Research and Data

Bibliography

The bibliography of the project is available on Zotero.

Models

  • 5 models trained in Transkribus
Model_nameDate_trainedBase_modelNumber_epochsNumber_GT_pagesTrain_setValidation_setNumber_linesNumber_wordsCER (training set)CR (validation set)Comments
LAD 1.015/08/2020Gothic Books (Hodel)5087158835470.27%4.52%Based only on the LAD manuscript (LAD 2013.051). Bias towards Genesis, Exodus, Mark and Matthew
LAD 1.122/08/2020LAD bible 1.050241951592963211.89%7.20%Based only on the LAD manuscript (LAD 2013.051). Same GT as 1.2 with a different base model
LAD 1.222/08/2020Charter Scripts (Hodel)5024195159296320.62%4.14%Based only on the LAD manuscript (LAD 2013.051). Same GT as 1.1 with a different base model
LAD 1.326/10/2020Gothic Books (Hodel)100393092516152580.51%3.01%Based only on the LAD manuscript (LAD 2013.051).
PBP 1.029/06/2021LAD bible 1.3502516915288402.04%12.76%Composite model based on Paris Bibles from around Europe in the 13th and 14th centuries.
PBP 2.0Upcoming          

Manuscripts

  • More than 320 manuscripts from 76 institutions in 14 countries
  • The versioned list of manuscripts is downloadable at Zenodo.

Transcriptions

  • 186 pages of ground truth
  • Versioned ground truth is downloadable at Zenodo.


Last update: 14/06/2023