The Austrian Academy of Sciences and Mistral have built a language model that suggests restorations for missing passages in ancient manuscripts, inscriptions and papyri. Scholars remain responsible for evaluating its proposals.
Apollo was trained on 600 million words of historical Greek texts. Researchers can submit a damaged fragment, and the model compares it with its training corpus to suggest likely missing words.
The system is available through a chatbot interface. The Austrian Academy of Sciences says it is intended to help researchers find overlooked details in historical sources and speed up scholarly work.
The model’s proposals do not replace the expertise of historians and philologists. Apollo generates statistical suggestions, while scholars decide which reading best fits the language and context of each document.
The Austrian Academy of Sciences says more than one million ancient papyri are held around the world, many of them still undeciphered. It describes Apollo as the first advanced language model developed specifically for Ancient Greek.