Plans and Planning in Narrative Generation: A Review of Plan-Based Approaches to the Generation of Story, Dis- course and Interactivity in Narratives

The last ten years have seen a significant increase in computationally relevant research seeking to build models of narrative and its use. These efforts have focused in and/or drawn from a range of disciplines, including narrative theory Many of these research efforts have been informed by a focus on the development of an explicit model of narrative and its function. Computational approaches from artificial intelligence (AI) are particularly well-suited to such modeling tasks, as they typically involve precise definitions of aspects of some domain of discourse and well-defined algorithms for reasoning over those definitions. In the case of narrative modeling, there is a natural fit with AI techniques. AI approaches often concern themselves with representing and reasoning about some real world domain of discourse – a microworld where inferences must be made in order to draw conclusions about some higher order property of the world or to explain, predict, control or communicate about the microworld's dynamic state. In this regard, the fictional worlds created by storytellers and the ways that we communicate about them suggest promising and immediate analogs for application of existing AI methods. One of the most immediate analogs between AI research and narrative models lies in the area of reasoning about actions and plans. The goals and plans that characters form and act upon within a story are the primary elements of the story's plot. At first glance, story plans have many of the same features as knowledge representations developed by AI researchers to characterize the plans formed by industrial robots operating to assemble automobile parts on a factory floor or by autonomous vehicles traversing unknown physical landscapes. As we will discuss below, planning representations have offered significant promise in modeling plot structure. Equally as significantly, however, is their ability to be used by intelligent algorithms in the automatic creation of plot lines. Just as AI planning systems can produce new plans to achieve an agent's goals in the face of a unanticipated execution context, so too may planning systems work to produce the plans of a collection of characters as they scheme to obtain, thwart, overcome or succeed.