A morphosyntactic authorship attribution study of the speeches of Demosthenes and Apollodorus
Vanessa B. Gorman, Robert J. Gorman
Abstract
Abstract Questions about authorship have plagued the corpus of Demosthenic orations since antiquity. In particular, scholars often assign certain speeches (usually 46, 49, 50, 52, 53 and 59; sometimes also 47 and 51) to Apollodorus, son of Pasion. We apply an innovative approach to the problem, using morphosyntactic information from dependency treebanks. From the treebank annotation we create input data for various well-established computational approaches to authorship attribution. The usefulness of the input data is first tested with clustering algorithms. We then make finer distinction with a logistic regression classifier. All steps are explained in detail for the benefit of those unfamiliar with computational stylometry. In broadest terms, our results are remarkably consistent with the common opinion about the orations, identifying 49, 50, 52 and 53 as written by a single author, who was not Demosthenes (presumably Apollodorus). We also discuss syntactic traits that are peculiarly ‘Apollodoran’ or ‘Demosthenic’. However, we demonstrate that the data point away from both authors for Dem. 46 and 51, while conclusions about 47 and 59 are ambiguous.