Makayla Souza-Wiggins¹, Joy J. Geng¹; ¹University of California, Davis, ²Center for Mind and Brain

Real-world scenes follow predictable spatial patterns, with some objects more likely to co-occur with other objects. Anchor objects (i.e., large, stable objects that predict smaller objects) are strong predictors of where smaller objects are likely to appear in a scene (e.g., a sponge in the kitchen sink). We hypothesized that under degraded visual conditions, observers will rely more heavily on predictable anchor objects to find targets. To test this, we developed 40 scenes in Unity and asked participants to search for a novel exemplar of an object within each scene under high- or low-visibility conditions. Within a scene, the target (sponge) could appear in a congruent location (by the sink), an incongruent location (on a chair), or be absent. Participants responded whether the target was present or absent (N= 51). There were significant interactions between visibility and congruency in both RT and accuracy due to larger congruency effects in the low-visibility condition (RT: β=55, p<0.001; accuracy: β=0.58, p<0.001). This suggests that when low visibility hinders target detection, we rely more on anchor-based spatial predictions as proxies to guide attention. This dependence produces a larger spatial congruency effect as expectations based on prior knowledge compensate for perceptual limitations. Dwell time on the anchors support this finding: under high-visibility conditions, only highly predictive congruent anchors (determined by a separate judgment task) were fixated longer. In contrast, all congruent anchors were fixated for longer under low-visibility conditions. This suggests that even less predictive anchors were relied upon under low-visibility to help disambiguate targets (β=0.14, p=0.015). Together, these results show that degraded visibility increases reliance on anchor objects for guiding attention, whereas under high-visibility conditions, anchors are used more selectively, specifically when they are highly predictive of the target.

Acknowledgements: NEI T32 Vision Training Grant