AfriHate: A Multilingual Collection of Hate Speech and Abusive Language Datasets for African Languages (ACL Anthology)

To effectively comprehend and manage hate speech and abusive language, sociocultural context is necessary. Moderation is frequently lacking or depends on out-of-context keyword detection in many regions of the Global South, which results in censorship and the disregard of specific hate campaigns against minorities. These problems are caused by a lack of community engagement and a shortage of high-quality data in the local languages. The researchers address this by presenting AfriHate, a multilingual collection of local speakers’ annotations of tweets in 15 African languages. The authors offer classification baselines and address the difficulties in building datasets, demonstrating that performance varies by language and that multilingual models perform better in low-resource settings.

https://aclanthology.org/2025.naacl-long.92