IN his paper, 'Does Tense Logic Rest Upon a Mistake?' (to appear

128 ANALYSIS context-dependence that if things had been different, 'the actual world' would have picked out some world other than the actual one. Tulane University, GRAEME FORBES 1983 New Orleans, Louisiana 70118, U.S.A. ACTUALITY AND CONTEXT DEPENDENCE II By MARTIN DAVIES I IN his paper, 'Does Tense Logic Rest Upon a Mistake?' (to appear in a Festschrift for Donald Davidson, edited by Bruce Vermazen and Merrill B. Hintikka), Gareth Evans considered three possible semantic treatments of temporal modification. One he rejected as involving an incoherent conception of assertion. The other two he admitted as coherent; but in each case he argued that there are important disanalogies between temporal modification and modality. His conclusion was that tense logic should not be regarded as genuinely analogous to modal logic: we should not be 'blinded by the structural parallels between time and modality to their substantial differences'. In Meaning, Quantification, Necessity: Themes in Philosophical Logic (Routledge and Kegan Paul, 1981; henceforth l MQN\ I stated the main points of Evans's argument. Graeme Forbes is still not convinced ('Actuality and Context Dependence I', ANALYSIS, this issue). In this note I shall try to make clearer what I take the argument to be. Of the two semantic accounts that Evans admitted as coherent, one involves the claim that an expression, such as 'Whitlam is angry', which admits significant temporal modification is strictly speaking ambiguous. The other involves the claim that temporal modifiers are context shifting operators. On the ambiguity account, 'Whitlam is angry' is assigned quite different semantic values according as it occurs standing alone or modified by a temporal operator. As it occurs standing alone, the expression is a context dependent sentence, and relative to any context it is assigned a semantic value appropriate to a complete sentence. This value has to provide for the evaluation of an utterance of the sentence in that context as TRUE or FALSE once-for-all, where TRUTH is the primary dimension of assessment for utterances, the aim of sincere assertion. As it occurs modified by a temporal operator the expression is assigned

ACTUALITY AND CONTEXT DEPENDENCE II 129 (even relative to a context) a quite different semantic value, akin to that of a predicate of times. If we are 'semantically fastidious' then we should remove this ambiguity by distinguishing between what Evans called a situation expression (akin to a predicate of times) and the context dependent sentence which can be represented as the result of applying the closure operator 'Now' to the situation expression. On the context shifting account, on the other hand, the expression 'Whitlam is angry' is a context dependent sentence in all its occurrences. But the semantic value assigned to a temporally modified sentence relative to a given context is not a function of the semantic value of the contained sentence relative to that same context; it depends upon the semantic value of the contained sentence relative to other contexts. I shall consider the two accounts in turn (following Evans, and reversing the order of presentation in MQN). II We can distinguish two lines of thought which lead to the idea that a 'semantically fastidious' representation of the context dependent sentence 'Whitlam is angry' will involve the closure operator 'Now'. First, the situation expression 'Whitlam is angry' admits significant temporal modification because (even relative to a context) it expresses a condition that obtains at some times and not at others. But such a temporally relative condition cannot be the condition for the TRUTH of an utterance of 'Whitlam is angry' standing alone, since assessment for TRUTH is a once-for-all matter. To obtain an expression which expresses (relative to a context) a condition that can be a TRUTH condition we need to apply the closure operator to the situation expression. Second, if our theory provides only for the temporally relative condition, then we are left asking the question 'What should [a sincere asserter] aim at?'. The answer, of course, is that as the language is actually used he should aim at producing an expression whose temporally relative condition obtains at the time of utterance. But in that case the expression uttered should be represented as involving reference to the time of utterance. Other conventions of use are possible, and if the convention were different then the representation would have to be correspondingly different. It may seem that this is a distinction without a difference, but in fact the difference becomes clear when we turn to modality. If an expression is to admit significant modal modification it needs to be contingent. We can avoid potentially confusing complications if we choose an expression which is free of temporal or spatial relativity. So let us take as our example 'Snow is white', construed

130 ANALYSISI as expressing a condition which obtains only contingently but is not temporally or spatially relative. Since 'Snow is white' already expresses a condition that can be the condition for absolute or once-for-all TRUTH of an utterance, we have nothing corresponding to the first line of thought. But it can seem plausible that there is, analogous to the second line of thought above, an argument for the idea that a 'semantically fastidious' representation of 'Snow is white' standing alone will involve the 'Actually' operator (which is formally analogous to the 'Now' operator). The conventions of use of our language could have been such that an utterance of 'Snow is white' is TRUE iff the expressed condition would have obtained (even) if kangaroos had no tails. In that case the representation of the expression would have involved a counterfactual conditional. As the language is actually used, the utterance is TRUE iff the expressed condition obtains as things actually are. But in that case should not the expression uttered be represented as involving reference to the actual world? If this argument is a good one then we shall have to agree that no deep disanalogy between the temporal and modal cases has been demonstrated. The main claim that Forbes makes is that a deep disanalogy cannot be demonstrated without appeal to metaphysical differences between times and possible worlds. I believe that Evans would have agreed. At one point he wrote: 'A sane view of possible worlds begins when we conceive of the actual world as the world in which all and only true propositions are true'. In MQN, I said essentially the same in reply to the argument set out in the last paragraph. (MQN p. 209. Christopher Peacocke drew the argument to my attention.) Forbes objects that the familiar rubric about truth reveals that notion as not conceptually prior to the notion of the actual world (Forbes p. 126). A short reply to that objection is that, as it occurs in the familiar rubric, the expression 'the world' does not refer to the actual world conceived as a member of a set of possible worlds. A more formal version of the rubric makes this clear: if _s can be used to assert that p, then s is true iff p. (Cf. MQN p.f27.) In the next section, I shall set out a little more fully what I take the order of conceptual priority to be. Ill My view about the location in the conceptual order of the notion of a possible world is similar to that expressed by Colin McGinn: 'if we are limning the true and ultimate structure of modal reality, we do better to leave the modal connectives primitive: paraphrasing them with possible worlds quantifiers misrepresents how things modally are'. ('Modal Reality', in Richard Healey (ed.), Reduction, Time and Reality, Cambridge University Press, 1981, p. 162.)

ACTUALITY AND CONTEXT DEPENDENCE D 131 Given the ideas of a sentence with a truth condition and of assertion as aiming at truth, we can employ the notion of possibility (expressed by the modal operator) to arrive at the idea of a condition which does not obtain but might have obtained, and with it the idea of a sentence which is not true but might have been true. Given also the notion of conjunction, we can arrive at the idea that various conditions might have obtained, and corresponding sentences been true, together. Here we have the idea of a possibility (or a way things might have been, or even a way 'the world' might have been). The notion of a possible world is an idealisation of this notion of a possibility: a possible world is the limit of a process of refinement of possibilities. (See Lloyd Humberstone, 'From Worlds to Possibilities', Journal of Philosophical Logic 10 (1981), and Graeme Forbes, 'Physicalism, Instrumentalism and the Semantics of Modal Logic', forthcoming in Journal of Philosophical Logic.) Possibilities and possible worlds are, on this conception, abstract objects rather like sets of propositions. We can think of possibilities and possible worlds as regions and points in 'logical space', but there are clear ontological differences between these 'regions' and 'points' and intervals and moments of time, or regions and points in physical space. (See McGinn op cit. pp. 151-2.) Thus we arrive at the idea of a collection W of possible worlds, and the idea of sentences being true or false relative to a member w of W. Amongst the possible worlds in W, one can be distinguished as the actual world: it is that world w such that the sentences that are true relative to w are precisely the sentences that are true (simpliciter). This is the conception of the actual world which, according to Evans, is needed for 'a sane view of possible worlds'. If what I have just sketched is a sane view of possible worlds then, clearly, a proper conception of assertion and truth does not require the use of 'Actually' as a closure operator. For to say that in assertion one aims at truth relative to the actual world is to say no more about assertion than that one aims at truth. It should be clear that, in offering these reflections, I am not disagreeing with Forbes's main claim. For whether or not they are taken in an eliminativist spirit, these reflections are metaphysical rather than semantic. IV Let us turn to the question whether the temporal and modal cases can be seen as analogous if temporal modifiers are treated as context shifting operators. It might seem that the cases can be seen as analogous: 'we can just take possible worlds to constitute the contexts' (Forbes p. 124). But then, of course, we face the question whether we can think of worlds as contexts, and Forbes gives a reason for saying that we cannot: 'The significant difference... is that we move through time and from place to place, but we do not

132 ANALYSIS move through the other possible worlds in which we exist' (p. 127). I think that Forbes is essentially right about this. I myself would try to highlight the significant difference by considering two facts about context dependent reference (e.g. the use of 'now' or 'here') in contrast to the use of 'actually'. First, in context dependent reference the object referred to (e.g. a time or a place) is a member of an extensive class (of times or places). If we say that in a use of 'actually' there is reference to the world in the sense of 'I and all my surroundings', then we have a sharp contrast with this first fact. We avoid that contrast if we say that in a use of 'actually' there is reference to the actual world that is a member of the set W. But, second, in context dependent reference there is an appropriate relation an en rapport (causal) relation or a locational relation between the token thought event (or token utterance) and the object referred to. On my conception, the members of W are abstract objects rather like sets of propositions, and we stand in no causal relations and so no en rapport relations to them. Nor can reference to a member of W be secured by the obtaining of a locational relation. For a token thought event or token utterance is located in countless different members of W. Perhaps we can find something analogous, in the case of abstract objects, to an en rapport relation to a physical object. This is a difficult matter, but in the case of a set I think that the closest analogue is having exhaustive knowledge of the membership of the set. (See MQN pp. 110-11.) In the case of a possible world, the closest analogue would then seem to be having exhaustive knowledge of the conditions that obtain at the world. But that is knowledge that we do not have, even for the actual world. (Hence the remark at MQN p. 206, quoted by Forbes at (a) on p. 124.) If a speaker uses 'now' or 'here' (or 'I' or 'you') in an utterance, then it is not sufficient for understanding of that utterance that an audience should merely know the meaning of the sentence uttered. He must be able to identify the time or place (or speaker or addressee) of the context of utterance; only then will he know what has been said. In central cases, this identification involves the audience thinking of the time or place (or person) under a demonstrative, rather than a merely descriptive, mode of presentation; and this is possible precisely because the audience stands in an appropriate en rapport or locational relation to the object. But one can hardly hold that understanding of an utterance involving 'actually' requires identification of the actual world from amongst the set W of possible worlds. These reflections, like those of the previous section, are of a metaphysical, rather than a purely semantic, character. So again, they involve no disagreement with Forbes's main claim. Nor, I think, would Evans have disagreed. Concerning the application of a context shifting account to modality he wrote: 'To think in this way

ACTUALITY AND CONTEXT DEPENDENCE II 133 about possible worlds seems to commit one to an unacceptable form of modal realism'. It seems then that on neither the ambiguity account nor the context shifting account of temporal modification is it genuinely analogous to modality. 1 Birkbeck College, MARTIN DAVIES 1983 Malet Street, London WC1E 7HX 1 I am grateful to Graeme Forbes and Christopher Peacocke for advice on this note. DOUBTING ONE'S METHODS By ALAN R. WHITE A RECENT attempt in this journal (R. A. Sorensen, 'Knowing, l\believing and Guessing', ANALYSIS 42.4, October 1982, pp. 212-3) to support the thesis that knowing that p implies believing that p is based on the supposition that if one wrongly believes that one's method of arriving at an answer is unreliable, then one does not know the answer (Sorensen says 'Julia did not know the answers to the examination since she was mistaken about [Sorensen means that she had a wrong view about] the reliability of the method' and 'Since Jean falsely believes that his method is unreliable, Jean does not know the correct answer to the English history questions'). But this supposition is really a variant on the thesis at issue and so begs the question. Moreover, the supposition is as wrong, and for the same reasons, as the thesis. Seeds of doubt about the reliability of his method may exist or be sown in the mind of one who uses a reliable method to arrive at an answer as easily, and in the same way, as seeds of doubt may exist or be sown about his knowledge of the answer in the mind of one who knows the answer. If someone is wrongly persuaded by a mathematician that his method of telling at a glance whether a large number, other than one ending in zero, is divisible by four, namely by checking whether the number represented by the last two digits is divisible by four, is mistaken, does he not know that, e.g. the number 1234567896 is divisible by four? If a physicist succeeds in making someone wrongly distrust the reliability of telescopes, does he not know that the distant object seen through his telescope is a building? If a sceptical philosopher's subtle arguments entice someone into falsely believing that sense perception is never a reliable method of discovering material objects, does he never know