Afternoons with DARIAH
On Tuesday, 3 November 2026, at 1:00 p.m., you are invited to a lecture by Sarah Oberbichler from the Centre for Modern and Digital History (C²DH) at the University of Luxembourg. The lecture, titled “LLMs for In-Depth Analysis: Where Are the Limits of Scaling?”, will take place in the reading room of the Institute of Contemporary History, Privoz 11, Ljubljana.
Lecture summary
Large language models promise to close the long-standing gap between quantitative and qualitative methods. They could let humanities researchers ask “why” and “how” questions across corpora far too large for close reading. This talk asks how far that promise holds, using a transnational study of how states shaped public reality after the catastrophic 1908 Messina earthquake, based on historical newspapers. Drawing on an LLM-based argument-mining pipeline, it shows that prompt-based methods scale easily but are epistemically fragile, while fine-tuning and reinforcement learning offer greater rigour but are bounded by the need for representative training data. Source criticism remains a key gap. The talk concludes that case studies are not made obsolete; they become essential for validation, which requires integrating human expertise and machine processing throughout the research process.
Lecturer
Sarah Oberbichler is a historian and researcher specialising in artificial intelligence and data mining in digital history at the Luxembourg Centre for Contemporary and Digital History (C²DH), University of Luxembourg. Her interdisciplinary research lies at the intersection of historical migration studies, environmental studies, transnational media studies, NLP, and (critical) AI studies. She is the author of two monographs and more than 30 academic articles and has given over over 50 invited talks and keynote lectures worldwide. She regularly contributes to public debates on the role of digital humanities and historiography in contemporary society.
