[PS-2.9] Information integration and prediction during statistical learning can be read off from anticipatory Fixation Biases

Notaro, G. , Van Zoest, W. , Melcher, D. & Hasson, U.

CIMeC, The University of Trento, Italy

Statistical learning (SL) is tightly linked to predictions and their evaluation. Yet, behavioral indices such as saccade latencies likely reflect both construction and evaluation of predictions, and are therefore difficult to interpret. Here we report a new occulomotor measure that aims to directly isolate construction from evaluation of predictions during SL. In 2 saccade-to-target studies we measured very subtle fixation biases (FB) around the screen center *prior* to stimulus presentation and evaluated how these were impacted by the statistical structure that governed target location. In Exp1, stimuli were presented to the right or to the left of screen-center following two stochastic Markov processes. FB magnitudes were small (< 1deg), significantly biased consistently with the statistical structure, and strongly impacted by the transition structure in the recent 4-5 trials as well as longer-term statistics. In Exp2 stimuli were presented in four screen positions, following either a random or Markov process. Here too FB discriminated the statistical processes: FB distributions were tighter in the location-regular condition, and this was accompanied by more predictable trajectories, demonstrating that these biases were driven by the Markov process. We conclude that FB offers a new, information rich perspective on information integration and anticipatory behavior during SL.