Researchers develop AI model to reconstruct ancient Greek papyrus fragments

Researchers at the Institute for Digital Antiquities have unveiled a specialized large‑language model designed to reconstruct missing text on ancient papyrus fragments. The AI system, dubbed PapyrusGPT, was trained on a curated corpus of over 200,000 digitized Egyptian, Greek, and Roman manuscripts, enabling it to predict plausible wording for deteriorated or incomplete sections. By integrating high‑resolution imaging, multispectral analysis, and contextual linguistic patterns, the model can suggest restorations that align with known historical syntax and vocabulary, offering scholars a new tool for piecing together fragmented documents without manual conjecture.

Initial trials on a collection of 3rd‑century BCE Egyptian medical papyri demonstrated that PapyrusGPT successfully filled gaps in 87 percent of test cases, producing reconstructions that matched expert‑validated transcriptions within a margin of error comparable to human scribes. The project, funded by a joint grant from the European Research Council and the National Science Foundation, aims to accelerate the study of ancient life by reducing the time required to decipher damaged texts. Researchers anticipate that broader deployment of the model will facilitate more comprehensive analyses of historical sources, potentially revealing previously obscured details about daily practices, trade, and scientific knowledge in antiquity.