Can Prompting LLMs Unlock Hate Speech Detection across Languages? A Zero-shot and Few-shot Study

TL;DR

This paper investigates the effectiveness of multilingual large language models in detecting hate speech across eight non-English languages using zero-shot and few-shot prompting, highlighting the importance of prompt design and generalization capabilities.

Contribution

It provides a comprehensive evaluation of LLM prompting techniques for hate speech detection across multiple languages, comparing them to traditional fine-tuned models and emphasizing prompt customization.

Findings

Prompting techniques are crucial for performance.

Zero-shot and few-shot methods generalize better on functional tests.

Prompt design significantly impacts multilingual hate speech detection.

Abstract

Despite growing interest in automated hate speech detection, most existing approaches overlook the linguistic diversity of online content. Multilingual instruction-tuned large language models such as LLaMA, Aya, Qwen, and BloomZ offer promising capabilities across languages, but their effectiveness in identifying hate speech through zero-shot and few-shot prompting remains underexplored. This work evaluates LLM prompting-based detection across eight non-English languages, utilizing several prompting techniques and comparing them to fine-tuned encoder models. We show that while zero-shot and few-shot prompting lag behind fine-tuned encoder models on most of the real-world evaluation sets, they achieve better generalization on functional tests for hate speech detection. Our study also reveals that prompt design plays a critical role, with each language often requiring customized prompting techniques to maximize performance.