AI Helps Researchers Study Old Russian Writing
Researchers at Yaroslav-the-Wise Novgorod State University (NovSU) used artificial intelligence and computer vision to study the evolution of ancient Russian writing. A neural network analyzed both the content of the texts and the graphical features of individual letters – their shape, slant and proportions.
The AI read and analyzed 1,249 birch-bark documents and around 27,000 individual letters. The neural network then assessed changes in writing and confirmed that the transition between different types of Old Russian script happened gradually rather than as a single abrupt shift.
What Did the AI Measure?
The experiments were initiated by Ivan Filippov, an assistant professor in NovSU's Department of Information Technologies and Systems. He wanted to answer a question: could changes in the language of medieval Novgorodians be detected not through a handful of striking examples, but by analyzing all the texts? The study drew on several independent layers of data, ranging from complete transcriptions to a rigorously selected set of 434 documents with stratigraphic, or archaeological, dates.
The algorithm examined the texts, looking at the distribution of letters, spelling variants and combinations within words. The neural network then analyzed the images themselves, measuring the proportions, slant, angularity and stroke complexity of each letter carved into the birch bark.
“The project's technical implementation is a cascaded AI system developed specifically for paleography. The first stage is character recognition. We initially used a basic classification model whose job was to distinguish one letter from another in a scan or photograph of a birch-bark fragment. The second stage is the analysis of intraclass drift. Once the neural network recognizes that it is looking at, say, the letter ‘A’ or ‘V,’ a second model takes over. It looks for differences within the same class by evaluating the specific features of each individual character,” Filippov said, describing the method.
The Neural Network Backs Up Researchers' Findings
AI allowed the researchers to refine the picture of how Old Russian writing developed. Linguist Andrey Zaliznyak devised a classification of writing samples based on the evolution of letterforms. Type B2 represents earlier forms, predominantly from the 11th and 12th centuries, while Type V marks the transition to 13th-century writing, when letters became broader and straighter and lost many archaic features.
Filippov says his work translates this philological model into quantitative data: the algorithm confirmed Zaliznyak's conclusions about the gradual nature of the changes, while making it possible for the first time to quantify the breadth of the transition across the entire corpus of texts.
“It turned out that the transformation of scribal habits stretched over more than a century and a half, with the most intense period of change occurring between 1190 and 1250. The changes affected neither vocabulary nor the conventional structure of the letters, but the very foundations of the writing system – the system of vowels and reduced vowels. Scribes gradually changed the rules they used to select characters within words,” Filippov said.
AI in Fundamental Research
In recent years, artificial intelligence has increasingly been used in fundamental research and the study of cultural heritage. It makes it possible to analyze large collections of historical data and identify patterns that are difficult to detect using traditional methods.
NovSU has already used AI to study Old Russian writing. In 2025, the university introduced a system for automatically recognizing and interpreting Old Russian birch-bark documents. The technology assists paleographers and archivists. The new study, in which researchers again turned to neural networks, continues the search for effective modern tools for investigating the past.
Neural networks can already be used to automatically recognize manuscripts, classify documents, identify patterns in large historical corpora and assist experts in dating finds. In the future, similar technologies could be used by national and regional archives and museums, including to create digital collections of historical documents.
