Proceedings of the Annual Meeting of the Cognitive Science Society

We examined the visual perception of joint actions, in whichtwo individuals coordinate their body movements in space andtime to achieve a joint goal. Animations of interacting actionpairs (partners in human interactions) and non-interacting ac-tion pairs (individual actors sampled from different interactionsequences) were shown in the experiment. Participants wereasked to rate how likely the two actors were interacting. Therating data were then analyzed using multidimensional scalingto recover a two-dimensional psychological space for repre-senting joint actions. A descriptive model based on ordinallogit regression with a sparseness constraint was developed toaccount for human judgments by identifying critical featuresthat signal joint actions. We found that identification of jointactions could be accomplished by assessing inter-actor correla-tions between motion features derived from body movementsof individual actions. These critical features may enable rapiddetection of meaningful inter-personal interactions in complexscenes.