Learning-dependent modulation of working memory

Category knowledge helps us make inferences and decisions but may also bias how information is perceived and remembered. Previous studies have reported categorical biases in working memory (WM), but mostly for highly familiar features with category structure established through life-long exposure (e.g., colour). Key unanswered questions are how quickly category biases arise following new learning and how biases are shaped by the nature of prior experience. Across three experiments, we asked how new category learning biases WM reports (total N=180). Participants learned to categorise novel shapes in a brief training session (10-15 minutes). Training either emphasised category prototypes (Experiment 1) or the category boundary (Experiment 2). Following learning, they performed a two-item WM task where category was irrelevant. New category learning modulated WM-guided behaviour in two ways. First, training improved WM accuracy for the most familiar areas of feature space. Second, we found category-dependent bias of WM reports following training that stressed category prototypes but not category boundaries, compared to a control experiment with no learning (Experiment 3). Category bias scaled with the distance from the category boundary and was driven by trials when memory items belonged to distinct categories. Our results show that even newly learned categories may act as priors for WM, but biases may depend on the specific nature of prior experiences.