In a piece that I wrote for the Digital Orientalist last year, I compiled a list of digital resources for Japanese palaeography that I had learned about and used through my involvement in the “Tackling Pandemics in Early Modern Japan” transcription project organized by the University of Cambridge in collaboration with the AI platform Minna de honkoku みんなで翻刻. In today’s post, I will introduce readers to Minna de honkoku and provide some reflections based on my use of the platform during the “Tackling Pandemics in Early Modern Japan” project.
Minna de honkoku was developed by Prof. Hashimoto Yuta and others at Kyoto University in 2017 and was updated and expanded in 2019. It is a platform that allows members of the general public, or in some cases select groups such as those involved in the “Tackling Pandemics in Early Modern Japan” project, to collaborate on the transcription of historical documents. The following video provides a detailed introduction to the platform.
An interview with Prof. Yuta Hashimoto explaining Minna de honkoku.
The Platform and its Interface
Minna de honkoku can be used on Google Chrome or Mozilla Firefox. After creating and logging into an account, the user is transported to their timeline. Here the user can see the contributions of others working on the project listed in chronological order with the most recent transcriptions appearing first. They can also navigate to other parts of the platform and to the documents which are being transcribed as part of the project. To access such documents, the user clicks on the “Projects” button, selects the relevant project, and then selects the document that they wish to work on.
An entry on the project’s timeline.
In doing so the user reaches the main section of the platform – the place where transcriptions are input. The interface is fairly simple, and is available in Japanese and English. On the right side of the screen is the image of a single page (or double page) of the document that one is transcribing. The user can zoom in and out of this image using their mouse or trackpad, and move around the image by clicking and dragging. On the left side of the screen the user can do several things:
- View the text which has already been transcribed.
- Input text or edit the transcription.
- View input history, which allows you to track changes to the document.
- View bibliographical information on the text (i.e. the digital repository it is from, its title, publisher details etc).
- View a guide explaining how to input different features of classical Japanese text.
Viewing a document in Minna de honkoku.
To start editing, the user clicks on the “Start Editing” button, and once one has finished editing, they click on the same button which now displays the words “Finish Editing.” Users can also choose to discard the edits or input they have made in a given session, by clicking the “Discard” button. After clicking the “Finish Editing” button, the user may choose to leave comments for other participants to view, share the transcription on the project’s timeline, ask for feedback from other participants, or demark the piece as complete. Each time the user transcribes something they receive points (based I believe of the number of characters transcribed) which correspond to the user’s level. Whilst this points system isn’t a prominent part of the platform, it certainly helped to motivate me and set personal goals at the beginning of the project.
AI Kuzushiji Recognition Tool
The system includes an AI recognition tool for cursive characters which allows users to select an individual character by drawing a box around it and see suggestions for the potential readings of that character. The tool compares the selected character to images from the Center for Open Data in the Humanities’ (CODH) dataset and the Toppan Printing Company’s dataset. The suggestions that it outputs include scores based on the program’s confidence that it is a given character. The user can additionally view the suggested characters in CODH’s dataset or the Denshi Kuzushiji Jiten Database by clicking the links next to the probability scores.
Using the AI Kuzushiji Recongition Tool on a character.
Using the AI Kuzushiji Recongition Tool on a character.
This tool is one of the platform’s principal innovations and makes the transcription of historical Japanese texts accessible to those with very little experience interacting with historical documents. Personally, I found it useful as a tool to check my own choices of characters particuarly when I was uncertain about the transcriptions I had made. The tool provides fairly accurate suggestions for characters, although there are some rare occassions when it fails to include the correct character within its list.
Collaboration and its many forms
At the very heart of Minna de honkoku is collaboration – users collaborate with each other to transcribe texts by leaving feedback or offering corrections. It is, therefore, a highly useful educational tool which facilitates peer (and in the case of our project, teacher) feedback. Nevertheless, Minna de honkoku’s collaborative elements extend far beyond collaborative transcription or collaborative learning – it is collaboration between the digital and the human which is at its heart. Unlike many optical character recognition (OCR) tools which decide upon the identity of and transcribe a character (or more often than not a more significant section of text) for the user automatically, Minna de honkoku’s AI Kuzushiji Recognition Tool requires the user to choose what they perceive to be the correct character (or refuse to choose any of the suggested characters). The user, therefore, maintains an active role in the transcription process, and is constantly interacting with and reading the text.
Prior to using Minna de honkoku, I had always thought that automated transcription was the future, but using the platform and its AI Kuzushiji Recognition Tool has highlighted the importance of active human-computer collaboration and interaction to me. I hope that more digital humanities projects will encourage active human participation with the digital world, rather than passive consumption. Indeed, the active interaction with historical texts that platforms like Minna de honkoku encourage may, I believe, provide a route to introduce more people to palaeography, historical Japanese, and manuscript studies.
Using Minna de honkoku over the past academic year has made it one of my favourite digital tools in the field of Japanese studies. Its simplicity, and the tools and features that it offers position it as an essential platform for use in both collaborative transcription projects and for courses in Japanese palaeography. On a personal level, using the platform has highlighted the importance of collaboration between humans and machines, and of active human participation in digital projects. My only wish is that the platform and its recognition tool or versions of them were avaliable for personal use.