Estimating Grammatical Gender Directions in Contextual Embeddings under Controlled and Natural Contexts

Huanping Xiao; Yingji Li

arxiv: 2606.30152 · v1 · pith:JRGKFDN7new · submitted 2026-06-29 · 💻 cs.CL · cs.AI

Estimating Grammatical Gender Directions in Contextual Embeddings under Controlled and Natural Contexts

Huanping Xiao , Yingji Li This is my paper

Pith reviewed 2026-06-30 06:05 UTC · model grok-4.3

classification 💻 cs.CL cs.AI

keywords contextual embeddingsgrammatical gendergender debiasingSpanish language modelsdirection estimationinanimate nounscontrolled contexts

0 comments

The pith

Unweighted controlled contexts isolate the purest grammatical gender direction in contextual embeddings.

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

The paper seeks to separate grammatical gender signals from semantic bias in contextual language models for languages like Spanish, where the two are conflated. It builds balanced datasets of inanimate nouns using controlled sentence templates and natural Wikipedia sentences, then applies a framework of centroid, SVM, and LDA estimators along with weighting options. Dual metrics assess how well each method suppresses grammatical gender leakage on inanimate nouns while keeping semantic gender distinctions intact for occupation words. Experiments show that unweighted controlled contexts produce the cleanest direction estimates and that the simple centroid method beats the two discriminative baselines.

Core claim

By constructing balanced datasets of inanimate nouns in both controlled templates and natural Wikipedia contexts, the authors estimate grammatical gender directions in contextual embeddings. They compare centroid, SVM, and LDA estimators under different weighting schemes and introduce dual-objective metrics that trade off leakage suppression against semantic preservation. The results establish that unweighted controlled contexts yield the purest grammatical gender direction and that the centroid estimator outperforms the discriminative baselines.

What carries the argument

The dual-objective evaluation framework that applies centroid, SVM, and LDA estimators to balanced datasets of inanimate nouns drawn from controlled templates and Wikipedia contexts.

If this is right

Debiasing techniques can now target contextual representations instead of being limited to static word embeddings.
Grammatical gender signals can be estimated independently of semantic gender distinctions.
Controlled sentence templates supply cleaner signals for direction estimation than natural text does.
A simple centroid calculation is sufficient and preferable to SVM or LDA for this estimation task.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same construction of controlled contexts could be applied to other gendered languages to test whether the purity advantage holds.
Directions obtained this way could be inserted into existing debiasing pipelines to measure downstream gains on bias reduction tasks.
The approach might extend to disentangling other grammatical features such as number or animacy from semantic content.

Load-bearing premise

The balanced datasets of inanimate nouns successfully isolate grammatical gender without residual semantic contamination that would distort the direction estimates.

What would settle it

Finding that the gender direction estimated from controlled contexts still correlates strongly with semantic features of occupation terms would show that isolation failed.

Figures

Figures reproduced from arXiv: 2606.30152 by Huanping Xiao, Yingji Li.

read the original abstract

Contextual language models conflate grammatical gender and social semantic bias in gendered languages such as Spanish. Existing gender debiasing approaches only operate on static word embeddings leaving contextual representations unexplored for this two dimensional gender disentanglement. To address the this issue, we make the first attempt to disentangle grammatical gender from semantic contamination for contextual embeddings. We construct both controlled templates and natural Wikipedia contexts to build balanced datasets of inanimate nouns, and design a framework equipped with centroid, Support Vector Machine (SVM) and Linear Discriminant Analysis (LDA) gender direction estimators as well as contamination-aware weighting strategies. A set of dual-objective evaluation metrics is proposed to balance the suppression of grammatical gender leakage on inanimate nouns and the preservation of semantic gender distinctions for occupation terms. The results reveal that unweighted controlled contexts yield the purest grammatical gender direction, and the centroid estimator achieves better performance than discriminative baselines.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper makes an initial pass at isolating grammatical gender directions in contextual embeddings via inanimate nouns in controlled templates and Wikipedia, but the isolation claim lacks the diagnostics needed to make the estimator comparisons reliable.

read the letter

The core contribution is extending static-embedding gender direction work to contextual models by building balanced inanimate-noun sets in both artificial templates and natural contexts, then comparing centroid, SVM, and LDA estimators plus some weighting. They also introduce dual metrics that penalize leakage on inanimates while trying to keep semantic distinctions on occupations. That setup is new for contextual embeddings and the controlled-vs-natural comparison is a reasonable way to test context effects.

The results favor unweighted controlled contexts and the centroid estimator, which is plausible on its face. The paper does a service by naming the two-dimensional problem explicitly and by moving beyond static embeddings.

The soft spot is the untested assumption that the inanimate-noun datasets actually isolate grammatical gender. Even inanimate nouns in Spanish can carry collocational or cultural associations that align with gender in embedding space, and the abstract gives no correlation check or ablation that would show the direction is free of semantic residue. Without that, the claim that controlled contexts are purest and that centroid beats the others rests on an unverified premise. Dataset sizes, exact metrics, and error bars are also missing from the summary, which makes it hard to judge how much the numbers move.

This is for people already working on debiasing in gendered languages who want a starting point for contextual cases. It is not yet strong enough for a serious referee without the missing validation steps, but the gap it targets is real enough that a revised version with diagnostics could be worth sending out.

Referee Report

2 major / 2 minor

Summary. The paper claims to make the first attempt at disentangling grammatical gender from semantic contamination in contextual embeddings for Spanish. It constructs balanced datasets of inanimate nouns via controlled templates and natural Wikipedia contexts, applies centroid/SVM/LDA gender direction estimators together with contamination-aware weighting, and proposes dual-objective metrics that trade off suppression of grammatical gender leakage on inanimates against preservation of semantic gender distinctions on occupation terms. The reported results are that unweighted controlled contexts produce the purest grammatical gender direction and that the centroid estimator outperforms the discriminative baselines.

Significance. If the isolation of grammatical gender holds, the work supplies a concrete framework and dual-objective evaluation protocol for contextual debiasing in gendered languages, extending prior static-embedding methods. The explicit comparison of controlled versus natural contexts and the inclusion of both centroid and discriminative estimators are constructive contributions that could be reused.

major comments (2)

[§3 / Abstract] §3 (Dataset Construction) and Abstract: The central claim that unweighted controlled contexts yield the purest direction and that centroid outperforms SVM/LDA rests on the assumption that the balanced inanimate-noun datasets isolate grammatical gender without residual semantic contamination. No diagnostic (e.g., pre/post-weighting correlation between noun semantics and estimated directions, or an independent semantic probe) is described that would falsify this isolation; if the assumption fails, both the purity ranking and the estimator comparison become unreliable.
[Evaluation / Abstract] Evaluation section: The dual-objective metrics are introduced to balance suppression on inanimates versus preservation on occupations, yet the abstract (and by extension the reported results) supplies no quantitative values, dataset sizes, error bars, or statistical tests. Without these numbers it is impossible to verify the claimed superiority of the centroid estimator or the ranking of context types.

minor comments (2)

[Abstract] Abstract: 'To address the this issue' contains a typographical error.
[Abstract / Introduction] Abstract: The claim of being the 'first attempt' should be supported by a brief literature comparison in the introduction rather than asserted only in the abstract.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment below, proposing revisions where the manuscript can be strengthened without misrepresenting our contributions.

read point-by-point responses

Referee: [§3 / Abstract] §3 (Dataset Construction) and Abstract: The central claim that unweighted controlled contexts yield the purest direction and that centroid outperforms SVM/LDA rests on the assumption that the balanced inanimate-noun datasets isolate grammatical gender without residual semantic contamination. No diagnostic (e.g., pre/post-weighting correlation between noun semantics and estimated directions, or an independent semantic probe) is described that would falsify this isolation; if the assumption fails, both the purity ranking and the estimator comparison become unreliable.

Authors: We agree that an explicit falsification diagnostic would strengthen the isolation claim. The framework relies on the established practice of using inanimate nouns (which carry no semantic gender) to isolate grammatical gender, combined with contamination-aware weighting to mitigate residual effects. We will add a pre/post-weighting correlation analysis between noun semantics and the estimated directions as a diagnostic in the revised §3. revision: yes
Referee: [Evaluation / Abstract] Evaluation section: The dual-objective metrics are introduced to balance suppression on inanimates versus preservation on occupations, yet the abstract (and by extension the reported results) supplies no quantitative values, dataset sizes, error bars, or statistical tests. Without these numbers it is impossible to verify the claimed superiority of the centroid estimator or the ranking of context types.

Authors: The Evaluation section reports the dual-objective metric values, dataset sizes, and comparisons, but the abstract summarizes results qualitatively. We will revise the abstract to include key quantitative values and dataset sizes. Error bars and statistical tests were not computed in the original experiments; we can add basic significance notes if feasible but cannot retroactively introduce them without new runs. revision: partial

Circularity Check

0 steps flagged

No circularity detected in derivation chain

full rationale

The paper constructs balanced datasets of inanimate nouns from controlled templates and Wikipedia contexts, then applies standard centroid, SVM, and LDA estimators along with contamination-aware weighting and dual-objective metrics. The headline results (unweighted controlled contexts being purest; centroid outperforming baselines) are presented as empirical comparisons on these constructed data. No equations, self-definitional reductions, fitted inputs renamed as predictions, or load-bearing self-citations appear in the abstract or described methods. The derivation chain relies on external ML estimators and data construction rather than reducing outputs to inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review; no explicit free parameters, axioms, or invented entities stated. The framework implicitly assumes that grammatical gender can be linearly separated in embedding space and that the chosen estimators capture it without semantic leakage.

pith-pipeline@v0.9.1-grok · 5673 in / 1072 out tokens · 22366 ms · 2026-06-30T06:05:09.126380+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

50 extracted references · 44 canonical work pages · 3 internal anchors

[1]

Social Bias in Multilingual Language Models: A Survey

Gamboa LCL, Feng Y, Lee MG. Social Bias in Multilingual Language Models: A Survey. In: Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. Suzhou, June 30, 2026 13/18 China: Association for Computational Linguistics; 2025. p. 27857-80. Available from: https://aclanthology.org/2025.emnlp-main.1416/. doi:10.18653/v1/2025....

work page doi:10.18653/v1/2025.emnlp-main.1416 2025
[2]

Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings

Bolukbasi T, Chang KW, Zou J, Saligrama V, Kalai A. Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings. arXiv preprint arXiv:160706520. 2016. Available from:https://arxiv.org/abs/1607.06520. doi:https://doi.org/10.48550/arXiv.1607.06520

work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1607.06520 2016
[3]

Measuring Bias in Contextualized Word Representations

Kurita K, Vyas N, Pareek A, Black AW, Tsvetkov Y. Measuring Bias in Contextualized Word Representations. In: Proceedings of the First Workshop on Gender Bias in Natural Language Processing; 2019. p. 166-72. Available from:https://aclanthology.org/W19-3823/. doi:https://doi.org/10.18653/v1/W19-3823

work page doi:10.18653/v1/w19-3823 2019
[4]

Semantics derived automatically from language corpora contain human-like biases

Caliskan A, Bryson JJ, Narayanan A. Semantics derived automatically from language corpora contain human-like biases. Science. 2017;356(6334):183-6. Available from: https://arxiv.org/abs/1608.07187. doi:https://doi.org/10.1126/science.aal4230

work page internal anchor Pith review Pith/arXiv arXiv doi:10.1126/science.aal4230 2017
[5]

Examining Gender Bias in Languages with Grammatical Gender

Zhou P, Shi W, Zhao J, Huang KH, Chen M, Cotterell R, et al. Examining Gender Bias in Languages with Grammatical Gender. In: Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP); 2019. p. 5276-84. Available from:https://aclanthology.o...

work page doi:10.18653/v1/d19-1531 2019
[6]

OSCaR: Orthogonal Subspace Correction and Rectification of Biases in Word Embeddings

Dev S, Li T, Phillips JM, Srikumar V. OSCaR: Orthogonal Subspace Correction and Rectification of Biases in Word Embeddings. In: Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing. Online and Punta Cana, Dominican Republic: Association for Computational Linguistics; 2021. p. 5034-50. Available from:https://aclanthology.o...

work page doi:10.18653/v1/2021.emnlp-main.411 2021
[7]

Bias and Fairness in Large Language Models: A Survey

Gallegos IO, Rossi RA, Barrow J, Tanjim MM, Kim S, Dernoncourt F, et al. Bias and Fairness in Large Language Models: A Survey. Computational Linguistics. 2024;50(3):649-704. Available from: https://aclanthology.org/2024.cl-3.8/. doi:10.1162/coli_a_00498

work page doi:10.1162/coli_a_00498 2024
[8]

A Survey on Fairness in Large Language Models; 2023

Li Y, Du M, Song R, Wang X, Wang Y. A Survey on Fairness in Large Language Models; 2023. Available from:https://arxiv.org/abs/2308.10149. arXiv:2308.10149. doi:10.48550/arXiv.2308.10149

work page doi:10.48550/arxiv.2308.10149 2023
[9]

Multilingual Denoising Pre-training for Neural Machine Translation

Liu Y, Gu J, Goyal N, Li X, Edunov S, Ghazvininejad M, et al. Multilingual Denoising Pre-training for Neural Machine Translation. Transactions of the Association for Computational Linguistics. 2020;8:726-42. Available from:https://aclanthology.org/2020.tacl-1.47/. doi:https://doi.org/10.1162/tacl_a_00343

work page doi:10.1162/tacl_a_00343 2020
[10]

MarIA and BETO are sexist: evaluating gender bias in large language models for Spanish

Garrido-Muñoz I, Martínez-Santiago F, Montejo-Ráez A. MarIA and BETO are sexist: evaluating gender bias in large language models for Spanish. Language Resources and Evaluation. 2023;58(4):1387-417. Available from: https://link.springer.com/article/10.1007/s10579-023-09670-3. doi:https://doi.org/10.1007/s10579-023-09670-3

work page doi:10.1007/s10579-023-09670-3 2023
[11]

Corbett GG. Gender. Cambridge Textbooks in Linguistics. Cambridge University Press; 1991

1991
[12]

Identifying and Reducing Gender Bias in Word-Level Language Models

Bordia S, Bowman SR. Identifying and Reducing Gender Bias in Word-Level Language Models. In: Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Student Research Workshop; 2019. p. 7-15. Available from: https://aclanthology.org/N19-3002/. doi:https://doi.org/10.18653/v1/N19-3002

work page doi:10.18653/v1/n19-3002 2019
[13]

Dresses and ties: the effect of grammatical gender and stereotypical semantic bias in three Spanish-speaking communities

Zunino GM, Aguilar M, Stetie NA, Martínez Rebolledo C, Hinojosa JA. Dresses and ties: the effect of grammatical gender and stereotypical semantic bias in three Spanish-speaking communities. Language and Cognition. 2025;17:e35. Available from: June 30, 2026 14/18 https://www.cambridge.org/core/journals/language-and-cognition/article/dresses-and -ties-the-e...

work page doi:10.1017/langcog.2025.4 2025
[14]

Sex, Syntax, and Semantics

Boroditsky L, Schmidt LA, Phillips W. Sex, Syntax, and Semantics. In: Language in Mind: Advances in the Study of Language and Thought; 2003. p. 61-79. Available from:https://direct.mit.edu/b ooks/edited-volume/1917/chapter-abstract/52672/Sex-Syntax-and-Semantics. doi:https://doi.org/10.7551/mitpress/4117.003.0010

work page doi:10.7551/mitpress/4117.003.0010 2003
[15]

What an Elegant Bridge: Multilingual LLMs are Biased Similarly in Different Languages

Mihaylov V, Shtedritski A. What an Elegant Bridge: Multilingual LLMs are Biased Similarly in Different Languages. In: Proceedings of the Fourth Workshop on Multilingual Representation Learning (MRL 2024); 2024. p. 22-9. Available from:https://aclanthology.org/2024.mrl-1.2/. doi:https://doi.org/10.18653/v1/2024.mrl-1.2

work page doi:10.18653/v1/2024.mrl-1.2 2024
[16]

The Causal Influence of Grammatical Gender on Distributional Semantics

Stańczak K, Du K, Williams A, Augenstein I, Cottrell R. The Causal Influence of Grammatical Gender on Distributional Semantics. Transactions of the Association for Computational Linguistics. 2024;12:1672-85. Available from:https://aclanthology.org/2024.tacl-1.90/. doi:https://doi.org/10.1162/tacl_a_00723

work page doi:10.1162/tacl_a_00723 2024
[17]

Investigating grammatical abstraction in language models using few-shot learning of novel noun gender

Sukumaran P, Houghton C, Kazanina N. Investigating grammatical abstraction in language models using few-shot learning of novel noun gender. In: Findings of the Association for Computational Linguistics: EACL 2024. St. Julian’s, Malta: Association for Computational Linguistics; 2024. p. 747-65. Available from:https://aclanthology.org/2024.findings-eacl.50/...

work page doi:10.18653/v1/2024.findings-eacl.50 2024
[18]

Beyond Content: How Grammatical Gender Shapes Visual Representation in Text-to-Image Models

Saeed M, Raza S, Vayani A, Abdul-Mageed M, Emami A, Shehata S. Beyond Content: How Grammatical Gender Shapes Visual Representation in Text-to-Image Models. Findings of the Association for Computational Linguistics: EMNLP 2025. 2025. Available from: https://aclanthology.org/2025.findings-emnlp.1343/. doi:10.18653/v1/2025.findings-emnlp.1343

work page doi:10.18653/v1/2025.findings-emnlp.1343 2025
[19]

Measuring Gender Bias in Word Embeddings of Gendered Languages Requires Disentangling Grammatical Gender Signals

Sabbaghi SO, Caliskan A. Measuring Gender Bias in Word Embeddings of Gendered Languages Requires Disentangling Grammatical Gender Signals. In: Proceedings of the 2022 AAAI/ACM Conference on AI, Ethics, and Society; 2022. p. 518-27. Available from: https://dl.acm.org/doi/10.1145/3514094.3534176. doi:https://doi.org/10.1145/3514094.3534176

work page doi:10.1145/3514094.3534176 2022
[21]

Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings

Bolukbasi T, Chang KW, Zou JY, Saligrama V, Kalai AT. Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings. In: Advances in Neural Information Processing Systems 29; 2016. p. 4349-57. Available from:https://proceedings.neurips.cc/paper _files/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html

2016
[23]

On Measuring Social Biases in Sentence Encoders

May C, Wang A, Bordia S, Bowman SR, Rudinger R. On Measuring Social Biases in Sentence Encoders. In: Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers)

2019
[24]

p. 622-8. Available from:https://aclanthology.org/N19-1063/. doi:https://doi.org/10.18653/v1/N19-1063. June 30, 2026 15/18

work page doi:10.18653/v1/n19-1063 2026
[25]

Lipstick on a Pig: Debiasing Methods Cover up Systematic Gender Biases in Word Embeddings But do not Remove Them

Gonen H, Goldberg Y. Lipstick on a Pig: Debiasing Methods Cover up Systematic Gender Biases in Word Embeddings But do not Remove Them. In: Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers); 2019. p. 609-14. Available from: https:/...

work page doi:10.18653/v1/n19-1061 2019
[26]

Bias in Bios: A Case Study of Semantic Representation Bias in a High-Stakes Setting

De-Arteaga M, Romanov A, Wallach H, Chayes J, Borgs C, Chouldechova A, et al. Bias in Bios: A Case Study of Semantic Representation Bias in a High-Stakes Setting. arXiv preprint arXiv:190109451
[27]

Bias in Bios: A Case Study of Semantic Representation Bias in a High-Stakes Setting

Available from:https://arxiv.org/abs/1901.09451. doi:https://doi.org/10.48550/arXiv.1901.09451

work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1901.09451 1901
[28]

Mitigating Gender Bias in Contextual Word Embeddings

Yarrabelly N, Damodaran V, Su FG. Mitigating Gender Bias in Contextual Word Embeddings. arXiv preprint arXiv:241112074. 2024. Available from:https://arxiv.org/abs/2411.12074. doi:https://doi.org/10.48550/arXiv.2411.12074

work page doi:10.48550/arxiv.2411.12074 2024
[29]

Assessing Social and Intersectional Biases in Contextualized Word Representations

Tan YC, Celis LE. Assessing Social and Intersectional Biases in Contextualized Word Representations. arXiv preprint arXiv:191101485. 2019. Available from:https://arxiv.org/abs/1911.01485. doi:10.48550/arXiv.1911.01485

work page doi:10.48550/arxiv.1911.01485 2019
[30]

Stereotype and Skew: Quantifying Gender Bias in Pre-trained and Fine-tuned Language Models

de Vassimon Manela D, Errington D, Fisher T, van Breugel B, Minervini P. Stereotype and Skew: Quantifying Gender Bias in Pre-trained and Fine-tuned Language Models. In: Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume. Online: Association for Computational Linguistics; 2021. p. 2232-4...

work page doi:10.18653/v1/2021.eacl-main.190 2021
[31]

How Does Grammatical Gender Affect Noun Representations in Gender-Marking Languages? In: Proceedings of the 23rd Conference on Computational Natural Language Learning (CoNLL); 2019

Gonen H, Kementchedjhieva Y, Goldberg Y. How Does Grammatical Gender Affect Noun Representations in Gender-Marking Languages? In: Proceedings of the 23rd Conference on Computational Natural Language Learning (CoNLL); 2019. p. 463-71. Available from: https://aclanthology.org/K19-1043/. doi:https://doi.org/10.18653/v1/K19-1043

work page doi:10.18653/v1/k19-1043 2019
[32]

Gendered Grammar or Ingrained Bias? Exploring Gender Bias in Icelandic Language Models

Friðriksdóttir SR, Einarsson H. Gendered Grammar or Ingrained Bias? Exploring Gender Bias in Icelandic Language Models. In: Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024). Torino, Italia: ELRA and ICCL; 2024. p. 7596-610. Available from: https://aclanthology.org/202...

2024
[33]

Does grammatical gender affect object concepts? Registered replication of Phillips and Boroditsky (2003)

Elpers N, Jensen G, Holmes KJ. Does grammatical gender affect object concepts? Registered replication of Phillips and Boroditsky (2003). Journal of Memory and Language. 2022;127:104357. Available from: https://www.sciencedirect.com/science/article/abs/pii/S0749596X22000444. doi:https://doi.org/10.1016/j.jml.2022.104357

work page doi:10.1016/j.jml.2022.104357 2003
[34]

Leveraging Large Language Models to Measure Gender Representation Bias in Gendered Language Corpora

Derner E, de la Fuente SS, Gutiérrez Y, Moreda P, Oliver N. Leveraging Large Language Models to Measure Gender Representation Bias in Gendered Language Corpora. In: Proceedings of the 6th Workshop on Gender Bias in Natural Language Processing (GeBNLP); 2025. p. 468-83. Available from: https://aclanthology.org/2025.gebnlp-1.39/. doi:https://doi.org/10.1865...

work page doi:10.18653/v1/2025.gebnlp-1.39 2025
[35]

EuroGEST: Investigating gender stereotypes in multilingual language models

Rowe J, Klimaszewski M, Guillou L, Vallor S, Birch A. EuroGEST: Investigating gender stereotypes in multilingual language models. In: Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing; 2025. p. 32074-96. Available from: https://aclanthology.org/2025.emnlp-main.1632/. doi:https://doi.org/10.18653/v1/2025.emnlp-main.1632

work page doi:10.18653/v1/2025.emnlp-main.1632 2025
[36]

Spanish Pre-trained BERT Model and Evaluation Data

Cañete J, Chaperon G, Fuentes R, Ho JH, Kang H, Pérez J. Spanish Pre-trained BERT Model and Evaluation Data. arXiv preprint arXiv:230802976. 2023. Available from: https://arxiv.org/abs/2308.02976. doi:https://doi.org/10.48550/arXiv.2308.02976. June 30, 2026 16/18

work page doi:10.48550/arxiv.2308.02976 2023
[37]

Conneau, K

Conneau A, Khandelwal K, Goyal N, Chaudhary V, Wenzek G, Guzmán F, et al. Unsupervised Cross-lingual Representation Learning at Scale. In: Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics; 2020. p. 8440-51. Available from: https://aclanthology.org/2020.acl-main.747/. doi:https://doi.org/10.18653/v1/2020.acl-main.747

work page doi:10.18653/v1/2020.acl-main.747 2020
[38]

CamemBERT: a Tasty French Language Model

Martin L, Muller B, Ortiz Suárez PJ, Dupont Y, Romary L, de la Clergerie É, et al. CamemBERT: a Tasty French Language Model. In: Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics; 2020. p. 7203-19. Available from: https://aclanthology.org/2020.acl-main.645/. doi:https://doi.org/10.18653/v1/2020.acl-main.645

work page doi:10.18653/v1/2020.acl-main.645 2020
[39]

Word embeddings are biased: But whose bias are they reflecting? AI & Society

Petreski D, Hashim IC. Word embeddings are biased: But whose bias are they reflecting? AI & Society. 2023;38:975-82. Available from:https://www.researchgate.net/publication/342995921 _Social_biases_in_word_embeddings_and_their_relation_to_human_cognition. doi:https://doi.org/10.1007/s00146-022-01443-w

work page doi:10.1007/s00146-022-01443-w 2023
[40]

Detecting gender bias in Arabic text through word embeddings

Mourad A, Abu Salem FK, Elbassuoni S. Detecting gender bias in Arabic text through word embeddings. PLOS ONE. 2025. Available from: https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0319301. doi:https://doi.org/10.1371/journal.pone.0319301

work page doi:10.1371/journal.pone.0319301 2025
[41]

Job Unfair: An Investigation of Gender and Occupational Bias in Free-Form Text Completions by LLMs

Casula C, Vecellio Salto S, Leonardelli E, Tonelli S. Job Unfair: An Investigation of Gender and Occupational Bias in Free-Form Text Completions by LLMs. In: Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. Suzhou, China: Association for Computational Linguistics; 2025. p. 22759-77. Available from: https://aclantholo...

work page doi:10.18653/v1/2025.emnlp-main.1159 2025
[42]

Bias Is a Subspace, Not a Coordinate: A Geometric Rethinking of Post-hoc Debiasing in Vision-Language Models

Zhao D, Li W, Shen Z, Qiu Y, Xu B, Chen H, et al. Bias Is a Subspace, Not a Coordinate: A Geometric Rethinking of Post-hoc Debiasing in Vision-Language Models. arXiv preprint arXiv:251118123. 2025. Available from:https://arxiv.org/abs/2511.18123. arXiv:2511.18123. doi:10.48550/arXiv.2511.18123

work page doi:10.48550/arxiv.2511.18123 2025
[43]

Grammatical Gender Associations Outweigh Topical Gender Bias in Cross-linguistic Word Embeddings

McCurdy K, Serbetci O. Grammatical Gender Associations Outweigh Topical Gender Bias in Cross-linguistic Word Embeddings. arXiv preprint arXiv:200508864. 2020. Available from: https://arxiv.org/abs/2005.08864. doi:https://doi.org/10.48550/arXiv.2005.08864

work page doi:10.48550/arxiv.2005.08864 2020
[44]

Discovering Low-rank Subspaces for Language-agnostic Multilingual Representations

Xie Z, Zhao H, Yu T, Li S. Discovering Low-rank Subspaces for Language-agnostic Multilingual Representations. In: Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing. Abu Dhabi, United Arab Emirates: Association for Computational Linguistics; 2022. p. 5617-33. Available from:https://aclanthology.org/2022.emnlp-main.379/....

work page doi:10.18653/v1/2022.emnlp-main.379 2022
[45]

Bias Is a Subspace, Not a Coordinate: A Geometric Rethinking of Post-hoc Debiasing in Vision-Language Models

Gupta S, Balia R, Angioni D, Brau F, Pintor M, Demontis A, et al. Bias Is a Subspace, Not a Coordinate: A Geometric Rethinking of Post-hoc Debiasing in Vision-Language Models. arXiv preprint arXiv:251118123. 2025. Available from:https://arxiv.org/abs/2511.18123. arXiv:2511.18123

work page arXiv 2025
[46]

Bias in language models: A survey

de Vassimon Manela D, Errington D, Fisher T, van Breugel B, Minervini P. Bias in language models: A survey. Cognitive Systems Research. 2021;70:100554. Available from: https://www.sciencedirect.com/science/article/abs/pii/S0925231221019408. doi:10.1016/j.cogsys.2021.100554

work page doi:10.1016/j.cogsys.2021.100554 2021
[47]

RobustDebias: Debiasing Language Models using Distributionally Robust Optimization

Gandhi D, Singh K, Hegde N. RobustDebias: Debiasing Language Models using Distributionally Robust Optimization. arXiv preprint arXiv:260200405. 2026. Available from: https://arxiv.org/abs/2602.00405. arXiv:2602.00405. doi:10.48550/arXiv.2602.00405. June 30, 2026 17/18

work page doi:10.48550/arxiv.2602.00405 2026
[48]

Universal Patterns of Grammatical Gender in Multilingual Large Language Models

Schröter A, Basirat A. Universal Patterns of Grammatical Gender in Multilingual Large Language Models. In: Proceedings of the 5th Workshop on Multilingual Representation Learning (MRL 2025). Suzhou, China: Association for Computational Linguistics; 2025. p. 34-46. Available from: https://aclanthology.org/2025.mrl-main.3/. doi:10.18653/v1/2025.mrl-main.3

work page doi:10.18653/v1/2025.mrl-main.3 2025
[49]

Do Multilingual Transformers Encode Paninian Grammatical Relations? A Layer-wise Probing Study

Kumar A, Sharma D, Krishnamurthy P. Do Multilingual Transformers Encode Paninian Grammatical Relations? A Layer-wise Probing Study. In: Proceedings of the Third Workshop on Quantitative Syntax (QUASY, SyntaxFest 2025). Ljubljana, Slovenia: Association for Computational Linguistics

2025
[50]

p. 124-30. Available from:https://aclanthology.org/2025.quasy-1.15/. doi:10.18653/v1/2025.quasy-1.15

work page doi:10.18653/v1/2025.quasy-1.15 2025
[51]

Causal Intervention and Counterfactual Reasoning for Multi-modal Fake News Detection

Chen Z, Hu L, Li W, Shao Y, Nie L. Causal Intervention and Counterfactual Reasoning for Multi-modal Fake News Detection. In: Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). Toronto, Canada: Association for Computational Linguistics; 2023. p. 627-38. Available from:https://aclanthology.org/20...

work page doi:10.18653/v1/2023.acl-long.37 2023
[52]

Typological Diversity

Ploeger E, Poelman W, de Lhoneux M, Bjerva J. What is “Typological Diversity” in NLP? In: Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. Miami, Florida, USA: Association for Computational Linguistics; 2024. p. 5681-700. Available from: https://aclanthology.org/2024.emnlp-main.326/. doi:10.18653/v1/2024.emnlp-main.3...

work page doi:10.18653/v1/2024.emnlp-main.326 2024

[1] [1]

Social Bias in Multilingual Language Models: A Survey

Gamboa LCL, Feng Y, Lee MG. Social Bias in Multilingual Language Models: A Survey. In: Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. Suzhou, June 30, 2026 13/18 China: Association for Computational Linguistics; 2025. p. 27857-80. Available from: https://aclanthology.org/2025.emnlp-main.1416/. doi:10.18653/v1/2025....

work page doi:10.18653/v1/2025.emnlp-main.1416 2025

[2] [2]

Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings

Bolukbasi T, Chang KW, Zou J, Saligrama V, Kalai A. Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings. arXiv preprint arXiv:160706520. 2016. Available from:https://arxiv.org/abs/1607.06520. doi:https://doi.org/10.48550/arXiv.1607.06520

work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1607.06520 2016

[3] [3]

Measuring Bias in Contextualized Word Representations

Kurita K, Vyas N, Pareek A, Black AW, Tsvetkov Y. Measuring Bias in Contextualized Word Representations. In: Proceedings of the First Workshop on Gender Bias in Natural Language Processing; 2019. p. 166-72. Available from:https://aclanthology.org/W19-3823/. doi:https://doi.org/10.18653/v1/W19-3823

work page doi:10.18653/v1/w19-3823 2019

[4] [4]

Semantics derived automatically from language corpora contain human-like biases

Caliskan A, Bryson JJ, Narayanan A. Semantics derived automatically from language corpora contain human-like biases. Science. 2017;356(6334):183-6. Available from: https://arxiv.org/abs/1608.07187. doi:https://doi.org/10.1126/science.aal4230

work page internal anchor Pith review Pith/arXiv arXiv doi:10.1126/science.aal4230 2017

[5] [5]

Examining Gender Bias in Languages with Grammatical Gender

Zhou P, Shi W, Zhao J, Huang KH, Chen M, Cotterell R, et al. Examining Gender Bias in Languages with Grammatical Gender. In: Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP); 2019. p. 5276-84. Available from:https://aclanthology.o...

work page doi:10.18653/v1/d19-1531 2019

[6] [6]

OSCaR: Orthogonal Subspace Correction and Rectification of Biases in Word Embeddings

Dev S, Li T, Phillips JM, Srikumar V. OSCaR: Orthogonal Subspace Correction and Rectification of Biases in Word Embeddings. In: Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing. Online and Punta Cana, Dominican Republic: Association for Computational Linguistics; 2021. p. 5034-50. Available from:https://aclanthology.o...

work page doi:10.18653/v1/2021.emnlp-main.411 2021

[7] [7]

Bias and Fairness in Large Language Models: A Survey

Gallegos IO, Rossi RA, Barrow J, Tanjim MM, Kim S, Dernoncourt F, et al. Bias and Fairness in Large Language Models: A Survey. Computational Linguistics. 2024;50(3):649-704. Available from: https://aclanthology.org/2024.cl-3.8/. doi:10.1162/coli_a_00498

work page doi:10.1162/coli_a_00498 2024

[8] [8]

A Survey on Fairness in Large Language Models; 2023

Li Y, Du M, Song R, Wang X, Wang Y. A Survey on Fairness in Large Language Models; 2023. Available from:https://arxiv.org/abs/2308.10149. arXiv:2308.10149. doi:10.48550/arXiv.2308.10149

work page doi:10.48550/arxiv.2308.10149 2023

[9] [9]

Multilingual Denoising Pre-training for Neural Machine Translation

Liu Y, Gu J, Goyal N, Li X, Edunov S, Ghazvininejad M, et al. Multilingual Denoising Pre-training for Neural Machine Translation. Transactions of the Association for Computational Linguistics. 2020;8:726-42. Available from:https://aclanthology.org/2020.tacl-1.47/. doi:https://doi.org/10.1162/tacl_a_00343

work page doi:10.1162/tacl_a_00343 2020

[10] [10]

MarIA and BETO are sexist: evaluating gender bias in large language models for Spanish

Garrido-Muñoz I, Martínez-Santiago F, Montejo-Ráez A. MarIA and BETO are sexist: evaluating gender bias in large language models for Spanish. Language Resources and Evaluation. 2023;58(4):1387-417. Available from: https://link.springer.com/article/10.1007/s10579-023-09670-3. doi:https://doi.org/10.1007/s10579-023-09670-3

work page doi:10.1007/s10579-023-09670-3 2023

[11] [11]

Corbett GG. Gender. Cambridge Textbooks in Linguistics. Cambridge University Press; 1991

1991

[12] [12]

Identifying and Reducing Gender Bias in Word-Level Language Models

Bordia S, Bowman SR. Identifying and Reducing Gender Bias in Word-Level Language Models. In: Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Student Research Workshop; 2019. p. 7-15. Available from: https://aclanthology.org/N19-3002/. doi:https://doi.org/10.18653/v1/N19-3002

work page doi:10.18653/v1/n19-3002 2019

[13] [13]

Dresses and ties: the effect of grammatical gender and stereotypical semantic bias in three Spanish-speaking communities

Zunino GM, Aguilar M, Stetie NA, Martínez Rebolledo C, Hinojosa JA. Dresses and ties: the effect of grammatical gender and stereotypical semantic bias in three Spanish-speaking communities. Language and Cognition. 2025;17:e35. Available from: June 30, 2026 14/18 https://www.cambridge.org/core/journals/language-and-cognition/article/dresses-and -ties-the-e...

work page doi:10.1017/langcog.2025.4 2025

[14] [14]

Sex, Syntax, and Semantics

Boroditsky L, Schmidt LA, Phillips W. Sex, Syntax, and Semantics. In: Language in Mind: Advances in the Study of Language and Thought; 2003. p. 61-79. Available from:https://direct.mit.edu/b ooks/edited-volume/1917/chapter-abstract/52672/Sex-Syntax-and-Semantics. doi:https://doi.org/10.7551/mitpress/4117.003.0010

work page doi:10.7551/mitpress/4117.003.0010 2003

[15] [15]

What an Elegant Bridge: Multilingual LLMs are Biased Similarly in Different Languages

Mihaylov V, Shtedritski A. What an Elegant Bridge: Multilingual LLMs are Biased Similarly in Different Languages. In: Proceedings of the Fourth Workshop on Multilingual Representation Learning (MRL 2024); 2024. p. 22-9. Available from:https://aclanthology.org/2024.mrl-1.2/. doi:https://doi.org/10.18653/v1/2024.mrl-1.2

work page doi:10.18653/v1/2024.mrl-1.2 2024

[16] [16]

The Causal Influence of Grammatical Gender on Distributional Semantics

Stańczak K, Du K, Williams A, Augenstein I, Cottrell R. The Causal Influence of Grammatical Gender on Distributional Semantics. Transactions of the Association for Computational Linguistics. 2024;12:1672-85. Available from:https://aclanthology.org/2024.tacl-1.90/. doi:https://doi.org/10.1162/tacl_a_00723

work page doi:10.1162/tacl_a_00723 2024

[17] [17]

Investigating grammatical abstraction in language models using few-shot learning of novel noun gender

Sukumaran P, Houghton C, Kazanina N. Investigating grammatical abstraction in language models using few-shot learning of novel noun gender. In: Findings of the Association for Computational Linguistics: EACL 2024. St. Julian’s, Malta: Association for Computational Linguistics; 2024. p. 747-65. Available from:https://aclanthology.org/2024.findings-eacl.50/...

work page doi:10.18653/v1/2024.findings-eacl.50 2024

[18] [18]

Beyond Content: How Grammatical Gender Shapes Visual Representation in Text-to-Image Models

Saeed M, Raza S, Vayani A, Abdul-Mageed M, Emami A, Shehata S. Beyond Content: How Grammatical Gender Shapes Visual Representation in Text-to-Image Models. Findings of the Association for Computational Linguistics: EMNLP 2025. 2025. Available from: https://aclanthology.org/2025.findings-emnlp.1343/. doi:10.18653/v1/2025.findings-emnlp.1343

work page doi:10.18653/v1/2025.findings-emnlp.1343 2025

[19] [19]

Measuring Gender Bias in Word Embeddings of Gendered Languages Requires Disentangling Grammatical Gender Signals

Sabbaghi SO, Caliskan A. Measuring Gender Bias in Word Embeddings of Gendered Languages Requires Disentangling Grammatical Gender Signals. In: Proceedings of the 2022 AAAI/ACM Conference on AI, Ethics, and Society; 2022. p. 518-27. Available from: https://dl.acm.org/doi/10.1145/3514094.3534176. doi:https://doi.org/10.1145/3514094.3534176

work page doi:10.1145/3514094.3534176 2022

[20] [21]

Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings

Bolukbasi T, Chang KW, Zou JY, Saligrama V, Kalai AT. Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings. In: Advances in Neural Information Processing Systems 29; 2016. p. 4349-57. Available from:https://proceedings.neurips.cc/paper _files/paper/2016/hash/a486cd07e4ac3d270571622f4f316ec5-Abstract.html

2016

[21] [23]

On Measuring Social Biases in Sentence Encoders

May C, Wang A, Bordia S, Bowman SR, Rudinger R. On Measuring Social Biases in Sentence Encoders. In: Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers)

2019

[22] [24]

p. 622-8. Available from:https://aclanthology.org/N19-1063/. doi:https://doi.org/10.18653/v1/N19-1063. June 30, 2026 15/18

work page doi:10.18653/v1/n19-1063 2026

[23] [25]

Lipstick on a Pig: Debiasing Methods Cover up Systematic Gender Biases in Word Embeddings But do not Remove Them

Gonen H, Goldberg Y. Lipstick on a Pig: Debiasing Methods Cover up Systematic Gender Biases in Word Embeddings But do not Remove Them. In: Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers); 2019. p. 609-14. Available from: https:/...

work page doi:10.18653/v1/n19-1061 2019

[24] [26]

Bias in Bios: A Case Study of Semantic Representation Bias in a High-Stakes Setting

De-Arteaga M, Romanov A, Wallach H, Chayes J, Borgs C, Chouldechova A, et al. Bias in Bios: A Case Study of Semantic Representation Bias in a High-Stakes Setting. arXiv preprint arXiv:190109451

[25] [27]

Bias in Bios: A Case Study of Semantic Representation Bias in a High-Stakes Setting

Available from:https://arxiv.org/abs/1901.09451. doi:https://doi.org/10.48550/arXiv.1901.09451

work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1901.09451 1901

[26] [28]

Mitigating Gender Bias in Contextual Word Embeddings

Yarrabelly N, Damodaran V, Su FG. Mitigating Gender Bias in Contextual Word Embeddings. arXiv preprint arXiv:241112074. 2024. Available from:https://arxiv.org/abs/2411.12074. doi:https://doi.org/10.48550/arXiv.2411.12074

work page doi:10.48550/arxiv.2411.12074 2024

[27] [29]

Assessing Social and Intersectional Biases in Contextualized Word Representations

Tan YC, Celis LE. Assessing Social and Intersectional Biases in Contextualized Word Representations. arXiv preprint arXiv:191101485. 2019. Available from:https://arxiv.org/abs/1911.01485. doi:10.48550/arXiv.1911.01485

work page doi:10.48550/arxiv.1911.01485 2019

[28] [30]

Stereotype and Skew: Quantifying Gender Bias in Pre-trained and Fine-tuned Language Models

de Vassimon Manela D, Errington D, Fisher T, van Breugel B, Minervini P. Stereotype and Skew: Quantifying Gender Bias in Pre-trained and Fine-tuned Language Models. In: Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume. Online: Association for Computational Linguistics; 2021. p. 2232-4...

work page doi:10.18653/v1/2021.eacl-main.190 2021

[29] [31]

How Does Grammatical Gender Affect Noun Representations in Gender-Marking Languages? In: Proceedings of the 23rd Conference on Computational Natural Language Learning (CoNLL); 2019

Gonen H, Kementchedjhieva Y, Goldberg Y. How Does Grammatical Gender Affect Noun Representations in Gender-Marking Languages? In: Proceedings of the 23rd Conference on Computational Natural Language Learning (CoNLL); 2019. p. 463-71. Available from: https://aclanthology.org/K19-1043/. doi:https://doi.org/10.18653/v1/K19-1043

work page doi:10.18653/v1/k19-1043 2019

[30] [32]

Gendered Grammar or Ingrained Bias? Exploring Gender Bias in Icelandic Language Models

Friðriksdóttir SR, Einarsson H. Gendered Grammar or Ingrained Bias? Exploring Gender Bias in Icelandic Language Models. In: Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024). Torino, Italia: ELRA and ICCL; 2024. p. 7596-610. Available from: https://aclanthology.org/202...

2024

[31] [33]

Does grammatical gender affect object concepts? Registered replication of Phillips and Boroditsky (2003)

Elpers N, Jensen G, Holmes KJ. Does grammatical gender affect object concepts? Registered replication of Phillips and Boroditsky (2003). Journal of Memory and Language. 2022;127:104357. Available from: https://www.sciencedirect.com/science/article/abs/pii/S0749596X22000444. doi:https://doi.org/10.1016/j.jml.2022.104357

work page doi:10.1016/j.jml.2022.104357 2003

[32] [34]

Leveraging Large Language Models to Measure Gender Representation Bias in Gendered Language Corpora

Derner E, de la Fuente SS, Gutiérrez Y, Moreda P, Oliver N. Leveraging Large Language Models to Measure Gender Representation Bias in Gendered Language Corpora. In: Proceedings of the 6th Workshop on Gender Bias in Natural Language Processing (GeBNLP); 2025. p. 468-83. Available from: https://aclanthology.org/2025.gebnlp-1.39/. doi:https://doi.org/10.1865...

work page doi:10.18653/v1/2025.gebnlp-1.39 2025

[33] [35]

EuroGEST: Investigating gender stereotypes in multilingual language models

Rowe J, Klimaszewski M, Guillou L, Vallor S, Birch A. EuroGEST: Investigating gender stereotypes in multilingual language models. In: Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing; 2025. p. 32074-96. Available from: https://aclanthology.org/2025.emnlp-main.1632/. doi:https://doi.org/10.18653/v1/2025.emnlp-main.1632

work page doi:10.18653/v1/2025.emnlp-main.1632 2025

[34] [36]

Spanish Pre-trained BERT Model and Evaluation Data

Cañete J, Chaperon G, Fuentes R, Ho JH, Kang H, Pérez J. Spanish Pre-trained BERT Model and Evaluation Data. arXiv preprint arXiv:230802976. 2023. Available from: https://arxiv.org/abs/2308.02976. doi:https://doi.org/10.48550/arXiv.2308.02976. June 30, 2026 16/18

work page doi:10.48550/arxiv.2308.02976 2023

[35] [37]

Conneau, K

Conneau A, Khandelwal K, Goyal N, Chaudhary V, Wenzek G, Guzmán F, et al. Unsupervised Cross-lingual Representation Learning at Scale. In: Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics; 2020. p. 8440-51. Available from: https://aclanthology.org/2020.acl-main.747/. doi:https://doi.org/10.18653/v1/2020.acl-main.747

work page doi:10.18653/v1/2020.acl-main.747 2020

[36] [38]

CamemBERT: a Tasty French Language Model

Martin L, Muller B, Ortiz Suárez PJ, Dupont Y, Romary L, de la Clergerie É, et al. CamemBERT: a Tasty French Language Model. In: Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics; 2020. p. 7203-19. Available from: https://aclanthology.org/2020.acl-main.645/. doi:https://doi.org/10.18653/v1/2020.acl-main.645

work page doi:10.18653/v1/2020.acl-main.645 2020

[37] [39]

Word embeddings are biased: But whose bias are they reflecting? AI & Society

Petreski D, Hashim IC. Word embeddings are biased: But whose bias are they reflecting? AI & Society. 2023;38:975-82. Available from:https://www.researchgate.net/publication/342995921 _Social_biases_in_word_embeddings_and_their_relation_to_human_cognition. doi:https://doi.org/10.1007/s00146-022-01443-w

work page doi:10.1007/s00146-022-01443-w 2023

[38] [40]

Detecting gender bias in Arabic text through word embeddings

Mourad A, Abu Salem FK, Elbassuoni S. Detecting gender bias in Arabic text through word embeddings. PLOS ONE. 2025. Available from: https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0319301. doi:https://doi.org/10.1371/journal.pone.0319301

work page doi:10.1371/journal.pone.0319301 2025

[39] [41]

Job Unfair: An Investigation of Gender and Occupational Bias in Free-Form Text Completions by LLMs

Casula C, Vecellio Salto S, Leonardelli E, Tonelli S. Job Unfair: An Investigation of Gender and Occupational Bias in Free-Form Text Completions by LLMs. In: Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. Suzhou, China: Association for Computational Linguistics; 2025. p. 22759-77. Available from: https://aclantholo...

work page doi:10.18653/v1/2025.emnlp-main.1159 2025

[40] [42]

Bias Is a Subspace, Not a Coordinate: A Geometric Rethinking of Post-hoc Debiasing in Vision-Language Models

Zhao D, Li W, Shen Z, Qiu Y, Xu B, Chen H, et al. Bias Is a Subspace, Not a Coordinate: A Geometric Rethinking of Post-hoc Debiasing in Vision-Language Models. arXiv preprint arXiv:251118123. 2025. Available from:https://arxiv.org/abs/2511.18123. arXiv:2511.18123. doi:10.48550/arXiv.2511.18123

work page doi:10.48550/arxiv.2511.18123 2025

[41] [43]

Grammatical Gender Associations Outweigh Topical Gender Bias in Cross-linguistic Word Embeddings

McCurdy K, Serbetci O. Grammatical Gender Associations Outweigh Topical Gender Bias in Cross-linguistic Word Embeddings. arXiv preprint arXiv:200508864. 2020. Available from: https://arxiv.org/abs/2005.08864. doi:https://doi.org/10.48550/arXiv.2005.08864

work page doi:10.48550/arxiv.2005.08864 2020

[42] [44]

Discovering Low-rank Subspaces for Language-agnostic Multilingual Representations

Xie Z, Zhao H, Yu T, Li S. Discovering Low-rank Subspaces for Language-agnostic Multilingual Representations. In: Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing. Abu Dhabi, United Arab Emirates: Association for Computational Linguistics; 2022. p. 5617-33. Available from:https://aclanthology.org/2022.emnlp-main.379/....

work page doi:10.18653/v1/2022.emnlp-main.379 2022

[43] [45]

Bias Is a Subspace, Not a Coordinate: A Geometric Rethinking of Post-hoc Debiasing in Vision-Language Models

Gupta S, Balia R, Angioni D, Brau F, Pintor M, Demontis A, et al. Bias Is a Subspace, Not a Coordinate: A Geometric Rethinking of Post-hoc Debiasing in Vision-Language Models. arXiv preprint arXiv:251118123. 2025. Available from:https://arxiv.org/abs/2511.18123. arXiv:2511.18123

work page arXiv 2025

[44] [46]

Bias in language models: A survey

de Vassimon Manela D, Errington D, Fisher T, van Breugel B, Minervini P. Bias in language models: A survey. Cognitive Systems Research. 2021;70:100554. Available from: https://www.sciencedirect.com/science/article/abs/pii/S0925231221019408. doi:10.1016/j.cogsys.2021.100554

work page doi:10.1016/j.cogsys.2021.100554 2021

[45] [47]

RobustDebias: Debiasing Language Models using Distributionally Robust Optimization

Gandhi D, Singh K, Hegde N. RobustDebias: Debiasing Language Models using Distributionally Robust Optimization. arXiv preprint arXiv:260200405. 2026. Available from: https://arxiv.org/abs/2602.00405. arXiv:2602.00405. doi:10.48550/arXiv.2602.00405. June 30, 2026 17/18

work page doi:10.48550/arxiv.2602.00405 2026

[46] [48]

Universal Patterns of Grammatical Gender in Multilingual Large Language Models

Schröter A, Basirat A. Universal Patterns of Grammatical Gender in Multilingual Large Language Models. In: Proceedings of the 5th Workshop on Multilingual Representation Learning (MRL 2025). Suzhou, China: Association for Computational Linguistics; 2025. p. 34-46. Available from: https://aclanthology.org/2025.mrl-main.3/. doi:10.18653/v1/2025.mrl-main.3

work page doi:10.18653/v1/2025.mrl-main.3 2025

[47] [49]

Do Multilingual Transformers Encode Paninian Grammatical Relations? A Layer-wise Probing Study

Kumar A, Sharma D, Krishnamurthy P. Do Multilingual Transformers Encode Paninian Grammatical Relations? A Layer-wise Probing Study. In: Proceedings of the Third Workshop on Quantitative Syntax (QUASY, SyntaxFest 2025). Ljubljana, Slovenia: Association for Computational Linguistics

2025

[48] [50]

p. 124-30. Available from:https://aclanthology.org/2025.quasy-1.15/. doi:10.18653/v1/2025.quasy-1.15

work page doi:10.18653/v1/2025.quasy-1.15 2025

[49] [51]

Causal Intervention and Counterfactual Reasoning for Multi-modal Fake News Detection

Chen Z, Hu L, Li W, Shao Y, Nie L. Causal Intervention and Counterfactual Reasoning for Multi-modal Fake News Detection. In: Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). Toronto, Canada: Association for Computational Linguistics; 2023. p. 627-38. Available from:https://aclanthology.org/20...

work page doi:10.18653/v1/2023.acl-long.37 2023

[50] [52]

Typological Diversity

Ploeger E, Poelman W, de Lhoneux M, Bjerva J. What is “Typological Diversity” in NLP? In: Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. Miami, Florida, USA: Association for Computational Linguistics; 2024. p. 5681-700. Available from: https://aclanthology.org/2024.emnlp-main.326/. doi:10.18653/v1/2024.emnlp-main.3...

work page doi:10.18653/v1/2024.emnlp-main.326 2024