Propaganda Across Text Types in Russian Wartime Narratives

Online disinformation surrounding the Russo-Ukrainian War illustrates the need for theoretically grounded and context-sensitive NLP approaches to information disorder. This paper investigates linguistic variation in propagandistic narratives about the Russo-Ukrainian War across state media, social media, and Wikipedia versions. Using Kullback–Leibler Divergence (KLD), an interpretable measure of distributional divergence, we identify distinctive lexical features across platforms in two resources: the Wartime Media Monitor corpus and a large-scale dataset of edits made to the original Russian Wikipedia to create a Russian Wikipedia Fork (RWFork), its alternative Kremlin-compliant version. Our findings show systematic cross-platform differences. Specifically, state media favors a demobilizational propaganda strategy aimed at pacifying the population, whereas social media attempts to mobilize its users. RWFork mirrors state-aligned lexical framing through the replacement of direct war terminology with euphemistic expressions and references to geopolitical control.