When learning to classify by relations is easier than by features

Relational reasoning is often considered more resource intensive than feature-based reasoning. This view implies that learning categories defined by relational regularities should be more difficult than learning categories defined by featural regularities. Unfortunately previous studies do not ground featural and relational information in a common perceptual substrate. After addressing this concern, a series of experiments compare learning performance for relation- and feature-based categories. Under certain circumstances we find faster learning for relation-based categories. The results suggest that mechanisms rooted in relational processes (e.g., relative stimulus judgement, analogical comparison) facilitate or hinder learning depending on whether the relational processes highlight or obscure the underlying category structure. Conversely, category learning affects relational processes by promoting relational comparisons that increase the coherence of acquired categories. In contrast to the largely independent research efforts in category learning and analogy research, our findings suggest that learning and comparison processes are deeply intertwined.