Authorial attribution and computational stylistics: if you can tell authors apart, have you learned anything about them?

Abstract

Within stylometrics, the disciplines of authorial attribution and descriptive stylistics hitherto have been pursued separately. The first has achieved mainstream status within literary studies, while the second is little recognized. A study of the plays of Thomas Middleton compared with a large control sample shows that it is possible to achieve a good classification of his work in 2,000 word segments as against those of his contemporaries, using frequencies of very common words. Such a result can serve as the basis both for testing Middleton's hand in some disputed plays and for a description of his style. Most of «The Revenger's Tragedy», part of «The Yorkshire Tragedy», and all of «The Second Maiden's Tragedy» prove to be in a style similar to that of Middleton's uncontested plays. This style, judging by an examination of instances in context of the ten word types most strongly correlated with the Middleton-other discriminant function, is (among other things) rich in deictics and poor in conjunctions, features readily accommodated to previous descriptions including Middleton's own. It is concluded that classification and description can be mutually supportive: the first confirms the validity of the second, while the second helps to establish the stylistic mechanism underlying a successful classification.