Variation in intonation: A stochastic dynamical system for pitch accents and its inversion

The literature on the pitch accents of American English (AE) reveals substantial variation across speakers and within accent categories, as well as variation in which pitch accent category is produced in a given discourse context. In this work we present a stochastic revision of a deterministic dynamical system theory of American English pitch accents. This theory generates F0 trajectories from a system of differential equations that govern the change in F0 over time, capturing the distinctions in peak alignment and scaling that characterize within-and across-category variation in AE pitch accents. The stochastic model has one free parameter which is set by the language’s phonological system. We also present a stochastic model of perception of pitch accents, which invokes the production model to generate hypotheses about the phonological free parameter describing the observed trajectory. We therefore aim to provide a framework in which variability can be explicitly modeled, and in which the interaction of phonology, production, and perception of prosody can also be modeled.