Immigration and the securitization of what? A semi-supervised approach for multi-class classification of debates in the German Bundestag 1990-2020

There is a broad set of research questions that require a thematic classification of parliamentary speeches according to a pre-established scheme. In our paper, we suggest an approach that moves beyond simple dictionary approaches and pure unsupervised topic modelling to strengthen validity and reproducibility. The semi-supervised approach we suggest combines (a) a matching step to classify (optimized) topic models in a reproducible manner by using features of hand-coded data and word embeddings as an intermediary, and (b) an aggregation step to group topics into a set of conceptual categories, the categorization scheme of the Comparative Agendas Project (CAP) in our case. The methodological concern to improve the classification of parliamentary debates takes the securitization of migration in parliamentary debates in Germany as a case: The 2015 refugee situation evoked many different framings of migration in the political domain – from humanitarian aspects to portraying migration as a security problem threatening different aspects of society and the state. This means parliamentarians are securitizing migration across different domains ranging from for example migration as a threat to culture and physical well-being to the social security system. According to securitization theory, “securitization” may justify emergency measures that are not acceptable under normal circumstances by claiming that a particular actor or event poses a threat to the survival of a certain object or actor. Securitization is thus an “extreme form of politicization” (Buzan, Wæver & De Wilde 1998, 23), with actual consequences for groups of people. As parliaments are the place where elected representatives discuss these emergency measures in liberal democracies, we deliberately depart from the focus of securitization research on media coverage and put parliamentary debates center-stage. We also move beyond qualitative methods and small-N studies that are predominant in securitization research. The securitization of migration is not new (Best 2010). So we investigate securitization across time: How did the securitization of migration in parliamentary speeches of the German Bundestag develop across time and what are the domains of securitization? The data basis for our research is “Unity Edition” of the GermaParl corpus that covers all parliamentary debates from 1990 up to 2020. It has been prepared in the context of the PolMine Project using a Framework for Parsing Plenary Protocols (frappp). Another resource are (manual) annotations of plenary speeches according to the coding scheme of the Comparative Agendas Project (CAP). These can serve as human input for the classification task we seek to optimize. The matching of topic models and hand-coded data with word embeddings as an intermediary is implemented in the R package MapCAP (not yet published). To perform the aggregation of topics into conceptual categories, the R package topicanalysis (https://github.com/PolMine/topicanalysis - experimental at this stage) will be developed further. The general background of this work is the PolMine Project (polmine.de).