Most hate speech detection datasets focus on English; this dataset enables models to learn culturally-specific patterns in Mexican Spanish video content, which is essential since hate speech is deeply tied to local context and language nuances.
MexHat is a new video dataset with ~1,000 annotated clips for detecting hate speech in Mexican Spanish. It addresses the lack of non-English resources by capturing linguistic and cultural context specific to Mexico, with annotations for both general categories (offensive vs. hate speech) and fine-grained hate speech subtypes.