Testing the sub-test

This paper pursues two objectives, one linguistic and one methodological in nature. First, it is concerned with a corpus-based analysis of the degree to which pairs of -ic/-ical adjectives (e.g. classic/classical) are synonymous. Second, it investigates whether Church et al.'s (1994) sub-test can be fruitfully applied to this phenomenon. As to the first issue, I conclude that individual -ic/-ical adjectives can be located on a continuum of semantic similarity, with some being virtually completely synonymous and others being strongly differentiated; several semantic and distributional distinctions between members of adjective pairs are pointed out on the basis of distinctive collocates. As to the second question, I demonstrate on the basis of a simulation that the sub-test is conceptually adequate, but suffers from its asymptotic approach, which is why Fisher-exact is argued to be a more adequate diagnostic.