Parallel and consecutive processing stages for social interaction recognition

Understanding a social interaction requires –at least– the recognition of social agents (i.e., people), their individual actions, and the relationship between them. In what sequence do these processing steps occur? In two behavioral experiments (n = 123), we presented visual scenes showing bodies (or sometimes chairs), in spatial relations that implied interaction (face-to-face) or not (back-to-back), and in postures that suggested different types of interactions (i.e., ‘fighting’ or ‘dancing’). We presented these scenes for different time-durations (17, 33, 50, 67, 167 ms) followed by masking, as a way to manipulate the available processing time. In three different tasks, participants had to report the object category (body/chair), the spatial relation (facing/non-facing) or the interaction category (fighting/dancing). Categorization of people and their spatial relation yielded comparable performance at every stimulus duration, which was consistently more accurate than performance in interaction categorization. This implies that it takes no longer to access the objects’ category (bodies vs. chairs) than to access the relation between objects (facing vs. non-facing); whereas substantially longer processing time is required in order to categorize social interactions, with comparable accuracy. The parallel between categorization of objects and spatial relations suggests that relations between objects are extracted as rapidly as categorical object information, providing the building blocks for social interaction recognition. These findings impose important constraints on future models of social interaction recognition.