A Quantitative Approach to Nationalist Discourse in 20th Century Germany

Abstract

How can we measure the expression of nationalist ideology in text? What determines the use of nationalist rhetoric by political elites? How are nationalist narratives and political violence connected? This dissertation investigates these questions in the context of 20th century Germany by applying new approaches from the realms of natural language processing and quantitative text analysis. The first part of this work describes the methodological toolbox applied in the empirical chapters, in particular the use of word embeddings and transformer models. It then demonstrates a system based on a pre-trained large language model that successfully improves historical texts by reducing errors due to optical character recognition inaccuracies. The second part of the dissertation investigates elements of nationalist ideology and its expression by the political elite of Germany. I show that the experience of the First World War led veterans in the Reichstag debates to engage in more war-related discourse. Turning to political dynamics outside the parliament, I compile a new geocoded data set of violent events during the final years of the Weimar Republic. These data are used to find suggestive evidence that propaganda speeches by Adolf Hitler increased local levels of violence in the spatio-temporal vicinity. The last chapter applies a word embedding technique to measure discourse around the abstract concept of a German homeland, and how it is expressed in debates in the German Bundestag. I find that Members of Parliament who experienced the expulsion after the end of World War II frame the homeland issue differently from their colleagues. In addition, I investigate whether the share of German refugees in an electoral district influences how parliamentarians address the concept of the German homeland.