Repeated Viewing of a Film Clip Changes Event Timescales in The Brain

1. Problem Statement

The human brain processes continuous experience by segmenting it into discrete events across multiple temporal scales (from milliseconds to minutes). While it is established that different brain regions operate on different "temporal receptive windows" (lower-level sensory areas on fast timescales, higher-order areas on slow timescales), it remains unclear how flexible these timescales are. Specifically, the authors investigate whether repeated exposure to a familiar stimulus alters the brain's internal representation of event structure. Do neural representations become finer (more detailed, shorter events) or coarser (more generalized, longer events) as familiarity increases? The study aims to determine if these changes depend on the narrative coherence of the stimulus and the stability of temporal regularities.

2. Methodology

Experimental Design & Stimuli

• Participants: 30 human participants (12 males, 18 females) undergoing fMRI.
• Stimuli: Three 90-second clips from the film The Grand Budapest Hotel.
• Procedure: Each clip was viewed six times.
• Conditions: The clips were manipulated to vary in narrative coherence and temporal stability:
  1. Intact: Original narrative order (coherent and stable).
  2. Scrambled-Fixed: Scenes randomly shuffled but kept in the same order for all 6 viewings (incoherent but stable).
  3. Scrambled-Random: Scenes shuffled differently for every viewing (incoherent and unstable).
• Behavioral Task: After the fMRI session, participants performed a free recall task, typing out details of the clips.

Neuroimaging & Analysis Pipeline

• Data Acquisition: 3-Tesla fMRI (TR = 1.5s) with multiband EPI sequences.
• Event Segmentation (HMM): The authors utilized a Hidden Markov Model (HMM) approach (Baldassano et al., 2017) applied to whole-brain searchlights.
  • The HMM was fit independently for each viewing, varying the number of events from 2 to 10 to capture a range of timescales (slow to fast).
  • This allowed the detection of event boundaries at multiple concurrent timescales within a single brain region.
• Metric: Within- vs. Between-Event Similarity:
  • To quantify the "strength" of event structure, the authors calculated the correlation of spatial activity patterns within HMM-defined events versus between events.
  • Higher values indicate distinct, stable event representations at that specific timescale.
  • Change Metric: The difference in this similarity value between the 1st viewing (unfamiliar) and the average of viewings 2–6 (familiar).
• Statistical Approach:
  • Permutation-based null hypothesis testing to determine significance.
  • Conjunction Analysis: Identified brain regions showing consistent directional changes (finer or coarser) across all three clip conditions, isolating stimulus-agnostic adaptation.
  • Brain-Behavior Correlation: Correlated changes in event structure with individual recall scores.

3. Key Contributions

• Methodological Innovation: Developed a framework to quantify event structure strength at multiple concurrent timescales within a single brain region, rather than assigning a single fixed timescale to a region.
• Demonstration of Flexibility: Provided empirical evidence that neural event timescales are not static architectural constraints but are dynamically reweighted by experience.
• Stimulus-Dependent Adaptation: Showed that the direction of change (fining vs. coarsening) depends on the stimulus structure (narrative coherence vs. disorder).
• Memory Link: Established a link between the change in neural event timescales and subsequent episodic memory performance.

4. Key Results

General Findings on Timescale Changes

• Stability vs. Flexibility: Most brain regions showed stable event structure across repetitions. However, specific regions exhibited significant shifts.
• Direction of Change:
  • Finer Segmentation: Some regions (e.g., lateral occipital cortex, fusiform gyrus) showed weaker slow-timescale structure and/or stronger fast-timescale structure with repetition, effectively dividing events into smaller sub-components.
  • Coarser Segmentation: Other regions (e.g., superior temporal sulcus, inferior frontal cortex) showed stronger slow-timescale structure and/or weaker fast-timescale structure, grouping distinct events into larger, more generalized units.

Condition-Specific Effects

• Intact (Coherent): Showed a topographically organized pattern. Visual regions (lateral occipital) became finer, while higher-order frontal regions became coarser.
• Scrambled-Fixed: Showed subtle changes, primarily in lateral temporal cortex.
• Scrambled-Random: The majority of regions (87%) shifted toward coarser representations. The lack of narrative structure appeared to drive the brain to integrate information over longer timescales to impose order.

Cross-Clip Generalization (Conjunction Analysis)

• Lateral Occipital Cortex (LOC): Consistently showed a weakening of slow-timescale structure (becoming finer) across all three conditions.
• Middle Temporal Gyrus (MTG): Consistently showed a strengthening of slow-timescale structure (becoming coarser) across all three conditions.
• Interpretation: These regions adapt their integration windows based on general familiarity mechanisms rather than specific clip content.

Relationship to Memory

• LOC: Participants with greater weakening of slow-timescale structure (finer segmentation) in later viewings had better recall. However, baseline (1st viewing) slow-timescale structure was a stronger predictor of recall than the change itself.
• MTG: Participants with less strengthening of slow-timescale structure in later viewings had better recall. This suggests that for MTG, maintaining finer boundaries (avoiding excessive abstraction) during repeated viewing supports detailed memory.
• Conclusion: The relationship between event structure and memory is region-specific and depends on whether the region is tracking visual details (LOC) or semantic integration (MTG).

5. Significance

• Dynamic Event Processing: The study challenges the view that temporal receptive windows are fixed. Instead, the brain actively reconfigures its event segmentation strategy based on familiarity, shifting the balance between fine-grained detail and coarse-grained abstraction.
• Mechanism of Learning: The findings suggest that familiarity allows the brain to either "zoom in" on subtle transitions (fine-tuning) or "zoom out" to integrate information into broader narratives (coarsening), depending on the region's function and the stimulus context.
• Implications for Memory: The results indicate that optimal memory encoding may require a dynamic interplay between initial coarse scaffolding (to organize the scene) and subsequent fine-grained processing (to encode details), or vice versa, depending on the brain region.
• Methodological Impact: The multi-timescale HMM approach provides a robust tool for future research into how the brain constructs narratives, processes routines, and adapts to variable temporal structures in naturalistic settings.