We studied metaphorical language in spoken discourse from a number of settings to explore the words and expressions that regularly appear in the cotext of both conventional and innovative metaphors. We found that expressions that we call "tuning devices" are frequent in all the data consulted. Tuning devices have a number of different functions, clustered around the central notion of suggesting to the hearer how to interpret a metaphor. Our data was drawn from 2 computerized corpora, 1 small enough to be handsearched as well as concordanced, the other very large. Both small and large corpora present research problems, and we argue that combining the 2, by using a small corpus as a starting point for searches in a large corpus, may reduce the disadvantages of each.