Function Word Adjacency Networks and Early Modern Plays

Tools

Barber, Ros. 2020. Function Word Adjacency Networks and Early Modern Plays. ANQ: A Quarterly Journal of Short Articles, Notes and Reviews, 33(2-3), pp. 204-213. ISSN 0895-769X [Article]

Preview

Text
2019-09-15_essay_function-wans-FINAL.pdf - Accepted Version
Available under License Creative Commons Attribution Non-commercial.
Download (259kB) | Preview

Official URL: https://www.tandfonline.com/doi/abs/10.1080/089576...

Abstract or Description

The Word Adjacency Network method underpinning the New Oxford Shakespeare’s attribution of the Henry VI plays to Christopher Marlowe as co-author has not been independently tested and is only now being subjected to critiques. The response of Segarra et al. (2019) to criticism by Pervez Rizvi (2018) barely alleviates concerns. This article demonstrates that sections of the plays designated as Shakespeare’s were not detected as Shakespeare’s by the method according to the authors’ own definitions, since his “relative entropy” score was often above zero, which according to Segarra et al. (2016) means the play is no more like Shakespeare’s style than it is like the combined style of all six playwrights tested. The disproportionate representation of Shakespeare in the underlying dataset, combined with a mathematical procedure intended to remove “background noise” may explain Shakespeare’s hovering around the zero line. A claimed concordance with the results of other stylometric tests giving parts of 1 Henry VI to Marlowe is demonstrably not present. The high success rates claimed for the method in Eisen at al. (2018) are based on a flawed validation process known as overfitting, an interpretive method altered to improve success percentages, and the effects of disparate canon sizes for which the equations fail to adequately compensate. It is argued that in the light of potential flaws in the method, and the authors’ misrepresentation of their results, the conclusions of both Segarra et al.’s 2016 article and Eisen et al.’s 2018 study should be set aside.

Item Type:

Article

Identification Number (DOI):

https://doi.org/10.1080/0895769X.2019.1655631