Constraint Interaction in Probabilistic Phonology: Deducing MaxEnt from Hayes and Zuraw's Shifted Sigmoids Generalization

Abstract This paper explores the implications of an empirical generalization on the rates of application of variable processes developed by B. Hayes and K. Zuraw (HZ). The analysis is framed within the harmony-based framework where candidates surface realizations have probabilities proportional to the harmony scores of their constraint violation profiles. The main result is that these harmony-based probabilities predict HZ’s generalization if and only if the harmony function is the Maximum Entropy (ME) harmony or else it can be construed as the ME harmony by simply rescaling the constraints. In conclusion, within harmony-based probabilistic phonology, the ME mode of constraint interaction follows in a principled way from HZ’s generalization.