Minguinho-zeze/

dialect-penalty-t2i

Paired Standard American English (SAE) and dialect prompt dataset from "Not Safe for All: Auditing the Dialect Penalty in Text-to-Image Safety Pipelines" (Findings of EMNLP 2026). Code, and all experiment results reported in the paper: https://github.com/minguinho26/dialect-penalty-t2i Dialects: AAVE, Chicano English (ChcE), Colloquial Singapore English (CollSgE), Indian English (IndE), Jamaican English (JamE). from datasets import… See the full description on the dataset page: https://huggingface.co/datasets/Minguinho-zeze/dialect-penalty-t2i.