'A wonderful project,' says Myrthe enthusiastically. 'Our aim was to make the many writings - mostly correspondence - of Abraham Kuyper accessible. In addition, we wanted to investigate how we could use AI in this, so that we could automatically convert scanned letters into readable files.' Not a superfluous luxury, given that the Historical Documentation Centre (HDC) of VU has no less than some 100 boxes of Kuyper archives under management and there is undiminished interest in his work. Moreover, some 4 kilometres of other archival material in the HDC awaits digital access.
'Because that is the beauty,' Myrthe says, 'that with the knowledge and experience from this project we will soon be able to unlock other archives faster and better,' she refers to the AI model they trained to decipher Kuyper's handwriting. 'We first fed Transkribus, a widely used European AI tool within archival institutions, with manual transcriptions,' she explains. 'That is a lot of work, because first you have to decipher documents manually, word by word. Especially with such tricky handwriting as Kuyper's, that is really monk's work. Then you automatically transcribe scanned documents and check how many errors they still contain. You feed that back to the model so that it can be refined again.'
This seems simple, but there is still a lot of precise manual work involved. 'We are now approaching completion,' says Myrthe. 'And then we will have an AI model that can transcribe Kuyper's handwriting with less than 10% margin of error; that's a very good score! This model can then also be used outside the VU archives, so that Kuyper's legacy becomes accessible elsewhere in the Netherlands and it no longer matters where such a document is located.'
'Because therein lies the added value of this project,' Myrthe explains. 'That we can unlock Kuyper's archive everywhere, but soon also other archives that we can train with this AI model in the same way. For example, I would very much like to continue with archives that contain a lot of correspondence from administrators and professors who were connected to VU. Because the letter and the answer can then be examined together, contacts can be investigated better and that provides a good insight into the past. In this way, we facilitate researchers with our work, so that they can use their work to extract and share the stories and insights from written texts'.