The interpretation of Zeta test results

摘要

Zeta has been described as the most powerful general-purpose authorship tool currently available. It is therefore of the utmost importance that Zeta test results be correctly interpreted, because incorrect interpretations can lead to incorrect authorship attributions. This article argues that the current method of interpreting Zeta results, pioneered by Craig and Kinney (2009) in Shakespeare, Computers, and the Mystery of Authorship and used in the Authorship Companion to The New Oxford Shakespeare, is unsound. The article provides theoretical arguments and a counterexample to demonstrate this. Moreover, the article argues that the validation of the Zeta method that has so far been performed is less strong than it appears, being based on a misunderstanding of what the results are telling us. It suggests that conclusions apparently drawn from Zeta, that shorter n-grams and function word skip bigrams make better authorial markers, are unwarranted. Finally, the article presents the results of new Zeta tests, which show that the method is less reliable than has been supposed.