Evidence for complementary learning systems within the hippocampus

Decades of research have established the hippocampus as central to episodic memory, but growing evidence suggests that it also contributes to structure learning, rapidly extracting regularities across experiences in support of prediction and generalization. These functions impose conflicting demands: episodic memory requires experiences to be separated, whereas structure learning requires integrating across them. The C-HORSE model, a biologically grounded neural network of the hippocampus, proposed an anatomical division of labour that resolves this tension. The trisynaptic pathway (TSP), characterized by high plasticity and strong pattern separation, supports rapid learning of individual episodes, consistent with prior theory and data. By contrast, the monosynaptic pathway (MSP), which learns more incrementally and with less separation, supports the extraction of regularities across experiences. This review evaluates how C-HORSE aligns with empirical findings on hippocampal involvement in structure learning. We review evidence from domains that benefit from quickly detecting regularities, including statistical learning, motor sequence learning, inference and category learning, surveying the role of the hippocampus in each and assessing evidence for functional dissociations across pathways in relation to model predictions. Together, this synthesis highlights strong convergence between C-HORSE and the empirical literature, identifies the MSP as a key contributor to structure learning and raises open questions for future experimental and modelling work. This article is part of the theme issue 'The role of hippocampal predictions in cognition: bridging perception and memory'.