AI Transcribes 32,763 Medieval Manuscripts in Four Months

Researchers in computational humanities at Inria report the creation of CoMMA, a massive corpus of 32,763 medieval manuscripts transcribed automatically using artificial intelligence. The project was completed in four months. To scale up their efforts, the team utilized digitized manuscript holdings and metadata aggregated through the Biblissima+ platform.
The manuscripts in the corpus date from approximately 800 to 1600 CE and are written in medieval Latin, Old French, and Italian. For automatic text recognition, the researchers employed Handwritten Text Recognition technology using the Kraken engine alongside CATMuS models. In total, the resulting corpus contains over 3.3 billion tokens, providing a vast dataset for scholars studying historical texts.