Abstract
Through the use of classical conditioning procedures, the perceived direction of rotation of a wire-frame cube, which is a bistable stimulus, can be trained to depend on stimulus position (Haijiang, Saunders, Stone, & Backus, 2006). Backus & Haijiang (2007) tested whether this effect was mediated by retinal position or position in the world. They trained two groups of subjects using consistent retinal or world positions, respectively, through trial-by-trial manipulation of fixation. Cue recruitment occurred quickly for retinal training but did not reach statistical significance for world training. We revisited the question of whether position in the world can be recruited as a cue in new experiments that were designed to facilitate the use of world position as a cue. A rotating wire-frame cube appeared at one of two world positions (screen locations) on each trial. On training trials, the cube's rotation direction was specified by disparity and occlusion cues in accordance with its position. On test trials there were no disambiguating cues. World and retinal position were deconfounded by manipulating subjects' fixation, using a marker that appeared at the start of each trial either to the left or right (Experiment 1) or at one of 16 locations (Experiment 2) relative to the cube. Our results show that the visual system can learn world position cues to a certain extent, but retinal position competed strongly as a determinant of apparent rotation direction, despite lack of consistent training in retinal coordinates. The relatively stronger learning of retinal position could be explained either by the behavior of specific mechanisms (early, retinotopically organized cortical areas such as MT may be trainable) or by appealing to computational principles (with location in the visual field being assumed by learning mechanisms to have greater ecological validity than location in the world).