{"id":"b1d6e2fa-d6e4-4665-a3be-7bf8dfded749","arxiv_id":"2608.05075","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"German elite discourse shifted toward intuition-based rhetoric between 2015 and 2025, with the sharpest drop in parliamentary language coinciding with the AfD's 2017 entry.","lead":"Using 4.5 million tweets and 59,170 German parliamentary speeches from 2015 to 2025, this study measured how often politicians use evidence-based versus intuition-based language. It finds that German political discourse moved toward intuition-based rhetoric after the far-right AfD entered parliament in 2017, with the AfD consistently scoring lowest on the evidence-minus-intuition scale.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 2017 breakpoint conflates the AfD's own entry with a chamber-wide shift: Table A8 cannot estimate a right-camp interaction, and no balanced-panel or AfD-excluded analysis isolates whether established parties changed.","rationale":"The reader's weakest assumption was EMI construct validity. That is a genuine concern, and the cross-embedding inconsistency (0.93 in the main text vs 0.80 and 0.65 in the supplement) plus the low single-rater ICCs justify caution. But the paper offers multiple validation layers, including dictionary validation, topic controls, component decomposition, bootstrapping, and historical face validity, so measurement problems alone do not fully undo the descriptive pattern. The more decisive threat to the headline is the 2017 composition confound, which is specific to the paper's central temporal claim and is visible in the model's own structure: no 2017 right-camp interaction can be estimated. The proposed AfD-exclusion and balanced-panel test would settle whether the 'broader chamber' shift survives once the AfD's own entry is removed. If it does not survive, the finding reduces to a composition change, not a party-level rhetorical shift, and the title and abstract overstate the result. This supports the reader's CONDITIONAL verdict while shifting the emphasis from measurement to identification.","tokens_in":54111,"tokens_out":9205,"duration_ms":128028,"concrete_test":"Re-fit the constrained breakpoint model (Eq. 2) on the 2015-2021 Bundestag corpus twice: (i) excluding all AfD speeches, and (ii) restricted to MPs with at least one speech before and after 1 July 2017 so delta_2017 is identified solely from within-speaker change. If delta_2017 remains about -0.4 and significant in both variants, the chamber-wide shift is real; if it collapses toward zero, the 2017 effect is the mechanical entry of low-EMI AfD members, and the abstract and title should be weakened to describe a composition shift rather than a shift in German parties.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The 2017 'chamber-wide' breakpoint is not identified as a party-level shift. The constrained breakpoint model (Methods Eq. 2, Table A7) estimates level_elec2017 = -0.427 (p < 0.001), and the paper reads this as a broad shift coinciding with the AfD's entry. But the AfD has no pre-2017 Bundestag speeches, so in Eq. 3 the AfD's leaning main effect and the post-2017 indicator are collinear; Table A8 accordingly reports no 'level elec2017 right' interaction, whereas 2021 and 2025 include right interactions. The uniform delta_2017 therefore mixes two quantities: the mechanical addition of roughly 90 low-EMI AfD members (mean -0.473, Extended Data Table A5) and any genuine within-speaker or within-party change among incumbent MPs. The paper itself describes Bundestag shifts as 'align[ing] temporally with changes in the composition of the Bundestag' (§2.3) and lists 'compositional adaptation' as a mechanism (Discussion). Without an analysis that excludes AfD or restricts to MPs present on both sides of 2017, the title's claim that German parties shifted is an overreach: a fall in the chamber's mean EMI is not evidence that established parties changed their rhetorical style.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper analyzes 4.5 million tweets and 59,170 Bundestag speeches from 2015–2025 to ask whether German political elites' rhetoric shifted from evidence- toward intuition-based language after the far-right AfD entered parliament in 2017. Using a distributed dictionary representation (DDR), the authors compute an Evidence-Minus-Intuition (EMI) score for each text and estimate mixed-effects models with election-year breakpoints. They report: (i) a consistent decline in EMI over time on both Twitter and in parliament, (ii) a persistent left-to-right ordering with the AfD lowest, and (iii) a sharp chamber-wide downward level shift in parliamentary EMI at the 2017 election breakpoint, together with a more gradual decline on Twitter. The manuscript interprets these patterns as an association between far-right parliamentary entry and a changing approach to truth in elite discourse, while noting that causal attribution is not possible.","tokens_in":54344,"tokens_out":5317,"duration_ms":69351,"significance":"If the measurement assumptions hold, this is a valuable cross-arena, multiparty extension of prior work on evidence-versus-intuition rhetoric in U.S. congressional speeches. The study leverages a large, heterogeneous corpus, two arenas with opposite failure modes, an EU DSA Article 40 data access, preregistered human validation, bootstrap stability checks, and topic-fixed-effect robustness analyses. The finding that parliamentary EMI declined most sharply at the exact moment of AfD's entry is a striking descriptive result that could motivate further causal designs. However, the paper's central claim—that 'German parties shifted'—is stronger than what the current identification supports, and the reported cross-embedding validation contains a numerical inconsistency. These issues are fixable but currently limit the strength of the conclusion.","major_comments":[{"comment":"The 2017 chamber-wide shift is not identified as a party-level shift among established parties. Because the AfD has no pre-2017 Bundestag speeches, the AfD main effect and the post-2017 indicator are collinear, which is why Table A8 reports no 'level elec2017 right' interaction. The constrained breakpoint level_elec2017 = -0.427 therefore mixes the mechanical entry of roughly 6,700 AfD speeches with mean EMI -0.473 (Table A5) with any genuine within-speaker or within-party change. The paper itself acknowledges this by describing the Bundestag shifts as aligning 'with changes in the composition of the Bundestag' (§2.3) and listing 'compositional adaptation' as a mechanism (Discussion). Given this, the abstract's and title's claim that 'German parties shifted' is an overreach. An analysis excluding AfD speeches or restricted to MPs observed on both sides of 2017 is required before the claim can be made.","section":"§2.3, Table A8, Eq. 3"},{"comment":"The main text and Methods state that fastText-based EMI scores correlate with the primary gensim-based scores at r = 0.93 (95% CI [0.91, 0.94]), but Supplementary Note 2.1 reports document-level correlations of 0.80 for parliamentary speeches and 0.65 for Twitter. This discrepancy is load-bearing because the r = 0.93 statistic is used to argue that results are not specific to the fine-tuned embedding model. If the Twitter correlation is in fact 0.65, the robustness claim is substantially weakened for the arena where the paper makes its main temporal comparisons. The authors must reconcile these numbers and report the correlation separately for each corpus.","section":"§2.2 and §4.5 vs. Supplementary Note 2.1"},{"comment":"The validity of EMI as a measure of epistemic orientation is moderate and deserves a stronger discriminant-validity check. The reported AUC of 0.72 for EMI is acceptable, but the single-rater ICCs (0.14–0.16) are low, and the mean cosine similarity between the evidence and intuition dictionary embeddings is 0.44 on Twitter, indicating substantial semantic overlap between the two poles of the difference score. Although the paper honestly reports these numbers, the interpretation of EMI as a stable epistemic-style signal would be strengthened by a placebo test using matched random word lists or by demonstrating that the 2017 shift and party ordering are not reproduced when EMI is computed from sentiment-matched or frequency-matched control dictionaries.","section":"§4.5, Supplementary Notes 2.2 and 2.12"}],"minor_comments":[{"comment":"The demographic description of the keyword-validation sample is inconsistent: the main text reports '35 male, 12 female' participants, while Supplementary Figure 25 and the accompanying text report '35 women and 12 non-binary/diverse'. The authors should correct this discrepancy.","section":"§4.3 vs. Supplementary Note 2.11"},{"comment":"The sentence 'we find a chamber-wide and largely party uniformly shift in 2017 shwon by the finding that both mainstream left and mainstream right shift positions once a radical right party enters parliament' is ungrammatical and contains a typo ('shwon'); it should be rewritten for clarity.","section":"Discussion, §3"},{"comment":"The figures mark the 2017 election breakpoint at 1 July 2017, and the text says breakpoints were set 'approximately one month before the election date.' The federal election took place on 24 September 2017, so 1 July is more than one month earlier. Please clarify how the breakpoint date was chosen.","section":"Figures 2 and 3, Methods Eq. 2"},{"comment":"The absence of a right-camp interaction for the 2017 breakpoint should be explained in a table note (e.g., collinearity due to no pre-2017 AfD speeches) so that readers do not interpret it as a null finding.","section":"Table A8"},{"comment":"Several figure captions and some text passages contain garbled character sequences (e.g., repeated '/uni000000...' fragments in the captions of Figures 1–4 and A1–A2). These rendering artifacts must be cleaned before submission.","section":"General"}],"recommendation":"major_revision","confidential_remarks":"The paper's title and abstract make a claim about 'German parties' shifting that the current analyses do not fully support, because the 2017 parliamentary shift is compositionally confounded with the AfD's own entry. This is likely to attract media and policy attention, so the editorial team should insist on either a within-MP analysis or a carefully qualified reformulation. I also recommend verifying the cross-embedding correlation discrepancy (r = 0.93 in the main text versus 0.80/0.65 in the supplement) before any acceptance decision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The headline finding is probably real; the headline claim is overstated. The paper tracks an Evidence-Minus-Intuition (EMI) measure through 4.5 million tweets and 59,170 Bundestag speeches (2015-2025) and documents a consistent drift toward intuition-based language in both arenas, with the AfD anchoring the low end throughout. That descriptive pattern is new for Germany and it extends the far-right contagion literature beyond positions and issue salience to a third, roughly orthogonal dimension of epistemic style.\n\nWhat's genuinely good: two arenas with opposite failure modes; document-level topic fixed effects that blunt the agenda-setting confound; a rolling-window specification that relaxes the piecewise-linear functional form; bootstrap checks on dictionary composition; a pre-registered document-level validation survey; and released code with data. The paper is also candid about its limits: it flags co-temporal shocks, lists composition as a mechanism, and disclaims causality.\n\nThe soft spots are real but mostly addressable. The largest is the title. The 2017 breakpoint (-0.427) comes from a model that cannot separate the mechanical entry of roughly 90 low-EMI AfD MPs from genuine within-party change among incumbents: the AfD has no pre-2017 speeches, so the right-camp interaction at 2017 is collinear and absent from Table A8. The paper itself says the shifts \"align temporally with changes in the composition of the Bundestag\" and lists \"compositional adaptation,\" yet the title and discussion claim that German parties shifted. The fix is straightforward: re-estimate without the AfD, or on a balanced panel of MPs seated on both sides of 2017. The rolling-window result showing all camps stepping down together mildly argues against a pure-composition story, but that is not a substitute for the clean test. The abstract's \"coincides with\" wording is fine; the title is not.\n\nSecond, the validation reporting is internally inconsistent: r=0.93 in the main text versus 0.80 (speeches) and 0.65 (Twitter) in the supplement for model agreement, and the keyword-survey demographics appear as 35 male/12 female in one place and 35 women/12 non-binary in another. These need a correction pass. The measurement itself is moderate: single-rater ICCs of 0.14-0.16, AUC of 0.72, and cross-dictionary similarity of 0.44 in the Twitter embedding space, which is a genuine limitation for short-tweet inference, though the aggregated analyses and average ICCs (0.62-0.65) blunt the damage.\n\nThis paper deserves a serious referee. The data work is substantial, the limitations are disclosed, and the load-bearing flaw is a too-strong title backed by a missing robustness analysis, not a broken research design. I'd engage with it, requesting the AfD-excluded and balanced-panel analyses plus a consistency pass on the validation numbers.","headline":"Real descriptive pattern, overstated headline: the 2017 chamber-wide shift is not separated from the AfD's mechanical entry, but the data work and disclosures deserve serious referee time.","tokens_in":54946,"tokens_out":8550,"would_cite":true,"duration_ms":92508,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper argues that German elite rhetoric moved toward intuition-based language after the AfD's 2017 entry, most sharply in the chamber.","keywords":["evidence minus intuition","epistemic orientation","far-right contagion","German Bundestag","political rhetoric","distributed dictionary representation","Twitter elite discourse","structural breakpoints"],"falsifier":"Fix topic, sentiment, and platform composition so that the 2017 breakpoint is the only thing varying, and re-estimate the EMI regressions on sentiment- and topic-matched speech segments, or replace the evidence and intuition dictionaries with matched neutral control dictionaries. If the chamber-wide 2017 drop persists with control dictionaries, EMI is measuring something other than epistemic orientation specifically; if it disappears once sentiment and topic are matched, the claimed rhetorical shift is an artifact of textual composition.","tokens_in":53866,"feed_emoji":"🏛️","tokens_out":7448,"duration_ms":81640,"temperature":0.7,"pith_summary":"This paper tries to establish that, between 2015 and 2025, the rhetorical style of German political elites moved away from evidence-based argument and toward intuition-based language, and that the sharpest single shift in parliamentary speech coincided with the entry of the far-right Alternative for Germany (AfD) in 2017. The authors trace this movement in two complementary arenas: 4.5 million tweets from over 2,000 German politicians and 59,170 plenary speeches in the Bundestag. If the result holds, the long-run decline of evidence-based political discourse previously documented in the United States is also present in a multiparty European democracy, and far-right visibility tracks not just what parties talk about but how they commit to truth. The paper is explicit that these are co-movements, not proof of causation.","feed_headline":"AfD's 2017 entry tracks sharp shift to intuition-based rhetoric","feed_subtitle":"4.5 million tweets and 59,170 Bundestag speeches show intuition-based rhetoric rising fastest in the chamber.","key_machinery":"The load-bearing instrument is the Evidence Minus Intuition (EMI) score: a text's averaged word embedding is compared by cosine similarity to averaged embeddings of a validated German evidence dictionary and a validated German intuition dictionary; the two similarities are z-scored across the corpus, and EMI is their difference. Platform-specific word2vec embeddings are trained on a historical parliamentary corpus and fine-tuned separately on tweets and contemporary speeches. Structural-break linear mixed-effects models with breakpoints at the 2017, 2021, and 2025 federal elections, plus a neural topic model used to add topic fixed effects, carry the temporal and party-level inferences.","core_discovery":"On the paper's own terms, the central discovery is that an Evidence Minus Intuition (EMI) score declines in both arenas over the decade: mean EMI on Twitter falls from 0.027 to -0.035, and in the Bundestag from 0.205 to -0.147. Right-leaning actors, above all the AfD, hold the lowest EMI scores throughout. In the chamber, a breakpoint regression shows a significant downward level shift at the 2017 election that is uniform across ideological camps, with further shifts at the 2021 and 2025 elections; on Twitter the decline is gradual and approximately linear. The estimates survive controls for document-level topic fixed effects, which the authors read as evidence that the shift is stylistic rather than a change in the topics under debate.","pith_inferences":["A testable extension implied by the paper's diffusion story: politicians' measurable network exposure to AfD content should predict within-actor EMI decline, not just aggregate co-movement.","Applying the same measurement to other countries with differently timed far-right entries would quasi-experimentally separate the 2017 event from concurrent shocks such as the COVID-19 pandemic and the war in Ukraine.","If EMI partly tracks emotional intensity or contested topics rather than epistemic posture, then fact-checking and platform moderation aimed at individual claims may address the symptom rather than the rhetorical drift; this could be tested by re-validating EMI on sentiment-matched texts.","The same dictionaries could be extended to mass-media or citizen discourse, a step the paper notes is outside its elite-only scope but that would show whether the shift begins with elites or with the public."],"forward_implications":["If the paper is right, the US-style decline in evidence-based congressional language has a multiparty European analogue within a single decade.","Far-right 'contagion' would extend beyond policy positions and issue salience to a third dimension: the epistemic style of how truth claims are made.","Mainstream and left parties converged downward toward the right's intuition-heavy baseline without the ordering of camps ever inverting.","Because the shift survives topic fixed effects, responses to far-right entry that target individual false claims may miss the level at which discourse is changing.","The contrast between discrete parliamentary shifts and gradual Twitter decline suggests different mechanisms: compositional adaptation in the chamber and continuous diffusion online."],"supporting_citations":[{"why":"Supplies the original EMI pipeline and the English evidence and intuition dictionaries that the German versions translate and validate.","marker":"[25]"},{"why":"Extends EMI across countries and links it to deliberative democracy, framing the paper's interpretation of the German decline.","marker":"[21]"},{"why":"Provides the fact-speaking versus belief-speaking distinction and the human-validation procedure adapted for document-level EMI.","marker":"[16]"},{"why":"Gives the causal regression-discontinuity evidence that radical-right parliamentary entry shifts mainstream parties, motivating the 2017 anchor.","marker":"[27]"},{"why":"Supplies the parliamentary speech corpus for contemporary Bundestag texts and for embedding pre-training.","marker":"[48]"},{"why":"Provides the original German politician Twitter dataset that the paper extends with additional handles and tweets.","marker":"[17]"}],"fun_headline_variants":["German elites' rhetoric shifts from evidence to intuition after AfD","Study: Bundestag speeches show decline in evidence-based language after 2017","Intuition-based rhetoric rises in German politics post-AfD","Far right's 2017 breakthrough linked to drop in factual language"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument depends on treating the EMI score, a difference of z-scored cosine similarities to evidence and intuition word lists, as a valid measure of a speaker's epistemic approach to truth, including in short tweets, rather than as a proxy for topic, sentiment, or platform style.","fun_headline_variants_meta":{"raw":{"variants":["German elites' rhetoric shifts from evidence to intuition after AfD","Study: Bundestag speeches show decline in evidence-based language after 2017","Intuition-based rhetoric rises in German politics post-AfD","Far right's 2017 breakthrough linked to drop in factual language"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000301,"raw_usage":{"total_tokens":1694,"prompt_tokens":861,"completion_tokens":833,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":477,"completion_tokens_details":{"reasoning_tokens":757}},"tokens_in":477,"tokens_out":833,"duration_ms":9696,"temperature":1.0,"reasoning_tokens":757,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T05:55:09.261059+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Fix topic, sentiment, and platform composition so that the 2017 breakpoint is the only thing varying, and re-estimate the EMI regressions on sentiment- and topic-matched speech segments, or replace the evidence and intuition dictionaries with matched neutral control dictionaries. If the chamber-wide 2017 drop persists with control dictionaries, EMI is measuring something other than epistemic orientation specifically; if it disappears once sentiment and topic are matched, the claimed rhetorical shift is an artifact of textual composition.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the original EMI pipeline and the English evidence and intuition dictionaries that the German versions translate and validate."},{"cited_title":", author Cohen, T","cited_arxiv_id":null,"evidence_quote":"Extends EMI across countries and links it to deliberative democracy, framing the paper's interpretation of the German decline."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the fact-speaking versus belief-speaking distinction and the human-validation procedure adapted for document-level EMI."},{"cited_title":"Epistemic orientation in parliamentary discourse is associated with deliberative democracy","cited_arxiv_id":"2604.19699","evidence_quote":"Gives the causal regression-discontinuity evidence that radical-right parliamentary entry shifts mainstream parties, motivating the 2017 anchor."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the parliamentary speech corpus for contemporary Bundestag texts and for embedding pre-training."},{"cited_title":", author Fraccaroli, N","cited_arxiv_id":null,"evidence_quote":"Provides the original German politician Twitter dataset that the paper extends with additional handles and tweets."}],"review_version":1}