Prior beliefs modulate rule-based learning in dynamic probabilistic environments: Insights from a hierarchical Bayesian model

Humans learn and utilize diverse rules to guide their decision-making. Previous studies have demonstrated that individuals possess symbolic flexibility—the ability to effectively acquire different symbol associations. However, it remains unclear as to whether individuals retain symbol flexibility when acquiring language rules (pseudowords) and non-language symbol rules (mathematical operators) in dynamically changing environments. This study integrates a probabilistic reversal learning task and a picture-word matching task to create dynamic environments with varying levels of probability of receiving a reward. Computational modeling of behavioral and fMRI data revealed that individuals exhibited symbol flexibility, effectively learning distinct symbol systems while displaying divergent learning patterns. Specifically, as the value of prior beliefs increased, response times became faster during non-language symbol rule learning but slower during language rule learning. The right caudate nucleus exhibited rule-specific activation: In language tasks, activation was significantly higher under high probability environments compared to no probability conditions; and in non-language symbol tasks, activation under no probability conditions surpassed that in language tasks. These findings highlight the neural and behavioral adaptability underlying symbol flexibility, demonstrating that prior beliefs dynamically shape learning strategies across rule types and environmental uncertainties.