Learning Narrative Structure from Annotated Stories
By Mark Finlayson
Narrative structure is an ubiquitous and intriguing phenomenon. By virtue of this structure humans are able to perform high-level processing of stories necessary for deep understanding; for example, we recognize the presence of 'villainy' or 'revenge', even if those words are not actually present in the text. Narrative structure is an anvil for forging new artificial intelligence and machine learning techniques, and is a window into abstraction and conceptual learning as well as into culture and its influence on cognition. I will discuss various components of my approach to learning narrative structure automatically.
First, I will present Analogical Story Merging (ASM), a new machine learning algorithm for extracting plot patterns from sets of stories. I demonstrate, for the first time, automatically learning a theory of narrative structure from text: ASM can learn a substantive portion of Vladimir Propp's influential theory of the structure of folktale plots. Second, I will discuss data and tools that support this extraction, including: the several deeply-semantically-annotated story corpora I have created; the Story Workbench, a general-purpose text annotation tool I have designed for collecting the corpus data; and the new linguistic representation schemes I have developed to capture important aspects of story semantics. I will highlight a new result, namely, the first-ever demonstration of the reliable double-blind coding of higher-level narrative by trained annotators. Third, I will discuss future applications for my techniques and tools, including in cultural studies, persuasive communication, understanding religious extremism, and understanding the structure of business and law cases, as well as an upcoming project funded by the NIH on investigating narratives that affect patient health care behavior.
Bio: Dr. Mark Finlayson is a Research Scientist at the Computer Science and Artificial Intelligence Laboratory at MIT. His research focuses on representing, extracting, and using higher-order semantic patterns in natural language, especially with regard to narrative. He received the B.S.E from the University of Michigan in 1998, and the M.S. and Ph.D. from MIT in 2001 and 2012, respectively, all in Electrical Engineering and Computer Science. He is general chair of the Computational Models of Narrative (CMN) Workshop series, now in its fifth iteration, and is lead guest editor of a special issue on Computational Models of Narrative, to be published by the Journal of Literary & Linguistic Computing in 2014.