Clustering of idiostyles using the Burrows’ Delta and the diagnostic potential of high-rank word forms

This article provides a linguistic interpretation of the most widely used and authoritative stylistic algorithm, Burrows’ Delta, using Russian language material drawn from novels from the era of Alexander II. The algorithm is based on intertextual distance analysis; the Delta measure is calculated as the absolute sum of the differences between the z-scores of two compared texts divided by the number of words. A technical feature of the algorithm is that its effectiveness depends largely on numerous small linguistic features. The paper poses three questions: 1) What enables Delta to accurately cluster texts from a single author? 2) What does Delta rely on for clustering different authors? 3) What, if anything, motivates individual authors’ preferences for the frequency of high-ranking word forms, and particularly function words and pronouns? The paper analyzes both simple relative frequencies of the most frequent function words and z-scores. We propose the following answers: 1) Delta clusters texts by the same author not only because they share similar frequency and z-score profiles, but also because these profiles differ significantly and consistently from those of other authors, making variation in the texts of the same author less significant; 2) variation also distinguishes authorial styles: an analysis of the 50 highest-ranking word forms shows that they provide a rich resource for identifying authorial style and clustering different authorial styles; and 3) the authors’ preferences regarding the use of high-ranking word forms can be motivated by three factors: the author’s lexicon as an element of idiostyle (reflected primarily in the frequency of prepositions), the structure of texts, their genre-stylistic characteristics (pronouns), and the interaction of these factors (with conjunctions contributing to the syntactic perspective of the text and particles to its modal perspective).