Sesotho Hate Speech Dataset with Type and Target Annotations

This dataset contains 8,485 Sesotho-language text records annotated using three hierarchical levels: Label (Hate or Non-Hate), Type (Identity-Attack, Insult, Threat or None), and Target (Gender, Race, Religion, Nationality, Disability, Other or None). It includes reversible numeric encodings for machine-learning applications. The dataset supports research on hate-speech detection, natural language processing and low-resource African languages. It contains offensive, discriminatory and violent language and should be used responsibly and ethically.