The Metastable Mind: Neural Underpinnings of Naturalistic Cognition Through the Synthesis of Event Segmentation and Metastable Neural States

Abstract: A multitude of findings and theories from cognitive, behavioural and computational neuroscience show that neural activity unfolds in a variety of meaningful temporal units. Behavioural research on event segmentation (ES) has shown that continuous experience is segmented into discrete events and sub-events, which aid real-time comprehension, memory, and decision-making. Computational neuroscience research observes and models ongoing brain activity as a series of stable population activity that occur across wide spatial and temporal scales, referred to as metastable neural activity (MNA). Through this review, we show that these isolated branches of literature, the cognitive theory of Event Segmentation (ES) and the mechanistic approach of metastability (MNA), actually study the same metastable neural states from different perspectives. While the behavioural branch offers a theory for the cognitive and behavioural utility of segmentation, the metastability literature provides the mechanistic account at the implementational level. We describe how metastable neural states act as the fundamental computational units of cognition and identify a number of core principles of how they operate. One is the spatio-temporally nested hierarchy of states, where longer-duration states in higher-order regions both constrain and are shaped by states in faster-operating regions. Another is that neural states are a reflection of underlying predictive models which shape perception, decision making, memory encoding and recall. And finally that neural states are periods of more modular processing, which are interspersed by boundaries where there is a reconfiguration of connectivity. Understanding how neural states emerge, interact, and shape cognition brings us closer to understanding the brain in its natural mode of operation.

• The paper clarifies that transient, metastable neural states serve as the units of naturalistic cognition by integrating event segmentation and dynamic brain activity.
• It employs data-driven methods like hidden Markov models to map neural state durations across cortical hierarchies, linking state transitions with behavioral events.
• The study proposes testable hypotheses on how state boundaries enhance prediction, memory encoding, and cognitive flexibility, offering translational insights.

The Metastable Mind: Neural Underpinnings of Naturalistic Cognition

Overview

"The Metastable Mind: Neural Underpinnings of Naturalistic Cognition Through the Synthesis of Event Segmentation and Metastable Neural States" (2605.31473) presents a comprehensive synthesis of two major lines of cognitive and computational neuroscience: Event Segmentation (ES) and Metastable Neural Activity (MNA). The central thesis is that these traditions, which have developed largely independently, actually converge on the same phenomenon—transient, quasi-stable neural states that underlie naturalistic cognition. The review articulates core principles rooted in empirical findings, offers a unified framework connecting cognition and neural implementation, and proposes multiple implications and testable hypotheses for future research.

Neural States: Definition, Scales, and Mechanisms

The concept of a neural state is formalized as a temporarily stable pattern of brain activity, observable at multiple timescales and spatial scales, from local neuronal ensembles to distributed large-scale cortical networks. Neural states are viewed as attractors in a high-dimensional state-space, with metastability referring to temporary proximity to these attractors, enabling flexible yet structured transitions.

Strong evidence for neural states is provided by diverse analytical methodologies, including hidden Markov models and data-driven state segmentation algorithms such as Greedy State Boundary Search. Empirically, neural states are observed from sub-second episodes in local circuits (e.g., on/off states, spiking packets) up to global brain states lasting minutes to hours, exhibiting hierarchical nesting. Importantly, neural states align with established spatial hierarchies; higher-order regions exhibit longer-duration states, constraining the temporal granularity of cognition in a manner paralleling the classic cortical processing hierarchy. Computational models link these dynamics to oscillatory coupling and propagating wave patterns, supporting an integrated spatiotemporal view of brain organization.

Transitions, Boundaries, and Event Structure

Transitions between neural states—boundaries—are conceptualized as brief windows of heightened network integration and reconfiguration of functional connectivity. The literature from MNA characterizes these periods as moments of maximized communication potential within the brain, while the ES framework identifies them behaviorally as event boundaries, often triggered by prediction error or contextual change.

The review emphasizes the convergence of these mechanisms: both theoretical strands support a view in which the brain operates via alternation between integration at boundaries and modular, segregated processing within stable states. Empirically, neural state boundaries are aligned with behaviorally salient event transitions and are modulated by predictability and familiarity, reflecting the continuous updating and reformation of generative internal models.

Functional Roles in Cognition

Perception, Decision, and Action

Neural state hierarchies organize not only perception but also decision making and action. Rapid sequences of local states encode fine-grained features (e.g., visual saccades), which are nested within more abstract object and scene representations, ultimately embedded in long-duration situation models maintained by prefrontal and default mode regions. State transitions at all levels structure behavioral output—decision difficulty is correlated with state duration and error rates in both humans and animals.

The review particularly highlights the evidence that attention samples the environment rhythmically, organized by neural oscillations (notably theta and alpha), temporal scaffolds that are reset and coordinated by state transitions. The bidirectional interplay across hierarchical levels—top-down constraints from goals and context, bottom-up salience signals—is intrinsic to the metastable state framework.

Predictive Processing

A central claim is that metastable neural states instantiate the brain’s generative models: within each state, predictions are formed and refined, and states persist so long as these predictions are consistent with sensory input. Upon surprise or contextual shift, boundaries are formed, and models are reconfigured. The review distinguishes between prediction at content and contextual levels, supporting a hierarchical Bayesian perspective: both low-level violations and high-level contextual changes can prompt state transitions, with Bayesian surprise being a strong predictor of event boundaries.

Empirical support is cited for anticipatory reinstatement phenomena, where upcoming event states are neurally pre-activated, and for the adaptive timing of boundary shifts with increased familiarity or expertise. Thus, predictions, memory encoding, and generative modeling are unified in the dynamics of state transitions.

Memory

The review synthesizes the hierarchical process memory (HPM) framework, mapping it onto metastable dynamics. Memory is not localized but is intrinsic to ongoing processing in the cortex, with integration windows corresponding directly to neural state durations at each level of the hierarchy. Empirically, boundaries are associated with hippocampal engagement and enhanced encoding; reinstatement and replay phenomena during rest and recall directly involve the recurrence of prior neural state patterns. Furthermore, memory is shown to be structured by the timing and nesting of state dynamics, crosscutting traditional distinctions between working, episodic, and semantic memory.

Synthesis and Core Principles

The authors extract several unifying principles:

1. Temporal and spatial hierarchy: Neural state durations increase along the cortical processing hierarchy, paralleling abstraction in cognitive representations.
2. Segregation and integration: Stable states are periods of modular specialization; boundaries are windows of high integration and reconfiguration.
3. States as generative models: Neural states are active, expectancy-based models of the environment, refined through accumulated experience.
4. Memory and states: State boundaries structure encoding, while the states themselves serve as substrates for memory storage and retrieval.
5. Bidirectional nesting: Higher-level states constrain fast local dynamics, but local states can precipitate changes in the global landscape.

A detailed scenario illustrates these principles via an everyday example, highlighting the multi-scale, dynamic nature of cognition and memory as mediated by metastable neural states.

Implications and Directions for Future Research

The synthesis presented generates concrete research directives:

• Cross-scale interactions: The mechanisms by which slow global states modulate fast local transitions require joint experimental-computational clarification.
• Oscillatory integration: The generalization of cross-frequency coupling and communication-through-coherence to slow, behaviorally relevant metastable dynamics is underexplored.
• Boundaries and generative models: Delineating the determinants of state transition thresholds (e.g., Bayesian surprise vs. surprisal) and their neurobiological implementation is critical.
• Within-state accumulation: Testing the prediction that predictive accuracy and integration increase within stable states across the hierarchy is a clear next step.
• Individual and clinical differences: Understanding how segmentation and metastable dynamics vary across individuals and clinical populations may clarify cognitive dysfunction.
• Neurostimulation and real-time intervention: Exploiting the window of integration at boundaries for targeted intervention is a promising translational direction.

Conclusion

The review robustly connects cognitive and neural levels by aligning the theories of event segmentation and metastable neural dynamics. The proposal that metastable neural states form the units of generative modeling, prediction, and memory provides a unified account with broad implications for cognitive neuroscience. Future work leveraging simultaneous multi-scale measurement, advanced analysis, and computational modeling will be essential to further interrogate and refine this unifying framework.

Citation: "The Metastable Mind: Neural Underpinnings of Naturalistic Cognition Through the Synthesis of Event Segmentation and Metastable Neural States" (2605.31473)