Phonetic Perceptual Reference Space for Prosodic Phonological Categories

The research program of the project is situated at the interface between phonology and phonetics. Its principal goal is to define the perceptual reference space for prosodic categories. The methodology applied towards this goal is both experimental and computational. In the experimental work classical paradigms such as Categorical Perception and the Perceptual Magnet Effect are applied to determine which prosodic categories posited by phonological theory have a distinctive representation in the perceptual phonetic reference space. The computational model that we are developing serves to formulate hypotheses and make predictions of test results.

The key ingredient of the model is the concept of context, which comprises the immediate segmental-phonetic context, prosodic context, local linguistic context, discourse context, and situative context, as well as speaker-specific information. It is assumed that the listener takes various types of context information into account when extracting abstract categories from the speech input. The model further posits that the listener enriches the underspecified categories with the context information to synthesize internally exemplars of the pertinent categories, and then performs a comparison between the internally synthesized exemplars and those available in the speech input. The comparison is assumed to be performed in the perceptual reference space. Exemplar Theory serves as a common framework for the experimental and computational aspects of the project.