LLMsAgainstHate @ NLU of Devanagari Script Languages 2025: Hate Speech Detection and Target Identification in Devanagari Languages via Parameter Efficient Fine-Tuning of LLMs

2024-12-22

AI Summary

Detecting online hate speech and identifying its targets in Devanagari-script languages—such as Hindi and Nepali—is hindered by scarce annotated data, the high computational cost of full fine-tuning large language models (LLMs), and a lack of specialized tooling. Method: This work pioneers the systematic application of parameter-efficient fine-tuning (PEFT) techniques—including LoRA and Adapter—to this low-resource multilingual setting, integrating them with multilingual LLMs (IndicBERT, mT5, and Phi-3) and leveraging cross-lingual transfer alongside task-oriented prompt design. Results: Evaluated on Thapa et al.’s (2025) bilingual Devanagari hate speech dataset, our approach achieves an F1 score of 86.3%, substantially outperforming both full fine-tuning and conventional classifiers. This study bridges a critical gap in fine-grained content safety for Devanagari languages and empirically validates the effectiveness and scalability of lightweight model adaptation for resource-constrained languages.

Technology Category

• Natural Language Processing: Safety and Robustness
• Machine Learning: Large Multimodal Models (LMMs)
• Computer Vision: Language and Vision

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Web Mining and Content Analysis: Large pretrained models with web data
• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling

Abstract

The detection of hate speech has become increasingly important in combating online hostility and its real-world consequences. Despite recent advancements, there is limited research addressing hate speech detection in Devanagari-scripted languages, where resources and tools are scarce. While large language models (LLMs) have shown promise in language-related tasks, traditional fine-tuning approaches are often infeasible given the size of the models. In this paper, we propose a Parameter Efficient Fine tuning (PEFT) based solution for hate speech detection and target identification. We evaluate multiple LLMs on the Devanagari dataset provided by (Thapa et al., 2025), which contains annotated instances in 2 languages - Hindi and Nepali. The results demonstrate the efficacy of our approach in handling Devanagari-scripted content.

Problem

• Hate Speech Detection
• Low-resource Languages
• Large Language Models Adaptation

Innovation

• Parameter-Efficient Fine-Tuning
• Hate Speech Detection
• Indic Languages