A dynamical approach to speech categorization: Two routes to learning

Learning to perceive a non-native speech sound reliably is a specific instance of the more general issue of category learning. Here we take a dynamical approach that provides a theoretically motivated way to understand individual differences in the process of learning to perceive new speech sounds. Fundamental to this approach is the focus on the initial categorization and discrimination abilities of individual perceivers and how these structure the form of learning over time. Two distinct patterns of learning were observed that were predictable based on initial perceptual abilities. In one pattern, subjects became more attuned to small acoustic distinctions between stimuli and could use that sensitivity to label tokens reliably. In the second pattern, subjects became less attuned to within-category acoustic distinctions, suggesting cognitive restructuring. Finally, we present a dynamical model that incorporates both modes of category learning, and also allows for failure to learn.