- Models that are trained with certain datasets affects which people the models work for
- Racial classifications can easily become widespread if the dataset is used frequently

Consequences of Racial Bias in NLP Datasets

It is important to analyze datasets to prevent the spread of racial bias because:

- The type of person creating the data determines who the data will work for (ex. certain facial recognition algorithms working better on lighter-skinned people)
- NLP research produces technology that affects people's lives. Because of this, the prevention of racially-motivated technologies is beneficial.

Importance of Analyzing Racial Bias in NLP

- Racial bias occurs in datasets when it associates names with a certain race
- In addition to what the dataset contains, it is important to consider who the data was authored by

University of Virginia, Charlottesville

- Race and language go hand-in-hand, yet not much research has been done towards race.
- This is the reason that race-related bias in NLP is overlooked.
- NLP systems can facilitate the encoding of racial biases

Racism in NLP

Anjalie Field, Su Lin Blodgett, Zeerak Waseem, and Yulia Tsvetkov. 2021. A Survey of Race, Racism, and Anti-Racism in NLP. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers), pages 1905–1925, Online. Association for Computational Linguistics.

A Survey of Race, Racism, and Anti-Racism in NLP

- racial stereotypes and prejudices are communicated through language
- language is very indicative of how racial categorizations are different in different contexts


Race and Language

Racial bias in NLP datasets

- Racial biases of model outputs are changed by model instances or architectures
- The word embedding associations that are centered around words related to race change based on what model architecture is used. 

Learn Before

Related

Learn After