A prior for direct gaze
26.512, Saturday, May 11, 2:45 - 6:45 pm, Vista Ballroom
Isabelle Mareschal1,2, Andrew Calder3, Colin Clifford1,2; 1School of Psychology, The University of Sydney, 2Australian Centre of Excellence in Vision Science, 3MRC Cognition and Brain Sciences Unit, Cambridge, UK
Introduction: Determining where someone else is looking is critical to successful social interactions. Surprisingly, however, very little is known about how othersÂ’ direction of gaze is interpreted. Here we apply a Bayesian framework to determine whether normal adults have a prior expectation for the direction of someone elseÂ’s gaze. Methods: Gaze discrimination thresholds were measured in normal adult participants using grayscale male and female synthetic faces whose eye deviations were independently controlled using Matlab procedures. Observers viewed the same face in a two-interval presentation and had to judge whether the gaze deviation in the second interval was to the left or to the right of the gaze in the first interval. Discrimination thresholds were measured around 5 baseline gaze deviations: direct, to the left and right of direct (e.g. ± 2deg) and extreme left and right deviations (e.g. ± 9deg). Observers performed the task in three different conditions: (a) no noise on the eyes; (b) noise on the eyes in both faces; and (c) a mixed condition with noise on the eyes in one face only. Results: Gaze discrimination thresholds were constant across the 5 baselines tested and raised when noise was added to the eyes. In the mixed condition, observersÂ’ biases were shifted in a direction consistent with the noisy stimulus being perceived as more direct. Likelihood functions were derived from the threshold data in conditions (a) and (b) and observersÂ’ data in the mixed conditions were fit with a Bayesian model. The model accounted for approximately 90% of the variance and estimates of the priorÂ’s peak were near direct. Conclusion: Our data suggest that the adult nervous system internally represents a prior for gaze as close to direct.