The relationship between overt attention and event perception during dynamic social scenes

When watching people perform everyday activities such as doing laundry viewers spontaneously segment the activities into a sequence of events. The timing of these event boundaries are surprisingly consistent across viewers. Neuroimaging evidence suggests that this consistency across viewers may be due to the use of changes in motion to identify points at which prediction error about the future form of an event rises above threshold and signals the beginning of a new event (Zacks et al, Nature Neuroscience; 2001). Brain regions thought to be associated with attentional control are also selectively active during event boundaries. Do changes in motion at event boundaries capture overt attention and can the gaze behaviour across viewers be used to predict event segmentation? In this study, participants viewed videos depicting an actor performing an everyday task (laundry, washing a car, or building a tent) while their eye movements were recorded. Participants initially free-viewed each video then subsequently watched the videos again while either parsing them into the smallest (fine) or largest (coarse) units that seem natural and meaningful. Analysis of gaze behaviour during free-viewing indicated significantly greater clustering of gaze across participants at fine event boundaries but less clustering at coarse boundaries. As previously reported, fine boundaries appear to be associated with changes in motion of the actor's hands and our results confirm that this relationship can also be observed as an increase in gaze to the hands and greater gaze clustering across viewers. These results both confirm the spontaneous segmentation of human events during free-viewing, and suggest the role overt attention and expectation play in parsing the flow of information in naturalistic dynamic social scenes. Meeting abstract presented at VSS 2012