Realist and moralist reasoning in climate politics: A RoBERTa-based classification of European Green Deal debates

This article examines how Members of the European Parliament justify their positions in EU climate politics by analysing political reasoning in legislative debates on the European Green Deal. It introduces the distinction between realist and moralist approaches to normative reasoning as an analytical framework that captures variation in climate policy justification beyond established frameworks focusing on party group alignments and roll-call behaviour. Using a fine-tuned XLM-RoBERTa-large model, the study classifies sentence-level reasoning from plenary speeches into realist and moralist categories with an accuracy of 0.93 and a macro-average F1-score of 0.86, demonstrating strong classification performance of the machine learning model. Analysis based on Shapley values provides interpretable insights into the textual features guiding these classifications. Statistical tests further show that realist and moralist reasoning are not reducible to party affiliation, region, or voting on climate legislation, with both reasoning types appearing among supporters and opponents of climate action. The article highlights the potential of machine-learning-based text analysis for studying political reasoning in legislative climate debates.