Personality and Emotion in Strong-Story Narrative Planning

Believable characters are core elements of a coherent story. Qualities that make story characters more believable include goals, beliefs, personality, and emotion. We propose computational models of emotion and personality by adapting the Ortony, Clore, and Collins (OCC) model of emotion and the Five-Factor personality model. Our models are formulated into multiagent strong-story narrative planning with the promise of being highly reusable and domain independent. We evaluate these models using multiple human subject studies. We show that our model's reasoning about character emotions matches the expectations of human readers, and using our emotion model, we can generate a larger set of stories than precedent narrative planners. We also demonstrate that human readers can perceive and recognize the personalities of story characters through their consistent behavior generated by our model. Our final experiment supports that human readers significantly find the behavior generated by our models of emotion and personality more believable than behavior that lacks either or both.

I. Introduction

Interactive narratives allow users to be part of a fictional world and influence the storyline through their actions and interactions with nonplayer characters (NPCs) [1]. The applications of interactive narratives can be found in education [2], [3], [4], training [5], [6], [7], therapy [8], [9], and entertainment [10], [11], [12].