Understanding the Impact of Differentiation and Unitization on the Perceptual Features Learned During Category Training

Abstract

Perceptual representations are a foundational aspect of all cognitive processes that involve input from the external environment. Yet there is ample evidence that these perceptual representations are altered by experience in systematic ways. This work focuses on understanding how perceptual representations are modified through two perceptual learning processes, differentiation and unitization, in the context of category learning. First, we review the empirical evidence for perceptual learning with a focus on the evidence for unitization and differentiation processes in the context of category learning. This section also includes a discussion of the role of differentiation and unitization learning processes in four computational models of perceptual learning. Second, we present a series of four experiments that measure the change in perceptual representations after learning category structures designed to promote differentiation and unitization in perceptual learning. Third, we investigate the impact of these category structures on the features inferred by a model that incorporates both differentiation and unitization perceptual learning processes. Fourth, we develop a modeling framework to directly compare the fit of computational models that assume different perceptual representations to the empirical results. Finally, we conclude by considering the implications and limits of these results.