Cultural Memory from Antagonism to Deliberation in (Social) Media: AI Approach

Cultural memory refers to the shared pool of knowledge, traditions, and collective experiences that are passed down through generations within a community or society. It shapes a group's identity, provides a sense of continuity across generations, and ensures socio-political stability. However, when memories of significant historical events are contested or divisive, they can fuel antagonism and polarization within or between communities. In this paper, we present a comprehensive exploratory computational approach for measuring and analyzing social media’s impact on the transformation of cultural memory. To investigate how commemorative narratives are being transformed with the rise of social media, we developed a large language model (LLM)-supported framework, theoretically anchored in the deliberative democracy. To identify narratives associated with specific commemorations, we use Greimas' structuralist actantial model, an analytical tool for understanding the roles and relationships between narrative functions (actants) in discourse. Computationally extracted narrative actants are cross-analyzed on a scale of antagonism, agonism, and deliberation and evaluated within a discourse functions analysis (epistemic, ethical, democratic, and ideological), adapted from the deliberative democracy theoretical framework. The approach is, to our knowledge, one of the first human-evaluated attempt to check validity of automated actant extraction and then demonstrated using social media and newspaper articles about the Slovenia’s commemoration of the Day of Resistance against Occupier as a case study. The results provide valuable insights into how different segments of the narrative in social and traditional media reinforce antagonism, foster agonistic debate, or create opportunities for deliberation. Special attention is given to identifying points of conflict and establishing common ground through computational analysis. We underscore the methodological novelties potentially leading to broader applicability of the demonstrated approaches, emphasizing the possibility to develop an automated pipeline for analyzing debates on divisive collective memories in different contexts, as well as for assessing the quality of other types of discourse.