Austrian Academy Develops Apollo to Reconstruct Damaged Ancient Greek Texts

The Austrian Academy of Sciences plans to present Apollo on September 23: an AI system built with Mistral and Sail Reply to suggest missing words and passages in damaged Ancient Greek documents. Its promise is not an automated historical verdict, but a faster starting point for scholars confronting fragments that can no longer be read whole.

A reconstruction task with little room for shortcuts

Ancient Greek writing does not include the word spaces modern readers expect. Restoring a torn papyrus can therefore require a specialist to identify word divisions, date the document, judge its social and political setting, and consult reference works before proposing what belongs in a gap. Apollo is meant to reduce the time spent on that initial reconstruction work, where expertise is scarce and the volume of material is large.

The model was trained on roughly 600 million historical Greek words drawn from manuscripts, papyri and inscriptions. That corpus is intended to let it respond differently to varieties of Greek: Apollo’s developers say it can draw on Homeric Greek when it encounters Homer and on Doric when it encounters an inscription in that dialect.

What the system is designed to provide

• Likely words or longer passages for gaps in damaged papyrus fragments.
• Several possible reconstructions for a scholar to weigh, rather than one final transcription.
• A chatbot interface that is scheduled to be free for academic users.

The crucial point is that human competence needs to remain.

Anna Dolganov, historian and papyrologist at the Austrian Academy of Sciences

Suggestions, not a single answer

That design choice is central. A language model predicts what is statistically likely, while historical reconstruction also depends on context that may be uncertain or contested. Apollo is built to return options for a scholar to assess, an approach intended to avoid presenting a model’s most probable completion as an authoritative reading of the past.

For classicists, the useful outcome may be time. Armand D’Angour, a professor at the University of Oxford, said that being offered three words that could fit a gap would speed the work considerably. Researchers behind Apollo expect that to free scholars to spend more time interpreting documents and less time generating a first set of plausible restorations.

A tool for accumulation, not lost masterpieces

The ambition should not be confused with a machine for recovering famous lost works. Many unreconstructed papyri are mundane records, including personal letters, marriage contracts and civil-service papers. Scholars interviewed about Apollo do not expect it to rewrite the broad outline of the ancient world; its more plausible contribution is a steady accumulation of smaller details about how people lived and worked.

The project’s first stage was reported complete on September 17, ahead of the Vienna presentation. The immediate test is whether academic users find its proposed alternatives useful enough to accelerate research while retaining the human scrutiny needed to distinguish a convincing suggestion from an error that merely sounds at home in an ancient text.