StorySpace : spatial semantic comparison of narrative

Abstract

It is well known that humans are far more adept than computers at identifying similarities between stories. Humans are able to communicate values and event patterns back and forth through these narratives. Parents communicate through the telling of "The Tortoise and the Hare" that hard work and determination can often trump talent, and that hubris can lead to one's downfall. It would be quite useful to develop a computational technique to apply this type of analysis to a story to relate to more generic cases. In this paper, I demonstrate the beginnings of a technique called Spatial Semantic Analysis of Narrative that identifies a "trajectory" for each story that enables comparison between them. These trajectories take into account the temporal progression of a story, which aims to provide a dimension of information beyond traditional "bag of words" comparisons. I present promising results when this technique is applied to a corpus of "how-to" articles scraped from the Internet as well as a corpus of Islamic texts annotated using Mark Finlayson's Story Workbench application. I also present next steps for improving the algorithm and allowing it to operate on standard untagged datasets.