Language Models for Cultural Heritage: a Case Study in Historical Tibetan

Artificial Intelligence and Large Language Models in particular appear to rapidly change the modern world, but how can we use them to unlock the past of pre-print cultures? In this talk I'll show how and when LLMs can be useful in the process of digitising handwritten manuscripts and turning them into searchable corpora. I'll argue that when these corpora are well-annotated in particular, they are unique resources not only for linguists, but also for researchers in the fields of philology, history, literature, religion or other social sciences and humanities. Using examples from an extremely difficult body of historical Tibetan texts, I will furthermore show the limits of LLMs and where exactly in the pipeline from manuscript to answering research questions other computational methods should be considered as well.