AI models struggle to detect implicit arguments in political discourse

A new research paper explores the challenges of detecting enthymemes, which are arguments with unstated premises or conclusions, particularly in controversial political discourse on social media. The study found that even fine-tuned models struggle to achieve high accuracy (F1 score around 0.67) when annotator disagreement is high, as is common with these types of arguments. The researchers propose alternative evaluation metrics that consider the full distribution of annotator votes and introduce a model that better captures this disagreement. They also observed that current models rely heavily on explicit argumentative structure, often missing implicit arguments that require reader inference, which has implications for content moderation systems trained on majority labels. AI

IMPACT Highlights limitations of current AI in understanding nuanced human argumentation, impacting content moderation and persuasive technology.

RANK_REASON Academic paper on NLP and discourse analysis. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• alphaXiv
• arXiv
• CatalyzeX
• Connected Papers
• DagsHub
• Gotit.pub
• Hugging Face
• Influence Flower
• Litmaps
• ScienceCast
• Scite

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →