Researchers are releasing a language model designed to help scholars reconstruct damaged Ancient Greek papyri. Apollo was trained on roughly 600 million words from manuscripts, inscriptions and papyrus fragments, and will be available free to academics through a chatbot.
Restoring a fragment normally requires specialists to identify word boundaries, date the document, recognize its dialect and weigh its historical setting. Apollo can use those signals to suggest statistically likely words for missing passages. Its training is broad enough to distinguish, for example, Homeric Greek from the Doric dialect used in some inscriptions.
The Austrian Academy of Sciences developed the model with French AI company Mistral and technology firm Sail Reply. Beyond reconstruction, the team expects it to help researchers find fragments relevant to narrow topics across large library collections.
Apollo does not produce a single authoritative restoration. It offers several possibilities for a scholar to assess, an important safeguard because language models can generate plausible but incorrect text. Researchers also caution that the likely gains are incremental: the tool may reveal details of everyday ancient life, rather than suddenly recovering famous lost works.