Uriel Gonzalez¹ (uriel.gonzalezbr@rutgers.edu), Jacob Feldman¹; ¹Rutgers, The State University of New Jersey

Understanding others’ actions and intentions is central to human perception, cognition, and social interaction. Classic research on biological motion shows that people can infer complex actions from reduced visual information (Johansson, 1973). Most studies have examined single agents; however, many social actions gain meaning only through interactions between two people (e.g., Ikeda, Destler, & Feldman, 2025). This study examines how people interpret biological-motion displays comprising two individuals, focusing on whether, and to what degree, such interpretations depend on their joint action. Observers viewed either both agents together or each one in isolation. Each subject (N = 145) viewed 24 short videos depicting social actions, presented as point-light displays (PLDs) or stick figures. The PLD and stick-figure stimuli were generated from naturalistic videos using OpenPose to extract joint positions. Each action appeared in one of three versions—full interaction, only agent A, or only agent B—counterbalanced so that no participant viewed the same action twice. After each video, participants provided free-text descriptions of the perceived action. A separate participant group viewed the full interactions, and their responses were used to derive reference labels. We used an AI-based text analysis procedure to compute semantic similarity among participants’ responses, quantifying (a) agreement among observers and (b) similarity between their descriptions and the reference labels. A hierarchical Bayesian model estimated posterior distributions across representation type, interaction condition, action function, and agent role. Observers showed higher agreement when viewing both agents interacting, demonstrating that social context enhances action understanding. Agreement among participants was more variable than similarity to the reference labels, which remained stable across conditions. When viewing only one agent, observers showed lower agreement and sometimes converged on a different action, indicating that the absence of a partner changes perceived meaning. Together, these findings shed light on how humans interpret moving and interacting human forms.