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

TL;DR

This paper investigates how large language models modify speeches to influence persuasiveness, revealing systematic stylistic changes rather than genuine optimization of persuasive impact, through analysis of linguistic features and rhetorical devices.

Contribution

Introduces a novel methodology and interpretable feature set to analyze linguistic shifts in LLM-modified speeches, focusing on persuasiveness and rhetorical strategies.

Findings

GPT-4o applies systematic stylistic modifications

Manipulates emotional lexicon and syntactic structures

Does not optimize persuasiveness in a human-like manner

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.