From Normalisation to Understanding: AI and the study of Old Norse

Tarrin Wills & Ellert Þór Jóhannsson.

Tuesday, September 8, 2026, at 16.30

Edda auditorium (E-103)

Old Norse texts are preserved in a variety of forms, and the relationship between original manuscript spelling, diplomatic editions, and normalised texts has long been an important issue in the study and teaching of the language. Normalised editions make medieval texts accessible to modern readers, while diplomatic editions preserve information about the manuscripts and provide the textual basis for much historical and lexicographical research. This tension is particularly visible in A Dictionary of Old Norse Prose (ONP), whose extensive citation material is based largely on unnormalised scholarly editions and can therefore be difficult for non-specialists to access.

The problem of normalisation is not new, and earlier attempts to automate it using computational and machine-learning methods have shown both its possibilities and its limitations. Recent developments in Large Language Models, however, have changed the picture considerably. In a pilot study we tested several LLMs on the normalisation of Old Norse material from the ONP database. The results demonstrate a substantial improvement in performance and suggest that current models can produce highly accurate normalisations of a wide range of texts.

This raises a broader question: what becomes possible if AI systems can reliably process Old Norse not only at the level of spelling, but also at the level of form and meaning? We consider potential applications including semantic searching, matching parallel passages, and identifying relationships between Old Norse texts and their translations. Such possibilities could open new ways of exploring a large and low-resourced textual heritage corpus.

At the same time, these developments raise important questions for Old Norse scholarship. What role will human expertise and dictionary editors have when AI can perform increasingly sophisticated linguistic tasks? What are the environmental costs of relying on such systems? And, perhaps most importantly, if medieval texts become accessible without learning the language itself, what will be the future role of learning Old Norse?

Tarrin Wills is a Senior Researcher and head of the Dictionary of Old Norse Prose at the University of Copenhagen. He has led and co-led digital editing and analysis projects in Old Norse and related fields including the Skaldic Project, Lexicon Poeticum and EAE.

Ellert Þór Jóhannsson is an Assistant Research Professor at the Árni Magnússon Institute for Icelandic Studies in Reykjavík. His research focuses on Icelandic lexical studies, lexicography and the history of the Icelandic lexis. From 2012 to 2021 he was one of the editors of the Dictionary of Old Norse Prose.

The talk will be delivered in English and is open to all.