- Email: [email protected]
Conjunction, explanation and relevance
Lingua
90 (1993) 2748.
Conjunction, Robyn
27
North-Holland
explanation and relevance
Carston*
Department of Phonetics and Linguistics, University College London, Gower Street, London WCIE 6BT. UK
1. Introduction
The ground-breaking work of Paul Grice (1967) has established a fundamental distinction: between the intrinsic semantics of linguistic expressions, on the one hand, and the pragmatic, context-dependent, content of an utterance, on the other. This judicious division of labour is well illustrated in the account of what is communicated by utterances of and-conjunctions, such as those in (1): (la) (lb) (lc) (Id) (le)
It’s spring in England and it’s autumn in New Zealand. He handed her the scalpel and she made the incision. We spent the day in town and I went to Harrods. She fed him poisoned stew and he died. I left the door open and the cat got in.
The word and is taken to be pretty well semantically empty; that is, it is taken to be the natural language equivalent of the truth-functional logical conjunction operator. Pragmatics takes care of the great variety of temporal, causeconsequence and other sorts of relations understood to hold between the states of affairs described. Some of these come through in the so-called asymmetrical examples in (lb)-(le). For instance, we take it that the making
*
This is a revised and expanded
version
of a paper
by the same name, which
appears
in the
fourth volume of UCL Working Papers in Linguistics, 1992. I am very grateful to Deirdre Wilson for useful and enjoyable discussions on the concerns of this paper. My debt to the work of Diane Blakemore is evident throughout the paper. I have also been helped and encouraged by Mira Ariel, Noel Burton-Roberts, John Durkin, Marcus Giaquinto, Dick Hudson, Neil Smith, Dan Sperber and Vladimir Zegarac.
0024-3841/93/$06.00
0
1993 -
Elsevier
Science Publishers
B.V. All rights reserved
28
R. Curston 1 Conjunction, explanution und relevuncr
of the incision in (lb) followed the handing over of the scalpel and that an interval of a few seconds intervened. A quite different temporal relation is understood to hold between the states of affairs described in (lc), the event of going to Harrods is interpreted as contained within the period of time spent in town. Different sorts of consequence relations are understood in (Id) and (le): while the feeding of poisoned stew is a sufficient cause for death, the leaving open of the door is just one of a range of factors contributing to the cat’s getting in. These few examples by no means exhaust the range of different temporal, cause-consequence and other relations that conjunctions can communicate. Some further possibilities will be mentioned in the next section and others have been discussed by Schmerling (1975) Posner (1980) and Wilson and Sperber (forthcoming a). According to the pragmatic approach, a hearer recovers these connections inferentially. This involves an interaction of the decoded semantic content with general knowledge assumptions about the way things connect up and relate in the world, an interaction constrained by some general criterion or criteria of rational communicative behaviour.
2. The relevance-based pragmatic account It is not my purpose here to argue for the superiority of one pragmatic approach over another. Rather, I shall simply assume the relevance-theoretic framework and work within it. The fundamental idea is that, in processing an utterance, a hearer is looking for an optimally relevant interpretation; that is, an interpretation that has the two properties given in (2): (2) An utterance, on a given interpretation, is optimally relevant iff: (a) it achieves enough effects to be worth the hearer’s attention; (b) it puts the hearer to no gratuitous effort in achieving those effects. (Wilson
and Sperber
forthcoming
a)
What Sperber and Wilson call the principle of relevance is the thesis that every utterance communicates the presumption of its own optimal relevance. The pragmatic criterion that hearers actually employ in utterance interpretation is one of consistency with the principle of relevance: an interpretation has this property if and only if a speaker could have rationally expected it to be optimally relevant to the hearer. Once the hearer has accessed an interpretation which meets this criterion, he looks no further but takes this to be the
R. Carston 1 Conjunction,
explanation
29
and relevance
interpretation the speaker intended. The implications of this definition are fully discussed elsewhere (Sperber and Wilson 1986, 1987) so I shall do no more here than give a brief outline of how this pragmatic criterion applies in the interpretation of one of the examples above. As with any utterance, there are many possible interpretations of (lb) which are compatible with the linguistically encoded, semantic, content. Two of these logical possibilities for (lb) are given in (3): (3a) He handed her the scalpel and a second or two later she made the incision with that scalpel. (3b) He handed her the scalpel and simultaneously she made the incision with her pocketknife. Now, obviously, although these are both possible and consistent, the first one is overwhelmingly more likely to be recovered by the hearer, and to have been intended by the speaker, than the second one. Why is this? The intuitive answer is that it is just the obvious, natural, normal, etc., assumption to make. Put in slightly more theoretical terms, most adults, within a particular cultural range, have a mental script concerning the making of incisions with scalpels, perhaps as a subpart of a broader surgical operation script, involving a co-operative interaction of a nurse and a doctor. So when we hear or read (lb), a chunk of stereotypical material of this sort, in memory, is activated. On this basis, we instantly assume that the making of the incision followed the handing over of the scalpel, that the scalpel was used for the making of the incision and that the interval of time between the two actions was a few seconds at most; that is, the proposition is enriched along the lines of (3a). The relevance-based pragmatic criterion accounts for why we can safely rely on stereotypic knowledge in this instance, rather than considering other possible lines of interpretation, such as (3b). l The second part of the definition of optimal relevance, concerning the minimising of processing effort, is crucial here. An important factor, contributing to the effort involved r
Some other
instance, instruction
pragmatic
the principle
accounts
of informational
to derive implicatures
give a lot of weight enrichment,
to stereotypic
in Levinson
which are stereotypical
interpretation
(1987: 65-68)
enrichments
too;
for
is essentially
an
of the information
encoded
by the utterance. However, this principle makes some false predictions and offers nothing by way of explanation for why hearers should enrich in a stereotypic way (when they do) instead of in more unusual (and interesting) ways. See Carston (1990), Carston Sperber (forthcoming a), for criticisms of Levinson’s account.
(forthcoming)
and Wilson and
30
in deriving
R. Carston / Conjunction, rxplanahn
contextual
effects,
is the accessing
and rekvancr
of contextual
assumptions.
Now, in this particular example, the very concepts encoded in the sentence give immediate access to a ready-made script, requiring no imaginative effort of construction on the part of the hearer/reader. So (3a) is simply the first interpretation to come to mind. Then, assuming it has an adequate range of effects that the speaker could have manifestly foreseen, no other possibility is considered. Any of the other logically possible interpretations, such as (3b), would be much less accessible, thus requiring considerably more effort to construct, and no one of these further possibilities would be more obviously available than dozens of others. Note that (3b) might have just as wide a range of effects as (3a), or, conceivably, an even richer array (for instance: she doesn’t expect help from him, he is too slow, there is something wrong with the scalpel, etc.). However, the guarantee is not one of maximal effects; all that hearers can expect from speakers is that their utterances will have enough effects to justify the call on their attention. It follows that a speaker who wanted to communicate something other than (3a) would not be able to do so by uttering (lb) and, if functioning rationally, would not attempt to do so. The relevance-based account of the different cause-consequence relations in (Id) and (le) would run in essentially the same way. Each would activate a stereotypic script or schema, with, in the case of (Id), a built-in assumption about poison causing death, and, in the case of (le), assumptions about factors enabling cats to get in. I am assuming that these various enrichments of the bare linguistic meaning are instances of pragmatically derived content which contributes to the proposition expressed by the utterance, to ‘what is said’ in Gricean terms. That is, they are not implicatures, contrary to the assumption of Grice and his followers. Sperber and Wilson (1986: 188-192) give a range of examples which show that, in addition to disambiguation and reference assignment, there are many sorts of pragmatic enrichment which a hearer may have to recover, in order to arrive at the proposition explicitly expressed. With regard to the specific case of conjunct relations, I have argued the case for enrichment of the proposition expressed, at length, elsewhere (Carston 1988 and forthcoming). There is more to conjunction enrichment than the story told so far. There are cases where the temporal, cause-consequence, teleological and other assumptions do not spring out of ready-made scripted knowledge, or where two readily available scripts provide contradictory information about causee consequence relations :
31
R. Carston 1 Conjunction. explanation and relevance
(4a) (4b) (4~) (4d)
She screamed at him and he hit her. He hit her and she screamed at him. ? She went to bed and brushed her teeth. He yelled and the light came on.
The immediate assumption in interpreting (4a) is that the hitting is a consequence of the screaming, while in (4b) the screaming is a consequence of the hitting. Arguably, we have fairly accessible scripts or schemas reflecting both scenarios, but what dictates which of the two schemas comes into play in each case? The intuitive answer is that the order in which the information is presented in the utterances is taken to reflect the order in which the events occurred. This secures the sequential interpretation and the cause-consequence relation follows, in accordance with the script whose event sequence matches that of the utterance. The oddness of (4~) seems to arise from a clash between this order-of-presentation assumption and the familiar script in which teeth-brushing precedes going to bed. However, as will be evident in the next section, there is no pragmatic maxim or discourse principle requiring that the order of information in utterances reflect the order of events in the world. So our account of how the sequentiality enrichment arises, when it does, is not yet complete. The sort of explanation I would look for would be a general cognitive one, lying with our information-processing capacities and their bid to minimise effort expenditure. It should then follow from considerations of optimal relevance that conjunct order reflects the order of events in the world.2 Finally, the case of (4d). This does not map onto a specific mental script (an act of yelling followed by a light coming on is not a stereotypic sequence of events) and yet there is a strong tendency to assume a causeeconsequence connection here too. This raises the possibility that there may be something more fundamental at work in the imputing of causal relations, a very general procedure, perhaps, which licenses the move to causeeconsequence assumptions when a sequential relation is taken to hold between two states of affairs. It might be that the input to such a move would have to be broader than temporal sequentiality since that is not the only sort of temporal relation which can accompany a consequence relation. Consider the following examples, first pointed out by Schmerling (1975), and discussed more fully by Wilson and Sperber (forthcoming a):
2
A preliminary
effort considerations
attempt
at accounting
was made in Carston
for the sequentiality (1990).
enrichment
in terms of processing
32
R. Carston / Conjunction, explanation and relevance
(5a) The lights were off and I couldn’t see. (5b) Susan is underage and can’t drink. (5~) I spoke to Sam at the party and discovered
that he was unwell.
Examples (5a) and (5b) are cases for which the time-span of the state of affairs described by the second conjunct fully overlaps the time-span of the state of affairs described by the first. That is, the conjuncts do not form a temporal sequence, yet the second is understood as a consequence of the first. (5~) plainly involves a consequential relationship too and there is some sort of temporal priority of the first mentioned state of affairs over the one mentioned second: some of the speaking had to have taken place before the discovery could be made. But this temporal relation is appropriately paraphrased by ‘and in the process’, rather than by ‘and then’, which would give a different reading, one on which the conversation with Sam would not be seen as bringing about the discovery. So, if there is some more general schema or heuristic at work in the interpretation of some of these examples, it would be accessed on the basis of a temporal relation along the lines of ‘the onset of P precedes the onset of Q’. An utterance whose temporal structure matched this would then give access to the further assumption, ‘Q is a consequence of P’. Note that this is not to be understood as a context-free rule. There are utterances which meet the temporal requirement but are not further enriched to a cause-consequence relation. Standard assumptions concerning the limits of human causal powers block a cause-consequence interpretation of (6) for instance: (6) I walked
to the front gate and a fire engine
came around
the corner.
If this accessing strategy exists, it is, like heuristics in general, a useful rule of thumb, effort-saving and often, though not always, correct. I believe that the existence of a heuristic of this sort would be very natural within the relevance-theoretic framework, since causal stories are far more likely to yield a crop of implications, that is, have cognitive effects, than are mere temporal contiguities.3 Certainly, this is all very speculative and calls for deeper consideration than I can give it here.
3
This apparent
of course,
drive toward
not restricted
seeing states of affairs as glued together
to utterance
interpretation.
It is quite fundamental
in a great causal to our cognitive
mesh is. life in
general: it affects our perception, reasoning and memory, and emerges at a very early stage in development. See Carston (in preparation) for a survey of some of the vast literature on this.
R. Carston / Conjunction, explanation and relevance
33
3. Conjunction versus juxtaposition Various arguments have been put forward for favouring a pragmatic account of conjunct relations over a richer lexical semantics for and. One of these focuses on the point that the very same temporal, consequential, etc., relations arise when the and is removed. Compare (7a,b,c) with their conjunctive equivalents in (1): (7a) He handed her the scalpel. She made the incision. (7b) She fed him poisoned stew. He died. (7~) I left the door open. The cat got in. We find, for instance, that (7a) communicates that he handed her the scalpel before she made the incision, just as (lb) does, and similarly for the other connections. So, the argument went, these imputed relations don’t seem to have anything to do with the meaning of and. Rather, they are the product of some quite general cognitive predisposition to forge certain connections and relations between states and events whenever it seems reasonable to do so. That is, rather than being arbitrary linguistic facts, they are a matter for a cognitively-based pragmatics. This is a plausible, if not knock-down, argument; putting it together with several other arguments made a strong case for the pragmatic treatment. However, while it may be the case that all the relations communicable by an and-conjunction are still communicated when the and is removed, the converse does not seem to be the case. That is, there are some relations that are communicated by juxtaposed sentences which apparently cannot be communicated when these sentences are conjoined by and. This was first pointed out by Herbert Clark with a pair of examples, essentially the same as those given in (8): (8a) John broke his leg. He slipped on a banana skin. (8b) John broke his leg and he slipped on a banana skin. For (8a), it is quite possible, in fact more or less inevitable, that we understand the event of slipping on a banana skin as the cause of the event of leg-breaking; that is, the event mentioned second is understood as having caused, and hence as having preceded, the event mentioned first. This does not seem to be possible for (8b), where the sentences are presented in the same order but conjoined with and. The same disparity arises between (9a) and (9b), and a range of other cases:
34
R. Carston / Cortjunction, explanation and relevance
(9a) Mary fell asleep at work; she was exhausted. (9b) Mary fell asleep at work and she was exhausted. This is just
one of the relations
or connections
between
two juxtaposed
sentences which are precluded when they are conjoined with and. In an interesting but largely neglected paper, published in this journal in 1980, BarLev and Palacas surveyed the extent of this phenomenon and the variety of its manifestations. They pointed out that, whatever the source of the meaning difference is, it cannot be to do with the order of the constituent clauses since this ordering is the same for the two members of each pair. So there is no possible resort here to any pragmatic maxim concerning the order in which we communicate material, such as Grice’s manner maxim of orderliness or Harnish’s mirroring maxim: ‘in so far as possible, make the order of saying reflect the order of events’ (Harnish 1976: 359). The (a) members of the pairs in (8) and (9) provide compelling evidence against the existence of any such maxims, assuming that no discourse principle should be so limited in its application as to apply only to cases of explicit conjunction.4 Bar-Lev and Palacas (1980: 141) went on to claim that the inescapable conclusion is that there is more to natural language and than simple truth-functionality. They proposed that this extra element of meaning that and encodes is their relation of semantic command: (10) Semantic command: The second conjunct (S”) is not prior to the first (S’) (chronologically or causally). This idea does have some attractive features. First, it seems to encompass the whole spectrum of temporal and consequence relationships which conjuncts can have (a sample of which has been given in (1) and (5)) and to exclude those that they cannot have; that is, ‘backwards’ sequence and consequence. Second, although Bar-Lev and Palacas did not say so, it is clear that this semantic account still leaves a vast amount of appropriate work for pragmatics, in determining, for any particular case of a conjunctive utterance, precisely which, if any, temporal and causeeconsequence connection is to be taken to pertain. The relevance-based pragmatic analysis given in section 2 would go through, without any obvious changes being required, taking as its input
4 This point is argued further and Sperber (forthcoming a).
in Carston
(forthcoming),
Carston
(in preparation).
and Wilson
R. Carston 1 Conjunction, explanation and relevance
35
the ‘semantic command’ characterisation of conjunction. Third, this semantics for and avoids a range of problems which other attempts to give a semantic account fall prey to. The semantic ambiguity account and the univocal multi-featured semantic account, the latter due to Cohen (1971), wind up having to proliferate a potentially infinite number of senses or semantic features, in their bid to account for the full set of subtly different temporal and consequence relations that different examples communicate. Furthermore, some of these senses or features contradict others, for instance, simultaneity and sequentiality.5 So Bar-Lev and Palacas’s account seems to be the most promising of the semantic options on offer. However, it is not adequate. It is really nothing more than a descriptive statement of what we have just observed, which is that the ‘backwards’ causal and temporal relations possible for juxtaposed sentences are excluded when they’re conjoined with and. Surely what we want to know is why this is the case; it just looks like the kind of fact that ought to be able to be explained rather than merely recorded. Second, the definition is probably redundantly broad, as it stands. Juxtaposed sentences seem to share part of the restriction, which ‘semantic command’ places on conjunctions: in the absence of a backward causal relation, it appears to be just as difficult to get a reverse temporal sequence for them as it is for conjunctions: (1 la) (11 b) (1 lc) (1 Id) (11 e)
He She She He He
handed her the scalpel and she made the incision. made the incision and he handed her the scalpel. made the incision; he handed her the scalpel. rode into the sunset and jumped onto his horse. rode into the sunset. He jumped onto his horse.
There is a strong tendency to take the incision-making in (1 lb) to have preceded the handing over of the scalpel, even though it does not match the familiar script. It is this that the semantic command requirement is a response to. However, the point here is that the same goes for (1 lc): its temporal interpretation goes the way of (1 lb) rather than (1 la), given that the linguistically encoded content allows for either. The examples in (1 Id) and (1 le) further demonstrate this: (1 le) is peculiar in the same way that (1 Id) is. It is not the case that the juxtaposed example is interpreted with the event described by the second sentence understood as preceding that of the first
5 Posner (1980) provides an excellent overview position and compelling criticisms against it.
of the ‘meaning-maximalist’
(i.e. rich semantic)
36
R. Carston / Conjunction.
e.uplunation and relevance
sentence, although this is the stereotypical sequence. To this extent. then, conjunctions and juxtapositions are similarly restricted, which suggests that there is a more general story (hinted at in section 2) to be told about temporal sequencing, which applies to both of them, and that this need not be incorporated into the lexical semantics of and. Third, and most important, there are counterexamples to this semantics for and. These fall into two classes. The first set concerns those for which a socalled ‘backwards’ causal relation is understood despite the presence of and. The second set involves different sorts of relationships altogether, having nothing to do with temporal or causal links between states of affairs. Examples of the first sort are given in (12) and ( 13):
(12) If the old king has died of a heart attack
and a republic has been formed, and the latter event has caused the former, then Tom will be upset. (13) A: Did Bill break the vase? B: Well, the vase broke and Bill dropped it. (example due to Larry Horn) The point about (12) is that on the semantic command analysis the conjunctive antecedent of this conditional should have a contradictory feel to it: we are told that a cause-consequence relation between the second conjunct and the first is precluded, and then that just such a relation holds. There is of course no such feeling of inconsistency or tension; the encoded content of the and is just the minimal truth-functional connection, which is perfectly compatible with the further specification of a backwards causal relation. Consider now the more interesting case of (13) focusing on B’s reply to A’s question. Here there is no linguistic specification of a backwards causal relation, as there is in (12), but the event of dropping the vase is readily understood as having preceded and caused the breaking of the vase. This should not be possible if and really does encode the property of ‘semantic command’. Certainly B’s utterance is an indirect, round-about, sort of way of communicating the causeeconsequence relation; it is an implicature rather than an aspect of the explicit content. 6 The hearer is required to do some extra inferential
6
work in order to arrive at it, and no doubt
The explicit/implicit
elsewhere: Sperber (this volume).
distinction
and Wilson
is an important
(1986), Carston
and contentious (1988), Carston
receives extra effects in
one, discussed
at some length
(1990) and Wilson
and Sperber
R. Carston 1 Conjunction, explanation and relevance
31
the process, as the definition of optimal relevance predicts. The example deserves further analysis but the important point for me here is that, whatever the effects this conjunctive response has, they are not derived via a contradiction, which is what the analysis of and in terms of semantic command would require. An instance of the second sort of counterexample, in which temporal cause-consequence relations are not at issue, is given in (14) : (14a) Language (14b) Language (14~) Language
and
is rule-governed; it follows regular patterns. is rule-governed and it follows regular patterns. follows regular patterns and it is rule-governed.
In (14a) the second sentence is understood as a spelling out of, or elaboration on, the first one. This relationship seems to be another one which is precluded when the two sentences are conjoined with and, in whichever order, as in (14b) and (14~). These examples are cited by Bar-Lev and Palacas (1980: 144) later in their paper, which is odd, since if their semantic command analysis makes any prediction about them it makes the wrong one. It predicts that the conjunction in (14b) should be able to be interpreted in the same way as the juxtaposed sentences, since all that semantic command precludes is backwards causal relations. It seems clear that this attempt to account for the constraints on the interpretation of conjunctions has to be abandoned. It is both too restrictive, since, in some contexts, backwards causal relations are possible, and not restrictive enough, since it does not account for why the conjunction in (14b) cannot be interpreted in the same way as the non-conjoined clauses in (14a).
4. Explanation If we do away with the semantic account we are back where we began with the problem of the non-equivalence of the pairs in (8) and (9), and, in addition, we now have the problem of (14a) and (14b). Let’s focus for a moment on the juxtaposed, non-conjoined cases, (8a) and (9a): what’s going on in these examples is that the second sentence is being understood as providing an explanation for the state of affairs described in the first. Now, an explanation of something standardly involves the citing of a cause or a reason for that thing; it is often prompted by a ‘why?’ question or a ‘how did it come to be so?’ sort of question. What is interesting here is how dominant this sort of interpretation is for the juxtaposed cases.
38
R. Carsron / Conjunction,
(Isa) (15b) (15~) (15d)
Sue Sue Max Max
explanation
and relevance
is happy today. She got/made a phone-call. is miserable today. She got/made a phone-call. can’t read; he’s a linguist. can’t read and he’s a linguist.
It is quite natural and immediate in interpreting (15a) to take the getting or making of the phone-call, mentioned in the second sentence, to be the reason for Sue’s feeling happy, though such a state of affairs is entirely neutral as regards any standard assumptions about causes of happiness. We are no less inclined to take it as explanatory of her misery, when it is presented in the different juxtaposition in (15b). Even more surprising is (1 SC), where the immediate interpretive strategy seems to involve taking the property of being a linguist as explanatory of illiteracy, even if this is subsequently dismissed or taken as a joke. Compare this with the conjunctive counterpart in (ISd), where the most accessible interpretation is one in which the conjuncts are understood as describing two contrasting facts about Max, an interpretation which is entirely in line with our standard assumptions about the way the properties relate to each other. Faced with these examples, we might well feel that what needs accounting for is the dominance of this ‘fact followed by explanation’ type of interpretive strategy for the juxtaposed cases. In this paper, however, I’m just going to accept that this is so, that we are question-asking, explanation-seeking creatures, and return to the main issue here, which is the preclusion of certain sorts of interpretation from the andconjunctions. Could it be that what it all comes down to is that the second conjunct cannot be interpreted as explanatory of the first’? I think this comes quite close to the answer and is worth pursuing to see how far it takes us. Stating the condition this way not only covers the restriction on the interpretation of (8b) and (9b), but it also extends to (14b) and (14~). Although there is no issue of causality in these cases, there is an issue of explanatoriness. The second clause in (14a), ‘it follows regular patterns’, is given as an explanation of the notion of ‘rule-governedness’; it is a conceptual, or analytic, explanation as opposed to a causal one. This distinction is demonstrated by the two different ways in which the ‘why?’ questions in (16) can be interpreted and answered: (16) Why is John a bachelor? (a) Because he doesn’t like women. (b) Because he’s unmarried.
Why is Mary so witty? (a) She wants to impress us. (b) She says unexpected things.
R. Carston / Conjunction,
explanation
John’s dislike of women would enter into a bachelorhood while his unmarriedness is part tion of what it is for him to be a bachelor and for the responses regarding Mary’s wittiness.
39
and relevance
causal explanation of his of the conceptual explanathe same distinction holds So asserting that the ‘fact
plus explanation’ interpretation is impossible for conjunctions mops up not just the prohibited backwards causal relations, as in (8b) and (9b), which the semantic command analysis sought to preclude, but also cases such as (14) about which it was silent. Furthermore, it is not at odds with the interpretation of (12) or (13B), as it does not absolutely forbid an interpretation in which a state of affairs described in the second conjunct is understood as being in a causal relation with that described in the first. What it does do is account for the rarity of this sort of relation, since understanding a speaker as giving a cause or a reason for something is, typically, to understand her as presenting an explanation of it. In the context of A’s question in (13), B’s conjunctive response, ‘the vase broke and Bill dropped it’, presents A with two facts, rather than a fact followed by an explanation of that fact, though, via an inferential process, a hearer will no doubt derive the implication that the dropping caused the breaking. Further evidence, in favour of explanation as the crucial notion in accounting for the constraints on the conjunct relations, comes from a consideration of the class of so-called discourse connectives, which Blakemore (1987) has studied in some detail. These include expressions such as so, therefore, moreover, after all and you see. These form an interesting class of connectives since, it is argued, they do not affect the truthconditional content of utterances in which they occur, unlike such connectives as because, while, before, in order that, etc., but rather have the function of instructing the hearer how the proposition they introduce is to be processed. That is, they indicate that this proposition is to be placed in a certain inferential relation with some other, usually the one expressed by the immediately preceding utterance. Now, what is of interest to us here is that some of these seem to be embeddable in an and-conjunction, while others do not:
(17a) (17b) (17~) (17d)
The road was icy and so she slipped. She’s good-looking and moreover her father’s rich. *She slipped and you see the road was icy. *He’ll pass the French exam and after all he is a native
speaker.
40
R. Carston 1 Conjunction,
(17e)
*Language all it follows
explanation
and relevance
is rule-governed and that is/in other words/you regular patterns. ’
Of course, in all five cases, taking
seejqfter
out the and to render them non-conjunctive
makes them all equally acceptable. On Blakemore’s analysis, what distinguishes after all and you see from the others is that the sort of inferential role they indicate for the following proposition is one of being a premise in an argument which has the preceding proposition as conclusion. For instance, taking (17d), after all indicates that the proposition ‘he is a native speaker of French’, together with some other accessible proposition(s), such as ‘native speakers of French will pass the French exam’, form the premises of an argument, from which ‘he’ll pass the French exam’ follows as conclusion. In other words, after ull and you see indicate that the proposition they introduce is to be processed as a piece of evidence for the preceding one, or, equivalently, as explanatory of it. As she puts it: ‘The proposition introduced by you see must be relevant as an explanation. That is, it is relevant as an answer to a question raised by the presentation of the first proposition. . The same point applies to the connection expressed by after all’ (Blakemore 1987: 123). It is entirely in keeping with the observations already made that it is just these explanation-indicating conncctives that should not embed comfortably in an and-conjunction. The prohibition seems to be wider than this might indicate. There are adverbials such as then, in the process, consequentl?), thus, and us u result, which can collocate with and in a conjunction. They encode the sorts of conjunct relations which, as we have seen, are readily inferred pragmatically. Interestingly, there does not seem to be an explanation-indicating adverbial, which can function in a similar way, so as to force a ‘fact plus explanation’ interpretation. The adverbial explanatorily cannot be used in this way, as can be seen from (18): (18a) *Mary fell asleep at work and, explanatorily, she was exhausted. (18b) *Language is rule-governed and, explanatorily, it follows regular patterns. So the idea that the explanatory relation is not possible between the second and the first conjuncts does seem to be strongly supported. At this point, ’
As Burton-Roberts
(forthcoming)
notes, phrases
such as that is, in orher words, ./or t~.umpk,
are standard markers of apposition. As such it is to be expected that they would not occur in coordinate constructions. A fuller investigation of the issues here would require consideration of the different properties of co-ordination, subordination and apposition.
41
R. Carston / Conjunction, explanation and relevance
someone
inclined
towards
a semantic
account
might
feel that
Bar-Lev
and
Palacas’s notion of semantic command, in (10) above, could be simply and satisfactorily recast along these lines: the second conjunct cannot be interpreted as explanatory of the first. This would not be the right way to go, in my view. It would be a very odd sort of semantic feature. Furthermore, as will be seen from examples (19) and (20) in the next section, ‘explanation’ is not the whole story after all, so we need a more flexible restriction. Third, and most important, postulating a semantic feature would make the constraint on the interpretation of conjunctions seem to be an entirely arbitrary matter: words encode certain concepts and that’s all there is to it. But it surely isn’t an arbitrary fact about and-conjunctions that they cannot be interpreted in this way. In the next section I am going to suggest that it follows from a couple of very simple observations, one syntactic and the other pragmatic.
5. Towards a new solution An and-conjunction is a single syntactic unit; that is, it constitutes one of the infinitely many sentences that the grammar generates. The juxtaposed sentences, on the other hand, are two syntactic units, two distinct outputs of the grammar. Second, the principle of relevance states that every utterance carries a presumption of its own optimal relevance,8 which raises the question of what an utterance is; that is, of what the processing unit carrying this presumption is. The simplest assumption to make here, surely, is that an utterance unit is in some fairly direct correspondence with a grammatical unit. I think we should go along with this ‘simplest assumption’ unless we find some good reason not to. Blakemore (1987: 120) makes the same point about conjunctions, though she backs it with relevance-theoretic considerations. It follows from the principle of relevance that a hearer is entitled to assume that he will not be required to expend processing effort gratuitously; that is, the effort demanded will be adequately rewarded by cognitive effects. A hearer presented with a conjoined sentence has to undertake the processing that follows from the extra lexical and syntactic structure involved in conjoining; and he can, therefore, expect effects over and above those that might follow from the
8
In fact, it states that ‘every act of ostensive
its own optimal relevance ostensive communication.
(Sperber
and Wilson
communication’ 1986: IS),
communicates but, of course,
the presumption utterances
of
are acts of
conjuncts taken individually. Blakemore concludes from this that it is the complex conjoined proposition that carries the presumption of optimal relevance; the constituent propositions (the conjuncts) may be relevant in their own right but there is no automatic presumption that they will have adequate effects individually. Both syntactic and pragmatic considerations, then, support the claim that an and-conjunction constitutes a single utterance or speech act, and so carries the presumption of optimal relevance as a whole, while the juxtaposed clauses constitute two processing units, two speech acts, each carrying the presumption individually. If this is right, then the exclusion of the explanation interpretation from conjunctions follows directly. An explanation is typically an answer to a ‘why?’ or ‘how come?’ question and, as Blakemore (1987: 123) has put it: ‘questions and answers are by their very nature planned as separate utterances, each one satisfying the principle of relevance individually’. That is, the processing of the and-conjunction as a single pragmatic unit rules out the interpretation of the first conjunct as an independent unit which can raise implicit questions, that are then answered in the second conjunct. Now, the way this is phrased is broader than it would be if the only sort of question-answer sequence precluded were the ‘why?‘/‘because’, explanatory sort, which we have concentrated on up to now. This broader characterisation seems justified when we consider two further examples, suggested by Deirdre Wilson: (19a) I ate somewhere (19b) I ate somewhere
nice last week; I ate at Macdonald’s. nice last week and I ate at Macdonald’s.
(20a) I met a great actress (20b) I met a great actress
at the party; I met Vanessa Redgrave. at the party and I met Vanessa Redgrave.
It would surely be stretching the concept of explanation too far to apply it to the second utterances in (19a) and (20a); they do not give a cause or a reason for the event described in the first utterance, nor do they provide an analysis of any of the concepts in the first utterance. They do have an amplificatory function, though, and are readily understood as responses to questions. prompted by the first utterance. In fact, the first clauses of each of (19a) and (20a) seem specifically designed to raise the questions ‘where?’ and ‘who‘?’ respectively, which the second clauses answer. Again, conjoining these with and, as in (19b) and (20b), knocks out that interpretation and causes a strikingly different one to come to mind.
43
R. Carston 1 Conjunction, explanation and relevance
Notice that these differences between the (a) and (b) versions could not be accounted for at all by the semantic analysis in terms of semantic command, nor by any other semantic analysis for that matter. They follow, however, from the simple observations just given, concerning the grammatical status of and-conjunctions and their processing as single utterance units, meeting the pragmatic criterion of consistency with the principle of relevance as a whole. It may be that questions of the ‘who?‘, ‘what?‘, ‘where?‘, ‘when?’ sort have to be more obviously provoked, as they are in these cases, than does the ‘why?’ or ‘how come?‘, explanation-requiring, sort of question, whose insistent appearance has already been noted, in discussing the examples in (15). g Let us now see how the proposed account fares when faced with some examples which exhibit rather different properties from those of the cases considered so far. (21a) Jim has a new girlfriend. (21b) Jim has a new girlfriend
He goes to New York every weekend. and he goes to New York every weekend.
There are various possible interpretations, including a cause-consequence one, on which (21a) and (21b) are essentially the same, schematically ‘P (and) so Q’. But what I’m concerned with here are interpretations of the juxtaposition in (21a) which are not possible for the conjunction in (21b). The solution just proposed, for the restrictions on conjunction interpretation, predicts that the second conjunct of (21b) will not be able to function as an explanation of the fact given in the first conjunct; this indeed seems to be the case. As we have seen, this is the sort of relation that juxtaposed utterances, on the other hand, frequently enter into. However, the backwards causal relation which hearers are likely to understand for the juxtaposition in (21a) is different from that discussed so far. It is not the case that Jim’s going to New York every weekend is a cause or a reason for his having a new girlfriend. Rather, it is his going to New York that gives the speaker grounds for her belief that Jim has a new girlfriend; that is, it plays a causal role in her having this belief. It is not too surprising that this is a readily accessible interpretation for the juxaposed utterances, 9 As Mira Ariel has pointed out to me, subordinate clauses introduced by because or since are explanations, so it may be overly strong to say ‘fact plus explanation’ interpretations are absolutely
excluded
from single utterances.
The glib response
material can always override pragmatic constraints. However, response and agree that the issue calls for deeper consideration.
to this is that semantically
encoded
I don’t think that is an adequate
R. Carston / Conjunckm,
44
since, as is widely acknowledged, hecause-clause, as in the examples (22a) Jim
has
a new
it is often the intended in (22):
girlfriend,
weekend. (22b) Jane has left, because
explanation and rekwance
because
he goes
interpretation
to New
York
of a
every
her son isn’t here.
The two scope possibilities for because-clauses are clearly evident in (22b): it can be interpreted either as giving the reason for Jane’s having left, or as providing evidence for the belief that she has left. Further context is required to resolve this ambiguity. This ‘reason for believing’ interpretation of the juxtaposed utterances in (21a) is easily handled within relevance theory, which claims that an utterance may communicate several propositions explicitly (see Sperber and Wilson (1986) and Wilson and Sperber this volume). So, for instance, the first utterance in (21a), used literally, would have at least the following explicatures: (23a) Jim has a new girlfriend. (23b) The speaker is saying that Jim has a new girlfriend. (23~) The speaker believes that Jim has a new girlfriend. So a ‘why?’ question can be raised in reaction to any of these and the second utterance taken as explanatory of any of them. In the case of (21a), the second utterance, ‘he goes to New York every weekend’, is clearly to be taken as an explanatory response to (23~) the higher-level explicature representing the speaker’s propositional attitude, rather than to (23a), the proposition expressed, which has been the target of the explanation in all the previous juxtaposed cases.
6. Conjunction and logical consequence This account of the different interpretations of the utterances also be given for those in (24) another pair from Bar-Lev (1980: 144145): (24a) These are his footprints; (24b) These are his footprints
he’s been here recently. and he’s been here recently.
in (21) could and Palacas
45
R. Carston / Conjunction, explanation and relevance
On the ‘because’ sort of interpretation of (24a) (perhaps not the most accessible one), the fact that the ‘he’ referred to has been here recently could be taken as the evidence on which the speaker bases her belief that these are ‘his’ footprints (as opposed to anyone else’s). This interpretation would directly parallel that of (21a), and, as our account predicts, it is impossible for (24b), as it was for (21b). This, though, is not the meaning that Bar-Lev and Palacas have in mind: they take the second utterance in (24a) to be ‘a conclusion stemming forward from’ the first one. They are unable to account for why this is not possible for the conjunctive counterpart in (24b), since it is a forward-directed relationship and so quite different from the backwards causal and temporal cases that their ‘semantic command’ definition rules out. However, it is not clear that the analysis I’ve given so far can cover this exclusion either, since it doesn’t seem to be a case where the problem is one of trying to interpret the second conjunct as an answer to a question raised by the first. Some new considerations seem to be called for. While the addition of so to the second of the juxtaposed utterances simply makes the conclusion relationship of (24a) more explicit, embedding it in the conjunction seems to induce interpretive difficulty : (25a) These are his footprints; (25b) ? These are his footprints
so, he’s been here recently. and so he’s been here recently.
This is not to say that there is any general problem with so in conjunctions. Consider some of the various forward-directed cause-consequence andconjunctions that have come up throughout this paper: (26a) She fed him poisoned stew and (so) he died. (26b) The road was icy and (so) she slipped. (26~) Jim has a new girlfriend and (so) he goes weekend.
to New
York
every
Each of these examples is quite acceptable with so, which merely makes the cause-consequence relation between the conjuncts more explicit. Each is paraphrasable as ‘P and, as a result of P, Q’. This distinguishes them from (24) and (25) where the relation in question is one of logical consequence, or conclusion. As Blakemore (1987: 87ff., and 1988: 183) has emphasised, ‘consequence’ can refer either to a causal effect or to a logical conclusion; that is, to a causal connection between the states of affairs described, as in the
R. Carston / Conjunction, explanation and relevance
46
examples
in (26), or to an inferential
relation,
of premise
and
conclusion,
between two propositional representations. It is evident, then, from examples (25a) and (26) that SO can signal either of these kinds of consequence. However, it looks as if, when it is embedded in an and-conjunction, it has just the causee consequence enrichment-inducing function. If this is right, it provides a clue to the puzzle of why the conjunction in (24b) is unable to express a logical conclusion relation. First, though, this claim about SO needs some substantiation. The examples in (27) can be seen as various attempts to formulate a conjunctive utterance that captures the relation of logical consequence in (24a), ((27a) = (24b), (27b) = (25b)): (27) These are his footprints (a) *and he’s been here recently. (b) ? and so he’s been here recently. (c) and so I know he’s been here recently. (d) and so, clearly/without doubt, he’s been here recently. Intuitions are fairly delicate in this region, but at least some people lo seem to share those marked in (27). Note that these judgements concern just the forward conclusion interpretation; (27a), for instance, although starred. is quite acceptable on a different interpretation, not discussed here. As already noted, although better than (27a), (27b) does not seem to be completely satisfactory. This can be accounted for as follows: SO cannot have its logical-consequence-indicating role here, and its causeeconsequence-indieating role would lead to the nonsensical interpretation, ‘these are his footprints and as a result he’s been here recently’. However. once we have it explicitly given that the state of affairs described by the second conjunct is a knowledge state or belief, as in (27~) we have a thoroughly acceptable cause-consequence interpretation: ‘P and as a result I know Q’. Example (27d), in which sentence adverbials indicate the sort of propositional attitude involved, shows that, as long as some explicit indication of beliefhood is given, the causeeconsequence enrichment can be made. What seems to be required is that the conjuncts describe two states of affairs between which a causal relation can reasonably be thought to hold; an identification of a set of footprints and a belief about a particular person having been in that place are two such likely states of affairs.
I0
Bar-Lev
identical
and Palacas
(1980: 145) make the same judgements
to those in (27), except that they use thus rather
for a set of examples
than so.
which are
41
R. Carston / Conjunction, explanation and relevance
The question,
now, is why so cannot
have the function,
has in the case of (25a), of indicating an inferential consequence in these conjunction cases. My suggestion tial relations can only be specified between processing
which
it clearly
relation of logical is that such inferenunits. Lexical items
with this sort of function provide a speaker with a means of indicating to a hearer how the utterance they introduce is to be processed, in the context of the immediately preceding utterance or an otherwise highly accessible set of assumptions. Blakemore (1987, 1988) conceives of the class of ‘discourse connectives’, which includes so and after all, as guiding hearers in their bid for an optimally relevant interpretation of an utterance, by imposing constraints on the inferential (or pragmatic) computations they should perform. Since an and-conjunction is a single utterance, carrying the optimal relevance presumption as a whole, it is not possible to specify this sort of processing instruction for a single conjunct, a proper subpart of the utterance, which is not guaranteed optimally relevant on its own. Thus, the only consequence relation SO can manifest in a conjunction is that of causal consequence. In other words, when it occurs in a conjunction so functions as a constraint on the proposition expressed, or, equivalently, on the truthconditional content of the utterance, rather than on its inferential relation with the context. l l Returning to the pair of examples in (24) that prompted all this, it is now no surpise that the conjunction in (24b) cannot be understood as expressing the conclusion-drawing relation, since this is a relation that holds just between utterances, not within them. The nearest conjunctive equivalents to (24a) are (27~) and (27d), but what is an implicit inferential relation in (24a) is explicitly described in (27~) and (27d). It is painfully apparent that the ideas presented in this paper especially in this last section, are highly speculative and raise many further questions. However, whether this particular analysis proves to be on the right track or not, the general pragmatic approach advocated here does embrace a much wider range of examples than the semantic analysis and it opens up new directions in which to look for a more adequate account.
I1 This marks a departure of so is to constrain the distinctions
developed
from Blakemore (1988: inferential computations
by Wilson
and
Sperber
185) who claims that the sole function a proposition may enter into. Using
(this
volume,
and
forthcoming
b), I am
proposing that while so is indeed a constraint, that is, it encodes a procedure rather than a concept, it may function at either the implicit (inferential) level or at the explicit (representational) level.
R. Carston / Conjuncrion.
explanukm
und relevance
References Bar-Lev,
2. and A. Palacas,
1980. Semantic
command
over pragmatic
priority.
Lingua
51. 137-
146. Blakemore,
D., 1987. Semantic
Blakemore,
D., 1988. ‘So’ as a constraint
tions: The interface
constraints
on relevance. on relevance.
between language
and reality,
Oxford:
Blackwell.
In: R. Kempson 183-195.
(ed.), Mental
representa-
Cambridge
University
Cambridge:
Press. Burton-Roberts,
N., forthcoming.
tics. Oxford:
Pergamon
Apposition.
In: The Encyclopedia
Press, and Aberdeen:
Aberdeen
Carston, R., 1988. Implicature, explicature and truth-theoretic Mental representations: The interface between language Cambridge University Press. Reprinted Oxford: Oxford University Press. Carston,
R.,
Linguistics Carston,
1990. Quantity
Carston, Cohen,
R., forthcoming.
and
Conjunction
Oxford:
generahsed
and pragmatic
Pergamon
R., in preparation.
semantics. In: R. Kempson (ed.). and reality, 1555181. Cambridge:
(ed.),
Pragmatics:
implicature.
UCL
A reader. Working
effects. In: The Encyclopedia
Press, and Aberdeen:
Conjunction,
L. J., 1971. Some remarks
In: Y. Bar-Hillel
in: S. Davis
and Linguis-
Press.
33351.
Papers
juxtaposition
of natural
University
of Language Press.
and relevance.
on Grice’s views about
(ed.), Pragmatics
Aberdeen
the logical particles
languages,
5&68.
of natural
Dordrecht:
language
Reidel.
Grice, H.P., 1967. Logic and conversation. The William James Lectures. In: H.P. G-ice. Studies in the way of words, 22240. Cambridge, MA: Harvard University Press. Harnish,
R., 1976. Logical
integrated
theory
form and implicature.
of linguistic
ability,
Levinson, S., 1987. Minimization celli-Papi (eds.), The pragmatic Posner,
R.,
J. Searle, Dordrecht Schmerling,
1980. Semantics F. Kiefer,
and pragmatics
M.
In: T. Bever, J. Katz, T. Langendoen
3 133391. New York
Bierwisch
(eds.).
of sentence Speech
connectives act
1989.
(eds.). An
: Crowell.
and conversational inference. In: J. Verschueren. perspective, 61-129. Amsterdam: Benjamins. theory
in natural and
M. Bertuclanguage.
pragmatic%
In:
1688203.
: Reidel. S., 1975. Asymmetric
(eds.), Syntax Sperber, Sperber,
in
I.
2, l-3
and Linguistics.
maxims
of Language
University
and semantics,
conjunction
and rules of conversation.
Vol. 3: Speech acts, 211-232.
New York:
In: P. Cole, J. Morgan Academic
Press.
D. and D. Wilson, 1986. Relevance: Communication and cognition. Oxford: D. and D. Wilson, 1987. Precis of Relevance: Communication and cognition.
and Brain Sciences
10, 697-710.
Wilson,
D. and D. Sperber,
this volume.
Linguistic
Wilson,
D. and D. Sperber,
forthcoming
a. Pragmatics
form and relevance.
Wilson,
D. and D. Sperber,
forthcoming
b. Varieties
and time. of non-truth-conditional
meaning.
Blackwell. Behavioral
90 (1993) 2748.
Conjunction, Robyn
27
North-Holland
explanation and relevance
Carston*
Department of Phonetics and Linguistics, University College London, Gower Street, London WCIE 6BT. UK
1. Introduction
The ground-breaking work of Paul Grice (1967) has established a fundamental distinction: between the intrinsic semantics of linguistic expressions, on the one hand, and the pragmatic, context-dependent, content of an utterance, on the other. This judicious division of labour is well illustrated in the account of what is communicated by utterances of and-conjunctions, such as those in (1): (la) (lb) (lc) (Id) (le)
It’s spring in England and it’s autumn in New Zealand. He handed her the scalpel and she made the incision. We spent the day in town and I went to Harrods. She fed him poisoned stew and he died. I left the door open and the cat got in.
The word and is taken to be pretty well semantically empty; that is, it is taken to be the natural language equivalent of the truth-functional logical conjunction operator. Pragmatics takes care of the great variety of temporal, causeconsequence and other sorts of relations understood to hold between the states of affairs described. Some of these come through in the so-called asymmetrical examples in (lb)-(le). For instance, we take it that the making
*
This is a revised and expanded
version
of a paper
by the same name, which
appears
in the
fourth volume of UCL Working Papers in Linguistics, 1992. I am very grateful to Deirdre Wilson for useful and enjoyable discussions on the concerns of this paper. My debt to the work of Diane Blakemore is evident throughout the paper. I have also been helped and encouraged by Mira Ariel, Noel Burton-Roberts, John Durkin, Marcus Giaquinto, Dick Hudson, Neil Smith, Dan Sperber and Vladimir Zegarac.
0024-3841/93/$06.00
0
1993 -
Elsevier
Science Publishers
B.V. All rights reserved
28
R. Curston 1 Conjunction, explanution und relevuncr
of the incision in (lb) followed the handing over of the scalpel and that an interval of a few seconds intervened. A quite different temporal relation is understood to hold between the states of affairs described in (lc), the event of going to Harrods is interpreted as contained within the period of time spent in town. Different sorts of consequence relations are understood in (Id) and (le): while the feeding of poisoned stew is a sufficient cause for death, the leaving open of the door is just one of a range of factors contributing to the cat’s getting in. These few examples by no means exhaust the range of different temporal, cause-consequence and other relations that conjunctions can communicate. Some further possibilities will be mentioned in the next section and others have been discussed by Schmerling (1975) Posner (1980) and Wilson and Sperber (forthcoming a). According to the pragmatic approach, a hearer recovers these connections inferentially. This involves an interaction of the decoded semantic content with general knowledge assumptions about the way things connect up and relate in the world, an interaction constrained by some general criterion or criteria of rational communicative behaviour.
2. The relevance-based pragmatic account It is not my purpose here to argue for the superiority of one pragmatic approach over another. Rather, I shall simply assume the relevance-theoretic framework and work within it. The fundamental idea is that, in processing an utterance, a hearer is looking for an optimally relevant interpretation; that is, an interpretation that has the two properties given in (2): (2) An utterance, on a given interpretation, is optimally relevant iff: (a) it achieves enough effects to be worth the hearer’s attention; (b) it puts the hearer to no gratuitous effort in achieving those effects. (Wilson
and Sperber
forthcoming
a)
What Sperber and Wilson call the principle of relevance is the thesis that every utterance communicates the presumption of its own optimal relevance. The pragmatic criterion that hearers actually employ in utterance interpretation is one of consistency with the principle of relevance: an interpretation has this property if and only if a speaker could have rationally expected it to be optimally relevant to the hearer. Once the hearer has accessed an interpretation which meets this criterion, he looks no further but takes this to be the
R. Carston 1 Conjunction,
explanation
29
and relevance
interpretation the speaker intended. The implications of this definition are fully discussed elsewhere (Sperber and Wilson 1986, 1987) so I shall do no more here than give a brief outline of how this pragmatic criterion applies in the interpretation of one of the examples above. As with any utterance, there are many possible interpretations of (lb) which are compatible with the linguistically encoded, semantic, content. Two of these logical possibilities for (lb) are given in (3): (3a) He handed her the scalpel and a second or two later she made the incision with that scalpel. (3b) He handed her the scalpel and simultaneously she made the incision with her pocketknife. Now, obviously, although these are both possible and consistent, the first one is overwhelmingly more likely to be recovered by the hearer, and to have been intended by the speaker, than the second one. Why is this? The intuitive answer is that it is just the obvious, natural, normal, etc., assumption to make. Put in slightly more theoretical terms, most adults, within a particular cultural range, have a mental script concerning the making of incisions with scalpels, perhaps as a subpart of a broader surgical operation script, involving a co-operative interaction of a nurse and a doctor. So when we hear or read (lb), a chunk of stereotypical material of this sort, in memory, is activated. On this basis, we instantly assume that the making of the incision followed the handing over of the scalpel, that the scalpel was used for the making of the incision and that the interval of time between the two actions was a few seconds at most; that is, the proposition is enriched along the lines of (3a). The relevance-based pragmatic criterion accounts for why we can safely rely on stereotypic knowledge in this instance, rather than considering other possible lines of interpretation, such as (3b). l The second part of the definition of optimal relevance, concerning the minimising of processing effort, is crucial here. An important factor, contributing to the effort involved r
Some other
instance, instruction
pragmatic
the principle
accounts
of informational
to derive implicatures
give a lot of weight enrichment,
to stereotypic
in Levinson
which are stereotypical
interpretation
(1987: 65-68)
enrichments
too;
for
is essentially
an
of the information
encoded
by the utterance. However, this principle makes some false predictions and offers nothing by way of explanation for why hearers should enrich in a stereotypic way (when they do) instead of in more unusual (and interesting) ways. See Carston (1990), Carston Sperber (forthcoming a), for criticisms of Levinson’s account.
(forthcoming)
and Wilson and
30
in deriving
R. Carston / Conjunction, rxplanahn
contextual
effects,
is the accessing
and rekvancr
of contextual
assumptions.
Now, in this particular example, the very concepts encoded in the sentence give immediate access to a ready-made script, requiring no imaginative effort of construction on the part of the hearer/reader. So (3a) is simply the first interpretation to come to mind. Then, assuming it has an adequate range of effects that the speaker could have manifestly foreseen, no other possibility is considered. Any of the other logically possible interpretations, such as (3b), would be much less accessible, thus requiring considerably more effort to construct, and no one of these further possibilities would be more obviously available than dozens of others. Note that (3b) might have just as wide a range of effects as (3a), or, conceivably, an even richer array (for instance: she doesn’t expect help from him, he is too slow, there is something wrong with the scalpel, etc.). However, the guarantee is not one of maximal effects; all that hearers can expect from speakers is that their utterances will have enough effects to justify the call on their attention. It follows that a speaker who wanted to communicate something other than (3a) would not be able to do so by uttering (lb) and, if functioning rationally, would not attempt to do so. The relevance-based account of the different cause-consequence relations in (Id) and (le) would run in essentially the same way. Each would activate a stereotypic script or schema, with, in the case of (Id), a built-in assumption about poison causing death, and, in the case of (le), assumptions about factors enabling cats to get in. I am assuming that these various enrichments of the bare linguistic meaning are instances of pragmatically derived content which contributes to the proposition expressed by the utterance, to ‘what is said’ in Gricean terms. That is, they are not implicatures, contrary to the assumption of Grice and his followers. Sperber and Wilson (1986: 188-192) give a range of examples which show that, in addition to disambiguation and reference assignment, there are many sorts of pragmatic enrichment which a hearer may have to recover, in order to arrive at the proposition explicitly expressed. With regard to the specific case of conjunct relations, I have argued the case for enrichment of the proposition expressed, at length, elsewhere (Carston 1988 and forthcoming). There is more to conjunction enrichment than the story told so far. There are cases where the temporal, cause-consequence, teleological and other assumptions do not spring out of ready-made scripted knowledge, or where two readily available scripts provide contradictory information about causee consequence relations :
31
R. Carston 1 Conjunction. explanation and relevance
(4a) (4b) (4~) (4d)
She screamed at him and he hit her. He hit her and she screamed at him. ? She went to bed and brushed her teeth. He yelled and the light came on.
The immediate assumption in interpreting (4a) is that the hitting is a consequence of the screaming, while in (4b) the screaming is a consequence of the hitting. Arguably, we have fairly accessible scripts or schemas reflecting both scenarios, but what dictates which of the two schemas comes into play in each case? The intuitive answer is that the order in which the information is presented in the utterances is taken to reflect the order in which the events occurred. This secures the sequential interpretation and the cause-consequence relation follows, in accordance with the script whose event sequence matches that of the utterance. The oddness of (4~) seems to arise from a clash between this order-of-presentation assumption and the familiar script in which teeth-brushing precedes going to bed. However, as will be evident in the next section, there is no pragmatic maxim or discourse principle requiring that the order of information in utterances reflect the order of events in the world. So our account of how the sequentiality enrichment arises, when it does, is not yet complete. The sort of explanation I would look for would be a general cognitive one, lying with our information-processing capacities and their bid to minimise effort expenditure. It should then follow from considerations of optimal relevance that conjunct order reflects the order of events in the world.2 Finally, the case of (4d). This does not map onto a specific mental script (an act of yelling followed by a light coming on is not a stereotypic sequence of events) and yet there is a strong tendency to assume a causeeconsequence connection here too. This raises the possibility that there may be something more fundamental at work in the imputing of causal relations, a very general procedure, perhaps, which licenses the move to causeeconsequence assumptions when a sequential relation is taken to hold between two states of affairs. It might be that the input to such a move would have to be broader than temporal sequentiality since that is not the only sort of temporal relation which can accompany a consequence relation. Consider the following examples, first pointed out by Schmerling (1975), and discussed more fully by Wilson and Sperber (forthcoming a):
2
A preliminary
effort considerations
attempt
at accounting
was made in Carston
for the sequentiality (1990).
enrichment
in terms of processing
32
R. Carston / Conjunction, explanation and relevance
(5a) The lights were off and I couldn’t see. (5b) Susan is underage and can’t drink. (5~) I spoke to Sam at the party and discovered
that he was unwell.
Examples (5a) and (5b) are cases for which the time-span of the state of affairs described by the second conjunct fully overlaps the time-span of the state of affairs described by the first. That is, the conjuncts do not form a temporal sequence, yet the second is understood as a consequence of the first. (5~) plainly involves a consequential relationship too and there is some sort of temporal priority of the first mentioned state of affairs over the one mentioned second: some of the speaking had to have taken place before the discovery could be made. But this temporal relation is appropriately paraphrased by ‘and in the process’, rather than by ‘and then’, which would give a different reading, one on which the conversation with Sam would not be seen as bringing about the discovery. So, if there is some more general schema or heuristic at work in the interpretation of some of these examples, it would be accessed on the basis of a temporal relation along the lines of ‘the onset of P precedes the onset of Q’. An utterance whose temporal structure matched this would then give access to the further assumption, ‘Q is a consequence of P’. Note that this is not to be understood as a context-free rule. There are utterances which meet the temporal requirement but are not further enriched to a cause-consequence relation. Standard assumptions concerning the limits of human causal powers block a cause-consequence interpretation of (6) for instance: (6) I walked
to the front gate and a fire engine
came around
the corner.
If this accessing strategy exists, it is, like heuristics in general, a useful rule of thumb, effort-saving and often, though not always, correct. I believe that the existence of a heuristic of this sort would be very natural within the relevance-theoretic framework, since causal stories are far more likely to yield a crop of implications, that is, have cognitive effects, than are mere temporal contiguities.3 Certainly, this is all very speculative and calls for deeper consideration than I can give it here.
3
This apparent
of course,
drive toward
not restricted
seeing states of affairs as glued together
to utterance
interpretation.
It is quite fundamental
in a great causal to our cognitive
mesh is. life in
general: it affects our perception, reasoning and memory, and emerges at a very early stage in development. See Carston (in preparation) for a survey of some of the vast literature on this.
R. Carston / Conjunction, explanation and relevance
33
3. Conjunction versus juxtaposition Various arguments have been put forward for favouring a pragmatic account of conjunct relations over a richer lexical semantics for and. One of these focuses on the point that the very same temporal, consequential, etc., relations arise when the and is removed. Compare (7a,b,c) with their conjunctive equivalents in (1): (7a) He handed her the scalpel. She made the incision. (7b) She fed him poisoned stew. He died. (7~) I left the door open. The cat got in. We find, for instance, that (7a) communicates that he handed her the scalpel before she made the incision, just as (lb) does, and similarly for the other connections. So, the argument went, these imputed relations don’t seem to have anything to do with the meaning of and. Rather, they are the product of some quite general cognitive predisposition to forge certain connections and relations between states and events whenever it seems reasonable to do so. That is, rather than being arbitrary linguistic facts, they are a matter for a cognitively-based pragmatics. This is a plausible, if not knock-down, argument; putting it together with several other arguments made a strong case for the pragmatic treatment. However, while it may be the case that all the relations communicable by an and-conjunction are still communicated when the and is removed, the converse does not seem to be the case. That is, there are some relations that are communicated by juxtaposed sentences which apparently cannot be communicated when these sentences are conjoined by and. This was first pointed out by Herbert Clark with a pair of examples, essentially the same as those given in (8): (8a) John broke his leg. He slipped on a banana skin. (8b) John broke his leg and he slipped on a banana skin. For (8a), it is quite possible, in fact more or less inevitable, that we understand the event of slipping on a banana skin as the cause of the event of leg-breaking; that is, the event mentioned second is understood as having caused, and hence as having preceded, the event mentioned first. This does not seem to be possible for (8b), where the sentences are presented in the same order but conjoined with and. The same disparity arises between (9a) and (9b), and a range of other cases:
34
R. Carston / Cortjunction, explanation and relevance
(9a) Mary fell asleep at work; she was exhausted. (9b) Mary fell asleep at work and she was exhausted. This is just
one of the relations
or connections
between
two juxtaposed
sentences which are precluded when they are conjoined with and. In an interesting but largely neglected paper, published in this journal in 1980, BarLev and Palacas surveyed the extent of this phenomenon and the variety of its manifestations. They pointed out that, whatever the source of the meaning difference is, it cannot be to do with the order of the constituent clauses since this ordering is the same for the two members of each pair. So there is no possible resort here to any pragmatic maxim concerning the order in which we communicate material, such as Grice’s manner maxim of orderliness or Harnish’s mirroring maxim: ‘in so far as possible, make the order of saying reflect the order of events’ (Harnish 1976: 359). The (a) members of the pairs in (8) and (9) provide compelling evidence against the existence of any such maxims, assuming that no discourse principle should be so limited in its application as to apply only to cases of explicit conjunction.4 Bar-Lev and Palacas (1980: 141) went on to claim that the inescapable conclusion is that there is more to natural language and than simple truth-functionality. They proposed that this extra element of meaning that and encodes is their relation of semantic command: (10) Semantic command: The second conjunct (S”) is not prior to the first (S’) (chronologically or causally). This idea does have some attractive features. First, it seems to encompass the whole spectrum of temporal and consequence relationships which conjuncts can have (a sample of which has been given in (1) and (5)) and to exclude those that they cannot have; that is, ‘backwards’ sequence and consequence. Second, although Bar-Lev and Palacas did not say so, it is clear that this semantic account still leaves a vast amount of appropriate work for pragmatics, in determining, for any particular case of a conjunctive utterance, precisely which, if any, temporal and causeeconsequence connection is to be taken to pertain. The relevance-based pragmatic analysis given in section 2 would go through, without any obvious changes being required, taking as its input
4 This point is argued further and Sperber (forthcoming a).
in Carston
(forthcoming),
Carston
(in preparation).
and Wilson
R. Carston 1 Conjunction, explanation and relevance
35
the ‘semantic command’ characterisation of conjunction. Third, this semantics for and avoids a range of problems which other attempts to give a semantic account fall prey to. The semantic ambiguity account and the univocal multi-featured semantic account, the latter due to Cohen (1971), wind up having to proliferate a potentially infinite number of senses or semantic features, in their bid to account for the full set of subtly different temporal and consequence relations that different examples communicate. Furthermore, some of these senses or features contradict others, for instance, simultaneity and sequentiality.5 So Bar-Lev and Palacas’s account seems to be the most promising of the semantic options on offer. However, it is not adequate. It is really nothing more than a descriptive statement of what we have just observed, which is that the ‘backwards’ causal and temporal relations possible for juxtaposed sentences are excluded when they’re conjoined with and. Surely what we want to know is why this is the case; it just looks like the kind of fact that ought to be able to be explained rather than merely recorded. Second, the definition is probably redundantly broad, as it stands. Juxtaposed sentences seem to share part of the restriction, which ‘semantic command’ places on conjunctions: in the absence of a backward causal relation, it appears to be just as difficult to get a reverse temporal sequence for them as it is for conjunctions: (1 la) (11 b) (1 lc) (1 Id) (11 e)
He She She He He
handed her the scalpel and she made the incision. made the incision and he handed her the scalpel. made the incision; he handed her the scalpel. rode into the sunset and jumped onto his horse. rode into the sunset. He jumped onto his horse.
There is a strong tendency to take the incision-making in (1 lb) to have preceded the handing over of the scalpel, even though it does not match the familiar script. It is this that the semantic command requirement is a response to. However, the point here is that the same goes for (1 lc): its temporal interpretation goes the way of (1 lb) rather than (1 la), given that the linguistically encoded content allows for either. The examples in (1 Id) and (1 le) further demonstrate this: (1 le) is peculiar in the same way that (1 Id) is. It is not the case that the juxtaposed example is interpreted with the event described by the second sentence understood as preceding that of the first
5 Posner (1980) provides an excellent overview position and compelling criticisms against it.
of the ‘meaning-maximalist’
(i.e. rich semantic)
36
R. Carston / Conjunction.
e.uplunation and relevance
sentence, although this is the stereotypical sequence. To this extent. then, conjunctions and juxtapositions are similarly restricted, which suggests that there is a more general story (hinted at in section 2) to be told about temporal sequencing, which applies to both of them, and that this need not be incorporated into the lexical semantics of and. Third, and most important, there are counterexamples to this semantics for and. These fall into two classes. The first set concerns those for which a socalled ‘backwards’ causal relation is understood despite the presence of and. The second set involves different sorts of relationships altogether, having nothing to do with temporal or causal links between states of affairs. Examples of the first sort are given in (12) and ( 13):
(12) If the old king has died of a heart attack
and a republic has been formed, and the latter event has caused the former, then Tom will be upset. (13) A: Did Bill break the vase? B: Well, the vase broke and Bill dropped it. (example due to Larry Horn) The point about (12) is that on the semantic command analysis the conjunctive antecedent of this conditional should have a contradictory feel to it: we are told that a cause-consequence relation between the second conjunct and the first is precluded, and then that just such a relation holds. There is of course no such feeling of inconsistency or tension; the encoded content of the and is just the minimal truth-functional connection, which is perfectly compatible with the further specification of a backwards causal relation. Consider now the more interesting case of (13) focusing on B’s reply to A’s question. Here there is no linguistic specification of a backwards causal relation, as there is in (12), but the event of dropping the vase is readily understood as having preceded and caused the breaking of the vase. This should not be possible if and really does encode the property of ‘semantic command’. Certainly B’s utterance is an indirect, round-about, sort of way of communicating the causeeconsequence relation; it is an implicature rather than an aspect of the explicit content. 6 The hearer is required to do some extra inferential
6
work in order to arrive at it, and no doubt
The explicit/implicit
elsewhere: Sperber (this volume).
distinction
and Wilson
is an important
(1986), Carston
and contentious (1988), Carston
receives extra effects in
one, discussed
at some length
(1990) and Wilson
and Sperber
R. Carston 1 Conjunction, explanation and relevance
31
the process, as the definition of optimal relevance predicts. The example deserves further analysis but the important point for me here is that, whatever the effects this conjunctive response has, they are not derived via a contradiction, which is what the analysis of and in terms of semantic command would require. An instance of the second sort of counterexample, in which temporal cause-consequence relations are not at issue, is given in (14) : (14a) Language (14b) Language (14~) Language
and
is rule-governed; it follows regular patterns. is rule-governed and it follows regular patterns. follows regular patterns and it is rule-governed.
In (14a) the second sentence is understood as a spelling out of, or elaboration on, the first one. This relationship seems to be another one which is precluded when the two sentences are conjoined with and, in whichever order, as in (14b) and (14~). These examples are cited by Bar-Lev and Palacas (1980: 144) later in their paper, which is odd, since if their semantic command analysis makes any prediction about them it makes the wrong one. It predicts that the conjunction in (14b) should be able to be interpreted in the same way as the juxtaposed sentences, since all that semantic command precludes is backwards causal relations. It seems clear that this attempt to account for the constraints on the interpretation of conjunctions has to be abandoned. It is both too restrictive, since, in some contexts, backwards causal relations are possible, and not restrictive enough, since it does not account for why the conjunction in (14b) cannot be interpreted in the same way as the non-conjoined clauses in (14a).
4. Explanation If we do away with the semantic account we are back where we began with the problem of the non-equivalence of the pairs in (8) and (9), and, in addition, we now have the problem of (14a) and (14b). Let’s focus for a moment on the juxtaposed, non-conjoined cases, (8a) and (9a): what’s going on in these examples is that the second sentence is being understood as providing an explanation for the state of affairs described in the first. Now, an explanation of something standardly involves the citing of a cause or a reason for that thing; it is often prompted by a ‘why?’ question or a ‘how did it come to be so?’ sort of question. What is interesting here is how dominant this sort of interpretation is for the juxtaposed cases.
38
R. Carsron / Conjunction,
(Isa) (15b) (15~) (15d)
Sue Sue Max Max
explanation
and relevance
is happy today. She got/made a phone-call. is miserable today. She got/made a phone-call. can’t read; he’s a linguist. can’t read and he’s a linguist.
It is quite natural and immediate in interpreting (15a) to take the getting or making of the phone-call, mentioned in the second sentence, to be the reason for Sue’s feeling happy, though such a state of affairs is entirely neutral as regards any standard assumptions about causes of happiness. We are no less inclined to take it as explanatory of her misery, when it is presented in the different juxtaposition in (15b). Even more surprising is (1 SC), where the immediate interpretive strategy seems to involve taking the property of being a linguist as explanatory of illiteracy, even if this is subsequently dismissed or taken as a joke. Compare this with the conjunctive counterpart in (ISd), where the most accessible interpretation is one in which the conjuncts are understood as describing two contrasting facts about Max, an interpretation which is entirely in line with our standard assumptions about the way the properties relate to each other. Faced with these examples, we might well feel that what needs accounting for is the dominance of this ‘fact followed by explanation’ type of interpretive strategy for the juxtaposed cases. In this paper, however, I’m just going to accept that this is so, that we are question-asking, explanation-seeking creatures, and return to the main issue here, which is the preclusion of certain sorts of interpretation from the andconjunctions. Could it be that what it all comes down to is that the second conjunct cannot be interpreted as explanatory of the first’? I think this comes quite close to the answer and is worth pursuing to see how far it takes us. Stating the condition this way not only covers the restriction on the interpretation of (8b) and (9b), but it also extends to (14b) and (14~). Although there is no issue of causality in these cases, there is an issue of explanatoriness. The second clause in (14a), ‘it follows regular patterns’, is given as an explanation of the notion of ‘rule-governedness’; it is a conceptual, or analytic, explanation as opposed to a causal one. This distinction is demonstrated by the two different ways in which the ‘why?’ questions in (16) can be interpreted and answered: (16) Why is John a bachelor? (a) Because he doesn’t like women. (b) Because he’s unmarried.
Why is Mary so witty? (a) She wants to impress us. (b) She says unexpected things.
R. Carston / Conjunction,
explanation
John’s dislike of women would enter into a bachelorhood while his unmarriedness is part tion of what it is for him to be a bachelor and for the responses regarding Mary’s wittiness.
39
and relevance
causal explanation of his of the conceptual explanathe same distinction holds So asserting that the ‘fact
plus explanation’ interpretation is impossible for conjunctions mops up not just the prohibited backwards causal relations, as in (8b) and (9b), which the semantic command analysis sought to preclude, but also cases such as (14) about which it was silent. Furthermore, it is not at odds with the interpretation of (12) or (13B), as it does not absolutely forbid an interpretation in which a state of affairs described in the second conjunct is understood as being in a causal relation with that described in the first. What it does do is account for the rarity of this sort of relation, since understanding a speaker as giving a cause or a reason for something is, typically, to understand her as presenting an explanation of it. In the context of A’s question in (13), B’s conjunctive response, ‘the vase broke and Bill dropped it’, presents A with two facts, rather than a fact followed by an explanation of that fact, though, via an inferential process, a hearer will no doubt derive the implication that the dropping caused the breaking. Further evidence, in favour of explanation as the crucial notion in accounting for the constraints on the conjunct relations, comes from a consideration of the class of so-called discourse connectives, which Blakemore (1987) has studied in some detail. These include expressions such as so, therefore, moreover, after all and you see. These form an interesting class of connectives since, it is argued, they do not affect the truthconditional content of utterances in which they occur, unlike such connectives as because, while, before, in order that, etc., but rather have the function of instructing the hearer how the proposition they introduce is to be processed. That is, they indicate that this proposition is to be placed in a certain inferential relation with some other, usually the one expressed by the immediately preceding utterance. Now, what is of interest to us here is that some of these seem to be embeddable in an and-conjunction, while others do not:
(17a) (17b) (17~) (17d)
The road was icy and so she slipped. She’s good-looking and moreover her father’s rich. *She slipped and you see the road was icy. *He’ll pass the French exam and after all he is a native
speaker.
40
R. Carston 1 Conjunction,
(17e)
*Language all it follows
explanation
and relevance
is rule-governed and that is/in other words/you regular patterns. ’
Of course, in all five cases, taking
seejqfter
out the and to render them non-conjunctive
makes them all equally acceptable. On Blakemore’s analysis, what distinguishes after all and you see from the others is that the sort of inferential role they indicate for the following proposition is one of being a premise in an argument which has the preceding proposition as conclusion. For instance, taking (17d), after all indicates that the proposition ‘he is a native speaker of French’, together with some other accessible proposition(s), such as ‘native speakers of French will pass the French exam’, form the premises of an argument, from which ‘he’ll pass the French exam’ follows as conclusion. In other words, after ull and you see indicate that the proposition they introduce is to be processed as a piece of evidence for the preceding one, or, equivalently, as explanatory of it. As she puts it: ‘The proposition introduced by you see must be relevant as an explanation. That is, it is relevant as an answer to a question raised by the presentation of the first proposition. . The same point applies to the connection expressed by after all’ (Blakemore 1987: 123). It is entirely in keeping with the observations already made that it is just these explanation-indicating conncctives that should not embed comfortably in an and-conjunction. The prohibition seems to be wider than this might indicate. There are adverbials such as then, in the process, consequentl?), thus, and us u result, which can collocate with and in a conjunction. They encode the sorts of conjunct relations which, as we have seen, are readily inferred pragmatically. Interestingly, there does not seem to be an explanation-indicating adverbial, which can function in a similar way, so as to force a ‘fact plus explanation’ interpretation. The adverbial explanatorily cannot be used in this way, as can be seen from (18): (18a) *Mary fell asleep at work and, explanatorily, she was exhausted. (18b) *Language is rule-governed and, explanatorily, it follows regular patterns. So the idea that the explanatory relation is not possible between the second and the first conjuncts does seem to be strongly supported. At this point, ’
As Burton-Roberts
(forthcoming)
notes, phrases
such as that is, in orher words, ./or t~.umpk,
are standard markers of apposition. As such it is to be expected that they would not occur in coordinate constructions. A fuller investigation of the issues here would require consideration of the different properties of co-ordination, subordination and apposition.
41
R. Carston / Conjunction, explanation and relevance
someone
inclined
towards
a semantic
account
might
feel that
Bar-Lev
and
Palacas’s notion of semantic command, in (10) above, could be simply and satisfactorily recast along these lines: the second conjunct cannot be interpreted as explanatory of the first. This would not be the right way to go, in my view. It would be a very odd sort of semantic feature. Furthermore, as will be seen from examples (19) and (20) in the next section, ‘explanation’ is not the whole story after all, so we need a more flexible restriction. Third, and most important, postulating a semantic feature would make the constraint on the interpretation of conjunctions seem to be an entirely arbitrary matter: words encode certain concepts and that’s all there is to it. But it surely isn’t an arbitrary fact about and-conjunctions that they cannot be interpreted in this way. In the next section I am going to suggest that it follows from a couple of very simple observations, one syntactic and the other pragmatic.
5. Towards a new solution An and-conjunction is a single syntactic unit; that is, it constitutes one of the infinitely many sentences that the grammar generates. The juxtaposed sentences, on the other hand, are two syntactic units, two distinct outputs of the grammar. Second, the principle of relevance states that every utterance carries a presumption of its own optimal relevance,8 which raises the question of what an utterance is; that is, of what the processing unit carrying this presumption is. The simplest assumption to make here, surely, is that an utterance unit is in some fairly direct correspondence with a grammatical unit. I think we should go along with this ‘simplest assumption’ unless we find some good reason not to. Blakemore (1987: 120) makes the same point about conjunctions, though she backs it with relevance-theoretic considerations. It follows from the principle of relevance that a hearer is entitled to assume that he will not be required to expend processing effort gratuitously; that is, the effort demanded will be adequately rewarded by cognitive effects. A hearer presented with a conjoined sentence has to undertake the processing that follows from the extra lexical and syntactic structure involved in conjoining; and he can, therefore, expect effects over and above those that might follow from the
8
In fact, it states that ‘every act of ostensive
its own optimal relevance ostensive communication.
(Sperber
and Wilson
communication’ 1986: IS),
communicates but, of course,
the presumption utterances
of
are acts of
conjuncts taken individually. Blakemore concludes from this that it is the complex conjoined proposition that carries the presumption of optimal relevance; the constituent propositions (the conjuncts) may be relevant in their own right but there is no automatic presumption that they will have adequate effects individually. Both syntactic and pragmatic considerations, then, support the claim that an and-conjunction constitutes a single utterance or speech act, and so carries the presumption of optimal relevance as a whole, while the juxtaposed clauses constitute two processing units, two speech acts, each carrying the presumption individually. If this is right, then the exclusion of the explanation interpretation from conjunctions follows directly. An explanation is typically an answer to a ‘why?’ or ‘how come?’ question and, as Blakemore (1987: 123) has put it: ‘questions and answers are by their very nature planned as separate utterances, each one satisfying the principle of relevance individually’. That is, the processing of the and-conjunction as a single pragmatic unit rules out the interpretation of the first conjunct as an independent unit which can raise implicit questions, that are then answered in the second conjunct. Now, the way this is phrased is broader than it would be if the only sort of question-answer sequence precluded were the ‘why?‘/‘because’, explanatory sort, which we have concentrated on up to now. This broader characterisation seems justified when we consider two further examples, suggested by Deirdre Wilson: (19a) I ate somewhere (19b) I ate somewhere
nice last week; I ate at Macdonald’s. nice last week and I ate at Macdonald’s.
(20a) I met a great actress (20b) I met a great actress
at the party; I met Vanessa Redgrave. at the party and I met Vanessa Redgrave.
It would surely be stretching the concept of explanation too far to apply it to the second utterances in (19a) and (20a); they do not give a cause or a reason for the event described in the first utterance, nor do they provide an analysis of any of the concepts in the first utterance. They do have an amplificatory function, though, and are readily understood as responses to questions. prompted by the first utterance. In fact, the first clauses of each of (19a) and (20a) seem specifically designed to raise the questions ‘where?’ and ‘who‘?’ respectively, which the second clauses answer. Again, conjoining these with and, as in (19b) and (20b), knocks out that interpretation and causes a strikingly different one to come to mind.
43
R. Carston 1 Conjunction, explanation and relevance
Notice that these differences between the (a) and (b) versions could not be accounted for at all by the semantic analysis in terms of semantic command, nor by any other semantic analysis for that matter. They follow, however, from the simple observations just given, concerning the grammatical status of and-conjunctions and their processing as single utterance units, meeting the pragmatic criterion of consistency with the principle of relevance as a whole. It may be that questions of the ‘who?‘, ‘what?‘, ‘where?‘, ‘when?’ sort have to be more obviously provoked, as they are in these cases, than does the ‘why?’ or ‘how come?‘, explanation-requiring, sort of question, whose insistent appearance has already been noted, in discussing the examples in (15). g Let us now see how the proposed account fares when faced with some examples which exhibit rather different properties from those of the cases considered so far. (21a) Jim has a new girlfriend. (21b) Jim has a new girlfriend
He goes to New York every weekend. and he goes to New York every weekend.
There are various possible interpretations, including a cause-consequence one, on which (21a) and (21b) are essentially the same, schematically ‘P (and) so Q’. But what I’m concerned with here are interpretations of the juxtaposition in (21a) which are not possible for the conjunction in (21b). The solution just proposed, for the restrictions on conjunction interpretation, predicts that the second conjunct of (21b) will not be able to function as an explanation of the fact given in the first conjunct; this indeed seems to be the case. As we have seen, this is the sort of relation that juxtaposed utterances, on the other hand, frequently enter into. However, the backwards causal relation which hearers are likely to understand for the juxtaposition in (21a) is different from that discussed so far. It is not the case that Jim’s going to New York every weekend is a cause or a reason for his having a new girlfriend. Rather, it is his going to New York that gives the speaker grounds for her belief that Jim has a new girlfriend; that is, it plays a causal role in her having this belief. It is not too surprising that this is a readily accessible interpretation for the juxaposed utterances, 9 As Mira Ariel has pointed out to me, subordinate clauses introduced by because or since are explanations, so it may be overly strong to say ‘fact plus explanation’ interpretations are absolutely
excluded
from single utterances.
The glib response
material can always override pragmatic constraints. However, response and agree that the issue calls for deeper consideration.
to this is that semantically
encoded
I don’t think that is an adequate
R. Carston / Conjunckm,
44
since, as is widely acknowledged, hecause-clause, as in the examples (22a) Jim
has
a new
it is often the intended in (22):
girlfriend,
weekend. (22b) Jane has left, because
explanation and rekwance
because
he goes
interpretation
to New
York
of a
every
her son isn’t here.
The two scope possibilities for because-clauses are clearly evident in (22b): it can be interpreted either as giving the reason for Jane’s having left, or as providing evidence for the belief that she has left. Further context is required to resolve this ambiguity. This ‘reason for believing’ interpretation of the juxtaposed utterances in (21a) is easily handled within relevance theory, which claims that an utterance may communicate several propositions explicitly (see Sperber and Wilson (1986) and Wilson and Sperber this volume). So, for instance, the first utterance in (21a), used literally, would have at least the following explicatures: (23a) Jim has a new girlfriend. (23b) The speaker is saying that Jim has a new girlfriend. (23~) The speaker believes that Jim has a new girlfriend. So a ‘why?’ question can be raised in reaction to any of these and the second utterance taken as explanatory of any of them. In the case of (21a), the second utterance, ‘he goes to New York every weekend’, is clearly to be taken as an explanatory response to (23~) the higher-level explicature representing the speaker’s propositional attitude, rather than to (23a), the proposition expressed, which has been the target of the explanation in all the previous juxtaposed cases.
6. Conjunction and logical consequence This account of the different interpretations of the utterances also be given for those in (24) another pair from Bar-Lev (1980: 144145): (24a) These are his footprints; (24b) These are his footprints
he’s been here recently. and he’s been here recently.
in (21) could and Palacas
45
R. Carston / Conjunction, explanation and relevance
On the ‘because’ sort of interpretation of (24a) (perhaps not the most accessible one), the fact that the ‘he’ referred to has been here recently could be taken as the evidence on which the speaker bases her belief that these are ‘his’ footprints (as opposed to anyone else’s). This interpretation would directly parallel that of (21a), and, as our account predicts, it is impossible for (24b), as it was for (21b). This, though, is not the meaning that Bar-Lev and Palacas have in mind: they take the second utterance in (24a) to be ‘a conclusion stemming forward from’ the first one. They are unable to account for why this is not possible for the conjunctive counterpart in (24b), since it is a forward-directed relationship and so quite different from the backwards causal and temporal cases that their ‘semantic command’ definition rules out. However, it is not clear that the analysis I’ve given so far can cover this exclusion either, since it doesn’t seem to be a case where the problem is one of trying to interpret the second conjunct as an answer to a question raised by the first. Some new considerations seem to be called for. While the addition of so to the second of the juxtaposed utterances simply makes the conclusion relationship of (24a) more explicit, embedding it in the conjunction seems to induce interpretive difficulty : (25a) These are his footprints; (25b) ? These are his footprints
so, he’s been here recently. and so he’s been here recently.
This is not to say that there is any general problem with so in conjunctions. Consider some of the various forward-directed cause-consequence andconjunctions that have come up throughout this paper: (26a) She fed him poisoned stew and (so) he died. (26b) The road was icy and (so) she slipped. (26~) Jim has a new girlfriend and (so) he goes weekend.
to New
York
every
Each of these examples is quite acceptable with so, which merely makes the cause-consequence relation between the conjuncts more explicit. Each is paraphrasable as ‘P and, as a result of P, Q’. This distinguishes them from (24) and (25) where the relation in question is one of logical consequence, or conclusion. As Blakemore (1987: 87ff., and 1988: 183) has emphasised, ‘consequence’ can refer either to a causal effect or to a logical conclusion; that is, to a causal connection between the states of affairs described, as in the
R. Carston / Conjunction, explanation and relevance
46
examples
in (26), or to an inferential
relation,
of premise
and
conclusion,
between two propositional representations. It is evident, then, from examples (25a) and (26) that SO can signal either of these kinds of consequence. However, it looks as if, when it is embedded in an and-conjunction, it has just the causee consequence enrichment-inducing function. If this is right, it provides a clue to the puzzle of why the conjunction in (24b) is unable to express a logical conclusion relation. First, though, this claim about SO needs some substantiation. The examples in (27) can be seen as various attempts to formulate a conjunctive utterance that captures the relation of logical consequence in (24a), ((27a) = (24b), (27b) = (25b)): (27) These are his footprints (a) *and he’s been here recently. (b) ? and so he’s been here recently. (c) and so I know he’s been here recently. (d) and so, clearly/without doubt, he’s been here recently. Intuitions are fairly delicate in this region, but at least some people lo seem to share those marked in (27). Note that these judgements concern just the forward conclusion interpretation; (27a), for instance, although starred. is quite acceptable on a different interpretation, not discussed here. As already noted, although better than (27a), (27b) does not seem to be completely satisfactory. This can be accounted for as follows: SO cannot have its logical-consequence-indicating role here, and its causeeconsequence-indieating role would lead to the nonsensical interpretation, ‘these are his footprints and as a result he’s been here recently’. However. once we have it explicitly given that the state of affairs described by the second conjunct is a knowledge state or belief, as in (27~) we have a thoroughly acceptable cause-consequence interpretation: ‘P and as a result I know Q’. Example (27d), in which sentence adverbials indicate the sort of propositional attitude involved, shows that, as long as some explicit indication of beliefhood is given, the causeeconsequence enrichment can be made. What seems to be required is that the conjuncts describe two states of affairs between which a causal relation can reasonably be thought to hold; an identification of a set of footprints and a belief about a particular person having been in that place are two such likely states of affairs.
I0
Bar-Lev
identical
and Palacas
(1980: 145) make the same judgements
to those in (27), except that they use thus rather
for a set of examples
than so.
which are
41
R. Carston / Conjunction, explanation and relevance
The question,
now, is why so cannot
have the function,
has in the case of (25a), of indicating an inferential consequence in these conjunction cases. My suggestion tial relations can only be specified between processing
which
it clearly
relation of logical is that such inferenunits. Lexical items
with this sort of function provide a speaker with a means of indicating to a hearer how the utterance they introduce is to be processed, in the context of the immediately preceding utterance or an otherwise highly accessible set of assumptions. Blakemore (1987, 1988) conceives of the class of ‘discourse connectives’, which includes so and after all, as guiding hearers in their bid for an optimally relevant interpretation of an utterance, by imposing constraints on the inferential (or pragmatic) computations they should perform. Since an and-conjunction is a single utterance, carrying the optimal relevance presumption as a whole, it is not possible to specify this sort of processing instruction for a single conjunct, a proper subpart of the utterance, which is not guaranteed optimally relevant on its own. Thus, the only consequence relation SO can manifest in a conjunction is that of causal consequence. In other words, when it occurs in a conjunction so functions as a constraint on the proposition expressed, or, equivalently, on the truthconditional content of the utterance, rather than on its inferential relation with the context. l l Returning to the pair of examples in (24) that prompted all this, it is now no surpise that the conjunction in (24b) cannot be understood as expressing the conclusion-drawing relation, since this is a relation that holds just between utterances, not within them. The nearest conjunctive equivalents to (24a) are (27~) and (27d), but what is an implicit inferential relation in (24a) is explicitly described in (27~) and (27d). It is painfully apparent that the ideas presented in this paper especially in this last section, are highly speculative and raise many further questions. However, whether this particular analysis proves to be on the right track or not, the general pragmatic approach advocated here does embrace a much wider range of examples than the semantic analysis and it opens up new directions in which to look for a more adequate account.
I1 This marks a departure of so is to constrain the distinctions
developed
from Blakemore (1988: inferential computations
by Wilson
and
Sperber
185) who claims that the sole function a proposition may enter into. Using
(this
volume,
and
forthcoming
b), I am
proposing that while so is indeed a constraint, that is, it encodes a procedure rather than a concept, it may function at either the implicit (inferential) level or at the explicit (representational) level.
R. Carston / Conjuncrion.
explanukm
und relevance
References Bar-Lev,
2. and A. Palacas,
1980. Semantic
command
over pragmatic
priority.
Lingua
51. 137-
146. Blakemore,
D., 1987. Semantic
Blakemore,
D., 1988. ‘So’ as a constraint
tions: The interface
constraints
on relevance. on relevance.
between language
and reality,
Oxford:
Blackwell.
In: R. Kempson 183-195.
(ed.), Mental
representa-
Cambridge
University
Cambridge:
Press. Burton-Roberts,
N., forthcoming.
tics. Oxford:
Pergamon
Apposition.
In: The Encyclopedia
Press, and Aberdeen:
Aberdeen
Carston, R., 1988. Implicature, explicature and truth-theoretic Mental representations: The interface between language Cambridge University Press. Reprinted Oxford: Oxford University Press. Carston,
R.,
Linguistics Carston,
1990. Quantity
Carston, Cohen,
R., forthcoming.
and
Conjunction
Oxford:
generahsed
and pragmatic
Pergamon
R., in preparation.
semantics. In: R. Kempson (ed.). and reality, 1555181. Cambridge:
(ed.),
Pragmatics:
implicature.
UCL
A reader. Working
effects. In: The Encyclopedia
Press, and Aberdeen:
Conjunction,
L. J., 1971. Some remarks
In: Y. Bar-Hillel
in: S. Davis
and Linguis-
Press.
33351.
Papers
juxtaposition
of natural
University
of Language Press.
and relevance.
on Grice’s views about
(ed.), Pragmatics
Aberdeen
the logical particles
languages,
5&68.
of natural
Dordrecht:
language
Reidel.
Grice, H.P., 1967. Logic and conversation. The William James Lectures. In: H.P. G-ice. Studies in the way of words, 22240. Cambridge, MA: Harvard University Press. Harnish,
R., 1976. Logical
integrated
theory
form and implicature.
of linguistic
ability,
Levinson, S., 1987. Minimization celli-Papi (eds.), The pragmatic Posner,
R.,
J. Searle, Dordrecht Schmerling,
1980. Semantics F. Kiefer,
and pragmatics
M.
In: T. Bever, J. Katz, T. Langendoen
3 133391. New York
Bierwisch
(eds.).
of sentence Speech
connectives act
1989.
(eds.). An
: Crowell.
and conversational inference. In: J. Verschueren. perspective, 61-129. Amsterdam: Benjamins. theory
in natural and
M. Bertuclanguage.
pragmatic%
In:
1688203.
: Reidel. S., 1975. Asymmetric
(eds.), Syntax Sperber, Sperber,
in
I.
2, l-3
and Linguistics.
maxims
of Language
University
and semantics,
conjunction
and rules of conversation.
Vol. 3: Speech acts, 211-232.
New York:
In: P. Cole, J. Morgan Academic
Press.
D. and D. Wilson, 1986. Relevance: Communication and cognition. Oxford: D. and D. Wilson, 1987. Precis of Relevance: Communication and cognition.
and Brain Sciences
10, 697-710.
Wilson,
D. and D. Sperber,
this volume.
Linguistic
Wilson,
D. and D. Sperber,
forthcoming
a. Pragmatics
form and relevance.
Wilson,
D. and D. Sperber,
forthcoming
b. Varieties
and time. of non-truth-conditional
meaning.
Blackwell. Behavioral