Novel Concentric-Circle Technique Interrogates Implicit Category Learning

Abstract

Prototype, exemplar, and boundary models compete to explain representational-level abstractions during human category learning. Vast majority of previous work use linear categories structures to evaluate learning. We present the development of a novel, circular category structure and leverage it to explore limitations of prototype, exemplar and boundary models. We find that circular categories are readily learned by human participants, and the induced representation is most likely a quadratic boundary. We deductively eliminate prototype theories as an explanation of these circular categories and show that exemplar models, though viable, provide a weaker explanation than boundary models which are best fitted to the present data. These circular category structures offer a promising new technique to studying implicit category learning.