Breaking the chains: Toward a neural-level account of episodic memory.

It has been suggested that episodic memory relies on the well-studied machinery of spatial memory. This influential notion faces hurdles that become evident with dynamically changing spatial scenes and an immobile agent. Here I propose a model of episodic memory that can accommodate such episodes via temporal indexing. Indices in the model have flexible duration, capable of exhibiting both fixed duration and broadening time fields akin to classical time cells. The latter cannot index episodes beyond short durations and are reminiscent of timing codes in scalar expectancy theory. Contrary to timing repetitive events, the present model focuses on the one-shot indexing of within-episode structure. Hippocampal indices are recruited by a combination of contextual inputs, lateral inhibition, and drive from temporal analogues of grid cells, functioning as an on-demand sequence generator and memory store. Indices learn connections to cortical representations, modulated by an amygdala signal. This architecture relies on biologically plausible, common network motifs, which can replay dynamically changing and spatially structured events, while an agent is immobile, suggests a mechanism for modulating the speed of recall, and can replay disjoint collections (i.e., broken chains) of indices with preserved temporal order. The model is embedded in an extensive review/perspective along two conceptual axes: first, how the model fits in with other accounts of time coding, serial order memory, and flexible temporal cognition and, second, how we can simultaneously reconcile the model framework with classical accounts of episodic memory à la Tulving, as well as with modern reinforcement learning and generative model accounts of hippocampal function. (PsycInfo Database Record (c) 2025 APA, all rights reserved).