Fear and Loathing on the Frontline: Decoding the Language of Othering by Russia-Ukraine War Bloggers

Abstract

Othering, the act of portraying outgroups as fundamentally different from the ingroup, often escalates into framing them as existential threats--fueling intergroup conflict and justifying exclusion and violence. These dynamics are alarmingly pervasive, spanning from the extreme historical examples of genocides against minorities in Germany and Rwanda to the ongoing violence and rhetoric targeting migrants in the US and Europe. While concepts like hate speech and fear speech have been explored in existing literature, they capture only part of this broader and more nuanced dynamic which can often be harder to detect, particularly in online speech and propaganda. To address this challenge, we introduce a novel computational framework that leverages large language models (LLMs) to quantify othering across diverse contexts, extending beyond traditional linguistic indicators of hostility. Applying the model to real-world data from Telegram war bloggers and political discussions on Gab reveals how othering escalates during conflicts, interacts with moral language, and garners significant attention, particularly during periods of crisis. Our framework, designed to offer deeper insights into othering dynamics, combines with a rapid adaptation process to provide essential tools for mitigating othering's adverse impacts on social cohesion.

Figures (15)

• Figure 1: Conceptualization of the othering process. Othering starts with the separation of ingroup and outgroup members, the creation of symbolic boundaries, and the subsequent pipeline of othering. Through the construction and affirmation of perceived threats, the outgroup is increasingly framed as a threat.
• Figure 2: Artificial Annotator Alignment process. Human-annotated data is first used to train a high-quality LLM. The HQ-LLM's annotations are compared with human annotations for alignment, and then the HQ-LLM is used to annotate a larger dataset. Finally, an open-source model is trained using the HQ-LLM-annotated data, optimizing both effectiveness and cost-efficiency.
• Figure 3: System Prompt % Gain Compared to No Additional Prompt and In-Context Learning Across Metrics (Accuracy, F1 Score, Precision, Recall).
• Figure 4: Venn diagram showing overlap between othering, fear speech, and hate speech in the Gab corpus. The diagram reveals that while fear speech and hate speech often co-occur with othering, many instances of othering occur without these explicit forms of conflict language.
• Figure 5: Temporal trends in the proportion of messages with othering language posted by (a) Russian war bloggers and (b) Ukrainian war bloggers from December 2021 to May 2023. The four classes of othering language are: Threats to Culture or Identity, Threats to Survival or Physical Security, Vilification/Villainization, and Explicit Dehumanization.