TraPrInq Handwritten Portuguese 16th-19th c.: the final model
The TraPrInq project (17.1.2022–16.7.2023) created an AI model for automatized transcription of handwritten documents. It is open access on the platform of digital paleography Transkribus.
Further information on the website of the project.
The CER of the model TraPrInq Handwritten Portuguese 16th-19th c. on Validation Set (VS) is 5,2%.
The VS (505 untrained pages) totalizes 107,760 words and the Training Set (TS, 5721 pages), 1.160 280 words.
Since mid-January 2022, the team of paleographers has transcribed about 1.3 million words from a very large number of hands and period from the early 16th to the early 19th century. On the previous training (model M-8), see here.
A few remarks on the VS results:
- 52 of the untrained pages (10,3% of the total of the VS) have a CER higher than 10% (from 10,02% to 20,66%).
- 177 pages (34,8% of the VS) have a CER between 5,2% and 10%.
- 276 pages (54,6% of the VS) have a CER inferior to 5,2%.
OpenEdition vous propose de citer ce billet de la manière suivante :
Hervé Baudry (16 juillet 2023). TraPrInq Handwritten Portuguese 16th-19th c.: the final model. e-Inquisition. Consulté le 24 mai 2025 à l’adresse https://doi.org/10.58079/uubp
What a fantastic project!
How can I get an update on the results of the international cooperative workforce to transcription of the Inquisition processes (Torre do Tombo)?
Thanks
Clea
Dear Clea, thank you for your interest in the TraPrInq Project.
The AI Public Model is available for all on the platform Transkribus.
About its use(s), please see on the page of the TraPrInq Observatory / Inquérito permanente (updated data) (https://traprinq.mozellosite.com/observatorio/).
Very best from the Team of the project.