A Computational Model of Narrative Generation for Surprise Arousal

Abstract

This dissertation describes work to develop a planning-based computational model of narrative generation designed to elicit surprise in the mind of a reader. To this end, my approach makes use of two narrative devices – flashback and foreshadowing. While surprise plays an important role for attention focusing, learning, and creativity, little effort has been made to build a computational framework for surprise arousal in narrative. In my computational model, flashback provides a backstory to explain what causes a surprising outcome, while foreshadowing gives hints about the surprise before it occurs. In this work I focus on the arousal of surprise emotion as a cognitive response which is based on a reader's cognitive appraisal of a given situation. In this dissertation I present Prevoyant, a planning-based computational model of surprise arousal in narrative generation, and analyze the effectiveness of Prevoyant. To build a computational model of the unexpectedness in surprise, I adopt a cognitive model of surprise based on expectation failure.

There are two contributions made by this dissertation. First, I present a computational framework for narrative generation designed to elicit surprise. The approach makes use of a two-tier model of narrative and draws on Structural Affect Theory, which claims that a reader’s emotions such as surprise or suspense are closely related to narrative structure. Second, I present a methodology to evaluate surprise in narrative generation using a planning-based approach based on the cognitive model of surprise causes. The results of the experiments that I conducted show strong support that my system effectively generates a discourse structure for surprise arousal in narrative.