Prior knowledge influences the neural mechanisms supporting memory-based inference

Abstract

Memory allows us to go beyond direct experience by enabling inference across separate events. Yet how prior knowledge shapes the mechanisms supporting such inferences remains unclear. Here, we show that alignment between ongoing experience and existing knowledge determines how the brain constructs cross-event inferences. Behaviorally, inferences across schema-congruent events did not require accurate retrieval of the individual episodes, whereas inferences across schema-incongruent events depended on successful retrieval of those episodes. EEG-based multivariate pattern analysis further showed that schema-congruent inference was supported by schema reinstatement during encoding, consistent with the formation of integrated memory representations. In contrast, schema-incongruent inference relied on context-specific reinstatement at retrieval, reflecting the flexible recombination of distinct memories. Together, these findings demonstrate that prior knowledge dynamically shapes memory toward integration or separation, revealing how the brain constructs inferences to support flexible cognition.