Sentiment analysis between gaining knowledge and falsification?

First-hand certificates are important sources on historical events and provide insights into individual lives and experiences. Certificates from those persecuted in the Holocaust are essential for research, as they document events and perspectives that are absent from, or distorted in, official documents or perpetrators’ accounts. They are also a central component of remembrance culture. Autobiographical testimonies are taking on particular significance, especially in light of the much-discussed end of eyewitness accounts and the recent focus on the search for new forms and points of reference.

As AI has already been used for several years to facilitate interviews with eyewitnesses, the question of the influence and impact of AI is increasingly arising with regard to the analysis of written personal testimonies. AI-supported transcription methods for the automated processing of audio or video source material are already partially established and make it possible to prepare thousands of hours of interviews, which lie dormant in archives and memorial sites, for research and analysis. Computer-assisted, and above all computational linguistic, methods of natural language processing (NLP) promise opportunities for in-depth research into individual and collective experiences, interpretations and memories. Generative AI, moreover, presents new challenges in terms of the low-threshold yet opaque nature of text analysis and production. Against this backdrop, a critical reflection on machine-based analytical methods with regard to the evaluation and interpretation of personal accounts is imperative.

The aim of this project, funded by NFDI4Memory as part of the Incubator Funds 2026, is to contribute to a critical reflection on the functioning and application scenarios of AI-supported analytical methods, particularly in relation to the evaluation and interpretation of personal accounts of the Holocaust and its aftermath, against the backdrop of the growing significance of such methods. The project examines the potential of existing tools and machine-based methods using the example of sentiment analysis, as well as emotion-based and emotion-historical approaches that are discussed in both educational outreach and research. Using the diaries of Theresienstadt survivor Martha Glass, lexicon-based approaches as well as machine learning methods are tested and evaluated to open up new avenues for historical research. In doing so, the project provides important impetus for digital source criticism and the reflective use of methods and tools when working with personal accounts. The project’s methodologies and findings are also being adapted for educational purposes in teaching and outreach.