Paper short abstract:

The paper describes Minna de Honkoku, a crowdsourcing platform for transcribing large volumes of historical documents with a built-in AI kuzushiji recognition function.

Paper long abstract:

Before the 19th century, the majority of documents published or written in Japan were written in kuzushiji, a cursive script that is no longer used in publications and is therefore difficult for modern readers to decipher. To access the knowledge contained in these historical documents, it is often necessary to transcribe a large number of kuzushiji documents in an efficient manner. However, the number of people with the ability to decipher kuzushiji is said to have fallen below 10,000.

Recently, advances in deep learning technology have made it possible for AI to automatically recognize kuzushiji characters. AI-based mobile applications, such as miwo (http://codh.rois.ac.jp/miwo/index.html.en), which automatically recognize kuzushiji characters, are now available. However, the accuracy of these AI programs is still far from perfect, resulting in many recognition errors. A more realistic approach to large-scale kuzushiji transcription is to combine human and machine intelligence.

Minna de Honkoku (https://honkoku.org/index_en.html) is a crowdsourcing platform for transcribing large volumes of historical Japanese documents. It has a built-in AI kuzushiji recognition function to support the transcription efforts of its participants. Since its release in 2019, the platform has completed the transcription of over 1,500 documents, totaling over 20 million characters.

This paper discusses two aspects of Minna de Honkoku. The first is its technical aspect as a Digital Humanities project, including the adoption of the International Image Interoperability Framework (IIIF) to facilitate collaboration with multiple digital archives, as well as its integration with more powerful AI through its collaboration with the miwo app.

The second aspect is Minna de Honkoku's engagement with the general public as a public history project. We have promoted our project through social media platforms such as Twitter and Facebook, as well as through a live broadcast program on Nico Nico Douga. As a result, the platform has attracted 2,000 volunteers as participants.