Are Stereotypes Leading LLMs' Zero-Shot Stance Detection ?

2025-10-22

AI Summary

This study investigates whether large language models (LLMs) exhibit stereotype-based biases in zero-shot stance detection—specifically, biases tied to sociolinguistic dialects (e.g., African American English) and textual complexity. We propose an automated annotation framework that jointly leverages readability metrics and dialect identification models to perform fine-grained attribute labeling on pre-annotated stance datasets. Using this framework, we systematically analyze spurious correlations between textual attributes and stance labels in LLM zero-shot predictions. Our empirical analysis reveals, for the first time, that LLMs significantly misattribute low-complexity texts as “pro-marijuana legalization” and erroneously associate African American English with “anti-Trump” stances. Beyond establishing a reproducible bias diagnostic paradigm, this work uncovers how representational biases in language models distort stance inference—providing critical empirical grounding for fairness evaluation and debiasing interventions. (149 words)

Technology Category

• Natural Language Processing: Ethics — Bias, Fairness, Transparency & Privacy
• Machine Learning: Large Multimodal Models (LMMs)
• Philosophy and Ethics of AI: Bias, Fairness & Equity

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Social Networks and Social Media: Fairness and bias in social network and social media analysis
• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling

Abstract

Large Language Models inherit stereotypes from their pretraining data, leading to biased behavior toward certain social groups in many Natural Language Processing tasks, such as hateful speech detection or sentiment analysis. Surprisingly, the evaluation of this kind of bias in stance detection methods has been largely overlooked by the community. Stance Detection involves labeling a statement as being against, in favor, or neutral towards a specific target and is among the most sensitive NLP tasks, as it often relates to political leanings. In this paper, we focus on the bias of Large Language Models when performing stance detection in a zero-shot setting. We automatically annotate posts in pre-existing stance detection datasets with two attributes: dialect or vernacular of a specific group and text complexity/readability, to investigate whether these attributes influence the model's stance detection decisions. Our results show that LLMs exhibit significant stereotypes in stance detection tasks, such as incorrectly associating pro-marijuana views with low text complexity and African American dialect with opposition to Donald Trump.

Problem

• Investigating stereotype bias in zero-shot stance detection
• Evaluating LLM bias towards social groups via dialect
• Assessing text complexity influence on stance classification

Innovation

• Automatically annotating posts with dialect attributes
• Measuring text complexity to assess bias influence
• Evaluating zero-shot stance detection stereotype patterns