EpiPalaeo: Unlocking inscriptions with AI models

Duration: September 2026 – March 2027

PIs: Jun. Prof. Dr. Jakob Jünger (University of Münster)

Research associates: Luana Moraes Costa (University of Münster), Patrick D. Brookshire (Academy of Sciences and Literature | Mainz)

Student Assistants: Amelie Stolz, Annabell Jendrilek, Emma Grimm, Josefine von Puttkamer, Philipp Baranyai (all from the University of Münster)

Funding: Heidelberg Academy of Sciences

Keywords: Epigraphic Text Recognition, Automated Image Recognition, Artificial Intelligence, Historical Communication Studies

Abstract

How can AI models contribute to the analysis of inscriptions? This is the focus of the EpiPalaeo pilot study. As part of the project, models for the segmentation, classification, and recognition of inscriptions will be developed. To this end, the research team is annotating an extensive corpus of images of inscriptions from the Middle Ages and the Early Modern Period, which is characterized by its diversity of scripts and objects. The results will lay the groundwork for digital epigraphic paleography and Epigraphic Text Recognition (ETR).

Inscriptions are a significant part of our cultural heritage and can be found almost everywhere in public spaces, such as in and on churches or secular buildings. Researching them offers a glimpse through a keyhole into the history of social and cultural contexts. These objects are often at risk of deterioration and can only be interpreted by experts who preserve them for basic research and for future generations. The project evaluates the extent to which automated inscription recognition can support the decipherment and preservation of these inscriptions. Powerful methods for handwritten text recognition are already available, whereas for ancient inscriptions, existing models primarily serve to restore and contextualise texts that have already been transcribed. These approaches cannot readily be transferred to medieval and early modern inscriptions: in addition to distinctive letterforms, ligatures, abbreviations and irregular layouts, as well as the materiality of the inscribed objects and their state of preservation, pose challenges for automated recognition.

The EpiPalaeo project is led by Jakob Jünger in the Digital Media & Computational Methods research group at the University of Münster and is a joint project between the Academy of Sciences and Literature in Mainz and the Heidelberg Academy of Sciences. It is being carried out in collaboration with the Academy’s Laboratory for Artificial Intelligence at the Mainz Academy. The Heidelberg Academy is also contributing its extensive photographic collections relating to the inscriptions of the town of Pforzheim, the former district of Mergentheim and the Hohenlohe district. The study forms part of the inter-academic project ‘Die Deutschen Inschriften’  (The German Inscriptions). The aim of the overall project is to compile an epigraphic reference work covering the German-speaking world from approximately 500 to 1650.