Analyzing Bias in False Refusal Behavior of Large Language Models for Hate Speech Detoxification

TL;DR

This paper investigates the false refusal behavior of large language models in hate speech detoxification, revealing biases towards certain groups and proposing a translation-based mitigation strategy to reduce refusals.

Contribution

It systematically analyzes biases causing false refusals in LLMs and introduces a simple cross-translation method to mitigate this issue.

Findings

LLMs disproportionately refuse toxic inputs targeting specific groups.

Multilingual datasets show lower false refusal rates but still exhibit biases.

Cross-translation significantly reduces false refusals while maintaining content integrity.

Abstract

While large language models (LLMs) have increasingly been applied to hate speech detoxification, the prompts often trigger safety alerts, causing LLMs to refuse the task. In this study, we systematically investigate false refusal behavior in hate speech detoxification and analyze the contextual and linguistic biases that trigger such refusals. We evaluate nine LLMs on both English and multilingual datasets, our results show that LLMs disproportionately refuse inputs with higher semantic toxicity and those targeting specific groups, particularly nationality, religion, and political ideology. Although multilingual datasets exhibit lower overall false refusal rates than English datasets, models still display systematic, language-dependent biases toward certain targets. Based on these findings, we propose a simple cross-translation strategy, translating English hate speech into Chinese for detoxification and back, which substantially reduces false refusals while preserving the original content, providing an effective and lightweight mitigation approach.