The Austrian Academy of Sciences releases the Apollo language model for filling in damaged ancient Greek texts
The Austrian Academy of Sciences, together with Mistral and Sail Reply, is releasing the Apollo model, trained on 600 million words of ancient Greek. It fills in missing passages in damaged papyri and manuscripts and is available free through a chatbot for academics, but the final choice of words remains with experts.
The Austrian Academy of Sciences, in collaboration with the French company Mistral and the technology company Sail Reply, is releasing the Apollo language model, described as the first advanced large language model for ancient Greek. The model is trained on approximately 600 million words of historical Greek drawn from manuscripts, papyri and inscriptions, and is available free to academics through a chatbot interface.
Apollo is intended to fill in missing or damaged passages in torn or worn texts by suggesting the statistically most likely words or passages. In doing so, the model distinguishes dialects and authorial style — it fills in Homeric Greek for a Homeric text and chooses Doric forms for an inscription in the Doric dialect. According to the experts quoted, such reconstruction has until now required specialists with rare qualifications who had to separate words (ancient Greek is written without spaces), date the text and take into account the social and political context of the period.
Experts quoted in the source expect the model to speed up research by allowing researchers to focus on interpreting historical documents instead of painstakingly reconstructing them. At the same time, they caution that Apollo is unlikely to lead to the discovery of fundamentally new literary works — many papyri that have yet to be processed are everyday documents such as personal letters, contracts or official records — but it may gradually fill in small details about life in antiquity and confirm existing scholarly assumptions.
To avoid the risk of introducing errors into the historical record, the model always offers a selection of possible words from which a human makes the final choice, according to one of the experts quoted. According to a representative of Sail Reply, the same approach could, if successful, also be applied to other ancient languages, such as Latin or Egyptian.
Why it matters
Manual reconstruction of damaged Greek texts has until now required rare specialists skilled in dating, dialects and historical context; the Apollo model makes this specialized knowledge accessible to a broader range of researchers and may speed up the processing of large numbers of papyri that have yet to be deciphered. According to the experts quoted, this involves gradually filling in small details, not discovering fundamentally new literary works, and the final decision on the words inserted remains with a human to avoid errors in the historical record.
Release card
Apollo
Rakouská akademie věd s firmami Mistral a Sail Reply
- Inputs
- text (filling in damaged ancient Greek texts)
- Price
- free for academics
- Availability
- Available free to academics through a chatbot interface
- Filling in missing or damaged passages in manuscripts, papyri and inscriptions in ancient Greek
- Distinguishing dialects and authorial style (e.g. Homeric Greek, Doric dialect) when suggesting completions
- Speeding up the identification of papyrus fragments relevant to specific research subfields
- Does not offer the discovery of fundamentally new literary works, because many papyri that have yet to be processed are everyday documents (letters, contracts, official records)
- A human must always make the final choice of the word to fill in from the options offered; the model does not complete the reconstruction on its own
The card summarizes information from the article and any dated corrections, with a link to the original source. It is not our assessment of the model. It does not yet have a dedicated editorial profile. Model selection and other announcements →
Two audiences, two different impacts
What this means
For individuals
Researchers working on ancient Greek texts (papyrology, classical philology) gain a freely available tool that suggests completions for damaged passages and may save them time spent on manual reconstruction, allowing them to focus on interpreting the content.
For a business
The project demonstrates a successful collaboration between an academic institution, Mistral and Sail Reply in building a language model on a narrowly specialized corpus (600 million words of historical Greek) — an example of collaboration that other companies may also consider when developing domain-specific AI tools for specialist institutions.
DevelopmentCheck the original
Event sources
only one source so far · 1 publisher, 1 independent. We count feeds from the same owner only once.