Academic institutions worldwide hold hundreds of thousands of fragments of Ancient Greek papyrus, many of which are so deteriorated that their original meaning has likely been lost. However, a new collaboration between the Austrian Academy of Sciences, French AI laboratory Mistral, and technology firm Sail Reply aims to accelerate the painstaking process of restoring these texts. On Wednesday, they will introduce Apollo, described as the world’s first advanced large language model specifically built for Ancient Greek.
Apollo has been trained on approximately 600 million historical Greek words sourced from manuscripts, inscriptions, and papyrus fragments. The model will be made available to researchers via a free chatbot interface, designed to help scholars quickly identify relevant fragments and explore new research directions. For damaged documents, Apollo predicts the most statistically probable words or passages to fill in missing sections, potentially uncovering previously hidden details about historical events and social practices.
Dimitris Vlitas, a partner at Sail Reply, noted that the ability to unlock knowledge in this manner was “unthinkable a year ago.” Historically, restoring a fragmented papyrus required experts to decipher word divisions in texts that lacked spacing, accurately date the document, assess socio-political contexts, and consult extensive reference materials to select appropriate language. Stephen Colvin, a professor of classics and historical linguistics at University College London, emphasized the rarity of such expertise, stating, “There are very few people in the world who are that good at Greek history.”
According to Anna Dolganov, a historian and papyrologist at the Austrian Academy of Sciences, Apollo internalizes this specialized knowledge and adapts its linguistic approach based on the text type. “When it sees Homer, it supplements Homeric Greek. When it sees an inscription in Doric dialect, it uses Doric dialect,” she explained. This capability allows academics to move past laborious reconstruction work and focus on interpreting the implications of historical documents.
Armand D’Angour, a professor of classical languages and literature at the University of Oxford, which houses the world’s largest collection of ancient papyri, welcomed the development. He suggested that having a machine propose the top three possible words for a gap would “speed up matters considerably.” However, experts caution that the model is unlikely to revolutionize the broad understanding of the ancient world, as many unrestored papyri consist of mundane records such as personal letters and civil service documents rather than significant literary works.
While Apollo may not yield new plays by famous tragedians like Sophocles, it could reveal subtle details about daily life in antiquity and reinforce existing scholarly theories. Vlitas added that if successful, the technique could be adapted for other ancient languages, including Latin and Egyptian, or applied to any discipline requiring the analysis of large corpora of material.
Addressing concerns that probabilistic AI models might introduce errors into the historical record, the creators ensured that Apollo proposes multiple word options for scholars to evaluate. Dolganov stressed the importance of maintaining human oversight: “The crucial point is that human competence needs to remain. If we become totally reliant on AI transcriptions and interpretations of historical material, that’s when the problems start.”
Leave a Reply