The field of hate speech detection is moving towards leveraging large language models and multimodal approaches to improve detection accuracy in low-resource languages and settings. Recent work has focused on developing novel frameworks and datasets that can handle multilingual hate speech detection, including the use of translation-based approaches and contrastive learning to jointly align audio and text representations. These advancements have shown promising results, with improved performance in detecting hate speech in languages such as Urdu, Spanish, and low-resource Indian languages. Furthermore, the use of large language models has been explored for toxic language detection in low-resource languages, including Serbian, Croatian, and Bosnian. Noteworthy papers in this area include:

• Multimodal Zero-Shot Framework for Deepfake Hate Speech Detection in Low-Resource Languages, which introduces a novel multimodal framework for hate speech detection in deepfake audio.
• ToxSyn-PT: A Large-Scale Synthetic Dataset for Hate Speech Detection in Portuguese, which presents a large-scale Portuguese corpus for fine-grained hate-speech classification.
• Large Language Models for Toxic Language Detection in Low-Resource Balkan Languages, which evaluates the performance of large language models in detecting toxic comments in low-resource languages.