You are here

Brunel and Transkribus: joined-up writing

 

Having submitted Brunel’s handwriting data to the Transkribus team earlier this year, we are now seeing some exciting developments to our Handwritten Text Recognition (HTR) project.


The project, which we are working on with cutting edge technology from Transkribus, aims to learn Brunel’s handwriting so that his numerous documents and letters can be transcribed using AI. We have been feeding handwriting samples to the software for analysis, putting the HTR model through some rigorous ‘training’. It is now achieving success rates of between 35-40%, almost four times more accurate than the initial tests.

With the first uploads under our transcribing belts, we are now seeing that the software can detect and successfully decipher certain words and phrases present in Brunel’s notoriously illegible handwriting. A HTR model with a higher level of accuracy will ultimately save countless hours of volunteer work in the Brunel Institute.

As well as building a profile for Isambard Kingdom Brunel we have been working towards another HTR model specifically for his father. Sir Marc’s HTR profile has also been successfully trained and again we are starting to see the program develop an understanding of his handwriting - which is far easier to read than Brunel’s. It will be interesting to see if this has makes a difference to the HTR success rate. Sir Marc was also bilingual and our collections hold several letters written in French. This presents us with an exciting challenge to build a model which can decipher in another language!