Transfer of statistical regularity in visual search

Abstract

People are able to take advantage of statistical regularities in scenes, using those regularities to bias their attention to the likely locations of items of interest. People also seem able to learn object-centered statistical regularities—for example, that the top of an object is the most likely target location. We show here that such regularities transfer to new spatial locations even in the absence of any explicit object and hence may not be truly object centered. Additionally, when transfer is measured on a new object with a new shape—the transfer is substantially reduced. The findings suggest that statistical information about likely target locations can be encoded in a configuration-based reference frame that is sensitive to the context established by the objects in the scene. The results lead to a new interpretation of earlier findings and have important implications for understanding the coordinate systems in which attentional priorities are represented.