Machine Learning and the Text of Aristotle

This article uses the Princeton-based AI Logion and its error detection algorithm to show that large language models can contribute to the textual criticism of Aristotle. We discuss a total of eight case studies from the Metaphysics, Poetics, and De motu animalium to demonstrate that Logion can (i) correctly identify corruptions in the transmitted text of Aristotle and (ii) suggest plausible emendations. Even where Logion’s suggested readings are not viable, they can alert the human philologist to problems in the text and thus initiate a search for new solutions. We conclude that language models like Logion can contribute to the current revival in the study of the Aristotelian text, provided we use them responsibly and hold on to the fact that, while machines may make intriguing suggestions, only human philologists can ultimately adjudicate philological problems.