Pattern separation is the key driver of expectation-modulated memory

The hippocampus uses pattern separation and pattern completion in a continuous manner to successfully encode and retrieve memories1,2. However, whether and how cognitive factors might modulate the dynamics between these types of computation is not well understood. Here we examine the role of expectation in shifting the hippocampus to perform pattern separation. Expectation can be built up through multiple contextual exposures leading to prediction (as in a learnt contingency) or through logical deduction based on a previous mnemonic response. Participants first learned a contingency between a cue and an object’s category (man-made or natural). Then, at encoding, one third of the cues that preceded the to-be-memorised objects violated the studied rule. At test, participants performed an old/new recognition task with old items (targets) and a set of parametrically manipulated (very similar to dissimilar) new foils for each object. We explored the effects of both contextual expectation, manipulated at initial encoding, and mnemonic-attribution expectation, driven by the mnemonic decisions taken on previous retrieval trials. For example, a target would be unexpected if in a previous trial a similar foil had been erroneously accepted as old. Memory was found to be better for foils of high and mid similarity to contextually unexpected targets at encoding, compared to expected ones. Additionally, violations of mnemonic-attribution expectation also yielded improved memory performance when the level of foil similarity was high. These results suggest that violations of both contextual expectation and mnemonic-attribution expectation engage pattern separation, resulting in better discrimination performance for these items. Importantly, this mechanism is engaged when input differentiation is required in order to make a correct recognition decision.