Deciphering Diplomatic Speech in the EU using Large Language Models
P4-S94-2
Presented by: Jens Wäckerle
International negotiation outcomes have far-reaching consequences and can teach us about elite interaction as well as the effect of domestic incentives on government behavior. While these negotiations typically take place behind closed doors, the availability of speech corpora of international negotiations in the European Union as well as the United Nations have opened this black box in recent years. Extracting meaningful information about negotiation tactics, such as delaying agreements, raising demands, making threats, or seeking compromise, from such corpora, is however, inherently difficult. First, the use of diplomatic language allows speakers to politely express hard stances. Second, it is often unclear which verbal expressions exactly serve as an indicator of particular negotiation strategies. Third, diplomatic rhetorical styles are moreover context-dependent and vary across different international negotiation arenas, thus rendering measurement difficult. We characterize the potential variation in negotiation tactics based on the literature and implement an LLM-based approach that overcomes the measurement challenges by leveraging case-knowledge of the international negotiation arena. We demonstrate our approach using a corpus of speeches in the Council of the European Union.
Keywords: Council of the European Union, Quantitative Text Analysis, Large Language Models, Negotiation Tactics