An adequate theory of natural language should include a solution to the semantic paradoxes. This is not the only condition of adequacy, but it is an important one. For the paradoxes are among the crucial tests of one's theory of natural semantics, syntax and phonetics, all of which figure in the paradoxes to some degree. The theory should either show how a correct conception of our language would block the derivations of paradox, or explain why they cannot be blocked even on a correct conception, and justify a relatively simple linguistic reform which would do so. Just as a theory of natural language should afford a solution to the paradoxes, one's solution typically will presuppose a conception of language. Hence much more is at stake than sometimes meets the eye when philosophers argue about classical paradoxes such as the Liar, the Grelling, and so on. It is simply not true that these arguments are 'five-fingered exercises' or 'sterile ornaments' of a philosophy of language (as I have heard them dismissed by two prominent philosophers, who in fairness should remain unnamed, not having committed themselves on the point in print). One's treatment of semantic paradox can crucially affect one's views about meaning, truth, propositions, sentences, and utterances, together with their interrelations and much else. This fact may occasionally be obscured by the belief that the classical paradoxes have long since been solved or quarantined by some means or other. But even if they have been adequately solved-which is far from clear-there is no guarantee that the same means would suffice for any new paradox that might turn up. Of course a sweeping rejection of all self-reference has nothing to fear from a new paradox. But naturallanguage solutions typically make a virtue of allowing some self-reference, in contrast, say, to a strict language-level approach. Furthermore, all such solutions have concentrated on absolute notions of truth and falsity, as expressed in one-place predicates like 'is true' and 'is false'. Relativised predicates like 'is true (false) in possible world W' and 'is true (false) with respect to language L' have been neglected. Yet paradox lurks in this neglected portion of the expressive power of natural language.