Harms of Gender Exclusivity and Challenges in Non-Binary Representation in Language Technologies

Abstract

Content Warning: This paper contains examples of stereotypes and associations, misgendering, erasure, and other harms that could be offensive and triggering to trans and nonbinary individuals.Gender is widely discussed in the context of language tasks and when examining the stereotypes propagated by language models. However, current discussions primarily treat gender as binary, which can perpetuate harms such as the cyclical erasure of non-binary gender identities. These harms are driven by model and dataset biases, which are consequences of the non-recognition and lack of understanding of non-binary genders in society. In this paper, we explain the complexity of gender and language around it, and survey non-binary persons to understand harms associated with the treatment of gender as binary in English language technologies. We also detail how current language representations (e.g., GloVe, BERT) capture and perpetuate these harms and related challenges that need to be acknowledged and addressed for representations to equitably encode gender information.

Welcome to the Modern World of Pronouns: Identity-Inclusive Natural Language Processing beyond Gender

“…While there are some works in NLP on genderinclusion (e.g., Dev et al, 2021) and gender bias in static (e.g., Bolukbasi et al, 2016;Gonen and Goldberg, 2019;Lauscher et al, 2020, inter alia) and contextualized (e.g., Kurita et al, 2019;Bordia and Bowman, 2019;Lauscher et al, 2021, inter alia) language representations as well as works focusing on specific gender bias in downstream tasks, e.g., natural language inference (Dev et al, 2020) and co-reference resolution (e.g., Rudinger et al, 2018;Webster et al, 2018), we are not aware of any work that deals with the broader field of identity-inclusion. Thus, there is no other NLP work that deals with a larger variety of pronouns and acknowledges pronouns as an open word class.…”

“…For surveys on the general topic of unfair bias in NLP we refer to Blodgett et al (2020) andShah et al (2020). Recently, Dev et al (2021) pointed broadly at the harms (Barocas et al, 2017) arising from gender-exclusivity in NLP. They surveyed queer individuals and assessed non-binary representations in existing data set and language representations.…”

“…According to Cao and Daumé III (2021), pronouns are a way of expressing referential gender. Referring to an indi- vidual with sets of pronouns they do not identify with, e.g., resulting in misgendering, is considered harmful (Dev et al, 2021).…”

“…3 This "social push" to respect diverse gender identities also affects aspects of NLP. Recent studies have pointed out the potential harms from the cur-rent lack of non-binary representation in NLP data sets, embeddings, and tasks (Cao and Daumé III, 2021;Dev et al, 2021), and the related issue of unfair stereotyping of queer individuals (Barikeri et al, 2021). However, the research landscape on modern pronoun usage is still surprisingly scarce, hindering progress for a fair and inclusive NLP.…”

Handling Bias in Toxic Speech Detection: A Survey

“…Existing literature in the area of gender debasing in NLP as well as in toxicity detection has evaluated gender as binary (male vs female). In their recent work, Dev et al [22] provided a general overview of how non-binary individuals are at risk of erasure and misgendering at the hands of existing language models. These harms trickle down to the task of toxicity detection as well, and unfortunately, its full extent has not been studied yet.…”