Introduction to Transkribus and the New User Interface:
The workshop
By Celeste Pedro
Abstract:
Transkribus is a software application that uses artificial intelligence to perform text recognition and transcription of handwritten and printed documents. With Transkribus, users can search and upload digitised corpus, use text recognition based on existing models or customised ones, and train, edit, export, and share their work.
It was developed in the context of an EU Horizon 2020 project based at the University of Innsbruck called READ, whose research consortium eventually created a European Cooperative Society to guarantee widespread facilitation of resources and continued research on software improvements and new technologies. This article aims to highlight the features of the user interface of Transkribus’ online platform.
Fig. 1: Landing page of the Read-Coop website.
Transkribus has been around for some years (click here to learn more about its history), and the community of users is growing and is actively involved in testing and commenting across social platforms. On the 13th of November, the Transkribus team offered a workshop in English to present its new user interface features. It lasted 1,5 hours plus Q&A, and around 190 people attended the webinar. The interest is high, and the improvements are pretty relevant.
The platform is structured into four main “tasks” that the user can perform over his collections or individual pages. It makes working with historical documents much easier, be it collecting information or searching and sharing transcribed editions. There are two recognition features: Texts and Layouts; the Train your own AI Model feature; and the Edit and Correct feature. There are also instrumental metadata enrichment possibilities.
This AI platform uses HTR (handwritten text recognition) technology – a machine learning tool, to facilitate the job of transcription with the most accuracy possible. This accuracy is possible if models are trained. Models are trained to recognise handwriting and other visual elements in documents. It can deliver a very accurate result in recognising different handwritings and typefaces if provided with sufficient examples (of letter/word/abbreviation shapes and corresponding transcription). Transkribus already has a significant database of public models and some super models that include multiple language recognition that users can apply to their documents. If not, they can train their custom models (here’s a video about it), a big task, considering it first takes transcribing an average of 10.000 words by hand to then perform recognition tests. Still, it is instrumental in research… and many texts we work with are longer than that anyway.
Fig. 2: Example of a public model for Carolingian hands.
During the workshop, the Transkribus team presented the new interface and what is to come for users of the platform:
Tags: tags add metadata to documents, and with Transkribus, one can do that to text units, such as the name of the person or place (adding a wiki data ID, for example), as well as selected areas of the page, called structural tags. Users can also customise tags in the configuration box. By selecting parts of images, users can use different recognition models for different areas of the image. Tags used in these ways can be very beneficial for information retrieval. Tags can also be exported in several ways.
Layouts: it takes at least 50 pages to train a layout recognition model, and layout models can include different languages as well; it all depends on how many examples are added. Transkribus already has a few models users can select and customise or alternatively build from scratch. Identifying and tagging areas of the layout is very easy with the new interface. Transkribus can also be used and trained to recognise text in tagged regions in forms, tables or diagrams with lines and multicolumn layouts.
Fig. 3: Me, playing with Transkribus, thinking, “Can I train a model to recognise text inside circular diagrams?”
Visit the Trankribus website or the app or have a quick try here! There will be news over the next two months. The much-awaited TEI export function is arriving and if you subscribe to a paid plan, you’ll also have access to Transkribus Sites, where you can set up your own website to share your work.
This new generation of the user interface suffered a really good design upgrade. It’s intuitive and organised, and there’s so much you can do to your documents, including collaborating with others on documents or sharing your work with the public.
Links in the text:
IPM Monthly 