Mark Finlayson -Eminent Scholar Chaired Associate Professor of Computer Science in the Knight Foundation Schoo

Date: January 11, 2024
Time: 12:00 PM - 1:00 PM
Location: 1889 Museum Road, Gainesville, Fl, 32611
Host: Bonnie Dorr
Admission: n/a

Narrative Natural Language Processing: Recent Advances and Future Prospects
Narratives are ubiquitous: they are found in every society and culture and used by nearly every person every single day. Narratives fundamentally shape our world and our perceptions of it, play a central role in the transmission of culture and the expression of implicit knowledge, and are one of the main ways we persuade others for good or for ill. Despite these observations, there remain many NLP tasks specific to narrative that have received relatively little attention, as well as a variety of unanswered questions regarding how narrative works, cognitively speaking. I present steps taken over the past ten years by researchers in my Cognition, Narrative, and Culture (Cognac) Laboratory to drive forward our understanding of narrative, both from a computational and cognitive point of view. First, I describe work on narrative event detection seeking to reveal event hierarchies, foreground/background status, and “key plot points”. Second, work on events leads naturally to a new, exact, and provably complete solution to timeline extraction, which is supporting new work on duration estimation and narrative level extraction. Third, I review our work on animacy and character analysis, which is leading to very interesting tests, at scale, of long-held linguistic assumptions. Fourth, I discuss several in-progress efforts, including detection of narrative motifs and the categorization and resolution of narrative reference. Finally, I show how all these advances support a vision of extracting higher-level structure from narratives, including common plot pieces, morals, and themes. Along the way, I point out various applications (to, e.g., argument analysis and mis-/dis-information detection) as well as next steps that I plan to explore in the context of currently popular Large Language Models.