Innsbruck researchers develop AI model to analyse and restore Latin texts

Researchers at the University of Innsbruck, the Austrian Academy of Sciences and international partners are developing an AI model to translate Latin, reconstruct missing passages and analyse large collections of historical texts, ORF reports.

The model, called Latinitas, is intended to help researchers make sense of extensive and often difficult-to-access records spanning antiquity to the modern era. Innsbruck philologist Martin Korenjak said the volume of surviving texts is too large for individual researchers or teams to examine in full.

The two-year project involves researchers from the Austrian Academy of Sciences and international partners, including AI companies Mistral AI and SAIL Reply. The model is to be trained on about 12 billion Latin words. The University of Innsbruck’s Institute of Classical Philology will test it and provide feedback to help refine the system.

Anna Dolganov, an ancient historian and papyrologist at the Austrian Academy of Sciences’ Austrian Archaeological Institute, secured a grant of 810,000 US dollars, about 722,700 euros, from the US organisation Schmidt Sciences. She was also responsible for an earlier project focused on Ancient Greek.

Latinitas builds on Apollo, an AI system recently presented to analyse Ancient Greek, fill gaps in fragments and help interpret and historically classify old texts.

The Latin model could also be used in schools, Korenjak said, helping students learn Latin and develop their AI skills. The project team expects initial results next year, and other historical languages could eventually be added.