Collective memories shape political narratives but are difficult to measure. Existing approaches largely rely on surveys, media, or textbooks, which face issues of scalability, accessibility, or representativeness. This paper introduces a framework that conceptualizes collective memories as context-specific group-role associations and leverages natural language processing tools and Wikipedia revision histories to measure their content and change over time. Focusing on the case of WWII and Holocaust memory in Poland, we provide evidence that our approach allows us to capture collective memory: In line with existing literature on contemporary Polish collective memory of WWII, WWII collective memory as estimated using Wikipedia data represents Poles as the main victims of the period, and Germans as the main perpetrators. This representation has been relatively stable over the past 20 years. These findings have important implications for empirical studies of collective memory. Firstly, they suggest that we can use Wikipedia data to capture relevant group-role associations that describe the content of collective memories. Secondly, they suggest that we can use Wikipedia to study collective memory content of different socio-linguistic groups over time.