Epigrafia greca e intelligenza artificiale. Il progetto Lacunae dell’Universita di Udine

This paper presents the first results of the LACUNAE project, which aims to develop digital support for the study and restoration of lacunae in ancient Greek inscriptions. While recent research has primarily addressed the problem from a linguistic and text ual perspective using Natural Language Processing and Large Language Models, this study proposes a shift in focus by approaching lacunae as material phenomena before treating them as textual ones. For the first time, Computer Vision (CV) is applied not to character recognition, but to estimating the spatial capacity of a lacuna. The method calculates the minimum and maximum number of characters that a gap may contain based on the proportional relati onship between the available space and the size of epichoric letter forms. Through image enhancement, binarization, and layout analysis, the system examines inscribed surfaces and models graphic variability in letter shapes. This approach makes it possible to objectify a crucial yet traditionally intuitive step in epigraphic practice — estimating the length of a restoration — and to provide quantitative constraints that assist scholars, without automating or determining the philological restoration itself