Attribution of Contested and Anonymous Ancient Greek Works

Authorship attribution has been a persistent problem in the Classical genre, as texts that reach us from antiquity are often corrupted, edited, or forged over the thousands of years since their initial production. Scholars have worked on identifying writers’ stylistic differences in an attempt to distinguish genuine texts from fakes, and to attribute an author to previously anonymous works. Increasing computing power allows the derivation of more complex features, giving us new information about each author’s linguistic signature and writing style. Our system is able to accurately predict the author of a complete anonymous work, as well as many text fragments that currently have contested authorship. We experimented with using semantic and lexical features, and explored both discriminative and generative classification algorithms. Our highest-performing system achieved an attribution accuracy of 85.7%.