Testimonies from Holocaust survivors are essential for Holocaust research, as they document events and perspectives that are not mentioned in official documents or reports by the perpetrators or are only mentioned in a distorted way. They also help to overcome the perception of passive victims and explore decision-making processes and agency. Additionally, they serve as a central component of remembrance culture and help keep memories alive for future generations. This is particularly important considering the much-discussed end of contemporary witness testimony.

Recently, in historical education and research, there has been a debate about the importance of emotion-based or emotion-historical approaches to texts in general as well as to memories of the Holocaust. Methods from computational linguistics and AI-supported transcription methods for the automated indexing of source material are established methods in digital history. But can these methods lead to false results and distortions?

A first test study on sentiment analysis of the Theresienstadt-diaries of Martha Glass classified them as positive, as many terms were interpreted as positive. This serves as the starting point for the analysis presented in this contribution. The central objectives are: 1) to answer the question of the extent to which the automated evaluation and analysis of biographical memories of the Holocaust can lead to false positive perceptions, 2) to develop a methodology to approach, examine, and understand possible misrepresentations, 3) to contribute to the discussion of the purposes for which and the ways in which AI can be used to answer historical questions in a source-critical way.