Exploring aspects of similarity between spoken personal narratives by disentangling them into narrative clause types

Sharing personal narratives is a fundamental aspect of human social behavior\nas it helps share our life experiences. We can tell stories and rely on our\nbackground to understand their context, similarities, and differences. A\nsubstantial effort has been made towards developing storytelling machines or\ninferring characters' features. However, we don't usually find models that\ncompare narratives. This task is remarkably challenging for machines since\nthey, as sometimes we do, lack an understanding of what similarity means. To\naddress this challenge, we first introduce a corpus of real-world spoken\npersonal narratives comprising 10,296 narrative clauses from 594 video\ntranscripts. Second, we ask non-narrative experts to annotate those clauses\nunder Labov's sociolinguistic model of personal narratives (i.e., action,\norientation, and evaluation clause types) and train a classifier that reaches\n84.7% F-score for the highest-agreed clauses. Finally, we match stories and\nexplore whether people implicitly rely on Labov's framework to compare\nnarratives. We show that actions followed by the narrator's evaluation of these\nare the aspects non-experts consider the most. Our approach is intended to help\ninform machine learning methods aimed at studying or representing personal\nnarratives.\n