On the uses of variable rules

Abstract

The introduction of variable rules ten years ago has provoked a variety of critical reactions; among these Kay & McDaniel's (1979) review appears as a clear and penetrating study of many issues neglected or unresolved in earlier discussions. It is refreshing to observe an approach to the problems of variation and sociolinguistics relatively free from the ideological constraints that other critiques have inherited from formal linguistics (e.g. Bickerton 1971, 1973; Gazdar 1976). They bring to the discussion of variable rules a clarity gained by several years' reflection on the early statistical approaches, together with a certain distance from current sociolinguistic methodological developments and problems. At the same time, there are some attendant disadvantages of such a distance; these appear in their treatment of the work that preceded probabilistic models, in their lack of attention to the interaction between the practical aspects of linguistic data analysis and the evolution of theoretical concerns, in their misunderstanding of certain mathematical facts, and in their neglect of the more recent developments over the past five years. The K & M analysis may best be evaluated as a reaction to the stage in variable rule analysis around 1971–4 when the first probabilistic models were being proposed and tested.