New paper unites event segmentation and metastable neural states theories

A new review paper from Dora Gozukara, Nasir Ahmad, Djamari Oetringer, and Linda Geerligs (Radboud University) proposes a synthesis of two long-separate neuroscience fields: Event Segmentation (ES) theory from cognitive and behavioral neuroscience, and Metastable Neural Activity (MNA) from computational neuroscience. Published on arXiv in May 2026, the 24-page work argues that both literatures actually describe the same underlying phenomenon — metastable neural states that serve as the fundamental computational units of cognition. The behavioral branch provides the cognitive theory of why the brain segments continuous experience into discrete events (to aid comprehension, memory, and decision-making), while the metastability literature offers the mechanistic implementation-level account of how stable population activity patterns emerge across wide spatio-temporal scales.

The authors identify three core principles governing these metastable neural states. First, they form a spatio-temporally nested hierarchy: longer-duration states in higher-order brain regions (e.g., prefrontal cortex) both constrain and are shaped by faster states in lower-level sensory regions. Second, neural states reflect underlying predictive models that actively shape perception, decision-making, memory encoding, and recall. Third, states are periods of relatively modular processing, interspersed by boundaries where large-scale connectivity reconfigures — these transitions are critical for updating predictions and segmenting experience. Understanding how metastable states emerge, interact, and shape cognition moves us closer to modeling the brain in its natural, dynamic mode of operation, with implications for both neuroscience and AI architectures that learn from naturalistic stimuli.