wefe.fetch_debias_multiclass() Dict[str, Union[List[str], list]][source]

Fetch the word sets used in the paper Black Is To Criminals as Caucasian Is To Police: Detecting And Removing Multiclass Bias In Word Embeddings.

This dataset contains gender (male, female), ethnicity (asian, black, white) and religion (christianity, judaism and islam) word sets. This helper allow accessing independently to each of the word sets (to be used as target or attribute sets in metrics) as well as to access them in the original format (to be used in debiasing methods). The dictionary keys whose names contain definitional sets and analogies templates are the keys that point to the original format focused on debiasing.


A dictionary in which each key correspond to the name of the set and its values correspond to the word set.


[1]: Thomas Manzini, Lim Yao Chong,Alan W Black, and Yulia Tsvetkov.
Black is to Criminal as Caucasian is to Police: Detecting and Removing Multiclass
Bias in Word Embeddings.
In Proceedings of the 2019 Conference of the North American Chapter of the
Association for Computational Linguistics:
Human Language Technologies, Volume 1 (Long and Short Papers), pages 615–621,
Minneapolis, Minnesota, June 2019. Association for Computational Linguistics.