Identifying Narrative Content in Podcast Transcripts

Yosra Abdessamed, Shadi Rezapour, Steven R. Wilson

Main: Computational Social Science and Cultural Analytics Oral Paper

Session 10: Computational Social Science and Cultural Analytics (Oral)
Conference Room: Marie Louise 2
Conference Time: March 20, 11:00-12:30 (CET) (Europe/Malta)
TLDR:
You can open the #paper-404-Oral channel in a separate window.
Abstract: As one of the oldest forms of human communication, narratives appear across a variety of genres and media. Computational methods have been applied to study narrativity in novels, social media, and patient records, leading to new approaches and insights. However, other types of media are growing in popularity, like podcasts. Podcasts contain a multitude of spoken narratives that can provide a meaningful glimpse into how people share stories with one another. In this paper, we outline and apply methods to process English-language podcast transcripts and extract narrative content from conversations within each episode. We provide an initial analysis of the types of narrative content that exists within a wide range of podcasts, and compare our results to other established narrative analysis tools. Our annotations for narrativity and pretrained models can help to enable future research into narrativity within a large corpus of approximately 100,000 podcast episodes.