Relevance Theory; Dan Sperber

Cognition. Cognitive Science. Ostensive-inferential human communication. Comprehension. Sub-tasks in the overall. Experimental methods of cognitive psychology

  • Enviado por: Vio
  • Idioma: inglés
  • País: España España
  • 32 páginas

RELEVANCE THEORY. deirdre wilson and dan sperber

1. Introduction

Relevance theory may be seen as an attempt to work out in detail one of Grice's central claims: that an essential feature of most human communication, both verbal and non-verbal, is the expression and recognition of intentions (Grice 1957; 1967; 1969; 1982; 1989a: 349-368). In developing this claim, Grice laid the foundations for an inferential model of communication, an alternative to the classical code model. According to the code model, a communicator encodes her intended message into a signal, which is decoded by the audience using an identical copy of the code. According to the inferential model, a communicator provides evidence of her intention to convey a certain meaning, which is inferred by the audience on the basis of the evidence provided. An utterance is, of course, a linguistically-coded piece of evidence, so that verbal comprehension involves an element of decoding. However, the linguistic meaning recovered by decoding is just one of the inputs to a non-demonstrative inference process which yields an interpretation of the speaker's meaning.[1]

The goal of inferential pragmatics is to explain how the hearer infers the speaker's meaning on the basis of the evidence provided. The relevance-theoretic account is based on another of Grice's central claims: that utterances automatically create expectations which guide the hearer towards the speaker's meaning. Grice described these expectations in terms of a Co-operative Principle and maxims of Quality, Quantity, Relation and Manner (truthfulness, informativeness, relevance and clarity) which speakers are expected to observe (Grice 1961; 1967; 1989a: 368-72): the interpretation a rational hearer should choose is the one that best satisfies those expectations. Relevance theorists share Grice's intuition that utterances raise expectations of relevance, but question several other aspects of his account, including the need for a Co-operative Principle and maxims, the restriction of pragmatics to the study of implicatures, the role of deliberate maxim violation in utterance interpretation, and the treatment of figurative utterances as deviations from a maxim or convention of truthfulness.[2] The central claim of relevance theory is that the expectations of relevance raised by an utterance are precise enough, and predictable enough, to guide the hearer towards the speaker's meaning. The aim is to explain in cognitively realistic terms what these expectations of relevance amount to, and how they might contribute to an empirically plausible account of comprehension. The theory has developed in several stages. A detailed version was published in Relevance: Communication and Cognition (Sperber & Wilson 1986a/95; 1987a,b) and updated in Sperber & Wilson 1995; 1998a; 2002; Wilson & Sperber 2002. Here, we will outline the main assumptions of the current version of the theory and discuss some of its implications for pragmatics.


2. Relevance and cognition

What sort of things may be relevant? Intuitively, relevance is a potential property not only of utterances and other observable phenomena, but of thoughts, memories and conclusions of inferences. In relevance-theoretic terms, any external stimulus or internal representation which provides an input to cognitive processes may be relevant to an individual at some time. According to relevance theory, utterances raise expectations of relevance not because speakers are expected to obey a Co-operative Principle and maxims or some other specifically communicative convention, but because the search for relevance is a basic feature of human cognition, which communicators may exploit. In this section, we will introduce the basic cognitive notion of relevance and the Cognitive Principle of Relevance, which lay the foundation for the relevance-theoretic approach to pragmatics.

When is an input relevant? Intuitively, an input (a sight, a sound, an utterance, a memory) is relevant to an individual when it connects with background information he has available to yield conclusions that matter to him: say, by answering a question he had in mind, improving his knowledge on a certain topic, settling a doubt, confirming a suspicion, or correcting a mistaken impression. In relevance-theoretic terms, an input is relevant to an individual when its processing in a context of available assumptions yields a positive cognitive effect. A positive cognitive effect is a worthwhile difference to the individual's representation of the world - a true conclusion, for example. False conclusions are not worth having. They are cognitive effects, but not positive ones (Sperber & Wilson 1995: 3.1-3.2).

The most important type of cognitive effect achieved by processing an input in a context is a contextual implication, a conclusion deducible from the input and the context together, but from neither input nor context alone. For example, on seeing my train arriving, I might look at my watch, access my knowledge of the train timetable, and derive the contextual implication that my train is late (which may itself achieve relevance by combining with further contextual assumptions to yield further implications). Other types of cognitive effect include the strengthening, revision or abandonment of available assumptions. For example, the sight of my train arriving late might confirm my impression that the service is deteriorating, or make me alter my plans to do some shopping on the way to work. According to relevance theory, an input is relevant to an individual when, and only when, its processing yields such positive cognitive effects.[3]

Intuitively, relevance is not just an all-or-none matter but a matter of degree. There is no shortage of potential inputs which might have at least some relevance for us, but we cannot attend to them all. Relevance theory claims that what makes an input worth picking out from the mass of competing stimuli is not just that it is relevant, but that it is more relevant than any alternative input available to us at that time. Intuitively, other things being equal, the more worthwhile conclusions achieved by processing an input, the more relevant it will be. In relevance-theoretic terms, other things being equal, the greater the positive cognitive effects achieved by processing an input, the greater its relevance will be. Thus, the sight of my train arriving one minute late may make little worthwhile difference to my representation of the world, while the sight of it arriving half an hour late may lead to a radical reorganisation of my day, and the relevance of the two inputs will vary accordingly.

What makes an input worth picking out from the mass of competing stimuli is not just the cognitive effects it achieves. In different circumstances, the same stimulus may be more or less salient, the same contextual assumptions more or less accessible, and the same cognitive effects easier or harder to derive. Intuitively, the greater the effort of perception, memory and inference required, the less rewarding the input will be to process, and hence the less deserving of our attention. In relevance-theoretic terms, other things being equal, the greater the processing effort required, the less relevant the input will be. Thus, relevance may be assessed in terms of cognitive effects and processing effort:


(1) Relevance of an input to an individual

a. Other things being equal, the greater the positive cognitive effects achieved by processing an input, the greater the relevance of the input to the individual at that time.

b. Other things being equal, the greater the processing effort expended, the lower the relevance of the input to the individual at that time.


Here is a brief and artificial illustration of how the relevance of alternative inputs might be compared in terms of effort and effect. Mary, who dislikes most meat and is allergic to chicken, rings her dinner party host to find out what is on the menu. He could truly tell her any of three things:


(2) We are serving meat.

(3) We are serving chicken.

(4) Either we are serving chicken or (72 - 3) is not 46.


According to the characterisation of relevance in (1), all three utterances would be relevant to Mary, but (3) would be more relevant than either (2) or (4). It would be more relevant than (2) for reasons of cognitive effect: (3) entails (2), and therefore yields all the conclusions derivable from (2), and more besides. It would be more relevant than (4) for reasons of processing effort: although (3) and (4) are logically equivalent, and therefore yield exactly the same cognitive effects, these effects are easier to derive from (3) than from (4), which requires an additional effort of parsing and inference (in order to work out that the second disjunct is false and the first is therefore true). Thus, (3) would be the most relevant utterance to Mary, for reasons of both effort and effect. More generally, when similar amounts of effort are required, the effect factor is decisive in determining degrees of relevance, and when similar amounts of effect are achievable, the effort factor is decisive.

This characterisation of relevance is comparative rather than quantitative: it makes clear comparisons possible in some cases (e.g. (2)-(4)), but not in all. While quantitative notions of relevance might be worth exploring from a formal point of view[4], it is the comparative rather than the quantitative notion that is likely to provide the best starting point for constructing a psychologically plausible theory. In the first place, it is highly unlikely that individuals have to compute numerical values for effort and effect when assessing relevance “from the inside”. Such computation would itself be effort-consuming and therefore detract from relevance. Moreover, even when individuals are clearly capable of computing numerical values (for weight or distance, for example), they generally have access to more intuitive methods of assessment which are comparative rather than quantitative, and which are in some sense more basic. In the second place, while some aspects of human cognitive processes can already be measured “from the outside” (e.g. processing time) and others may be measurable in principle (e.g. number of contextual implications), it is quite possible that others are not measurable at all (e.g. strength of implications, level of attention). As noted in Relevance (124-32), it therefore seems preferable to treat effort and effect as non-representational dimensions of mental processes: they exist and play a role in cognition whether or not they are mentally represented; and when they are mentally represented, it is in the form of intuitive comparative judgements rather than absolute numerical ones. The same is true of relevance, which is a function of effort and effect.[5], [6]

Given the characterisation of relevance in (1), aiming to maximise the relevance of the inputs one processes is simply a matter of making the most efficient use of the available processing resources. No doubt this is something we would all want to do, given a choice. Relevance theory claims that humans do have an automatic tendency to maximise relevance, not because we have a choice in the matter - we rarely do - but because of the way our cognitive systems have evolved. As a result of constant selection pressure towards increasing efficiency, the human cognitive system has developed in such a way that our perceptual mechanisms tend automatically to pick out potentially relevant stimuli, our memory retrieval mechanisms tend automatically to activate potentially relevant assumptions, and our inferential mechanisms tend spontaneously to process them in the most productive way. Thus, while we are all likely to notice the sound of glass breaking in our vicinity, we are likely to attend to it more, and process it more deeply, when our memory and inference mechanisms identify it as the sound of our glass breaking, and compute the consequences that are likely to be most worthwhile for us. This universal tendency is described in the First, or Cognitive, Principle of Relevance (Sperber & Wilson 1995, section 3.1-3.2):


(5) Cognitive Principle of Relevance

Human cognition tends to be geared to the maximisation of relevance.


It is against this cognitive background that inferential communication takes place.


3. Relevance and communication

The universal cognitive tendency to maximise relevance makes it at possible, at least to some extent, to predict and manipulate the mental states of others. Knowing of your tendency to pick out the most relevant stimuli in your environment and process them so as to maximise their relevance, I may be able to produce a stimulus which is likely to attract your attention, to prompt the retrieval of certain contextual assumptions and to point you towards an intended conclusion. For example, I may leave my empty glass in your line of vision, intending you to notice and conclude that I might like another drink. As Grice pointed out, this is not yet a case of inferential communication because, although I did intend to affect your thoughts in a certain way, I gave you no evidence that I had this intention. Inferential communication is not just a matter of intending to affect the thoughts of an audience; it is a matter of getting them to recognise that one has this intention. When I quietly leave my glass in your line of vision, I am not engaging in inferential communication, but merely exploiting your natural cognitive tendency to maximise relevance.

Inferential communication - what relevance theory calls ostensive-inferential communication for reasons that will shortly become apparent - involves an extra layer of intention:


(6) Ostensive-inferential communication

a. The informative intention:

The intention to inform an audience of something.

b. The communicative intention:

The intention to inform the audience of one's informative intention.[7]


Understanding is achieved when the communicative intention is fulfilled - that is, when the audience recognises the informative intention. (Whether the informative intention itself is fulfilled depends on how much the audience trusts the communicator. There is a gap between understanding and believing. For understanding to be achieved, the informative intention must be recognised, but it does not have to be fulfilled.)

How does the communicator indicate to the audience that she is trying to communicate with them in this overt, intentional way? Instead of covertly leaving my glass in your line of vision, I might touch your arm and point to my empty glass, wave it at you, ostentatiously put it down in front of you, stare at it meaningfully, or say “My glass is empty”. More generally, ostensive-inferential communication involves the use of an ostensive stimulus, designed to attract an audience's attention and focus it on the communicator's meaning. Relevance theory claims that use of an ostensive stimulus may create precise and predictable expectations of relevance not raised by other stimuli. In this section, we will describe these expectations and show how they may help the audience to identify the communicator's meaning.

The fact that ostensive stimuli create expectations of relevance follows from the definition of an ostensive stimulus and the Cognitive Principle of Relevance. An ostensive stimulus is designed to attract the audience's attention. Given the universal tendency to maximise relevance, an audience will only pay attention to a stimulus that seems relevant enough. By producing an ostensive stimulus, the communicator therefore encourages her audience to presume that it is relevant enough to be worth processing. This is not a case of Gricean co-operation. Even a deceptive communicator manifestly intends her audience to assume that her stimulus is relevant enough to be worth processing - why else would he pay attention? This is the basis for the Second, or Communicative, Principle of Relevance, which applies specifically to ostensive-inferential communication:


(7) Communicative Principle of Relevance

Every ostensive stimulus conveys a presumption of its own optimal relevance.


The Communicative Principle of Relevance and the notion of optimal relevance are the key to relevance-theoretic pragmatics.

An ostensive stimulus, then, creates a presumption of relevance. The notion of optimal relevance is meant to spell out what the audience of an act of ostensive communication is entitled to expect in terms of effort and effect:


(8) Optimal relevance

An ostensive stimulus is optimally relevant to an audience iff:

a. It is relevant enough to be worth the audience's processing effort;

b. It is the most relevant one compatible with communicator's abilities and preferences.


According to clause (a) of this definition of optimal relevance, the audience is entitled to expect the ostensive stimulus to be at least relevant enough to be worth processing. Given the argument of the last section that a stimulus is worth processing only if it is more relevant than any alternative input available at the time, this is not a trivial claim. Indeed, in order to satisfy the presumption of relevance conveyed by an ostensive stimulus, the audience may have to draw stronger conclusions than would otherwise have been warranted. For example, if you just happen to notice my empty glass, you may be entitled to conclude that I might like a drink. If I deliberately wave it about in front of you, you would generally be justified in drawing the stronger conclusion that I would like a drink.

According to clause (b) of the definition of optimal relevance, the audience of an ostensive stimulus is entitled to even higher expectations than this. The communicator wants to be understood. It is therefore in her interest - within the limits of her own capabilities and preferences - to make her ostensive stimulus as easy as possible for the audience to understand, and to provide evidence not just for the cognitive effects she aims to achieve in her audience but also for further cognitive effects which, by holding his attention, will help her achieve her goal. For instance, the communicator's goal might be to inform her audience that she has begun writing her paper. It may be effective for her, in pursuit of this goal, to volunteer more specific information and say, “I have already written a third of the paper.” In the circumstances, her audience would then be entitled to understand her as saying that she has she has written only a third of the paper, for if she had written two thirds (say), she would normally be expected to say so, given clause (b) of the definition of optimal relevance.

Communicators, of course, are not omniscient, and they cannot be expected to go against their own interests and preferences in producing an utterance. There may be relevant information that they are unable or unwilling to provide, and ostensive stimuli that would convey their intentions more economically, but that they are unwilling to produce, or unable to think of at the time. All this is allowed for in clause (b) of the definition of optimal relevance, which states that the ostensive stimulus is the most relevant one (i.e. yielding the greatest effects, in return for the smallest processing effort) that the communicator is willing and able to produce (see Sperber & Wilson 1995: section 3.3 and 266-78).

This approach sheds light on some cases where a communicator withholds relevant information, and which seem to present problems for Grice. Suppose I ask you a question and you remain silent, for example. Silence in these circumstances may or may not be an ostensive stimulus. When it is not, we would naturally take it as indicating that the addressee was unable or unwilling to answer the question. If you are clearly willing to answer, I am entitled to conclude that you are unable, and if you are clearly able to answer, I am entitled to conclude that you are unwilling. When the silence is ostensive, we would like to be able to analyse it as merely involving an extra layer of intention, and hence as communicating - or implicating - that the addressee is unable or unwilling to answer. Given the presumption of relevance and the definition of optimal relevance in (8), this is possible in the relevance-theoretic framework.[8] In Grice's framework, by contrast, the co-operative communicator's willingness to provide any required information is taken for granted, and the parallels between ostensive and non-ostensive silences are lost. On a Gricean account, violation of the first Quantity maxim (“Make your contribution as informative as required”) is invariably attributed to the communicator's inability - rather than unwillingness - to provide the required information. Unwillingness to co-operate is a violation of the Co-operative Principle, and suspension of the Co-operative Principle should make it impossible to convey any conversational implicatures at all.[9] We have argued that, although much communication is co-operative in the sense that the communicator is willing to provide the required information, co-operation in this sense is not essential for communication, as it is for Grice (Wilson & Sperber 1981; Smith & Wilson 1992; Sperber 1994).

This relevance-theoretic account of cognition and communication has practical implications for pragmatics. As noted above, verbal comprehension starts with the recovery of a linguistically-encoded sentence meaning, which must be contextually enriched in a variety of ways to yield a full-fledged speaker's meaning. There may be ambiguities and referential ambivalences to resolve, ellipses to interpret, and other underdeterminacies of explicit content to deal with.[10] There may be implicatures to identify, illocutionary indeterminacies to resolve, metaphors and ironies to interpret. All this requires an appropriate set of contextual assumptions, which the hearer must also supply. The Communicative Principle of Relevance and the definition of optimal relevance suggest a practical procedure for performing these subtasks and constructing a hypothesis about the speaker's meaning. The hearer should take the linguistically encoded sentence meaning; following a path of least effort, he should enrich it at the explicit level and complement it at the implicit level until the resulting interpretation meets his expectation of relevance:


(9) Relevance-theoretic comprehension procedure

a. Follow a path of least effort in computing cognitive effects: Test interpretive hypotheses (disambiguations, reference resolutions, implicatures, etc.) in order of accessibility.

b. Stop when your expectations of relevance are satisfied.


Given clause (b) of the definition of optimal relevance in (8), it is reasonable for the hearer to follow a path of least effort because the speaker is expected (within the limits of her abilities and preferences) to make her utterance as easy as possible to understand. Since relevance varies inversely with effort, the very fact that an interpretation is easily accessible gives it an initial degree of plausibility. It is also reasonable for the hearer to stop at the first interpretation that satisfies his expectations of relevance, because there should never be more than one. A speaker who wants her utterance to be as easy as possible to understand should formulate it (within the limits of her abilities and preferences) so that the first interpretation to satisfy the hearer's expectation of relevance is the one she intended to convey.[11] An utterance with two apparently satisfactory competing interpretations would cause the hearer the unnecessary extra effort of choosing between them, and the resulting interpretation (if there were one) would not satisfy clause (b) of the definition of optimal relevance.[12]

Thus, when a hearer following the path of least effort arrives at an interpretation that satisfies his expectations of relevance, in the absence of contrary evidence, this is the most plausible hypothesis about the speaker's meaning. Since comprehension is a non-demonstrative inference process, this hypothesis may well be false; but it is the best a rational hearer can do.


4. Relevance and comprehension

In many non-verbal cases (e.g. pointing to one's empty glass, failing to respond to a question), use of an ostensive stimulus merely adds an extra layer of intention recognition to a basic layer of information that the audience might have picked up anyway. In other cases (e.g. inviting someone out to a drink by pretending to raise a glass to one's lips), the communicator's behaviour provides no direct evidence for the intended conclusion, and it is only the presumption of relevance conveyed by the ostensive stimulus which encourages the audience to devote the necessary processing resources to discovering her meaning. Either way, the range of meanings that can be non-verbally conveyed is necessarily limited by the range of concepts the communicator can evoke in her audience by drawing attention to observable features of the environment (whether preexisting or produced specifically for this purpose).

In verbal communication, speakers manage to convey a very wide range of meanings despite the fact that there is no independently-identifiable basic layer of information for the hearer to pick up. What makes it possible for the hearer to recognise the speaker's informative intention is that utterances encode logical forms (conceptual representations, however fragmentary or incomplete) which the speaker has manifestly chosen to provide as input to the hearer's inferential comprehension process. As a result, verbal communication can achieve a degree of explicitness not available in non-verbal communication (compare pointing in the direction of a table containing glasses, ashtrays, plates, etc., and saying, “My glass is empty”).

Although the decoded logical form of an utterance is an important clue to the speaker's intentions, it is now increasingly recognised that even the explicitly-communicated content of an utterance goes well beyond what is linguistically encoded.[13] Grice talked of his Co-operative Principle and maxims mainly in connection with the recovery of implicatures, and he seems to have thought of them as playing no significant role on the explicit side. His few remarks on disambiguation and reference assignment - which he saw as falling on the explicit rather than the implicit side - suggest that he thought of them as determined by sentence meaning and contextual factors alone, without reference to pragmatic principles or speakers' intentions,[14] and many pragmatists have followed him on this. There has thus been a tendency, even in much of the recent pragmatic literature, to treat the “primary” processes involved in the recovery of explicit content as significantly different from - i.e. less inferential, or less directly dependent on speakers' intentions or pragmatic principles than - the “secondary” processes involved in the recovery of implicatures.[15]

Relevance theory treats the identification of explicit content as equally inferential, and equally guided by the Communicative Principle of Relevance, as the recovery of implicatures. The relevance-theoretic comprehension procedure (“Follow a path of least effort in computing cognitive effects: test interpretive hypotheses in order of accessibility, and stop when your expectations of relevance are satisfied.”) applies in the same way to the resolution of linguistic underdeterminacies at both explicit and implicit levels. The hearer's goal is to construct a hypothesis about the speaker's meaning which satisfies the presumption of relevance conveyed by the utterance. This overall task can be broken down into a number of sub-tasks:


(10) Sub-tasks in the overall comprehension process

a. Constructing an appropriate hypothesis about explicit content (in relevance-theoretic terms, explicatures) via decoding, disambiguation, reference resolution, and other pragmatic enrichment processes.

b. Constructing an appropriate hypothesis about the intended contextual assumptions (in relevance-theoretic terms, implicated premises).

c. Constructing an appropriate hypothesis about the intended contextual implications (in relevance-theoretic terms, implicated conclusions).


These sub-tasks should not be thought of as sequentially ordered. The hearer does not first decode the logical form of the sentence uttered, then construct an explicature and select an appropriate context, and then derive a range of implicated conclusions. Comprehension is an on-line process, and hypotheses about explicatures, implicated premises and implicated conclusions are developed in parallel against a background of expectations (or anticipatory hypotheses) which may be revised or elaborated as the utterance unfolds.[16] In particular, the hearer may bring to the comprehension process not only a general presumption of relevance, but more specific expectations about how the utterance will be relevant to him (what cognitive effects it is likely to achieve), and these may contribute, via backwards inference, to the identification of explicatures and implicated premises.[17] Thus, each sub-task in (10a-c) above involves a non-demonstrative inference process embedded within the overall process of constructing a hypothesis about the speaker's meaning.

To take just one illustration, consider the exchange in (11):


(11) a. Peter: Did John pay back the money he owed you?

b. Mary: No. He forgot to go to the bank.


Here is a schematic outline of how Peter might use the relevance-theoretic comprehension procedure to construct hypotheses about the explicatures and implicatures of Mary's utterance, “He forgot to go to the bank”:



(a) Mary has said to Peter, “Hex forgot to go to the BANK1 / BANK 2.”

[Hex = uninterpreted pronoun]
[BANK1 = financial institution]
[BANK2 = river bank]

Embedding of the decoded (incomplete) logical form of Mary's utterance into a description of Mary's ostensive behaviour.

(b) Mary's utterance will be optimally relevant to Peter.


Expectation raised by recognition of Mary's ostensive behaviour and acceptance of the presumption of relevance it conveys.


(c) Mary's utterance will achieve relevance by explaining why John has not repaid the money he owed her.

Expectation raised by (b), together with the fact that such an explanation would be most relevant to Peter at this point


(d) Forgetting to go to the BANK1 may make one unable to repay the money one owes.


First assumption to occur to Peter which, together with other appropriate premises, might satisfy expectation (c). Accepted as an implicit premise of Mary's utterance.


(e) John forgot to go to the BANK1.


First enrichment of the logical form of Mary's utterance to occur to Peter which might combine with (d) to lead to the satisfaction of (c). Accepted as an explicature of Mary's utterance.


(f) John was unable to repay Mary the money he owes because he forgot to go to the BANK1.

Inferred from (d) and (e), satisfying (c) and accepted as an implicit conclusion of Mary's utterance


(g) John may repay Mary the money he owes when he next goes to the BANK1.

From (f) plus background knowledge. One of several possible weak implicatures of Mary's utterance which, together with (f), satisfy expectation (b)




Peter assumes in (12b) that Mary's utterance, decoded as in (12a), is optimally relevant to him. Since what he wants to know at this point is why John did not repay the money he owed, he assumes in (c) that Mary's utterance will achieve relevance by answering this question. In the situation described, the logical form of the utterance provides easy access to the contextual assumption in (d) (that forgetting to go to the bank may prevent one from repaying money one owes). This could be used as an implicit premise in deriving the expected explanation of John's behaviour, provided that the utterance is interpreted on the explicit side (via disambiguation and reference resolution) as conveying the information in (e): that John forgot to go to the BANK1. By combining the implicit premise in (d) and the explicit premise in (e), Peter arrives at the implicit conclusion in (f), from which further, weaker implicatures, including (g) and others, can be derived. The resulting overall interpretation satisfies Peter's expectations of relevance. On this account, explicatures and implicatures (i.e. implicit premises and conclusions) are arrived at by a process of mutual parallel adjustment, with hypotheses about both being considered in order of accessibility.[18]

This schematic outline of the comprehension process is considerably oversimplified.[19] In particular, it omits a range of lexical-pragmatic processes which contribute in important ways to the construction of explicatures. Consider the word bank in (11b). In interpreting this utterance, Peter would probably take Mary to be referring not just to a banking institution but to a specific type of banking institution: one that deals with private individuals, and in particular, with John. Unless the denotation of bank is narrowed in this way, the explicit content of Mary's utterance will not warrant the conclusion in (12f), which is needed to satisfy Peter's expectation of relevance. (It is hard to see how the fact that John had forgotten to go to the World Bank, say, or the European Investment Bank, might explain his failure to repay the money he owed.) By the same token, in interpreting the phrase go to the bank, he would take Mary to be referring not merely to visiting the bank but to visiting it in order to get money, and, moreover, to get money in the regular way (legally, rather than, say, by robbing the bank). Unless the explicit content of the utterance is narrowed in this way, it will not warrant the conclusion in (12f), which is needed to satisfy Peter's expectation of relevance.

Some of these stereotypical narrowings have been described in the pragmatic literature as generalised conversational implicatures, and analysed as default interpretations, derivable via default rules.[20] Despite the richness and subtlety of much of the literature on generalised conversational implicature, relevance theory takes a different approach, for two main reasons. In the first place, as noted above, it treats lexical narrowing as a type of pragmatic enrichment process which contributes to explicatures rather than implicatures.[21] Like all enrichment processes, lexical narrowing is driven by the search for relevance, which involves the derivation of cognitive effects, and in particular of contextual implications. By definition, a contextual implication must follow logically from the explicatures of the utterance and the context. Sometimes, as in (11b), in order to yield an expected implication, the explicit content of the utterance must be enriched to a point where it warrants the expected conclusion. In any framework where implicated conclusions are seen as logically warranted by explicit content, there is thus good reason to treat lexical narrowings as falling on the explicit rather than the implicit side.[22]

In the second place, lexical narrowing is a much more flexible and context-dependent process than appeals to generalised implicature or default interpretations suggest. Barsalou (1989, 1992) surveys experimental evidence which suggests that even apparently stereotypical narrowings of terms such as bird, animal, furniture, food, etc. vary considerably across situations, individuals and times, and are strongly affected by discourse context and considerations of relevance. In Barsalou's view, his results are best explained by the assumption that lexical items give access not to ready-made prototypes but to a vast array of encyclopaedic information which varies in accessibility from occasion to occasion, with different subsets being selected ad hoc to determine the occasion-specific interpretation of a word. On this approach, bank in (11b) might be understood as conveying not the encoded concept BANK1 but the related concept BANK*, with a more restricted encyclopedic entry and a narrower denotation, constructed ad hoc for this particular occasion.

In Barsalou's view, the construction of ad hoc concepts is affected by a variety of factors, including context, the accessibility of encyclopedic assumptions and considerations of relevance. However, he makes no concrete proposal about how these concepts might be derived, and in particular about how the construction process is triggered and when it stops. The relevance-theoretic comprehension procedure may be seen as a concrete hypothesis about how such a flexible, relevance-governed lexical interpretation process might go. The hearer treats the linguistically-encoded word meaning (e.g. BANK1 in (11b)) as no more than a clue to the speaker's meaning. Guided by his expectations of relevance, and using contextual assumptions made accessible by the encyclopedic entry of the linguistically-encoded concept (e.g. that forgetting to go to the bank where one keeps one's money may make one unable to repay money one owes), he starts deriving cognitive effects. When he has enough effects to satisfy his expectations of relevance, he stops. The results would be as in (12) above, except that the contextual assumption in (d), the explicature in (e) and the implicatures in (f) and (g) would contain not the encoded concept BANK1 but the ad hoc concept BANK*, with a narrower denotation, which would warrant the derivation of the cognitive effects required to satisfy the hearer's expectations of relevance.

The effect of such a flexible interpretation process may be a loosening rather than a narrowing of the encoded meaning. This is another way in which lexical pragmatic processes differ from default, or stereotypical, narrowing. Clear cases of loosening include extending a prominent brand name (e.g. Hoover , Xerox, Kleenex) to members of less prominent brands; other good examples are approximations based on well-defined terms such as square, painless or silent, but the phenomenon is very widespread. Consider bank in (11b). Given current banking practice, the word is often loosely used to apply not only to banking institutions but also to the cash dispensers found in supermarkets and stations. Indeed, in order to satisfy his expectations of relevance in (11b), Peter would probably have to take it this way. (If John regularly gets his money from a cash dispenser, the claim that he forgot to go to the BANK1, might be strictly speaking false, and in any case would not adequately explain his failure to repay Mary.) Thus, bank in (11b) might be understood as expressing not the encoded concept BANK1, but an ad hoc concept BANK**, which shares with BANK1 the salient encyclopedic attribute of being a place one goes to in order to access money from one's account. The interpretation of a quite ordinary utterance such as (11b) might then involve both a loosening and a narrowing of the encoded meaning.

Loose uses of language present a problem for Grice's framework. Strictly speaking, faces are not square, rooms are generally not silent, and to describe them as such would violate his maxim of truthfulness (“Do not say what you believe to be false”). However, these departures from truthfulness do not fall into any of the categories of maxim-violation recognised by Grice (Grice 1989b: 30). They are not covert violations, like lies, designed to deceive the hearer into believing what was said. They are not like jokes and fictions, which suspend the maxims entirely. Given their intuitive similarities to metaphor and hyperbole, it might be tempting to analyse them, like tropes, as overt violations (floutings) of the maxim of truthfulness, designed to trigger the search for a related implicature (in this case, a hedged version of what was said). The problem is that these loose uses of language would not be generally perceived as violating the maxim of truthfulness at all. They do not have the striking quality that Grice associated with floutings, and which he saw as resulting in figurative or quasi-figurative interpretations. While we are all capable of realising on reflection that they are not strictly and literally true, these departures from truthfulness pass unattended and undetected in the normal flow of discourse. Grice's framework thus leaves them unexplained.[23]

Loose uses of language are not the only problem for Grice's maxim of truthfulness. There are questions about how the maxim itself is to be understood, and a series of difficulties with the analysis of tropes as overt violations of the maxim (for detailed discussion, see Wilson & Sperber 2002). Notice, too, that the intuitive similarities between loose talk, metaphor and hyperbole cannot be captured within this framework, since metaphor and hyperbole are seen as overt violations of the maxim of truthfulness, while loose uses of language are not. We have argued that the best solution is to abandon the maxim of truthfulness and treat whatever expectations of truthfulness arise in utterance interpretation as resulting not from an independent maxim, norm or convention of truthfulness, but as by-products of the more basic expectation of relevance. On this approach, loose talk, metaphor and hyperbole involve no violation of any maxim, but are merely alternative routes to achieving optimal relevance. Whether an utterance is literally, loosely or metaphorically understood will depend on the mutual adjustment of context, context and cognitive effects in the effort to satisfy the hearer's overall expectation of relevance[24]

To illustrate this unified approach, consider the exchange in (13):


(13) a. Peter: What do you think of Martin's latest novel?

b. Mary: It puts me to sleep.


In Grice's framework, Mary's utterance in (13b) should have three distinct interpretations: as a literal assertion, a hyperbole or a metaphor.[25] Of these, Peter should test the literal interpretation first, and move to a figurative interpretation only if the literal interpretation blatantly violates the maxim of truthfulness. Yet there is now a lot of experimental evidence suggesting that literal interpretations do not have to be tested and rejected before figurative interpretations are considered;[26] indeed, in interpreting (13b), it would probably not even occur to Peter to wonder whether Mary literally fell asleep.

The relevance-theoretic analysis takes these points into account. In the first place, there is no suggestion that the literal meaning must be tested first. As with bank in (11b), the encoded conceptual address is treated merely as a point of access to an ordered array of encyclopedic information from which the hearer is expected to select in constructing a satisfactory overall interpretation. Whether this interpretation is literal or loose will depend on which types of information he selects. In processing (13b), Peter will be expecting to derive an answer to his question: that is, an evaluation of the book. In the circumstances, the first contextual assumption to occur to him is likely to be that a book which puts one to sleep is extremely boring and unengaging. Having used this assumption to derive an answer to his question, thus satisfying his expectations of relevance, he should stop. Just as in interpreting bank in (11b), it does not occur to him to wonder whether John gets his money from a bank or a cash dispenser, so in interpreting (13b), it should not occur to him to wonder whether the book literally puts Mary to sleep or merely bores her greatly. Just as the mutual adjustment process in (13) yields an explicature containing the ad hoc concept BANK**, which has undergone simultaneous narrowing and loosening, so the mutual adjustment process for (13b) should yield an explicature containing the ad hoc concept PUT TO SLEEP*, which denotes not only literal cases of putting to sleep, but other cases that share with it the encyclopedic property of being extremely boring and unengaging. Only if such a loose interpretation fails to satisfy his expectations of relevance would Peter be justified in spending the effort required to explore further contextual assumptions, and moving towards a more literal interpretation.[27]

Typically, the explicit content of loose uses in general, of metaphors in particular, exhibits a certain degree of indeterminacy. Compare, for instance, the results of using the word square literally in a geometric statement to convey the concept SQUARE, using it loosely in the phrase a square face to convey the concept SQUARE*, and using it metaphorically in the phrase a square mind to convey the concept SQUARE**. In relevance theory, this relative indeterminacy of explicatures is linked to the relative strength of implicatures.

A proposition may be more or less strongly implicated by an utterance. It is strongly implicated (or is a strong implicature) if its recovery is essential in order to arrive at an interpretation that satisfies the expectations of relevance raised by the utterance itself. It is weakly implicated if its recovery helps with the construction of an interpretation that is relevant in the expected way, but is not itself essential because the utterance suggests a range of similar possible implicatures, any one of which would do (Relevance, chapter 1, sections 10-12, chapter 4, section 6). For instance, (11b), “He forgot to go to the bank”, strongly implicates (12f), John was unable to repay Mary the money he owes because he forgot to go to the BANK1, since without this implication, (11b) is not a relevant reply to (11a), “Did John pay back the money he owed you?”. (11b) also encourages the audience to derive a further implicature along the lines of (12g), John may repay Mary the money he owes when he next goes to the BANK1, but here the audience must take some responsibility for coming to this conclusion rather than, say, the conclusion that John will repay Mary the money he owes when he next goes to the BANK1, or some other similar conclusion.

Typically, loose uses, and metaphorical uses in particular, convey an array of weak implicatures. Thus, the utterance, “John has a square mind”, weakly implicates that he is somewhat rigid in his thinking, that he does not easily change his mind, that he is a man of principle, and so on. None of these implicatures is individually required for the utterance to make sense, but, on the other hand, without some such implicatures, it will make no sense at all. If the word square is understood as conveying the concept SQUARE**, which combines with contextual information to yield these implications, then the concept SQUARE** itself will exhibit some indeterminacy or fuzziness, and the utterance as a whole will exhibit a corresponding weakness of explicature. Loose uses and metaphors typically exhibit such fuzziness, for which relevance theory provides an original account.

The distinction between strong and weak implicatures sheds some light on the variety of ways in which an utterance can achieve relevance. Some utterances, (technical instructions, for instance) achieve relevance by conveying a few strong implicatures. Other utterances achieve relevance by weakly suggesting a wide array of possible implications, each of which is a weak implicature of the utterance. This is typical of poetic uses of language, and has been discussed in relevance theory under the heading of poetic effect (Sperber & Wilson 1986a/1995, chapter 4, sections 6-9; Pilkington 2000).

In Grice's framework (and indeed in all rhetorical and pragmatic discussions of irony as a figure of speech before Sperber & Wilson 1981) the treatment of verbal irony parallels the treatments of metaphor and hyperbole. For Grice, irony is an overt violation of the maxim of truthfulness, and differs from metaphor and hyperbole only in the kind of implicature it conveys (metaphor implicates a simile based on what was said, hyperbole implicates a weakening of what was said, and irony implicates the opposite of what was said). Relevance theory rejects not only the Gricean analysis of irony but the more general assumption that metaphor, hyperbole and irony should be given parallel treatments.

Grice's analysis of irony as an overt violation of the maxim of truthfulness is a variant of the classical rhetorical view of irony as literally saying one thing and figuratively meaning the opposite. There are well-known arguments against this view. It is descriptively inadequate because ironical understatements, ironical quotations and ironical allusions cannot be analysed as communicating the opposite of what is literally said. It is theoretically inadequate because saying the opposite of what one means is patently irrational, and on this approach it is hard to explain why verbal irony is universal and appears to arise spontaneously, without being taught or learned (Sperber & Wilson 1981; 1998b; Wilson & Sperber 1992).

Moreover, given the relevance-theoretic analysis of metaphor and hyperbole as varieties of loose use, the parallelism between metaphor, hyperbole and irony cannot be maintained. While it is easy to see how a speaker aiming at optimal relevance might convey her meaning more economically by speaking loosely rather than using a cumbersome literal paraphrase, it is hard to see how a rational speaker could hope to convey her meaning more economically by choosing a word whose encoded meaning is the opposite of the one she intends to convey (or how a hearer using the relevance-theoretic comprehension procedure could understand her if she did). Some alternative explanation of irony must be found.

According to the explanation proposed by relevance theory, verbal irony involves no special machinery or procedures not already needed to account for a basic use of language, interpretive use, and a specific form of interpretive use, echoic use. An utterance may be interpretively used to (meta)represent another utterance or thought that it resembles in content. The best-known type of interpretive use is reported speech or thought. An utterance is echoic when it achieves most of its relevance not by expressing the speaker's own views, nor by reporting someone else's utterances or thoughts, but by expressing the speaker's attitude to views she tacitly attributes to someone else.[28] To illustrate, suppose that Peter and Mary are leaving a party, and one of the following exchanges occurs:


(14) Peter: That was a fantastic party.

(15) Mary: a. [happily] Fantastic.

b. [puzzled] Fantastic?

c. [scornfully] Fantastic!


In (15a), Mary echoes Peter's utterance in order to indicate that she agrees with it; in (15b), she indicates that she is wondering about it; and in (15c) she indicates that she disagrees with it. The resulting interpretations might be as in (16):


(16) a. She believes I was right to say/think that the party was fantastic.

b. She is wondering whether I was right to say/think that the party was fantastic.

c. She believes I was wrong to say/think that the party was fantastic.


Here, the basic proposition expressed by the utterances in (15) (that the party was fantastic) is embedded under an appropriate higher-order speech-act or propositional-attitude description indicating, on the one hand, that the basic proposition is being used to interpret views Mary attributes to someone else, and, on the other, Mary's attitude to these views. In order to understand Mary's meaning, Peter has to recognise not only the basic proposition expressed but the fact that it is being attributively used, and Mary's attitude to the attributed views.

The attitudes conveyed by use of an echoic utterance may be very rich and varied. The speaker may indicate that she endorses or dissociates herself from the thought or utterance she is echoing: that she is puzzled, angry, amused, intrigued, sceptical, and so on, or any combination of these. On the relevance-theoretic account, verbal irony involves the expression of a tacitly dissociative attitude - wry, sceptical, bitter or mocking - to an attributed utterance or thought. Consider Mary's utterance in (15c) above. This is clearly both ironical and echoic. Relevance theory claims that it is ironical because it is echoic: verbal irony consists in echoing a tacitly attributed thought or utterance with a tacitly dissociative attitude.[29]

This approach sheds light on many cases of irony not dealt with on the classical or Gricean accounts. Consider Mary's utterance “He forgot to go to the bank” in (11b) above. There are situations where this might well be ironical, even though it is neither blatantly false nor used to convey the opposite of what was said. Suppose Peter and Mary both know that John has repeatedly failed to repay Mary, with a series of pitifully inadequate excuses. Then (11b) may be seen as an ironical echo in which Mary tacitly dissociates herself from the latest excuse in the series. Thus, all that is needed to make (11b) ironical is a scenario in which it can be understood as a mocking echo of an attributed utterance or thought.[30]

One implication of this analysis is that irony involves a higher order of metarepresentational ability than metaphor. On the relevance-theoretic account, as illustrated in (16) above, the interpretation of echoic utterances in general involves the ability to recognise that the speaker is thinking, not directly about a state of affairs in the world, but about another thought or utterance that she attributes to someone else. This implication of our account is confirmed by experimental evidence from the literature on autism, child development and right hemisphere damage, which shows that the comprehension of irony correlates with second-order metarepresentational abilities, while the comprehension of metaphor requires only first-order abilities.[31] This fits straightforwardly with the relevance-theoretic account of irony, but is unexplained on the classical or Gricean accounts.[32]

Another area in which metarepresentational abilities play an important role is the interpretation of illocutionary acts. Consider the exchange in (17):


(17) a. Peter: Will you pay back the money by Tuesday?

b. Mary: I will pay it back by then.


Both (17a) and (17b) express the proposition that Mary will pay back the money by Tuesday. In the interrogative (17a), this proposition is expressed but not communicated (in the sense that Peter does not put it forward as true, or probably true)[33]: in relevance-theoretic terms, it is not an explicature of Peter's utterance. Yet intuitively, (17a) is no less explicit an act of communication than (17b). Relevance theory claims that what is explicitly communicated by (17a) is the higher-order speech-act description in (18):


(18) Peter is asking Mary whether she will pay back the money by Tuesday;


Like all explicatures, (18) is recovered by a mixture of decoding and inference based on a variety of linguistic and non-linguistic clues (e.g. word order, mood indicators, tone of voice, facial expression): in relevance-theoretic terms, it is a higher-level explicature of (17a).[34] In (17b), by contrast, the explicatures might include both (19a), the BASIC EXPLICATURE, and higher-level explicatures such as (19b) and (19c):


(19) a. Mary will pay back the money by Tuesday.

b. Mary is promising to pay back the money by Tuesday.

c. Mary believes she will pay back the money by Tuesday.


Thus, an utterance may convey several explicatures, each of which may contribute to relevance and warrant the derivation of implicatures.[35]

On this approach, verbal irony has more in common with illocutionary and attitudinal utterances than it does with metaphor or hyperbole. The recognition of irony, like the recognition of illocutionary acts and expressions of attitude, involves a higher order of metarepresentational ability than the recognition of the basic proposition expressed by an utterance, whether literal, loose or metaphorical. More generally, on both Gricean and relevance-theoretic accounts, the interpretation of every utterance involves a high degree of metarepresentational capacity, since comprehension rests on the ability to attribute both informative and communicative intentions. This raises the question of how pragmatic abilities are acquired, and how they fit into the overall architecture of the mind.


5. Relevance theory and mental architecture

Grice's analysis of overt communication as involving the expression and recognition of intentions treats comprehension as a variety of mind-reading, or theory of mind (the ability to attribute mental states to others in order to explain and predict their behaviour).[36] The link between mind-reading and communication is confirmed by a wealth of developmental and neuropsychological evidence.[37] However, mind-reading itself has been analysed in rather different ways. Philosophers often describe it as an exercise in reflective reasoning (a central thought process, in the terms of Fodor 1983), and many of Grice's remarks about pragmatics are consistent with this. Thus, his rational reconstruction of how conversational implicatures are derived is a straightforward exercise in “belief-desire” psychology, involving the application of general-purpose reasoning mechanisms to premises based on explicit hypotheses about the relations between mental states and behaviour:

  He said that P; he could not have done this unless he thought that Q; he knows (and knows that I know that he knows) that I will realise that it is necessary to suppose that Q; he has done nothing to stop me thinking that Q; so he intends me to think, or is at least willing for me to think, that Q. (Grice, 1989b: 30-31)


In our own early work, we also treated pragmatic interpretation as a central, inferential process (as opposed to part of a peripheral language module), albeit a spontaneous, intuitive rather than a conscious, reflective one (Relevance, chapter 2; Wilson & Sperber 1986). More recently, there has been a tendency in the cognitive sciences to move away from Fodor's sharp distinction between modular input processes and relatively undifferentiated central processes and towards an increasingly modular view of the mind.[38] In this section, we will consider how the relevance-theoretic comprehension procedure might fit with more modular accounts of inference, and in particular of mind-reading.[39]

One advantage of dedicated inferential mechanism or module is that it can take advantage of regularities in its own particular domain, and contain special-purpose inferential procedures which are justified by these regularities, but only in this domain. Thus, in modular accounts of mind-reading, standard “belief-desire” psychology may be replaced by special-purpose inferential procedures (“fast and frugal heuristics”, in the terms of Gigerenzer et al. 1999) attuned to the properties of this particular domain. Examples discussed in the literature on mind-reading include an Eye Direction Detector which attributes perceptual and attentional states on the basis of direction of gaze, and an Intentionality Detector which interprets self-propelled motion in terms of goals and desires (Leslie 1994; Premack & Premack 1994; Baron-Cohen1995). In mechanisms of this “fast and frugal” type, regularities in the relations between mental states and behaviour are not registered as explicit premises in an inference process, but function merely as tacit underpinnings for the working of the device.

Most approaches to mind-reading, whether modular or non-modular, have tended to take for granted that there is no need for special-purpose inferential comprehension procedures, because the mental-state attributions required for comprehension will be automatically generated by more general mind-reading mechanisms which apply across the whole domain.[40] We believe that there are serious problems with the view that speakers' meanings can be inferred from utterances by the same procedures used to infer intentions from actions. In the first place, the range of actions an agent can reasonably intend to perform in a given situation is in practice quite limited. Regular intention attribution is greatly facilitated by the relatively narrow range of actions available to an agent at a time. By contrast, as noted above (section 3), the range of meanings a speaker can reasonably intend to convey in a given situation is virtually unlimited. It is simply not clear how the standard procedures for intention attribution could yield attributions of speakers' meanings except in easy and trivial cases (for further discussion, see Sperber 2000; Sperber & Wilson 2002).

In the second place, as noted above (section 4), inferential comprehension typically involves several layers of metarepresentation, while in regular mind-reading a single level is generally enough. This discrepancy between the metarepresentational capacities required for inferential comprehension and regular mind-reading is particularly apparent in child development. It is hard to believe that two-year-old children, who fail, for instance, on regular first-order false belief tasks, can recognise and understand the peculiar multi-levelled representations involved in verbal comprehension, using nothing more than a general ability to attribute intentions to agents in order to explain their behaviour. For these reasons, it is worth exploring the possibility that, within the overall mind-reading module, there has evolved a specialised sub-module dedicated to comprehension, with its own proprietary concepts and mechanisms (Sperber 1996; 2000; forthcoming; Origgi & Sperber 2000; Wilson 2000; Sperber & Wilson 2002).

If we are right, the Communicative Principle of Relevance in (7) above (“Every ostensive stimulus conveys a presumption of its own optimal relevance”) describes a regularity specific to the communicative domain. Only acts of ostensive communication create legitimate presumptions of optimal relevance, and this might form the basis for a a special-purpose comprehension device. On this modular account, the relevance-theoretic comprehension procedure in (9) above (“Follow a path of least effort in computing cognitive effects: test interpretive hypotheses in order of accessibility; stop when your expectations of relevance are satisfied”) could be seen as “fast and frugal heuristic”, a special-purpose inferential procedure which automatically computes a hypothesis about the speaker's meaning on the basis of the linguistic and other evidence provided, via mutual adjustment of context, content and cognitive effects.

This approach allows for varying degrees of sophistication in the expectations of relevance with which an utterance is approached. In the terms of Sperber (1994), a child with limited metarepresentational capacity might start out as a Naively Optimistic interpreter, who accepts the first interpretation he finds relevant enough regardless of whether it is one the speaker could plausibly have intended. A Cautious Optimist, with enough metarepresentational capacity to pass first-order false belief tasks, might be capable of dealing with mismatches of this type, but unable to deal with deliberate deception. A Sophisticated Understander has the metarepresentational capacity to deal simultaneously with both mismatches and deception (Sperber 1994; Bezuidenhout & Sroda 1998; Wilson 2000; Happé & Loth 2002).

The complexity of the inferences required on the Gricean account of communication has sometimes been seen as an argument against the whole inferential approach. We are suggesting an alternative view on which, just as children do not have to learn their language but come with a substantial innate endowment, so they do not have to learn what ostensive-inferential communication is, but come with a substantial innate endowment.


6. Conclusion: an experimentally testable cognitive theory

Relevance theory is a cognitive psychological theory. In particular, it treats utterance interpretation as a cognitive process. Like other psychological theories, it has testable consequences: it can suggest experimental research, and is open to confirmation, disconfirmation or fine-tuning in the light of experimental evidence. Of course, as with other theories of comparable scope, its most general tenets can be tested only indirectly, by evaluating some of their consequences. Thus, the Cognitive Principle of Relevance (the claim that human cognition tends to be geared to the maximisation of relevance) suggests testable predictions only when combined with descriptions of particular cognitive mechanisms (for perception, categorisation, memory, or inference, for example). Given a description of such a mechanism, it may be possible to test the relevance-theoretic claim that this mechanism contributes to a greater allocation of cognitive resources to potentially relevant inputs, by comparing it with some alternative hypothesis, or at least the null hypothesis.

The Communicative Principle of Relevance (the claim that every ostensive stimulus conveys a presumption of its own optimal relevance) is a law-like generalisation which follows from the Cognitive Principle of Relevance, together with a broadly Gricean view of communication as a process of inferential intention-attribution. The Communicative Principle of Relevance could be falsified by finding genuine communicative acts which did not convey a presumption of their own optimal relevance (but conveyed instead, say, a presumption of literal truthfulness, or maximal informativeness, or no such presumption at all). When combined with descriptions of specific types and properties of communicative acts (and in particular of utterances), the Communicative Principle yields precise predictions, some of which have now been tested experimentally.

Throughout this survey, we have tried to point out cases where the predictions of relevance theory differ from those more or less clearly suggested by alternative frameworks (on the interpretation of ostensive silences, for example, the order in which literal and metaphorical interpretations are accessed, the contribution of pragmatic principles to explicit communication, the nature of lexical-pragmatic processes such as narrowing and loosening, the lack of parallelism between metaphor and irony, and so on), and we have drawn attention to many cases where the relevance-theoretic analyses have been experimentally tested and their predictions confirmed. Here we will give two further illustrations of how the basic notion of optimal relevance, characterised in terms of the effort and effect, yields testable predictions.

As noted above (section 2), relevance theory does not provide an absolute measure of mental effort or cognitive effect, and it does not assume that such a measure is available to the spontaneous workings of the mind. What it does assume is that the actual or expected relevance of two inputs can quite often be compared. These possibilities of comparison help individuals to allocate their cognitive resources, and communicators to predict and influence the cognitive processes of others. They also make it possible for researchers to manipulate the effect and effort factors in experimental situations.

Thus, consider a conditional statement such as “If a card has a 6 on the front, it has an E on the back.” In the Wason selection task (Wason 1966), the most famous experimental paradigm in the psychology of reasoning, participants are presented with four cards showing (say) a 6, a 4, an E and an A, and asked which of these cards should be turned over to reveal the hidden letter or number and check whether the conditional statement is true or false. By 1995, literally thousands of experiments with such materials had failed to elicit a majority of correct responses (where the correct response is to select the 6 and the A cards). Most people choose the 6 alone, or the 6 and the E. In “Relevance theory explains the selection task” (1995), Sperber, Cara and Girotto argued that participants derive the testable implications of the conditional statement in order of accessibility, stop when their expectations of relevance are satisfied, and select cards on the basis of this interpretation. Using this idea, Sperber et al. were able, by varying the content and the context of the conditional statement, to manipulate the effort and effect factors so as to elicit correct or incorrect selections at will.

Typically, the way a conditional statement If P then Q achieves relevance is by making it possible to derive the consequent Q in cases where the antecedent P is satisfied. With the conditional “If a card has a 6 on the front, it has an E on the back”, this leads to selection of the card with a 6. Another common way for a conditional statement to achieve relevance is by creating the expectation that the antecedent P and the consequent Q will be true together. In the present case, this leads to selection of the both the 6 and the E cards. Of course, it is also possible to use a conditional statement to derive the implication that its antecedent will not co-occur with the negation of its consequent. If participants chose cards on this basis, they would correctly select the 6 and A cards. However, in most contexts this implication is relatively costly in terms of effort, yields no further effects, and would not be derived by a hearer looking for optimal relevance. What Sperber et al. did was manipulate the effort and effect factors, first separately and then together, so that this implication was easier and/or more rewarding to derive, and the correct cards were therefore more and more likely to be chosen. The most successful condition was one in which the statement “If a card has a 6 on the front, it has an E on the back” was presented as being made by a technician who had just repaired a machine which was supposed to print cards with this specification, but which had malfunctioned and printed cards with a 6 on the front and an A on the back. Here, the statement achieved relevance by implying that there would be no more cards with a 6 and an A rather than an E, and a majority of participants made the correct selection. This and other experiments with the selection task (see also Girotto, Kemmelmeir, Sperber, & van der Henst 2001, Sperber and Girotto forthcoming) showed that performance on this task was determined not by domain-general or domain-specific reasoning mechanisms (as had been argued by most reseachers) but by pragmatic factors affecting the interpretation of conditional statements. It also confirmed that the interpretation of conditionals is governed by the twin factors of effort and effect, which can act either separately or in combination.[41]

Here is a second example of how the interaction of effort and effect can be experimentally investigated, this time in the production rather than the interpretation of utterances. Suppose a stranger approaches and asks me the time. I look at my watch and see that it is 11. 58 exactly. How should I reply? A speaker observing Grice's maxims (and in particular the maxim of truthfulness), addressing an audience who expects her to observe these maxims, should say, “11.58”. In this situation, a speaker who violated the maxim of truthfulness by rounding up the time to 12.00 would create, not some metaphorical or ironical implicature, but the mistaken assumption that she meant to convey that it was (exactly) 12.00. By contrast, a speaker aiming at optimal relevance would have every reason to round up the time (thus reducing her hearer's processing effort) unless, in her view, some cognitive effect would be lost by speaking loosely. It should therefore be possible, on the relevance-theoretic approach, to elicit stricter or looser answers by manipulating the scenario in which the question is asked, so that the stricter answer does or does not yield relevant implications. This prediction has been experimentally tested, and the relevance-theoretic analysis confirmed: strangers in public places asked for the time tend either to round up or to give answers that are accurate to the minute, depending on subtle clues as to what might make it relevant to the person making the request to know the time (van der Henst, Carles & Sperber 2002).

At the moment, the main obstacle to experimental comparisons of relevance theory with other pragmatic theories is that the testable consequences of these other theories have often not been made explicit. It is true that much pragmatic research has been carried out in a philosophical or linguistic tradition where the goal of achieving theoretical generality, and reliance on intuitions, have led to a certain reluctance to get one's hands dirty by getting down to actualexperimentation. Relevance theorists have been trying to combine theoretical generality with all the possibilities of testing provided by the careful use of linguistic intuitions, observational data, and the experimental methods of cognitive psychology.

[1] On the distinction between decoding and inference, see Relevance (Sperber & Wilson 1986a/1995): chapter 1, sections 1-5, chapter 2. On the relation between decoding and inference in comprehension, see Blakemore 1987; forthcoming; this volume; ; Wilson & Sperber 1993; Wilson 1998; Carston 1999; forthcoming; Origgi & Sperber 2000; Wharton 2000; 2001; Breheny 2002; Recanati 2002. On the role of demonstrative and non-demonstrative inference processes in comprehension, see Relevance: chapter 2; Sperber & Wilson 2002; Recanati 2002; Carston forthcoming.


[2] For early arguments against these aspects of Grice's framework, see Sperber & Wilson 1981; Wilson & Sperber 1981. For discussion and further references, see below.


[3] The notion of a cognitive effect (or contextual effect) has been revised several times. For early accounts, see Wilson & Sperber 1981, 1986b. For the standard definitions, see Relevance: chapter 2, section 7, and especially footnote 26. For the notion of a positive cognitive effect, see Sperber & Wilson 1995: 3.1-3.2. We leave open the possibility that there may be still further types of positive cognitive effect (improvements in memory or imagination, for example (cf. Wilson & Sperber 2002).


[4] For some suggestions about how this might be done, see Relevance: 124-32. Formal notions of relevance are currently being explored by Merin 1997; Blutner 1998 (which brings together ideas from Horn 1984; 1992, Levinson 1987; 2000, Hobbs et al. 1993, and Sperber & Wilson); van Rooy 1999; 2001. For some alternative notions of relevance, see references in Sperber & Wilson 1986b; Wilson 1999.


[5] On the distinction between comparative and quantitative concepts, see Carnap 1950; Relevance: 79-81, 124-32. On comparative and quantitative notions of relevance, see Relevance: chapter 3, sections 2, 5 and 6. For some factors affecting comparative assessments of relevance, see Relevance: chapter 3, sections 2, 6; Sperber & Wilson 1996.


[6] It is sometimes suggested that the lack of a quantitative notion of relevance makes the theory untestable. In fact, there is now a considerable experimental literature on relevance theory, and many procedures for testing and manipulating effort, effect and relevance (for discussion, see footnote 5 and section 6 below.)


[7] This is the simpler of two characterisations of ostensive-inferential communication in Relevance: 29, 58, 61. The fuller characterisation involves the notions of manifestness and mutual manifestness. In particular, we argue that for communication to be truly overt, the communicator's informative intention must become not merely manifest to the audience (i.e. capable of being recognised and accepted as true, or probably true), but mutually manifest to communicator and audience. On the communicative and informative intentions, see Relevance: chapter 1, sections 9-12; on the notion of mutual manifestness, see Garnham & Perner 1990; Sperber & Wilson 1990.


[8] On the use of silence as an ostensive stimulus, see Morgan & Green 1987: 727; Sperber & Wilson 1987: 746-7.


[9] The analysis of scalar implicatures is another case where Gricean analyses tend to lose the symmetry between unwillingness and inability to provide relevant information. For discussion, see Sperber & Wilson 1995: 276-8; Green 1995; Matsumoto 1995; Carston 1997; and section 6 below.


[10] For discussion and illustration, see Carston (this volume). On the notion of explicit content, see section 4 below.


[11] Notice, incidentally, that the hearer's expectations of relevance may be readjusted in the course of comprehension. For example, it may turn out that the effort of finding any interpretation at all would be too great: as a result, the hearer would disbelieve the presumption of relevance and terminate the process, with his now null expectations of relevance trivially satisfied.


[12] It is sometimes suggested (e.g. by Morgan and Green 1987: 726-7) that puns and deliberate equivocations present a problem for this approach. We would analyse these as cases of layering in communication, a widespread phenomenon which fits straightforwardly with our account. Just as the failure to provide relevant information at one level may be used as an ostensive stimulus at another, so the production of an utterance which is apparently uninterpretable at one level may be used as an ostensive stimulus at another (see Sperber & Wilson 1987: 751; Tanaka 1992).


[13] By `explicitly communicated content' (or explicature), we mean a proposition recovered by a combination of decoding and inference, which provides a premise for the derivation of contextual implications and other cognitive effects (Relevance: 176-93; Carston this volume; forthcoming). Despite many terminological disagreements (see footnotes 23 and 24), the existence of pragmatic contributions at this level is now widely recognised (see e.g. Wilson & Sperber 1981; 1998a; 2002; Kempson & Cormack 1982; Travis 1985; 2001; Relevance: chapter 4, sections 2-3; Kempson 1986; 1996; Blakemore 1987; Carston 1988, 2000; forthcoming; Recanati 1989; forthcoming; Bach 1994a, 1994b; 1997; Stainton 1994; 1997; Bezuidenhout 1997; Levinson 2000; Fodor 2001.


[14] In his `Retrospective Epilogue', and occasionally elsewhere, Grice seems to acknowledge the possibility of intentional pragmatic contributions to `dictive content' (Grice 1989a: 359-68). See Carston forthcoming; Wharton in preparation for discussion.


[15] On the distinction between primary and secondary pragmatic processes, see Breheny 2002; Recanati 2002; Carston this volume; forthcoming, Sperber & Wilson 2002. Some of the literature on generalised conversational implicature and discourse pragmatics tacitly invokes a similar distinction (cf. Hobbs 1985; Lascarides & Asher 1993; Lascarides, Copestake & Briscoe 1996; Levinson 2000). See also footnotes 22-3.


[16] See, for example, Relevance: chapter 4, sections 3-5, esp. pp 204-208; Wilson & Sperber 2002.


[17] A hearer's expectations of relevance may be more or less sophisticated. In an unsophisticated version, presumably the one always used by young children, what is expected is actual optimal relevance. In a more sophisticated version, used by competent adult communicators who are aware that the speaker may be mistaken about what is relevant to the hearer, or in bad faith and merely intending to appear relevant, what is expected is a speaker's meaning that it may have seemed to the speaker would seem optimally relevant to the hearer. Adult communicators may nevertheless expect actual optimal relevance by default. Here we will ignore these complexities, but see Sperber 1994; Wilson 2000; and section 5 below.


[18] For expository purposes, we have chosen an example in which the linguistic content of the discourse, and in particular the preceding utterance (“No”), creates a fairly precise expectation of relevance, allowing the interpretation process to be strongly driven by expectations of effect. In an indirect answer such as (ib), where the linguistic form of the utterance is compatible with two different lines of interpretation, considerations of effort, and in particular the accessibility of contextual assumptions capable of yielding the expected conclusions, play a more important role. In a discourse-initial utterance such as (ii), or in a questionnaire situation, considerations of effort are likely to play a decisive role in narrowing down the possible lines of interpretation:


(i) a. Peter: Did John pay back the money he owed?

b. Mary: He forgot to go to the bank.

(ii) He forgot to go to the bank.


[19] For one thing, we have used English sentences to represent the assumptions and assumption schemas that Peter entertains at different stages of the comprehension process, which we assume he does not represent in English but in some conceptual representation system or language of thought.


[20] See for example Horn 1984, 1992; Levinson 1987, 2000; Hobbs et al. 1993; Lascarides, Copestake & Briscoe 1996; Blutner 1998, 2002; Lascarides & Copestake 1998.


[21] As noted above (footnote 16), there is some debate about how the explicit-implicit distinction should be drawn (see, for example, Horn 1992; Relevance: chapter 4, sections 1-4; Wilson & Sperber 1993; Bach 1994a; 1994b; 1997; Levinson 2000; Carston 2002; this volume; forthcoming). The issue is partly terminological, but becomes substantive when combined with the claim that explicit and implicit communication involve distinct pragmatic processes (as it is in much of the literature on generalised implicatures: e.g. Levinson 2000).


[22] Levinson (2000: 195-6) discusses a number of possible criteria for distinguishing explicatures from implicatures, provides arguments against each, and concludes that the distinction is unjustified. But there is no reason to expect a criterion to be provided for each theoretical distinction. (We would not expect the defenders of a distinction between generalised and particularised implicatures to provide a criterion, although we would expect them to characterise these notions clearly and provide sound supporting evidence). Our notion of an explicature is motivated, among other things, by embedding tests which suggest that certain pragmatic processes contribute to truth-conditional content, while others do not (Wilson & Sperber 1986: xxx; 2002). The allocation of pragmatically inferred material between explicatures and implicatures is constrained, on the one hand, by our theoretical definitions of explicature and implicature (Relevance: 182), and, on the other, by the fact that the implicated conclusions which satisfy the hearer's expectations of relevance must be warranted by the explicit content of the utterance, together with the context. For further discussion, see Relevance: chapter 4, section 3; Sperber & Wilson 1998a; Carston 2000, this volume; Wilson & Sperber 1997; 2000. For some experimental work, see Gibbs & Moise 1997; Matsui 1998; 2000; Nicolle & Clark 1999; Wilson & Matsui 2000.


[23] Since lexical loosening is widely acknowledged as one of the factors driving semantic change, it might be argued that from a synchronic point of view, these are simply cases of polysemy. However, we are interested in the pragmatic micro-processes underlying these semantic changes, and we will largely abstract away from the question of whether Hoover , or square, or silent has acquired an extra stable sense. Notice, though, that the variation in interpretations of a word such as square or silent applied to different objects in different circumstances is so great as to make purely semantic or default-pragmatic explanations seem infeasible (for discussion, see Horn 1984; Lakoff 1987; Sweetser 1990; Hobbs et al. 1993; Recanati 1995; Carston 1997; this volume; forthcoming; Sperber & Wilson 1998a; Traugott 1998; Wilson 1998; Asher & Lascarides 2001; Papafragou 2001; Wilson & Sperber 2002).


[24] For early arguments against the maxim of truthfulness, see Wilson & Sperber 1981. For detailed critiques of frameworks based on maxims or conventions of truthfulness, discussion of some existing accounts of loose use, and justification of an alternative, relevance-theoretic account, see Wilson & Sperber 2002. For experimental evidence, see Matsui 1998; 2000; Wilson & Matsui 2000; Van der Henst, Carles & Sperber forthcoming.


[25] For Grice, metaphor and hyperbole involve different types of interpretation process, and may indeed be mutually exclusive: see Grice 1989b: 34.


[26] See, for example, Gibbs 1994; Noveck 2000; Glucksberg 2001. Glucksberg's view that the interpretation of metaphor involves the construction of a broader category than the encoded meaning fits well with our analysis based on loose use.


[27] While the claim that metaphor is a variety of loose use has been part of the theory for some time (see e.g. Sperber & Wilson 1985/6; Relevance chapter 4, sections 7-8; 1990b), the details of this analysis are new. For discussion, see Carston 1997; this volume; forthcoming; Recanati 1995; Sperber & Wilson 1998a; Wilson & Sperber 2002.


[28] On the notion of interpretive use, see Relevance: chapter 4, section 7; Blass 1990; Gutt 1991; Sperber 1997; Wilson 2000; Noh 2001; Papafragou 2001. On the notion of echoic use, see Relevance: chapter 4, section 9; Blakemore 1994; Carston 1996; forthcoming; Wilson 2000; Noh 1998.


[29] The relevance-theoretic account of irony was first proposed in Sperber and Wilson 1981. It was extended and developed in Relevance: chapter 4, sections 7, 9; Sperber & Wilson 1990b; 1998b; Wilson and Sperber 1992; Curcò 1998. For critical discussion, see Clark & Gerrig 1984; Kreuz & Glucksberg 1989; Kumon-Nakamura, Glucksberg & Brown 1995; Martin 1992; and the papers by Seto, Hamamoto and Yamanashi in Carston & Uchida 1998. For responses, see Sperber 1984; Sperber & Wilson 1998a.


[30] This approach has been experimentally tested: see Jorgensen, Miller & Sperber 1984; Happé 1993; Gibbs 1994; Kreuz and Glucksberg 1989; Kumon-Nakamura, Glucksberg and Brown 1995; Langdon, Davies & Coltheart 2002.


[31] On the development of metaphor and irony, see Winner 1988. On the relation between irony, metaphor and metarepresentational abilities, see Happé 1993; Langdon, Davies & Coltheart 2002. For further discussion of the relation between communicative and metarepresentational abilities, see below, section 5.


[32] Levinson (2000: 239) interprets us (mistakenly) as claiming that ironies “are implicatures interpreted as `echoes' of what someone might have said: they are distinctly not explicatures”. He objects that our account does not allow for the fact that ironical use of a referential expression may make a difference to truth conditions (as in his nice example “If you need a car, you may borrow my Porsche” [used to refer to the speaker's VW]). In fact, such examples provide strong confirmation of our account, on which irony is closely related to mention, quotation and other types of metalinguistic use. It is well known that metalinguistic use of a word may make a difference to truth conditions (see Horn 1989; Sperber & Wilson 1981; Relevance: chapter 4, section 7; Carston 1996; forthcoming; Cappelen & Lepore 1997; Noh 2000; Wilson 2000.)


[33] For discussion, see Relevance: chapter 1, sections 9-12.


[34] In the relevance-theoretic framework, mood indicators are among the items seen as carrying procedural rather than conceptual meaning. For discussion, see Blakemore (this volume).


[35] On higher-level explicatures, see Blakemore 1991; Wilson & Sperber 1993; Ifantidou 2001. On the analysis of non-declarative utterances, see Relevance: chapter 4, section 10; Wilson & Sperber 1988; Wilson 2000; Noh 2001. For critical discussion, see Bird 1994; Harnish 1994.

[36] See, for example, Davies and Stone 1995a; 1995b; Whiten 1991; Carruthers & Smith 1996; Malle, Moses & Baldwin 2001.


[37] See, for example, Perner, Frith, Leslie & Leekam 1989; Happé 1993; Baron-Cohen 1995; Mitchell, Robinson & Thompson 1999; Happé & Loth 2002; Papafragou 2002 and the papers in Mind & Language 17.1-2 (2002).


[38] We are using “module” in a somewhat looser sense than Fodor's, to mean a domain-or task-specific autonomous computational mechanism (for discussion, see Sperber 1996: chapter 6).


[39] See, for example, Leslie 1991; Hirschfeld & Gelman 1994; Barkow, Cosmides & Tooby 1995; Sperber 1996; forthcoming. For critical comments, see Fodor 2000.


[40] For explicit defence of this position, see Bloom 2000; 2002. For experimental evidence for a more modular approach, see Happé & Loth 2002.


[41] For other applications of relevance theory in the experimental study of reasoning, see Politzer & Macchi (2000), Van der Henst (1999) Van der Henst, Sperber, & Politzer (in press).