AI framework analyzes 1,600+ Holocaust testimonies, challenges structured vs. free-form dichotomy

Researchers at Hebrew University have published a paper on arXiv introducing a scalable AI framework for comparing oral history archives. The team applied discourse segmentation, topic modeling, and large language models (LLMs) to analyze over 1,600 Holocaust survivor testimonies from two major collections: the USC Shoah Foundation (structured, interviewer-guided) and the Yale Fortunoff Video Archive (free-form, open-ended). Their goal was to quantitatively test a long-held distinction in Holocaust studies.

Their analysis generally corroborated the structural differences—Shoah testimonies showed higher topic coherence and more interviewer-driven question patterns, while Fortunoff testimonies had more survivor-led narrative flow. However, they also found significant overlaps, both within individual interviews (e.g., structured sections within a free-form testimony) and across common narrative arcs. This complicates the simple 'structured vs. free-form' dichotomy. The framework is scalable and replicable, promising broader applications for digital oral history, narrative analysis, and citizen-science annotation platforms.