Machine Learning and Sigillography: Using Decision Trees to Date British Seal Matrices

The availability of software libraries for machine learning, in combination with the growing number of digital sigillographic information resources, enables scholars to explore how this technology can help date medieval seal matrices. This paper presents and evaluates a pioneering system based on archival seal impression data, courtesy of the Digisig Project. The system constructs a decision tree that can be employed to date seal matrices. To explore the potential practical applications of the method, we compare its outputs with the dates assigned to seal matrices by the cataloguers of the Portable Antiquities Scheme and the Schøyen Collection catalogue. This method does not replace human cataloguers, but it can quickly and cost-effectively identify sections of existing catalogues that might benefit from revision.