Rank‑Turbulence Delta and interpretable approaches to stylometric Delta measures

Abstract This article repositions Burrows’s Delta as a flexible family of distance measures for exploratory and unsupervised stylometry, where interpretability and stability are as important as predictive accuracy. We introduce two probabilistic extensions, Rank-Turbulence Delta and Jensen–Shannon Delta, by reinterpreting uncentred standardized word-frequency vectors as non-negative representations that can be normalized into probability distributions and compared using information and rank-based divergences. Building on this representation, we derive token-level decompositions for classical Delta, Cosine Delta, Jensen–Shannon Delta, and Rank-Turbulence Delta, enabling each document distance to be expressed as a sum of interpretable lexical contributions. We evaluate the proposed framework in clustering and nearest-neighbour attribution experiments on four literary corpora in English, German, French, and Russian, including an extended Russian benchmark drawn from SOCIOLIT (639 works by 89 authors, XVIII–XXI centuries). To substantiate interpretability, we additionally verify that top-contributing tokens are stable under small perturbations (mfw variation and bootstrap resampling) and that removing these tokens reduces inter-author separation in the expected direction. Overall, the results show that probabilistic and rank-based geometries extend Delta without abandoning its core intuition, while providing a reproducible bridge between aggregate distances and philologically meaningful lexical signals.