Greer Gillies¹, Keisuke Fukuda², Jonathan Cant³; ¹University of Toronto, ²University of Toronto, Mississauga, ³University of Toronto, Scarborough

Ensemble coding operates across a range of visual features (e.g., average orientation, average color) and stimuli (e.g., triangles and faces). However, ensemble displays usually contain stimuli from only one category. It is unclear if the visual system can extract summary statistics (1) across different stimulus categories to make comparisons, or (2) from single displays containing multiple stimulus types. In Experiment 1, participants compared average orientation (same/different) across two ensembles that were made up of either the same or different stimulus types (faces, shoes, triangles). We found that participants were just as accurate when the stimulus types were the same for both ensembles compared to when they were different, suggesting that ensemble abilities are governed by domain-general mechanisms. In Experiments 2 and 3, participants were shown a single ensemble that was made up of either one stimulus type (e.g., triangles) (pure condition) or two (e.g., triangles and faces) (mixed condition), and then determined whether the direction of a subsequent probe arrow was representative of the average orientation. There were no differences across stimulus types in the pure condition. However, in the mixed condition, stimulus specificities emerged. Specifically, when the ensemble contained triangles, participants had a hard time rejecting the probes that matched the orientation of just the triangle subset, and further evidence of this bias was found via a continuous-report task in Experiment 4. In summary, when every item within an ensemble is from the same category, the ability to extract summary statistics is similar across ensembles composed of different stimuli, but having more than one type of stimulus present in an ensemble interferes with the extraction/report of summary statistics. This suggests that ensemble coding abilities are governed by a largely domain-general mechanism, but specificities do emerge under certain possibly relating to the ease of computing shape-based orientation values.