Investigating the conditions under which movie viewers' event models influence their attention

Abstract

Film, television, and video play a major role in daily information consumption, making it important to understand how we comprehend dynamic scenes and how this shapes visual attention. Although event models—mental representations of ongoing events—are known to guide attention during reading and static scene viewing, their influence on attentional selection in films is minimal or absent, a pattern known as the Tyranny of Film. This project provides an existence proof that event models can influence selective visual attention in commercial film clips. Based on the Scene Perception & Event Comprehension Theory and Biased Competition Theory, I propose two conditions under which event models should guide attention: 1) The target must be prioritized differently because of clear differences in viewers’ event models (Differential Event Model Criterion); and 2) The target must have low bottom-up default gaze probability, making it unlikely to attract attention automatically (Low Default Gaze Probability Criterion). I validated the first criterion using linguistic and moment-to-moment spatial importance measures, and the second using DeepGaze IIE predictions, identifying eight film clips that met both criteria. In a within-subjects Eye-Tracking study (N = 44), participants viewed each clip without context and then with context. Targets were fixated four times more often in the Context condition, whereas fixation rates in the No-context condition were at chance. Moreover, a priori event-based importance predicted attention more strongly than a priori default-gaze predictions when participants were aware of the context. These findings demonstrate that when both criteria are satisfied, event models can influence attentional selection during film viewing, overcoming the Tyranny of Film and offering strong support for SPECT.