Pith. sign in

REVIEW 4 major objections 6 minor 1 cited by

Information Gain-Guided Causal Intervention for Autonomous Debiasing Large Language Models

T0 review · 4 major / 6 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read An LLM can be debiased by rewriting its training data until the biased feature yields zero information gain about the answer, which improves zero-shot transfer across tasks.

desk verdict A coherent debiasing recipe with a clean criterion, but the paper never isolates the rewriting step from the balancing objective, so the 'across tasks' claim is only partly supported. read the letter →

arxiv 2504.12898 v3 pith:KBRIR3JW submitted 2025-04-17 cs.CL cs.AI

Zhouhao Sun , Xiao Ding , Li Du , Yunpeng Xu , Yixuan Ma , Yang Zhao , Bing Qin , Ting Liu This is my paper · ORCID
classification cs.CLcs.AI
keywords datasetbiaslargelanguagemodelsinformationgaincausalinterventionautomaticdebiasinginstructiontuningzero-shottransferdatarewriting
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

The paper argues that large language models trained on instruction-tuning data learn dataset biases—correlations between non-causal features and answers—and that these biases hurt generalization to new datasets of the same task. It claims that a dataset is debiased with respect to a biased feature exactly when that feature contributes zero information gain about the answer, and that this condition can be met by rewriting examples so that each value of the biased feature appears with each answer equally often. The proposed ICD framework identifies biased features automatically, rewrites a minimal set of examples using the LLM itself, and then fine-tunes on the rebalanced data. The reported result is consistent improvement on zero-shot transfer sets (SNLI, MRPC, IMDB, Squadv2) and challenge sets (HANS, PAWS) with no loss on in-domain accuracy. If correct, the work turns debiasing from a per-task manual exercise into a data-level goal that any LLM can apply, and debiased datasets transfer to other models.

What carries the argument

The load-bearing object is the information gain-guided data debiasing goal: for any biased feature B, the dataset must satisfy IG(Y,B)=0, equivalently N(y_i,b_k)=N(y_j,b_k) for every feature value and answer pair. To reach this state, the framework first identifies biased features automatically with a causal-guided active learning method, then performs a causal intervention do(B=b_k) by using the LLM to rewrite samples via few-shot prompting, and selects which samples to rewrite by solving a linear program whose objective is minimal rewriting subject to the count-balance constraint. The rewriting step is what carries the empirical weight: it alters the joint distribution of biased feature and answer without, by assumption, changing the semantics that determine the correct answer.

What would settle it

A control experiment that rewrites the same number of samples without targeting any biased feature; if transfer gains match ICD's, the improvement is not due to bias removal. A second check is human annotation of rewritten samples for whether the gold answer still holds; if a substantial fraction change meaning, the debiasing explanation is unsupported.

Watch

Extended reading notes

Core claim

The central claim is that predictive but non-causal features ('biased features') in instruction-tuning datasets can be eliminated by enforcing the information-theoretic condition IG(Y,B)=0—the biased feature B provides no information about the answer Y. The paper derives the operational form P(Y|B)=P(Y), realizes it as the count-balancing requirement N(y_i,b_k)=N(y_j,b_k), and achieves this by a causal intervention do(B=b_k) that rewrites a fraction of examples with the LLM itself, solving a linear program to minimize the number of rewrites. Trained on the rewritten data, the LLM no longer relies on the biased feature and transfers better to held-out datasets of the same task, while keeping in-domain performance and even improving general benchmarks such as TruthfulQA. The authors frame this as the first strict criterion for an unbiased instruction-tuning dataset.

Load-bearing premise

The method assumes that few-shot rewriting changes only the targeted biased feature and leaves every other property of the sample that determines the answer unchanged.

Editorial extensions

If this is right

  • ICD improves zero-shot transfer accuracy over vanilla fine-tuning on SNLI, MRPC, IMDB, and Squadv2 while preserving in-domain accuracy on MNLI, QQP, SST2, Squadv1, and TriviaQA.
  • On challenge sets HANS and PAWS, which remove dataset biases, ICD outperforms both vanilla fine-tuning and the prior-knowledge baseline Razor.
  • Datasets debiased by one LLM (Llama3.1-8B) can be used to fine-tune a different LLM (Gemma2-9B) with the same debiasing benefit.
  • General ability benchmarks MMLU, BBH, and TruthfulQA do not decline after debiasing; TruthfulQA and BBH improve.
  • The IG=0 condition provides a formal, task-agnostic definition of what an unbiased instruction-tuning dataset must satisfy.

Reading between the lines

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

  • If the count-balancing criterion is the right target, the same principle could be applied to any biased feature in any dataset, turning debiasing into a data-curation step that does not require knowing the downstream task.
  • Because the paper reports no check that rewritten examples remain semantically equivalent, some of the transfer gains may come from data augmentation rather than bias removal; a random-rewriting control would separate these explanations.
  • The framework inherits the limitation that bias identification requires a task with standard answers, so open-ended or generative tasks still lack an automatic debiasing route within this approach.
  • The linear-programming formulation suggests a general principle: among all datasets satisfying the independence condition, prefer the one closest to the original; this framing could support future work on minimal-intervention debiasing.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper proposes ICD, an information gain-guided causal intervention debiasing framework for LLMs. The method first fine-tunes a pretrained LLM on the original instruction-tuning data, then uses the CAL method to identify biased features, and rewrites a subset of the training data with the LLM itself so that, within each value of the biased feature, the answer counts are approximately balanced. This realizes the stated goal that the biased feature provides zero information gain for the answer (IG(Y,B)=0). The debiased dataset is then used for standard supervised fine-tuning. Experiments across NLI, paraphrase identification, sentiment analysis, and QA report improved zero-shot transfer on SNLI, MRPC, HANS, and PAWS, along with gains on MMLU, BBH, and TruthfulQA, while in-domain accuracy is roughly maintained. The paper also shows that data debiased by Llama3.1-8B can be used to fine-tune Gemma2-9B with similar benefits.

Significance. If the causal attribution is established, the paper would provide a principled, automatic debiasing method with a clean information-theoretic objective: a derivation from IG(Y,B)=0 to the count-balancing rule N(y_i,b_k)=N(y_j,b_k). The strengths include the algebraic clarity of the derivation, the use of standard deviations over three runs, the automatic nature of the pipeline, and the demonstration that a debiased dataset transfers across different base LLMs. The main empirical claim, however, is broader than the evidence: the transfer gains on sentiment and QA are within noise, and the experiments do not isolate the information-gain balancing from the generic effect of LLM rewriting. The paper is therefore of interest but requires additional controls and a more measured claim before it can be accepted.

major comments (4)
  1. [Section 3.3 and Section 5.1] The experiments do not include a control that isolates the causal intervention (Eq. 6 balancing) from the rewriting operation itself. The only manipulation is the few-shot LLM rewriting described in Section 3.3, and the paper compares it only against vanilla SFT and a prior-knowledge baseline. Without a condition that performs the same number of LLM rewrites while not enforcing the information-gain goal (e.g., random paraphrasing or rewriting to balance a non-identified feature), the improvements on SNLI, MRPC, HANS, and PAWS could be attributed to generic data augmentation or to a shift in example difficulty rather than to the removal of the B-Y association. This is load-bearing because the paper's central claim is specifically about the information gain-guided criterion.
  2. [Section 3.3, footnote 3, and Appendix D] There is no semantic-equivalence check for the rewritten examples. Footnote 3 concedes that semantics tightly tied to the biased feature must change (e.g., substituting the person in the answer), and Appendix D checks only that the targeted feature changed (lexical overlap rate, negation presence, or popularity as judged by the biased LLM). The paper should verify that the rewritten samples retain their gold answer and remain plausible, for example through human evaluation or a second-model consistency check, and should report the rate at which rewriting had to change the answer. Without such a check, the observed transfer gains may reflect changes in content difficulty or new artifacts rather than debiasing.
  3. [Table 1] The abstract and Section 5.1 (point 2) claim that ICD achieves consistent improvement across all transfer test sets, but the SA and QA transfer gains are negligible: Llama IMDB +0.1, Gemma IMDB +0.3, Llama Squadv2 +0.4, Gemma Squadv2 +0.3, several of which are within one standard deviation of the vanilla baseline. The claim of improved generalizability should be tempered to the NLI and PI tasks (SNLI, MRPC, HANS, PAWS), or significance testing should be provided to support the cross-task claim.
  4. [Section 5.1 and Appendix E] The comparison with Razor is not a fair automatic-debiasing comparison for QA, because Appendix E states that Razor cannot be applied to QA and is therefore only applied to MNLI, QQP, and SST2; the Razor rows for QA are effectively identical to vanilla. The text should explicitly state this limitation when claiming that ICD outperforms Razor on transfer test sets. Additionally, since the proposed method relies on CAL for biased-feature identification, the paper should include a CAL-based control (e.g., CAL's in-context learning debiasing or a CAL-identified feature without the count-balancing rewrite) to separate the contribution of the new information-gain objective from the contribution of feature identification.
minor comments (6)
  1. [Section 3.2, Eq. (5)] The step from Eq. (4) to Eq. (5) relies on the additional assumption that P(Y) is uniform ("there is no prior knowledge about Y"); this is a modeling choice, not a consequence of IG(Y,B)=0, and should be stated as an assumption.
  2. [Section 5.1] The text refers to "IDB" in the discussion of general ability; this should be "ICD".
  3. [Appendix C] There are typos in the lexical overlap classes: "low lexical overelap" appears twice, and the third class should read "high lexical overlap" instead of "low lexical overelap (bigger than 0.6)".
  4. [Section 3.3] The details of the few-shot rewriting prompt and the retry mechanism are deferred to the supplemental material; including the prompt (or a pointer to it) in an appendix would improve reproducibility.
  5. [Appendix D] For popularity bias, checking the feature using the biased LLM itself is circular; the paper should validate this popularity judgment against human annotation or another reference.
  6. [Table 1] The claim of "maintaining performance on the in-domain datasets" should acknowledge the small drops on MNLI (Llama 84.3 to 83.8, Gemma 76.2 to 75.7) and QQP (Llama 86.3 to 85.7, Gemma 87.0 to 86.5), even if these are within the reported standard deviations.

Circularity Check

0 steps flagged · score 2.0 of 10

No circular reduction: the IG=0 to count-balancing derivation is algebraically direct and the transfer gains are externally evaluated; the only self-referential dependency is the CAL bias-identification citation.

full rationale

The paper's claimed derivation is self-contained: Section 3.2 derives the debiasing goal from IG(Y,B)=0 (Eq. 1) through H(Y|B)=H(Y) (Eq. 2) to the count-balancing condition N(yi,bk)=N(yj,bk) (Eq. 6), and the proof in Appendix B is a direct algebraic manipulation. The uniform-prior assumption used to reach Eq. 5 is a stated design choice, not an input-output inversion. Section 3.3 implements Eq. 6 by LLM rewriting that is checked against the targeted biased feature (Appendix D), and the headline results are measured on external held-out transfer and challenge sets (SNLI, MRPC, IMDB, Squadv2, HANS, PAWS) plus general-ability benchmarks (MMLU, BBH, TruthfulQA). No reported 'prediction' is a fitted parameter, and no evaluation set is used to construct Eq. 6. The only self-referential dependency is Section 3.1's use of CAL (Sun et al. [35]), the same group's prior method, described as 'currently the only automatic method for identifying interpretable biased features'; this is a component-level dependency rather than a circular reduction, because CAL's output is an input feature set and the transfer improvements are externally evaluated. Appendix A flags that CAL cannot handle non-standard-answer tasks, which is a scope limitation, not a circular step. Potential concerns about the lack of a rewriting-only control and small effect sizes are experimental-attribution issues, not circularity. Accordingly, no circular step meets the evidence bar; the minor self-citation warrants a low non-zero score.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central claim rests on the CAL-identified biased feature, the uniform prior over answers, and the assumption that LLM rewriting alters only the intended feature. These are domain assumptions rather than fitted parameters. The chosen discretization thresholds and per-task feature selection are hand choices that the paper does not justify with sensitivity experiments.

free parameters (4)
  • lexical overlap thresholds = 0.4 and 0.6
    The value space of the lexical-overlap biased feature is discretized into low/medium/high using hand-chosen thresholds in Appendix C.
  • retry limit for rewriting = 3
    Section 3.3: rewriting is repeated until the intervention objective is met or until three failed attempts; this is an implementation choice not derived from the theory.
  • number of lexical overlap classes = 3
    Appendix C divides the overlap rate into three bins; no sensitivity analysis is given for this choice.
  • most prominent biased feature per task = the first feature in Appendix G
    Appendix C states the first biased feature listed is used for each task; this selection is not justified by a quantitative prominence criterion.
assumptions (5)
  • domain assumption Predictive non-causal features (biased features) cause poor out-of-distribution generalization when their correlation with answers shifts.
    Section 2 states that biased features lack a causal relation with answers and that distribution shift in these features reduces generalization; this motivates the whole debiasing objective.
  • ad hoc to paper The answer distribution P(Y) should be uniform because there is no prior knowledge about Y.
    Section 3.2 introduces the uniform assumption to convert P(Y|B)=P(Y) into P(yi|B)=P(yj|B); it is a design choice, not a consequence of information theory.
  • domain assumption CAL correctly identifies the biased features that matter for generalization.
    Section 3.1 relies entirely on the authors' prior CAL method for feature identification; no independent verification of these features is provided.
  • domain assumption LLM-based rewriting preserves causal content and changes only the targeted biased feature.
    Footnote 3 in Section 3.3 acknowledges that semantics tied to the biased feature must also be modified; the paper provides no check that rewriting does not introduce new spurious correlations.
  • standard math Sampling frequency approximates probability in the debiasing goal.
    Section 3.2 replaces probabilities with counts N(yi,bk) to make the goal operational.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Information Gain-Guided Causal Intervention for Autonomous Debiasing Large Language Models." pith.science (2026). https://pith.science/paper/KBRIR3JW

@misc{pith2026250412898,
  author       = {Pith},
  title        = {Pith review of: Information Gain-Guided Causal Intervention for Autonomous Debiasing Large Language Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KBRIR3JW}},
  note         = {Machine review of arXiv:2504.12898}
}
read the original abstract

Despite significant progress, recent studies indicate that current large language models (LLMs) may still capture dataset biases and utilize them during inference, leading to the poor generalizability of LLMs. However, due to the diversity of dataset biases and the insufficient nature of bias suppression based on in-context learning, the effectiveness of previous prior knowledge-based debiasing methods and in-context learning based automatic debiasing methods is limited. To address these challenges, we explore the combination of causal mechanisms with information theory and propose an information gain-guided causal intervention debiasing (ICD) framework. To eliminate biases within the instruction-tuning dataset, it is essential to ensure that these biases do not provide any additional information to predict the answers, i.e., the information gain of these biases for predicting the answers needs to be 0. Under this guidance, this framework utilizes a causal intervention-based data rewriting method to automatically and autonomously balance the distribution of instruction-tuning dataset for reducing the information gain. Subsequently, it employs a standard supervised fine-tuning process to train LLMs on the debiased dataset. Experimental results show that ICD can effectively debias LLM to improve its generalizability across different tasks.

Figures

Figures reproduced from arXiv: 2504.12898 by the authors.

Figure 1
Figure 1. The examples in dataset before and after data debiasing. By intervening example 2, an [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. (a) Three steps of Information Gain-Guided Causal Intervention Debiasing Framework. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 4
Figure 4. The density plot curve for the model confidence of answer contradiction when experi￾mented with Gemma2-9B. Prior Knowledge-Based Debiasing Method Because there are no training-based automatic debiasing methods for LLMs. So we choose the latest prior knowledge-based debiasing method Razor [47] and adapt it to the instruction tuning setting of LLMs. More details can be found in the Appendix E. 5 Experimental Results 5… view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. EmoPerso: Enhancing Personality Detection with Self-Supervised Emotion-Aware Modelling

    cs.CL 2025-09 conditional novelty 5.0 of 10

    EmoPerso improves MBTI personality detection by training an emotion head on heuristic pseudo-labels and using cross-attention with reasoning chains, achieving 81.07% Macro-F1 on Kaggle and 68.60% on Pandora.

Reference graph

Works this paper leans on

65 extracted references · 50 canonical work pages · cited by 1 Pith paper

  1. [1]

    Gpt-4 technical report

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

    arXiv 2023

  2. [2]

    A large annotated corpus for learning natural language inference

    Samuel R Bowman, Gabor Angeli, Christopher Potts, and Christopher D Manning. A large annotated corpus for learning natural language inference. In Proceedings of the 2015 Confer- ence on Empirical Methods in Natural Language Processing. Association for Computational Linguistics, 2015

    work page 2015

  3. [3]

    Fairflow: Mitigating dataset biases through undecided learning for natural language understanding

    Jiali Cheng and Hadi Amiri. Fairflow: Mitigating dataset biases through undecided learning for natural language understanding. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 21960–21975, 2024

    work page 2024

  4. [4]

    Information Theory: Coding Theorems for Discrete Memoryless Systems

    Imre Csiszár and János Körner. Information Theory: Coding Theorems for Discrete Memoryless Systems. Cambridge University Press, 2015

    work page 2015

  5. [5]

    Mitigate extrinsic social bias in pre-trained language models via continuous prompts adjustment

    Yiwei Dai, Hengrui Gu, Ying Wang, and Xin Wang. Mitigate extrinsic social bias in pre-trained language models via continuous prompts adjustment. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 11068–11083, 2024

    work page 2024

  6. [6]

    Automatically constructing a corpus of sentential para- phrases

    William B Dolan and Chris Brockett. Automatically constructing a corpus of sentential para- phrases. In Proceedings of the Third International Workshop on Paraphrasing (IWP2005) , 2005

    work page 2005

  7. [7]

    Towards stable natural language understanding via information entropy guided debiasing

    Li Du, Xiao Ding, Zhouhao Sun, Ting Liu, Bing Qin, and Jingshuo Liu. Towards stable natural language understanding via information entropy guided debiasing. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages 2868–2882, 2023

    work page 2023

  8. [8]

    The llama 3 herd of models

    Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Amy Yang, Angela Fan, et al. The llama 3 herd of models. arXiv preprint arXiv:2407.21783, 2024

    arXiv 2024

Show all 65 references
  1. [9]

    The capacity for moral self-correction in large language models

    Deep Ganguli, Amanda Askell, Nicholas Schiefer, Thomas Liao, Kamil˙e Lukoši¯ut˙e, Anna Chen, Anna Goldie, Azalia Mirhoseini, Catherine Olsson, Danny Hernandez, et al. The capacity for moral self-correction in large language models. arXiv preprint arXiv:2302.07459, 2023

    2023 arXiv

  2. [10]

    Measuring massive multitask language understanding

    Dan Hendrycks, Collin Burns, Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song, and Jacob Steinhardt. Measuring massive multitask language understanding. Proceedings of the International Conference on Learning Representations (ICLR), 2021

    2021

  3. [11]

    LoRA: Low-rank adaptation of large language models

    Edward J Hu, yelong shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. LoRA: Low-rank adaptation of large language models. In International Conference on Learning Representations, 2022

    2022

  4. [12]

    Triviaqa: A large scale distantly supervised challenge dataset for reading comprehension

    Mandar Joshi, Eunsol Choi, Daniel S Weld, and Luke Zettlemoyer. Triviaqa: A large scale distantly supervised challenge dataset for reading comprehension. In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages ...

    2017

  5. [13]

    A survey on popularity bias in recommender systems

    Anastasiia Klimashevskaia, Dietmar Jannach, Mehdi Elahi, and Christoph Trattner. A survey on popularity bias in recommender systems. User Modeling and User-Adapted Interaction, 34 (5):1777–1834, 2024

    2024

  6. [14]

    Identifying and mitigating annotation bias in natural language under- standing using causal mediation analysis

    Sitiporn Sae Lim, Can Udomcharoenchaikit, Peerat Limkonchotiwat, Ekapol Chuangsuwanich, and Sarana Nutanong. Identifying and mitigating annotation bias in natural language under- standing using causal mediation analysis. In Findings of the Association for Computational Linguis...

    2024

  7. [15]

    Truthfulqa: Measuring how models mimic human falsehoods

    Stephanie Lin, Jacob Hilton, and Owain Evans. Truthfulqa: Measuring how models mimic human falsehoods. In Proceedings of the 60th Annual Meeting of the Association for Computa- tional Linguistics (Volume 1: Long Papers), pages 3214–3252, 2022

    2022

  8. [16]

    Trustworthy llms: a survey and guideline for evaluating large language models’ alignment

    Yang Liu, Yuanshun Yao, Jean-Francois Ton, Xiaoying Zhang, Ruocheng Guo, Hao Cheng, Yegor Klochkov, Muhammad Faaiz Taufiq, and Hang Li. Trustworthy llms: a survey and guideline for evaluating large language models’ alignment. In Socially Responsible Language Modelling Research, 2023

    2023

  9. [17]

    Causality inspired representation learning for domain generalization

    Fangrui Lv, Jian Liang, Shuang Li, Bin Zang, Chi Harold Liu, Ziteng Wang, and Di Liu. Causality inspired representation learning for domain generalization. In 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 8036–8046. IEEE, 2022

    2022

  10. [18]

    Feature-level debiased natural language understanding

    Yougang Lyu, Piji Li, Yechang Yang, Maarten de Rijke, Pengjie Ren, Yukun Zhao, Dawei Yin, and Zhaochun Ren. Feature-level debiased natural language understanding. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 37, pages 13353–13361, 2023

    2023

  11. [19]

    Learning word vectors for sentiment analysis

    Andrew Maas, Raymond E Daly, Peter T Pham, Dan Huang, Andrew Y Ng, and Christopher Potts. Learning word vectors for sentiment analysis. In Proceedings of the 49th annual meeting of the association for computational linguistics: Human language technologies, pages 142–150, 2011

    2011

  12. [20]

    Right for the wrong reasons: Diagnosing syntactic heuristics in natural language inference

    Tom McCoy, Ellie Pavlick, and Tal Linzen. Right for the wrong reasons: Diagnosing syntactic heuristics in natural language inference. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 3428–3448, 2019

    2019

  13. [21]

    Effect of negation in sentences on sentiment analysis and polarity detection

    Partha Mukherjee, Youakim Badr, Shreyesh Doppalapudi, Satish M Srinivasan, Raghvinder S Sangwan, and Rahul Sharma. Effect of negation in sentences on sentiment analysis and polarity detection. Procedia Computer Science, 185:370–379, 2021

    2021

  14. [22]

    Biases in large language models: Origins, inventory and discussion

    Roberto Navigli, Simone Conia, and Björn Ross. Biases in large language models: Origins, inventory and discussion. ACM Journal of Data and Information Quality, 2023

    2023

  15. [23]

    Adversarial nli: A new benchmark for natural language understanding

    Yixin Nie, Adina Williams, Emily Dinan, Mohit Bansal, Jason Weston, and Douwe Kiela. Adversarial nli: A new benchmark for natural language understanding. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 4885–4901, 2020

    2020

  16. [24]

    In-contextual bias suppression for large language models

    Daisuke Oba, Masahiro Kaneko, and Danushka Bollegala. In-contextual bias suppression for large language models. arXiv preprint arXiv:2309.07251, 2023

    2023 arXiv

  17. [25]

    Causality

    Judea Pearl. Causality. Cambridge university press, 2009

    2009

  18. [26]

    Hypothesis only baselines in natural language inference

    Adam Poliak, Jason Naradowsky, Aparajita Haldar, Rachel Rudinger, and Benjamin Van Durme. Hypothesis only baselines in natural language inference. In Proceedings of the Seventh Joint Conference on Lexical and Computational Semantics, pages 180–191, 2018

    2018

  19. [27]

    Squad: 100,000+ questions for machine comprehension of text

    P Rajpurkar. Squad: 100,000+ questions for machine comprehension of text. arXiv preprint arXiv:1606.05250, 2016

    2016 arXiv

  20. [28]

    Know what you don’t know: Unanswerable ques- tions for squad

    Pranav Rajpurkar, Robin Jia, and Percy Liang. Know what you don’t know: Unanswerable ques- tions for squad. In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 784–789, 2018. 11

    2018

  21. [29]

    Learning from others’ mistakes: Avoiding dataset biases without modeling them

    Victor Sanh, Thomas Wolf, Yonatan Belinkov, and Alexander M Rush. Learning from others’ mistakes: Avoiding dataset biases without modeling them. In International Conference on Learning Representations, 2020

    2020

  22. [30]

    Self-diagnosis and self-debiasing: A proposal for reducing corpus-based bias in nlp

    Timo Schick, Sahana Udupa, and Hinrich Schütze. Self-diagnosis and self-debiasing: A proposal for reducing corpus-based bias in nlp. Transactions of the Association for Computational Linguistics, 9:1408–1424, 2021

    2021

  23. [31]

    Towards debiasing fact verification models

    Tal Schuster, Darsh Shah, Yun Jie Serene Yeo, Daniel Roberto Filizzola Ortiz, Enrico Santus, and Regina Barzilay. Towards debiasing fact verification models. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joi...

    2019

  24. [32]

    Natural language understand- ing with the quora question pairs dataset

    Lakshay Sharma, Laura Graesser, Nikita Nangia, and Utku Evci. Natural language understand- ing with the quora question pairs dataset. arXiv preprint arXiv:1907.01041, 2019

    1907 arXiv

  25. [33]

    Recursive deep models for semantic compositionality over a sentiment treebank

    Richard Socher, Alex Perelygin, Jean Wu, Jason Chuang, Christopher D Manning, Andrew Y Ng, and Christopher Potts. Recursive deep models for semantic compositionality over a sentiment treebank. In Proceedings of the 2013 conference on empirical methods in natural language proce...

    2013

  26. [34]

    Bias in news summarization: Measures, pitfalls and corpora

    Julius Steen and Katja Markert. Bias in news summarization: Measures, pitfalls and corpora. In Findings of the Association for Computational Linguistics ACL 2024, pages 5962–5983, 2024

    2024

  27. [35]

    Causal-guided active learning for debiasing large language models

    Zhouhao Sun, Li Du, Xiao Ding, Yixuan Ma, Yang Zhao, Kaitao Qiu, Ting Liu, and Bing Qin. Causal-guided active learning for debiasing large language models. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , page...

    2024

  28. [36]

    Challenging big-bench tasks and whether chain-of-thought can solve them

    Mirac Suzgun, Nathan Scales, Nathanael Schärli, Sebastian Gehrmann, Yi Tay, Hyung Won Chung, Aakanksha Chowdhery, Quoc Le, Ed Chi, Denny Zhou, et al. Challenging big-bench tasks and whether chain-of-thought can solve them. In Findings of the Association for Compu- tational Lin...

    2023

  29. [37]

    Generalized preference optimization: A unified approach to offline alignment

    Yunhao Tang, Zhaohan Daniel Guo, Zeyu Zheng, Daniele Calandriello, Remi Munos, Mark Rowland, Pierre Harvey Richemond, Michal Valko, Bernardo Avila Pires, and Bilal Piot. Generalized preference optimization: A unified approach to offline alignment. In Forty-first International ...

    2024

  30. [38]

    Gemma 2: Improving open language models at a practical size

    Gemma Team, Morgane Riviere, Shreya Pathak, Pier Giuseppe Sessa, Cassidy Hardin, Surya Bhupatiraju, Léonard Hussenot, Thomas Mesnard, Bobak Shahriari, Alexandre Ramé, et al. Gemma 2: Improving open language models at a practical size. arXiv preprint arXiv:2408.00118, 2024

    2024 arXiv

  31. [39]

    Llama 2: Open foundation and fine-tuned chat models

    Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, et al. Llama 2: Open foundation and fine-tuned chat models. arXiv preprint arXiv:2307.09288, 2023

    2023 arXiv

  32. [40]

    Mind the trade-off: Debiasing nlu models without degrading the in-distribution performance

    Prasetya Ajie Utama, Nafise Sadat Moosavi, and Iryna Gurevych. Mind the trade-off: Debiasing nlu models without degrading the in-distribution performance. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 8717–8729, 2020

    2020

  33. [41]

    Towards debiasing nlu models from unknown biases

    Prasetya Ajie Utama, Nafise Sadat Moosavi, and Iryna Gurevych. Towards debiasing nlu models from unknown biases. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 7597–7610, 2020

    2020

  34. [42]

    Click-conversion multi-task model with position bias mitigation for sponsored search in ecommerce

    Yibo Wang, Yanbing Xue, Bo Liu, Musen Wen, Wenting Zhao, Stephen Guo, and Philip S Yu. Click-conversion multi-task model with position bias mitigation for sponsored search in ecommerce. In Proceedings of the 46th International ACM SIGIR Conference on Research and Development i...

    2023

  35. [43]

    Finetuned language models are zero-shot learners

    Jason Wei, Maarten Bosma, Vincent Zhao, Kelvin Guu, Adams Wei Yu, Brian Lester, Nan Du, Andrew M Dai, and Quoc V Le. Finetuned language models are zero-shot learners. In International Conference on Learning Representations, 2022. 12

    2022

  36. [44]

    A broad-coverage challenge corpus for sentence understanding through inference

    Adina Williams, Nikita Nangia, and Samuel R Bowman. A broad-coverage challenge corpus for sentence understanding through inference. In NAACL-HLT, 2018

    2018

  37. [45]

    Counterfactual debiasing for fact verification

    Weizhi Xu, Qiang Liu, Shu Wu, and Liang Wang. Counterfactual debiasing for fact verification. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 6777–6789, 2023

    2023

  38. [46]

    Mitigating biases for instruction-following language models via bias neurons elimination

    Nakyeong Yang, Taegwan Kang, Stanley Jungkyu Choi, Honglak Lee, and Kyomin Jung. Mitigating biases for instruction-following language models via bias neurons elimination. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long...

    2024

  39. [47]

    Razor: Sharpening knowledge by cutting bias with unsupervised text rewriting

    Shuo Yang, Bardh Prenkaj, and Gjergji Kasneci. Razor: Sharpening knowledge by cutting bias with unsupervised text rewriting. In Proceedings of the AAAI Conference on Artificial Intelligence, 2025

    2025

  40. [48]

    Paws: Paraphrase adversaries from word scrambling

    Yuan Zhang, Jason Baldridge, and Luheng He. Paws: Paraphrase adversaries from word scrambling. 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), p...

    2019

  41. [49]

    Lima: Less is more for alignment

    Chunting Zhou, Pengfei Liu, Puxin Xu, Srinivasan Iyer, Jiao Sun, Yuning Mao, Xuezhe Ma, Avia Efrat, Ping Yu, Lili Yu, et al. Lima: Less is more for alignment. Advances in Neural Information Processing Systems, 36, 2024. A Limitations Although our method can automatically debia...

    2024

  42. [50]

    • The abstract and/or introduction should clearly state the claims made, including the contributions made in the paper and important assumptions and limitations

    Claims Question: Do the main claims made in the abstract and introduction accurately reflect the paper’s contributions and scope? Answer: [Yes] Justification: The main claims can be supported by the experiments Guidelines: • The answer NA means that the abstract and introducti...

  43. [51]

    Limitations

    Limitations Question: Does the paper discuss the limitations of the work performed by the authors? Answer: [Yes] Justification: We provide the limitations of this work in Appendix A. Guidelines: • The answer NA means that the paper has no limitation while the answer No means t...

  44. [52]

    Guidelines: • The answer NA means that the paper does not include theoretical results

    Theory assumptions and proofs Question: For each theoretical result, does the paper provide the full set of assumptions and a complete (and correct) proof? Answer: [Yes] 16 Justification: We provide the full set of assumptions and a complete proof in Appendix B and Section 3.2...

  45. [53]

    We also submit the code and data in the supplemental material

    Experimental result reproducibility Question: Does the paper fully disclose all the information needed to reproduce the main ex- perimental results of the paper to the extent that it affects the main claims and/or conclusions of the paper (regardless of whether the code and da...

  46. [54]

    Guidelines: • The answer NA means that paper does not include experiments requiring code

    Open access to data and code 17 Question: Does the paper provide open access to the data and code, with sufficient instruc- tions to faithfully reproduce the main experimental results, as described in supplemental material? Answer: [Yes] Justification: We have provided the the...

  47. [55]

    Some other details such as the biased features and prompts for data rewriting are provided in appendix and supplemental material

    Experimental setting/details Question: Does the paper specify all the training and test details (e.g., data splits, hyper- parameters, how they were chosen, type of optimizer, etc.) necessary to understand the results? Answer: [Yes] Justification: We have provided the training...

  48. [56]

    Guidelines: • The answer NA means that the paper does not include experiments

    Experiment statistical significance Question: Does the paper report error bars suitably and correctly defined or other appropriate information about the statistical significance of the experiments? Answer: [Yes] Justification: We have reported the standard deviation across 3 r...

  49. [57]

    Guidelines: • The answer NA means that the paper does not include experiments

    Experiments compute resources Question: For each experiment, does the paper provide sufficient information on the com- puter resources (type of compute workers, memory, time of execution) needed to reproduce the experiments? Answer: [Yes] Justification: We have provided the co...

  50. [58]

    We also follow the licenses when using publicly available datasets

    Code of ethics Question: Does the research conducted in the paper conform, in every respect, with the NeurIPS Code of Ethics https://neurips.cc/public/EthicsGuidelines? Answer: [Yes] Justification: Our research does not involve human subjects or participants. We also follow th...

  51. [59]

    Guidelines: • The answer NA means that there is no societal impact of the work performed

    Broader impacts Question: Does the paper discuss both potential positive societal impacts and negative societal impacts of the work performed? Answer: [NA] Justification: We focus on the generalization abilities of LLMs, which is a foundational research and does not concentrat...

  52. [60]

    Guidelines: • The answer NA means that the paper poses no such risks

    Safeguards Question: Does the paper describe safeguards that have been put in place for responsible release of data or models that have a high risk for misuse (e.g., pretrained language models, image generators, or scraped datasets)? Answer: [NA] Justification: We use publicly...

  53. [61]

    We also follow the the license and terms of use of corresponding LLMs and data

    Licenses for existing assets Question: Are the creators or original owners of assets (e.g., code, data, models), used in the paper, properly credited and are the license and terms of use explicitly mentioned and properly respected? Answer: [Yes] Justification: We cite every re...

  54. [62]

    Guidelines: • The answer NA means that the paper does not release new assets

    New assets Question: Are new assets introduced in the paper well documented and is the documentation provided alongside the assets? Answer: [Yes] Justification: We provide the code to create the debiased dataset and debiased LLMs in the supplemental material. Guidelines: • The...

  55. [63]

    Guidelines: • The answer NA means that the paper does not involve crowdsourcing nor research with human subjects

    Crowdsourcing and research with human subjects Question: For crowdsourcing experiments and research with human subjects, does the paper include the full text of instructions given to participants and screenshots, if applicable, as well as details about compensation (if any)? A...

  56. [64]

    • Depending on the country in which research is conducted, IRB approval (or equivalent) may be required for any human subjects research

    Institutional review board (IRB) approvals or equivalent for research with human subjects Question: Does the paper describe potential risks incurred by study participants, whether such risks were disclosed to the subjects, and whether Institutional Review Board (IRB) approvals...

  57. [65]

    Answer: [NA] Justification: We use LLM for writing and editing

    Declaration of LLM usage Question: Does the paper describe the usage of LLMs if it is an important, original, or non-standard component of the core methods in this research? Note that if the LLM is used only for writing, editing, or formatting purposes and does not impact the ...

    2025

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.