TRIDIS (Tria Digita Scribunt) is a Handwritten Text Recognition model trained on and for medieval and Early Modern manuscripts. While trained specifically for legal, administrative, and memorial writings from the Late Middle Ages, it may also be useful for a more diverse range of materials including literature and treatises. The model was originally trained on a dataset of legal documents of 2,950 pages (245 thousand lines) and a synthetic dataset of 300 thousand lines. TRIDIS recognizes Latin, Old French, and Old Spanish including a variety of Latin Script families.
The authors of the model have an accompanying paper “Hand written Text Recognition for Documentary Medieval Manuscripts” and the training data are also available for download from the project’s homepage.
dh+lib Review
This post was produced through a cooperation between Carla Brooks, Amy Gay, Miranda Phair, and Michelle Speed (Editors-at-Large), Caitlin Christian-Lamb and Ruth Carpenter (Editors for the week), Claudia Berger, Nickoal Eichmann-Kalwara, Linsey Ford, Pamella Lach, Molly McGuire, Hillary Richardson, Christine Salek, and Rachel Starry (dh+lib Review Editors), and Tom Lee (Technical Editor).