Pith. sign in

REVIEW 5 major objections 6 minor 39 references

SALAD: Improving Robustness and Generalization through Contrastive Learning with Structure-Aware and LLM-Driven Augmented Data

T0 review · 5 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 SALAD combines POS-tag-masked positives and LLM-generated counterfactual negatives in a contrastive loss to make fine-tuned language models more robust to spurious correlations.

desk verdict Unverified label preservation in POS-masked positives is the load-bearing flaw; the combination is new and the gains are consistent, but the numbers are provisional until that assumption is checked. read the letter →

arxiv 2504.12185 v1 pith:CZCUEP7L submitted 2025-04-16 cs.CL cs.AI

Suyoung Bae , Hyojun Kim , YunSeok Choi , Jee-Hyong Lee This is my paper · ORCID
classification cs.CLcs.AI
keywords datalanguagelearningsaladapproachaugmentedcontrastivegeneralization
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

When a large language model is fine-tuned for a task such as sentiment analysis, it can latch onto accidental patterns in the training data, for instance a director's name that happens to appear in positive reviews. This hurts performance when the model meets new data where those patterns do not hold. SALAD is a training procedure designed to counter this. First, it tags each word with its grammatical role, such as noun, verb, or adjective. Using a normally trained model, it measures how much accuracy drops when all words of a given role are removed, and declares roles with almost no effect to be non-causal. It then replaces a few of those non-causal words with a special [UNK] token, creating a positive example that keeps the sentence's meaning. Second, it asks an LLM, GPT-4o-mini, to rewrite the sentence just enough to flip its label, creating a negative example. Third, it trains the model with two losses: the usual classification loss and a triplet loss that pulls the positive example close to the original sentence in the model's internal representation while pushing the negative example away. The hope is that the model learns which sentence components actually drive the label and stops depending on coincidental wording. The paper tests SALAD on sentiment classification, sexism detection, and natural language inference, and reports higher accuracy on out-of-distribution and cross-domain test sets than standard fine-tuning and several augmentation baselines.
Extended reading notes

Core claim

Section 5.1 states that 'SALAD demonstrates improved overall accuracy compared to standard fine-tuning baseline across all tasks' and reports overall accuracy of 93.05 on sentiment classification, 88.31 on sexism detection, 79.57 on NLI, and 91.61 on cross-domain sentiment. If these results are correct, SALAD is a new state-of-the-art training method for these benchmarks, improving robustness to spurious correlations and OOD generalization.

Load-bearing premise

The method's positive samples are built from a non-causal POS tag set G, which is determined by the average accuracy reduction R_Tj of a standard fine-tuned model (Eq. 1, Section 3.1). The central premise is that this reduction reliably identifies which grammatical roles are causally irrelevant to the label, and that the chosen threshold (stated as 0.1 in Section 4.3 but 1% in Appendix B.4) separates causal from non-causal tags across all three tasks. If this premise fails, the positive samples will mask informative words, and the contrastive objective will push the model toward distorted representations.

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

5 major / 6 minor

Summary. The paper proposes SALAD, a fine-tuning method that augments the standard cross-entropy loss with a triplet contrastive loss. Positive samples are generated by randomly masking a small number of tokens belonging to a POS-tag set that is deemed non-causal, where the set is derived from the average accuracy reduction of a standard fine-tuned model when each POS tag is removed (Eq. 1, Section 3.1). Negative samples are counterfactual sentences generated by an LLM that are prompted to flip the label by changing only causal words. The method is evaluated on sentiment classification, sexism detection, and natural language inference, with additional cross-domain and prompt-based few-shot experiments. The main claimed results are improved in-domain robustness, particularly on counterfactual test sets, and improved out-of-distribution and cross-domain accuracy relative to several data-augmentation and robust-learning baselines.

Significance. The empirical program is broad: three tasks, in-domain and out-of-distribution evaluation, cross-domain transfer, prompt-based few-shot evaluation, component ablations, and prompt-instruction analysis. If the results hold, SALAD would be a useful contribution to mitigating spurious correlations in fine-tuned PLMs. The paper also provides several useful analyses, including a comparison of LLM prompt templates for counterfactual generation and a diversity/overlap/BERTScore characterization of the generated data, which strengthen the empirical narrative. However, the central mechanism currently rests on an unverified assumption about label preservation under POS-based masking, and several reporting practices, including an inconsistent threshold, potential test-set tuning of hyperparameters, missing variance information, and incomplete cross-domain baselines, weaken the evidence. These issues are addressable within the scope of the manuscript.

major comments (5)
  1. [§4.3 vs. §B.4, Fig. 3] The threshold used to define the non-causal tag set is inconsistent: Section 4.3 states that the threshold is 0.1 across all datasets, while Appendix B.4 and Figure 3 state that θ is estimated to be 1% and that non-causal tags are those with an average accuracy reduction of less than 1%. These thresholds differ by an order of magnitude and can change the tag set G materially. Please state the correct threshold, report the measured accuracy-reduction values for each POS tag, and reconstruct the tag sets with a justified, consistent threshold.
  2. [§4.3, §B.3, Table 9] The hyperparameters k and λ are selected per dataset, and Section B.3 states that k=8 was chosen because it 'showed significant performance improvement for the CF-IMDB dataset, particularly on the out-of-distribution dataset (ODD)'. If k, λ, and the threshold are selected using the same ODD test sets whose accuracies are then reported in Tables 1-4, the reported gains are partly selection artifacts. The manuscript does not specify a validation protocol that keeps the reported test sets out of hyperparameter selection. Please either describe a protocol that uses only held-out validation data or report results with hyperparameters fixed a priori.
  3. [Tables 1-4, 13; §4.3] All results are reported as averages over three random seeds without standard deviations or significance tests. Several claimed improvements are small relative to the strongest baselines, for example 93.05 vs. 91.49 overall on sentiment in Table 1, 91.61 vs. 89.92 on cross-domain in Table 4, and 88.31 vs. 89.47 on sexism in Table 2. Without variance information or significance testing, the reader cannot tell whether SALAD is actually better than the comparison methods. Please report per-seed results or standard deviations and, where possible, significance tests.
  4. [Table 4] The cross-domain experiment omits several baselines that appear in the main single-domain tables, including AugGPT, Human-CAD, CORE-CAD, and DISCO. The statement that 'except in IMDB→SST2, all accuracy achieves the best performance' is therefore only relative to a subset of the considered baselines. Please add the missing baselines to Table 4, or explicitly restrict the claim to the compared methods.
  5. [§5.1, Table 2] The discussion of the sexism task should be more careful: in Table 2, SALAD's overall accuracy is 88.31, which is below Human-CAD's 89.47, and on the counterfactual test set SALAD is 88.47 vs. Human-CAD's 91.80. The paper acknowledges this in passing, but the abstract's general claim that SALAD improves robustness and generalization should be qualified by this comparative result, especially since Human-CAD is not included in the cross-domain experiments.
minor comments (6)
  1. [§3.3, Eq. (3)] The symbol α is used both as the margin in the triplet loss in Eq. (3) and as the scaling factor for k in Sections 3.1 and 4.3; please rename one of them to avoid ambiguity.
  2. [§3.1] There is a typo in 'non-casual words' in the paragraph following Eq. (1); the intended word is 'non-causal'.
  3. [Table 1] The column header 'FindFood' should read 'FineFood' for consistency with the rest of the paper.
  4. [Table 10 vs. §3.1] Table 10 is headed 'Causal tag set G', but Section 3.1 defines G as the non-causal tag set; please align the notation in the table with the text.
  5. [Fig. 3] The axis labels in Figure 3 render with garbled characters, making the accuracy-reduction values unreadable; the figure should be regenerated with a standard font.
  6. [Appendix C] The sentence 'Table 4 reports the performance of applying SALAD...' appears to refer to Figure 4 in the appendix rather than Table 4 in the main text; please correct the cross-reference.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; SALAD's contrastive data construction is an empirical heuristic and its benchmark claims rest on held-out evaluations, not on constructional equivalences.

full rationale

The central mechanism of SALAD is the non-causal POS tag set G computed via Eq. (1) from a standard fine-tuned model. This is an operational feature-selection step, not a derivation of the reported accuracies. The paper does not claim that Eq. (1) logically entails the test-set results; the OOD, cross-domain, and counterfactual test sets are external to the fitting procedure. The threshold discrepancy (0.1 in Section 4.3 vs. 1% in Appendix B.4) and the task-dependent choices of k and lambda are hyperparameter and design decisions; these would be correctness or reproducibility concerns, not circularity. The triplet loss is explicitly adapted from C2L, but C2L is an external prior work and no load-bearing self-citation chain is present. No uniqueness theorem is imported from the authors' own prior work, and no known empirical pattern is merely renamed. Every step of the pipeline is an empirical heuristic whose success is established by held-out benchmarks; therefore the derivation chain does not reduce to its own inputs by construction.

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

The method introduces no new theoretical entities, but its empirical claims rest on several fitted hyperparameters and domain assumptions about POS tags, LLM reliability, and contrastive learning.

free parameters (5)
  • alpha (scaling factor for k) = 0.18
    Derived by setting k=8 for CF-IMDB, which was chosen after observing significant OOD improvement; applied to all datasets via average non-causal token count (Appendix B.3).
  • k (number of non-causal tokens masked) = 8 (CF-IMDB, IMDB), 1 (CF-NLI, Sexism, SST2), 5 (FineFood)
    Selected per dataset based on experimental performance; Table 9.
  • lambda (contrastive loss weight) = 0.9 (CF-IMDB, CF-NLI, IMDB), 0.3 (Sexism), 0.1 (SST2, FineFood)
    Tuned separately for each dataset to achieve the best performance (Section 4.3, Table 9).
  • theta (threshold for causal/non-causal POS) = 0.1 in main text, 1% in Appendix B.4
    Determines the non-causal POS tag set; value chosen from ablation and stated inconsistently.
  • Prompt instruction template = Instruction 4
    Selected as the most effective among four variants via experiments in Appendix C; includes causal word information.
assumptions (4)
  • domain assumption POS tags whose removal causes low average accuracy reduction are non-causal and safe to mask.
    Section 3.1, Eq. (1). If incorrect, positive samples will mask label-relevant words and hurt learning.
  • domain assumption A standard fine-tuned model f provides a reliable estimate of which POS tags are causally related to the label.
    Section 3.1. The tag sets G are derived from f's accuracy reduction, so f must be a trustworthy estimator of causal structure.
  • domain assumption LLM-generated sentences with flipped labels are valid negative examples that preserve the original sentence structure.
    Section 3.2 and Appendix C. No human verification is performed on all generated data; the method depends on GPT-4o-mini following the prompt.
  • domain assumption Triplet contrastive learning with anchor, masked positive, and counterfactual negative improves robustness and generalization.
    Section 3.3. This is the standard assumption of contrastive representation learning applied here.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SALAD: Improving Robustness and Generalization through Contrastive Learning with Structure-Aware and LLM-Driven Augmented Data." pith.science (2026). https://pith.science/paper/CZCUEP7L

@misc{pith2026250412185,
  author       = {Pith},
  title        = {Pith review of: SALAD: Improving Robustness and Generalization through Contrastive Learning with Structure-Aware and LLM-Driven Augmented Data},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CZCUEP7L}},
  note         = {Machine review of arXiv:2504.12185}
}
read the original abstract

In various natural language processing (NLP) tasks, fine-tuning Pre-trained Language Models (PLMs) often leads to the issue of spurious correlations, which negatively impacts performance, particularly when dealing with out-of-distribution data. To address this problem, we propose SALAD}(Structure Aware and LLM-driven Augmented Data), a novel approach designed to enhance model robustness and generalization by generating structure-aware and counterfactually augmented data for contrastive learning. Our method leverages a tagging-based approach to generate structure-aware positive samples and utilizes large language models (LLMs) to generate counterfactual negative samples with diverse sentence patterns. By applying contrastive learning, SALAD enables the model to focus on learning the structural relationships between key sentence components while minimizing reliance on spurious correlations. We validate our approach through experiments on three tasks: Sentiment Classification, Sexism Detection, and Natural Language Inference. The results demonstrate that SALAD not only improves model robustness and performance across different environments but also enhances generalization to out-of-distribution datasets and cross-domain scenarios.

Figures

Figures reproduced from arXiv: 2504.12185 by the authors.

Figure 1
Figure 1. Overview of SALAD. Our proposed method consists of three steps. First, we use a tagging-based method to generate positive data based on the structure where shortcuts occur (Sec. 3.1). Next, we use an LLM to generate counterfactual data to capture complex and diverse sentence patterns (Sec. 3.2). Finally, contrastive learning is applied to effectively capture key sentence structural patterns between our augmented dat… view at source ↗
Figure 2
Figure 2. Experiments on defining k: The value of 8 shows significant performance improvement for the CF￾IMDB dataset, particularly on the out-of-distribution dataset (ODD). Dataset k λ CF-IMDB (Kaushik et al., 2020) 8 0.9 CF-NLI (Kaushik et al., 2020) 1 0.9 Sexism (Samory et al., 2021) 1 0.3 SST2 (Socher et al., 2013) 1 0.1 IMDB (Maas et al., 2011) 8 0.9 FineFood (McAuley and Leskovec, 2013) 5 0.1 [PITH_FULL_IMAGE:figures/f… view at source ↗
Figure 3
Figure 3. is an ablation study on the results of calcu￾lating accuracy reductions in sentiment and sexism classification datasets. We estimate θ to be 1%, defining the non-causal tag set as the POS infor￾mation for which the score is less than 1%. The Causal tag sets calculated for each dataset used in our experiment are listed in [PITH_FULL_IMAGE:figures/full_fig_p013_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Performance variations of SALAD on datasets generated for each instruction. The number following “SALAD" corresponds to the instructions associated with each number used in [PITH_FULL_IMAGE:figures/full_fig_p014_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

39 extracted references · 8 canonical work pages

  1. [1]

    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...

  2. [2]

    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...

  3. [3]

    Nabiha Asghar. 2016. https://arxiv.org/abs/1605.05362 Yelp dataset challenge: Review rating prediction . CoRR, abs/1605.05362

    arXiv 2016

  4. [4]

    Steven Bird. 2006. https://doi.org/10.3115/1225403.1225421 NLTK : The N atural L anguage T oolkit . In COLING:2006:4, pages 69--72, Sydney, Australia. Association for Computational Linguistics

    arXiv 2006

  5. [5]

    Tom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert - Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel M. Ziegler, Jeffrey Wu, Clemens Winter, Christopher Hesse, Mark Chen, Eric Sigler, Mateusz Litw...

    2020

  6. [6]

    Zeming Chen, Qiyue Gao, Antoine Bosselut, Ashish Sabharwal, and Kyle Richardson. 2023. https://doi.org/10.18653/v1/2023.acl-long.302 DISCO : Distilling counterfactuals with large language models . In ACL:2023:long, pages 5514--5528, Toronto, Canada. Association for Computational Linguistics

    doi:10.18653/v1/2023.acl-long.302 2023

  7. [7]

    Seungtaek Choi, Myeongho Jeong, Hojae Han, and Seung-won Hwang. 2022. https://doi.org/10.1609/aaai.v36i10.21296 C2l: Causally contrastive learning for robust text classification . Proceedings of the AAAI Conference on Artificial Intelligence, 36(10):10526--10534

    work page doi:10.1609/aaai.v36i10.21296 2022

  8. [8]

    Haixing Dai, Zhengliang Liu, Wenxiong Liao, Xiaoke Huang, Yihan Cao, Zihao Wu, Lin Zhao, Shaochen Xu, Wei Liu, Ninghao Liu, Sheng Li, Dajiang Zhu, Hongmin Cai, Lichao Sun, Quanzheng Li, Dinggang Shen, Tianming Liu, and Xiang Li. 2023. https://arxiv.org/abs/2302.13007 Auggpt: Leveraging chatgpt for text data augmentation . Preprint, arXiv:2302.13007

    arXiv 2023

Show all 39 references
  1. [9]

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. https://doi.org/10.18653/v1/N19-1423 BERT : Pre-training of deep bidirectional transformers for language understanding . In NAACL:2019:1, pages 4171--4186, Minneapolis, Minnesota. Association for Computati...

    2019 doi

  2. [10]

    Ning Ding, Shengding Hu, Weilin Zhao, Yulin Chen, Zhiyuan Liu, Hai-Tao Zheng, and Maosong Sun. 2021. Openprompt: An open-source framework for prompt-learning. arXiv preprint arXiv:2111.01998

    2021 arXiv

  3. [11]

    Tanay Dixit, Bhargavi Paranjape, Hannaneh Hajishirzi, and Luke Zettlemoyer. 2022. https://doi.org/10.18653/v1/2022.findings-emnlp.216 CORE : A retrieve-then-edit framework for counterfactual data generation . In FINDINGS:2022:emnlp, pages 2964--2984, Abu Dhabi, United Arab Emi...

    2022 doi

  4. [12]

    Tianyu Gao, Adam Fisch, and Danqi Chen. 2021. https://doi.org/10.18653/v1/2021.acl-long.295 Making pre-trained language models better few-shot learners . In ACL:2021:long, pages 3816--3830, Online. Association for Computational Linguistics

    2021 doi

  5. [13]

    Beliz Gunel, Jingfei Du, Alexis Conneau, and Veselin Stoyanov. 2021. https://openreview.net/forum?id=cu7IUiOhujH Supervised contrastive learning for pre-trained language model fine-tuning . In International Conference on Learning Representations

    2021

  6. [14]

    Divyansh Kaushik, Eduard Hovy, and Zachary Lipton. 2020. https://openreview.net/forum?id=Sklgs0NFvr Learning the difference that makes a difference with counterfactually-augmented data . In International Conference on Learning Representations

    2020

  7. [15]

    Prannay Khosla, Piotr Teterwak, Chen Wang, Aaron Sarna, Yonglong Tian, Phillip Isola, Aaron Maschinot, Ce Liu, and Dilip Krishnan. 2020. Supervised contrastive learning. Advances in neural information processing systems, 33:18661--18673

    2020

  8. [16]

    Smith, and Yejin Choi

    Alisa Liu, Swabha Swayamdipta, Noah A. Smith, and Yejin Choi. 2022. https://doi.org/10.18653/v1/2022.findings-emnlp.508 WANLI : Worker and AI collaboration for natural language inference dataset creation . In FINDINGS:2022:emnlp, pages 6826--6847, Abu Dhabi, United Arab Emirat...

    2022 doi

  9. [17]

    Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, and Veselin Stoyanov. 2019. https://arxiv.org/abs/1907.11692 Roberta: A robustly optimized bert pretraining approach

    2019 arXiv

  10. [18]

    Maas, Raymond E

    Andrew L. Maas, Raymond E. Daly, Peter T. Pham, Dan Huang, Andrew Y. Ng, and Christopher Potts. 2011. https://aclanthology.org/P11-1015 Learning word vectors for sentiment analysis . In ACL:2011:1, pages 142--150, Portland, Oregon, USA. Association for Computational Linguistics

    2011

  11. [19]

    Nishtha Madaan, Inkit Padhi, Naveen Panwar, and Diptikalyan Saha. 2021. https://doi.org/10.1609/aaai.v35i15.17594 Generate your counterfactuals: Towards controlled counterfactual generation for text . Proceedings of the AAAI Conference on Artificial Intelligence, 35(15):13516--13524

    2021 doi

  12. [20]

    Julian John McAuley and Jure Leskovec. 2013. From amateurs to connoisseurs: modeling the evolution of user expertise through online reviews. In Proceedings of the 22nd international conference on World Wide Web, pages 897--908

    2013

  13. [21]

    Nathan Ng, Kyunghyun Cho, and Marzyeh Ghassemi. 2020. https://doi.org/10.18653/v1/2020.emnlp-main.97 SSMBA : Self-supervised manifold based data augmentation for improving out-of-domain robustness . In EMNLP:2020:main, pages 1268--1283, Online. Association for Computational Li...

    2020 doi

  14. [22]

    Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, et al. 2019. Pytorch: An imperative style, high-performance deep learning library. Advances in neural information processing systems, 32

    2019

  15. [23]

    Slav Petrov, Dipanjan Das, and Ryan McDonald. 2012. http://www.lrec-conf.org/proceedings/lrec2012/pdf/274_Paper.pdf A universal part-of-speech tagset . In LREC:2012:1, pages 2089--2096, Istanbul, Turkey. European Language Resources Association (ELRA)

    2012

  16. [24]

    o ck, and Claudia Wagner. 2021. “call me sexist, but

    Mattia Samory, Indira Sen, Julian Kohne, Fabian Fl \"o ck, and Claudia Wagner. 2021. “call me sexist, but...”: Revisiting sexism detection using psychological scales and adversarial samples. In Proceedings of the international AAAI conference on web and social media, volume 15...

    2021

  17. [25]

    Timo Schick and Hinrich Sch \"u tze. 2021. https://doi.org/10.18653/v1/2021.eacl-main.20 Exploiting cloze-questions for few-shot text classification and natural language inference . In EACL:2021:main, pages 255--269, Online. Association for Computational Linguistics

    2021 doi

  18. [26]

    Logan IV, Eric Wallace, and Sameer Singh

    Taylor Shin, Yasaman Razeghi, Robert L. Logan IV, Eric Wallace, and Sameer Singh. 2020. https://doi.org/10.18653/v1/2020.emnlp-main.346 A uto P rompt: E liciting K nowledge from L anguage M odels with A utomatically G enerated P rompts . In EMNLP:2020:main, pages 4222--4235, O...

    2020 doi

  19. [27]

    Manning, Andrew Ng, and Christopher Potts

    Richard Socher, Alex Perelygin, Jean Wu, Jason Chuang, Christopher D. Manning, Andrew Ng, and Christopher Potts. 2013. https://aclanthology.org/D13-1170 Recursive deep models for semantic compositionality over a sentiment treebank . In EMNLP:2013:1, pages 1631--1642, Seattle, ...

    2013

  20. [28]

    Lifu Tu, Garima Lalwani, Spandana Gella, and He He. 2020. https://doi.org/10.1162/tacl_a_00335 An empirical study on robustness to spurious correlations using pre-trained language models . Transactions of the Association for Computational Linguistics, 8:621--633

    2020 doi

  21. [29]

    Tianlu Wang, Rohit Sridhar, Diyi Yang, and Xuezhi Wang. 2022. https://doi.org/10.18653/v1/2022.findings-naacl.130 Identifying and mitigating spurious correlations for improving robustness in NLP models . In FINDINGS:2022:naacl, pages 1719--1729, Seattle, United States. Associa...

    2022 doi

  22. [30]

    Zhao Wang and Aron Culotta. 2020. https://doi.org/10.18653/v1/2020.findings-emnlp.308 Identifying spurious correlations for robust text classification . In FINDINGS:2020:emnlp, pages 3431--3440, Online. Association for Computational Linguistics

    2020 doi

  23. [31]

    Zhao Wang and Aron Culotta. 2021. https://doi.org/10.1609/aaai.v35i16.17651 Robustness to spurious correlations in text classification via automatically generated counterfactuals . Proceedings of the AAAI Conference on Artificial Intelligence, 35(16):14024--14031

    2021 doi

  24. [32]

    Jason Wei and Kai Zou. 2019. https://doi.org/10.18653/v1/D19-1670 EDA : Easy data augmentation techniques for boosting performance on text classification tasks . In EMNLP:2019:1, pages 6382--6388, Hong Kong, China. Association for Computational Linguistics

    2019 doi

  25. [33]

    Jiaxin Wen, Yeshuang Zhu, Jinchao Zhang, Jie Zhou, and Minlie Huang. 2022. https://doi.org/10.18653/v1/2022.findings-emnlp.170 A uto CAD : Automatically generate counterfactuals for mitigating shortcut learning . In FINDINGS:2022:emnlp, pages 2302--2317, Abu Dhabi, United Arab...

    2022 doi

  26. [34]

    Adina Williams, Nikita Nangia, and Samuel Bowman. 2018. https://doi.org/10.18653/v1/N18-1101 A broad-coverage challenge corpus for sentence understanding through inference . In Proceedings of the 2018 Conference of the North A merican Chapter of the Association for Computation...

    2018 doi

  27. [35]

    Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pierric Cistac, Tim Rault, Remi Louf, Morgan Funtowicz, Joe Davison, Sam Shleifer, Patrick von Platen, Clara Ma, Yacine Jernite, Julien Plu, Canwen Xu, Teven Le Scao, Sylvain Gugger, Mari...

    2020 doi

  28. [36]

    Tongshuang Wu, Marco Tulio Ribeiro, Jeffrey Heer, and Daniel Weld. 2021. https://doi.org/10.18653/v1/2021.acl-long.523 Polyjuice: Generating counterfactuals for explaining, evaluating, and improving models . In ACL:2021:long, pages 6707--6723, Online. Association for Computati...

    2021 doi

  29. [37]

    Linyi Yang, Jiazheng Li, Padraig Cunningham, Yue Zhang, Barry Smyth, and Ruihai Dong. 2021. https://doi.org/10.18653/v1/2021.acl-long.26 Exploring the efficacy of automatically generated counterfactuals for sentiment analysis . In ACL:2021:long, pages 306--316, Online. Associa...

    2021 doi

  30. [38]

    Weinberger, and Yoav Artzi

    Tianyi Zhang*, Varsha Kishore*, Felix Wu*, Kilian Q. Weinberger, and Yoav Artzi. 2020. https://openreview.net/forum?id=SkeHuCVFDr Bertscore: Evaluating text generation with bert . In International Conference on Learning Representations

    2020

  31. [39]

    Jing Zhou, Yanan Zheng, Jie Tang, Li Jian, and Zhilin Yang. 2022. https://doi.org/10.18653/v1/2022.acl-long.592 F lip DA : Effective and robust data augmentation for few-shot learning . In ACL:2022:long, pages 8646--8665, Dublin, Ireland. Association for Computational Linguistics

    2022 doi

Pith tools

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