Pith. sign in

REVIEW 5 major objections 4 minor 32 references

COLA-GEC: A Bidirectional Framework for Enhancing Grammatical Acceptability and Error Correction

T0 review · 5 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read Grammatical acceptability and error correction improve each other in a single bidirectional framework.

desk verdict Useful bidirectional GEC/COLA idea, but impossible F0.5 values in Table 6 and shared critic/GEC training data make the headline claims unverified. read the letter →

arxiv 2507.11867 v1 pith:KKSG5ESS submitted 2025-07-16 cs.CL

Xiangyu Yang , Xinying Qiu This is my paper · ORCID
classification cs.CL
keywords grammaticalerrorcorrectionacceptabilityjudgmentbidirectionalknowledgetransferdynamiclosslabelconfidenceweightedlearningmultilingualevaluationpunctuationanalysisdual-criticdecoding
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 tries to show that grammatical acceptability judgment (COLA) and grammatical error correction (GEC) are mutually reinforcing: GEC error-correction pairs can be converted into acceptability training data that improves COLA models, and COLA acceptability signals can be folded into GEC training to push corrections toward acceptable output. It builds a bidirectional framework in which GEC-augmented critic models (G-Cola) are trained with the two tasks' data, and the critic's judgments enter the GEC loss through a dynamic weighting term. If the claim holds, the two tasks stop being developed in isolation and can share grammatical knowledge across languages, which matters because both are core to language-learning technology. The reported results include state-of-the-art F0.5 scores on MuCGEC and CoNLL-14 and consistent COLA gains across English, Chinese, German, and Arabic.

What carries the argument

The central mechanism is a dynamic loss function built from Label Confidence Weighted Learning: the standard GEC cross-entropy loss at each timestep is multiplied by the square root of the COLA critic's accuracy times a sigmoid-transformed acceptability score, so corrections the critic deems unacceptable incur larger penalties. Carrying the argument on the other side is the data augmentation step that turns GEC error-correction pairs into binary acceptability judgments (corrected sentences labeled acceptable, error sentences unacceptable), which is what lets a single critic share grammatical knowledge between the tasks. The GEC model is then plugged into a dual-critic decoding framework, where the acceptability signal is combined with existing critics during inference.

What would settle it

Retrain the G-Cola critic using only GEC error-correction pairs that were withheld from the GEC model's training corpora, keeping everything else fixed, and rerun the MuCGEC, FCGEC, and CoNLL-14 evaluations; if the F0.5 advantage over the dual-critic baseline shrinks or disappears, the central claim of shared-knowledge transfer is not supported.

Watch

Extended reading notes

Core claim

The paper claims that a bidirectional knowledge transfer between grammatical acceptability and grammatical error correction improves both tasks. On the acceptability side, converting GEC error-correction pairs into binary acceptability labels and adding them to existing COLA corpora improves accuracy and Matthews correlation across multiple languages, including a multilingual model that surpasses the previous state-of-the-art on the MELA benchmark. On the correction side, a COLA critic scores each candidate correction, and the score is multiplied into a label-confidence-weighted cross-entropy loss so that grammatically questionable corrections are penalized more heavily; combining this G-Cola GEC model with a dual-critic decoding framework yields new best F0.5 results on MuCGEC (49.91) and CoNLL-14 (70.84). The paper also reports an error analysis indicating that punctuation errors are a persistent bottleneck.

Load-bearing premise

The load-bearing premise is that the COLA critic supplies an independent grammaticality signal during GEC training even though it was trained on the same GEC corpora as the GEC model, so the reported transfer gains could be an artifact of data overlap rather than shared grammatical knowledge.

Editorial extensions

If this is right

  • GEC corpora can serve as cheap acceptability-judgment training data, improving COLA accuracy and MCC for English, Chinese, German, and Arabic without new linguistic annotation.
  • Acceptability-guided training raises GEC precision and F0.5 on CoNLL-14 and MuCGEC enough to set new state-of-the-art scores.
  • The benefit transfers to a low-resource language: the German G-Cola GECdi model far surpasses the text-only German rule/baseline system.
  • Ablations show that the gains are additive: adding GEC-enhanced G-Cola to both a base BART GEC model and a dual-critic GEC model improves F0.5 consistently across Chinese and English test sets.
  • Punctuation errors are a principal remaining weakness; removing them from BEA-19 dev evaluation raises the model's F0.5 from 72.25 to 74.40.

Reading between the lines

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

  • A direct consequence the paper leaves implicit: if acceptability signals transfer across languages, a critic trained on one language's GEC data might bootstrap GEC training for another language, lowering data requirements for low-resource GEC.
  • The dynamic loss depends on the critic's accuracy term, so one could test a curriculum that iteratively retrains the critic on the GEC model's own corrected outputs, potentially compounding gains.
  • Because the critic and GEC model share training corpora, a clean test of the mechanism would hold out entire GEC datasets from the critic's training and check whether the MuCGEC and CoNLL-14 gains survive; the paper does not perform this separation.
  • The punctuation bottleneck suggests a targeted synthetic augmentation: generating punctuation-specific error-correction pairs (or training a dedicated punctuation head) might recover the missing F0.5 more directly than generic acceptability fine-tuning.
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 / 4 minor

Summary. The paper proposes COLA-GEC, a bidirectional framework linking grammatical acceptability judgment (COLA) and grammatical error correction (GEC). It augments COLA training data with GEC error-correction pairs in multiple languages and introduces a dynamic loss that uses a COLA model's judgment of the GEC model's output to reweight the GEC training loss. The authors report state-of-the-art results on MuCGEC and CoNLL-14, consistent improvements in ablations, and an error analysis focusing on punctuation errors. The paper also provides code and detailed corpus statistics.

Significance. The cross-task transfer idea is timely and potentially valuable for both COLA and GEC, especially for low-resource languages. The authors provide reproducibility assets (a code link and corpus statistics) and a concrete ablation design. However, the main empirical claims are currently undermined by internal data inconsistencies and by a training-data overlap that clouds the interpretation of the COLA signal. If these issues are resolved, the framework could be a useful contribution.

major comments (5)
  1. [Section 5.4, Table 6] The NaSGEC-Exam row in Table 6 reports F0.5 = 67.33 for GEC+G-CoLA with P=53.67 and R=25.13, and F0.5 = 79.32 for GECdi+G-CoLA with P=63.92 and R=32.10. Using the standard F0.5 formula (1.25*P*R)/(0.25*P+R), these values are approximately 43.7 and 52.3, respectively. The reported numbers appear to be copied from the CoNLL-14 row, corrupting the ablation claim of 'substantial gains' on NaSGEC-Exam.
  2. [Section 5.4, Table 6] Other F0.5 values in Table 6 are also arithmetically inconsistent with the reported P/R. For example, the CoNLL-14 base GEC row (P=63.02, R=41.26) yields F0.5 ≈ 57.0, not 54.08; the BEA-19 base row (P=64.41, R=51.11) yields ≈ 61.2, not 61.12; and the MuCGEC GECdi row (P=56.83, R=33.71) yields ≈ 50.0, not 49.77. The authors should either report exact computations or correct the numbers.
  3. [Section 5.1 vs. Section 5.4] The baseline GECdi F0.5 for MuCGEC is 48.61 in Table 3 but 49.77 in Table 6; for CoNLL-14 it is 69.6 and 70.11, respectively. The claimed improvement from adding G-CoLA therefore changes by up to 1.16 F0.5 points depending on the table. The MuCGEC 'new best' margin shrinks from 1.30 to 0.14 points when using Table 6's baseline. The authors must clarify which table is correct and re-evaluate the SOTA claims.
  4. [Section 3.1, Tables 8-10] The COLA critic (G-Cola) is trained on the same GEC corpora used to train and fine-tune the GEC model (e.g., MuCGEC, FCGEC, NaSGEC for Chinese; W&I+LOCNESS, FCE, NUCLE for English; Falko/MERLIN for German). Because the critic has memorized the gold corrections for these training examples, its 'grammaticality' judgments may simply reflect that memorization rather than an independent grammatical signal. To validate the claimed mechanism, the authors should provide an ablation where the COLA model is trained on data disjoint from the GEC model's training data (e.g., a held-out GEC set) and show that the gains persist.
  5. [Abstract and Section 5.2] The abstract and Section 5.2 state that the approach achieves state-of-the-art results on several multilingual benchmarks. In fact, the reported results are state-of-the-art only for MuCGEC and CoNLL-14; on BEA-19 the F0.5 of 74.03 is below Unsupervised GEC (76.5), on FCGEC the F0.5 of 58.6 is below EPOGEC (59.63), and on German the F0.5 of 70.80 is below MultimodelGEC (76.30). The claims should be scoped to the datasets where the method actually sets a new state of the art.
minor comments (4)
  1. [Section 3.2, Eq. (1)] The COLA_score definition is confusing: when Logits_0 > Logits_1 (the model predicts 'unacceptable'), the score approaches 1, meaning a high score indicates unacceptability. This contradicts the name 'COLA_score' and the text in Section 3.2 that describes the model as producing acceptability scores. Consider renaming it to a 'penalty score' or clarifying the sign convention.
  2. [Table 1] Some rows in Table 1 have dev/test sizes exceeding the training size (e.g., French: train 500, dev 521, test 521; Icelandic: train 500, dev 1,194, test 1,194). These are likely typos and should be corrected.
  3. [Section 4, German Implementation] There is a typo in 'HuggingFace'sBart-GermanWepretrainedonFalko' — missing spaces around 'Bart-German' and 'We'.
  4. [Section 5.4] The paper does not report variance or significance tests for the improvements, which is important because several margins are small (e.g., 0.14 F0.5 on MuCGEC when using the Table 6 baseline).

Circularity Check

1 steps flagged · score 4.0 of 10

The COLA critic used in the dynamic loss is trained on the same GEC corpora as the GEC model, so its grammaticality signal is partly fitted to the gold corrections; the bidirectional claim is partially confounded, though external test sets keep the result from being forced.

  1. fitted input called prediction [Section 3.2 (Framework Overview and Eq. 2), Section 4 (G-Cola GEC Training), Tables 7-10.]
    "We add GEC data to enhance pretrained COLA model using corpora discussed in Section 3.1. ... The Chinese model used Lang8 (590K sentences, 89.5% error rate) and HSK (157K sentences, 60.8% error rate) for pre-training, followed by fine-tuning on FCGEC, MuCGEC, and NaSGEC datasets. ... The English model pretrained on Clang8 (2.37M sentences, 57.8% error rate) and fine-tuned on W&I+LOCNESS, FCE, and NUCLE datasets."

    The GEC model's cross-entropy loss (Eq. 2) is reweighted by COLA_score produced by a critic that was itself fine-tuned on the same GEC corpora (FCGEC/MuCGEC/NaSGEC for Chinese; W&I+LOCNESS/FCE/NUCLE for English; Falko/MERLIN for German) whose gold corrections appear as Y_t in that same loss. By construction, the critic labels the gold-corrected training sentences as acceptable (1) and the error sentences as unacceptable (0), so its acceptability judgments are fitted to the exact target corrections the GEC model is trained to emit. The dynamic loss therefore encodes the same gold-correction information already present in the cross-entropy term; the claimed 'grammatical acceptability signal' is not an independent source of grammatical knowledge.

full rationale

The GEC-to-COLA direction is externally grounded: GEC-augmented COLA models are evaluated on held-out CoLA test sets with no overlap with the augmentation data, so that half of the bidirectional claim is not circular. The COLA-to-GEC direction is partially confounded: the critic that supplies COLA_score in Eq. (2) is trained on the same GEC corpora used to train the GEC model, making its 'acceptability' judgment a learned function of the same gold corrections rather than an independent grammaticality signal. This is a data-overlap/fitted-input issue, not a formal equivalence, and the test-set numbers could in principle have gone either way; I therefore score it 4 rather than 6+. The self-citations (LCWL [17], MELA [30]) are not load-bearing in the sense that the paper's central results are numerically evaluated rather than assumed from those works; they are normal antecedents. Separately, the manuscript contains arithmetic inconsistencies in Table 6 (e.g., NaSGEC-Exam F0.5 = 67.33 with P = 53.67 and R = 25.13 does not match the F0.5 formula, and the values appear copied from the CoNLL-14 row; MuCGEC and CoNLL-14 GECdi baselines also differ between Tables 3 and 6). That is a correctness/verifiability risk, not circularity, and does not raise the circularity score, but it should be flagged for the authors and reviewers.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

The central empirical claims rest on domain assumptions about the transferability of binary acceptability labels derived from GEC pairs and on the validity of the COLA-score weighting in the loss. The dynamic loss introduces a hand-chosen accuracy scalar Acc; GEC-pair filtering uses unreported quality thresholds. No invented entities are introduced.

free parameters (2)
  • Acc (COLA model accuracy in dynamic loss) = unreported, model-specific
    Eq. (2) multiplies the GEC loss by sqrt(Acc), where Acc is the COLA model's measured accuracy. The value is a hand-chosen scalar hyperparameter per language/model and is not reported.
  • GEC-pair filtering thresholds = unreported
    Section 3.1 selects 'high-quality sentence pairs' from GEC datasets, but the quality filters and their thresholds are not specified, affecting the size and nature of the augmented COLA corpora.
assumptions (3)
  • domain assumption Binary acceptability labels derived from GEC pairs (corrected sentence = acceptable, error sentence = unacceptable) are valid training signal for acceptability judgment.
    This assumes the GEC gold correction is always grammatical and the source error always ungrammatical, which may not hold for all annotations or register-dependent usage.
  • domain assumption The COLA model's acceptability score is a meaningful training signal for GEC, and the dynamic loss using it is valid.
    No theoretical or empirical justification is given for the specific weighting sqrt(Acc)*COLA_score; it is taken from the authors' prior LCWL work.
  • standard math ERRANT/ChERRANT F0.5 computed from automatic edits is a faithful measure of correction quality.
    Standard metric, but the paper does not account for metric variance.

how reviews work

0 comments
Cite this review

Pith. "Pith review of COLA-GEC: A Bidirectional Framework for Enhancing Grammatical Acceptability and Error Correction." pith.science (2026). https://pith.science/paper/KKSG5ESS

@misc{pith2026250711867,
  author       = {Pith},
  title        = {Pith review of: COLA-GEC: A Bidirectional Framework for Enhancing Grammatical Acceptability and Error Correction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KKSG5ESS}},
  note         = {Machine review of arXiv:2507.11867}
}
read the original abstract

Grammatical Error Correction (GEC) and grammatical acceptability judgment (COLA) are core tasks in natural language processing, sharing foundational grammatical knowledge yet typically evolving independently. This paper introduces COLA-GEC, a novel bidirectional framework that enhances both tasks through mutual knowledge transfer. First, we augment grammatical acceptability models using GEC datasets, significantly improving their performance across multiple languages. Second, we integrate grammatical acceptability signals into GEC model training via a dynamic loss function, effectively guiding corrections toward grammatically acceptable outputs. Our approach achieves state-of-the-art results on several multilingual benchmarks. Comprehensive error analysis highlights remaining challenges, particularly in punctuation error correction, providing insights for future improvements in grammatical modeling.

Figures

Figures reproduced from arXiv: 2507.11867 by the authors.

Figure 1
Figure 1. Bidirectional Framework of COLA-GEC Incorporating COLA judgements into GEC training: The COLA model evaluates corrected sentences from the GEC model by producing scores for two categories: "0" (grammatically unacceptable) and "1" (grammatically accept￾able). We transform these using a sigmoid function: COLA_score = 1 1 + e −(Logits_0−Logits_1) (1) where Logits0 and Logits1 represent scores for categories 0 and 1 res… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

32 extracted references · 32 canonical work pages

  1. [1]

    In: Proceedings of the 2018 EMNLP Workshop W-NUT: pp

    Boyd, A.: Using Wikipedia Edits in Low Resource Grammatical Error Correction. In: Proceedings of the 2018 EMNLP Workshop W-NUT: pp. 79–84

    work page 2018

  2. [2]

    In: Proceedings of BEA,(2019) pp

    Bryant, C., Felice, M., Andersen, Ø.E., Briscoe, T.: The BEA-2019 Shared Task on Grammatical Error Correction. In: Proceedings of BEA,(2019) pp. 52–75

    work page 2019

  3. [3]

    Cao, H., Yuan, L., Zhang, Y., Du, Z., Sun, Y., and Fu, J. (2023). Unsupervised Grammatical Error Correction Rivaling Supervised Methods. In *Proceedings of the 2023 EMNLP*, pp 3072–3088. 12 Yang et al

    work page 2023

  4. [4]

    In: Proceedings of BEA (2013), pp

    Dahlmeier, D., Ng, H.T.: Building a Large Annotated Corpus of Learner English: The NUS Corpus of Learner English. In: Proceedings of BEA (2013), pp. 22–31

    work page 2013

  5. [5]

    In: Findings of ACL 2023, pp

    Fang, T., Hu, J., Wong, D.F., Chao, L.S., Chang, T.H.: Improving Grammatical Error Correction with Multimodal Feature Integration. In: Findings of ACL 2023, pp. 9328–9344

    work page 2023

  6. [6]

    Fei, Y., Cui, L., Yang, S., Lam, W., Lan, Z., Shi, S.: Enhancing Grammatical Error CorrectionSystemswithExplanations.In:ProceedingsofACL2023),pp.7489–7501 (2023)

    work page 2023

  7. [7]

    Revisiting Acceptability Judgements

    Hu, H., Zhang, Z., Huang, W., Lai, J.Y.K., Li, A., Patterson, Y.: Revisiting Accept- ability Judgements: CoLAC - Corpus of Linguistic Acceptability in Chinese. arXiv preprint arXiv:2305.14091 (2023)

    work page Pith review arXiv 2023

  8. [8]

    In: Proceedings of the NoDaLiDa, (2023)

    Jentoft, M., Samuel, D.: NoCoLA: The Norwegian Corpus of Linguistic Acceptabil- ity. In: Proceedings of the NoDaLiDa, (2023)

    work page 2023

Show all 32 references
  1. [9]

    In: Proceedings of the ACL 2020

    Kaneko, M., Mita, M., Kiyono, S., Suzuki, J., Inui, K.: Encoder-Decoder Models Can Benefit from Pre-trained Masked Language Models in Grammatical Error Cor- rection. In: Proceedings of the ACL 2020

    2020

  2. [10]

    In: Proceedings of ACL 2015, pp

    Lau, J.H., Clark, A., Lappin, S.: Unsupervised Prediction of Acceptability Judge- ments. In: Proceedings of ACL 2015, pp. 1618–1628

    2015

  3. [11]

    University of Chicago Press, Chicago (1993)

    Levin, B.: English Verb Classes and Alternations: A Preliminary Investigation. University of Chicago Press, Chicago (1993)

    1993

  4. [12]

    In: Proceedings of COLING 2025, pp

    Liang, J., Yang, H., Gao, S., Quan, X.: Edit-Wise Preference Optimization for Grammatical Error Correction. In: Proceedings of COLING 2025, pp. 3401–3414

    2025

  5. [13]

    In: Findings of the ACL 2024, pp

    Liu, Y., Li, Z., Jiang, H., Zhang, M., Wang, S., Fu, J., Zhang, Y.: Towards Bet- ter Utilization of Multi-Reference Training Data for Chinese Grammatical Error Correction. In: Findings of the ACL 2024, pp. 3044–3052

    2024

  6. [14]

    5114–5125

    Mikhailov, V., Shamardina, T., Ryabinin, M., Pestova, A., Smurov, I., Artemova, E.:RuCoLA:RussianCorpusofLinguisticAcceptability.In:ProceedingsofEMNLP 2022, pp. 5114–5125

    2022

  7. [15]

    In: Proceedings of the 24th Nordic Conference on Computational Linguistics (NoDaLiDa) 2023, pp

    Nielsen, D.: ScandEval: A Benchmark for Scandinavian Natural Language Process- ing. In: Proceedings of the 24th Nordic Conference on Computational Linguistics (NoDaLiDa) 2023, pp. 185–201

    2023

  8. [16]

    In: Proceedings of SRC Student Research Conference (2025)

    Pham, H.N.A.: Comparing Grammar Error Correction Techniques in English Learners’ Writing. In: Proceedings of SRC Student Research Conference (2025)

    2025

  9. [17]

    In: Proceedings of EMNLP 2024, pp

    Qiu, X., Zhang, J.: Label Confidence Weighted Learning for Target-level Sentence Simplification. In: Proceedings of EMNLP 2024, pp. 999–1010

    2024

  10. [18]

    In: Proceedings of the LREC 2024, pp

    Someya, T., Sugimoto, Y., Oseki, Y.: JCoLA: Japanese Corpus of Linguistic Ac- ceptability. In: Proceedings of the LREC 2024, pp. 7428–7435

    2024

  11. [19]

    In: Findings of EMNLP 2021, pp

    Trotta, D., Guarasci, R., Leonardelli, E., Tonelli, S.: Monolingual and Cross- Lingual Acceptability Judgments with the Italian CoLA Corpus. In: Findings of EMNLP 2021, pp. 2922–2931

    2021

  12. [20]

    In: Proceedings of the 10th Workshop on NLP for Computer Assisted Language Learning, pp

    Volodina, E., Mohammed, Y.A., Klezl, J.: DaLAJ – a Dataset for Linguistic Ac- ceptability Judgments for Swedish. In: Proceedings of the 10th Workshop on NLP for Computer Assisted Language Learning, pp. 28–37. LiU Electronic Press (2021)

    2021

  13. [21]

    Transactions of the Association for Computational Linguistics, 7, 625–641 (2019)

    Warstadt, A., Singh, A., Bowman, S.R.: Neural Network Acceptability Judgments. Transactions of the Association for Computational Linguistics, 7, 625–641 (2019)

    2019

  14. [22]

    Transactions of the Association for Computational Linguistics, 8, 377–392 (2020)

    Warstadt, A., Parrish, A., Liu, H., Mohananey, A., Peng, W., Wang, S., Bowman, S.R.: BLiMP: The Benchmark of Linguistic Minimal Pairs for English. Transactions of the Association for Computational Linguistics, 8, 377–392 (2020)

    2020

  15. [23]

    In: Proceedings of EACL 2021, pp

    Xiang, B., Yang, C., Li, Y., Zhang, Y., Qiu, X.: CLiMP: A Benchmark for Chinese Language Model Evaluation. In: Proceedings of EACL 2021, pp. 2784–2790. Title Suppressed Due to Excessive Length 13

    2021

  16. [24]

    In: Findings of EMNLP 2022, pp

    Xu, L., Wu, J., Peng, J., Wang, Y., Wang, Y., Zhang, H., Li, Z.: FCGEC: Fine- Grained Corpus for Chinese Grammatical Error Correction. In: Findings of EMNLP 2022, pp. 1900–1918

    2022

  17. [25]

    In: Proceedings of the 49th ACL (2011)

    Yannakoudakis, H., Briscoe, T., Medlock, B.: A New Dataset and Method for Automatically Grading ESOL Texts. In: Proceedings of the 49th ACL (2011)

    2011

  18. [26]

    In: Proceedings of the CoNLL 2013: Shared Task, pp

    Yuan, Z., Felice, M.: Constrained Grammatical Error Correction using Statistical Machine Translation. In: Proceedings of the CoNLL 2013: Shared Task, pp. 52–61

    2013

  19. [27]

    In: Proceedings of NAACL (2022)

    Zhang, Y., Li, J., Li, Z., Zhang, M., Huang, F.: MuCGEC: A Multi-Reference Multi-Source Evaluation Dataset for Chinese Grammatical Error Correction. In: Proceedings of NAACL (2022)

    2022

  20. [28]

    In: Proceed- ings of EMNLP 2022, pp

    Zhang,Y.,Zhang,B.,Li,Z.,Bao,Z.,Li,C.,Zhang,M.:SynGEC:Syntax-Enhanced Grammatical Error Correction with a Tailored GEC-Oriented Parser. In: Proceed- ings of EMNLP 2022, pp. 2518–2531

    2022

  21. [29]

    In: Findings of the ACL 2023, pp

    Zhang, Y., Li, J., Li, Z., Zhang, M., Huang, F.: NaSGEC: A Multi-Domain Chinese Grammatical Error Correction Dataset from Native Speaker Texts. In: Findings of the ACL 2023, pp. 9987–10000

    2023

  22. [30]

    In: Proceedings of ACL 2024, pp

    Zhang, Y., Chen, J., Li, A., Hu, H., Qiu, X.: MELA: Multilingual Evaluation of Linguistic Acceptability. In: Proceedings of ACL 2024, pp. 1932–1947

    2024

  23. [31]

    In: Proceedings of AAAI (2020), pp

    Zhao, Z., Wang, H.: Improving Neural Grammatical Error Correction via Dynamic Masking. In: Proceedings of AAAI (2020), pp. 1226–1233

    2020

  24. [32]

    In: Findings of the EMNLP 2023, pp

    Zhou, H., Liu, Y., Li, Z., Zhang, M., Wang, S., Wang, D., Fu, J.: Improving Seq2Seq Grammatical Error Correction via Decoding Interventions. In: Findings of the EMNLP 2023, pp. 7393–7405. 7 Appendix T able 7.G-COLA Monolingual Dataset Statistics Language Source Train Dev Test ...

    2023

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.