Epistemic orientation in parliamentary discourse is associated with deliberative democracy

The correlation between evidence-based speech and democratic health is the headline, but the more consequential methodological question is whether an LLM-derived score can reliably distinguish 'evidence' from 'intuition' in political rhetoric across seven languages and eight decades without systematic bias baked in from training data. The paper's causal claims, if any, deserve scrutiny that the summary doesn't invite.

The reliability problem here connects directly to coverage from mid-April: 'Diagnosing LLM Judge Reliability' found that even when aggregate consistency looks high, one-third to two-thirds of documents show logical inconsistencies in pairwise comparisons. Applying that finding to EMI scores across 15 million speeches is not a small concern. If the judge is inconsistent at the document level, a country-level aggregate could still look stable while masking substantial noise. The 'Context Over Content' paper from the same period adds another wrinkle, showing LLM judges shift verdicts based on framing rather than substance, which is a particular hazard when scoring politically charged speech.

Watch whether the authors release per-country, per-decade reliability diagnostics for the EMI scorer itself. If the score variance within clearly authoritarian periods is as high as within democratic ones, the correlation finding weakens considerably.