Victoria L. Jacoby¹ (), Christine Massey¹, Philip Kellman; ¹University of California, Los Angeles

Perceptual learning of complex classifications has largely used trials in which a learner makes a categorization response to a single presented instance. Some prior research, however, suggests simultaneous comparisons of items from different categories can enhance processes of perceptual learning that facilitate successful categorization. Little work has directly compared how the concurrent learning of multiple perceptual classifications may differ when training emphasizes active comparison of items as opposed to individual item classification. We tested the effectiveness of paired comparisons for learning the differential diagnosis of 10 categories of benign and cancerous skin lesions. A Paired Comparisons condition, in which a learner was presented on each trial with a category label and required to choose between instances from two different categories, was compared to Single-Classification and Dual-Classification conditions, where instances of one or two categories were presented for classification on each trial. Novice participants were given a 40-min learning period before completing immediate and one-week delayed posttests on the classification of novel exemplars. Importantly, these assessments used the standard single-classification trial format regardless of the learning condition that preceded it. Results: In the learning phase, participants in the Paired Comparisons condition were able to complete significantly more trials than those in the classification conditions (large effect sizes), suggesting the relative ease of this trial type. In the posttests, classification accuracy reliably favored the Paired Comparisons condition, both immediately following learning, as well as after a one-week delay (all medium effect sizes). The Single- and Dual-Classification conditions did not reliably differ from each other on any learning or assessment measures. These results indicate that training based on paired comparisons can promote more efficient learning and generalization of complex categories than the more standard classification-based approaches.

Acknowledgements: This research was supported by National Institutes of Health award R01CA236791 to PK and CM. Any opinions, findings, and conclusions are those of the authors and do not necessarily represent the views of the NIH.