Revolution in handwriting recognition: New opportunities for research!

Revolution in handwriting recognition: New opportunities for research!

Researchers at the University of Passau have launched an innovative project on error methodology in automatic handwriting recognition. Professors Malte Rehbein and Alexander Werth are leading the project, which is part of the Volkswagen Foundation’s “Awakening” funding line. The project duration extends from 2025 to 2027 and is called “Methodology of the Inaccurate”.

The central goal of this project is to investigate the extent to which incorrect data can enable scientific work. Automatically transcribed historical manuscripts from council minutes from the 17th to 19th centuries are used, which have an accuracy of around 90%. The project will compare these transcriptions with manually transcribed data, achieving 100% accuracy.

Science Center “Methodikum”

The project is part of the “Methodikum” science center, which was founded by chairs for multilingual computational linguistics, computational humanities and German linguistics. The aim of the “Methodikum” is basic methodological research in the humanities as well as the support of computer-aided and digital methods.

The relevance of automatic handwriting recognition is also underscored by the challenges faced by machines. Humans are able to decipher handwriting, whereas this ability is much more complex for machines. Tobias Hodel from the Zurich State Archives reports on the progress in automated handwriting recognition in projects such as READ, which is funded by the European Commission and aims to transcribe large amounts of handwritten documents.

Technological support and level of development

A central tool in this context is the free Transkribus software, which not only enables automatic recognition and transcription, but also the search of historical documents. This software has established itself as a valuable tool for archives, libraries and documentation institutions.

Using Transkribus requires creating training data through manual transcription, which is time-consuming but can significantly improve the quality of recognition. Current results show that some models of the software can achieve a Character Error Rate (CER) of less than 1%, which represents remarkable accuracy for specific text corpora. Transkribus also offers various tools for creating your own models, adapted to the respective needs of the user.

As technology has advanced, there have been significant improvements in handwriting recognition in recent years. The quality of the automatic transcriptions depends primarily on the font used and the number of hands used. Despite these advances, the challenge remains that machines are currently unable to achieve a 0% error rate, with acceptable levels for human transcriptions being below 10%.

Overall, the development in automated handwriting recognition shows how the combination of human and machine intelligence can enrich scientific work and significantly advance the digitization of historical holdings. The increasing efficiency of transcription, supported by HTR technology, opens up new opportunities for historical research and the development of valuable information from archival materials. Collaboration between scientists and modern technologies has the potential to revolutionize research in the humanities.

For further information and insights into the technical aspects of handwriting recognition, see the project pages University of Passau, DHC and Bop recommended.