Queer Bias in Natural Language Processing

Towards More Expansive Frameworks of Gender and Sexuality in NLP Bias Research

Abstract

Research in the growing field of NLP bias has made significant progress related to characteristics such as race and (binary) gender. However, bias with respect to queer communities and experiences has been critically underexplored. In this paper, I review sources of bias and describe the unique risks that biased NLP systems pose for queer individuals. I break down the social and computational factors which act as barriers to research in queer bias and discuss the importance of continued involvement with queer stakeholders within the research process. I then review common models of gender and sexuality in NLP bias research and argue how cis- and heteronormative assumptions as the standard in NLP academic frontiers continues to perpetuate research which excludes queer experiences. Finally, I review emerging methods and successes in evaluating queer bias in NLP systems, setting out recommendations on how to expand from these works and pointing towards a framework for future work in queer bias.