The classical problem of induction is often phrased in terms of finding some reason to expect that nature is uniform. In Fact, Fiction, and Forecast (1954) Goodman showed that we need in addition some reason for preferring some uniformities to others, for without such a selection the uniformity of nature is vacuous. Thus, suppose that all examined emeralds have been green. Uniformity would lead us to expect that future emeralds will be green as well. But now we define a predicate grue: x is grue if and only if x is examined before time T and is green, or x is examined after T and is blue. Let T refer to some time around the present. Then if newly examined emeralds are like previous ones in respect of being grue, they will be blue. We prefer blueness as a basis of prediction to grueness, but why? Goodman argued that although his new predicate appears to be gerrymandered, and itself involves a reference to a difference, this is just a parochial or language-relative judgement, there being no language-independent standard of similarity to which to appeal. Other philosophers have not been convinced by this degree of linguistic relativism. What remains clear is that the possibility of these ‘bent’ predicates puts a decisive obstacle in face of purely logical and syntactical approaches to problems of confirmation.