Research Summary: LLM Analysis of 150+ years of German Parliamentary Debates on Migration Reveals Shift from Post-War Solidarity to Anti-Solidarity in the Last Decade

Recent research utilized Large Language Models (LLMs) to analyze over 150 years of German parliamentary debates on migration. This analysis reveals a notable shift in discourse, moving from post-war solidarity towards anti-solidarity sentiment in the last decade. The study also validates the effectiveness of LLMs, specifically models like GPT-5, in providing scalable and accurate annotation for complex political discourse, affirming their utility for historical and contemporary data analysis.

Why it matters

This research provides a technological approach to understanding long-term shifts in political discourse, which can inform strategic planning and policy development. The proven capability of LLMs to analyze extensive historical data efficiently offers a new tool for monitoring evolving societal sentiments and their implications for governance and public opinion.

Key insights

• Analysis of German parliamentary debates on migration indicates a shift from post-war solidarity to anti-solidarity discourse over the last decade.
• Traditional analysis of political discourse has been limited by the need for extensive manual annotation.
• Large Language Models (LLMs) offer a scalable and effective alternative for annotating political discourse.
• The research employed a theory-driven annotation scheme to evaluate LLM performance on subtypes of solidarity and anti-solidarity.
• Evaluations considered various LLM aspects, including model size, prompting strategies, fine-tuning, and data types (historical vs. contemporary).
• Specific LLMs, such as GPT-5 and gpt-oss, demonstrated strong performance in supporting valid downstream inference from their annotations.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2509.07274