💬 NLP

Epistemic orientation in parliamentary discourse is associated with deliberative democracy

This paper introduces a scalable Evidence-Minus-Intuition (EMI) score to analyze 15 million parliamentary speeches across seven countries, revealing that a stronger orientation toward evidence-based reasoning is positively associated with higher levels of deliberative democracy and effective governance over time.

Original authors: Segun Aroyehun, Stephan Lewandowsky, David Garcia

Published 2026-04-22

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Segun Aroyehun, Stephan Lewandowsky, David Garcia

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a parliament not as a room full of politicians shouting over each other, but as a giant, global kitchen where the recipe for a country's future is being cooked up.

This paper is about what kind of ingredients those politicians are using in their recipes. Are they using fresh, measurable data (like "2 cups of flour" and "300 degrees heat"), or are they relying on gut feelings and old family myths (like "it just feels right" or "my grandmother always said so")?

Here is the breakdown of the study in simple terms:

1. The New Tool: The "Truth-O-Meter" (The EMI Score)

The researchers invented a new way to measure the "flavor" of political speech. They call it the Evidence-Minus-Intuition (EMI) score.

How it works: They used super-smart AI (Large Language Models) to read 15 million speeches from parliaments in seven different countries (USA, Germany, Italy, Iceland, Poland, Turkey) spanning from 1946 to 2025.
The AI's Job: The AI acted like a taste-tester. For every sentence, it asked: "Is this speaker citing facts, data, and logic?" (Evidence) OR "Are they relying on feelings, beliefs, and gut instincts?" (Intuition).
The Score: If a speech is heavy on facts, the score goes up. If it's heavy on feelings and opinions, the score goes down.

2. The Big Discovery: Facts and Democracy are Best Friends

The researchers wanted to know: Does the way politicians talk actually matter for how well a country runs?

They found a strong, positive link between the EMI Score and Deliberative Democracy.

The Analogy: Think of a democracy like a team sport.
- When players (politicians) talk to each other using facts and evidence, they are like teammates passing the ball and calling out plays. They can solve problems, agree on strategies, and win games together. This is a healthy democracy.
- When players talk using only gut feelings and personal beliefs, they are like players arguing about the rules of the game while standing in different corners. They can't agree on anything, and the game falls apart. This is a sick democracy.

The Result: In every country they studied, when politicians started using more evidence-based language, the quality of their democracy improved. Conversely, when they stopped using facts, the democracy started to struggle. It's a two-way street: good democracy encourages facts, and using facts builds good democracy.

3. The Bonus Effect: Clearer Laws, Fewer Surprises

The study also looked at Governance—specifically, how clear and predictable the laws are.

The Analogy: Imagine the law as a road map.
- High EMI (Evidence-based): The map is drawn with GPS precision. Everyone knows exactly where the road goes, where the potholes are, and how to get to the destination. The government enforces the rules consistently.
- Low EMI (Intuition-based): The map is a scribble on a napkin. "Turn left when you feel like it." The rules change depending on who is driving. This leads to chaos and corruption.

The Result: Countries where politicians spoke more in facts and data had laws that were clearer, more transparent, and enforced more predictably.

4. Why This Matters

In the past, we worried that politicians were just "lying" or "twisting the truth." This study suggests something deeper is happening: The very style of their thinking is changing.

The Trend: In some places (like the US), the "Evidence-O-Meter" has been dropping over time. Politicians are relying more on emotion and intuition and less on hard data.
The Warning: The authors warn that if we lose our shared "evidence-based" language, we lose the ability to solve problems together. Without a shared set of facts, we can't agree on what the problem is, let alone how to fix it.

The Bottom Line

This paper tells us that truth isn't just a moral virtue; it's a practical tool.

Just as a builder needs a level and a tape measure to build a sturdy house, a democracy needs evidence and data to build a stable society. When politicians stop using the "tools of truth" and start building on "shifting sands of opinion," the whole structure becomes shaky.

In short: If you want a democracy that works, you need politicians who argue with facts, not just feelings.

1. Problem Statement

Democratic deliberation relies on the pursuit of truth, which can be approached through two distinct epistemic orientations:

Evidence-based reasoning: Grounded in verifiable facts, data, and logical analysis.
Intuition-based reasoning: Rooted in beliefs, values, subjective interpretation, and emotional appeals.

While previous computational studies (primarily in the US) suggested a decline in evidence-based language in political discourse, these findings were limited by:

Linguistic and Institutional Constraints: Reliance on language-specific dictionaries or resources prevented cross-country and cross-linguistic comparison.
Scalability: Existing methods could not easily process millions of speech segments across diverse political systems.
Scope: It remained unclear how the epistemic nature of discourse correlates with broader measures of democratic health (deliberative democracy) and governance quality (transparency and predictability of laws) globally.

The study aims to develop a scalable, multilingual metric to measure Epistemic Orientation and analyze its relationship with Deliberative Democracy and Governance Quality across seven countries from 1946 to 2025.

2. Methodology

A. Data Collection and Preprocessing

Corpus: 15 million text segments extracted from parliamentary transcripts of seven countries: USA, West Germany/Germany, Italy, Iceland, Poland, and Turkey.
Timeframe: 1946–2025 (adjusted for specific political transitions, e.g., post-1989 for Poland, post-1982 constitution for Turkey).
Preprocessing Pipeline:
- Filtering: Removed speeches by session chairs and procedural interventions (e.g., agenda management, rule discussions) using an LLM classifier (Llama-3.1-8B) rated on a 0–4 procedural scale.
- Attribution: Ensured segments were attributable to specific speakers (removing annexes).
- Noise Reduction: Filtered out low-information speeches (dominated by lists of names/numbers) based on lexical composition ratios.
- Segmentation: Long speeches were chunked into ~150-token segments (minimum 50 tokens) to create uniform analysis units.

B. Measurement: The EMI Score

The authors introduced the Evidence–Minus–Intuition (EMI) score, calculated using a hybrid approach combining Large Language Models (LLMs) and semantic embeddings:

LLM Ratings: Three open-weight multilingual LLMs (Llama-3.1-8B, Qwen2.5-7B, Apertus-8B) rated each text segment on two scales (0–4):
- Extent of Evidence-based language (verifiable facts, data, causal reasoning).
- Extent of Intuition-based language (gut feelings, anecdotes, moral appeals, beliefs).
Embedding-Based Similarity:
- Conceptual Anchors: 15 seed terms per category (evidence vs. intuition) were defined with dictionary definitions and translated into target languages.
- Vector Space: Text segments were embedded using the multilingual mGTE model.
- Calculation: Cosine similarity was computed between the text segment and the averaged vectors of the "evidence" and "intuition" anchors.
Final Score: The LLM ratings and embedding similarities were standardized (z-transformation) and averaged.
- $EMI = \text{Evidence Score} - \text{Intuition Score}$
- Positive EMI: Dominance of evidence-based discourse.
- Negative EMI: Dominance of intuition-based discourse.

C. Statistical Analysis

Dependent Variables:
- Deliberative Democracy Index (DDI): A composite measure of electoral and deliberative aspects (reasoned justification, common good appeals, respect for counterarguments).
- Transparent Laws and Predictable Implementation (TPL): A governance index measuring the clarity and enforceability of laws.
Models: Linear regression models with Country and Year Fixed Effects to control for time-invariant cross-national differences and temporal shocks.
Lagged Analysis: Models were run with contemporaneous variables and one-year lags to assess temporal ordering.
Controls: Judicial independence, clientelism, log GDP per capita, and DDI (when analyzing TPL).

3. Key Contributions

Scalable Multilingual Metric: Developed a robust, scalable method (EMI) to measure epistemic orientation across seven languages and political systems, overcoming the limitations of dictionary-based approaches.
Hybrid Validation: Validated the hybrid LLM+Embedding approach against human annotations, achieving an AUC of 0.825, significantly outperforming previous word-embedding-only methods (AUC = 0.791).
Cross-National Empirical Evidence: Provided the first systematic analysis linking parliamentary discourse quality to democratic and governance outcomes across diverse regimes (liberal democracies, electoral democracies, and competitive autocracies).
Governance Linkage: Extended the analysis beyond deliberation to show that epistemic orientation predicts the quality of law implementation (transparency and predictability).

4. Key Results

EMI and Deliberative Democracy (DDI):
- There is a positive and significant correlation between EMI and DDI across all seven countries (Pooled $r = 0.207$ , $p < 0.05$ ).
- Bidirectional Relationship:
  - Higher EMI predicts higher DDI in the same year ( $b = 0.160$ ) and the following year ( $b = 0.116$ ).
  - Higher DDI predicts higher EMI in the same year ( $b = 0.296$ ) and the following year ( $b = 0.265$ ).
- Temporal Trends: Countries with rising DDI (e.g., post-war West Germany, Iceland) generally show stable or rising EMI. Countries experiencing democratic backsliding (e.g., Turkey, Poland) show concurrent declines in both DDI and EMI.
EMI and Governance (TPL):
- EMI is a positive and significant predictor of the Transparent Laws and Predictable Implementation index ( $b = 0.407$ , $p < 0.05$ ).
- This relationship holds even after controlling for judicial independence, clientelism, economic development, and deliberative democracy.
- Lagged models confirm that evidence-oriented discourse in one year predicts better law implementation in the subsequent year ( $b = 0.436$ ).

5. Significance and Conclusion

Reinforcing Mechanism: The findings suggest a reinforcing cycle where conditions supporting deliberation (high DDI) encourage evidence-based discourse, which in turn strengthens democratic quality.
Governance Implications: The study posits that evidence-oriented discourse leads to clearer, more internally consistent policy formulation. This reduces ambiguity in implementation, directly enhancing the predictability and transparency of governance.
Resilience of Democracy: The paper concludes that maintaining a shared evidentiary basis in parliamentary discourse is not merely an academic ideal but a functional necessity for the resilience and responsiveness of democratic governance.
Future Directions: The authors suggest extending this framework to other public discourse domains (news, social media, judicial communications) and conducting fine-grained analyses within specific institutional settings.

In summary, the paper demonstrates that how politicians speak (evidence vs. intuition) is inextricably linked to how well democracies function and govern, providing a quantitative tool to monitor the epistemic health of political systems globally.