Strategic Signaling and U.S. Public Statements in Secessionist Conflicts

What shapes the United States’ decision to issue public statements on secessionist conflicts, and how can large language models (LLMs) transform the study of such diplomatic signaling? This paper investigates the conditions under which the United States, as a major power, strategically deploys verbal cues as instruments of foreign policy. It theorizes that public statements about secessionist movements are driven by three core factors: (1) alliance and rivalry relations, (2) economic ties, and (3) conflict characteristics. Using secessionist conflicts between 1945 and 2020, the study applies LLMs to extract, classify, and scale resolve in U.S. conflict-related statements. As a methodological contribution to the emerging best-practice debate, the paper systematically benchmarks domain-specific models (ConfliBERT and ConfliLLAMA) against general-purpose BERT and LLaMA models to evaluate their performance in political text classification and resolve detection. The empirical analysis then links LLM-generated measures to the analysis on the possible factors which shape the diplomatic signaling preferences. Initial findings indicate that security relations, media salience, and economic stakes play a decisive role in shaping U.S. public rhetoric. Beyond its substantive results, the paper reflects on how LLM-driven measurement reshapes data construction, inference, and knowledge production in the study of international conflict.