Artificial intelligence transcribes historic diaries of child prodigy, Pet Marjorie

The writings of Marjorie Fleming, who died aged eight exactly 200 years ago, have been transcribed using using Handwritten Text Recognition, a form of artificial intelligence – the first dataset from the National Libraries of Scotland as part of an initiative which could bring centuries-old documents to a much wider audience.

It ‘trained’ the artificial intelligence to learn Fleming’s original handwritten diaries by transcribing 100 pages before it went on to complete her works.

And for the person at the forefront of the project, Professor Melissa Terras, University of Edinburgh, it was a chance to return to her Lang Toun roots.She said: “As someone from Kirkcaldy, we had always heard about Marjorie Fleming in school.

“It has been great to use this emergent technology to be able to share the poignant original source materials with a wider audience.

“We look forward to creating more datasets from the National Library’s collections in this way, and understanding their needs, while also advising other libraries and archives how best to utilise artificial intelligence in their aim to digitise, and provide access to, the past.”

Now, the new data set has been released on the NLS Data Foundry.

Fife Free Press 150th anniversary: Deadlines met and friendships...Revealed: How 'Operation Inferno' mock scenario prepared Fife's...

Traditionally, manuscripts are transcribed by a person, word by word, to be used by researchers - a time consuming, costly task.

Technology which allows computer systems to read historical writing have made significant advances, and the NLS has a commitment to having “one third digital” by 2025.

A three-year research council has brought together teams from the Universities of Edinburgh and Glasgow, and the NLS.

PhD student, Joe Nockels, had to train Transkribus on Marjorie’s handwriting.

The system has a first pass on 100 pages of images of text, and the user then helps correct this.

Learning from its own mistakes, it builds a model of handwriting that can then be applied across a whole series of documents in the same handwriting, with results of up to 96% accuracy.

The output is a digital transcription of handwritten material that can then be read, used for future research, or further analysed by computer methods - much more efficient, and quicker, than transcribing by hand

Joe said: “I had not come across Fleming before applying Transkribus to her writings, but it soon became a delight to read her precocious poems and diary entries.

“On top of that, this dataset can now be fully searched by researchers on a word-level and the model we built can be used by anyone working with similar historical material. It’s been very rewarding.”

Born in Kirkcaldy, Marjorie moved to Edinburgh, and is best known for the diaries she kept for the last 18 months of her short life.

The 1934 edition of her diaries has a tribute from Robert Louis Stevenson saying, "Marjory Fleming was possibly – no, I take back possibly – she was one of the noblest works of God."

Her grave is in Abbotshall Church, Kirkcaldy.

https://data.nls.uk/data/digitised-collections/marjory-fleming/

Thank you for reading this article on our free-to-read website. We're more reliant on your support than ever as the shift in consumer habits brought about by Coronavirus impacts our advertisers.

Please consider purchasing a subscription to our print newspaper to help fund our trusted, fact-checked journalism.

https://www.localsubsplus.co.uk/nord/dm/FFP/V