The Austrian Academy of Science, in partnership with French AI lab Mistral and technology services firm Sail Reply, is releasing Apollo on Wednesday. This large language model for Ancient Greek is trained on approximately 600 million historical Greek words to help scholars restore tattered papyrus fragments and accelerate academic research.
The Release of Apollo: An Advanced Large Language Model for Ancient Greek
Academic institutions globally hold vast collections of Ancient Greek papyrus fragments. While many are too damaged to read, researchers can traditionally restore the rest by filling in missing words. To speed up this laborious process, the Austrian Academy of Science is launching Apollo on Wednesday, developed in partnership with French AI lab Mistral and technology services firm Sail Reply. According to Dimitris Vlitas, partner at Sail Reply, unlocking knowledge in this way was unthinkable a year ago.
Trained on roughly 600 million historical Greek words sourced from inscriptions, manuscripts, and papyri, the model will be freely available to academics via a chatbot interface. The system is designed to help scholars quickly identify fragments relevant to specific sub-disciplines. Where documents are torn, Apollo fills in the blanks using statistically likely words or passages, which could reveal hidden details about historical practices and events.
Overcoming the Complexities of Ancient Greek Reconstruction
Restoring tattered papyrus has historically demanded immense expertise. According to Stephen Colvin, a professor of classics and historical linguistics at University College London, There are very few people in the world who are that good at Greek history.
Scholars must first identify word divisions—since Ancient Greek writing contains no gaps—accurately date the document, weigh socio-political contexts, and consult reference materials.
Apollo bakes this specialized knowledge directly into the system. Anna Dolganov, a historian and papyrologist at the Austrian Academy of Science, notes that When it sees Homer, it supplements Homeric Greek. When it sees an inscription in Doric dialect, it uses Doric dialect.
By automating painstaking reconstruction work, academics expect the tool to let them focus on broader historical implications rather than basic transcription.
Scholarly Expectations and the Limits of AI in Classical Studies
Classics researchers view the new technology with optimism while maintaining realistic expectations about its scope. Armand D’Angour, a professor of classical languages and literature at the University of Oxford—home to the world’s largest ancient papyrus collection—states that I think it’s very exciting,
adding that If I had a machine telling me, ‘Here are the three possible words that could fit into that gap,’ it would speed up matters considerably.
Despite the excitement, experts caution that the model will not rewrite the foundational history of antiquity. Many un-restored papyri consist of mundane texts like marital contracts, personal letters, and civil service papers. As Colvin points out, If you were a layperson, you might think suddenly we’ll get a few new plays by Sophocles, but that’s not going to happen.
Even so, D’Angour notes that Every time something is produced, it adds a tiny element of knowledge about the ancient world,
potentially substantiating existing assumptions.
Future Applications and Safeguarding Historical Accuracy
If the launch succeeds, Vlitas suggests the underlying technique could extend to other ancient languages like Egyptian or Latin, or to any academic discipline relying on large-scale corpus indexing. This follows other notable AI achievements, such as OpenAI models solving a 200-year-old math problem, and Google DeepMind releasing a dataset mapping genetic mutations in molecular biology.
To prevent probability-based language models from polluting the historical record with errors, Apollo outputs a selection of options rather than a single definitive answer. Emphasizing the need for human oversight, Dolganov warns that The crucial point is that human competence needs to remain,
adding, If we become totally reliant on AI transcriptions and interpretations of historical material, that’s when the problems start.
Related reading
Discover more from Archyworldys
Subscribe to get the latest posts sent to your email.