Rules in the mist: Emerging probabilistic rules in uncertain categorization

Abstract

In this study, we explored the development of rules in probabilistic category learning, focusing on how knowledge acquired with uncertain feedback conditions transfers to a categorization task with similarity judgments. Using the Probabilistic Categorization Task (PCT) across two experiments, we examined whether rule-based knowledge learned under probabilistic feedback could be applied in the subsequent transfer phase. In Experiment 1, participants learned a unidimensional categorization rule with feedback reliability set at 70 %, 80 %, and 90 %. The findings indicated a strong correlation between feedback reliability during training and transfer phase performance, particularly in the 80 % and 90 % conditions. Experiment 2 expanded this approach by introducing a more complex categorization rule (XNOR), requiring participants to integrate two features. Here, participants trained with 80 % and 90 % reliable feedback successfully applied the learned rules in a similarity judgment task, proportionally to feedback reliability. Altogether, we argue that these findings question dual-system theories positing category learning as a sequential or competitive process between implicit and explicit systems. Instead, our results support the idea that a single either explicit rule-based or implicit similarity-based systems can effectively adapt to probabilistic settings, either independently or in close interaction with each other.