Evaluation of a computational model of surprise arousal in narratives

In this paper we analyze the effectiveness of Prevoyant, a planning-based computational model of surprise arousal suggested in our previous paper. This evaluation was performed by measuring the effect of stories produced by this method on several factors contributing to surprise in narratives (e.g., expectation failures, the importance of story events, the emotional valence of a reader, and resolution) and their relation to story interest. To measure the ratings of a reader´s surprise and story interest, we conducted a pilot study, where three different discourse types - chronological, flashback, and flashback with foreshadowing - were compared. A one-way ANOVA and a priori pair-wise comparisons confirmed that the mean ratings of surprise and interestingness increased at the .01 and .05 level, respectively. While the number of subjects in this study was relatively small, the significance of our results argues for support to the claim that the computational model we evaluated can effectively manipulate readers´ sense of surprise in their experience of a narrative.