Pith. sign in

REVIEW 3 major objections 5 minor 50 references

Analyzing values about gendered language reform in LLMs' revisions

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read LLMs neutralise gendered role nouns in revision, and context modulates how often.

desk verdict Solid empirical study of LLM revision behavior with a real but fixable coding ambiguity in the nonbinary-referent condition. read the letter →

arxiv 2505.21378 v1 pith:QBL3KIK5 submitted 2025-05-27 cs.CL

Jules Watson , Xi Wang , Raymond Liu , Suzanne Stevenson , Barend Beekhuizen This is my paper
classification cs.CL
keywords genderedlanguagereformrolenounsLLMtextrevisionneutralizationvaluealignmentideologiesgender-neutralcontexteffects
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper asks whether large language models, when asked to revise everyday sentences, treat gendered role nouns the way feminist and trans-inclusive language reforms recommend. Based on prompts built from naturalistic 'About Me' sentences and 50 role-noun sets (neutral/feminine/masculine variants such as firefighter/firewoman/fireman), the authors claim that four instruction-tuned LLMs follow a neutralization strategy: gendered variants are replaced by neutral forms more often than neutral forms are replaced. They further claim this strategy is context-dependent, being stronger for nonbinary and they/them referents, stronger when gender is explicitly declared, and shaped by stereotypical gender associations of the sentence, and that models' justifications express coherent value themes such as inclusive and modern versus natural and standard. The paper positions these results as evidence that value alignment transmits sociolinguistic reform values into routine LLM use, while applying them unevenly across contexts.

What carries the argument

The central object is a controlled revision prompt: a preamble naming the referent through pronoun usage, pronoun declaration, or gender declaration; a sentence from the AboutMe corpus containing one of three role-noun variants; and an instruction to revise. The argument runs on a logistic regression predicting whether the role noun is revised, with predictors coding the starting variant's gender, the preamble group, explicitness, and sentence-level stereotype scores computed as completion probabilities from one of the LLMs. A second component is the justification analysis, in which theme word sets (inclusive, modern, professional, standard, natural) built by embedding expansion are used to compare value language across conditions. These pieces jointly demonstrate neutralization and its contextual modulation.

What would settle it

One concrete check: collect human judgments of the perceived referent gender for each of the nine preambles, or probe the models' own representations of the referent, and test whether the prompts coded as neutral are actually treated as nonbinary; alternatively, counterbalancing the template wording within each coded category should leave the regression coefficients unchanged if the grouping captures the intended social meaning.

Watch

Extended reading notes

Core claim

The paper's central claim is that instruction-tuned LLMs exhibit neutralization: when revising text, they remove masculine and feminine role-noun variants in favor of gender-neutral forms. Logistic regressions over 14,229 prompt instances show significant positive effects for original_masc and original_fem, meaning gendered starting nouns are more likely to be revised, and the revisions are almost always gender-neutral. The remaining hypotheses are confirmed with qualifications: gendered variants are revised more for neutral prompts (nonbinary or they/them referents), more for incongruent gendered prompts, more for explicit pronoun or gender declarations, and feminine variants are revised more in masculine-stereotyped contexts while masculine variants are not revised more in feminine contexts. In justifications, removing gendered forms is justified by inclusive and modern themes, removing neutral forms by natural and standard themes, and inclusivity is emphasized for nonbinary people and explicit pronoun declarations. Together these results are claimed to show that LLMs encode sociolinguistically patterned values about gendered language reform rather than a single uniform rule.

Load-bearing premise

The argument assumes the nine prompt preambles were validly grouped into neutral, feminine, and masculine categories, such that a they/them pronoun user is effectively treated as nonbinary by the models; if that grouping is wrong, the finding that gendered variants are revised more for neutral prompts could be an artifact of the coding rather than evidence of value alignment.

Editorial extensions

If this is right

  • If the central claim holds, LLM revision tools will systematically shift user-facing text toward gender-neutral role nouns, making neutral vocabulary more common in everyday writing.
  • Because the effect is stronger for nonbinary referents and explicit pronoun or gender declarations, models treat neutral language as required for some people but optional for others, an uneven application of reform language.
  • The stereotype modulation means revisions can both challenge and reinforce gendered associations: feminine role nouns tend to be corrected in masculine-typed contexts, while masculine role nouns in feminine-typed contexts are left alone.
  • Justifications attach value themes to choices, with inclusive and modern themes used for removing gendered forms and natural and standard themes used for removing neutral forms, so the rationales themselves can spread reform-motivating or reform-resisting values.
  • Explicit prompts increase revisions to gendered variants as well as neutral ones, so making gender salient does not always lead to neutralization.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the result holds, an implication the paper leaves implicit is that routine LLM revision may avoid misgendering but can also produce degendering, erasing a referent's chosen gendered term even when that term is wanted.
  • One extension: test the stereotype asymmetry with matched contexts of equal masculine and feminine stereotype strength, since the current design may not have contained enough strongly feminine contexts to reveal a symmetric effect.
  • Another extension: the justification themes could serve as an interpretable audit signal in production revision systems, flagging when a model justifies removing neutral language on naturalness grounds.
  • Because the prompt grouping conflates gender identity with pronoun choice, a follow-up with orthogonal preambles (for example, a nonbinary person who uses she/her) would separate referent-gender effects from pronoun effects.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper investigates whether four instruction-tuned LLMs (GPT-4o, Llama-3.1-8B-Instruct, Gemma-2-9B-it, Mistral-Nemo) exhibit a neutralization strategy when revising gendered role nouns, and whether contextual factors (referent gender, explicitness of gender information, and stereotypical context) modulate this behavior. The authors construct 14,229 prompt instances from 527 AboutMe sentences, nine preambles, and 50 role-noun sets, and fit logistic regressions with random intercepts to predict whether a role noun is revised. They also analyze the justifications generated by the models, grouping value-related adjectives into themes and testing frequency differences with chi-square tests. They report broad support for overall neutralization, with significant effects for referent gender, explicitness, and context stereotypicality, and they discuss implications for value alignment in LLMs.

Significance. If the findings hold, this paper provides a large-scale, sociolinguistically grounded empirical study of value alignment in LLM text revision. The methodology is transparent: hypotheses are pre-specified, the prompt corpus is described in detail, mixed-effects logistic regression is used for the revision analysis, and the response-extraction heuristic is evaluated at 94% accuracy on a human-annotated sample. The paper also candidly discusses limitations and ethical considerations. The main caveats are the unvalidated grouping of preambles into three gender categories and the use of independence-assuming chi-square tests on clustered data; these affect the strength of the contextual-modulation claims, particularly H2a and the justification analyses.

major comments (3)
  1. [Section 4.2] The coding of prompt_neut is load-bearing for H2a and H2b but is not validated. The group includes the 'Pronoun Usage their' preamble, and singular 'their' in English is often a generic pronoun used when gender is unknown, not necessarily an indicator of a nonbinary referent. If the models treat this condition as 'gender unknown' rather than 'nonbinary', the significant original_gend:prompt_neut interaction in Table 4 may reflect a default-to-neutral strategy in the absence of gender information, which is a different claim from trans-inclusive alignment. The paper provides no manipulation check, human annotation, or probing evidence that the nine preambles are perceived in the intended three-way grouping. Please re-run the analysis with the 'their' preamble as a separate level (or excluding it) and report how the H2a and H2b conclusions change; alternatively, provide evidence that models distinguish 'their' from generic usage in these prompts.
  2. [Sections 5.2-5.3] The justification analyses use 2x2 chi-square tests on pooled responses without accounting for clustering. Each stimulus sentence contributes multiple responses (9 preambles x 3 role-noun variants x 4 models), so the observations are not independent. This likely inflates the reported significance levels. The revision analysis in Section 4 uses mixed-effects logistic regression with random intercepts; please apply a similar approach (e.g., generalized linear mixed models with random intercepts for sentence and model) to the theme-frequency analyses, or report intraclass correlations and verify that the substantive conclusions are robust to the non-independence.
  3. [Table 3 and Section 4.2] The variable original_gend is referenced in the regression formula but never defined. The paper should specify how it is coded (e.g., a dummy for either feminine or masculine starting variant) and how it relates to original_masc and original_fem. In addition, the model omits a main effect for prompt_neut; although prompt_neut may serve as the reference category for the prompt_masc/prompt_fem dummies, this should be stated explicitly, because the interpretation of the original_gend:prompt_neut interaction depends on the coding scheme.
minor comments (5)
  1. [Section 3.3] The phrase 'instruction-finedtuned' contains a typo; it should be 'instruction-fine-tuned'.
  2. [Table 3] The H2a row says 'original_mask:prompt_fem' but the intended predictor is 'original_masc:prompt_fem'.
  3. [Figure 2] The color coding is only partially explained in the text; please add a complete legend or explain the meaning of 'purple and green bars' and 'red and yellow bars' at their first mention.
  4. [Section 5.2] The description of '2x2 chi-square tests' is slightly inconsistent for H2b, where the comparison is between nonbinary and the combined woman/man condition; please clarify the exact contingency tables used for each test.
  5. [Section 8] The paper states 'Upon publication, we plan to release code and data' without a repository link; including an anonymized repository or a statement about data availability in the version of record would improve reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: reported effects are inferred from LLM behavior using independently computed predictors; self-citations are background, not load-bearing.

full rationale

The paper's derivation chain is not circular. The dependent variables are LLM revision choices and justification theme frequencies, measured from held-out prompt instances (14,229 per model) that are not used to fit any parameter later reported as a prediction. The regression in Table 3 is an inferential model of those choices, not a fit-then-predict loop. The context_fem/context_neut predictors are computed from a different, non-instruction-tuned model (llama-3.1-8B base) and are therefore not derived from the outcome models' outputs. Hypotheses H1a-H4a and H1b-H3b are stated before analysis and are tested against observed frequencies; no equation defines a target quantity in terms of itself. Self-citations (Watson et al. 2023a, 2025) supply stimulus sets and prior empirical expectations, but the current findings are independently measured and would stand or fall on the reported data. The unvalidated grouping of the nine preambles into prompt_neut/prompt_fem/prompt_masc (Section 4.2) is a measurement-validity concern, not a circularity: the coding is fixed ex ante from the preamble wording, and the outcome is the models' revision behavior, so the grouping does not construct the result by definition.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

The paper pulls role noun sets and language-ideology framing from Watson et al. (2025), and contextual hypotheses from sociolinguistic literature. The only notable free parameter is the BERT theme-expansion neighborhood size; no invented entities are introduced.

free parameters (1)
  • k (nearest neighbors for BERT theme expansion) = 10
    The theme word sets in the justifications analysis are expanded by taking the 10 nearest neighbors to seed embeddings (Appendix D). This hyperparameter is chosen by hand and affects which words are counted for each theme; if changed, the chi-square results could shift.
assumptions (4)
  • domain assumption LLM value alignment encodes contemporary feminist and trans-inclusive language reform norms (neutralization as a core strategy).
    The paper treats this as an expectation driving H1a (Section 4.1). If models do not encode these norms, the interpretation of neutralization as value alignment weakens.
  • domain assumption The AboutMe stimulus sentences are valid naturalistic contexts where the role noun refers to the friend/persona in the prompt preamble.
    Section 3.1 states stimulus sentences are sampled from AboutMe biographies and the preamble frames the friend as the author. If the role noun refers to another individual, the referent-gender manipulations (H2a) would not test what they claim.
  • domain assumption The heuristic METEOR-based algorithm correctly identifies the revised sentence and justification in LLM output.
    Appendix B reports 94% accuracy on a sampled evaluation. The dependent variable (whether a role noun was revised) and the justification theme counts depend on this segmentation; systematic errors could bias results.
  • standard math Logistic regression with random intercepts for sentence and role-noun set is an appropriate model for the repeated-measures data.
    Section 4.2 specifies the regression structure. The cluster-randomized design makes this a reasonable choice, though it assumes the random effects are normally distributed and the log-odds are linear in the predictors.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Analyzing values about gendered language reform in LLMs' revisions." pith.science (2026). https://pith.science/paper/QBL3KIK5

@misc{pith2026250521378,
  author       = {Pith},
  title        = {Pith review of: Analyzing values about gendered language reform in LLMs' revisions},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QBL3KIK5}},
  note         = {Machine review of arXiv:2505.21378}
}
read the original abstract

Within the common LLM use case of text revision, we study LLMs' revision of gendered role nouns (e.g., outdoorsperson/woman/man) and their justifications of such revisions. We evaluate their alignment with feminist and trans-inclusive language reforms for English. Drawing on insight from sociolinguistics, we further assess if LLMs are sensitive to the same contextual effects in the application of such reforms as people are, finding broad evidence of such effects. We discuss implications for value alignment.

Figures

Figures reproduced from arXiv: 2505.21378 by the authors.

Figure 1
Figure 1. Prompt setup and sample LLM responses. 2021). In particular, feminist and trans-inclusive language reforms encourage a strategy of neutral￾ization – using neutral terms instead of gendered ones – to include women and nonbinary people (Cameron, 2012; Zimman, 2017). Our overarching expectation is that, through their value alignment steps, LLMs will similarly follow this strategy. However, properties of the usage conte… view at source ↗
Figure 2
Figure 2. Revision patterns. For each of the three starting role noun variants, the bars show which variant or [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

50 extracted references · 41 canonical work pages

  1. [1]

    Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, and 1 others. 2023. Gpt-4 technical report. arXiv preprint arXiv:2303.08774

    arXiv 2023

  2. [2]

    Asif Agha. 2003. The social life of cultural value. Language & Communication, 23(3-4):231--273

    work page 2003

  3. [3]

    Y Gavriel Ansara and Peter Hegarty. 2014. Methodologies of misgendering: Recommendations for reducing cisgenderism in psychological research. Feminism & Psychology, 24(2):259--270

    work page 2014

  4. [4]

    Satanjeev Banerjee and Alon Lavie. 2005. METEOR : An automatic metric for MT evaluation with improved correlation with human judgments. In Proceedings of the ACL Workshop on Intrinsic and Extrinsic Evaluation Measures for Machine Translation and/or Summarization, pages 65--72

    2005

  5. [5]

    Marion Bartl and Susan Leavy. 2024. https://doi.org/10.18653/v1/2024.gebnlp-1.18 From showgirls' to performers': Fine-tuning with gender-inclusive language for bias reduction in LLM s . In Proceedings of the 5th Workshop on Gender Bias in Natural Language Processing (GeBNLP), pages 280--294, Bangkok, Thailand. Association for Computational Linguistics

    work page doi:10.18653/v1/2024.gebnlp-1.18 2024

  6. [6]

    Marion Bartl, Thomas Brendan Murphy, and Susan Leavy. 2025. Adapting psycholinguistic research for llms: Gender-inclusive language in a coreference context. arXiv preprint arXiv:2502.13120

    work page Pith review arXiv 2025

  7. [7]

    aid and abet

    Sandra L Bem and Daryl J Bem. 1973. Does sex-biased job advertising “aid and abet” sex discrimination? 1. Journal of Applied Social Psychology, 3(1):6--18

    work page 1973

  8. [8]

    Su Lin Blodgett, Solon Barocas, Hal Daum \'e III, and Hanna Wallach. 2020. Language (technology) is power: A critical survey of "bias" in NLP . Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics

    work page 2020

Show all 50 references
  1. [9]

    Stephanie Brandl, Ruixiang Cui, and Anders S gaard. 2022. How conservative are language models? adapting to the introduction of gender-neutral pronouns. Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Lan...

    2022

  2. [10]

    Deborah Cameron. 2012. Verbal hygiene. Routledge

    2012

  3. [11]

    Yang Trista Cao and Hal Daum \'e III. 2020. Toward gender-inclusive coreference resolution. Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics

    2020

  4. [12]

    Anne Curzan. 2014. Fixing English: Prescriptivism and language history. Cambridge University Press

    2014

  5. [13]

    Geraldine Damnati. 2024. From benchmark assessments to in-use evaluations: A n even wider gap to bridge at the era of generative AI . Mila Workshop: NLP in the era of generative AI, cognitive sciences, and societal transformation

    2024

  6. [14]

    Sunipa Dev, Masoud Monajatipoor, Anaelia Ovalle, Arjun Subramonian, Jeff M Phillips, and Kai-Wei Chang. 2021. Harms of gender exclusivity and challenges in non-binary representation in language technologies. In Proceedings of the 2021 Conference on Empirical Methods in Natural...

    2021

  7. [15]

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. Bert: Pre-training of deep bidirectional transformers for language understanding. Proceedings of the Conference of the North American Chapter of the Association for Computational Linguistics: Human Languag...

    2019

  8. [16]

    Susan Ehrlich and Ruth King. 1992. Gender-based language reform and the social construction of meaning. Discourse & Society, 3(2):151--166

    1992

  9. [17]

    Magdalena Formanowicz, Sylwia Bedynska, Aleksandra Cis ak, Friederike Braun, and Sabine Sczesny. 2013. Side effects of gender-fair language: How feminine job titles influence the evaluation of female applicants. European Journal of Social Psychology, 43(1):62--71

    2013

  10. [18]

    Gemma Team , Morgane Riviere, Shreya Pathak, Pier Giuseppe Sessa, Cassidy Hardin, Surya Bhupatiraju, L \'e onard Hussenot, Thomas Mesnard, Bobak Shahriari, Alexandre Ram \'e , and 1 others. 2024. Gemma 2: Improving open language models at a practical size. arXiv preprint arXiv...

    2024 arXiv

  11. [19]

    Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Alex Vaughan, and 1 others. 2024. The llama 3 herd of models. arXiv preprint arXiv:2407.21783

    2024 arXiv

  12. [20]

    Valentin Hofmann, Pratyusha Ria Kalluri, Dan Jurafsky, and Sharese King. 2024. AI generates covertly racist decisions about people based on their dialect. Nature, 633(8028):147--154

    2024

  13. [21]

    Tamanna Hossain, Sunipa Dev, and Sameer Singh. 2023. https://doi.org/10.18653/v1/2023.acl-long.293 MISGENDERED : Limits of large language models in understanding pronouns . In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Lo...

    2023 doi

  14. [22]

    Aaron Hurst, Adam Lerer, Adam P Goucher, Adam Perelman, Aditya Ramesh, Aidan Clark, AJ Ostrow, Akila Welihinda, Alan Hayes, Alec Radford, and 1 others. 2024. Gpt-4o system card. arXiv preprint arXiv:2410.21276

    2024 arXiv

  15. [23]

    Judith T Irvine. 1989. When talk isn't cheap: Language and political economy. American Ethnologist, 16(2):248--267

    1989

  16. [24]

    Samantha Jackson, Barend Beekhuizen, Zhao Zhao, and Rhonda McEwen. 2024. GPT -4- T rinis: Assessing GPT -4’s communicative competence in the E nglish-speaking majority world. AI & Society, pages 1--17

    2024

  17. [25]

    Kai Jacobsen, Charlie E Davis, Drew Burchell, Leo Rutherford, Nathan Lachowsky, Greta Bauer, and Ayden Scheim. 2024. Misgendering and the health and wellbeing of nonbinary people in C anada. International Journal of Transgender Health, 25(4):816--830

    2024

  18. [26]

    Lee Jiang. 2023. Resistance to singular “they” in R eddit communities. Master's thesis, University of Toronto

    2023

  19. [27]

    Rishi Kant. 2025. https://www.reuters.com/technology/artificial-intelligence/openais-weekly-active-users-surpass-400-million-2025-02-20/ Open AI 's weekly active users surpass 400 million . Reuters

    2025

  20. [28]

    Hadas Kotek, Rikker Dockum, and David Sun. 2023. Gender bias and stereotypes in large language models. In Proceedings of the ACM Collective Intelligence Conference, pages 12--24

    2023

  21. [29]

    Paul V Kroskrity. 2004. Language ideologies. A Companion to Linguistic Anthropology, 496:517

    2004

  22. [30]

    Anne Lauscher, Archie Crowley, and Dirk Hovy. 2022. Welcome to the modern world of pronouns: Identity-inclusive natural language processing beyond gender. Proceedings of the 29th International Conference on Computational Linguistics

    2022

  23. [31]

    Li Lucy, Suchin Gururangan, Luca Soldaini, Emma Strubell, David Bamman, Lauren Klein, and Jesse Dodge. 2024. https://aclanthology.org/2024.acl-long.400 A bout M e: Using self-descriptions in webpages to document the effects of E nglish pretraining data filters . In Proceedings...

    2024

  24. [32]

    Gunnar Lund, Kostiantyn Omelianchuk, and Igor Samokhin. 2023. https://doi.org/10.18653/v1/2023.bea-1.13 Gender-inclusive grammatical error correction through augmentation . In Proceedings of the 18th Workshop on Innovative Use of NLP for Building Educational Applications (BEA ...

    2023 doi

  25. [33]

    Alonzo Martinez. 2023. https://www.forbes.com/sites/alonzomartinez/2023/06/22/an-employers-guide-to-inclusive-language/ An employer’s guide to inclusive language . Forbes Magazine

    2023

  26. [34]

    Mistral AI Team . 2024. https://mistral.ai/news/mistral-nemo Mistral nemo

    2024

  27. [35]

    Annabelle Mooney and Betsy Evans. 2015. Language, society and power: An introduction. Routledge

    2015

  28. [36]

    Brittney O'Neill. 2021. He, (s)he/she, and they: Language ideologies and ideological conflict in gendered language reform. Working papers in Applied Linguistics and Linguistics at York, 1:16--28

    2021

  29. [37]

    I ’m fully who I am

    Anaelia Ovalle, Palash Goyal, Jwala Dhamala, Zachary Jaggers, Kai-Wei Chang, Aram Galstyan, Richard Zemel, and Rahul Gupta. 2023. “ I ’m fully who I am”: Towards centering transgender and non-binary voices to measure biases in open language generation. In Proceedings of the 20...

    2023

  30. [38]

    Brandon Papineau, Rob Podesva, and Judith Degen. 2022. ‘ S ally the congressperson’: The role of individual ideology on the processing and production of english gender-neutral role nouns. In Proceedings of the Annual Meeting of the Cognitive Science Society, volume 44

    2022

  31. [39]

    Sabine Sczesny, Magda Formanowicz, and Franziska Moser. 2016. Can gender-fair language reduce gender stereotyping and discrimination? Frontiers in psychology, page 25

    2016

  32. [40]

    Michael Silverstein. 1985. Language and the culture of gender: At the intersection of structure, usage, and ideology. In Semiotic mediation, pages 219--259. Elsevier

    1985

  33. [41]

    Elizabeth Stokoe and Frederick Attenborough. 2014. Gender and categorial systematics. Handbook of language, gender and sexuality, pages 161--179

    2014

  34. [42]

    Yolande Strengers, Lizhen Qu, Qiongkai Xu, and Jarrod Knibbe. 2020. Adhering, steering, and queering: Treatment of gender in natural language generation. In Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems, pages 1--14

    2020

  35. [43]

    Eva Vanmassenhove, Chris Emmery, and Dimitar Shterionov. 2021. Neu T ral rewriter: A rule-based and neural approach to automatic rewriting into gender-neutral alternatives. Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing

    2021

  36. [44]

    Julia Watson, Barend Beekhuizen, and Suzanne Stevenson. 2023 a . https://doi.org/10.18653/v1/2023.acl-long.375 What social attitudes about gender does BERT encode? leveraging insights from psycholinguistics . In Proceedings of the 61st Annual Meeting of the Association for Com...

    2023 doi

  37. [45]

    Lee, Barend Beekhuizen, and Suzanne Stevenson

    Julia Watson, Sophia S. Lee, Barend Beekhuizen, and Suzanne Stevenson. 2025. https://aclanthology.org/2025.coling-main.80/ Do language models practice what they preach? examining language ideologies about gendered language reform encoded in LLM s . In Proceedings of the 31st I...

    2025

  38. [46]

    Julia Watson, Sarah Walker, Suzanne Stevenson, and Barend Beekhuizen. 2023 b . Communicative need shapes choices to use gendered vs. gender-neutral kinship terms across online communities. In Proceedings of the Annual Meeting of the Cognitive Science Society, volume 45

    2023

  39. [47]

    Langdon Winner. 1980. https://www.osti.gov/biblio/5525771 Do artifacts have politics? Daedalus, 109(1):121--136

    1980

  40. [48]

    Lal Zimman. 2017. Transgender language reform: Some challenges and strategies for promoting trans-affirming, gender-inclusive language. Journal of Language and Discrimination, 1(1):84--105

    2017

  41. [49]

    online" 'onlinestring :=

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint eprinttype howpublished institution journal key month note number organization pages publisher school series title type volume year doi pubmed url lastchecked label extra.label sort.label short.list...

  42. [50]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.