Four Fundamental Dimensions Underlying the Perception of Human Actions

Abstract

We evaluate the actions of other individuals based upon a variety of movements that can reveal critical information to guide decision making and behavioural responses. These signals can convey a range of information about the actor, including their goals, intentions and internal mental states. Although progress has been made to identify cortical regions involved in action processing, the organising principles underlying our representation of actions still remains unclear. In this paper we aimed to investigate the conceptual space that underlies action perception by assessing which characteristics are fundamental to the perception of actions executed by other human individuals. We recorded 240 different actions using motion-capture and used this data to animate a volumetric avatar that performed the different actions. Two-hundred and thirty participants then viewed these actions and rated the extent to which each action demonstrated 23 different action characteristics (e.g., avoiding-approaching, pulling-pushing, weakpowerful, etc.). We analysed this data using Exploratory Factor Analysis to examine the latent factors underlying visual action perception. The best fitting model was a four-dimensional model with oblique rotation. We named the factors: unfriendly-friendly, feeble-formidable, unplanned-planned, and adduction-abduction. The first two factors of friendliness and formidableness explained approximately 22% of the variance each compared to planned and abduction that explained approximately 7-8% of the variance each, as such we interpret this representation of action space as having 2+2 dimensions. A closer examination of the first two factors suggests a potential overlap with the way we evaluate facial traits and emotions, whilst the last two factors of planning and abduction appear unique to actions.

Gabriel Mackie

Research Assistant

My research interests include distributed robotics, mobile computing and programmable matter.