Computer scientists at FAU receive funding of approximately 300,000 euros for two new projects
When deciphering historic documents, even experts can come up against their limits – for instance if ancient wooden tablets are charred, scrolls have decomposed or notes have been taken in a rare type of shorthand. In two new projects, computer scientists from Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU) aim to develop AI-assisted technologies to make it easier to decipher sources such as these. The Schmidt Sciences Humanities and AI Virtual Institute (HAVI) is providing approximately 770,000 euros in funding for the project over the next three years, part of which will be allocated to FAU.
Project 1: Deciphering the lost art of shorthand
Particularly in the 19th and early 20th century, many manuscripts were written in the rare Gabelsberger form of shorthand. This includes manuscripts from the mathematician and philosopher Kurt Gödel, the physicist Erwin Schrödinger, the legal expert and politician Carl Schmitt, the author Erich Kästner and the Archbishop of Munich Cardinal Faulhaber. “The manuscripts and diaries include material of historical and intellectual relevance. However, nowadays only very few people are able to read this writing, meaning that large quantities of archived material are virtually unaccessible,” explains PI Dr. Vincent Christlein from the Pattern Recognition Lab at FAU. Together with Nikolaus Weichselbaumer from the University of Mainz, Christlein hopes to train and develop AI-assisted handwriting recognition to be able to make shorthand writing machine-readable as well.
Project 2: Deciphering text in damaged manuscripts
Ancient manuscripts are often illegible after having been badly damaged. One well-known example are the papyrus scrolls and wooden tablets from Herculnaeum near Pompeii which survived the eruption of Mount Vesuvius in the year AD 79 but are badly charred. “Modern imaging procedures like multispectral imaging or x-ray computed tomography make remnants of ink more visible and therefore more legible. The main problem is that there are holes and rips in the carrier material and that individual papyrus layers are often unable to be separated from each other without causing even more damage,” explains Vincent Christlein. This is also the case with the Manichaean papyrus codices found in Medinet Madi in Egypt, which are also included in the project. The researchers from Erlangen, the USA, and Italy are working on an AI tool that not only works on the level of text recognition, but also diagnoses damage and makes suggestions for the content of missing passages.
Developing AI with and for the humanities
“We have been working from the outset with experts from the humanities, who have given us feedback on the training data,” explains Christlein. “This allows us to develop valid evaluation criteria that link methods from computer science with editing practices common in the humanities.” First of all, the AI models are trained using existing sources from shorthand and papyrus libraries. The aim of both projects, however, is to create universal tools that decipher largely inaccessible sources and thereby make a valuable contribution to research in the humanities.
Schmidt Sciences Humanities and AI Virtual Institute (HAVI) Award 2025
Both projects are being financed within the framework of the Humanities and AI Virtual Institute (HAVI) Award 2025. With the award, the Schmidt Sciences foundation is funding research teams from across the globe who are exploring new avenues for encouraging dialog between artificial intelligence and the humanities, from archeology and art history to literature, linguistics, and film studies. The aim of the funding program is to unlock historical manuscripts using AI, but at the same time to use humanistic questions, methods and values to drive forward the design and use of AI.