Testing Burrows' Delta on Ancient Greek Authors

Abstract

This paper tests the effectiveness of Burrow’s Delta Method on a corpus of selected prose writings in ancient Greek. When tested on a corpus of fourteen and eight authors, the method yields good results with relatively small samples (1000, 3000, and 5000 words) and different word frequency vectors (100, 200, 500 words), but its performance is worse with texts of similar genres (oratory, historical or medical writings). We conclude that it is the generic proximity that influences the results of classification most. However, in cases where confusion is more likely, such as the writings of Demosthenes and Aeschines, the method proves effective for shortlisting potential authors. Shortlists can give an adequate idea of a sample’s nearest neighbors while leaving some freedom for the researcher in interpreting the results.