Narrative planning techniques have shown great potential in their ability to automatically generate plots that meet human authorial goals for their fictional world and characters. However, stories produced by existing techniques are limited in expressivity related to belief-based failed actions in stories, which is pervasive in various media ranging from Star Wars to Looney Tunes.

This dissertation outlines a principled means to generate storylines with failed actions and advance a broader goal of automatically creating more expressive, natural, and compelling narratives. It builds upon existing work in narrative generation using artificial intelligence planning algorithms. While traditional planning techniques rely on producing sound, causally coherent plans that avoid or recover from failure, our approach explicitly plans for action failure in the generated plans.

The approach described by this dissertation involves (1) defining a knowledge representation and an associated plan generation algorithm for intentionally generating stories where characters attempt an action and fail and (2) extending current methods for the specification of narrative planning problems included in this expanded representation involving desired actions, action failures, and the intent dynamics that surround them. We evaluated our approach to gauge human comprehension of the computationally generated plots and the capability of novice storywriters to use the enhanced representation to guide the planner towards desired stories consisting of failed actions.