Hierarchical coding of local and global positions in episodic memory for naturalistic sequences

Life unfolds continuously across multiple timescales, yet how the brain encodes hierarchical event sequences remains poorly understood. Using magnetoencephalography, we investigated the neural coding of hierarchical positions in naturalistic sequence memory. Participants viewed video sequences with a nested structure and later recalled their serial order. Behavioral analyses revealed transposition errors for items occupying identical global or local position. Multivariate classification and subspace analyses revealed separable neural representations of global and local positions. Local positions were better decoded in high-dimensional space, whereas global positions exhibited greater pattern separability in low-dimensional space. Brain regions encoding global positions exhibited longer neural timescales and reduced sensitivity to low-level event transitions compared to regions encoding local positions. These findings support a parallel hierarchical organization of content and context, a novel framework that enables integration of item and order across hierarchical levels and provides principles for hierarchical representations in both human memory and artificial intelligence.