Document Type
Poster
Publication Date
Fall 2020
Abstract
Writing styles are often viewed as unique to their writers–a compositional fingerprint of sorts. An analytical tool based upon this assumption is stylometry: the statistical analysis of the variations in the literary styles of works, often used to determine the most likely author of a particular work. Stylometric techniques abound in a multitude of fields, including history, literary studies, and even courts of law. Stylometry is often used as a form of evidence as to the identities of authors of written material pertaining to legal cases, a famous example being the conviction of the Unabomber based upon stylistic similarities between his earlier essays and his famous manuscript [1]. Thus, stylometric techniques are ascribed a lot of power. But, what if stylometry isn’t as dependable as it is assumed to be? What if a writer’s so-called “unique” style can be easily imitated to fool stylometric tools? In this project, we aim to analyze the ability of AI to generate text stylometrically consistent with the writer upon whom it was trained.
Recommended Citation
Lawson, Rebecca, "GPT-2: Girl Detective Analyzing AI-Generated Nancy Drew with Stylometry" (2020). IPHS 200: Programming Humanity. Paper 32.
https://digital.kenyon.edu/dh_iphs_prog/32
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.