Categorization of narrative semantics for use in generative multidocument summarization

Abstract

The generative summarization of textual stories has been one of the goals of computational narratology since attempts at full semantic NLU in the '70s. Our NLP group has recently created several systems for multidocument news summarization, but using purely statistical methods. Between these poles, there may be an unexplored avenue where knowledge of story structure can give partial, yet useful semantic understanding to a news reader. Such knowledge can then lead to summaries more informed than those based on solely statistical means. This student paper represents work in progress on a two-module system: The first module categorizes news articles into their underlying dramatic structures; the second will attempt to use this understanding to create and execute a generative plan, concisely retelling the story to form a surface-level summary. © Springer-Verlag 2004.