Hyderabadi Pearls at Multilingual Counterspeech Generation : HALT : Hate Speech Alleviation using Large Language Models and Transformers (ACL Anthology)

The potential of employing fine-tuned Large Language Models (LLMs) to produce counter-narratives (CNs) in order to counter hate speech (HS) is examined in this research. Using the ML_MTCONAN_KN dataset, it provides hate speech and counter-narrative pairs in different languages with an emphasis on English and Basque. The performance of Mistral, Llama, and a Llama-based LLM optimized on a Basque language dataset for CN generation is compared in this paper. In addition to conventional metrics like ROUGE-L, BLEU, BERTScore, and others, the created CNs are assessed using JudgeLM, an LLM that assesses other LLMs in open-ended situations. The findings show that optimized LLMs may provide contextually appropriate CNs of high quality for low-resource languages that are on par with human-generated answers. This makes a substantial contribution to the fight against hate speech online in a variety of linguistic contexts.

https://aclanthology.org/2025.mcg-1.8