OCR4all

Modern tool for historical texts

When Gutenberg invented the printing press nearly 600 years ago, it was considered a technological revolution. Over the centuries, more and more people gained access to books, newspapers and in this way also education. At the end of the 20th century came what we call the "digital revolution". In 2019, almost everyone owns a smartphone with internet access and thus has access to all the knowledge in the world. All the knowledge in the world? Well, whilst for example the majority of medieval and early modern texts have now been digitised and are available online in the form of high-resolution scans, the works are far from being readable for everyone. The problem of reception lies in converting the old scripts into text that computers can read. Researchers at Uni Würzburg have now succeeded in developing text recognition software that transforms digitised historical scripts into a modern form and thus makes them readable for computers and non-specialists. The tool is called OCR4all, an abbreviation for Optical Character Recognition – and one that is usable by and for everyone. The research project thus functions not only as a bridge between text and recipient, it also forms an interdisciplinary interface between the humanities and computer science. One person who was involved from the beginning is Maximilian Wehner. In addition to his doctorate, he has been working since 2019 as a research associate at the Chair of Artificial Intelligence and Applied Computer Science and as a lecturer for Older German Literature at the Institute for German Philology. He has answered some questions for us about the new tool.

Sprachrohr: What is the secret that enables OCR4all to recognise characters and letters that other text recognition programmes have been unable to recognise until now?

M. Wehner: The secret of the tool probably lies in its overall conception. There have been certain technologies and concepts from the field of artificial intelligence for some years now that theoretically make it possible to access very early prints from the 15th and 16th centuries and convert them into machine-readable texts. Important keywords here are so-called deep learning and neural networks. Unlike other applications, however, OCR4all aims precisely at combining certain technologies, tools and modules in a closed workflow that previously only existed separately from one another – and to design them in such a way that they can be operated independently by users without computer science training. And following this premise, solutions emerged for certain problems that had not previously been discussed or tested in this form. For example, each user can now use our software to train so-called work-specific recognition models that incorporate the particularities of the work to be processed, whether it is the layout design, print quality or the often very specific characters and letters used, continuously learn and thus achieve character and text recognition rates of over 99%. 

Sprachrohr: How long did the development of OCR4all take in total? 

M. Wehner: Development is far from complete! Of course, there has been a freely available version for some time. But the more frequently the software is used, the more users – whether computer scientists, DH specialists or literary scholars – provide feedback and suggestions, the more it becomes clear that in the most positive sense there is still room for improvement. Perhaps there is also something like pioneering spirit at play, but it is simply nice to see how the applications have grown from a very specific user circle here at the university and are suddenly being used worldwide.

Sprachrohr: Do you think it is early or late that such a tool appeared in 2019 and thus more than 25 years after the "Digital Revolution"? 

M. Wehner: Everything in its time! Sometimes certain things – and so too technical innovations – probably require certain development contexts that do not always and everywhere spring up from the ground. And then you always need people who have the ambition to work together on problems for a long time that cannot be solved overnight.

Sprachrohr: How many people make up the research team and how can one imagine the interdisciplinary collaboration? 

M. Wehner: It is not possible to quantify this precisely, as over the years, in addition to university staff, many Student Staff have also worked on OCR4all. Overall, however, a group comes together for which it certainly takes more than two hands. The question of interdisciplinary collaboration is then the much more exciting one. Because without it, OCR4all would never have existed. Basically, you have to imagine the collaboration as a colourful mixture of – in our case – literary studies requirements and the search for technical solutions from the computer science side, which constantly fertilise each other, leading to innovative developments, insights and adaptations of data and research questions on both sides.

Sprachrohr: What was your role in the development process? 

M. Wehner: With regard to interdisciplinary collaboration, one could perhaps argue that I always stood somewhat between literary studies and computer science. Although I studied German philology, at some point alongside my actual Student Staff position in the research project "Narragonien digital", for which OCR4all was co-developed, I somewhat accidentally also took on a position in computer science. At first, of course, a completely new and unknown field for me – but eventually I began to understand the technical background of the software and, in addition to intensive testing phases, could repeatedly serve through literary studies consultation. So I was co-responsible for the interdisciplinary exchange of the different research fields! And this activity continues, since I became a research associate at the AI Chair after my state examination, of course in an even more intensive way.

Sprachrohr: Since you were also involved in the research project "Narragonien digital": what role did OCR4all play in the digital processing of the moral satire "Narrenschiff" from the 15th century? 

M. Wehner: A very considerable one. Before the software existed, the various editions of the Narrenschiff had to be typed out manually – line by line, page by page, with abbreviations and special characters from the Latin, French and English of the 15th and early 16th centuries, which cannot be found on any normal keyboard.

Sprachrohr: As the name suggests, OCR4all is freely accessible to everyone (on the "GitHub" platform). How do you assess the reach of the tool in the coming years? Who would you recommend using OCR4all? 

M. Wehner: The reach has already achieved a global level today – even if it is of course still limited to very subject-specific user groups. However, we work every day to make OCR4all better and more well-known. The software is of course recommended to anyone interested in knowledge stocks that have not yet been or have not been sufficiently developed, that is to say researchers from all disciplines, but also students or private individuals who have difficulty reading digitised prints from the late Middle Ages and early modern period, but want to access the contents – and in a script that both they and the computer can easily read and further process. So anyone interested in works on botany, medicine, astrology, philosophy or social satires of this period and can only obtain scans of the works but no text files from the pages of major libraries should feel called upon!

Sprachrohr: Do you know where the tool has already been used outside the research projects of JMU? 

M. Wehner: Almost every day I am in contact with users from all over – once with researchers from Vietnam, then again with staff from a research project in Scotland. And all of them find the most diverse application texts and fields for our software.

Sprachrohr: What opportunities/access do you see for interested students to be able to participate in similar digitisation projects at JMU?

M. Wehner: Many, because there are many projects and many different tasks to be accomplished in them – and committed students are always in demand everywhere. So it is important to keep your eyes open, look for notices on noticeboards at chairs and institutes about research projects, lectures, workshops on topics, methods and strategies of digitisation in the university research field. And in the end, you have to have the courage to apply for positions, ask project leaders about vacant places and trust yourself to be able to handle tasks in the field of digitisation, DH and computer science that you have not yet dealt with and that initially have nothing to do with your own field of study. I speak from personal experience. It is always worth it!

We thank you for this detailed interview and the exciting insight behind the scenes of OCR4all. The tool was developed at the Chair of Artificial Intelligence and Applied Computer Science at Uni Würzburg and at the Centre for Philology and Digitality Kallimachos (ZPD). Via the ZPD's website (https://www.uni-wuerzburg.de/zpd/services-aktivitaeten/tools/) you can access GitHub and thus the tool. Enquiries can be directed to maximilian.wehner@uni-wuerzburg.de Maria Romanov 

Maria Romanov