Agnes Rigo¹ (), Daryl Fougnie¹; ¹Division of Science and Mathematics, New York University Abu Dhabi

Our environment contains far more visual information than we can process, making prioritization critical. Environmental regularities can guide attention and improve visual processing through statistical learning. A key question is how complex the information guiding attention can be. What happens when the regularities are not directly present in the visual input, but instead must be internally constructed? To address this question, we employed a contextual cueing paradigm in which the color and shape of a pre-cue explicitly defined the target while implicitly predicting target spatial location in a subsequent visual search task. For half of the color-shape pairings, the target appeared in one of eight positions with 100% consistency (mapped), whereas the remaining pairings were unmapped and did not predict target location. Critically we manipulated the relation between color and shape information such that it was either present as a fully formed object (Experiment 1) or required mentally combining features from two distinct objects, either in separate (Experiment 2) or overlapping spatial positions (Experiment 3) (all n=20). Experiment 1 produced a robust contextual-cueing effect: mapped trials were significantly faster than unmapped trials (t(19)=4.760, p<.001). Critically, this effect remained even after excluding participants who reported noticing the mapping, indicating that the guidance happened implicitly. Experiments 2 and 3, in contrast, showed no evidence of statistical learning: reaction times were equivalent for mapped and unmapped trials (Experiment 2, t(19)=1.022, p=.320; Experiment 3, t(19)=.430, p=.672), except among participants who explicitly reported noticing the mapping. Thus, while participants were able to explicitly combine color and shape (search performance was >90%) they showed no evidence of guidance when these features were split across objects. These findings suggest surprising limits on statistical learning: it succeeds when regularities are visible and contained within an object, but fails when they must be mentally constructed from separate objects.

Acknowledgements: This work was supported by the NYUAD Research Institute grant CG012 under the NYUAD Center for Brain and Health.