Kelly McEvoy¹, Dick Dubbelde², Sarah Shomstein¹; ¹The George Washington University, ²Georgetown University

Attentional selection operates on an object-based representation using both simple geometric shapes (e.g., rectangles) as well as semantically meaningful objects in real-world scenes (e.g., a cup on a table). While previous work suggests that both low-level (boundaries) and high-level properties (object meaning) contribute to object-based attention, the relative contribution of these features remains unclear. Here, we characterize the relative contribution of a consistent object border (i.e., object outline) and consistent object semantic information (i.e., object meaning) to object-based attentional selection in real-world objects. We adapted a two-rectangle task in which following a brief exogenous cue (150 ms) and a delay (70 ms), the attended object abruptly changed according to one of four conditions: an object with the same border but different semantic category, an object with a different border but same semantic category, an ‘outlier’ object with neither the border nor category in common, or no change at all. We predicted that both objects with the same border or same category contribute to the perception of objecthood but to varying extents, while ‘outlier’ objects contribute to an overall smaller object-based effect (OBE) compared to border and category. Objects that undergo no change will mimic classic object-based attentional selection (e.g., same object advantage). As predicted, OBEs were observed for all object change conditions, excluding outlier objects, with a significantly larger OBE for same category objects. These results were replicated in several follow-up experiments. Overall, object-based attentional guidance was modulated by the degree of low- and high-level changes in object properties. Our results suggest that object-based attention persists despite altering object properties in real-world objects, informing current models of attentional mechanisms and extending them to more naturalistic environments.

Acknowledgements: NSF BCS 2022572 to SS