Abstract
<title>Abstract</title> <p>Sleep is widely believed to consolidate and transform newly encoded episodic memories into more stable and generalized representations, yet the underlying neural mechanisms remain debated. Here, we present a biologically grounded neurocomputational model that links hippocampal episodic memory with a cortical sensory-semantic–lexical network to explain how non–rapid eye movement (non-REM) and rapid eye movement (REM) sleep differentially shape memory consolidation. The model integrates a theta–gamma-coded hippocampal sequence memory with a cortical network that encodes semantic features and lexical forms using exclusively Hebbian and anti-Hebbian plasticity rules. During simulated non-REM sleep, hippocampal replay drives Hebbian potentiation of heteroassociative synapses from semantic to sensory cortical layers. This mechanism transforms strictly autobiographical, context-bound episodic sequences into generalized, semantized representations that can be retrieved independently of the hippocampus when multiple object-level cues are present. In contrast, REM sleep simulations—characterized by hippocampal disengagement, cortical theta modulation, and increased noise—produce fragmented and recombined memory patterns, consistent with dream-like associative and creative processes. The model reproduces key empirical findings on sleep-dependent abstraction and generates testable predictions about changes in cortical connectivity and cue-dependent retrieval after sleep. By relying solely on physiologically plausible Hebbian mechanisms and oscillatory dynamics, this framework provides a parsimonious account of how sleep promotes the transformation of episodic sequential experiences into more semantized experiences while preserving hippocampal specificity.</p>