Expertise and Politicization in COVID discourse

• Applicants: Professor Dr. Sebastian Haunss; Professor Dr. Sebastian Padó
• Subject Area: Applied Linguistics, Computational Linguistics ; Empirical Social Research ; Political Science
• Term: since 2024
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 550916093

Public discourse on specific topics takes place in a multidimensional linguistic space. Two central dimensions are (a) how expert-specific vs. generally understandable the discussions are; and (b) how politicized these discussions are, i.e. how prominent political as opposed to subject-specific aspects are. These two dimensions have a major impact on the general perception of such discourses in society. This situation gives rise to a number of research questions (RQs) at the interface between political science and (computational) linguistics, namely: (RQ1) What are the linguistic means by which the two dimensions are expressed? How can these two dimensions be measured computationally in such a way that the measurements are genre- and topic-unspecific and also as reliable as possible for short inputs (e.g. single sentences)? (RQ2) How dynamic are these two dimensions over time within specific established forums and across different forums? (RQ3) How consistent is the behavior of individual discourse participants in multi-logical communication? Are variations in behavior related to strategic interests of these participants? With this project, we propose to analyze public communication in Germany on one of the most important topics of recent years, namely the COVID pandemic, from these perspectives. We first look at two canonical "official" sources of pandemic information, viz. the bulletins of the Robert Koch Institute and the Bundesgesundheitsblatt. These provide a good entry point as they are clearly located on the expert side. Our hypothesis is that the bulletins become more politicized over time, whereas this is not the case for the Bundesgesundheitsblatt (RQ2). To this end, we annotate COVID policy-related political statements and justifications in the text. We then move on to political discourses from different forums and compare debates in the Bundestag plenary with discussions in committees, especially the Health Committee. Our hypothesis is that this communication is generally politicized, but is more expert-oriented in the committees (RQ2). At the same time, we assume that individual discussion participants show clear differences with regard to both the degree of politicization and the degree of expertness and will analyze these with regard to the backgrounds of the participants (RQ3). In order to scale these analyses to the large amounts of text at hand, we develop computational linguistic models to measure the two dimensions for our analyses. We compare different approaches for models with respect to their basic suitability, to specific criteria such as their fairness and explainability, and to their ability to work well on short texts, which is important in order to be able to reliably analyze the contributions of individual speakers (RQ1/3).

• DFG Programme: Research Grants