Yin-ting Lin¹ (), Lisa M. Heisterberg¹, Mackenzie J. Siesel¹, Andrew B. Leber¹; ¹The Ohio State University

Visual working memory (WM) is limited, but we can strategically use grouping cues to improve performance. Work has revealed better performance for items forming an illusory shape (Kanizsa illusion) versus randomly oriented items (Allon et al., 2019). However, the mechanism behind this benefit is still unclear. Some have proposed that Kanizsa illusions improve memory by reducing storage demands, but existing studies have found mixed evidence (Heisterberg, 2021; McCollough, 2011; see also Diaz et al., 2021). Notably, the studies that found reductions in storage used longer stimulus encoding intervals. We propose that participants need these long durations to integrate the illusory shape and thus reduce their memory load, and we test this in the present study. Participants performed an orientation change detection task with EEG recording. The memory array consisted of either one “pacman” stimulus (single), three randomly oriented stimuli (proximity), or three stimuli that form an illusory triangle (Kanizsa). Critically, we manipulated encoding time (short vs. long) between subjects. Results show a Kanizsa benefit in accuracy for both groups. Importantly, we used both the contralateral delay activity (CDA) and a multivariate classifier to examine the amount of information stored in WM. There was a greater CDA amplitude for proximity versus single condition for both groups, showing that CDA is sensitive to WM load. However, we did not see a statistically reliable difference between Kanizsa and proximity conditions for either group. The multivariate classifier trained on single and proximity conditions was able decode WM load reliably above chance. Interestingly, the multivariate analyses suggest that there is a reduction in WM load for the long group but not for the short group. Our findings raise the possibility that Kanizsa benefits do not always result from reduced storage demands. Rather, chunking may reduce WM load only when there is sufficient encoding time.