Structural and contextual biases interact to shape duration perception

Abstract Human time perception is flexible and shaped by both structural constraints and contextual influences. Disentangling these sources of bias is essential for understanding the predictive mechanisms underlying temporal perception, yet no unified model currently integrates them. Here, we quantified precisely the structural and contextual biases in a duration discrimination task to constrain models of duration perception. Using a two-interval duration discrimination task, participants judged which of the two stimuli lasted longer. Stimuli were both visual, both auditory, or one of each modality. Consistent with previous findings, auditory stimuli were perceived longer than visual stimuli of equal duration, reflecting intrinsic properties of audiovisual neural processing. Contextual biases were manipulated through different duration distributions from which stimuli were drawn, a procedure known to influence temporal judgments. Our results show that duration discrimination relies on distinct representations of each stimulus distribution and is best explained by a combination of Bayesian inference and rescaling. While Bayesian inference accounts for contextual effects that bring perceived durations closer to the mean of each distribution, rescaling mechanisms have the opposite effects. These findings challenge existing accounts of contextual effects in time perception and suggest that the brain normalizes temporal representations to adapt to environmental statistics. Fundings This work was supported by the FRC and the UNAFAM association, and scientific expertise is provided by the FRC’s Scientific Advisory Board, and the Anneliese Maier Award to Pascal Mamassian from the Alexander von Humboldt Foundation. Data availability Behavioral data and modeling codes will be open-access upon publication.