Abstract

This paper presents preliminary results on the use of Large Language Models (LLMs), specifically ChatGPT and Claude, to assist with one of the most fundamental tasks in epigraphic research: identifying the typology of Greek inscriptions. As the scale of available material continues to expand—most notably through large online corpora such as the Packard Humanities Institute (PHI)—the ability to process and classify the existing epigraphic corpus at large has become increasingly valuable for studies of Greek epigraphic cultures.

The project evaluates the reliability of ChatGPT and Claude in distinguishing among major categories of epigraphic texts. The presentation will focus on the workflow that has been adopted, as well as the criteria used to assess their performance. While the results indicate that both models classify inscriptions effectively when the texts preserve recognizable formulae or structural cues such as decree-formulae, honorific language, or dedicatory invocations.

The presentation will outline the workflow, highlight areas where the two models converge or diverge in their typological decisions, and discuss the methods used to evaluate reliability. While these results remain preliminary, they demonstrate that LLMs can offer meaningful assistance in large-scale typological processing when used with appropriate constraints and scholarly oversight. Moreover, when combined with curated datasets and stable metadata from established corpora, these automated classifications can form the basis for a more robust attribution of epigraphic types and, thus, the analysis of Greek epigraphic cultures.

Back to Day 2 programme