors (definite or demonstrative NPs, pronouns of different types: definite ordinary pronouns, demonstrative pronouns [this vs. that], zeros, etc.) according as the status of the referent is foregrounded or backgrounded at the stage that the discourse has reached when the anaphor is used. (10) And (b) THE ONTOLOGICAL NATURE, IN TERMS OF ENTITY ORDER, OF THE REFERENT assumed by the use of particular types of anaphors--only pronouns are discussed in this context.

But in this latter respect, is there really a one-to-one relationship between type of pronoun and type of entity order of its assumed referent, as Dik implies there is? In his brief survey of entity-order types (1997b: section 10.3.2), Dik provides only one example of an anaphor type correlating with each entity-order type that he distinguishes. That is, one = `property ([f.sub.i])', (11) him/her/it = `spatial entity ([x.sub.i])', the X-ing (where X is a verb), or, presumably, it/that, = `SoA ([e.sub.i])', so = `possible fact ([X.sub.i])', and this/that or this/that N, where N is an illocutionary noun = `illocution (or perlocution) ([E.sub.i])'. Presumably, the pronoun there would be given Mackenzie's (1992)p variable (for "place"), and then would be given t, for "time." Dik (1997b: 220) is a little too rigid, to my mind, regarding the relationship (which he appears to see as fixed) between given anaphor types and given entity-order denotations: (12) for while the personal pronouns he/she/they and their variant forms do indeed have first-order referents, the pronoun it may refer to a range of entity types, in addition to the second-order SoA illustrated in his example (12b) (Dik's annotations):

(6) John saw [Bill win][sub.i] and Peter saw [it.sub.i]/[that.sub.i] too.

It may have a first-order referent (either inanimate as in `the table', or animate: animals, insects, etc., plants ... or even human: babies). But it can also have third-order referents (propositional contents), as in (7) (my example):

(7) Jack said he'd just won 30,000 [pounds sterling] on the lottery, though Mary didn't believe it.

In some cases, it may have a [PHI]-order "property" referent. This depends largely on the type of host verb that acts as predicator in the anaphoric clause (in [8], it is verbs of appearance that are illustrated):

(8) Mary said she felt weak and ill. She certainly {looked/sounded} it.

This and other examples shows that the type of entity order denoted by a given anaphor is only a preference: it can be overridden by various properties of the anaphoric predication as a whole, among other factors, particularly by those of the host predicator.

The pronoun so can also have as referent other entity-order types than the propositional-content one claimed to be the case by Dik, which he illustrated in his example (12c):

(9) John thought that [[Bill would win].sub.i] and Peter thought [so.sub.i] too.

It may denote a manner, as in the proclitic occurrences so-arranged,, so-called, etc. See also the indefinite determiner such (cf. Mackenzie [1997]'s FG account of this indefinite anaphor in written English, an account that is fully compatible with the view of anaphora argued for in the present article). I would say that so retrieves intensional "type" referents (nonpresupposed propositional contents), whereas it prototypically has as potential referents ones that have extensionality, or are propositions that have been mutually validated by the discourse partners: they can be treated as (actual) facts for the purposes of the ensuing discourse, not as possible facts, which is, I believe, why Dik considers so to have this kind of reference. So is indefinite and intensional (and not fully nominal, rather it is adverbial), whereas it is both definite and extensional, as well as being fully nominal in character: see Cornish (1992) for further evidence in favor of these analyses.

The distal demonstrative pronoun that can refer to four of the five types of entity order isolated by Dik: entity order 1 (inanimate entities), as in (10B):

(10) A I'm going to cook cous-cous tonight.
 B: Really? I've never liked that.

Entity order 2 (SoAs):

(11) [Child puts hand near flames coming from coal fire]
 Parent: Johnny, don't DO that!

Entity order 3 (propositional content):

(12) A: Did you see in the paper that the prime minister has called a
 General Election for next month?

 B: What? I didn't know that/That's incredible!

Entity order 4 (illocutions [and potentially also, perlocutions]):

(13) Mother-in-law to daughter-in-law: We're coming to visit you next
 Sunday.

 Daughter-in-law (jokingly): Is that a threat or a promise?!

Distal that, unlike its proximal counterpart, has a preference for use in negative, distancing contexts, or where the referent is conceived as a hypothetical, not an actual, object or situation. Owing to its demonstrative character, it also has a definite focalizing effect in relation to it, its unaccented counterpart. Its proximal counterpart this, which has exactly the same range of denotation types as that, can be used to encapsulate a whole segment of discourse and to treat it as a compacted entity that is to be the topic of a new discourse segment.

It is, in the last analysis, the properties of the anaphoric predication that determine the actual entity order of the pronoun's referent. First and foremost among these is the nature of the host predicator. That is, verb of physical contact/movement, etc. (e.g. hit) + it [arrow right] `[x.sub.i]'; verb of cognition or propositional attitude (e.g. believe)+ it [arrow right] `[X.sub.i]'; activity verb (e.g. do) + it [arrow right] `[e.sub.i]'; appearance verb (e.g. look) + it [arrow right] `[f.sub.i]'; it + illocutionary predicator (e.g. be a lie/threat) [arrow right] `[E.sub.i]'. (13)

4. The formalism used by Dik to represent anaphors, and the three-way relationship between anaphor type, anaphoric predication, and discourse context

The formalism that Dik adopts in representing the underlying structure of anaphors is problematic. Anaphors are indexicals: language-particular form types, each of which possesses a specific set of semantico-referential properties. Where pronominal in character, they encode values in respect of the categories of person, number, gender, case, and definiteness as well as features like "already categorized" vs. "not categorized/ uncategorizable"--a value encoded by the neuter pronouns in particular, in languages such as French, German, Spanish, and those from the Slavic family. An example of the latter distinction would be the existence in French of a gender-and number-variable demonstrative pronoun (celui/celle-ci `this one (masc.sg.)'/`this one (fem. sg.)' vs. that of a neuter, invariant demonstrative pronoun (ceci/cela/ca `this'/`that'), as well as the opposition between the gender- + number-variable ordinary object pronouns le/la/les `he'/`it';/`she'/`it';/`them' and the neuter "oblique" pronouns y `there'/`to/at it/them' and en `of/from it/them'. A nice contrast involving the "discrete" pronoun la and the "nondiscrete" neuter demonstrative pronoun ca is cited by Manoliu-Manea (1990: 92; example [7], from Les fleurs bleues by R. Queneau):

(14) -- Ou'est-ce qu'on fait quand y a de la neige sur la peniche?
 `What are you supposed to do when there is snow on the
 canal-boat (barge)?'
 -- On la pousse dans l'eau et ca fait floc.
 `You push it [la: fem.sg] into the water and it [ca: neut.]
 goes "plop"'

Here, there is subtle shift in the conception of the snow referred to by the use initially of the ordinary gender- + number-variable pronoun la, which assumes that its referent has been categorized (via the invocation in this context of the basic-level category noun neige, whose feminine gender corresponds to the gender value that this pronoun expresses); and subsequently by that of the neuter demonstrative pronoun ca. The "discrete" pronoun la has as its referent the first-order entity `the particular quantity of snow that was on the barge at the time of utterance', while the neuter demonstrative ca highlights the dynamic property ascribed to that referent by the immediately preceding predication as a whole (that is, `being pushed into the water'). Its discourse referent is thus somewhat distinct, corresponding as it does to the second-order entity `the snow that was previously on the barge making contact with the water'.

These properties, then, derive from the language system and so may legitimately be claimed to be included in underlying clause representations (where a given anaphor has been selected from the lexicon together with its term frame structure).

But Dik also represents in the structures underlying anaphors (a) the variable symbolizing the entity order of their actual referent (an entity order that may be interpreted in fact as indicating the TYPE of entity denoted by a token of the anaphor form in question) and (b) the index of that referent as encoded in a cooccurring underlying clause structure (presumably where there is no syntactic, semantic, or pragmatic restriction inhibiting this relationship). (15) provides a sample of Dik's representations of anaphors.

(15) Dik's (1997b: chapter 10) underlying representations of a sample
 of anaphors
 one: ([Af.sub.i])
 it: ([Ax.sub.i]) or ([Ae.sub.i])
 the X-ing: ([Ae.sub.i])
 so: (A[X.sub.i])
 that/that N (illocution): (A[E.sub.i])

An anonymous Linguistics reader points out that these are not lexical representations but are intended to represent the USE of anaphors in actual utterances. However, there is surely an equivocation here, since pronouns are viewed within FG as being basic and not derived terms, thus already available within the lexicon. Thus conceived, the representations given in (15) necessarily involve a mixing of two kinds of information placed on the same level within these representations: on the one hand, the A operator and the entity-order variable (specifying the lexically defined value of each anaphor type), and on the other, the referential index, marking the intended reference of a token of the form type in question, as established within some specific utterance context.

(16) presents a sample representation of an underlying two-clause minitext, where an anaphor (here so) is resolved (Dik 1997b: chapter 10, [33]):

(16) Representation of example (9) (simplified)
 think (John) ([X.sub.i]: [Bill would win]) and
 think (Peter) (A[X.sub.i]) too

In fact, with the exception of pronoun types such as one and the third-person personal pronouns, which unambiguously encode a particular entity-order type of denotation, the potential entity-order features of a given anaphor are not system-determined--unlike the grammatical features mentioned a moment ago; instead, they are DISCOURSE-determined. In such instances, there must therefore be a meshing between the output of the rules governing the construction of underlying clause structures, on the one hand, and the effect of the principles, tendencies, etc., that regulate the construction of discourse, on the other--contextual reference being clearly part of the latter. 914) In the case of entity-order type denotation, as we saw earlier (section 3), the range of entity orders that a given anaphor may have is constrained, not wholly by rules of clause structure (as Dik's representations seem to assume), but also, and mainly, by principles of discourse structure framed within the meta-principle of interpretative coherence. This point reinforces the need to implement fully the distinction (recognized in standard FG as well as in the recent FDG model) between the very general instructions for the creation of discourse provided by the verbal content of a given clause, and the much richer, more specific interpretation yielded via the integration of this configuration into a relevant context, at the utterance level.

This may be seen in examples like (3a) and (3b) above, where the same anaphoric form (the man) was able to have two different referents evoked via an identical antecedent-trigger predication--the initial sentences of each example--depending on which one was the target of a predication by the anaphoric clause as a whole: that is, in (3a), `having been quite badly hurt (in the circumstances previously evoked)', and in (3b), `needing to be deranged (to have done such a thing)'. We cannot know what the actual referent of a given anaphor might be until we have access to the discourse as well as to its context. (15) That is, to determine a given anaphor's intended referent, we need to take discourse pragmatics on board. The contradiction that I raised in section 2 of this article, between the conception of anaphora as involving a quasi-grammatical correlation between an antecedent expression in surrounding discourse and an anaphor (a relationship manifested in the sharing of indices between the two expressions), and the view of anaphora as involving the retrieval of a salient discourse representation encoded in the discourse partners' current discourse models, makes itself felt here. In fact, the coindexing device used by Dik is only conceivable where there IS a candidate antecedent expression in the surrounding cotext. Where there is not--as in the cases of exophora illustrated in (2a), (2b), (2d), and (2e), which I would argue are the basic instances of anaphora--this device is totally inoperative, the relationship here being of necessity between a salient discourse representation and an anaphoric expression, and not between two isolatable cooccurring expressions. (16)

The index of a given referent, whether it be an argument, predicate, predication, proposition, or clause, should not, in my view, already be marked as such in the representation of the anaphoric term, as Dik maintains it should be. After all, in the case of discourse anaphora, coreference or noncoreference is not something that is automatically assigned, as it may well be in grammaticalized instances of anaphora (sentence anaphora): for example, in infinitival subject control, relative clauses, reflexive predications, and so on. Here, the choice of index (i.e. coreferent) for a given anaphor is more or less fully determined by the syntactico-predicative context in which the anaphor occurs. I therefore believe that this index should not be specified in anaphoric terms at the level of the clause representation, at least under the interpretation where an anaphor is said to be in relation to an antecedent expression elsewhere in the cotext--as in Dik's definition II given in section 2 above. Thus a given anaphor is selected with its term-structure frame directly from the lexicon, where of course no indication of its actual reference can possibly be marked. This occurs only when the text constituted by a clause structure, in combination with surrounding clause structures, is converted into discourse by being integrated with a context. It should be left up to the (as yet unspecified) discourse component or domain of the model to provide heuristics for specifying the index--but see the tentative suggestions in section 5 as to how this might be done, as well as the existence within the FDG model (Hengeveld 2002) of a "communicative component" ensuring the updating of a given discourse context.

(17) below is an attempt to give more detailed specifications of the structures underlying given anaphor types (as contained in the lexicon).

(17) Proposal for revision of method of representing the structure underlying anaphors (PRIOR to their discourse resolution) in FG

it : (d1 A[alpha]: <inanimate> ([alpha]))
so : (i Af/X: <intensional> (f/X))
that : (dist A[alpha]: <inanimate> ([alpha]))
he : (d1 Ax: <human> & <male> (x))
they : (d/i m Ax: <set> (x))

Note that I have replaced the specific entity-order type included by Dik in the representations of it, so, and that in (15) either by the variable symbol "[alpha]" (the case of it and that in [17]) or by an alternative between two entity-order values (the case of so in [17]). Together with the presence of a selectional predicate at the core of each representation, this is intended to allow for a more flexible setting of the actual entity order of referent assigned to a token of each anaphor type through the influence of various contextual features at the utterance level (see section 5 for a suggestion as to how this might be formulated). The "A" symbol (designating an anaphoric term operator) may be interpreted as constituting an instruction to the addressee to look for a prior discourse entity whose representation can match the features already specified in the anaphor.

Particularly important here is the selection restriction imposed on a potential referent for the anaphor: see Cornish (1999: subsection 3.2.2) for some discussion, including reference to Dik's (1997a: section 4.4) very interesting conception of the rationale and operation of selection restrictions. Under Dik's view of the operation of selection restrictions, there is in fact a division of labor between the language system and the use of that system in context. Belonging to the former is the restriction by a given predicate on insertion in certain of its argument positions to terms bearing a specified semantic feature, a restriction that can be viewed as being part of that predicate's inherent meaning: see Cornish (i.p.: section 2.2) and the references cited therein on this point. But in terms of the USE of the language in context, the violation in purely system-internal terms of such a restriction via the presence of a term in an argument position whose value for the selectional predicate at issue is the opposite of the one corresponding to the semantic property associated with the inserted term can give rise to certain metaphorical readings in a given context. (17) Although Dik himself doesn't invoke it, it is Sperber and Wilson's (1995) relevance principle that is strongly suggested by this account of the "creative" operation of selection restrictions: the selection of the context-induced metaphorical interpretation constituting a "reward" for the extra processing cost incurred by the addressee is clearly what he has in mind here. In such an instance, we are clearly dealing with the addressee's perspective, since it is only in context, through the latter's attempt to arrive at a coherent interpretation at the discourse level--see my distinction between the levels of text and discourse under (1) above--that such an interpretation may be available.

Likewise, in terms of anaphora, the referent, which is not automatically yielded by the anaphor qua (potentially) referring expression, may be viewed as an argument to the essentially predicative conditions contained within the anaphor. (18) As in the case of a canonical predicate, the selection restriction carried by a given anaphor type constrains the possible choice of an argument/referent that will "saturate" it, thereby giving rise to a potential secondary reference, just as the saturation of a predicate by one or more arguments gives rise to a potential predication. As already noted, the "[alpha]" variable in the representations of the pronouns it and that in (17) ranges over the five entity-order types recognized so far in FG; however, as already indicated, we need to make allowance for more than this number.

5. The steps involved in the resolution of an anaphor within a discourse context

Under this conception of anaphor resolution from the addressee's perspective, (19) then, there are three distinct stages in the process (see also the algorithms for discourse anaphor resolution presented, for example, in Asher and Wada 1988; and in Passonneau 1996 and Walker 1998 within a centering-theory framework):

1. A given anaphor type is selected from the lexicon to fill the argument position of a predicator, and an anaphoric clause is formed.

2. A "co-composition" (Pustejovsky 1995: section 7.2) process is triggered, whereby certain relevant properties of the anaphoric clause as a whole are combined: the semantic class of the predicator assigns a denotational category to the anaphor, the predicator and its INTERNAL argument(s) are combined, then the output of this combination is integrated with the clause's referential features (its aspect, tense, mood, and modality assignments). These features collectively assign an entity-order type to the anaphor. See Cornish (1999: chapter 3) for some discussion of these relationships.

Where this is not already specified unambiguously (as in the case of this/that, it, so, etc.), then the entity-order type is filled in contextually.

Where it clashes with a specific entity-order type already contributed by the anaphor (as potentially in the case of the personal pronouns he/she/they, as well as one), anomaly is predicted.

Where it matches this value, then the resolution process may continue.

Where there is a small range of entity-order types already contributed by the anaphor (as in the case of so, for example--see [17]), then the anaphoric predication as a whole will be able to select the one conforming to the entity-order type that it compositionally specifies.

This general process is very similar to the way in which the selection restrictions of a predicate operate in the standard theory of FG (see Dik 1997a: subsection 4.4 as well as the brief discussion above) when a term is inserted in its argument position(s).

3. Consideration of the wider discourse context in which the clause is to be inserted will enable the anaphor's now quite specific indexical character to match the properties of an appropriate referent. I will list some of the relevant factors that impinge on this final stage of the resolution process here:

i. the relative topic status of potentially matching discourse referents;

ii. the type of discourse connection between the anaphoric clause and the segment in which the candidate discourse referents were last evoked;

iii. the existence or nonexistence of purely syntactic conditions relating the two clauses, conditions that may rule in or rule out certain matching processes; (20)

iv. the discourse status of the anaphoric clause in relation to the one where a candidate referent was last evoked: that is, whether the former clause is part of a background or a foreground segment, or effects a return pop over an intervening background segment to a previously active foreground segment, which it now continues, and so on. As an illustration, consider the interpretation of the pronouns in the last sentence of the following attested extract:

 (18) ... He [Kenny Rogers] grew up with four brothers and three sisters,
 the son of a labourer and a cleaning lady, in a poor area of Houston,
 Texas. My father was an alcoholic, but it wasn't disruptive because he
 was a wonderful man with a great sense of humour. The worst he did for
 our family was use money for alcohol rather than food or clothes. But
 he earned it, and had the right to get something out of life. He
 didn't drink for the last four years. His parents were not keen on him
 being a musician, and the early years were tough ... (Radio Times
 1999: 18).

Observe, first, that there are two discourse segments (21) in this extract, an outer or containing segment where it is the journalist who conducted the interview who is the speaker (or locutionary source), and an inner, embedded segment corresponding to the direct-speech section, where it is the interviewee, Kenny Rogers, who takes on the role of locutionary source. The direct-speech segment is explicitly delimited graphically in the original text via the opening and closing of the inverted commas (here via the use of italic script), and via the switch from third-person to first-person pronouns in reference to the interviewee.

Note also that the local discourse topics of each segment are distinct: for the main discourse segment, this is "Kenny Rogers," whereas for the embedded discourse segment, it is "Kenny Rogers's father." Once the direct-speech segment is terminated, it is popped from the highest position in the focus stack (according to Grosz and Sidner's [1986] hierarchy of focus spaces associated with given discourse segments), and its contents are therefore no longer available for anaphors (here the possessive determiner his and the third-person pronoun him in the final sentence) to pick up. And this corresponds to intuition, since these two anaphors are unambiguous in referring to Kenny Rogers rather than to Kenny Rogers's father, the topic of the intervening direct-speech segment.

This is also an instance where the criterion specified in 3(i) comes into play (i.e. anaphors' sensitivity to topicality differences in their potential referents). These anaphors, in conjunction with the content of their host predicator and the closing of the inverted commas (here, italic script) at the end of the immediately preceding sentence, effect a return pop to the main, interrupted segment, which is about Kenny Rogers himself.

Using (18) as a source of data for various other types of anaphor (specifically, the predicational it in line 3, and the first-order it in line 6), we can illustrate how the various types of resolution heuristics specified above operate. Taking first the it in line 3, this is the subject (and, more to the point, external argument--[A.sub.1] in FG terms) of the adjectival predication `not (be disruptive)'; as such, its denotation type would be that of a "second-order" entity, a type that is compatible with the range of entity types denotable by this pronoun (as [17] indicates).

Consideration of the wider discourse context yields the property `being an alcoholic', which has just been asserted of the local discourse topic of the segment being developed (Kenny Rogers's father). Now, this property can be construed in one of two ways: either as an attribute (the property `being an alcoholic' qua property), thus a zero-order entity; or as a dynamic SoA having a duration, localization in time and space, and so on.

Interestingly, it is the imposition of a second-order entity denotation type by the predicator (be) disruptive that selects the second of these two interpretation types from within this potential antecedent. Such a matching (the reference of it and the dynamically construed property in question) yields a perfectly coherent predication when combined with the negated adjectival predication (NEG Xi: (PAST [e.sub.i]: {disruptive [A]} (d1[Ae.sub.j]: <inanimate >)[PHI])).

Moreover, the property applied to Kenny Rogers's father of being an alcoholic is part of the focus space that is active at the point when the clause ... but it wasn't disruptive is processed, so that the pronoun it, which is specialized in accessing high-focus referents (see Gundel et al.'s [1993] givenness hierarchy), may readily retrieve it.

The next occurrence of the pronoun it is found in line 6 in the clause But he earned it ... Here, the pronoun is the direct object (i.e. [A.sub.2]) of the verbal predicate earn, so that it is construed as denoting a first-order entity, more specifically, due to the sense of earn, a definite amount of money. Again, a referent specifically concerning `money' has very recently been evoked--in the immediately preceding sentence--and so forms part of the focus space that is inherited by its following predication through the connection effected via the conjunction but and the occurrence of a pronoun, as subject of the verb, that maintains the previously existing topic, `Kenny Rogers's father'. For these reasons, this second occurrence of the pronoun it is interpreted as `the money that KR's father used to buy alcohol', and not `the alcohol that KR's father used [the] money [which he earned] to buy', which out of context is a type of referent that, potentially, this pronoun may well have.

These assignments of referents to pronouns within a naturally occurring text demonstrate the crucial role in this process played by the predicator of the indexical predication, through its transferring to the anaphor a specific selectional constraint that will enable it to retrieve a suitable part of the currently existing discourse representation as its referent. But this retrieval is by no means a direct, deterministic procedure, for as we have seen, the anaphor concerned, in conjunction with its immediate discourse-predicational context, is responsible for altering, adapting that salient discourse representation in relation to its new context.

This brief discussion has also shown that anaphora does not involve a relation between an anaphor and the referent of its antecedent (trigger) (in my terminology), but between an anaphor and the result of the processing of the relevant textual segment containing that trigger: in the case of the first it analyzed above, the pronoun is not interpreted as equivalent to simply `being alcoholic' qua general concept, but to `KR's father's being alcoholic during the period when KR and his brothers and sisters grew up'; likewise, the second instance of it that we examined is not interpreted as referring in general terms to `money', but to `the money that KR's father used to buy alcohol', a much more specific referent.

These referents are not available at the level of text, but at that of discourse: see my earlier distinction between these two notions under (1) in section 2; thus, in order to deal satisfactorily with discourse anaphora, more than simply the textual record of a communicative act needs to be taken into account--in particular also, a discourse model, where the results of the earlier processing of textual segments relative to a given context and their integration need to be available as the further context in terms of which subsequent textual segments are processed.

6. Conclusion

There should be a much more extensive division of labor in the FG account of discourse anaphora, as between the concerns of underlying clause structure (the representation of anaphoric terms as well as of the relevant properties of the anaphoric clause as a whole), and the broader discourse context in which the anaphoric predication is set. It is this dialectical relation between bottom-up and top-down processes that is instrumental in resolving an anaphor's reference--but also in specifying its precise function within the structure of the discourse as a whole. In this way, the gap between, on the one hand, the sparse, indirect set of instructions provided via the language system in the shape of a given text, and, on the other, the richer, more determinate context-bound interpretation that is discourse may more perspicuously be modelled. In this kind of way too, speaker's and addressee's perspectives may be reconciled at the level of the text in the form of a sequence of underlying clause structures, by structuring the latter in such a way that the speaker's anticipation and guidance of the addressee's intended interpretation may be represented.

"Prior mention" as a condition for the existence of discourse anaphora has, I believe, no specific role as such to play in an account of this phenomenon. Instead, the more general notion of antecedent trigger is, I would argue, required in order to bring exophora as well as indirect (e.g. associative) anaphora within the purview of this device.

Further progress along these lines would involve the following: first, more elaborate specification of the various anaphor types (along the lines I have suggested in [17]) needs to be undertaken, and the semantic-pragmatic effects on a given anaphor of the composition of the elements making up the anaphoric predication need to be made explicit. Furthermore, work needs to be done on the representation of discourse referents within a discourse model, taking account of discourse-dynamic relations such as coherence relations among clauses, discourse topic structures, and hierarchical discourse segmentation. I have attempted this for a selection of naturally occurring data using Asher's (1993) segmented discourse representation theory format in my recent book (1999: subsection 5.4.3). (22)

The picture painted above does, I believe, make for a more complete and more realistic view of the complex domain of discourse anaphora than the (necessarily) programmatic account given by Dik (1997b) in chapter 10.

Received 23 January 2001

Revised version received CNRS UMR 5610

20 November 2001

Notes

(1.) This article is an expanded and revised version of a paper presented at the 8th International Conference on Functional Grammar, which took place from 6 to 9 July, 1998, at the Vrije Universiteit, Amsterdam. A somewhat shorter version appeared under the title "The Functional Grammar conception of discourse anaphora: a (constructive) critique" as Working Paper in Functional Grammar 73, September, 2000 (pp. 1-17). I would like to thank Lachlan Mackenzie for his advice on an earlier version of the text published as WPFG 73, as well as two anonymous Linguistics readers for their detailed comments on the version submitted to the journal. Correspondence address: Dept. des Sciences du Langage, Universite de Toulouse-Le Mirail, 5 Allees Antonio-Machado, 31058 Toulouse Cedex, France. E-mail: cornish@univ-tlse.2.fr.

(2.) Witness the lengthy chapter 18 in vol. 2 of the revised theory (Dik 1997b) and the variety of work now being done on discourse-related issues--as represented in the conference at which the paper forming the basis of this article was presented. See also the range of interesting work within FG on issues connected with discourse contained in Connolly et al. (1997), and in Hannay and Bolkestein (1998). Following the 9th International Conference on Functional Grammar held in Madrid from 20 to 23 September, 2000, the outline of a new development in the model proper is being elaborated, to be called functional discourse grammar (see Hengeveld 2002).

(3.) By utterance acts, I mean here the acts of producing an utterance in some specific spatio-temporal context. The term is close to Austin's (1962) notion locutionary act.

(4.) I quote from Werth (1999: 46): "Let me at this point redraw the distinction I drew earlier [...] between the terms `text' and `discourse', based on context. Thus a text consists of the language itself, without taking into account the surrounding context. A discourse, however, is a language event: it is the language together with the context which supports it": and (1999: 302) (note the phrasing): "Analyzing the discourse of which [example] (6) is the textual expression ..." (emphases added within the latter quotation). However, Werth does not include under the heading text the nonverbal features that I include under it (see the text above) and does not consider a purely nonverbal act of communication--e.g. one conducted exclusively by means of gestures--as discourse, whereas I do.

(5.) For example, conversations, interviews, meetings, letters, sports commentaries, and so on.

(6.) See Cornish (1999: subsection 2.4.1) for justification for and illustration of this construct.

(7.) Cornish (1999: chapter 3) examines in detail various key aspects of this predication (termed there the indexical segment) and their role in the functioning of anaphora.

(8.) Note however the slight difference in the sense of the matrix verb told under each interpretation of the out-of-context ambiguous pronoun she in (4): where the NP Mary is understood as being coreferential with she, it has the sense `informed'; but where the NP Jane is coreferential with the pronoun, told has a value close to `ordered', `enjoined' (as in Mary told Jane to see a doctor, where only the second of the two interpretations indicated above is possible, the anaphor being the zero "subject" of see a doctor).

(9.) See McCawley (1991), Oakhill and Garnham (1992), and Cornish (1997) for presentations and analyses of corpora of attested utterances from a range of genres and registers, in connection with an account of discourse anaphora.

(10.) See Cornish (1998) for a discussion of Dik's conception of the structure of anaphoric chains in discourse, and the analysis of a short English newspaper article in terms of it.

(11.) Personally, I think `type' is a better characterization of the potential denotations of one.

(12.) This is in essence the same point that Liedtke (1998:111) makes in connection with Dik's view of the relationship between sentence form (mood types: declarative, interrogative, imperative, and exclamative) and type of illocutionary force (assertion/ statement, question, command, exclamation), Dik assuming a direct, one-to-one correlation between the various mood or sentence types, on the one hand, and the respective illocutionary force types, on the other. As analogously in the case of the relation between pronoun form types and entity-order types, Liedtke argues in favor of a much looser, nondeterministic relation between sentence form types and the potential illocutionary forces that they may be used to express.

(13.) Note that, where a subclass of propositional attitude verbs, such as guess, imagine, know, is constructed with so as propositional anaphor rather than with it, the semantic status of the verb changes: it no longer has its full lexical sense but takes on a parenthetical, nonassertive semantic value whereby the speaker withholds a full commitment to the truth of the proposition accessed via the pronoun. See Cornish (1992) for fuller discussion of this issue.

(14.) Others have argued in favor of a "division of labor" between, on the one hand, pragmatic principles that are separate from the clause grammar and deal with such phenomena as topic-focus assignment and illocutionary-force specification: these principles have a determining influence on grammatical coding; and on the other, the grammatical rules and constraints that govern morphosyntactic form, the province of a "grammatical module." See in particular Van den Berg (1998) for an elaboration of such a framework that takes account of the specifics of the FG model of grammar.

(15.) Of course the speaker knows what the intended reference is, since s/he actually initiated it, but obviously the speaker too is part and parcel of the utterance context, which is precisely the point I am making here (recall the brief discussion at the beginning of section 2 on this issue).

(16.) However, as an anonymous Linguistics reader points out, it might well be possible to specify the relation of intended coreference between the mental representation of a given referent within the speaker's (and the addressee's) respective discourse models, and a given anaphoric expression occurring in the cotext. This would be possible under Hengeveld's (2002) Junctional discourse grammar model, where a communicative component creates an ongoing record of the discourse developed prior to the incoming utterance.

(17.) See in this respect Dik's (1997a: 95) invented example (38): Rust eats iron, where the selectional predicates associated with eat are <animate> for the [A.sub.1] and <food> for the [A.sub.2].

(18.) See also Passonneau (1996: 241), who likewise suggests that discourse entities "(...) [are] arguments of a type of predicate corresponding to the lexical head of the evoking NP." This configuration is evidently not restricted to anaphoric expressions, under Passonneau's account.

(19.) A perspective that is nonetheless anticipated and guided by the speaker, who is primarily responsible for making the text available, on the basis of which the addressee, with his/her cooperation, will create discourse.

(20.) See in particular Van Hoek (1997) for a functionally oriented account of the traditionally syntactically defined constraints regulating intraclausal anaphora. Van Hoek's work is framed within Langacker's cognitive grammar system, an approach that has much in common with FG.

(21.) That is, basic discourse units, defined in part by their implementing a particular discourse purpose or goal relative to some more global discourse purpose: see Grosz and Sidner (1986) for both the term discourse segment and its definition and illustration.

(22.) I believe SDRT to be broadly compatible with the developing FG clause-structure framework.

References

Asher, Nicholas (1993). Reference to Abstract Objects in Discourse. Dordrecht: Kluwer Academic.

--; and Wada, Hajime (1988). A computational account of syntactic, semantic and discourse principles for anaphora resolution. Journal of Semantics 6(3/4), 309-344.

Austin, John L. (1962). How To Do Things With Words. Oxford: Clarendon.

Connolly, John H.; Vismans, Roel M.; Butler, Christopher S.; and Gatward, Richard A. (eds.) (1997). Discourse and Pragmatics in Functional Grammar. Berlin and New York: Mouton de Gruyter.

Cornish, Francis (1992). So be it: the discourse-semantic roles of so and it. Journal of Semantics 9, 163-178.

--(1997). Non-standard anaphora, discourse integration, and coherence. Verbum 19(1-2), 5-23.

--(1998). Les "chaines topicales": leur role dans la gestion et la structuration du discours. Cahiers de Grammaire 23, 19-40.

--(1999). Anaphora, Discourse, and Understanding. Evidence from English and French. Oxford: Clarendon.

--(i.p.). "Downstream" effects on the predicate in a functional grammar clause derivation. Journal of Linguistics 38 (2).

Dik, Simon C. (1997a). The Theory of Functional Grammar, Part 1: The Structure of the Clause, 2nd ed. Berlin and New York: Mouton de Gruyter.

--(1997b). The Theory of Functional Grammar, Part 2: Complex and Derived Constructions. Berlin and New York: Mouton de Gruyter.

Fretheim, Thorstein; and Gundel, Jeanette K. (eds.) (1996). Reference and Referent Accessibility. Amsterdam and Philadelphia: Benjamins.

Grosz, Barbara J.; and Sidner, Candace L. (1986). Attention, intentions, and the structure of discourse. Computational Linguistics 12, 175-204.

Gundel, Jeanette K.; Hedberg, Nancy; and Zacharski, Ron (1993). Cognitive status and the form of referring expressions in discourse. Language 69(2), 274-307.

Hannay, Mike; and Bolkestein, A. Machtelt (eds.) (1998). Functional Grammar and Verbal Interaction. Amsterdam and Philadelphia: Benjamins.

Hengeveld, Kees (2002). The architecture of a functional discourse grammar. In A New Architecture for Functional Grammar, Maria A. Gomez-Gonzalez and J. Lachlan Mackenzie (eds.). Berlin and New York: Mouton de Gruyter.

Lambrecht, Knud (1994). Information Structure and Sentence Form. Topic, Focus, and the Mental Representations of Discourse Referents. Cambridge: Cambridge University Press.

Liedtke, Frank (1998). Illocution and grammar: a double-level approach. In Functional Grammar and Verbal Interaction, Mike Hannay and A. Machtelt Bolkestein (eds.), 107-1127. Amsterdam and Philadelphia: Benjamins.

Mackenzie, J. Lachlan (1992). Places and things. In Layered Structure and Reference in a Functional Perspective, Michael Fortescue, Peter Harder, and Lars Kristoffersen (eds.), 253-276. Amsterdam and Philadelphia: Benjamins.

--(1997). Grammar, discourse and knowledge: the use of such in written English. In Studies in English Language and Teaching, J. Aarts, I de Monnick, and H. Wekker (eds.), 85-105. Amsterdam: Rodopi.

Manoliu-Manea, Maria (1990). French neuter demonstratives: evidence for a pragmasemantic definition of pronouns. In Variation and Change in French, John N. Green and Wendy Ayres-Bennett (eds.), 89- 115. London and New York: Routledge.

McCawley, James D. (1991). A linguistic flea circus. Indiana University Linguistics Club mimeo.

Oakhill, Jane; and Garnham, Alan (1992). Linguistic prescriptions and anaphoric reality. Text 12, 161-182.

Passonneau, Rebecca J. (1996). Using centering to relax Gricean informational constraints on discourse anaphoric noun phrases. Language and Speech 39(2-3), 229-264.

Pustejovsky, James (1995). The Generative Lexicon. Cambridge, MA, and London: MIT Press.

Radio Times (1999). Interview with Kenny Rogers. Radio Times 7, August 13, 18.

Sperber, Dan; and Wilson, Deirdre (1995). Relevance Theory: Cognition and Communication, 2nd ed. Oxford: Blackwell.

Van den Berg, Marinus (1998). An outline of a pragmatic functional grammar. In Functional Grammar and Verbal Interaction, Mike Hannay and A. Machtelt Bolkestein (eds.), 77-103. Amsterdam and Philadelphia: Benjamins.

Van Hock, Karen (1997). Anaphora and Conceptual Structure. Chicago: University of Chicago Press.

Walker, Marilyn A. (1998). Centering, anaphora resolution, and discourse structure. In Centering Theory in Discourse, Marilyn A. Walker, Aravind K. Joshi, and Ellen F. Prince (eds.), 401-435. Oxford: Clarendon.

Werth, Paul N. (1999). Text Worlds: Representing Conceptual Space in Discourse. Harlow: Pearson Education (Longman).

Wilson, Deirdre (1992). Relevance and reference. University College London Working Papers in Linguistics, 1-25.

Universite de Toulouse--Le Mirail CNRS UMR 5610