Comparative Phonotactics

1 Two kinds of phonotactics I contrast the notions of “absolute phonotactics” and “comparative phonotactics.” Absolute phonotactics is the study of well-formedness in phonology. The topic has a long history, but was laid out with particular cogency by Chomsky and Halle (1965), who noted that speakers have phonotactic judgments even of words they have never heard before; thus blick [blɪk] is non-existent but well-formed, while bnick [bnɪk] is non-existent and ill-formed. Further work (Scholes 1966, Chomsky and Halle 1968:416-418, and many others) has suggested that absolute phonotactics is gradient; for example, there are words (e.g. poik [pɔɪk]) that sound neither terrible nor perfect. In the approach to be taken here, gradience is captured by assigning words numerical scores, which have an explicit interpretation in the theory of probability. For comparative phonotactics, we assume two populations of strings, which we can call A and B, and a grammar whose outputs are likewise probabilities, but which specifies whether any given string will belong to Population A or Population B. That is, there will be a probability assigned to the outcome “this form belongs to A”, a probability assigned to the outcome “this form belongs to B”, and the two probabilities sum to one. In order to do this well, we will need to select constraints that single out traits that successfully distinguish the A and B populations; perhaps exceptionlessly, perhaps only probabilistically. This article explores the question of whether comparative phonotactics is a useful idea for phonology. I cover relevant theoretical background, then go through two case studies, and finally address some general theoretical questions raised by the concept.