Participants are asked to classify a YouTube user’s comment using the corresponding video title and description as contextual information. The challenge is organized into three subtasks:
- Subtask (A) Stereotype Detection: A binary classification task checking whether the comment contains a stereotype about LGBTQIA+ people.
- Subtask (B) Hate Speech Classification: A multiclass classification task distinguishing whether the comment expresses (i) explicit hate, (ii) implicit hate, or (iii) non-hateful or supportive content toward LGBTQIA+ individuals.
- Subtask (C) Target & Identity Identification: A multilabel, multiclass task that, for hateful comments only, identifies both the target type (individual or group) and any specific LGBTQIA+ identities referenced in the comment.
The task covers 4 languages across different cultural and geographical contexts:
- English: A high-resource global lingua franca with broad online coverage, enabling comparisons with many existing NLP benchmarks.
- Italian: A Romance mid‑resource language that extends earlier work and offers continuity within European research on LGBTQIA+ hate speech.
- Dutch: A mid‑resource Germanic language underrepresented in queer-focused NLP, providing Northern/Western European perspectives.
- Persian: A low‑resource Indo‑Iranian language that introduces culturally distinct patterns of discourse rarely covered in queer research. Note that the Persian data will be released for testing only.