Workshop: Evelina Overlingaite and Luigi Acerbi

Luigi Acerbi

Internal representations of temporal statistics and feedback calibrate motor-sensory interval timing

People exhibit a variety of biases and illusions in time perception - some instances may be explained in terms of probabilistic inference, but others seem to defy a simple explanation. The underlying assumption of many experiments is that people acquire the correct temporal statistics of the task, but this has never been verified for complex distributions of durations (e.g. highly skewed). In our recent work, we investigated how various distributions of time intervals and shapes of performance feedback affected people responses in a sensorimotor timing task. In particular, we studied how well participants learnt or approximated the experimentally provided distribution of intervals - with the idea that the internally acquired statistics might differ from their objective counterparts.  We found that the responses of the subjects in the task were well explained by a Bayesian ideal observer model that takes into account that people learn smoothed approximations of the experimental distributions. Moreover, we were able to non-parametrically reconstruct the subjective priors acquired by the participants, and we found that people were generally able to learn mean, standard deviation and skewness of the true distributions.  Our results show quantitatively what are the capabilities of human learning for complex, skewed temporal distributions of intervals with the help of corrective feedback, filling a gap in the literature which focussed only on symmetrical, simple distributions. This research is a starting point for addressing within a sound framework complex phenomena of temporal biases and adaptation.