Statistical Stylometrics and the Marlowe-Shakespeare Authorship Debate

I. Authorship Attribution Paradigm Historians, literary scholars, psychologists, and – more recently – computational linguists have long sought a reliable methodology for analyzing texts to determine the identity of their author. Since at least the late nineteenth century (see Mendenhall, 1887; Mascol, 1888a/b), one tool used in the investigation of authorship has been the extraction of statistical tendencies from the documents and comparison of these data in order to group the documents appropriately. Underlying the search for such a methodology is the critical assumption that some statistically quantifiable characteristic or set of characteristics inherent in a single author’s use of written language could be isolated that would be consistent across works by that author, but differ between different authors. Thus, the feature(s) could be used as a sort of fingerprint to distinguish between works by distinct authors and to identify the likely author of an anonymously published work. This endeavor is generally referred to as stylometry. With the advent of natural language processing and machine learning techniques in recent decades, the field has advanced greatly, and numerous researchers have addressed many specific