Politicians vs ChatGPT. A study of presuppositions in French and Italian political communication

Abstract: 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 LLMs.

• The paper reveals that ChatGPT exhibits a higher frequency of potentially manipulative presuppositions than human politicians.
• It finds that ChatGPT favors change-of-state verbs while politicians predominantly use definite descriptions to anchor their assumptions.
• The study underscores the need for stricter AI governance to prevent inadvertent persuasive manipulation in sensitive political discourse.

Understanding Presuppositions in Political Communication: Politicians versus ChatGPT

The study entitled "Politicians vs ChatGPT: A study of presuppositions in French and Italian political communication" embarks on an empirical comparison between human-generated political discourse and AI-generated text using ChatGPT 3.5. The focal point of the analysis is presuppositions in political communication, given their potential as a linguistic feature of manipulation. Recognized often in the literature as tools for implicit persuasion, presuppositions allow the embedding of assumed knowledge into discourse, potentially impacting the audience's beliefs in subtler ways than explicit statements might.

Key Findings of the Study

The researchers systematically examined speeches from prominent French and Italian politicians—Emmanuel Macron, Marine Le Pen, Nicola Zingaretti, and Giorgia Meloni—contrasted against text generated by ChatGPT prompts designed to emulate these political figures. The findings are multi-faceted, centering around the frequency, form, and function of presuppositions in these various text sources.

1. Frequency of Potentially Manipulative Presuppositions (PMPs):
   • It was observed that texts generated by ChatGPT contain a higher average frequency of PMPs compared to those delivered by politicians. This result signals the AI's tendency toward creating content with presupposed beliefs or conditions that can border on manipulation.
2. Form of Presupposition Triggers:
   • The distribution of presupposition triggers, specifically change-of-state verbs and definite descriptions, differed starkly between AI and human-generated texts. ChatGPT showed a marked preference for change-of-state verbs, while politicians more frequently utilized definite descriptions to anchor their suppositions.
3. Function of Presuppositions in Discourse:
   • The analysis revealed divergences in the discourse functions of presuppositions, where stance-taking was predominantly found within AI-generated texts. In contrast, politicians' speeches tended to diversify across several discourse functions, including criticism and self-promotion.

Implications and Future Directions

The study emphasizes the critical need for scrutiny over AI-generated communication, especially in politically-sensitive contexts. While LLMs such as ChatGPT exhibit sophisticated linguistic capabilities, they also reveal a propensity toward creating text that might inherently possess manipulative qualities due to vagueness and repetition of slogans.

For political communication, the findings are significant in illustrating the potential for AI systems to replicate, inadvertently or otherwise, the manipulative tactics found in human political rhetoric. This capability poses both opportunities for AI in generating persuasive narratives and distinct challenges, particularly regarding misinformation and biased discourse.

At the theoretical level, this research delineates the nuanced understanding of pragmatic competencies in AI systems. ChatGPT's ability to generate presuppositions suggests a form of linguistic creativity previously attributed primarily to human authors. Given the magnitude and impact of political journalism, applying LLMs in such domains necessitates stringent model governance to avoid the unintentional enhancement of persuasive propaganda through computational linguistics.

Future research could expand the corpus size, integrate non-European languages, and explore evolving AI models like ChatGPT 4.0 to measure progression in capturing pragmatic subtleties. Additionally, a more comprehensive analysis examining how AI's use of presuppositions varies with different political personalities could further illuminate AI's potential replicability of human-like discourse patterns.

Conclusion

The study achieves a compelling exploration of presuppositions within political speech, contrasting human articulacy against AI-generated narratives. It highlights the intricate dance of persuasion and manipulation within political discourse and the emergent role of AI in shaping communicative strategies. As AI continues to evolve and integrate more into the fabric of digital communication, maintaining an informed perspective on its influence and underpinnings remains indispensable.