Can Prompting LLMs Unlock Hate Speech Detection across Languages? A Zero-shot and Few-shot Study
Faeze Ghorbanpour, Daryna Dementieva, Alexander Fraser
TL;DR
This study probes hate speech detection across eight non-English languages using multilingual instruction-tuned LLM prompts, comparing zero-shot and few-shot prompting to fine-tuned encoders. It evaluates a broad suite of prompts and prompts combinations on real-world datasets and the multilingual HateCheck functional benchmark, revealing that prompt design and language-specific strategies critically affect performance. While prompting generally trails fine-tuned encoders on real-world data, it shows stronger generalization on functional tests and can be competitive in low-data scenarios; few-shot prompts often boost functional performance. The work highlights the potential and limits of prompting-based detection for multilingual hate speech, underscoring the need for language-aware prompts and suggesting encoder fine-tuning as the preferred approach when ample labeled data are available.
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.