Research into manuscript fragments has expanded enormously over the past twenty years, and recent studies of fragments — whether those tightly bound in books or excised leaves collected by scholars and connoisseurs over the centuries — have opened up rich new perspectives onto ancient and medieval texts, manuscript production, and the history of writing. For modern eyes, whose instincts are trained for cultural preservation, medieval codices were almost ruthlessly recycled in the premodern period. Pages were cut up and reused not only in book bindings but also as support linings in many other objects, such as musical instruments. It is striking how indiscriminate medieval scribes and craftspeople were in their re-use: they used any kind of codex — be it a school text, a bible, a legal codex, or a liturgical book — without appearing to attend to the content it carried.
The digital turn in manuscript studies has brought new possibilities for the study of fragments. Fragments belonging to a single codex have been reunited in a virtual format, their disparate remains gathered and displayed in a single location. Fragments uncovered in unexpected places — the inside of a violin case, the back of keyboard keys — can be studied with an endoscope without dismantling the historic object preserving them. The digital turn now brings a new possibility: scientifically reconstructing the lost page by using the visual evidence of extant fragments.
Over the next two years, a project led by Dr. Anna Dorofeeva at the Institute for Digital Humanities in Göttingen will develop an app for reconstructing lost manuscript pages from their fragmentary remains. With generous funding from the State of Lower Saxony, the project “Fragment Autocomplete” will move into the conjectural space of the fragment’s missing page. Using artificial intelligence, the team, consisting of Mo Basuony and Dr. Jeremy Thompson, will design and train a model to analyze a manuscript fragment and to make substantive hypothetical inferences about the lost page from which it comes. This project builds on the success of the Weltwissen initiative (2021–2027) financed by the Volkswagen Foundation for advancing the disciplines of digital paleography and imaging science at the University of Göttingen.
Manuscript scholars have of course long tried to visualize missing pages. They have reconstructed, sometimes physically, often mentally, the models on which surviving codices depend. In the cases of illegible and damaged texts, even where only a few letters are decipherable, they have exposed the manuscript to different wavelengths of light and profited from every trace of informative data: every ink trace; every smudge; and, now in our digital age, every pixel. “Fragment Autocomplete” continues in this tradition, endeavouring to produce a tool that will help manuscript researchers digitally visualize the fruit of these efforts.
Training a model will entail the study of a vast corpus of fragments and manuscripts with an eye to critical data about mise-en-page. This work will be supported by Fenius, an app developed in the Institute for Digital Humanities that allows researchers to trace page layouts on top of a digital image and record information about the disposition of lines, text, images, and columns. The challenges are considerable. The team will have to consider diverse script styles, initials, illustrations, diagrams, and musical staves. The team will work by inpainting — feeding an AI model an image and training it to expand that image into a complete one. Besides local manuscript holdings at the State and University Library in Göttingen, the project will draw heavily from the significant corpus of fragments curated by Dr. William Duba at Fragmentarium, with which “Fragment Autocomplete” will closely collaborate. To optimize model training, the team will also create their own fragments in a digital environment by cropping images of complete pages and using them to train a model to accurately reconstruct the missing part. At the end, the AI-model will be made over to Fragmentarium for full implementation on its own platform. If successful, the tool will offer manuscript scholars something they have long had to do without: a reliable method for reasoning systematically about what is no longer there. With it, the lost page need not stay lost.
If you have questions or wish to collaborate, please contact the team.