The Anatomy of Speech Persuasion: Linguistic Shifts in LLM-Modified Speeches

2025-06-23

AI Summary

This study investigates how large language models (LLMs) internally conceptualize “persuasiveness” in public speaking—a dimension poorly understood in current LLM research. Method: Leveraging authentic transcripts from the French “180-Second PhD” speech competition, we design controlled prompt-based tasks to enhance or attenuate persuasiveness, using GPT-4o as the test model. We introduce an interpretable feature set integrating rhetorical devices (e.g., rhetorical questions, parallelism) and discourse markers (e.g., “admittedly,” “the crux lies in”), coupled with fine-grained linguistic analysis. Contribution/Results: We find that GPT-4o modulates persuasiveness systematically—not via human-like logical argumentation or credibility construction—but primarily through affective lexical polarity, syntactic choices (e.g., ratios of interrogatives/exclamatives), and prosodic cues (e.g., pause placement). This work provides the first attributable, quantifiable, and interpretable analysis of LLM behavior along the persuasiveness dimension, establishing a novel methodology for LLM stylistic modeling and human–AI rhetorical collaboration.

Technology Category

• Natural Language Processing: Sentiment Analysis, Stylistic Analysis, and Argument Mining
• Machine Learning: Large Multimodal Models (LMMs)
• Cognitive Modeling & Cognitive Systems: Affective Computing

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information

Abstract

This study examines how large language models understand the concept of persuasiveness in public speaking by modifying speech transcripts from PhD candidates in the "Ma These en 180 Secondes" competition, using the 3MT French dataset. Our contributions include a novel methodology and an interpretable textual feature set integrating rhetorical devices and discourse markers. We prompt GPT-4o to enhance or diminish persuasiveness and analyze linguistic shifts between original and generated speech in terms of the new features. Results indicate that GPT-4o applies systematic stylistic modifications rather than optimizing persuasiveness in a human-like manner. Notably, it manipulates emotional lexicon and syntactic structures (such as interrogative and exclamatory clauses) to amplify rhetorical impact.

Problem

• How LLMs modify speeches to enhance persuasiveness
• Analyzing linguistic shifts in GPT-4o modified speeches
• Identifying systematic stylistic changes in persuasive speech generation

Innovation

• GPT-4o modifies speech persuasiveness systematically
• Analyzes rhetorical devices and discourse markers
• Manipulates emotional lexicon and syntactic structures