Political emotionalization in large text corpora: detecting affective cues with MORESPulse and its scientific uses

This paper presents MORESPulse, a freely available, theory-grounded, and expert-validated multilingual large language model based on XLM-RoBERTa that detects the textual manifestation of five primary emotions (anger, fear, sadness, joy, and disgust) in political texts across seven European languages. We illustrate its scientific potential through a longitudinal analysis of the Hungarian Parliamentary Corpus from 1998 to 2022, covering more than five million sentences across six electoral cycles. Emotional language has increased steadily since 2010, driven primarily by opposition Members of Parliament (MPs) and smaller party representatives, while government MPs show a comparatively stable and more restrained emotional tone. Anger and joy are the most prevalent emotions across the corpus, while disgust remains negligible. Emotional intensity also varies systematically by policy topic: macroeconomics, immigration, and health generate higher levels of emotional language, while foreign trade consistently produces the lowest. These findings suggest that emotionalization is a consistent and enduring feature of Hungarian parliamentary discourse rather than an episodic rhetorical response, with implications for how scholars understand affective dynamics in democratizing and backsliding political contexts.