Artificial Paleography: Computational Approaches to Identifying Script Types in Medieval Manuscripts

Abstract

Artificial intelligence (AI) is a vibrant research domain in which systems are developed that can reason and act like humans.1 In recent years, the endeavor to reproduce human intelligence in software has led to the introduction of many well-known computer applications that are increasingly, and yet sometimes almost unnoticeably, becoming a part of our everyday lives. Representative examples include face recognition on social media, spam filters for email clients, or recommendation systems in online stores. It is to these highly practical applications that AI currently owes its high visibility—as well as its at times controversial status, as exemplified by the ethical debate sparked by the introduction of self-driving vehicles.2 Nevertheless, these highly useful, practical applications make it easy to forget that AI also addresses more theoretical issues. Being able to reproduce human intelligence, even if only for specific tasks, can help advance our understanding of the working of the human mind itself—as famous physicist Richard Feynman is credited with saying, “What I cannot create, I do not understand.”3 The humanities, which can be broadly defined as the study of the products of the human mind, in this respect seem a privileged partner for AI.4 In the field of digital humanities, various forms of AI have played a role of increasing importance for a number of decades; now that computer technologies are maturing at a rapid pace, we expect to see the emergence of many more collaborations between the humanities and AI in the future. Here we focus on paleography, the scholarly study of historical handwriting, which, apart from being a long-standing discipline in its own right, also remains a crucial auxiliary science in medieval studies for codicologists, literary scholars, and historians alike. Paleography is an interesting case for the application of AI. Whereas most medievalists have at least a superficial reading competency for common script forms, experienced paleographers are typically still required to solve more complex tasks, such as dating, localizing, or authenticating specific scripts. Thus, the field of paleography is dominated by expert-based approaches and driven by the opinions of small groups of highly trained individuals.