The advantages and disadvantages of corpus-based and intuition-based studies are well-known, cf. for instance Aarts (1991), Leech (1991, 1992); in fact, the debate is a long-standing one. The influence of behaviourism on Bloomfield and his consequent suspicion of meaning and of intuitions in his later work (1935: 139-44) meant that the post-Bloomfieldians concentrated on texts and corpora. J. R. Firth was much happier about accepting semantics as a legitimate part of linguistic studies but still preferred to interpret meaning in terms of context (1957: 19); indeed, it was this insistence on the contextual study of language that led him to the notion of ‘collocation’ (1957: 194-5). Contexts can of course be invented, but it is much easier to look for them in texts, and it is not surprising to find direct or indirect pupils of Firth like Halliday (1966), Sinclair (1966) and Mitchell (1975) developing the concept further.