Speaking Volumes: Introducing the UNGA Speech Corpus

Abstract

Many theoretical conclusions core to the study of international politics rely on having access to, and understanding, the rhetoric of international actors. One important development in advancing the empirical study of international relations (IR) theory, therefore, is the availability of machine-analyzable speech data. A collection of fine-grained textual representations of states’ speeches in the context of an important international organization, such as the United Nations General Assembly (UNGA), is needed to understand the ideas, preferences, and values that states put forth in their statements. An especially promising use for such a corpus of texts would be to measure states’ preferences based on their statements. In an effort to add to the burgeoning field of text-as-data in IR, I present the UNGA Speech Corpus, a collection of over 34,000 speeches delivered by states in the UNGA from 1993 to 2018. I use it to improve on recent work that links text to preferences in IR by combining a structural topic model with locally trained word embeddings to estimate the policy positions of states on specific topics. I then show how these “topic scores” can help scholars to improve their analyses of exigent international issues, such as global climate change governance.