Computational Support for Trope Analysis of Textual Narratives

Narrative tropes are repeated patterns of recognizable communicative elements across stories. Tropes are a set of patterns that aid readers in story comprehension. They are also a reflection of socio-cultural norms that are formally or informally present in the particular context for authors and readers. Trope-based analyses are common in media studies but are limited to close readings and individual analyst perspectives. Distant reading of tropes is challenging due to the lack of precise definitions and the variety of forms in which tropes manifest in language, and over space and time within story worlds. This paper presents a trope labeled dataset of scripts, an initial analysis framework, and a system for computational support of trope analysis. For highlighting the challenges of developing computational models, we present a trope prediction algorithm on a movie script dataset based on a model trained with human-annotated tropes from TVTropes.

Duke Scholars

Author Arnav Jhala Engineering Graduate and Professional Programs