Abstract Memory must balance preserving episode-specific details with extracting generalizable structure. Here, we test whether spontaneous hippocampal reactivation during postencoding rest actively transforms memories by strengthening semantic gist at the expense of episodic detail. Twenty-four participants encoded Deese–Roediger–McDermott word lists presented in male or female voices, creating orthogonal semantic (list theme) and episodic (voice source) information. Using high-resolution fMRI, we tracked list-specific reactivation in hippocampal subfields during a 90-sec delay period. Most consistently in CA1, reactivation frequency showed opposing relationships with subsequent memory: Greater reactivation predicted increased false recognition of semantic lures (gist extraction; dz = 0.52) and decreased accuracy for voice discrimination (source memory; dz = −0.65), yielding a large, bidirectional effect within CA1 (gist minus source, paired dz = 1.00). This bidirectional pattern was most prominent in anterior-to-mid hippocampus, consistent with its role in schematic processing. These findings reveal that hippocampal reactivation implements a computational trade-off, actively constructing abstract and generalizable representations while competitively weakening episode-specific details. This mechanism could explain how the brain rapidly extracts meaning from experience, with false memories emerging as a natural consequence of adaptive memory transformation rather than retrieval failure.
Hindy et al. (Sun,) studied this question.