Pith. sign in

REVIEW 3 major objections 2 minor 46 references

The paper claims that instruction tuning on human-crafted gender-inclusive instructions, guided by a linguistic system prompt, can make gender inclusivity an inherent feature of Polish and multilingual LLMs, reducing masculine-default Polis

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

Abstract promises gender-inclusive Polish LLM tuning with the IPIS dataset, while the full text is an unrelated quantum transformer paper; no evidence for the declared claims is present.

T0 review reviewed 2026-08-05 challenge →

load-bearing objection The abstract and the full text are two different papers; the declared gender-inclusivity study has no evidence base in this submission. the 3 major comments →

arxiv 2508.18466 v1 pith:CYJVJAOZ submitted 2025-08-25 cs.CL

Integrating gender inclusivity into large language models via instruction tuning

Alina Wr\'oblewska , Bartosz \.Zuk This is my paper
classification cs.CL
keywords gender inclusivityinstruction tuningPolish languagelarge language modelsgender biasmasculine defaultIPIS datasetnatural language generation
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The reading

The paper sets out to establish that instruction tuning on a purpose-built dataset—IPIS, a collection of human-crafted gender-inclusive proofreading and Polish-to-English translation instructions—can make gender inclusivity an inherent property of large language models generating Polish. Poland's grammatical-gender system historically defaults to masculine forms for mixed-gender groups, and the authors argue that LLMs trained on Polish text inherit and amplify this bias. The proposed remedy couples the IPIS dataset with a system prompt derived from linguistic theory, and the abstract names five models to be tuned: the multilingual Llama-8B, Mistral-7B, and Mistral-Nemo, and the Polish-specific Bielik and PLLuM. The full text supplied with this submission, however, is a different paper on quantum transformers and contains no experiment, dataset description, or metric supporting the abstract's claims, so the evidence for the stated contribution is not present in this manuscript.

Core claim

The central claim is that IPIS-tuning—supervised instruction tuning on human-crafted gender-inclusive proofreading instructions (Polish) and translation instructions (Polish-to-English), guided by an explicitly worded system prompt grounded in linguistic theory—integrates gender inclusivity as an inherent feature of the model rather than a post-hoc generation constraint. If correct, the tuned models would produce measurably less masculine-default Polish, the form that currently dominates references to men, women, and mixed-gender groups. The provided full text does not describe or report the experiments, so within this submission the claim rests on the abstract alone.

What carries the argument

The central object is the IPIS dataset paired with a theory-derived system prompt. IPIS supplies human-crafted proofreading and translation instructions that model inclusive Polish forms; the system prompt encodes explicit gender-inclusive guidelines informed by a linguistic framework. The mechanism is instruction tuning: fine-tuning the named multilingual and Polish LLMs on these instructions so that inclusive usage becomes the models' default behavior. The dataset and prompt carry the argument in the sense that all claimed bias reduction is attributed to them.

Load-bearing premise

The load-bearing premise is that instruction tuning on the IPIS dataset and its linguistically motivated system prompt actually changes the five models' generation toward gender-inclusive Polish forms—a premise the attached full text, which is a different paper on quantum transformers, does not put to any test.

What would settle it

Run a held-out set of Polish prompts that require gendered agreement (references to women and mixed-gender groups) through each of the five named models, before and after IPIS tuning, and count masculine-default versus inclusive forms in the outputs; if the tuned models show no significant reduction in masculine-default forms relative to their baselines, the central claim is refuted. A cheaper antecedent check: locate the described experiment sections in the manuscript, since the supplied full text reports none.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • If the paper is right, Polish-language chatbots, translators, and generators built on the tuned models would stop defaulting to masculine forms when referring to women or mixed-gender groups.
  • Gender-inclusive behavior would live in the model weights, not in a post-processing wrapper, so it would persist across prompts and tasks without extra decoding constraints.
  • The IPIS instruction set would double as a reusable benchmark resource for measuring and countering masculine-default bias in other gendered languages.
  • The approach would offer a systematic, theory-grounded alternative to ad-hoc rule-based gender correction in Polish text generation.

Where Pith is reading between the lines

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

  • I infer that the instruction-tuning recipe, if validated, can be ported to other Slavic languages with similar masculine-default conventions (e.g., Czech, Slovak, Ukrainian) by translating and localizing the proofreading instructions; the paper does not extend its claim that far.
  • I infer that a decisive evaluation is easy to specify: take each of the five named models, run a held-out benchmark of Polish prompts that force grammatical-gender agreement, compare tuned versus baseline outputs on the rate of masculine-default forms; the abstract's promise of 'measurably less' bias implies exactly such a metric.
  • I infer that calling the behavior 'inherent' commits the approach to more than prompt-following—it implies the inclusive default survives distribution shift and prompt rephrasing—which is a stronger, testable property the manuscript does not demonstrate.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 2 minor

Summary. This submission presents an abstract claiming that the authors IPIS-tune Llama-8B, Mistral-7B, Mistral-Nemo, Bielik, and PLLuM on a human-crafted Polish gender-inclusive proofreading/translation dataset, with the goal of reducing masculine-default generation. However, the full text following the abstract is arXiv:2508.18464v2, “Vectorized Attention with Learnable Encoding for Quantum Transformer,” which describes a vectorized quantum self-attention mechanism, quantum perplexity, and experiments on the Brown corpus. None of the declared models, the IPIS dataset, the instruction-tuning protocol, or any gender-bias metric appears anywhere in the full text. The paper’s central claim is therefore unsupported by the submitted evidence.

Significance. If the claimed IPIS-tuning study were executed as described and evaluated appropriately, it would address a relevant social and technical problem in Polish NLP, where masculine-default forms are a documented bias. The proposal is non-circular in principle: it uses an external human-crafted dataset and compares tuned models against baselines. However, as submitted, the paper contains no dataset description, no tuning details, no evaluation metric, and no result for any of the five named models. The only evidence for the abstract’s claim is the abstract itself. The significance of the actual full text (quantum transformer) is a separate matter and is not the declared contribution.

major comments (3)
  1. [Full text vs. abstract] The full text is a different paper. From the Introduction through the Appendix, the content concerns VQDP, VNQE, quantum perplexity, Qiskit/IBM Kingston, and Brown-corpus language modeling. I could find no occurrence of 'IPIS', 'Polish', 'gender', 'Bielik', 'PLLuM', 'Llama', or 'Mistral' in the body. Consequently, the claimed IPIS instruction-tuning experiments and the claim that tuned models generate less masculine-default Polish have no supporting content. This is not a correctable weakness in an otherwise sound derivation; the evidence base for the declared central claim is absent.
  2. [Abstract-only evaluation] If I restrict attention to the abstract, it states an intention ('aims to integrate') rather than a demonstrated result. It does not specify the size or composition of the IPIS dataset, the instruction format, the PEFT/full fine-tuning setting, the number of training steps, or the evaluation metric for gender bias. Without these, the claim that five models become 'inherently' inclusive cannot be checked. The abstract also does not report any quantitative result, so no inference about effectiveness is possible.
  3. [Generalization claim] The abstract implies that instruction tuning on proofreading and translation instructions generalizes to inherent generation behavior outside the training distribution. The submission provides no test of this transfer (e.g., nontrivial free-generation prompts, coreference resolution, or diagnostic templates). Even if the missing experiments were added, this extrapolation would need explicit evidence. As submitted, it is an unevaluated assumption.
minor comments (2)
  1. [Title and abstract] The title and abstract describe a gender-inclusivity paper, while the body is a quantum-transformer paper. This mismatch should be resolved; one of the two is incorrect.
  2. [Typographical errors in full text] The quantum-transformer text contains typos such as 'theoery' (§2.2), 'matrix multiplcation' (§3.1), and 'a end-to-end' (§5). These are presentation issues, though they are in the unrelated portion of the submission.

Circularity Check

0 steps flagged

No circular step can be exhibited; the declared gender-inclusivity paper is backed only by a mismatched full text (a quantum-transformer paper), making the claims unsupported rather than circular.

full rationale

The submission's abstract claims IPIS instruction tuning of Llama-8B, Mistral-7B, Mistral-Nemo, Bielik, and PLLuM to mitigate masculine-default Polish, but the supplied full text is arXiv:2508.18464v2, 'Vectorized Attention with Learnable Encoding for Quantum Transformer,' by different authors. That full text contains no mention of IPIS, Polish, gender, Bielik, PLLuM, instruction tuning, or any model named in the abstract. Consequently, there is no derivation chain in which a prediction reduces by construction to its inputs: no IPIS dataset description, no tuning protocol, no evaluation metric, and no held-out comparison are present to inspect. The abstract's stated design—human-crafted inclusive proofreading instructions, tuned models compared against baselines—would, if implemented as claimed, be non-circular, since evaluation would rest on external human judgments or separate benchmarks rather than on a fitted parameter renamed as a prediction. The quantum-transformer full text, taken on its own, is also not circular: it benchmarks against NanoGPT, Q-LSTM, Quixer, and a hybrid QT on the Brown corpus, and Appendix A gives a self-contained shot-consistency derivation. Its citations to Balewski et al. (2024, 2025) are published, parameter-free encoding schemes with stated assumptions; they are not invoked as uniqueness theorems forbidding alternatives. The serious deficiency here is an evidentiary mismatch—the declared paper's evidence base is absent—not a circularity of the kind this review targets. Therefore the circularity score is 0.

Axiom & Free-Parameter Ledger

0 free parameters · 3 axioms · 0 invented entities

The central claim rests on domain premises about Polish grammar and about transfer of instruction tuning; none are evidenced in the submitted text. The full text introduces additional entities (VQDP, VNQE, quantum self-attention circuits), but these belong to a different paper and are not attributed to the declared study. No free parameters can be extracted because the declared paper's methods and tables are absent; the quantum paper's hyperparameters (Table 5) are not attributable to the declared research.

axioms (3)
  • domain assumption Masculine grammatical forms are the generic default in contemporary Polish, producing gender-imbalanced LLM outputs.
    Stated in the abstract as motivation; plausible and consistent with Polish linguistics, but the submission provides no model-output evidence.
  • domain assumption Instruction tuning on human-crafted gender-inclusive proofreading instructions makes inclusivity an inherent model property rather than a task-specific fit.
    The abstract's central mechanism; unverifiable because no tuning or evaluation appears in the submitted full text.
  • domain assumption A system prompt with explicit Polish gender-inclusive guidelines contributes to the effect beyond the tuning data.
    The abstract states this design choice; no ablation is provided anywhere in the submission.

reviewed 2026-08-05 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Integrating gender inclusivity into large language models via instruction tuning." pith.science (2026). https://pith.science/paper/CYJVJAOZ

@misc{pith2026250818466,
  author       = {Pith},
  title        = {Pith review of: Integrating gender inclusivity into large language models via instruction tuning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CYJVJAOZ}},
  note         = {Machine review of arXiv:2508.18466}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Imagine a language with masculine, feminine, and neuter grammatical genders, yet, due to historical and political conventions, masculine forms are predominantly used to refer to men, women and mixed-gender groups. This is the reality of contemporary Polish. A social consequence of this unfair linguistic system is that large language models (LLMs) trained on Polish texts inherit and reinforce this masculine bias, generating gender-imbalanced outputs. This study addresses this issue by tuning LLMs using the IPIS dataset, a collection of human-crafted gender-inclusive proofreading in Polish and Polish-to-English translation instructions. Grounded in a theoretical linguistic framework, we design a system prompt with explicit gender-inclusive guidelines for Polish. In our experiments, we IPIS-tune multilingual LLMs (Llama-8B, Mistral-7B and Mistral-Nemo) and Polish-specific LLMs (Bielik and PLLuM). Our approach aims to integrate gender inclusivity as an inherent feature of these models, offering a systematic solution to mitigate gender bias in Polish language generation.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

46 extracted references · 22 canonical work pages

  1. [1]

    Elham Albaroudi, Taha Mansouri, and Ali Alameer. 2024. https://doi.org/10.3390/ai5010019 A Comprehensive Review of AI Techniques for Addressing Algorithmic Bias in Job Hiring . AI, 5(1):383--404

    doi:10.3390/ai5010019 2024

  2. [2]

    Chantal Amrhein, Florian Schottmann, Rico Sennrich, and Samuel L \"a ubli. 2023. https://doi.org/10.18653/v1/2023.acl-long.246 Exploiting biased models to de-bias text: A gender-fair rewriting model . In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 4486--4506, Toronto, Canada. Assoc...

    doi:10.18653/v1/2023.acl-long.246 2023

  3. [3]

    Maarten De Backer and Ludovic De Cuypere. 2012. https://doi.org/10.1016/j.langsci.2011.10.001 The interpretation of masculine personal nouns in german and dutch: a comparative experimental study . Language Sciences, 34(3):253--268

    work page doi:10.1016/j.langsci.2011.10.001 2012

  4. [4]

    Angana Borah and Rada Mihalcea. 2024. https://doi.org/10.18653/v1/2024.findings-emnlp.545 Towards implicit bias detection and mitigation in multi-agent LLM interactions . In Findings of the Association for Computational Linguistics: EMNLP 2024, pages 9306--9326, Miami, Florida, USA. Association for Computational Linguistics

    work page doi:10.18653/v1/2024.findings-emnlp.545 2024

  5. [5]

    Tuhin Chakrabarty, Vishakh Padmakumar, and He He. 2022. https://doi.org/10.18653/v1/2022.emnlp-main.460 Help me write a poem - instruction tuning as a vehicle for collaborative poetry writing . In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 6848--6863, Abu Dhabi, United Arab Emirates. Association for Compu...

    doi:10.18653/v1/2022.emnlp-main.460 2022

  6. [6]

    Prafulla Kumar Choubey, Anna Currey, Prashant Mathur, and Georgiana Dinu. 2021. https://doi.org/10.18653/v1/2021.emnlp-main.123 GFST : G ender-filtered self-training for more accurate gender in translation . In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 1640--1654, Online and Punta Cana, Dominican Republi...

    work page doi:10.18653/v1/2021.emnlp-main.123 2021

  7. [7]

    Zhibo Chu, Zichong Wang, and Wenbin Zhang. 2024. https://arxiv.org/abs/2404.01349 Fairness in large language models: A taxonomic survey . Preprint, arXiv:2404.01349

    Pith/arXiv arXiv 2024

  8. [8]

    Tommaso Di Noia , Nava Tintarev, Panagiota Fatourou, and Markus Schedl. 2022. https://doi.org/10.1145/3512728 Recommender Systems under European AI Regulations . Communications of the ACM, 65(4):69--73

    work page doi:10.1145/3512728 2022

  9. [9]

    Agnieszka Fale \'n ska, Christine Basta, Marta Costa-juss \`a , Seraphina Goldfarb-Tarrant, and Debora Nozza, editors. 2024. https://aclanthology.org/2024.gebnlp-1.0/ Proceedings of the 5th Workshop on Gender Bias in Natural Language Processing (GeBNLP) . Association for Computational Linguistics, Bangkok, Thailand

    work page 2024

  10. [10]

    GEC. 2016. https://edoc.coe.int/en/gender-equality/6947-gender-equality-glossary.html Gender Equality Glossary . Gender Equality Commission, Council of Europe

    work page 2016

  11. [11]

    GNL-EU. 2018. https://www.europarl.europa.eu/cmsdata/151780/GNL_Guidelines_EN.pdf Gender-Neutral Language in the European Parliament . European Parliment

    work page 2018

  12. [12]

    Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Alex Vaughan, Amy Yang, Angela Fan, Anirudh Goyal, Anthony Hartshorn, Aobo Yang, Archi Mitra, Archie Sravankumar, Artem Korenev, Arthur Hinsvark, Arun Rao, Aston Zhang, Aurelien Rodriguez, Austen Gregerson, Ava S...

    Pith/arXiv arXiv 2024

  13. [13]

    Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen

    Edward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. 2021. https://arxiv.org/abs/2106.09685 Lora: Low-rank adaptation of large language models . Preprint, arXiv:2106.09685

    Pith/arXiv arXiv 2021

  14. [14]

    Albert Q. Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lucile Saulnier, Lélio Renard Lavaud, Marie-Anne Lachaux, Pierre Stock, Teven Le Scao, Thibaut Lavril, Thomas Wang, Timothée Lacroix, and William El Sayed. 2023. https://arxiv.org/abs/2310.0...

    Pith/arXiv arXiv 2023

  15. [15]

    Mateusz Klimaszewski and Alina Wr \'o blewska. 2021. https://aclanthology.org/2021.emnlp-demo.7 COMBO : State-of-the-art morphosyntactic analysis . In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing: System Demonstrations, pages 50--62, Online and Punta Cana, Dominican Republic. Association for Computational Linguistics

    work page 2021

  16. [16]

    Aobo Kong, Shiwan Zhao, Hao Chen, Qicheng Li, Yong Qin, Ruiqi Sun, Xin Zhou, Enzhi Wang, and Xiaohang Dong. 2024. https://doi.org/10.18653/v1/2024.naacl-long.228 Better zero-shot reasoning with role-play prompting . In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technolo...

    doi:10.18653/v1/2024.naacl-long.228 2024

  17. [17]

    Sandra Martinkov \'a , Karolina Stanczak, and Isabelle Augenstein. 2023. https://doi.org/10.18653/v1/2023.bsnlp-1.17 Measuring gender bias in W est S lavic language models . In Proceedings of the 9th Workshop on Slavic Natural Language Processing 2023 (SlavicNLP 2023), pages 146--154, Dubrovnik, Croatia. Association for Computational Linguistics

    work page doi:10.18653/v1/2023.bsnlp-1.17 2023

  18. [18]

    Mistral AI team . 2024. https://mistral.ai/news/mistral-nemo Mistral NeMo . Accessed: Jan 20, 2025

    work page 2024

  19. [19]

    Praneeth Nemani, Yericherla Deepak Joel, Palla Vijay, and Farhana Ferdouzi Liza. 2024. https://doi.org/10.1016/j.nlp.2023.100047 Gender bias in transformers: A comprehensive review of detection and mitigation strategies . Natural Language Processing Journal, 6:100047

    work page arXiv 2024

  20. [20]

    Mara Nunziatini and Sara Diego. 2024. https://aclanthology.org/2024.eamt-1.48/ Implementing gender-inclusivity in MT output using automatic post-editing with LLM s . In Proceedings of the 25th Annual Conference of the European Association for Machine Translation (Volume 1), pages 580--589, Sheffield, UK. European Association for Machine Translation (EAMT)

    work page 2024

  21. [21]

    Krzysztof Ociepa, Łukasz Flis, Remigiusz Kinas, Adrian Gwoździej, Krzysztof Wróbel, SpeakLeash Team , and Cyfronet Team . 2024 a . https://huggingface.co/speakleash/Bielik-11B-v2.3-Instruct Bielik-11b-v2.3-instruct model card . Accessed: 2025-01-27

    work page 2024

  22. [22]

    Krzysztof Ociepa, Łukasz Flis, Krzysztof Wróbel, Adrian Gwoździej, and Remigiusz Kinas. 2024 b . https://arxiv.org/abs/2410.18565 Bielik 7B v0.1: A Polish Language Model -- Development, Insights, and Evaluation . Preprint, arXiv:2410.18565

    arXiv 2024

  23. [23]

    Anaelia Ovalle, Ninareh Mehrabi, Palash Goyal, Jwala Dhamala, Kai-Wei Chang, Richard Zemel, Aram Galstyan, Yuval Pinter, and Rahul Gupta. 2024. https://doi.org/10.18653/v1/2024.findings-naacl.113 Tokenization matters: Navigating data-scarce tokenization for gender inclusive language technologies . In Findings of the Association for Computational Linguisti...

    work page doi:10.18653/v1/2024.findings-naacl.113 2024

  24. [24]

    Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. 2002. https://aclanthology.org/P02-1040 B leu: a method for automatic evaluation of machine translation . In Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics, pages 311--318, Philadelphia, Pennsylvania, USA. Association for Computational Linguistics

    work page 2002

  25. [25]

    Consortium PLLuM. 2025. Pllum: A family of polish large language models

    work page 2025

  26. [26]

    Maja Popovi \'c . 2015. https://doi.org/10.18653/v1/W15-3049 chr F : character n-gram F -score for automatic MT evaluation . In Proceedings of the Tenth Workshop on Statistical Machine Translation, pages 392--395, Lisbon, Portugal. Association for Computational Linguistics

    doi:10.18653/v1/w15-3049 2015

  27. [27]

    Piotr Przyby a. 2022. https://gitlab.clarin-pl.eu/syntactic-tools/lambo LAMBO: Layered Approach to Multi-level BOundary identification

    work page 2022

  28. [28]

    Vipul Raheja, Dimitris Alikaniotis, Vivek Kulkarni, Bashar Alhafni, and Dhruv Kumar. 2024. https://doi.org/10.18653/v1/2024.naacl-long.56 m E d IT : Multilingual text editing via instruction tuning . In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: ...

    doi:10.18653/v1/2024.naacl-long.56 2024

  29. [29]

    Vipul Raheja, Dhruv Kumar, Ryan Koo, and Dongyeop Kang. 2023. https://doi.org/10.18653/v1/2023.findings-emnlp.350 C o E d IT : Text editing by task-specific instruction tuning . In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 5274--5291, Singapore. Association for Computational Linguistics

    doi:10.18653/v1/2023.findings-emnlp.350 2023

  30. [30]

    Chiara Reali, Yulia Esaulova, and Lisa Von Stockhausen. 2015. https://doi.org/10.1017/S0142716414000010 Isolating stereotypical gender in a grammatical gender language: Evidence from eye movements . Applied Psycholinguistics, 36(4):977–1006

    work page doi:10.1017/s0142716414000010 2015

  31. [31]

    Riccardo Regis. 2020. https://doi.org/10.1093/acrefore/9780199384655.013.460 Personal nouns (agent nouns) in the romance languages

    arXiv 2020

  32. [32]

    Andy Rosenbaum, Saleh Soltan, Wael Hamza, Yannick Versley, and Markus Boese. 2022. https://aclanthology.org/2022.coling-1.18/ LINGUIST : Language model instruction tuning to generate annotated utterances for intent classification and slot tagging . In Proceedings of the 29th International Conference on Computational Linguistics, pages 218--241, Gyeongju, ...

    work page 2022

  33. [33]

    Danielle Saunders, Rosie Sallis, and Bill Byrne. 2022. https://doi.org/10.18653/v1/2022.findings-acl.301 First the worst: Finding better gender translations during beam search . In Findings of the Association for Computational Linguistics: ACL 2022, pages 3814--3823, Dublin, Ireland. Association for Computational Linguistics

    doi:10.18653/v1/2022.findings-acl.301 2022

  34. [34]

    Beatrice Savoldi, Jani c a Hackenbuchner, Luisa Bentivogli, Joke Daems, Eva Vanmassenhove, and Jasmijn Bastings, editors. 2024. https://aclanthology.org/2024.gitt-1.0/ Proceedings of the 2nd International Workshop on Gender-Inclusive Translation Technologies . European Association for Machine Translation (EAMT), Sheffield, United Kingdom

    work page 2024

  35. [35]

    Karolina Stanczak and Isabelle Augenstein. 2021. https://arxiv.org/abs/2112.14168 A Survey on Gender Bias in Natural Language Processing . Preprint, arXiv:2112.14168

    Pith/arXiv arXiv 2021

  36. [36]

    Lin Sun, Kai Zhang, Qingyuan Li, and Renze Lou. 2024. https://doi.org/10.1609/aaai.v38i17.29873 Umie: Unified multimodal information extraction with instruction tuning . Proceedings of the AAAI Conference on Artificial Intelligence, 38(17):19062--19070

    doi:10.1609/aaai.v38i17.29873 2024

  37. [37]

    Tony Sun, Andrew Gaut, Shirlyn Tang, Yuxin Huang, Mai ElSherief, Jieyu Zhao, Diba Mirza, Elizabeth Belding, Kai-Wei Chang, and William Yang Wang. 2019. https://doi.org/10.18653/v1/P19-1159 Mitigating gender bias in natural language processing: Literature review . In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, p...

    doi:10.18653/v1/p19-1159 2019

  38. [38]

    o rg Tiedemann, Mikko Aulamo, Daria Bakshandaeva, Michele Boggia, Stig-Arne Gr \

    J \"o rg Tiedemann, Mikko Aulamo, Daria Bakshandaeva, Michele Boggia, Stig-Arne Gr \"o nroos, Tommi Nieminen, Alessandro Raganato\, Yves Scherrer, Raul Vazquez, and Sami Virpioja. 2023. https://doi.org/10.1007/s10579-023-09704-w Democratizing neural machine translation with OPUS-MT . Language Resources and Evaluation, 58:713--755

    doi:10.1007/s10579-023-09704-w 2023

  39. [39]

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, ukasz Kaiser, and Illia Polosukhin. 2017. Attention is All you Need . In Advances in Neural Information Processing Systems, volume 30. Curran Associates, Inc

    work page 2017

  40. [40]

    Zhao, Kelvin Guu, Adams Wei Yu, Brian Lester, Nan Du, Andrew M

    Jason Wei, Maarten Bosma, Vincent Y. Zhao, Kelvin Guu, Adams Wei Yu, Brian Lester, Nan Du, Andrew M. Dai, and Quoc V. Le. 2022. https://arxiv.org/abs/2109.01652 Finetuned language models are zero-shot learners . Preprint, arXiv:2109.01652

    Pith/arXiv arXiv 2022

  41. [41]

    Alina Wróblewska, Martyna Lewandowska, Aleksandra Tomaszewska, Karolina Saputa, and Maciej Ogrodniczuk. 2025. https://doi.org/10.31286/JP.001040 Koncepcja form równościowych z asteryskiem inkluzywnym . Język Polski, CV(2)

    work page doi:10.31286/jp.001040 2025

  42. [42]

    Benfeng Xu, An Yang, Junyang Lin, Quan Wang, Chang Zhou, Yongdong Zhang, and Zhendong Mao. 2023. https://arxiv.org/abs/2305.14688 Expertprompting: Instructing large language models to be distinguished experts . Preprint, arXiv:2305.14688

    Pith/arXiv arXiv 2023

  43. [43]

    Shengyu Zhang, Linfeng Dong, Xiaoya Li, Sen Zhang, Xiaofei Sun, Shuhe Wang, Jiwei Li, Runyi Hu, Tianwei Zhang, Fei Wu, and Guoyin Wang. 2024. https://arxiv.org/abs/2308.10792 Instruction tuning for large language models: A survey . Preprint, arXiv:2308.10792

    arXiv 2024

  44. [44]

    Yue Zhang, Leyang Cui, Deng Cai, Xinting Huang, Tao Fang, and Wei Bi. 2023. https://arxiv.org/abs/2305.13225 Multi-task instruction tuning of llama for specific scenarios: A preliminary study on writing assistance . Preprint, arXiv:2305.13225

    Pith/arXiv arXiv 2023

  45. [45]

    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 INTEGERS output.state before.all mid.sentence after.sentence after.block STRING...

  46. [46]

    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 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

This paper was first reviewed by deepseek-v4-flash on August 5, 2026.