Modality-General Sensitivity of Pupil Responses to Regularity Violations

Hamit Başgöl, F. J. Raab, Peter Dayan, Volker H. FranzPsyArXiv 2025 · DOI

TL;DR

This study reveals that pupil dilation, a physiological marker of arousal, responds similarly to violations of statistical regularities in both visual and auditory stimuli. This suggests a domain-general mechanism for detecting unexpected events, even though subtle modality-specific differences exist.

What Problem Does This Paper Address?

The brain is remarkably adept at detecting patterns and regularities in sensory input. When these regularities are violated, it triggers a response, often reflected in physiological markers like pupil dilation. While pupil dilation has been extensively studied in response to auditory regularity violations, its behavior in the visual domain and, crucially, its cross-modal consistency have been less clear. This paper aims to bridge this gap by systematically comparing pupil responses to visual and auditory sequences and transitions, investigating whether the brain's detection of statistical structure and its violations is a general process or modality-specific.

Methods

The researchers employed a paradigm involving sequences of visual dots moving on a screen and auditory tones. They created "regular" sequences (repeating patterns) and "random" sequences. The core of the experiments involved transitions between these sequences:

  • Regular to Random (REG-RAND): Violating an established pattern.
  • Random to Regular (RAND-REG): Establishing a new pattern.
  • Regular to Regular (REG-REG): Transitioning between two different regular patterns.

Pupil dilation was measured using eye-tracking technology. To isolate the effects of regularity violations from other factors, participants performed an incidental task (e.g., detecting a gap) to maintain attention without directly engaging with the statistical properties of the sequences. They also analyzed pupil events (phasic dilations and constrictions) to get a more granular view of the responses.

Datasets

The study did not use pre-existing datasets for training or evaluation. Instead, they generated novel visual and auditory sequences for their experiments. The data collected from participants in Experiments 1, 2, and 3 form the basis of their analysis.

Key Insights and Ideas

  • Pupil Dilation as a Universal Surprise Detector: The most striking finding is that pupil dilation reliably signals violations of statistical regularities, regardless of whether the stimuli are visual or auditory. This points to a fundamental, domain-general mechanism in the brain for detecting unexpected events.
  • Emergence vs. Violation: The pupil dilates significantly when a regular pattern is violated (REG-RAND, REG-REG), but not when a new regularity emerges (RAND-REG). This suggests the brain is more sensitive to disruptions of established predictive models than to the formation of new ones.
  • Modality-General Arousal: The strong cross-modal similarity in pupil responses, particularly for REG-RAND transitions, suggests that pupil-linked arousal reflects a general state of surprise or uncertainty that transcends sensory modalities.
  • Subtle Modality Differences: While the overall patterns are similar, there are nuances. For instance, transitions between two regular sequences (REG-REG) showed some modality-specific differences in the size of dilation events, suggesting that while the detection of the violation is general, the magnitude of the response can be influenced by modality.
  • Pupil Events Offer Granularity: Analyzing individual pupil events (phasic dilations) provided a more detailed picture, revealing that even when overall pupil size changes are subtle, underlying event-based responses can still be present and show modality-specific characteristics.

Key Results

  • Experiment 1 & 2 (Visual): Confirmed that transitions from regular to random visual sequences (REG-RAND) and between two regular sequences (REG-REG) elicited pupil dilation, while random to regular transitions (RAND-REG) did not. The REG-RAND transition resulted in a larger dilation than REG-REG.
  • Experiment 3 (Cross-modal): Demonstrated strong cross-modal similarity in pupil dilation for REG-RAND transitions, both in time-series data and event analysis. For REG-REG transitions, while overall patterns were similar, there were some differences in the size of dilation events between visual and auditory modalities.
  • Baseline Trials: Even during baseline (REG5) trials without explicit transitions, pupil responses showed some cross-modal similarity, suggesting a general level of uncertainty or arousal might be tracked.
  • Surprise Modeling: A hierarchical Chinese restaurant process model indicated that REG-RAND transitions led to more sustained surprise compared to REG-REG transitions, aligning with the observed pupil dilation patterns.

Limitations and Future Work

  • REG-REG Inconsistency: The authors note that the REG-REG transition's pupil response was less consistent across experiments and modalities than REG-RAND. This might be due to the nature of the regularity itself (e.g., visual regularities being harder to learn than auditory ones) or subtle differences in the experimental setup.
  • Signal-to-Noise Ratio: The weaker and less stable REG-REG responses might indicate a lower signal-to-noise ratio for these transitions, potentially due to the difficulty in learning and violating these specific types of regularities.
  • Pupil as a Composite Readout: The authors acknowledge that pupil size is influenced by many factors beyond prediction errors, including arousal, effort, and circadian phase. While their experimental design aimed to minimize these, they remain potential confounds.
  • Future Directions: The authors suggest future research should:
  • Manipulate expected uncertainty independently of transition type.
  • Investigate how learning speed and pattern familiarity influence responses.
  • Model pupil traces as a generative process combining event-level responses with tonic state variables.
  • Explore how visual modality can surpass auditory modality in certain tasks.
  • Investigate the role of luminance and stimulus properties on pupil responses.
  • Examine how pupil size dynamically evolves in multimodal sequences.

Summary updated

Abstract

Pupil dilation responses are reliable physiological markers of arousal in reaction to unexpected events across sensory modalities. However, a direct comparison across modalities has not yet been conducted. In this study, we address this gap by examining pupil responses to regular and random sequences of visual dots and auditory tones, as well as transitions between these sequences. In Experiment 1, we showed dilation responses when a regular sequence of visual dots changed to a random sequence, but not to the reverse transition. A transition between two different regular sequences led to less dilation on average. In Experiment 2, we replicated these findings, confirming the reliability of the observed dilation patterns. In Experiment 3, we compared the pupil responses when sequences and transitions were presented in visual versus auditory modalities. We observed strong cross-modal similarity in pupil sizes, particularly for transitions between regular and random sequences. Notably, these similarities persisted even during baseline trials, when no explicit transition occurred. We also decomposed the time-series pupil size to approximate phasic pupil dilation events. The patterns of dilation events aligned with the overall dilation, with modality-specific dilation event size differences primarily in the context of transitions between different regular sequences. Overall, our findings suggest that pupil-linked arousal reflects imperfect inference of statistical structure and its violations, exhibiting substantial similarity across modalities.
BibTeX
@misc{basgol2025modality,
  author = {Hamit Başgöl and F. J. Raab and Peter Dayan and Volker H. Franz},
  title = {{Modality-General Sensitivity of Pupil Responses to Regularity Violations}},
  year = {2025},
  doi = {10.31234/osf.io/um9fe\_v1},
  howpublished = {PsyArXiv},
}
Figure