Turku Book Fair

The READ project and its technology were demonstrated to the public at the National Archive’s section at the Turku book fair in the beginning of October 2019. The themes for this year’s book fair were theatre, cultural impacts of climate change, and literacy. Not surprisingly, our section was mainly connected to the theme of literacy.

Court records on display at the book fair

The public could search for their ancestors or other interesting topics from our new interface. We also had some original court books from the 19th century on display. Unfortunately the first version of the search interface does not include the Turku region but luckily many visitors had relatives all around Finland. Some visitors managed even to find new information about their ancestors, for example a house being sold about 150 years ago.

Visitors testing the court record interface and ScanTent at National Archives's stand 

We also demonstrated how you can easily scan books and all kinds of documents with the ScanTent and DocScan app. People could also try how the computer reads their own handwriting. As we currently do not have any model for modern Finnish handwriting, we used a model developed for the war journals from the Continuation War (1941-1944). The results were quite mixed and sometimes even comical. It was interesting to see that sometimes the model read for an example a young girl’s handwriting pretty much perfectly but struggled horribly with older people who had learnt to write at a much closer time to the model’s examples.

The ScanTent and DocScan demo was popular and kept as busy at the book fair. During the weekend, we probably scanned a couple of hundred people’s handwriting and many times there was a queue to the scanning spot. The demo acted also as an introduction to the HTR process – we demonstrated that by correcting the transcription and gathering more writing examples, the model could have been trained to understand the visitor’s handwriting more accurately.

A handwriting sample where the HTR model performed fairly well

... and one where the result wasn't that great
Sampo Viiri

Comments