Apollo: Specialized LLM Launched to Restore and Interpret Ancient Greek Papyri
The Austrian Academy of Sciences, alongside Mistral AI and Sail Reply, has launched Apollo, an AI model trained on 600 million ancient Greek words. Available freely through a chatbot interface, the tool assists scholars in completing missing text and interpreting severely damaged ancient papyrus fragments. Restoring damaged ancient manuscripts historically required scarce domain experts to manually deduce missing words from continuous script and complex dialects. Apollo dramatically accelerates this tedious workflow and provides a framework that can be expanded to other ancient languages like Latin or Ancient Egyptian. Apollo automatically adapts to specific linguistic styles, switching to Homeric Greek or Doric dialect depending on the text context. To prevent AI hallucinations from corrupting historical records, the model proposes multiple probabilistic candidates rather than single definitive answers, keeping human expertise central to final decisions.
## BACKGROUND
Papyrology is the study of ancient manuscripts written on papyrus, which was the primary writing material in ancient Mediterranean civilizations like Greece and Rome. Ancient Greek manuscripts were typically written without spaces between words or punctuation, making the task of identifying word boundaries and reconstructing fragmented texts extremely labor-intensive.