Academic libraries worldwide hold numerous Ancient Greek papyrus fragments, many of which are damaged and difficult to interpret. Researchers are now utilizing artificial intelligence to aid in the restoration of these texts. On September 22, 2026, the Austrian Academy of Science announced the release of Apollo, described as the world's first advanced large language model specifically for Ancient Greek. This model was developed in collaboration with French AI lab Mistral and technology services firm Sail Reply, and is trained on approximately 600 million historical Greek words from various sources, including manuscripts and inscriptions.
Apollo will be accessible to academics via a chatbot interface, aimed at helping scholars quickly identify relevant papyrus fragments for their research. The AI is designed to fill in missing words or phrases in damaged texts, potentially uncovering new insights into historical events and practices. Dimitris Vlitas, a partner at Sail Reply, noted that the ability to unlock knowledge in this manner was previously unimaginable.
Traditionally, restoring damaged papyrus required extensive expertise, including identifying word divisions in the continuous text of Ancient Greek, dating the documents, and understanding socio-political contexts. Stephen Colvin, a professor at University College London, emphasized the rarity of scholars with such specialized knowledge. Apollo incorporates this expertise, adapting to various dialects and contexts as needed.
Academics anticipate that Apollo will streamline the reconstruction process, allowing them to concentrate on the implications of the texts rather than the reconstruction itself. Armand D'Angour, a professor at the University of Oxford, expressed enthusiasm about the potential for the AI to suggest possible words for gaps in the texts, significantly speeding up the work.
While Apollo may not drastically alter the overall understanding of ancient history, it could provide new details about daily life in antiquity and support existing scholarly theories. D'Angour noted that each new piece of information contributes to a broader understanding of the ancient world.
If successful, the techniques used in Apollo may be applied to other ancient languages, such as Latin or Egyptian, or to different academic fields that could benefit from analyzing large datasets. Concerns about the accuracy of AI-generated text remain, as there is a risk of introducing errors into the historical record. To mitigate this, Apollo is designed to offer multiple word options for scholars to choose from, ensuring that human expertise remains integral to the process. Anna Dolganov, a historian at the Austrian Academy of Science, stressed the importance of maintaining human oversight in the use of AI for historical interpretation.