Morphologically-conditioned tonotactics in multilevel Maximum Entropy grammar

Abstract

This paper presents a novel approach to probabilistic morphologically-conditioned tonotactics, featuring a case study of Mende, in which tonotactics vary by lexical category. This variation in surface tone patterns is modeled via indexed weight adjustments (i.e., varying slopes) for each constraint in a Maximum Entropy Harmonic Grammar, quantifying the degree to which each lexical class follows basic tonotactic principles in a common base grammar. Approaching morphologically-conditioned phonotactics as indexed weight adjustments of a base grammar offers a solution to the existing stalemate between single grammar (e.g., indexed constraints) and multiple grammar (e.g., Stratal OT; cophonologies) models of lexically-sensitive phonological patterns.