Revising Ekphrasis: Using Topic Modeling to Tell the Sister Arts’ Story

Abstract

For the past 20 years, the story of ekphrasis—poetry to, for, and about the visual arts—has been told as a long-standing, gendered contest between rival media, fraught with political, cultural, and religious anxieties. Although skeptical of the necessity of gendered rivalry as a principle of ekphrastic creation, literary scholars have struggled to present a compelling alternative model that sufficiently accounts for the genre’s representational complexity.

This talk begins by asking if computational methods might offer new insights into the canon and tradition of ekphrasic poetry and suggests how topic modeling—one form of computational text analysis—might begin to refocus the aperture of our critical lens on the genre’s conventions.

Oriented toward the non-expert, this presentation will assume no prior knowledge of topic modeling or social network analysis. I will provide a gentle introduction that builds toward an understanding of the potential uses for topic modeling and network analysis as a means for exploring large collections of poetic texts.

Poetic collections, dense and rich with figurative language, require revising how we as humanists interpret topic modeling results. Therefore, this presentation will also address how changes in interpretation affect the questions we might ask and the assumptions we can make about “topics” generated by latent Dirichlet allocation (LDA)—one type of topic modeling algorithm.