Exploring how high-level relations function in working memory: Unique mechanism of social grouping compared to nonsocial grouping.

Humans efficiently perceive high-level relations and use them to facilitate grouping. For instance, interactive agents can be grouped into coherent social units (social grouping), while nonsocial relations like physically compatible shapes or semantic relatedness form nonsocial grouping. In this study, we examined how grouping based on high-level relations functions in working memory (WM). While previous studies have demonstrated that both social and nonsocial relations enhance WM capacity for the relations themselves, these findings provide only a partial answer to how high-level relations function in WM. A critical yet unresolved issue is whether and how high-level relations organize content orthogonal to these relations. We addressed this question from the perspective of feature binding, investigating the impact of social and nonsocial grouping on WM for bound features. Participants memorized bound colors corresponding to social or nonsocial grouping cues. Their responses were fit to the swap model to isolate distinct representational components (the bias and standard deviation of memory, swap rate, and guess rate). Across 13 experiments, we found that grouping based on high-level relations can organize content orthogonal to these relations, but only when the contents themselves are chunkable. Moreover, social and nonsocial grouping functions differently in WM. While social grouping cues led to both enhanced accuracy (absolute memory bias) and increased confusion (swap rate) during maintenance, nonsocial grouping only resulted in heightened confusion during encoding. Together, our results provide a comprehensive account of how grouping based on high-level relations functions in WM and reveal qualitative differences between social and nonsocial grouping. (PsycInfo Database Record (c) 2025 APA, all rights reserved).