More Than Words: The Contestation of Social Normality and the Problem of Regime Types in International Relations

Do authoritarian states really have a separate conception of rules and normativity compared to democracies (Ginsburg 2020)? Or are undemocratic regimes unable to propose a radically different normative vision because they operate under hegemonic conditions within the international order? Did the normative vision of autocracies change and adapt over time since the end of the Cold War? In the following paper, I will use text-mining techniques to analyze if speech production in the UN actually highlights structural differences between autocracies and democracies in the way regimes conduct their political affairs in this forum. I will work on two large corpora: the UNGA (United Nations General Assembly) General Debate corpus (Jenkins and Oth, 2023) and the UNSC (United Nations Security Council) corpus (Schönfeld and Oth, 2020). The paper positions itself within the context of the debate on the ongoing contestation of the International Liberal Order (LIO) (Lake et al., 2021, Ikenberry, 2018, Jahn, 2018). It tries, however, to shed some light on the nature and extent of this contestation in relation to political speech and liberal normativity within political debate. In the paper, I will try to investigate if substantial semantic differences emerge in relation to democratic levels and regime types in a period of time that goes between 1992 and 2022, highlighting structural changes that happened over this time period. To do that, I will use three main models. Quasi-Poisson regression: the model regresses certain features of interest that have to do with liberal normativity (Democracy, International law, Human rights, etc.) on the V-Dem liberal democracy score. I also include an offset that takes into account different lengths of the speeches in my datasets. The main idea of the model is to investigate if the use of liberal and democratic semantics also is a predictor of democracy levels. Structural topic modeling: In this stage, I use a topic model algorithm (STM package) on the speeches of the USA and Russia. I created a dummy variable that codes speeches in my corpora that have to do with political crises. I choose the war on terror for the US and the Ukrainian crisis for Russia. The idea of the model is to investigate if speeches that have to do with political crises have a larger or a lower distribution of normativity-related topics. ALC embedding: Finally, I propose three ALC embedding models to investigate the distributional differences in the semantics of my key target words (Namely: Democracy, Law, and Sovereignty) which have to do with liberal normativity. In this model, I create a semantic difference score for democracy and autocracy making possible a comparative analysis of the speeches between 1992 and 2022