Evaluating VLMs on Multimodal Aristotelian Persuasion Tasks

2026-08-02

AI Summary

This study addresses the lack of systematic evaluation of current vision-language models (VLMs) on complex multimodal reasoning tasks, particularly in understanding persuasive content grounded in Aristotle’s rhetorical triad—Logos (logical appeal), Ethos (ethical credibility), and Pathos (emotional resonance). The work formalizes this rhetorical framework into a computable multimodal benchmark and introduces the first evaluation suite specifically targeting these three dimensions. Leveraging the ImageArg dataset, the authors fine-tune and evaluate the Qwen family of VLMs, using F1 score as the primary metric. Results reveal that Qwen3 significantly outperforms others on Logos and Pathos tasks, whereas Qwen2 achieves superior performance on the more challenging Ethos detection, highlighting distinct capabilities and limitations of contemporary VLMs in high-level semantic understanding of persuasive multimodal content.

Abstract

Vision Language Models (VLMs) have demonstrated exceptional performance across various tasks. However, they have not yet been thoroughly evaluated on more complex tasks. The Persuasion Model, conceived by Aristotle, resembles a triangle shape, which highlights its inherent challenges related to personal biases. To assess the progress of VLMs on these complex tasks, we use the ImageArg datasets, focusing on the Logos, Ethos, and Pathos detection tasks. Our findings indicate that models from the Qwen family achieve improved F1 scores, with Qwen3 performing exceptionally well on the Logos and Pathos tasks, while Qwen2 exhibits competitive performance on the more complex Ethos detection task. We release the code to foster research in this direction.

Problem

• Vision Language Models
• Multimodal Persuasion
• Aristotelian Rhetoric
• Ethos Detection
• Logos and Pathos

Innovation

• Vision Language Models
• Aristotelian Persuasion
• Multimodal Reasoning
• Ethos-Logos-Pathos
• ImageArg Dataset