[1] The attribution of anonymous or disputed literary works has traditionally been the province of connoisseurship: the trained reader, steeped in an author's habits of thought and phrasing, pronounces a text authentic or spurious on the strength of a cultivated intuition. [2] Over the past several decades, stylometry—the statistical study of measurable features of style—has advanced a rival claim to authority, one that its proponents present as freeing attribution from the subjectivity of individual taste. [3] The method proceeds not from an author's memorable turns of phrase but from the frequencies of unremarkable function words—articles, prepositions, conjunctions—that writers deploy unconsciously and therefore, it is argued, cannot readily counterfeit or suppress. [4] By this reasoning, the fingerprint of authorship lies precisely in what a writer does not think about.
[5] The quantitative approach has resolved several long-standing disputes to the satisfaction of most scholars, yet its foundations are less secure than its air of objectivity implies. [6] The choice of which features to count, the selection of comparison texts, and the threshold at which a statistical difference is deemed significant all require judgments that the method cannot itself supply. [7] A writer's function-word frequencies, moreover, are not perfectly stable: they shift with genre, with the conventions of collaboration, and across a career. [8] Stylometrists have grown alert to these confounds, but the corrections they introduce reintroduce the very interpretive discretion the method was meant to eliminate. [9] The discipline has thus not so much replaced the connoisseur's judgment as relocated it, from the assessment of a text's texture to the design of the procedure that assays it.
