{"id":"6f71d800-4eef-483e-bbb7-e90c19fa15b0","arxiv_id":"2411.18468","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Vocabulary, grammar, and short word sequences in French presidential speeches from 1958 to 2022 show a clear stylistic break around 1995 and make de Gaulle the most distinct speaker.","lead":"This paper analyzes 9,202 speeches by the eight French presidents from de Gaulle to Macron, about 20 million tagged words. It finds that presidential speaking style shifts over time and splits into two eras, with de Gaulle the most stylistically distant and Macron closest to Hollande.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Genre and archive confounds, acknowledged in §2, are not controlled in the Figure 1 distance tree; the apparent 'effet du temps' may be an artifact of changing speech settings and incomplete early corpora.","rationale":"The reader's weakest assumption identifies corpus comparability and genre effects as the main threat. My stress-test agrees and sharpens it: the manuscript itself, at the end of Section 2, states that exogenous factors should be neutralized by studying 'interventions présentées dans un cadre à peu près semblable', but the paper's headline Figure 1 does not do this. This is not an external ideological objection; it is an internally acknowledged methodological gap applied to the paper's most synthetic result. The concern is load-bearing because the abstract and Section 5 explicitly conclude that 'la figure illustre l’effet du temps' and that de Gaulle is the most distant from all others; if the distance tree is driven by genre mix and archival selectivity, those conclusions are overstated. The concrete test is straightforward and feasible with the existing corpus: select a genre that spans the whole period and rebuild the tree. The paper deserves credit for stating the limitation and pointing to prior controlled analyses, but the current manuscript does not supply the controlled evidence. Therefore the reader's CONDITIONAL verdict is appropriate; I do not see a need to move to REJECT because the descriptive tables (characteristic vocabulary, modals, POS) are partially independent and the claimed chronological trend is plausible. My verdict recommendation is UNCHANGED relative to the reader's CONDITIONAL, with the condition being the genre-controlled recomputation.","tokens_in":10980,"tokens_out":2132,"duration_ms":23281,"concrete_test":"Recompute the intertextual distance matrix and the Figure 1 tree using only a single recurring genre available across all presidents, e.g., televised New Year wishes (vœux) or major set-piece addresses, with corpora balanced by genre and roughly equal text counts. If the 1995 split and the de Gaulle–Hollande maximal distance persist in this controlled subcorpus, the 'effet du temps' survives the genre confound; if the tree flattens or rearranges, the central claim must be qualified as genre- and archive-driven.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 2 explicitly concedes that 'outre la question du genre... certaines caractéristiques de ces textes peuvent refléter l’influence des circonstances' and that 'il faudrait pouvoir aussi neutraliser ces facteurs exogènes'. Yet the central synthesis in Figure 1 aggregates all speeches per president and computes intertextual distance (Labbé & Labbé 2006) on the full, unbalanced corpus. Early presidents (de Gaulle, Pompidou, Giscard) have incomplete archives with few texts (460, 137, 191) and short mean lengths (908, 1 923, 3 504 words), while later presidents contribute thousands of heterogeneous interventions (press conferences, interviews, formal addresses). If the surviving early speeches are disproportionately ceremonial or set-piece, and later corpora include more varied, interactive, or unscripted formats, the lexical-frequency profiles will diverge over time for reasons unrelated to personal style. The claimed major break in 1995 and de Gaulle's isolation may therefore reflect corpus composition and genre drift rather than a genuine chronological evolution of presidential style. The manuscript's own prior work on a controlled genre (Labbé et al. 2020 on vœux) is not used to validate Figure 1, so the central claim rests on an uncontrolled comparison.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents a corpus of 9,202 French presidential speeches (1958-2022) spanning eight presidents, enriched with lemmas and part-of-speech tags. It then characterizes each president through overused vocabulary items, grammatical categories, short word sequences, and a dendrogram of intertextual distances computed between aggregated presidential corpora. The main analytical claims are that presidential rhetoric evolves along a chronological axis, that a major stylistic break occurs around 1995, that the two terms of Mitterrand and of Chirac are stylistically close, and that de Gaulle is the most isolated president.","tokens_in":11212,"tokens_out":3352,"duration_ms":31941,"significance":"The corpus itself is a substantial and potentially reusable resource, and the paper offers many concrete, falsifiable observations (e.g., the rise of the modal 'pouvoir', the shift from 'je' to 'nous', the characteristic n-grams of each presidency). If the central distance-tree result were robust, the paper would be a useful contribution to French political stylometry. The authors also make the corpus available and transparently flag incomplete early archives. However, the central synthesis rests on a comparison of corpora that differ strongly in archive completeness, genre composition, and average text length, as the authors themselves acknowledge. Until this uncontrolled variation is addressed, the 'effet du temps' in Figure 1 cannot be separated from corpus-construction effects.","major_comments":[{"comment":"The distance tree is computed on per-president aggregates of all speeches, but Table 1 shows that de Gaulle, Pompidou, and Giscard d'Estaing have incomplete and small corpora (460, 137, and 191 texts; mean lengths 908, 1,923, and 3,504 words), whereas later presidents contribute thousands of heterogeneous interventions (press conferences, interviews, formal addresses). Section 2 explicitly concedes that 'outre la question du genre... certaines caractéristiques de ces textes peuvent refléter l'influence des circonstances' and that 'il faudrait pouvoir aussi neutraliser ces facteurs exogènes'. The claimed 'effet du temps' and the 1995 break may therefore be an artifact of genre and archive drift rather than a genuine stylistic evolution. The authors should either recompute the distance tree on a controlled genre (e.g., the presidential vœux used in Labbé et al. 2020) or demonstrate through subsampling and genre-stratified analyses that the chronological pattern persists. As it stands, the central claim is not fully supported.","section":"§5, Fig. 1 and Table 1"},{"comment":"The method for identifying 'vocables caractéristiques' is only referenced to Labbé & Savoy 2021; no statistic, threshold, or normalization is defined in this paper, and no confidence intervals or significance tests are provided. Given the strongly unequal corpus sizes (e.g., 417k words for de Gaulle versus 5.6M for Mitterrand), the top overused lemmas and modal-verb rates may be unstable and dominated by rare-event noise. The authors should report the underlying score, effect sizes with confidence intervals (e.g., bootstrap), or a permutation-based null model so that the characteristic-vocabulary lists can be evaluated.","section":"§3, Tables 2 and 3"},{"comment":"The intertextual distance is cited to Labbé & Labbé 2006 but not defined in this manuscript; the phrase 'taux de recouvrement' is too vague for replication, and no robustness assessment of the dendrogram is given. In particular, the validation based on the closeness of Mitterrand's and Chirac's two terms is anecdotal. Please provide the distance formula, the preprocessing details (stopword handling, lemma versus form), and a measure of cluster stability (bootstrap or perturbation) so that the reader can assess whether the major 1995 split and de Gaulle's isolation are stable features.","section":"§5, distance calculation"}],"minor_comments":[{"comment":"The English abstract and the author list render 'Hollande' as 'Holland'; the correct name should be used consistently throughout the English text.","section":"Abstract"},{"comment":"The sentence defining the distance bounds appears to have the endpoints reversed: it says 0 corresponds to 'les deux n'ont rien en commun' and 1.0 to 'textes identiques,' but the following sentence states that many common words with similar frequencies yield a 'distance faible.' Please correct either the endpoint labels or the description.","section":"§5, first paragraph"},{"comment":"The paper states '10 x 10 = 100 valeurs' for the symmetric distance matrix; since Mitterrand and Chirac are each split into two terms, the matrix has 10 rows and 10 columns but only 45 pairwise distances + 10 self-similarities, or 55 distinct off-diagonal values. Please state the exact number of distinct pairs used in the clustering.","section":"§5, matrix size"},{"comment":"The 'Longueur' column values are hard to read in the typeset table (e.g., '417 146' and '20 027 931'); adding thousands separators consistently and aligning decimal points would improve readability.","section":"Table 1"}],"recommendation":"major_revision","confidential_remarks":"The archival and genre confounds are acknowledged in the manuscript itself, so this is not a hidden flaw; however, the central dendrogram claim is not robust without an additional controlled analysis. The paper should be revised to either add such an analysis or substantially weaken the claims in §5. The resource and descriptive observations are valuable and could merit publication in a specialized journal after the requested revisions."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a competent descriptive study that extends your own prior work on French presidential speeches. The new material—Macron's corpus, POS and n-gram profiles for all eight presidents, and the intertextual distance tree—is useful and mostly checkable. The central claim, a stylistic break around 1995, is plausible but rests on a corpus comparison that doesn't control for genre or archive completeness, so I'd treat the tree as a hypothesis rather than a settled result.\n\nWhat the paper does well: The corpus is large (9,202 texts, 20M labeled words) and the authors are transparent about its limits. Table 1 openly flags the early corpora as incomplete and the text in §2 concedes that genre and circumstances could affect the features. The characteristic vocabulary and n-gram tables give concrete, falsifiable descriptions—e.g., de Gaulle's 'honneur' and 'Algérie', Sarkozy's 'on' and negation, Hollande's 'nous' and 'pouvoir'. The distance tree has a nice internal check: Mitterrand's two mandates and Chirac's two mandates cluster tightly, which suggests the method captures stable authorial signal.\n\nThe soft spots are real but not fatal. The main one is Figure 1. It aggregates all speeches per president, so early presidents with 137–460 short, mostly archival texts are compared against later presidents with thousands of varied interventions. The 'effet du temps' could be partly an artifact of genre drift and incomplete early archives. The authors acknowledge this in §2 but don't act on it; they even mention the vœux corpus as a controlled alternative but don't use it to validate the tree. There are also no significance tests or confidence intervals, so we don't know how robust the 1995 split is. The characteristic-vocabulary method is cited to a previous paper rather than described, which is fine for a sequel but makes the table harder to audit on its own.\n\nThe reader's conditional verdict and moderate confidence seem right. The stress-test concern about genre/archive confounds is legitimate and lands. But the paper is honest, clearly written, and offers a useful synthesis for anyone working on French political discourse or presidential rhetoric. I'd send it to peer review with a request for a controlled genre comparison or at least a subsampling analysis to test whether the chronological split survives.","headline":"A competent, transparent descriptive study of French presidential style that extends prior work; the 1995 stylistic break is plausible but not fully secured because genre and archive confounds are acknowledged yet not controlled.","tokens_in":11745,"tokens_out":3353,"would_cite":true,"duration_ms":27602,"reading_group":"maybe","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 claims that French presidential speech style changed along a strong chronological axis, with a sharp break around 1995, and that de Gaulle is the most stylistically isolated president.","keywords":["stylometry","French presidential speeches","Fifth Republic","intertextual distance","characteristic vocabulary","part-of-speech","n-grams","presidential rhetoric"],"falsifier":"Restrict the analysis to one speech genre available for every president, such as televised New Year wishes, and recompute the distance tree; if the two-block break around 1995 and de Gaulle's position as the most isolated president do not appear, the chronological finding is an artifact of mixed genres and incomplete early archives.","tokens_in":10776,"feed_emoji":"🎤","tokens_out":10155,"duration_ms":88445,"temperature":0.7,"pith_summary":"This paper sets out to show that the public speech of French presidents changed in measurable, systematic ways over the sixty-six years of the Fifth Republic. It assembles a corpus of 9,202 speeches containing more than 20 million tagged words, gives each president a lexical fingerprint from vocabulary, grammatical categories, and recurring short phrases, and then computes how far the presidents' overall speech styles are from one another. The central finding is that the distances line up with time: presidents further apart in years sound more different, the main break occurs around 1995, and Charles de Gaulle is the most stylistically isolated of the eight. If this holds, presidential rhetoric becomes a quantifiable historical record that tracks institutional change as well as individual personality.","feed_headline":"French presidents' speech styles shift with time, split at 1995","feed_subtitle":"9,202 speeches show presidential style drifting with the decades, not with party.","key_machinery":"The engine of the analysis is the intertextual distance between two textual profiles, a value from 0 to 1 based on vocabulary overlap and the closeness of word frequencies, where 0 means nothing in common and 1 means identical texts. The paper applies it to ten profiles: the complete speech output of each of the eight presidents, with Mitterrand's and Chirac's two mandates treated as separate profiles. The resulting distance matrix is turned into a phylogenetic-style tree by automatic classification, with branch lengths representing stylistic distance. A second mechanism, characteristic-vocabulary computation, flags lemmas, parts of speech, and two- or three-word sequences that one president uses significantly more often than the others; the unit of counting is the 'vocable', a lemma paired with its grammatical tag.","core_discovery":"On its own terms, the paper establishes that the rhetoric of French presidents is ordered by time. Using a corpus of 9,202 speeches and more than 20 million words tagged for lemmas and parts of speech, it computes an intertextual distance between the complete set of speeches of each president, splitting Mitterrand's and Chirac's two terms into separate profiles. The resulting tree places the two terms of each of those presidents close together, which the paper reads as validation of the measure, and it arranges the presidents roughly along a chronological axis. The main rupture occurs in 1995: de Gaulle, Pompidou, Giscard d'Estaing, and Mitterrand form one block, while Chirac, Sarkozy, Hollande, and Macron form another. De Gaulle is the farthest from everyone, with the largest distance to Hollande; Pompidou is his nearest neighbour; Hollande is not a stylistic continuation of Mitterrand; and Macron sits closest to Hollande, at least up to 2022. The paper also identifies per-president markers: overused lemmas, preferred modal verbs, characteristic parts of speech, and two- or three-word sequences that survive the comparison with the other presidents.","pith_inferences":["An implication the paper leaves implicit is that the chronological axis should survive on a matched-genre corpus; if New Year wishes alone do not reproduce the 1995 split, genre mix explains the result.","The rise of 'pouvoir' and 'faire' in later presidents could be compared with non-presidential public speech to test whether it is a general shift in French public language rather than an individual trait.","The changing names for Europe ('marché commun', 'communauté européenne', 'Union Européenne') suggest the 1995 break may be partly lexical; removing Europe-related lemmas from the distance calculation would show how much of the split they carry."],"forward_implications":["A president's place on the 1958–2022 timeline can be inferred from vocabulary alone, because intertextual distance tracks chronology more than party.","The closeness of Mitterrand's two terms and Chirac's two terms validates the distance measure for authorship recognition in institutional speech.","The identified fingerprints give checkable predictions: de Gaulle overuses 'honneur' and 'peuple', Sarkozy favors 'on' and negation, Hollande and Macron favor 'nous', and each president has signature two- or three-word phrases.","The break at 1995 divides the Fifth Republic's rhetoric into two eras, meaning studies of French political discourse should not treat 1958–2022 as one homogeneous period."],"supporting_citations":[{"why":"Defines the intertextual distance measure used to build the presidential similarity tree.","marker":"Labbé & Labbé 2006"},{"why":"Provides the characteristic-vocabulary method and the earlier stylistic portraits this study extends to all eight presidents.","marker":"Labbé & Savoy 2021"},{"why":"Supplies the statistical principle for deciding when a word is overused by one author relative to others.","marker":"Muller 1977"},{"why":"Gives the automatic classification techniques applied to the distance matrix.","marker":"Baayen 2008"},{"why":"Provides the phylogenetic-tree representation used in the presidential similarity figure.","marker":"Paradis 2011"},{"why":"Offers comparable State of the Union evidence that time and exceptional events shape presidential style.","marker":"Savoy 2015; 2017"},{"why":"Supports the reading of pronoun choice (je, nous, on) as a stylistic and psychological marker.","marker":"Pennebaker 2011"}],"fun_headline_variants":["French presidential speech style splits at 1995","De Gaulle to Macron: rhetoric ordered by time, not party","9,202 speeches show 1995 turning point in presidential style","Time, not party, drives French presidents' speaking styles"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the eight corpora are comparable: early presidents' archives are incomplete and the speeches mix many genres, so if archival gaps or speech settings drive the lexical differences, the chronological tree reflects those factors rather than personal style.","fun_headline_variants_meta":{"raw":{"variants":["French presidential speech style splits at 1995","De Gaulle to Macron: rhetoric ordered by time, not party","9,202 speeches show 1995 turning point in presidential style","Time, not party, drives French presidents' speaking styles"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000182,"raw_usage":{"total_tokens":1395,"prompt_tokens":1111,"completion_tokens":284,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":727,"completion_tokens_details":{"reasoning_tokens":215}},"tokens_in":727,"tokens_out":284,"duration_ms":3600,"temperature":1.0,"reasoning_tokens":215,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T11:09:12.410398+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Restrict the analysis to one speech genre available for every president, such as televised New Year wishes, and recompute the distance tree; if the two-block break around 1995 and de Gaulle's position as the most isolated president do not appear, the chronological finding is an artifact of mixed genres and incomplete early archives.","supporting_citations":[{"cited_title":"Une approche identique s’applique afin de définir les catégories grammaticales (voir le tableau","cited_arxiv_id":null,"evidence_quote":"Provides the characteristic-vocabulary method and the earlier stylistic portraits this study extends to all eight presidents."},{"cited_title":"La figure 1 illustre le résultat obtenu nous permettant de mieux visualiser les relations entre présidents","cited_arxiv_id":null,"evidence_quote":"Provides the phylogenetic-tree representation used in the presidential similarity figure."}],"review_version":1}