«Indicators and procedures : nevertheless and but1 DIANE BLAKEMORE University of Salford (Received August ; revised January ...»
() Her husband is in hospital but she’s seeing other men.
Kitis claims that examples like () show that and functions as an emotional device that registers the speaker’s involvement. Her aim is to explain not only how and comes to have this function, but also why it is used in preference to but which is the ‘ prototypical adversative or contrastive connective ’ (Kitis : ).
The fact that () can be interpreting an emotional attitude does not necessarily show that and encodes emotional involvement. As Blakemore & Carston () show, it is possible to explain the contrast between () and () without abandoning a minimal truth-functional semantics for and.
According to their argument, an and conjunction like () is processed as a single unit of relevance so that its relevance hinges on the fact that the two conjuncts are true . In other words, it is the that is understood to give rise to attitudinal eﬀects. In contrast, but can only have what Kitis describes as a ‘ back-tracking ’ function in () because it is processed as two individual units of relevance, or, in other words, because the hearer ﬁrst processes the segment in (a) and draws an inference such as the  I do not want to suggest here that and is always used in utterances which express conjoined propositions. In some cases, where it is stressed, for example, and can be used discourse initially in the same way as but. In such cases, it plays a similar role to moreover or furthermore, and the proposition it introduces is expected to achieve relevance individually.
For further discussion of discourse initial uses of discourse connectives, see Blakemore ().
one in () and processes the segment in (b) which contradicts and leads to the elimination of ().) () (a) Her husband is in hospital.
(b) She’s seeing other men.
() She is not seeing other men.
This analysis of () is based on the analysis of but developed in Blakemore (, ), according to which the segment introduced by but communicates (explicitly or implicitly) a proposition that contradicts and leads to the elimination of a proposition which the speaker believes is manifestly inferrable from a mutually manifest phenomenon, which may be coded communicative behaviour, as in examples ()–() above, or simply something in the physical environment, as in ().
In discussing this analysis, Rieber () claims that it simply does not make sense for a speaker to intentionally communicate a proposition (say, ()) which she does not want the hearer to derive. However, that is not what is really going on here. A hearer who recognizes that he is expected to abandon () will also recognize that the inference from (a) to () is illegitimate and hence that he is expected to abandon the contextual premises which were needed for its derivation, for example, ().
() If someone’s husband is in hospital, then she will not be seeing other men.
Indeed, it is possible that the speaker’s intention in producing the utterance in () was to get the hearer to abandon this assumption, or to communicate that this assumption, which might be regarded as a particular instance of a social or cultural generalization, is in fact false.
The suggestion, then, is that in uttering the but segment, the speaker is communicating that she is attributing to the hearer the derivation of an assumption that is not justiﬁed. In some cases, the speaker’s grounds for attributing the hearer with having derived this assumption may be mistaken.
The speaker of (), for example, might have been mistaken in thinking that the hearer would have wanted to eat the whole pizza. Nevertheless the hearer will have understood the utterance if he recognizes that he is being attributed with this assumption.
 Carston (, forthcoming) has shown that a similar (pragmatic) explanation can be given for a range of interpretive discrepancies between conjoined and non-conjoined utterances, for example, the discrepancy in (i) ﬁrst noted by H. Clark (cited in Gazdar ) and cases like those in (ii) discussed by Bar-Lev & Palacas ().
(i) (a) John broke his leg. He slipped on the ice.
(b) John broke his leg and he slipped on the ice.
(ii) (a) Language is rule governed. It follows regular patterns.
(b) Language is rule governed and it follows regular patterns.
For further discussion, see Blakemore & Carston ().
There are cases, however, in which the speaker uses but in the denial of an assumption which she manifestly does attribute to the hearer. For example, a speaker may produce () on the understanding that the contextual assumption necessary for the derivation of the assumption of (a), namely (b), is not held either by herself or the hearer. Nevertheless the relevance of the but segment (and the success of the joke) clearly depends on the hearer’s recognition that both these assumptions are being communicated as representations of thoughts which are being attributed to someone else.
() She’s a linguist, but she’s quite intelligent.
() (a) She’s not intelligent.
(b) All linguists are unintelligent.
According to Rieber () but is an indicator in the sense that what it causes the hearer to notice – namely, that there is a contrast between the two propositions asserted – is something that he can see for himself. However, according to the analysis outlined in this section, a hearer ‘ sees ’ that there is a contrast between two propositions only in the sense that he recognizes that he is expected to perform certain kinds of inferences, or that he is expected to have followed a particular inferential route – a route which ends with the abandonment of an assumption derived by inference from an assumption which is presumed to be highly accessible. In other words, but causes the hearer to notice something that he can see for himself only in the sense that it leads him to this inferential process.
This analysis does not assume, as Rieber’s does, that the contrastive import of but is expressed in a distinct proposition whose truth is indicated or suggested rather than asserted. At the same time, it does not assume, as Bach’s () analysis does, that the contrastive import of but is part of what is said. Identifying what Bach calls the contrastive import of but is simply a matter of making the right sort of inferences and deriving the right kind of eﬀects. This means that Rieber’s example in () (repeated below) can be explained in terms of the attribution of an inferential process to Tom rather than the attribution of a propositional content.
() Tom thinks that Sheila is rich but she is unhappy. However, I have always thought that all rich people are unhappy.
. P According to the view of linguistic meaning suggested in the preceding section, there are two diﬀerent ways in which linguistic meaning can act as the input to the inferential processes involved in utterance comprehension.
On the one hand, expressions may encode which are the constituents of the conceptual representations that undergo inferential computations. And on the other hand, they may encode , or the means for increasing the salience of a particular kind of inferential computation.
Bach () has argued that this distinction is in fact vacuous since ‘ after all, in some way or other anything one utters ‘‘ constrains ’’ the inferential phrase of comprehension ’ ( : ). It is true that the inferences a hearer derives from an utterance depend on its conceptual content in the sense that this is what interacts with the context in the derivation of contextual eﬀects.
However, the contextual eﬀects the hearer derives also depend on the contextual assumptions which he uses in their derivation and on the type of inferential computation he performs. For example, the (b) segment in () (adapted from Hobbs ) can be interpreted either as evidence for the proposition that Tom can open Bill’s safe or as an implication of the proposition that Tom can open Bill’s safe.
() (a) Tom can open Bill’s safe. (b) He knows the combination.
In the ﬁrst case, the proposition expressed by (b) is functioning as a premise in an inference which has the proposition expressed by the (a) segment as a conclusion. In the second case, it is understood as a conclusion in an inference which has the proposition expressed by the (a) segment as a premise. The claim that linguistic meaning can encode constraints on the inferential phase of comprehension means that there are linguistic expressions (you see and so, for instance) which encode information about which of these inferential procedures yields the intended interpretation.* As I have shown in earlier work (Blakemore , ), the distinction between these two kinds of linguistic meaning can be justiﬁed in both cognitive and communicative terms."! The cognitive justiﬁcation follows from the assumption, fundamental to Relevance Theory, that utterance interpretation involves performing computations over conceptual representations. This means that a language can be expected to encode not only the constituents of the conceptual representations which undergo computations, but also information about the computations or inferential procedures in which these representations are involved.
Within Relevance Theory the fact that languages developed coded means for encoding inferential procedures can be explained in communicative terms. According to Sperber & Wilson’s () communicative principle of relevance, a hearer who recognizes that a speaker has made her intention to convey information manifest is entitled to assume that that speaker is  These are not the only interpretations for this sequence. For instance, a speaker may have intended to communicate not only that (b) is evidence for the truth of the proposition expressed by (a), but also that this proposition is an assumption which the speaker knows to be true. This interpretation would be indicated by the use of after all. For further discussion, see Blakemore ().
 For a full discussion of the distinction, see Wilson & Sperber ().
being . In other words, in making her intention to communicate manifest the speaker is communicating her belief that, ﬁrst, the utterance will achieve a level of relevance high enough for it to be worth processing, and, moreover, that this level of relevance is the highest level that she is capable of given her interests and preferences. Since the degree of relevance increases with the number of eﬀects derived but decreases with the amount of processing eﬀort required in deriving them, the use of an expression which encodes a procedure for identifying the intended contextual eﬀects would be consistent with the speaker’s aim of achieving relevance for a minimum cost in processing.
The idea that linguistic meaning may encode constraints on relevance has been applied to the analysis of a range of non-truth conditional discourse connectives."" At the same time, however, the investigation of the role of inference in comprehension has suggested that the role of procedural meaning is not, as I suggested in my earlier work, restricted to the recovery of implicit content. As both Sperber & Wilson () and Carston () have demonstrated, the recovery of the explicit propositional content of utterances involves inferential computations constrained by pragmatic principles, and thus that explicit content is ‘ much more inferential and hence worthy of pragmatic investigation ’ than is assumed by pragmatists in the Gricean tradition ( : , ). If there are linguistic expressions and structures which constrain the inferences hearers make in deriving the implicit content of utterances, then it is possible that there are also linguistically encoded constraints on the inferences involved in the identiﬁcation of explicit content. And indeed, as both Wilson & Sperber (, ) and Clark (, ) have shown, both non-truth conditional expressions, for example, illocutionary adverbials and mood indicators, and truth conditional expressions, for example, pronouns, can be analysed in terms of constraints on explicit content."# The present paper represents a return to the subject of my earlier work – linguistically encoded constraints on implicit content. On the other hand, it aims to go beyond this work by showing that the notion of a semantic constraint on implicit content that I developed in this work is too narrow.
According to Relevance Theory, the inferential processes involved in assessing relevance involve performing deductive inferences over conceptual representations (that is, the proposition expressed together with contextual assumptions) in the derivation of contextual eﬀects. Contextual eﬀects can be classiﬁed according to the kind of inferential process involved when newly presented information P is brought together with, or , old information C. Thus on the assumption that the human deductive device is  See, for example, Blass (, ) ; Haegeman () ; Itani () ; Higashimori ().
 See also Rouchota (a, b, c) on mood indicators, and Ifantidou-Trouki () and Blakemore () on parentheticals.
as described by Sperber & Wilson ( : –), we can distinguish three
types of contextual eﬀect :
(i) The derivation of contextual implications : the derivation of a new assumption in a deduction which crucially involves the synthesis of P and C.
(ii) Strengthening existing assumptions : the eﬀect derived when an assumption in C is independently derived from a new set of premises that includes P, or in other words, when P is involved in a ‘ backwards ’ inference.
(iii) Contradiction & elimination : the eﬀect derived when a contradiction between P and C is resolved by eliminating C.
This might suggest that if there are linguistic means for constraining these processes, then they could be expressions or structures which directly specify the kind of contextual eﬀect that is intended – for example, the elimination of an existing assumption or the strengthening of an existing assumption. On the other hand, since the particular eﬀect achieved will depend on the contextual assumptions used as premises in this deduction, it is equally possible that a linguistic expression or structure could constrain relevance by directly specifying the properties of the contextual assumptions which are intended to be used.
It seems that the term ‘ semantic constraint on relevance ’, as it is developed in my earlier work (Blakemore , ), would have to be construed in the ﬁrst of these ways. Consider, for example, my analysis of so as it is used in ().
() Tom can open Bill’s safe. So he knows the combination.