Stylometry

Stylometry is the quantitative analysis of writing style used to make inferences about authorship, provenance, or stylistic development. It treats text as a data object and seeks measurable, repeatable features that can differentiate authors or track changes over time. Commonly analyzed features include word frequencies, function word usage, character and word n-grams, punctuation patterns, syntax, and lexical richness.

Methods typically involve feature extraction followed by statistical or machine learning modeling. Supervised approaches train classifiers on texts with known authorship, while unsupervised methods cluster texts by similarity. Popular techniques include naive Bayes, support vector machines, logistic regression, and more recently neural network models. Robust stylometry often uses a combination of lexical, syntactic, and stylistic features and applies cross-validation to assess generalization.

Applications span authorship attribution of disputed texts (forensic linguistics), plagiarism detection, literary analysis, and historical document authentication. Stylometry has famously been used to examine the Federalist Papers, dating anonymous works, and revealing trends in an author's development. It also informs digital forensics, where attribution of online content and bot detection may rely on stylistic cues.

Limitations include the influence of topic, genre, translation, and period on style, as well as data scarcity for some authors. Short texts, dialect variation, and deliberate obfuscation reduce accuracy. The approach assumes that style is stable and distinctive, an assumption that may fail for evolving or collaborative authorship. Evaluation relies on careful splits and cross-domain testing to avoid overfitting. Ethical considerations include privacy and misattribution risks, and bias introduced by training data. Transparent reporting of methods and datasets is essential for reproducibility.