Named Entity Linking for the longitudinal study of the parliamentary discourse in the German Bundestag

Parliamentarians have various reasons to explicitly refer to actors, organizations or locations in legislative speeches. They might refer to their constituencies to emphasize their "geographic representation" (Zittel et al. 2019), refer explicitly to interest groups as resources for their arguments (Fraussen et al. 2018: 455) or refer to entities in the public discourse trying to influence the legislative process (Melenhorst 2015). In consequence, analyzing these references can shed light on issues of agenda-setting, party competition, representation and beyond. Since the amount of data which can be processed manually is limited, advances in the field of Natural Language Processing might provide an avenue towards automating such tasks. In this regard, Named Entity Linking is a particularly promising technique. It facilitates the identification, disambiguation and classification of the mentions of real-world entities in textual data. However, developing a methodologically robust workflow for large collections of legislative textual data is still no easy endeavor, in particular concerning usability, reliability and validity in the context of social science research. The proposed contribution will follow up on these considerations and showcase possibilities and limitations of Named Entity Linking for the study of legislative speech. As such, it also serves as an example case for the toolset developed by the authors of this proposal in the project "Linking Textual Data", namely the R packages "LinkTools" and "dbpedia". With this toolset we set out to make approaches of Named Entity Linking – currently DBpedia Spotlight (Mendes et al. 2011) in particular – more accessible in social science applications, thereby broadening the potential audience for an approach only used rarely (for example Glaser et al. 2022) in substantive analyses in the field until now. As a substantive example, this contribution will focus on how explicit references to actors or organizations are used strategically in parliament. Drawing on the notion of the "mediatization of politics" (Van Santen et al. 2015: 45-46) and related approaches, parliamentary actors might be incentivized to refer to the public discourse by mentioning such entities to aid their political goals (Melenhorst 2015). How this is actually used in parliament might vary between parliamentary groups and over time, for example due to dynamic opportunity structures in the media arena (Van Aelst and Walgrave 2016: 509). If so, this should reflect in the number and composition of actors and organizations referred to by politicians in parliament. Analytically, empirical insights could be combined with survey data (popularity measures, public issue salience) to broaden the understanding of such potential strategic opportunities. The example uses an updated version of the GermaParl2 corpus of plenary debates in the German Bundestag (Blätte and Leonhardt 2023) which has been enhanced with disambiguated entity annotations and facilitates longitudinal analyses from 1949 to 2023. Although it is used for German debates here, the proposed workflow can be applied to other languages as well. One core aspect of our contribution is that we offer tools for a generic workflow for the approach of Named Entity Linking relevant for parliamentary data and applicable for many research questions.