Blaming Across the Aisle: Political Contrasting and Blame Attribution in the Danish Parliament
Markus Lundsfryd Jensen, Rune Egeskov Trust, Kenneth Christian Enevoldsen, Sara Kolding
cs.CL
Quick verdict
Claims Danish parliamentary blame fell to 2016 then rose, amplified by right-wing extremity; credible but silver-label circularity and n=13 caveats temper it.
Summary
This empirical study examines blame attribution in Danish parliamentary speech from 1997 to 2026. The authors build BlameBERT, the first blame classifier for the Danish Parliament, using an annotation-efficient pipeline: 500,500 sampled sentences are machine-translated into English, labeled by the DEBATE zero-shot NLI model under five hypothesis templates (DIAL agreement levels), and a mmBERT encoder is LoRA-fine-tuned on the most conservative DIAL-5 subset. On a 424-sentence hand-labeled test set (two author-annotators, Cohen's Kappa .676), it reaches macro-F1 0.80 (blame-class F1 0.75). Applied to roughly 4.94 million sentences, the analysis finds a banana-shaped trajectory: blame declines until around April 2016, then rises significantly and linearly during 2019-2026. Opposition parties blame substantially more than governing parties ('political contrasting'), the dampening effect of governing is weaker for right-wing parties, and ideological extremity amplifies blame mainly on the right, intensifying in recent years. Threshold sensitivity analyses and a ManifestoBERTa topic-control check support robustness. Main limitations: silver labels inherit translation and zero-shot bias; the gold test set is drawn from the pipeline's own DIAL-1 labels and annotated by the authors; key ideological interactions rest on only 13 parties; blame targets remain unidentified.
Review complete
Assessment at a glance
6
risk flags identified
● high 0
● medium 3
● low 3
4 strengths · 4 weaknesses · 5 author questions
AI text signal = Binoculars zero-shot detector score over 20 sampled passages (Fast-DetectGPT co-reported) — absence does not prove human authorship, and a flag does not imply misconduct.
Levels reflect verification complexity — not a quality score. Evolve never rates your paper.
Key claims under review
- Blame attribution in the Danish Parliament declined until around April 2016, then entered a significant, sustained rise during 2019-2026.
- Opposition parties blame substantially more than governing parties (political contrasting; exp(-0.478)~0.62 for governing left parties, exp(-0.515)~0.60 in recent years), in both the full period and 2019-2026.
- Ideological extremity amplifies blame far more strongly on the right (wing x wingness b=0.554 full period; b=0.817, ~126% per unit, in 2019-2026), indicating an ideologically asymmetric hardening concentrated on the right.
- The conclusions are robust to classification thresholds (tau=0.50-0.875, Table 12) and to controlling for speech topic via ManifestoBERTa (Appendix A.10).
Strengths
First purpose-built resource for Danish blame detection, fully released
BlameBERT (LoRA-tuned mmBERT, DIAL-5) plus code (GitHub), model and dataset (Hugging Face), and W&B logs; fair baselines against Qwen3 embedding (macro-F1 .67) and generative Qwen 3.5 (.75 under a disclosed abstain strategy with 68/424 null outputs, Table 6).
Rare 29-year longitudinal scope with merge-point validation
~4.94M analyzed sentences across 1997-2026; Appendix A.1 validates the ParlSpeechV2/SFTP seam via KS test (D=0.0847, p=.810) and a post-2019 indicator (b=-58.2, ns), plus monthly-count density/boxplot comparisons.
Robustness discipline uncommon in this literature
Candid limitations section
Weaknesses
Circular gold standard
Operative-construct inconsistency between label rule and inference threshold
Headline ideological interactions rest on 13 clusters from one source variable
Interpretive gap: blame targets are unknown
Full Review Report
$1.99 · one-timeUnlock the complete preliminary analysis Evolve produced for this paper.
Report preview
Scope & Claims
The paper asks whether blame attribution in Danish parliamentary speech has risen over 1997-2026 (~4.94M analyzed sentences, 13 parties) and how it varies with government status and ideology. Four central claims: (1) a banana-shaped temporal trajectory - decline to a minimum around April 2016, then a significant sustained rise in 2019-2026 (H1 quadratic b=0.0135, p=.0347; H1.1 linear b=0.0921, p<.001); (2) 'political contrasting': opposition blames substantially more than government (b=-0.478, p<.001; governing left parties blame ~62% as much as opposition); (3) an ideologically asymmetric hardening: wingness amplifies blame far more on the right (wing x wingness b=0.554 full period; b=0.817, ~126% per unit, in 2019-2026); (4) conclusions survive threshold sensitivity (Table 12) and topic controls (A.10). Code, model, and dataset are released.
The remaining sections cover Methodology Assessment, Statistical Evidence, Results & Reproducibility, Limitations & Risks, and Verdict & Recommendations.
- All 4 strengths & 4 weaknesses, analyzed in detail
- 6 risk flags with severity, section & reviewer questions
- Methods & statistical evidence assessment
- 5 questions a peer reviewer would ask
- Full six-dimension evaluation report & recommendations
One-time purchase · this paper only
One-time purchase, this paper only. AI-generated preliminary analysis — not peer review.
Frequently asked questions
What is BlameBERT and how accurate is it?
BlameBERT is a fine-tuned mmBERT text classifier that detects whether a sentence from the Danish Parliament contains blame (a causal claim combined with negative sentiment). It was trained on silver labels created by machine-translating Danish sentences to English and labeling them with the DEBATE zero-shot NLI model. On a 424-sentence hand-labeled test set it reaches macro-F1 0.80 (blame-class precision 0.72, recall 0.79), ahead of zero-shot Qwen embedding (0.67) and generative (0.75) baselines.
Has political rhetoric in the Danish Parliament become more blame-heavy over time?
According to this study, yes - but only recently. Blame rates declined from 1997 until around April 2016, then rose significantly and continuously from 2019 to 2026, producing a banana-shaped 29-year trajectory across roughly 4.94 million parliamentary sentences.
What does 'political contrasting' mean in this paper?
It is the authors' term for the finding that opposition parties attribute blame substantially more than governing parties - modeling estimates governing parties blame about 60-62% as much as comparable opposition parties - consistent with the incentive for challengers to highlight incumbents' failures and differentiate themselves.
Are the paper's data, model, and code publicly available?
Yes. Code is on GitHub (Lundsfryd/BlameBERT), the model on Hugging Face (Lundsfryd/BlameBERT), the dataset at Hugging Face datasets/runetrust/blame-folketinget-dk, and experiment tracking logs on Weights & Biases via links in the paper's appendices.
About this review
This is an AI-generated preliminary review by Evolve.Pub, produced from the openly available text of the paper. It is not peer review and does not represent the authors' views. The canonical record of the paper is at the publisher link above.
- All figures (2-13) could not be analyzed from the extracted text; only captions and in-text statistics were available.
- PDF extraction may have merged table cells; the precision/recall discrepancy between Section 2.2 text and Tables 1/4 may partly reflect macro- vs blame-class metrics rather than error, though it is undocumented.
- Appendix content (A.3 W&B links, per-DIAL details, figures 6-13) was only partially extracted and not independently verified.
- Review is based solely on the embedded full text of arXiv 2609.26346; no external verification of repository contents was performed.
Go deeper on this paper
Wrote a paper? Stress-test it before reviewers do.
Upload your manuscript for the same structured scrutiny — reasoning, statistics, evidence, figures, and reproducibility.