Proceedings of the Annual Meeting of the Cognitive Science Society

Research on contrastive category learning has revealed a robust tendency for learners to develop caricaturized representations (elsewhere: ideals or extreme points) to support successful discriminative classification. These representations are defined by extreme values on some task-relevant dimension and are often indicated as highly representative of their categories. Work in this area has elaborated the task constraints and contexts necessary for these representations to emerge, but little research has scrutinized whether caricatured representations extend beyond a category’s known range of feature values. To these ends, across two experiments, we investigated whether the most representative items for a category can extend beyond the training set. Data from pairwise typicality comparisons following learning suggests that caricatured categories may be supported by representations that extend past the feature range present in training. The findings are better explained by certain representational frameworks (e.g., adaptive reference points, boundaries) than others (e.g., exemplars, clusters).