Shachar Lando^1,2 (), Roy Luria¹, Halely Balaban²; ¹Tel Aviv University, ²Open University Israel

Perceptual grouping enables effective information compression in visual working memory (VWM), but the precise computations that underlie grouping remain unclear. Recently, it has been suggested that VWM relies on a content-independent pointer system to establish unique representational instances, and to keep them in line with environmental changes. In three experiments, we examined whether, and under what conditions, a group-specific pointer might be assigned to the product of the grouping process. We recorded EEG and monitored the contralateral delay activity (CDA), a neural marker that rises with the number of representations in VWM. Notably, the CDA also reflects pointer dynamics, exhibiting a sharp drop in amplitude when a pointer is invalidated. Participants performed change-detection tasks where three Pacmen moved either independently or as a Kanizsa triangle (movement was task-irrelevant). In all studies, the Kanizsa induced information-compression, reflected by lower CDA amplitudes and higher accuracy rates than for three independently-moving items. In 50% of trials, all Pacmen abruptly changed trajectory, which in the grouped condition broke the Kanizsa formation. We contrasted different tasks (holding the animation sequence constant), and examined the resulting pointer dynamics. When the test involved judging the orientation (grouping-relevant feature) of only a single Pacman, after the trajectory-change the CDA gradually rose (reflecting de-grouping) without a drop. This indicates that the pointers remained intact when the group broke, meaning the group was not allocated its own pointer. Similar results were found when the test required judging the (grouping-irrelevant) location of the entire Kanizsa. Yet, when the test involved judging the (grouping-relevant) orientation of the entire Kanizsa, the trajectory-change triggered a CDA-drop, suggesting that now there was a group-specific pointer. The results suggest that seemingly-similar VWM-grouping can be supported by different pointer dynamics in different contexts, depending on how perceptual factors and task demands join to signal the relevant unit.