Bayesian priors in estimates of object location in virtual reality

We investigated the category bias in spatial memory, which reveals the influence of a region (i.e., a spatial category) on memory for specific locations within the region's bounds. The standard approach to investigating the category bias employs a static dot-in-circle task, in which observers indicate the location of a single dot from memory after a brief interval. The agreement in the literature is that these location estimates result from Bayesian principles; however, the priors in the dot-in-circle task are geometric prototypes (the central angular value of each quadrant and two-thirds of the radius from the center of the circle to its circumference). These geometric prototypes are not "true" priors in that they are not pre-existing statistical likelihoods of a target's location before other evidence is considered. In this paper, we tested the category bias with items for which informative priors exist (e.g., a vase, which is expected to be in the center of a table) and found that people favor them over geometric prototypes for estimating angular but not radial target positions. Our work contributes to the literature by showing that localizing common everyday objects in a circular space is not restricted to the use of cues intrinsic to the space. This is important because the majority of the empirical data on the category bias derives from locating targets that have little to no semantic information.