Prior knowledge influences the neural mechanisms supporting memory-based inference

Memory-based inference allows individuals to integrate information acquired across separate episodes to support novel decisions and reasoning. Although prior knowledge, such as schemas, is known to influence learning and memory, its impact on the neural mechanisms underlying inference remains unclear. In this study, we investigated how schema congruency affects the encoding and retrieval of overlapping events and how these processes contribute to memory-based inference. Thirty-nine participants encoded AB associations, consisting of picture-word pairs presented on either schema-congruent or schema-incongruent backgrounds. These were followed by BC associations involving the same word paired with a new picture on a neutral background. At test, participants were asked to infer the indirect AC association. While overall inference accuracy did not differ as a function of schema congruency, behavioral and neural data revealed distinct mechanisms. Inference for schema-incongruent events depended on accurate retrieval of both AB and BC associations, whereas schema-congruent inferences did not. To investigate the neural processes involved, we trained hierarchical multivariate pattern classifiers on EEG data to detect schema and context reinstatement during task performance. For schema-congruent events, successful inference was predicted by schema reinstatement during BC encoding, consistent with integrating overlapping information into a unified memory trace. In contrast, successful inference for schema-incongruent events was predicted by context reinstatement during AC retrieval, reflecting a reliance on flexible recombination of separate memory representations. These findings demonstrate that schema congruency modulates the neural basis of memory-based inference. Congruent events are integrated during encoding, whereas incongruent events rely on retrieval-based inference. Significance StatementTo make informed decisions, we often need to combine information from separate events. This study shows that prior knowledge, or schemas, affects how the brain forms these connections to make inferences about the world. Using EEG and machine learning, we found that when new information fits with existing schemas, people usually integrate memories during learning. In contrast, when information conflicts with prior knowledge, successful inference depends on retrieving and merging separate memories at the time of the decision. These findings reveal that the brain can switch between encoding and retrieval strategies based on prior knowledge, providing new insight into how memory supports reasoning, decision-making, and adaptive behavior.