Politicians vs ChatGPT: A study of presuppositions in French and Italian political communication
Open accessAbstract
This paper aims to provide a comparison between texts produced by French and Italian politicians on polarizing issues, such as immigration and the European Union, and their chatbot counterparts created with ChatGPT 3.5. In this study, we focus on implicit communication, in particular on presuppositions and their functions in discourse, which have been considered in the literature as a potential linguistic feature of manipulation. This study also aims to contribute to the emerging literature on the pragmatic competences of Large Language Models. Our results show that, on average, ChatGPT-generated texts contain more questionable presuppositions than the politicians’ texts. Furthermore, most presuppositions in the former texts show a different distribution and different discourse functions compared to the latter. This may be due to several factors inherent in the ChatGPT architecture, such as a tendency to be verbose and repetitive in longer texts, as exemplified by the occurrence of political slogans mainly formed by change-of-state verbs as presupposition triggers (e.g., dobbiamo costruire il nostro futuro, ‘we must build our future’).
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
In Garassino et al. (2024), CSV and DEF were the dominant presupposition triggers in GPT-3.5 data, with divergent distribution between LLM-generated and politician-authored speeches.
“Garassino et al. (2024) also observed the dominance of CSV and DEF as presupposition triggers in GPT-3.5 data, along with their divergent distribution in LLM-generated versus politician-authored speeches.”
From ChatGPT for President! Presupposed content in politicians versus GPT-generated texts (2025) · cites this paper as Garassino et al. (2024)
The Abstract says presuppositions in ChatGPT 3.5 texts have a different distribution from those in politicians’ texts, and it identifies change-of-state verbs as triggers in some chatbot-generated slogans. It does not say that CSV and DEF were the dominant triggers in the GPT-3.5 data; searches for CSV, DEF, and definite descriptions found no supporting passage in our copy. (LLM, conf=0.82)
Experimental — accuracy not yet measuredIn Garassino et al. (2024), the overall frequency of PMPs was higher in GPT-generated texts than in politician-authored speeches.
“In that study, the overall frequency of PMPs was likewise higher in GPT-generated texts.”
From ChatGPT for President! Presupposed content in politicians versus GPT-generated texts (2025) · cites this paper as Garassino et al. (2024)
LLM-generated texts tend to hallucinate when overly long.
“Following Garassino et al. (2024), who observed that LLM-generated texts tend to hallucinate when overly long”
From ChatGPT for President! Presupposed content in politicians versus GPT-generated texts (2025) · cites this paper as Garassino et al., 2024
The Abstract says ChatGPT-generated texts can be verbose and repetitive when longer and contain more questionable presuppositions than the politicians’ texts. It does not say longer texts tend to hallucinate; searches for hallucinations and longer texts found no passage establishing that link. (LLM, conf=0.76)
Experimental — accuracy not yet measuredIn Garassino et al. (2024), CRT was fairly frequent in the GPT-3.5 dataset.
“Notably, in Garassino et al. (2024), CRT was fairly frequent in the GPT-3.5 dataset”
From ChatGPT for President! Presupposed content in politicians versus GPT-generated texts (2025) · cites this paper as Garassino et al. (2024)
The Abstract (page unavailable) identifies a ChatGPT 3.5 dataset and mentions change-of-state presupposition triggers, but gives no frequency for CRT. Searches for “CRT” and for trigger frequency returned no passage in our copy establishing that CRT was fairly frequent. (LLM, conf=0.70)
Experimental — accuracy not yet measuredIn Garassino et al. (2024), STK was the most frequent discourse function of PMPs in the GPT data, while CRT also featured prominently in the politicians' texts.
“Similarly, STK emerged as the most frequent discourse function of PMPs in the GPT data, while CRT also featured prominently in the politicians ' texts.”
From ChatGPT for President! Presupposed content in politicians versus GPT-generated texts (2025) · cites this paper as Garassino et al. (2024)
The Abstract says presuppositions in ChatGPT and politicians’ texts have different discourse functions, but it does not identify which functions were most frequent. Searches for STK, CRT, and their relative frequencies found no relevant passage in our copy; the available passages do not verify either specific claim. (LLM, conf=0.70)
Experimental — accuracy not yet measured
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Cites
The publisher hasn’t deposited this paper’s reference list.
Research works printed but not linked: 15
- Agresti, Alan. 2019. An Introduction to Categorical Data Analysis (3rd edition). Hoboken, NJ: John Wiley & Sons.
- Cominetti, Federica & Gregori, Lorenzo & Lombardi Vallauri, Edoardo & Panunzi, Alessandro. 2022. IMPAQTS: un corpus di discorsi politici italiani annotato per impliciti linguistici. In Cresti, Emanuela & Moneglia, Massimo (eds), Corpora e Studi Linguistici. Atti del LIV Congresso Internazionale di Studi della Società di Linguistica Italiana, 151−164. Milano: Officinaventuno. https://doi.org/10.17469/O2106SLI000009 · DOI
- Garassino, Davide & Masia, Viviana & Brocca, Nicola. 2019. Tweet as you speak. The role of implicit strategies and pragmatic functions in political communication: Data from a diamesic comparison. Rassegna Italiana di Linguistica Applicata 2/3. 187–208. https://doi.org/10.5167/uzh-220284 · DOI
- Hoek, Jet & Scholman, Merel C.J. 2017. Evaluating discourse annotation: some recent insights and new approaches. In Proceedings of the 13th Joint ISO-ACL Workshop on Interoperable Semantic Annotation (Isa-13), 1−13. https://aclanthology.org/W17-7401/ (last accessed on 18.01.2024)
- Hoek, Jet & Scholman, Merel C.J. & Sanders, Ted J.M. 2021. Is there less annotator agreement when the discourse relation is underspecified?. In Proceedings of the First Workshop on Integrating Perspectives on Discourse Annotation, 1–6. Association for Computational Linguistics. http://doi.org/10.22028/D291-38635 · DOI
- Kiparsky, Carol & Kiparsky, Paul. 1971. Fact. In Steinberg, Danny D. & Jakobovitz, Leon A. (eds), Semantics: An Interdisciplinary Reading, 345−369. Cambridge: Cambridge University Press.
- Lombardi Vallauri, Edoardo. 2009. La struttura informativa. Forma e funzione negli enunciati linguistici. Roma: Carocci. http://hdl.handle.net/11590/179329
- Loewen, Shawn & Plonsky, Luke. 2016. An A-Z of Applied Linguistics Research Methods. London: Palgrave.
- Masia, Viviana & Garassino, Davide & Brocca, Nicola & de Saussure, Louis. 2023. Recalling presupposed information. Evidence from the online processing of presuppositions in political tweets. Pragmatics & Cognition 30:1. 92–119. https://doi.org/10.1075/pc.22011.mas · DOI
- Mercier, Hugo & Sperber, Dan. 2017. The Enigma of Reason. Boston, MA: Harvard University Press.
- Reboul, Anne. 2017. Communication and Cognition in the Evolution of Language. Oxford: Oxford University Press. https://doi.org/10.1093/acprof:oso/9780198747314.001.0001 · DOI
- Stalnaker, Robert C. 1973. Pragmatic Presuppositions. In Munitz, Milton Karl & Unger, Peter K. (eds), Semantics and Philosophy, 471−482. New York: New York University Press.
- Venables, William N. & Ripley, Brian D. 2002. Modern Applied Statistics with S (3rd edition). New York: Springer.
- Wickham, Hadley. 2016. ggplot2: Elegant Graphics for Data Analysis. New York: Springer.
- Vol. 1 No. 1 (2024): AI-Linguistica
Cited by
Sorted by citations
ChatGPT as a voting application in direct democracy
Reveilhac, M., & Morselli, D. (2025) · Swiss Political Science Review · doi:10.1111/spsr.12650 · Cited by 6 · see citing works
ChatGPT for President! Presupposed content in politicians versus GPT-generated texts
Garassino, D., Brocca, N., & Masia, V. (2025) · Applied Corpus Linguistics · doi:10.1016/j.acorp.2025.100156 · Cited by 2 · see citing works
Self-citationThe pragmatic competence of ChatGPT. An AI-assisted research project
Vasilescu, A. (2025) · Revue roumaine de linguistique · doi:10.59277/rrl.2025.3-4.15 · Cited by 1 · see citing works
Mayr, P. (2025) · Lebende Sprachen · doi:10.1515/les-2025-0014 · Cited by 1 · see citing works
From persuasion techniques to pragmatic mechanisms
Põldvere, N., Trnavac, R., & Alvestad, S. S. (2026) · Pragmatics and Society · doi:10.1075/ps.25118.pol
- Authors
- Garassino, D., Masia, V., Brocca, N., & Delorme Benites, A.
- Affiliations
- ZHAW Zurich University of Applied Sciences (CH)
- Roma Tre University (IT)
- Universität Innsbruck (AT)
- Venue
- AI-Linguistica. Linguistic Studies on AI-Generated Texts and Discourses
- Published
- 2024-07-08
- Volume
- 1
- Issue
- 1
- ISSN
- 2943-0070 (ISSN-L)
- OpenAlex
- W4400389449
- Cited by
- 4
- References
- 0
- Citations by year
- 2026: 1 · 2025: 3
- FWCI
- 11.15
- Type
- journal article
- Language
- en
- Open access
- diamond
- doi.org
- Licence
- Locations
- AI-Linguistica Linguistic Studies on AI-Generated Texts and Discourses (publishedVersion, cc-by-nc-sa)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- Zurich Open Repository and Archive (University of Zurich) (submittedVersion, cc-by-nc-nd)
- Topics
- Social Sciences › Arts and Humanities › Philosophy › Linguistics and Discourse Analysis
- Social Sciences › Linguistics and Language › Cultural and political discourse analysis
- Social Sciences › Arts and Humanities › Literature and Literary Theory › Discourse Analysis in Language Studies
- Keywords
- ChatGPT, large language models, political discourse, presupposition, implicit communication, discourse functions, political slogans, change-of-state verbs, immigration, European Union, manipulation, French politicians, Italian politicians, AI-generated text
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
