This paper reports on web-as-corpus research that seeks to explain why some semantically opposed word pairs have special status as canonical antonyms (for example: cold-hot), while other pairs do not (icy-scorching, cold-fiery, freezing-hot, etc.). In particular, it reports on the findings of Jones, Paradis, Murphy and Willners (2007), and extends their retrieval procedure to include the previously overlooked ‘ancillary’ function of antonymy (Jones 2002). The primary assumptions are that a language’s most canonical ‘opposites’ can be reasonably expected to co-occur with highest fidelity in those constructions associated most closely with the key discourse functions of antonymy, and that, given their low frequency in language, an extremely large corpus is needed in order to identify such patterns of co-occurrence.