ABSTRACT: In this talk, I address the problem of phonological relationships that seem to be intermediate between contrast and allophony. Allophony (in which two sounds are predictably distributed all of the time) and contrast (in which two sounds are unpredictably distributed at least some of the time) have traditionally been assumed to be binarily and categorically distinct. There are, however, a large number of intermediate relationships in the world’s languages that apparently affect both phonological and perceptual patterns. Such relationships have generally been marginalized in the phonological literature and given names such as “quasi-allophones” or “marginal contrasts.”

I present a new model of phonological relationships that redefines these relationships along a continuum, from fully predictable to never predictable. Intermediate relationships can be defined at all points along this continuum, and the precise relationship between any two sounds in a language can be quantified. The continuum is largely based on the information-theoretic concept of entropy, a measure of the uncertainty that exists in the choice among any set of items. This model can be used to describe the synchronic state of phonological relationships and to track changes in progress or variation across speakers. Additionally, using uncertainty as the basis for modeling phonological relationships provides a cognitively based explanation for the observation that contrastive sounds are perceptually more distinct than allophonic ones, or more precisely, that the more unpredictably distributed a pair of sounds is, the more perceptually distinct that pair will be.