Code-Mixed Telugu-English Hate Speech Detection

TL;DR

The paper investigates hate speech detection in Telugu, a low-resource language, by evaluating multiple transformer-based architectures with LoRA fine-tuning and by translating Telugu text to English to exploit richer English embeddings. Hindi-Abusive-MuRIL emerges as the most robust model across both original and translated data, highlighting the value of abusive-language pretraining for cross-lingual transfer. Translation generally raises accuracy but can decrease F1, underscoring the trade-offs in preserving Telugu-specific nuances. The work advocates hybrid multilingual and monolingual fine-tuning and points to future directions in expanding Telugu datasets and refining cross-lingual representations for robust hate speech detection.

Abstract

Hate speech detection in low-resource languages like Telugu is a growing challenge in NLP. This study investigates transformer-based models, including TeluguHateBERT, HateBERT, DeBERTa, Muril, IndicBERT, Roberta, and Hindi-Abusive-MuRIL, for classifying hate speech in Telugu. We fine-tune these models using Low-Rank Adaptation (LoRA) to optimize efficiency and performance. Additionally, we explore a multilingual approach by translating Telugu text into English using Google Translate to assess its impact on classification accuracy. Our experiments reveal that most models show improved performance after translation, with DeBERTa and Hindi-Abusive-MuRIL achieving higher accuracy and F1 scores compared to training directly on Telugu text. Notably, Hindi-Abusive-MuRIL outperforms all other models in both the original Telugu dataset and the translated dataset, demonstrating its robustness across different linguistic settings. This suggests that translation enables models to leverage richer linguistic features available in English, leading to improved classification performance. The results indicate that multilingual processing can be an effective approach for hate speech detection in low-resource languages. These findings demonstrate that transformer models, when fine-tuned appropriately, can significantly improve hate speech detection in Telugu, paving the way for more robust multilingual NLP applications.

Figures (1)

• Figure 1: Distribution of hate and non-hate speech in the training and testing datasets. The dataset is balanced to ensure fair evaluation of model performance.