Open Discourse: Towards the first fully Comprehensive and Annotated Corpus of the Parliamentary Protocols of the German Bundestag

Abstract

Open Discourse is the first temporally all-embracing and comprehensive project, which processes every word ever spoken in the parliamentary sessions of the German Bundestag in a machine-readable way. This paper serves as an introduction to the database, making a connection to existing national and international parliamentary text corpora projects while showing the versatile applicability for political research. The data collection and processing of Open Discourse implies a maintainable and coherent data structure that comes with a lot of contextual political information and is very promising for the intersection of modern quantitative text analysis and differentiated political science research plans. By including every legislative session, almost 900,000 speeches are included in thedatabase, ranging from the inaugural session in 1949 to this year (2022) in the ongoing 20th legislative period.

Uncovering the Secret Life of We-Pronouns in the German Parliament

“…We used a newly available corpus of speeches held in the German federal parliament (“Deutscher Bundestag”) from its establishment in 1949 to 2021 (Richter et al, 2023, opendiscouse.de). From digitalized parliamentary protocols, the corpus extracted individual speeches and their speakers.…”

“…Speech time , the total word count in each term (also left-skewed and logarithmized), differentiated frontbenchers and backbenchers, assuming that frontbenchers have more speech time compared to backbenchers. In addition, based on the speech corpus (Richter et al, 2023), we added the variables party (levels: “Linke” ∼ left, “SPD” ∼ socialdemocratic, “Grüne” ∼ green, “FDP” ∼ liberal, “CDU” ∼ conservative, “AfD” ∼ far-right), power (levels: government party, opposition party), age , gender (1 = female , 0 = male ), and academic title (1 = Dr title , 0 = none ; see , Appendix C, Table C1).…”

Computational scaling of political positions from textual data using word embeddings

“…I conducted computational experiments to show that the automatic scaling method proposed works in principle. Using speeches from the 19 th German Bundestag (Richter et al, 2020), I devise four dictionaries to produce two political issue axes in a Word2Vec embedding space. Please note that the integration with transformers has not happened yet; for now, I used static word embeddings because they are faster to train but still sufficient to get a first impression of whether the method can work in principle.…”

BUNDESTAG-MINE: Natural Language Processing for Extracting Key Information from Government Documents

“…Another tool for political discourse analysis is [4], who assign specific words to 30 different semantic classes. Furthermore, [5,14] have analyzed parliamentary procedures for the German parliament since 1949. SentiArt [10,11] analyze German political party programs, considering their semantic similarity and complexity, their main themes, the emotion potential and their readability.…”