Fixing the Federalist: Correcting Results and Evaluating Editions for Automated Attribution

In the history of authorship attribution, the analysis of The Federalist Papers plays an important role. However, most previous non-traditional (stylistic) authorship studies have been flawed, mainly due to the use of improper editions, as documented by Rudman (2005). Our goal in this work was to perform a correct study by using a revised corpus of the Federalist papers based in large part on Rudman’s critique. We used machine-learning techniques for analyzing the use of lexical features for authorship attribution of the papers. Another goal of our study was to explore how different corruptions of the corpus may affect the accuracy of the classification results, and the differences between them.