1. Expressivity markers in free indirect thought

Within relevance theory it has been argued that the meanings of a subset of the expressions labelled ‘discourse markers’, including standard cases of Gricean conventional implicature (however, therefore, moreover), should be analysed in terms of the way they constrain or direct pragmatic inference rather than the way they contribute to the conceptual content of the utterances that contain them (Blakemore Reference Blakemore1987, Reference Blakemore2002). This distinction between procedural and conceptual meaning reflects the representational/computational approach to cognition which underpins Sperber & Wilson's (Reference Sperber and Wilson1995) relevance theoretic approach to utterance interpretation. However, it is also justified in terms of the pragmatic principle which, according to relevance theory, is essential for the explanation of how human communication is achieved. According to this principle, every act of ostensive communication communicates a guarantee of optimal relevance so that it may be assumed that the communicator has aimed to achieve the highest level of relevance he is capable of within the parameters of his abilities and interests. Since the degree of relevance increases with the number of cognitive effects derived and decreases with the amount of processing effort required for their derivation, the use of an expression which encodes a procedure for identifying the intended cognitive effects would be advantageous from a processing point of view: it provides a means of ensuring the identification of the intended cognitive effects for a minimum cost in effort.Footnote ²

On this approach, expressions which encode procedures rather than concepts are intrinsically linked to communication: their function is to reduce the effort which the hearer must invest in recovering the intended interpretation of an utterance. In real life, communicators produce public representations (utterances) of their private thoughts, and hearers are intended to use the linguistic properties of these utterances and contextual assumptions in the derivation of interpretations of those thoughts. In fiction, authors may represent fictional acts of communication in which fictional speakers are represented as producing utterances which readers are meant to assume are interpreted by fictional hearers. Clearly, there is a limit to the extent to which these representations of fictional acts of communication resemble actual acts of communication. An author may attempt an orthographic indication of a character's accent (see Fludernik Reference Fludernik1993: 259–260), or orthographic representations of features such as emphatic stress. But in general the representation is a less than verbatim record of the utterance which the character is supposed to have made.Footnote ³ Nevertheless, according to the relevance theoretic framework of this paper, an author who is representing a character engaged in the act of communication must be assumed to be representing someone who is aiming at optimal relevance. This means that it is not particularly surprising that fictional communicators are represented as using expressions which encode procedures, or, in other words, expressions which are inherently communicative in the sense outlined above. The only difference is that in such cases the expression is linked to a represented act of communication rather than an actual one.

However, in free indirect thought (FIT) these ‘intrinsically communicative’ or audience-directed devices are used by writers in their representations of thoughts which must be attributed to characters who are not engaged in an act of communication at all. For example, in (1) well must be attributed to Mansfield's character Linda Burnell, who is represented as simply thinking (underlining in this example and the ones which follow is my own):

1. (1) And what made it doubly hard to bear was, she did not love her children. … Even if she had the strength she would never had nursed and played with the little girls. No, it was as though a cold breath had chilled her through and through on each of those awful journeys; she had no warmth left to give them. As to the boy – well, thank heaven, mother had taken him.

   (Katherine Mansfield, ‘At the Bay’, Collected Stories, p. 223)

Similarly, in (2), anyway must be attributed to Grace's character, Rua, who is engaged not in the act of communication, but in the act of thought:

1. (2) Not long after sunrise he'd thrown his line out and spent about an hour before catching two fish, then he'd scaled and cleaned them, washed up, taken his time coming back. Probably getting on for eight o'clock. Anyway, it was all his own doing, getting her to come here. Why had he?

   (Patricia Grace, Dogside Story, p. 75)

Discourse markers are amongst a whole range of expressions and structures found in FIT which are said to function as markers of expressivity or subjectivity (Banfield Reference Banfield1982, Fludernik Reference Fludernik1993). These include a heterogeneous class of items, usually called interjections, some of which are considered to be marginal to language. Examples include ah in (3), oh in (4), good heavens in (5) and Linda Burnell's no in (1):

1. (3) Linda was astonished at the confidence of this little creature … Ah, no, be sincere. That was not what she felt; it was something far different, it was something so new, so …

   (Katherine Mansfield, ‘At the Bay’, Collected Stories, p. 223)

2. (4) Oh, how marvellous to have a brother! In her excitement Leila felt that if there had been time, if it hadn't been impossible, she couldn't have helped crying because she was an only child and no brother had ever said ‘Twig?’ to her; no sister would ever say, as Meg said to Jose at that moment, ‘I've never known your hair go up more successfully than it has tonight’.

   (Katherine Mansfield, ‘Her First Ball’, Collected Stories, p. 337)

3. (5) And once too in his studio, where the Consul was so obviously not going to arrive, M. Laruelle had shown her some stills of his old French films, one of which it turned out – good heavens! – she'd seen in New York.

   (Malcolm Lowry, Under the Volcano, p. 267)

Since these expressions signify feelings or states of minds rather than concepts which can be explicated, they cannot be straightforwardly accommodated in a translational semantics for natural language (cf. Kaplan Reference Kaplan1997; Wharton Reference Wharton2003a, Reference Whartonb). Thus Kaplan (Reference Kaplan1997) argues that, in contrast with words such as fortnight and feral, these expressions encode information about the contexts in which they are appropriate, or, in other words, that they require a semantics of use rather than a semantics of meanings. And within a relevance theoretic framework, Wharton (Reference Wharton2003a, Reference Whartonb) has argued that while interjections such as ah and oh may not be properly linguistic, they nevertheless have a coded element which is procedural in the sense that they are used by speakers in order to encourage the hearer to construct conceptual representations of the emotions and attitudes they wish to communicate.Footnote ⁴ However, the interjections in (3)–(5) above are all attributable to characters who are engaged not in acts of communication but in acts of thought. While one can see why a communicator might use an expression which encourages a hearer/reader to construct a meta-representation of the emotions he is experiencing, it seems difficult to see how such an expression could play a role in thought.

The question, then, is how can we justify the use of expressions which, according to the relevance theoretic approach outlined above, are not constituents of thoughts in representations of the thoughts of characters who are engaged in private thought rather than communication?

There is no question that this must be justified. The authors cited above cannot be accused of doing something which is somehow illegitimate. As we shall see in this paper, these expressions play an important role in creating an impression of being able to gain entry to a character's mind, or of being able to witness him/her as s/he is actually having the thoughts in much the same way that we are able to witness a speaker as he constructs utterances as public representations of his own thoughts. This effect is often encapsulated in the description of FIT as ‘inner speech’ (cf. Chatman Reference Chatman1978, Ehrlich Reference Ehrlich1990). However, as I have just argued, this is not ‘speech’ in the sense of an act of communication intended for a hearer. It is not represented speech, but represented thought.

Here we have the key to the answer to the question I have just posed. The point is that these are public representations of thoughts, and these cannot be confused with the private thoughts they represent. The fact that the use of expressions such as well, anyway, ah and oh gives us access to the minds of the characters whose thoughts are being represented does not mean that these expressions are actually constituents of the thoughts being represented.

It might seem that the point I am making here is no different from the point that utterances made by communicating speakers should not be confused with the thoughts they represent, and hence that the presence of expressions such as well, anyway, ah and oh in FIT deserves no more explanation than their presence in any utterance which communicates the speaker's thoughts. However, this paper assumes that there is a difference between the two cases. An utterance made in the course of real-life (or, indeed, simulated) communication is evidence of the communicator's intention that the audience should derive an interpretation which is a representation of that communicator's thoughts. The speaker is communicating his thoughts. As we shall see, he may entertain a thought as a representation of another person's thoughts, in which case the utterance can be said to be used attributively. But it is still the speaker who is entertaining this thought as a representation of another thought.

A FIT representation, on the other hand, is simply intended as a representation of someone else's thoughts. In producing a FIT representation, the author cannot be understood to be communicating his thoughts, but must simply be taken to be representing thoughts which are being entertained by another. In other words, a FIT representation is unmediated by the thoughts of a communicating speaker.

In this paper, I argue that the use of discourse markers and expressive interjections in FIT representations must be explained in terms of an approach to FIT which is able to capture this difference. In section 2, I outline the relevance theoretic approach to attributively used utterances and argue that it does not capture the difference between communicated thought and represented thought just outlined. In section 3, I address the question of whether the use of expressive devices in FIT can serve as a means of distinguishing FIT representations from indirect thought representations. In section 4, I argue for a relevance theoretic account of FIT representations which departs from the one suggested by Wilson (Reference Wilson2006) in terms of tacit attributive use. And in section 5, I show how this approach serves as the basis for the explanation of the use of discourse markers, expletives and interjections in FIT representations. Section 6 concludes the paper.

2. Relevance and meta-representation

Within relevance theory, Wilson (1995, Reference Wilson and Sperber2000) have shown that the ability to form meta-representations or representations of representations is fundamental to the explanation of linguistic communication. In the first place, since an utterance is a public representation which has a propositional form, it can be used to represent another representation which has a propositional form – or, in other words, a thought. However, utterances do not communicate thoughts by duplicating them. All we can say is that utterances interpret the thoughts they are used to communicate (Sperber & Wilson Reference Sperber and Wilson1995: 230). An utterance is an interpretation of a thought to the extent that its propositional form resembles the communicator's thought, or, in other words, to the extent that it shares logical and contextual implications with that thought. The more implications they share, the closer the resemblance, and identity is simply a special case of resemblance, in which the propositional form of the utterance shares all the implications of the thought in every context.

The hearer's task in interpreting this utterance is to use the utterance as a means of deriving a representation of the speaker's thought – a thought about a thought. However, once again it is not assumed that the representation derived by the hearer duplicates that thought. The hearer must use pragmatic inference to bridge the gap between the linguistically encoded meaning of the utterance and its interpretation. Since the linguistic evidence provided by the utterance may give the hearer varying degrees of responsibility for deriving an interpretation which satisfies his expectations of relevance, it is possible that he will use the utterance as a means for deriving assumptions which are not identical with ones envisaged by the speaker. That is, the thoughts derived by the hearer will simply be assumed to be an interpretation of the speaker's thoughts, and communication will succeed to the extent that it results in the enlargement of the mutual cognitive environment of speaker and hearer.Footnote ⁵

However, as Sperber & Wilson show, the thought represented by an utterance may itself be entertained by the speaker as a representation of another (attributed) thought. Consider the summary of a speech, or the minutes of a meeting, or a report made by a speaker of another person's utterance or thought attributed made by another. Once again the intention is not necessarily to replicate that utterance or thought. Even so-called direct representations do not necessarily reproduce the original exactly. The following was heard in a reading of a story for children (BBC Radio 7):

1. (6) ‘Who said you could put your bottom on my chair’, he said – except he didn't use the word bottom.

Similar examples are given by Wilson (Reference Wilson and Sperber2000). In the narrratology literature, the idea that direct discourse is a verbatim representation of the original has also been exposed as a myth (see Sternberg Reference Sternberg1982, Fludernik Reference Fludernik1993). Thus Fludernik (Reference Fludernik1993: 409–410) cites both non-literary (oral) and literary examples of cases in which a narrator attributes a thought or statement to characters who have not uttered a word.

1. (7) And I just felt like, ‘This is where I belong’. (from Tannen Reference Tannen1989: 152)

2. (8) An approving murmur arose from the heads of tribes; as saying, ‘There you have him! Hold him!’

   (Charles Dickens, Our Mutual Friend I, xi, p. 187)

As Wilson (Reference Wilson and Sperber2000) points out, languages have a variety of devices available to speakers for indicating attributive intentions – for example, parenthetical clauses such as ‘he thought’ or ‘he said’, hearsay sentence adverbials such as ‘allegedly’ or ‘according to Pam’, hearsay particles, prepositions, and modals.Footnote ⁶ However, in many cases, the speaker's intentions are not indicated by the form of his utterance and the hearer will have to use pragmatic inference to work out that an utterance is intended as a representation of an attributed thought. For example, in (9) below, the speaker's use of the sentence adverbial according to Pam indicates that the thought which is interpreted by the segment in (b) is being entertained as a representation of Pam's thought, and hence that the hearer should recover the higher-level explicature in (11). However, in (10), the attribution is tacit and the hearer is intended to infer that the utterance is being used attributively on the basis of the context and the Principle of Relevance:

1. (9) (a) I thought it was really nice. (b) But according to Pam, I looked ridiculous.

2. (10) [The speaker, who has been trying on a dress in a shop, emerges from the changing cubicle and looks questioningly at her friend who prepares herself to say something tactful]

   Friend: Well, …

   Speaker: I look ridiculous.

3. (11) Pam thinks I look ridiculous.

According to Wilson (Reference Wilson2006), free indirect speech and thought can be treated along the same lines as (10), as a tacitly attributive uses of language. Thus she argues that the second segment of (12) will be misunderstood if it is not understood that the speaker is thinking about what the students are thinking rather thinking directly about a particular state of affairs.

1. (12) (a) The students were thoughtful. (b) If they didn't act now, it would be too late.

   (from Wilson Reference Wilson2006: 1730, ex. (12))

In fact, as Banfield (Reference Banfield1982: 76–88), Ehrlich (1982: 76–88) and Fludernik (Reference Fludernik1993: 165, 240–241, 285–297) have pointed out, authors who provide FIT representations may use parentheticals to explicitly indicate the source of the represented thought and its mode of representation. The example in (13) is cited by Ehrlich (Reference Ehrlich1990), while (14)–(16) are cited in Blakemore (Reference Blakemore2008):Footnote ⁷

1. (13) Her shoes were excellent, he observed.

   (Virginia Woolf, To the Lighthouse, p. 22)

2. (14) Human relations were all like that, she thought, and the worst (if it had not been for Mr Bankes) were between men and women.

   (Virginia Woolf, To the Lighthouse, p. 107)

3. (15) Yvonne knew where she was now, but the two alternatives, the two paths, stretched out before her on either side like the arms – the oddly dislocated thought struck her – of a man being crucified.

   (Malcolm Lowry, Under the Volcano, p. 319)

4. (16) So the ‘other’ had come again. And now gone, he thought: but no, not quite, for there was still something there, in some way connected with it, or here, at his elbow, or behind his back, in front of him now …

   (Malcolm Lowry, Under the Volcano, p. 96)

However, whether or not the source of the represented thought is indicated by a parenthetical, it seems that FIT representations must be distinguished from utterances such as (9) or (10). Notice that although the speaker of (10) will be understood to be using his utterance attributively, the attributed thought it is being used to represent is nevertheless being viewed from the speaker's perspective. As Reinhart (Reference Reinhart and Rauh1983) and Ehrlich (Reference Ehrlich1990) have pointed out, whereas in free indirect thought representations it has to be assumed that the person whose thought is being represented actually had that thought (or one something like it), this is not the case in indirect thought representations. Thus in (10), the speaker may have simply inferred that Pam had the thought that he looked ridiculous on the basis of his observations of Pam's behaviour.

Moreover, whereas in tacit indirect thought the speaker may use an expressive to indicate that the attributed thought is being seen from his own point of view, in free indirect thought, expressives must be attributed to the character whose thoughts are represented. For example, while Pam can be attributed with the thought that the speaker looks ridiculous in (17), the expletive can be attributed to the speaker:

1. (17) [The speaker emerges from a changing cubicle in a clothes shop and looks questioningly at Pam who prepares herself to say something tactful]

   Pam: Well, …

   Speaker: Oh God I look ridiculous.

Similarly, in (18), while the speaker will be understood to be attributing the thought that she looks ridiculous to someone else (Pam), she must be understood as the source of the discourse marker so:

1. (18) [The speaker emerges from a changing cubicle in a clothes shop and sees the look of shocked disapproval on her friend's face]

   So I look ridiculous. Is there anything in this shop you approve of?

In contrast, in the constructed FIT examples in (19) and (20), My God and so must be attributed to the person whose thoughts are being represented, and we are intended to view these thoughts from the perspective of that person:

1. (19) My God, how ridiculous he looked.

2. (20) So he had come.

The fact that that in tacitly attributed utterances, expletives and discourse markers can be attributed to the speaker who is attributing the represented thought rather than to the person whose thought is being represented suggests that we are intended to understand that this thought is being viewed from the perspective of the speaker rather than the source of the attributed thought. However, the fact that the expressives in (19) and (20) must be attributed to the character whose thought is being represented suggests that we are intended to view this thought from the perspective of this person rather than the speaker who is responsible for representing it.

3. Procedural meaning in indirect thought reports

The presence of expressive devices in free indirect thought representations is often contrasted with their absence in indirect thought representations. Thus for example, while good heavens is acceptable in the FIT passage in (5), it cannot be interpreted as falling under the scope of the verb think in (21):

1. (21) ?She thought that good heavens she had seen it in New York.

However, it seems that the question of whether these devices can occur in indirect discourse is far from straightforward. Thus both Fludernik (Reference Fludernik1993) and Potts (Reference Potts2007) have shown that expletives and expressive epithets can fall under the scope of certain verbs in indirect speech reports:

1. (22) Mr Sparkler, growing rash on his late success, observed that Edward had, biggod, a long bout of it, …

   (Charles Dickens, Little Dorrit I, xxiv, p. 762, cited by Fludernik Reference Fludernik1993: 239; my underlining)

2. (23) My father screamed that he would never allow me to marry that bastard Webster. However, I would never have gone out with him if I had thought he was a bastard.

   (adapted from example due to A. Kratzer, cited by Potts Reference Potts2007: 7, ex. (15); my underlining)

Moreover, as Rieber, (Reference Rieber1997), Bach (Reference Bach1999) and Potts (Reference Potts2005) have shown, certain discourse markers which have been analysed in procedural terms may fall within the scope of indirect thought reports. (24) is based on an example from Wilson & Sperber (Reference Wilson and Sperber1993), while (25) is from Potts (Reference Potts2005):

1. (24) Henry thought that Sue had had holiday and so he should have one too.

2. (25) Marv thinks that Shaq is huge but [that he is] agile.

If expressives and discourse markers fall within the scope of indirect thought reports, as these examples suggest, then it would seem that there is one less argument from distinguishing free indirect discourse from standard indirect discourse.Footnote ⁸

Examples such as (22) and (23) might be dismissed as unproblematic since they can be treated as cases of implicit direct quotation (see Hall Reference Hall2007 for discussion). This, however, is not an option in indirect thought reports, since the subject cannot be understood to have produced an utterance, and it seems that the assumption that ejectives and expressive epithets cannot fall within the scope of indirect thought reports must hold. Notice that in contrast with (23), the speaker's dissociation from the epithet in the second segment of the indirect thought report in (26) is odd:

1. (26) My father thinks that I shouldn't marry that bastard Webster. ?However, I would never have gone out with him if I had thought he was a bastard.

However, this leaves us with the examples in (24) and (25) where, it seems, so and but fall within the scope of the indirect thought report. If, as Bach (Reference Bach1999) has argued, the speaker of these utterances is attributing thought contents, then it would seem that so and but would have to be analysed as contributing to those contents rather than as audience directed devices for constraining interpretation. And indeed, Potts (Reference Potts2005: 39) has argued that the fact that but must be regarded as part of the utterance's truth conditional or ‘at issue’ content explains why the entirety of its content is attributed, not to the speaker, but to the subject of the verb of saying or thinking when it is used in an indirect speech or thought report.

Bach's (Reference Bach1999) solution focuses on utterances containing but, which, he claims, express multiple propositions, all of which are part of its truth conditional content. Thus an utterance such as (27) expresses the propositions in (28):

1. (27) Shaq is huge but agile.

2. (28)

   1. (a) Shaq is huge.

   2. (b) Shaq is agile.

   3. (c) There is a certain contrast between being huge and agile.

The fact that but (c) does not seem to affect the truth of (27), Bach argues, is explained by saying that it is usually part of common ground and hence not salient in assessments of truth or falsity. However, as Hall (Reference Hall2007) points out, Bach's account does not explain why the proposition that there is a contrast happens to be backgrounded in all cases except for speech and thought reports.

In fact, it seems that there are uses of but and so which become unacceptable when they are embedded in indirect thought reports. Consider, for example, a situation in which Jane believes that Henry has gone shopping for food and sees him return empty-handed. In this context, Jane might produce the utterance in (29):

1. (29) So we've got nothing for dinner.

Alternatively, Henry might anticipate Jane's utterance and produce the indirect thought report in (30):

1. (30) You think that we've got nothing for dinner.

However, he cannot produce the unacceptable indirect thought report in (31):

1. (31) ??You think that so we've got nothing for dinner.

Similar examples can be constructed for but. Thus in a context in which Henry has asked Jane to distribute the handouts for a lecture, Jane's utterance in (32) is acceptable, as is Henry's indirect thought report in (33). However, the indirect thought report in (34) is unacceptable:

1. (32) But there are not going to be enough.

2. (33) You think that there aren't going to be enough.

3. (34) ??You think that but there aren't going to be enough.

The point is, of course, that whatever explanation is given for the unacceptability of (31) or (34) would have to be consistent with the acceptability of (29) or (32).

It seems that this explanation can be found in a procedural analysis of so and but in the indirect thought reports in (24) and (25). As we have seen in section 2, indirect thought reports are reports of what the speaker thinks about the subject's thoughts, and these may include reports of how the subject thinks specific propositions are related in an inference. Given that the relevance of such a report lies in the speaker's interpretation of what the hearer thinks, then it is in the speaker's interests that the hearer is able to identify this interpretation on the basis of the utterance he makes. Since the function of words such as but and so is to make particular inferential routes accessible, one might expect a speaker who wishes to represent a subject as thinking that propositions are related in a particular kind of inference to use these words in order to ensure that the hearer derives the intended interpretation of his utterance for a minimum cost in processing. The interpretation recovered by the hearer on the basis of the meanings of each of these expressions is an interpretation of the speaker's interpretation of the subject's thoughts/inferences. However, this is not to say that the hearer will attribute the speaker with the thought that the subject has a thought which contains constituents corresponding to but or so. The hearer will attribute the speaker with the thought that the subject can be attributed with the thought that the propositions which are included within the scope of the thought report are related in a particular kind of inference. The role of so or but is simply to indicate what kind of inferential relationship this is.

Returning now to (29)–(31), we can see that the speaker of (29) (Jane) is not expressing a thought about an inferential relationship, but is simply expressing a thought (that they have nothing for dinner) and indicating how it achieves relevance (as a contextual implication from an assumption which is assumed to be mutually manifest). It is, of course, possible for someone (Henry) to attribute him with the thought which she has expressed (as in (30)). However, since so does not contribute to the content of the thought which Jane expresses in (29), but simply constrains the hearer's interpretation of it, it cannot be part of the thought content which Henry is attributing to her in (31). But nor is it being used to indicate how Jane thinks two propositions are related in an inference. It is not surprising, then, that (31) is unacceptable. The examples in (32)–(34) can be explained along similar lines.

4. The representation of thought in fiction: a relevance theoretic re-assessment

In contrast with the indirect thought reports in (31) and (34), the free indirect thought reports in (35) and (36) are acceptable:

1. (35) So there would be nothing for dinner, she thought.

2. (36) But there were not going to be enough, she thought.

This brings us back to the question with which we began, and the main focus of this paper: how do we account for the use of intrinsically communicative devices in the representation of the thoughts of a subject not engaged in the act of communication? In this section, I develop an account of free indirect style which is based on Sperber & Wilson's relevance theoretic approach to communication, but which incorporates the insights of Banfield (Reference Banfield1982), who argues that the interpretation of a free indirect discourse text must be contrasted with the interpretation of communicated discourse since its interpretation does not require the identification of a (speaking) communicator and a hearer, but simply the identification of the point of view (self) of the character whose thoughts and utterances are being represented.

It might seem that such an account is impossible. Sperber & Wilson's theory is a theory of communication: interpretation is constrained by the expectations of relevance which follow from the hearer's recognition that an act of ostensive communication has been performed. Banfield, on the other hand, distinguishes narration and representation in fiction (the expressive dimension of language) from communication (text), where the speaker's very presence guarantees the possibility of I (cf. Banfield Reference Banfield, Herman, Jahn and Ryan2005: 396).

However, as we have seen, in relevance theory communication is defined in pragmatic terms: the question of whether an act of ostensive communication has been performed is not determined linguistically. Banfield's distinction between communication and the expressive dimension of language is, in contrast, a linguistic distinction corresponding to the distinction between language whose interpretation depends on the deictic centres speaker and hearer (I and you), on the one hand, and language whose interpretation depends on the deictic centre self (the point of view of the character whose thoughts are being represented).

It seems that there is a similar distinction underlying Schlenker's (Reference Schlenker2004) proposals for the analysis of indexicals in free indirect discourse and the historical present. Schlenker distinguishes between the context of utterance (the point at which a thought is expressed) and context of thought (the point at which a thought originatesFootnote ⁹), arguing that indexicals must be classified according to the context with respect to which they are evaluated. In particular, whereas the interpretation of tenses and pronouns depend on the context of utterance, the interpretation of other indexicals (e.g. now, here, yesterday, this, that) depends on the context of thought. In ordinary (e.g. face-to-face) communication (Banfield's I speaking to you) these two contexts coincide. However, in free indirect discourse thought they do not, thus creating the impression that another person's thoughts are being articulated through another person's mouth.

Following Banfield and Schlenker, it seems that we could say that as they are used in free indirect discourse, expressives depend for their interpretation on the context of thought rather than the context of utterance. However, as we shall see, not all aspects of the interpretation of a represented thought depend on the context of thought, where this is defined as the point where the thought originates. Many of the conclusions that we draw from such a representation depend on contextual assumptions derived from our interpretation of other character's thoughts, and, in some cases, on the point of view of the author (so-called ‘authorial intrusion’). Moreover, as Fludernik (Reference Fludernik1993) shows, point of view is not always established linguistically. In some cases, it is not possible to identify a reflecting character whose point of view is being represented.Footnote ¹⁰ Further, while expressive devices may evoke expressivity, the actual subject of consciousness must be inferred pragmatically on the basis of contextual assumptions made accessible by the text. Finally, as we have seen in section 3, expressives are not restricted to free indirect discourse: they may be embedded within the scope of indirect speech and thought reports. Fludernik concludes that expressive elements must be treated as signals used to ‘evoke subjectivity rather than a mere surface structure of underlying actual consciousness or self’ (1993: 398; emphasis in the original).

My proposal is that the distinction Banfield draws between communication and representation should in fact be drawn in pragmatics. Moreover, in contrast with Banfield, I do not interpret this distinction as a distinction between a non-communicative use of language and a communicative use. Within a relevance theoretic framework, the representation of thought (and speech) in fiction is a variety of ostensive communication.

Clearly, we would not expect the representation of the thought processes of someone who does not have a communicative intention to capture the sort of steps a communicator might take in order to satisfy a hearer's expectations of optimal relevance. Hence the incoherence of thought representations such as the one in (37):

1. (37) Yet in the Earthly Paradise, what had he done? He had made few friends. He had acquired a Mexican mistress with whom he quarrelled, numerous beautiful Mayan idols he would be unable to take out of the country, and he had – M. Laruelle wondered if it was going to rain.

   (Malcolm Lowry, Under the Volcano, p. 16)

However, while we might not expect optimal relevance from M. Laruelle, we do expect optimal relevance from the author who is representing this thought. Lowry may not be communicating in the sense that he is telling us what M. Laruelle was thinking: he is simply showing us M. Laruelle's thoughts. However, as Sperber & Wilson (Reference Sperber and Wilson1995: 52–53) have argued, even where an act of showing provides strong direct evidence for the basic layer of information (as in the case of opening a cupboard to let someone see the contents), there is an intention to draw some attention to the fact that the act was intentional and hence that the audience can assume that by paying attention they will discover relevant information. Thus the author must be assumed to communicate a guarantee that the effort invested by the reader in processing his text will be rewarded.

However, as Banfield has said, the author cannot be said to be speaking in fiction in the sense that he is communicating his thoughts.Footnote ¹¹ Putting it in her terms, the author does not ‘speak’ in the text, but rather ‘creates a fictional world … out of language, whether or not the language is attributed to a narrator’ (Banfield Reference Banfield, Herman, Jahn and Ryan2005: 397). In texts with a narrator, the author can be said to represent or show us the thoughts of a narrator who is communicating with a narratee. In free indirect style texts, where there is no narrator, the author can be said to represent or show us the thoughts of a character who is either engaged in a fictional act of communication with another character (free indirect speech) or simply engaged in private thought. It is the latter sort of case which is the focus of this paper.

However, whether the reader is interpreting a text with a narrator or not, the reward for the effort he invests cannot lie in the improvement to the mutual cognitive environment of author and reader, since the author is showing us or representing the thoughts/utterances of another rather than communicating his own thoughts. Instead the reward lies in the relationship which is created between the reader and the fictional individuals in the fictional world which the author is representing. And this is more like the relationship that one might have with an individual whose life one is watching on film – except, of course, that in the sort of case under discussion in this paper, one has the impression of watching a film of an individual's private mental life. As Dillon & Kirchhoff (Reference Dillon and Kirchhoff1976: 438) say, ‘FIS [free indirect style] is more often a strategy through which the narrator appears to withdraw from the scene and thus present the illusion of a character acting out his mental state in an immediate relationship with the reader’.

While the author may not have a ‘voice’ in the text, he is nevertheless responsible for deciding what is shown to the reader and when it is shown. This means that the relationship just described is orchestrated by the author so that while readers are sometimes encouraged to adopt an empathetic attitude to the characters whose thoughts are being represented, in other cases they are encouraged to adopt a negative attitude towards them. This is achieved in a variety of ways. In some cases, the author simply lets a character reveal his own defects. For example, in ‘Mr Reginald Peacock's Day’, Mansfield's representation of Mr Peacock will lead most readers to the conclusion that he is an extremely vain man, so that his claim that he cannot stand vain men makes him seem all the more ridiculous. In other cases, an author may create a feeling of dissociation from one character by encouraging us to see him through the eyes of another. Thus in ‘Prelude’, Mansfield has us see Stanley Burnell do his exercises through the eyes of his wife Linda (see Blakemore Reference Blakemore2008 for further discussion). And as Dillon & Kirchoff (Reference Dillon and Kirchhoff1976) show, an author may interrupt the representation of the thoughts of another with an utterance which represents his own view (authorial intrusion). Thus they argue that in the following extract from Conrad's Secret Agent, the utterance introduced by the sentence adverbial as a matter of fact indicates a view which contrasts with that of the character whose thoughts are being represented in the preceding segment: Mrs Verloc, who has just killed her husband, is panicking and hysterical, while the narrator is cool and detached:

1. (38) She looked up mechanically at the clock. She thought it must have stopped. She could not believe that only two minutes had passed since she had looked at it last. Of course not. It had been stopped all the time. As a matter of fact, only three minutes had elapsed from the moment she had drawn the first deep, easy breath after the blow, to this moment when Mrs Verloc formed the resolution to drown herself in the Thames.

   (Joseph Conrad, The Secret Agent, p. 238)

More generally, the representation of a character's thought(s) must be interpreted in the context of the whole text, which, as Ehrlich (Reference Ehrlich1990) as emphasized, may represent the perspectives of a range of different characters. Thus the conclusions we derive from the representation of a given character's thoughts may be affected by the interpretation of the representations of the thoughts of another, or, indeed, by the thoughts revealed by the sort of authorial intrusion discussed by Dillon & Kirchoff.

5. FIT and procedural meaning

In section 4, I have argued that what Banfield calls representation can be distinguished from an act of communication performed by a speaker who wishes to communicate his own thoughts. The latter communicates a guarantee of the relevance of an interpretation which the hearer assumes is an interpretation of the speaker's thoughts. The act of representation communicates a guarantee of the relevance of an interpretation which the reader will take to be an interpretation of the thoughts of someone other than the speaker – an individual whose identity must be established on the basis of pragmatic inference. This interpretation is unmediated by the thoughts of the speaker/writer responsible for producing the representation. In this section, I will show how this distinction allows us to account for the use of expressions which, according to relevance theory, do not occur as constituents of thoughts, in the representations of the thoughts of non-communicating subjects.