The name of that project is Decoding Hidden work that is underway as part of this project on the manuscripts of the Main Collection that will be discuss in this blog post. The Main Manuscript Collection is estimat to contain approximately 700,000 manuscript pages, making it one of the largest collections of folklore material in western Europe. This would have been a significant challenge for digitization if Transkribus had not develop the AI technology for handwriting recognition in the last few years, a technology that Decoding Hidden Heritages relies heavily on and which allows the project to make its own handwriting recognition engines from a basis bas on particular scribes in the collection.
Retraining the language models
Since our researchers start working with the Turkey Phone Number List Transkribus software at the beginning of October, we have manag to make three known handwriting recognition engines that work at an accuracy rate of 95%+, one for Seosamh Ó Dálaigh, one for Seán Ó hEochaidh and one for Liam Mac Coisdealbha, three of the most enthusiastic collectors who work for the Commission. Seosamh Ó Dálaigh in the process of collecting folklore from.
That the collectors were
Thomás Mac Gerald (Márthain Parish, Cork Dhuibhne) and a manuscript he wrote down from a recording he made of Thadhg Ó Guithín (Baile na Abha, Dún Chaoin, Cork Dhuibhne) being transcrib in Transcribus. An accurate B2C Database transcription faithful to the manuscript page, handwriting and dialect of the collector requir by Transkribus to train the engine. It was found that transcribing about half a page in this way was enough to train a handwriting model with a fairly satisfactory accuracy rate (90%+). The approach of the following project is to transcribe a large number of pages automatically, correcting them gradually with the help of assistants (Emma McGee, Kate Ní Ghallchóir and Róisín.
