The Epistle to Cangrande through the Lens of Computational Authorship Verification

The Epistle to Cangrande, also known as Epistle XIII, may be one of the most interesting and debated documents in the biography of Dante Alighieri. Indeed, the Epistle is characterized by some very unique, most interesting peculiarities that make it stand out among other documents: not only it is one of the very few evidences of a medieval author’s auto-exegesis, hence representing a precious specimen of a rare phenomenon, but it is also the only texts survived until our times where Dante discusses his masterpiece, the Commedia. Thus, the importance of this literary work is fully evident. Nevertheless, it may be inexact to appoint Dante as the actual author of the letter. Many scholars have raised doubts about the supposed Dantean paternity of the epistle, presenting many powerful arguments why it should be considered a forgery (or a partial one); on the other side, many academics have countered with various proofs that the Epistle should be, indeed, ascribed to Dante. The dispute itself has lasted over a century. In this landscape, it seemed imperative to gain a new perspective on the authorship problem. To this aim, it seemed appropriate to turn to a methodology that, despite being considerably far from the traditional literary traditions and practises, has already helped humanists with many similar cases with interesting results: Computational Authorship Analysis. This practice’s purpose is to enlighten and inspect the stylistic traits of a certain textual piece (or of a certain author) considering some quantitative elements found in said piece (or in the textual production of said author). More specifically, the Epistle’s case can be thought as an Authorship Verification problem, a sub-task of Authorship Analysis which aims to determine whether the author of a certain set of documents is also the author of an anonymous text. The most prolific partnership was started with the CNR (Centro Nazionale di Ricerca, National Research Center) of Pisa in order to achieve this goal, in a proper collaboration between humanities and hard sciences. This dissertation will explain the systems and procedures modelled in said research, show the results obtained from these methods and comment on them accordingly. In particular, tracing similar studies in the field of Machine Learning, the problem was approximated to a binary classification task, where the candidate author’s texts (in this case, Dante’s Latin works) are taken as positive training samples, and other authors’ as negative training samples. After the definition of a set of features, a linear classifier (Logistic Regression) was trained over the corpus, and then asked to infer whether the Epistle XIII was written by Dante or not