A little proposition: contextualise 'this'

Most linguists and philosophers of language take the view that getting to the truth conditions of an utterance is a two-stage business in which first the contribution of the words and the syntax of the sentence are computed, and then information from the context is added to get to a complete proposition, something evaluable for truth or falsity. Emma Borg's book argues against this position, partly on linguistic grounds, but mostly from an ideological position. It maintains that if the human syntax-semantic computational engine needs to consult context to get to a proposition, then - since contexts can be open-ended and involve any amount of non-linguistic knowledge or concepts such as beliefs and intentions - that engine cannot be a "module" in Jerry Fodor's sense: a task-specific, informationally encapsulated, probably hard-wired, mental organ. Borg thinks Fodor is right about modularity, hence the human syntax-semantics engine must deliver something evaluable for truth without contextual consultation, she argues.

So how does Borg deal with overtly indexical constructs such as demonstratives? Following David Kaplan, she makes a distinction between "character", the invariant instructions for rummaging in the context that distinguish "this" from "that" and other indexicals, and "content", the indicated object located by following these instructions. But this kind of content is outside the semantics module and so cannot be part of the truth conditions delivered. On Borg's account, the truth conditions are themselves conditional and just include something like the character ("singular concepts"): so the truth conditions for "that's mine" (=P) are something like "if X utters P, and 'that' therein refers to Y, then P is true if and only if Y is X's". This does not sound right to me: it is fine for our semantic theory to deliver such conditional T-sentences - the condition just captures the context dependence -but to know just what is to be evaluated for truth given a particular utterance token, we need to know whether the condition is satisfied in that context, and that involves seeing whether a relation of reference does actually hold between "that" and Y. I do not fully understand Borg's account here, I confess: she holds that it is not necessary to locate the referent and claims that "one can have a genuinely singular thought about an object even in situations where one cannot identify the object of that thought in any substantial (non-linguistic) way". I do not see how one can be sure that this thought is "about an object" in such a case.

Covert indexicals such as "Paula can't continue" or "John is too short" - in which the relevant missing component is filled in from the context ("continue running", "too short to be a basketball player") and which have always seemed to provide strong arguments for a two-stage interpretation process - are analysed by Borg as instead having what she calls "liberal" truth conditions. She argues that in all these cases there is some syntactic clue that something is missing and that the missing items are existentially quantified over. Thus "Paula can't continue" means roughly "There's something that Paula can't continue". This has the odd consequence that many of these sentences will be always true, or always false, construed thus: there are many things that Paula cannot continue - being 32, living, standing... if no further spatio-temporal restriction is imposed on possible candidates. Borg accepts this conclusion, arguing that the way such constructions are interpreted is akin to non-literal meanings such as irony: the conveyed pragmatically derived interpretation differs from the literal one, in particular being more restricted as to context. Her anecdotal evidence for this is that people can distinguish the liberal from the restricted interpretations ("I said I'd do the review: I didn't say when..."). I find this conclusion implausible: there are many relevant differences between the interpretation of ellipsis and the interpretation of non-literal meaning, in particular the fact that in the latter case (on many accounts) the literal meaning has to be processed and rejected in order to infer the intended meaning. Surely a more likely account of these cases is that the contextual input is simply being resolved differently, either co-operatively local or bloody-mindedly global.

Borg writes extremely well. She knows her way around the landscape, and her willingness to argue against orthodoxy is exhilarating. This is a serious contribution to the literature. But the strong sense is that her argument is driven by ideology rather than by linguistic plausibility. And I was left with a nagging doubt as to whether her ideological premise - modularity - was as incompatible with a two-stage theory as she believes: is it so unlikely that a specialised computational module could deliver underspecified, pre-propositional representations (propositional "primal sketches") to a more general reasoning component? What the situation semanticists used to call the "efficiency" of language would be well served by such a module.

Stephen Pulman is professor of general linguistics, Oxford University.