Authors: Roman Fleischmann¹, David Meijer¹, Robert Baumgartner¹, Burcu Bayram², Ulrich Pomper², Michelle Spierings³
¹Austrian Academy of Sciences, Acoustics Research Institute
²Department of Cognition, Emotion, and Methods in Psychology, University of Vienna
³Department of Behavioral and Cognitive Biology, University of Vienna
Belief updating in statistical learning is well described by Bayesian inference models and accompanied by task-evoked pupil dilations, a proxy for locus coeruleus (LC) noradrenaline (NA) arousal system activity. LC-NA arousal is hypothesized to modulate belief updating by suppressing prior beliefs in favor of incoming evidence. The current study investigates whether LC-NA modulated belief updating, which drives conscious statistical learning, can also be observed during ongoing implicit decision making in auditory perception. Further, Bayesian inference models are often implied to be applicable in a generalized fashion over different domains and modalities in learning and perception. To this end, we investigate whether such a mechanism generalizes across spatial and temporal tasks. We devised a latent state discrimination task, targeting the spatial and temporal auditory domain separately. Two groups of participants were presented with auditory sequences of unpredictable length, providing evidences for, and sporadically changing between, two distinct latent states: temporal acceleration and deceleration in the first experiment and clockwise and counterclockwise spatial motion in the second. Participants were tasked with inferring and tracking latent states and reporting the last current state of the sequence in a 2AFC design. To extract per-stimulus estimates of pupil dilation despite the experiments’ rapid presentation (0.45-5 Hz) and the sluggish pupillary response we fitted a deconvolution-based general linear model to the pupil traces with a free gain parameter reflecting pupil dilation. We also fitted a Bayesian observer model to participants’ answers to estimate information gain, a latent variable quantifying the difference between prior and posterior beliefs. A linear mixed effects regression demonstrates a positive correlation with pupil dilation, while domain has little effect on pupil dilation, suggestive of a generalized arousal-mediated mechanism of perceptual inference within the auditory modality.
