XAI in Computational Linguistics: Understanding Political Leanings in the Slovenian Parliament

AI can reveal hidden patterns in political discourse: Machine learning models successfully predicted whether Slovenian parliamentarians leaned left or right based solely on their speeches about migration, demonstrating that political ideology leaves measurable linguistic fingerprints that algorithms can detect.

Explainability tools provide genuine insights beyond predictions: By using explainability techniques to understand why models made their predictions, researchers identified specific keywords and phrases—such as "people" and "unity" for left-leaning speakers versus "nationality" and "illegal migrants" for right-leaning speakers—that reveal how different political groups frame the same issue differently.

AI transparency serves research beyond engineering: This work demonstrates that explainable AI isn't just useful for improving algorithms; it can be a legitimate tool for social scientists and humanities researchers to systematically analyze political language and ideology in ways that complement traditional qualitative research methods.

Language choices reflect political values: The study found that left-leaning parliamentarians framed migration through concepts of shared humanity and inclusion, while right-leaning parliamentarians emphasized national identity and legal status—showing how political orientation shapes not just policy positions but fundamental vocabulary choices.

Context matters for responsible AI application: By focusing on a specific debate (European migration crisis) in a specific parliament over a defined period (2014-2020), the research demonstrates the importance of bounded, contextual AI analysis rather than claiming universal patterns about political language.