Image and Interpretation: Using Artificial Intelligence to Read Ancient Roman Texts

The ink and stylus tablets discovered at the Roman Fort of Vindolanda are a unique resource for scholars of ancient history. However, the stylus tablets have proved particularly difficult to read. This paper describes a system that assists expert papyrologists in the interpretation of the Vindolanda writing tablets. A model-based approach is taken that relies on models of the written form of characters, and statistical modelling of language, to produce plausible interpretations of the documents. Fusion of the contributions from the language, character, and image feature models is achieved by utilizing the GRAVA agent architecture that uses Minimum Description Length as the basis for information fusion across semantic levels. A system is developed that reads in image data and outputs plausible interpretations of the Vindolanda tablets.