How to (Machine) Read a Novel: Modeling Events and Information in Literary Fiction
Novels pose unique challenges that tend to be overlooked by current natural language processing systems. The number of different characters and sheer length of the text limit the effectiveness of state of the art coreference approaches. The extensive use of figurative language complicates how we define and identify events. The patterns and structures through which information is propagated between characters is highly distinct from how information is diffused across social media, the primary focus for this kind of research. In this webinar, I’ll discuss some of the difficulties that arise when building models to analyze fiction as well as how these challenges in turn make the literary domain potentially relevant to broader questions in the field of NLP. In particular, I’ll present two recent research projects, one on identifying events in fiction and the other on literary information propagation.
Matthew Sims is a postdoctoral scholar and lecturer in the School of Information’s MIDS program. His research is focused on developing new natural language processing approaches for understanding narrative. He received an M.Phil. from Cambridge University and a Ph.D. from the University of Chicago.