Ryan Dubnicek's Preliminary Exam
PhD candidate Ryan Dubnicek will present his dissertation proposal, "Toward Useful Genre: Computational Modeling and Literary Theory." His preliminary examination committee includes Professor J. Stephen Downie, Glen Layne-Worthey, Associate Director, HathiTrust Research Center, Assistant Professor Zoe LeBlanc, Professor Ted Underwood (Chair), and Associate Professor David Bamman (University of California, Berkeley).
Abstract
Genre has long served as a fundamental tool for organizing and understanding cultural expression, yet scholars have struggled to arrive at a single, stable definition. This dissertation proposes that resolving this debate may be less productive than reframing it: rather than asking which theory of genre is most valid, we might instead ask which theories are most useful — that is, grounded in literary scholarship and implementable within computational pipelines designed to meet the practical needs of digital collections and their users. I argue that computational modeling offers a unique opportunity to evaluate genre theories not by their philosophical coherence alone, but by how well their principles translate into concrete classification tasks at scale.
In service of this argument, I propose three experiments, each centering a different theory of genre and modeling approach. Chapter 1 applies supervised machine learning to fiction identification in large digital libraries, evaluating genre-as-social-action plus family resemblance as guiding theoretical frameworks. Chapter 2 turns to film, using unsupervised clustering across text and image data to assess the strength of formalist genre theory in comparison to socially-derived labels from scholars and viewers. Chapter 3 engages large language models and controlled vocabularies to evaluate automated genre assignment to library catalog records, approaching genre as an exemplar of quality cataloging practice. Across all three experiments, I attend not only to the accuracy of model outputs but to their interpretability–how well models surface and explain the features that drive their decisions, and what those decisions reveal about the theories that motivated them. Taken together, these experiments advance a bidirectional understanding of genre theory and computational practice, in which implementation choices illuminate theory, and theoretical commitments shape what we build and how we evaluate it.
Questions? Contact Ryan Dubnicek.