Our first workshop with an invited guest instructor was held online.
Tobias Hodel, professor for Digital Humanities at the University of Bern, gave an introduction to Transkribus and the possibilities of Handwritten Text Recognition (HTR) for Old Norse manuscripts. The group invited Tobias because of a general interest in both using HTR systems and to gain a better understanding of how they work.
We started off with the use of the Transkribus software: Document management, layout recognition, text recognition and data export. Then we discussed how to fine-tune layout recognition to cope specifically with Icelandic manuscripts, and how we could train our own text recognition models for Old Norse texts. We learned to interpret error rates and learning curves, and how to recognize good and overfitted models.
In the discussion we came up with interesting ideas, how the Menota catalogue data could be re-used to train a two-layered Transkribus model with abbreviation marks and their corresponding expansions, and how the Transkribus APIs could be used to import existing spatial data for layout recognition.
We also learned about other exciting initiatives, like e-scriptorium, OCR4all, HTR-United and Kraken.