Toward Automated Qualitative Data Analysis: The Case of Narrative NLP

Specific discourse types present their own special challenges across the spectrum of NLP techniques.

The narrative discourse form presents numerous interesting situations that both challenge the capabilities of existing techniques, and also suggest novel, NLP tasks that are specifically relevant to narrative. Importantly, narrative is especially useful and prevalent when it comes to qualitative data collected in the course of education, learning sciences, and social science research. I present recent progress in the FIU Cognition, Narrative, and Culture (Cognac) Laboratory on NLP as applied to narrative. First, story detection, a variation of the text classification task where the goal is to identify whether a text contains a narrative. Second, animacy and character detection, where the goal is to determine whether a referent is animate and is acting as a “character”. We see that this approach requires some narratological sophistication to be successful. Third, new improvements in sub-event and event relationship detection on narrative texts that take advantage of certain important features of narrative discourse. And, fourth, new approaches to timeline extraction that significantly improve our ability to extract, organize, and characterize timelines of events. This collection of results represents concrete steps toward our ability to extract meaning from unstructured narrative text and points the way forward to NLP approaches that hopefully will realize the dream of bringing the full potential of automated language understanding into the hands of qualitative researchers.

Author: Mark Finlayson