Features influencing judgment on the believability of the narrative of events

Jae-Myoung Yu and Cheongtag Kim

Abstract

An event is a sequence of actions. People use scripts, typical sequences of events, when they judge the believability of the narrative of an event. This mental process is repeatedly confirmed by behavioral experiments, but is not yet computationally modeled. To build a model of human judgment on the believability of events based on scripts, we should know which features of an event affect believability judgment and how they can be quantified. For the purpose, we first gathered scripts of mundane events, generated a variety of event scenarios based on those data, and computed a number of features that are possibly related to the believability of each scenario. Participants were asked to rate the believability of each event scenario in 5-point Likert scale. Linear regression analysis revealed two major features influencing believability judgment, the consistency between actions and smooth unfolding of an event. Consistency is measure by the probability at which two action occur in observed order, and smooth unfolding indicates the first-order Markovian transient probability of an action.