The Rolling Stylometry method is designed to detect stylistic takeovers. To give an example: imagine a text that is supposedly written by more than author, and you want to pinpoint the stylistic change between the individual authorial “signals”. In the case of the earliest Polish translation of the Bible knowns as Queen Sophia’s Bible, one might be interested in tracing the five scribal hands evidenced in the manuscript – were they “just” the scribes, or were they also involved in actual translatorial? In Fig. 1 above, one can notice that indeed, the change of a scribe (vertical dashed lines) is usually correlated by stylistic change (different colors of the strip).
The general idea is fairly simple: a text to be analyzed (e.g., an anonymous text to be attributed) is chunked into equal-sized blocks (partially overlapping). Then, instead of attributing the text in its entirety, the goal is to perform an independent similarity test for each chunk, and to inspect the results as a sequence of ordered stylistic signals. Arguably, any classification method can be combined with this procedure. However, so far, the method implementation in stylo includes support vector machines (SVM), nearest shrunken centroids (NSC), and Delta in its classical Burrowsian flavor.
An overview of the project is provided here, a more detailed description, supplemented by applicable/reusable snippets of code, can be found in this blog post. For a comprehensive explanation of the method’s theoretical assumptions, features, and implementation refer to the following paper: