"I'm fully who I am": Towards Centering Transgender and Non-Binary Voices to Measure Biases in Open Language Generation

Transgender and non-binary (TGNB) individuals disproportionately experience discrimination and exclusion from daily life. Given the recent popularity and adoption of language generation technologies, the potential to further marginalize this population only grows. Although a multitude of NLP fairness literature focuses on illuminating and addressing gender biases, assessing gender harms for TGNB identities requires understanding how such identities uniquely interact with societal gender norms and how they differ from gender binary-centric perspectives. Such measurement frameworks inherently require centering TGNB voices to help guide the alignment between gender-inclusive NLP and whom they are intended to serve. Towards this goal, we ground our work in the TGNB community and existing interdisciplinary literature to assess how the social reality surrounding experienced marginalization by TGNB persons contributes to and persists within Open Language Generation (OLG). By first understanding their marginalization stressors, we evaluate (1) misgendering and (2) harmful responses to gender disclosure. To do this, we introduce the TANGO dataset, comprising of template-based text curated from real-world text within a TGNB-oriented community. We discover a dominance of binary gender norms within the models; LLMs least misgendered subjects in generated text when triggered by prompts whose subjects used binary pronouns. Meanwhile, misgendering was most prevalent when triggering generation with singular they and neopronouns. When prompted with gender disclosures, LLM text contained stigmatizing language and scored most toxic when triggered by TGNB gender disclosure. Our findings warrant further research on how TGNB harms manifest in LLMs and serve as a broader case study toward concretely grounding the design of gender-inclusive AI in community voices and interdisciplinary literature.

[1]  B. Mustanski,et al.  Systems of cissexism and the daily production of stress for transgender and gender diverse people. , 2021, International journal of transgender health.

[2]  Ho-Chun Herbert Chang,et al.  Towards WinoQueer: Developing a Benchmark for Anti-Queer Bias in Large Language Models , 2022, ArXiv.

[3]  Jiliang Tang,et al.  Detecting Harmful Online Conversational Content towards LGBTQIA+ Individuals , 2022, ArXiv.

[4]  Xi Victoria Lin,et al.  OPT: Open Pre-trained Transformer Language Models , 2022, ArXiv.

[5]  Dirk Hovy,et al.  Welcome to the Modern World of Pronouns: Identity-Inclusive Natural Language Processing beyond Gender , 2022, COLING.

[6]  Kai-Wei Chang,et al.  On Measures of Biases and Harms in NLP , 2021, AACL/IJCNLP.

[7]  Weizhu Chen,et al.  What Makes Good In-Context Examples for GPT-3? , 2021, DEELIO.

[8]  K. Balsam,et al.  “Development of the Gender Minority Stress and Resilience Measure.”: Correction. , 2015, Psychology of Sexual Orientation and Gender Diversity.

[9]  Federico Bianchi,et al.  Measuring Harmful Sentence Completion in Language Models for LGBTQIA+ Individuals , 2022, LTEDI.

[10]  Li Dong,et al.  Why Can GPT Learn In-Context? Language Models Secretly Perform Gradient Descent as Meta-Optimizers , 2023, ACL.

[11]  B. Munson,et al.  Perceiving gender while perceiving language: Integrating psycholinguistics and gender theory. , 2021, Wiley interdisciplinary reviews. Cognitive science.

[12]  Ashwin,et al.  Rebuilding Trust: Queer in AI Approach to Artificial Intelligence Risk Management , 2021, ArXiv.

[13]  Po-Sen Huang,et al.  Challenges in Detoxifying Language Models , 2021, EMNLP.

[14]  Anaelia Ovalle,et al.  Harms of Gender Exclusivity and Challenges in Non-Binary Representation in Language Technologies , 2021, EMNLP.

[15]  Pushpak Bhattacharyya,et al.  A Large-Scale Dataset for Motivational Dialogue System: An Application of Natural Language Generation to Mental Health , 2021, 2021 International Joint Conference on Neural Networks (IJCNN).

[16]  Yang Trista Cao,et al.  Toward Gender-Inclusive Coreference Resolution: An Analysis of Gender and Bias Throughout the Machine Learning Lifecycle* , 2021, CL.

[17]  Goran Glavas,et al.  RedditBias: A Real-World Resource for Bias Evaluation and Debiasing of Conversational Language Models , 2021, ACL.

[18]  Li Lucy,et al.  Gender and Representation Bias in GPT-3 Generated Stories , 2021, NUSE.

[19]  Dirk Hovy,et al.  HONEST: Measuring Hurtful Sentence Completion in Language Models , 2021, NAACL.

[20]  Kai-Wei Chang,et al.  Societal Biases in Language Generation: Progress and Challenges , 2021, ACL.

[21]  Stella Biderman,et al.  GPT-Neo: Large Scale Autoregressive Language Modeling with Mesh-Tensorflow , 2021 .

[22]  Inioluwa Deborah Raji,et al.  You Can't Sit With Us: Exclusionary Pedagogy in AI Ethics Education , 2021, FAccT.

[23]  Kai-Wei Chang,et al.  BOLD: Dataset and Metrics for Measuring Biases in Open-Ended Language Generation , 2021, FAccT.

[24]  Charles Foster,et al.  The Pile: An 800GB Dataset of Diverse Text for Language Modeling , 2020, ArXiv.

[25]  Hinrich Schütze,et al.  It’s Not Just Size That Matters: Small Language Models Are Also Few-Shot Learners , 2020, NAACL.

[26]  Hui Liu,et al.  Mitigating Gender Bias for Neural Dialogue Generation with Adversarial Learning , 2020, EMNLP.

[27]  Sabrina Burtscher,et al.  "But where would I even start?": developing (gender) sensitivity in HCI research and practice , 2020, MuC.

[28]  Nanyun Peng,et al.  Towards Controllable Biases in Language Generation , 2020, FINDINGS.

[29]  Qiongkai Xu,et al.  Adhering, Steering, and Queering: Treatment of Gender in Natural Language Generation , 2020, CHI.

[30]  E. Fosch-Villaronga,et al.  Queering machines , 2020 .

[31]  Jeremy Blackburn,et al.  The Pushshift Reddit Dataset , 2020, ICWSM.

[32]  J. Weston,et al.  Queens Are Powerful Too: Mitigating Gender Bias in Dialogue Generation , 2019, EMNLP.

[33]  Munmun De Choudhury,et al.  The Language of LGBTQ+ Minority Stress Experiences on Social Media , 2019, Proc. ACM Hum. Comput. Interact..

[34]  B. Mustanski,et al.  Coping with discrimination: The insidious effects of gender minority stigma on depression and anxiety in transgender individuals. , 2019, Journal of clinical psychology.

[35]  Nanyun Peng,et al.  The Woman Worked as a Babysitter: On Biases in Language Generation , 2019, EMNLP.

[36]  Omer Levy,et al.  RoBERTa: A Robustly Optimized BERT Pretraining Approach , 2019, ArXiv.

[37]  Ilya Sutskever,et al.  Language Models are Unsupervised Multitask Learners , 2019 .

[38]  A. Narayanan,et al.  Fairness and Machine Learning Limitations and Opportunities , 2018 .

[39]  Bronwyn M. Bjorkman Singular they and the syntactic representation of gender in English , 2017 .

[40]  Radha Poovendran,et al.  Deceiving Google's Perspective API Built for Detecting Toxic Comments , 2017, ArXiv.

[41]  Arvind Narayanan,et al.  Semantics derived automatically from language corpora contain human-like biases , 2016, Science.

[42]  T. Joiner,et al.  Suicidal Ideation in Transgender People: Gender Minority Stress and Interpersonal Theory Factors , 2017, Journal of abnormal psychology.

[43]  Aidong Zhang,et al.  A Survey on Context Learning , 2017, IEEE Transactions on Knowledge and Data Engineering.

[44]  S. Reisner,et al.  Expecting Rejection: Understanding the Minority Stress Experiences of Transgender and Gender-Nonconforming Individuals , 2016, Transgender health.

[45]  Adam Tauman Kalai,et al.  Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings , 2016, NIPS.

[46]  A. Ramos-Soto,et al.  On the role of linguistic descriptions of data in the building of natural language generation systems , 2016, Fuzzy Sets Syst..

[47]  Kevin A. McLemore A Minority Stress Perspective on Transgender Individuals’ Experiences With Misgendering , 2016 .

[48]  I. Hickie,et al.  Moderator Assistant: A Natural Language Generation-Based Intervention to Support Mental Health via Social Media , 2015 .

[49]  A. Joshi,et al.  Likert Scale: Explored and Explained , 2015 .

[50]  P. Hegarty,et al.  Misgendering in English language contexts: Applying non-cisgenderist methods to feminist research , 2013 .

[51]  E. Coleman,et al.  Stigma, mental health, and resilience in an online sample of the US transgender population. , 2013, American journal of public health.

[52]  Ruth Filik,et al.  “They” as a gender-unspecified singular pronoun: Eye tracking reveals a processing cost , 2007, Quarterly journal of experimental psychology.

[53]  Jeanette M. Silveira Generic masculine words and thinking , 1980 .

[54]  M. C. Templin Certain language skills in children : their development and interrelationships , 1957 .