Pith. sign in

REVIEW 3 major objections 4 minor 37 references

Bangla Grammatical Error Detection Leveraging Transformer-based Token Classification

T0 review · 3 major / 4 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read The paper shows that Bangla grammatical error detection can be cast as a token-classification problem, with a BanglaBERT-based intersection ensemble and rule-based fixes reaching a Levenshtein distance of 1.04 on the shared test set.

desk verdict A workmanlike Bangla GED system with a useful ablation, but the headline score is compromised by explicit test-set tuning and unreconciled tables. read the letter →

arxiv 2411.08344 v1 pith:6JG5Y7E3 submitted 2024-11-13 cs.CL cs.LG

Shayekh Bin Islam , Ridwanul Hasan Tanvir , Sihat Afnan This is my paper · ORCID
classification cs.CLcs.LG
keywords BanglagrammaticalerrordetectiontokenclassificationBERTensemblelearningLevenshteindistancesequencelabelinglow-resourceNLPrule-basedpost-processing
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

Bangla grammatical error detection—locating the exact character spans of spelling, punctuation, and grammatical mistakes—is an under-built task for a language with hundreds of millions of speakers. The paper tries to establish that the task can be solved as a four-class token classification problem on top of a pre-trained Bangla transformer, avoiding the need to hand-write a comprehensive Bangla grammar. The final recipe combines an intersection ensemble of the three best checkpoints of BanglaBERT-base and BanglaBERT-large with confidence thresholding, label smoothing, Unicode normalization and reverse normalization, and deterministic rules for punctuation and spelling. On the shared challenge test set, the authors report a Levenshtein distance of 1.04, which is lower than the distances they report for multilingual baselines and for single BanglaBERT checkpoints. A reader who wants to build a Bangla typing assistant would use this as evidence that a span-detection pipeline, not a full error-correction generator, is the right starting point.

What carries the argument

The load-bearing mechanism is the checkpoint-intersection ensemble. For each model size, the three individually best checkpoints of the same finetuning run predict error spans, and the final prediction is the intersection of the base-model ensemble and the large-model ensemble; intersection suppresses false-positive spans at the expense of recall. The token classifier uses a four-class BIO-like scheme (O = no error, B = error begins, I = error continues, M = a character is missing after this token), with a separate proxy label on the [CLS] token when more than 30% of the text's offsets are errors. Around the ensemble sit four supporting mechanisms: confidence thresholding near 0.8 that keeps the model from flagging a token unless it is sure; label smoothing (0.1 for base, 0.2 for large) to regularize a model trained on only about 20,000 texts; Unicode normalization before tokenization and reverse normalization after prediction via minimum-edit-distance alignment, so spans are reported on the original text; and deterministic post-processing for spaces before punctuation, missing end-of-sentence punctuation, and dictionary- and named-entity-filtered spelling errors.

What would settle it

Run the authors' final pipeline exactly as described but select every hyper-parameter—threshold, label-smoothing factor, checkpoint count, normalization on/off, ensemble type—using only the development set, with the test set touched exactly once at the end; if the resulting Levenshtein distance is materially above the reported 1.04, the central claim's number depends on test-set adaptation rather than on the method's inherent quality.

Watch

Extended reading notes

Core claim

On its own terms, the paper claims that a properly regularized and ensembled BanglaBERT token classifier is a strong detector of Bangla grammatical errors. The discovery is not a new model architecture but a combination recipe: model the task with O/B/I/M token labels, finetune BanglaBERT-base and BanglaBERT-large, take the intersection of the three best checkpoints from each size, intersect those two results, apply a confidence threshold near 0.8, smooth labels during training, normalize Unicode before tokenizing and reverse-normalize the predicted spans through edit-distance alignment, and finish with deterministic rules for spaced punctuation, missing end punctuation, and dictionary-filtered spelling errors. With that recipe the paper reports a Levenshtein distance of 1.04 on the test set, below the 1.1844 of the best single BanglaBERT-large checkpoint and below the 1.3552 of the DeBERTa-V3-large baseline.

Load-bearing premise

The reported test score assumes the final hyper-parameters were chosen from the development set only; the paper's Section 5.8 says they were chosen on 'the dev set and the test set,' and if that is literal, the headline Levenshtein number is partly the result of peeking at the test set and is not a clean out-of-sample measurement.

Editorial extensions

If this is right

  • A production Bangla typing assistant can highlight the predicted error spans directly, because the system's output is a set of character offsets rather than a corrected sentence.
  • Intersection ensembling of checkpoints is more valuable than scaling up the model alone: the reported three-checkpoint intersection (1.054) beats the single-checkpoint large model (1.0648) on the same test set.
  • Unicode normalization with reverse mapping is a necessary component for low-resource token classifiers, since it avoids out-of-vocabulary tokens while keeping predictions aligned with the user's original text.
  • Adding deterministic punctuation rules on top of the neural model yields a small but consistent improvement, indicating that hybrid learned-plus-hand-coded pipelines suit this task.
  • The recipe is language-agnostic in its machinery; the only Bangla-specific assets are the pre-trained model, the spelling-error database, and the normalization rules.

Reading between the lines

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

  • The success of intersection ensembling implies that false positives dominate the error budget; reporting span-level precision and recall separately would reveal whether the ensemble is rejecting weak candidates or hiding real errors.
  • The paper's observation that synthetic data helps the dev set but hurts the test set suggests a distribution shift between the splits; a follow-up could use domain-adaptive self-training or per-source confidence thresholds instead of discarding synthetic data.
  • The same token-classification recipe should transfer to other low-resource languages that have an ELECTRA-style pre-trained model and a small span-annotated error corpus, since none of the core steps depend on Bangla-specific morphology.
  • A tokenizer with a built-in byte-level offset map could replace the edit-distance reverse-normalization step, removing an extra alignment and its hand-written corrections.
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

3 major / 4 minor

Summary. The paper addresses Bangla grammatical error detection as a token-classification problem. It fine-tunes BanglaBERT-base and BanglaBERT-large, compares them with XLM-RoBERTa-base and DeBERTa-V3-large, and combines the best checkpoints through an intersection ensemble. The system adds confidence thresholding, Unicode normalization with reverse mapping, deterministic punctuation correction, and a large spelling-error database with a named-entity filter. The authors report a best Levenshtein distance of 1.04 (abstract) or 1.054 (Table 6) on the Bhashabhrom test set and conclude that their system outperforms the internal baselines.

Significance. If the reported evaluation were valid, this would be a useful empirical contribution to a low-resource language task: it systematically ablates regularization, ensembling, thresholding, normalization, and deterministic post-processing, and it uses an external public benchmark. The paper does not provide code or machine-checked proofs, so its value rests on the reliability of the test-set numbers. That reliability is currently undermined by the explicit statement in Section 5.8 that hyperparameters were selected using both the dev and test sets, and by incompatible numbers across tables. The central claim can be repaired by a clean rerun with dev-only selection, but as written the headline score is not an unbiased out-of-sample result.

major comments (3)
  1. [§5.8, Abstract, Table 6] The central claim is the Levenshtein score of 1.04 (abstract) or 1.054 (Table 6) on the Bhashabhrom test set. Section 5.8 states: 'We apply all the aforementioned techniques with the hyper-parameters that perform best in the dev set and the test set.' Read literally, the final configuration was selected using test-set labels, so the reported score is not an unbiased out-of-sample estimate and cannot be compared with baseline numbers that were not selected on the test set. The abstract/Table 6 discrepancy (1.04 vs 1.054) must also be reconciled. Please rerun the evaluation with all hyperparameter and checkpoint selections made on the dev set only and report the resulting test score.
  2. [Tables 3, 5, 7, 8, 9] The published tables cannot be reconciled for what should be the same configuration of BanglaBERT-large (punctuation post-processing, no normalization, no thresholding, standard cross-entropy loss): Table 3 reports 1.1844, Table 5 reports 1.2212, Table 7 reports 1.164, Table 8 (threshold 0.0) reports 1.1892, and Table 9 (without normalization) reports 1.130. These inconsistencies make it impossible to verify the size of the claimed improvements from ensembling, label smoothing, thresholding, and normalization. Please provide a single consistency-checked table with exact configurations and clarify which entries are re-runs versus the same run reused across ablation settings.
  3. [§5.4, §5.8] The ensemble is built from 'the three best checkpoints' of BanglaBERT-base and BanglaBERT-large, but the paper never states whether these checkpoints are chosen on the dev set or the test set. Given the Section 5.8 statement, test-set selection appears possible, and the checkpoint-selection criterion must be made explicit and restricted to dev data for the headline claim to be valid.
minor comments (4)
  1. [§3.1.2] The sentence 'We incorporate a label for the [CLS] token when the proportion of toxic offsets in the text exceeds 30%' appears to have been copied from the toxic spans detection literature; the current task is grammatical error detection and has no toxic offsets. Please replace this with the actual class-labelling rule for the [CLS] token or remove the sentence.
  2. [§5.3] There is a typo in 'the models fail to capture these errors an we notice' — 'an' should be 'and'.
  3. [§5.4] The sentence 'Hence, We take the intersection...' has an unnecessarily capitalized 'We'; please fix the capitalization.
  4. [Abstract and §4.1] The abstract says the dataset contains 'over 25,000 texts' while Section 4.1 says 'around 25000 texts'; please make the description consistent.

Circularity Check

2 steps flagged · score 6.0 of 10

Final reported Levenshtein score is selected on the test set, making the central performance claim a fitted quantity rather than an out-of-sample prediction.

  1. fitted input called prediction [Section 5.8 (The Final System); also Section 5.4, Tables 3, 5, 6, 7, 8, 9]
    "We apply all the aforementioned techniques with the hyper-parameters that perform best in the dev set and the test set."

    The headline score (1.054 in Table 6, 1.04 in the abstract) is the test-set Levenshtein distance of a configuration whose hyperparameters, backbone, ensemble strategy, smoothing factor, threshold, and normalization were selected by comparing test-set scores across Tables 3, 5, 6, 7, 8, and 9. The 'best' configuration is therefore defined as the one minimizing test LD, and the reported test LD is the value of that selection criterion. It is not an out-of-sample prediction; the result is statistically forced by construction because the same data are used to choose the configuration and to evaluate it.

  2. fitted input called prediction [Section 5.4 (Ensemble Strategy), Tables 5 and 6]
    "Hence, We take the intersection of the three best checkpoints of BanglaBERT-base and BanglaBERT-large and combine the two predictions by intersection again to achieve the best Levenstein score in the test set."

    The ensemble design is chosen after inspecting test-set LD values in Tables 5 and 6, where 'Single-checkpoint 1.0648' and 'Three-checkpoints 1.054' are both test-set numbers. The final 'Three-checkpoints' score is the winning value of a test-set comparison, not an unbiased estimate of how the ensemble generalizes. This is a second, explicit instance of using the test set as a model-selection signal and then reporting the selected score as the system's performance.

full rationale

The paper is an empirical systems paper with no formal derivation chain, so the relevant circularity question is whether the reported performance number is an honest out-of-sample evaluation. Section 5.8 states that the final system uses 'the hyper-parameters that perform best in the dev set and the test set,' and the surrounding sections show that backbone choice (Table 3), ensemble strategy (Tables 5 and 6), label smoothing (Table 7), thresholding (Table 8), and Unicode normalization (Table 9) were all compared on the test set. The headline Levenshtein score is therefore a selection result rather than a prediction: the configuration was chosen to minimize test error, and the same test error is then reported as the achievement. This is a clear case of fitted input being called a prediction, and it directly undermines the central claim of outperforming baselines. Separately, the paper contains a copy-paste artifact in Section 3.1.2 about a '[CLS] token when the proportion of toxic offsets in the text exceeds 30%,' which is imported from toxic-span detection and is not a meaningful description of Bangla grammatical error labelling; this is a correctness and reporting risk, though not itself circular. The baseline numbers are also inconsistent across Tables 3, 5, 7, 8, and 9 for comparable BanglaBERT-large configurations, making the claimed improvements hard to verify. These issues are significant, but they are evaluation-integrity problems rather than self-citation or definitional circularity. The score of 6 reflects that the central reported number is, by the paper's own account, fitted to the test set.

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

The paper introduces no new theoretical entities. It relies on pretrained models, external spelling data, and rule-based post-processing; the central claim is an empirical performance number. The main parameters are standard hyperparameters tuned on the dev set, with the additional concern that the authors state some were tuned with the test set in mind.

free parameters (5)
  • confidence threshold = 0.8
    Chosen from the dev set (Section 5.6) and applied at inference; the final performance depends on this value.
  • label smoothing factor = 0.1 for BanglaBERT-base, 0.2 for BanglaBERT-large
    Tuned separately per model (Section 5.5); affects the model outputs used in the ensemble.
  • number of checkpoints in ensemble = 3
    Section 5.4: intersection of the three best checkpoints of each model; changing this count changes the intersection and the final score.
  • CLS label proportion threshold = 30%
    Hand-set threshold from the toxic span detection literature (Chhablani et al., 2021), used to assign a proxy label; Section 3.1.2.
  • maximum sequence length = 384
    Tokenization setting in Section 4.3; affects how longer sentences are truncated and therefore the predicted spans.
assumptions (4)
  • domain assumption BanglaBERT pretrained weights provide useful representations for Bangla grammatical error detection.
    The entire approach is built on BanglaBERT (Section 3.2.1); if its representations were inadequate, the reported scores would not hold.
  • domain assumption The Bhashabhrom dataset annotations are correct and the Levenshtein distance is the appropriate official evaluation metric.
    The evaluation takes the dataset and metric from Murad et al. (2023) at face value (Section 4.1, 4.2).
  • domain assumption The DPCSpell spelling-error database, Bangla dictionary filter, Wikipedia title filter, and BNLP NER model collectively separate true spelling errors from named entities.
    The deterministic spelling component in Section 3.5.2 relies on these external resources being accurate and complete.
  • domain assumption Character offsets can be reliably mapped between normalized and original text using minimum Levenshtein alignment from NLTK.
    The reverse normalization step (Section 3.3.2) depends on this alignment being correct; misalignment would corrupt the final error spans.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Bangla Grammatical Error Detection Leveraging Transformer-based Token Classification." pith.science (2026). https://pith.science/paper/6JG5Y7E3

@misc{pith2026241108344,
  author       = {Pith},
  title        = {Pith review of: Bangla Grammatical Error Detection Leveraging Transformer-based Token Classification},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6JG5Y7E3}},
  note         = {Machine review of arXiv:2411.08344}
}
read the original abstract

Bangla is the seventh most spoken language by a total number of speakers in the world, and yet the development of an automated grammar checker in this language is an understudied problem. Bangla grammatical error detection is a task of detecting sub-strings of a Bangla text that contain grammatical, punctuation, or spelling errors, which is crucial for developing an automated Bangla typing assistant. Our approach involves breaking down the task as a token classification problem and utilizing state-of-the-art transformer-based models. Finally, we combine the output of these models and apply rule-based post-processing to generate a more reliable and comprehensive result. Our system is evaluated on a dataset consisting of over 25,000 texts from various sources. Our best model achieves a Levenshtein distance score of 1.04. Finally, we provide a detailed analysis of different components of our system.

Figures

Figures reproduced from arXiv: 2411.08344 by the authors.

Figure 1
Figure 1. Transformer-based Token Classification Mod [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. LSTM-CRF Token Classification Models 3.3 Post-processing 3.3.1 Error Span Generation Biasing the method to prefer the original tokens, unless the model is very confident about an error, produces superior results (Alikaniotis and Raheja, 2019). We choose the confidence threshold that performs best in the dev set, often ranging from 0.7 to 0.8. Once our model generates the token-level labels, we further process the ou… view at source ↗
Figure 3
Figure 3. Effect of confidence thresholding on the dev [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Separate head for missing errors 7 Conclusion In this work, we present our approach to solving the Bangla grammatical error detection problem that outperforms multiple baselines. We formal￾ize the problem as a sequence labeling problem and apply different transformer-b…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

37 extracted references · 22 canonical work pages

  1. [1]

    ABA Abdullah and Ashfaq Rahman. 2003. A generic spell checker engine for south asian languages. In Conference on Software Engineering and Applications (SEA 2003), pages 3--5

    work page 2003

  2. [2]

    Tanvirul Alam, Akib Khan, and Firoj Alam. 2020. Punctuation restoration using transformer models for high-and low-resource languages. In Proceedings of the Sixth Workshop on Noisy User-generated Text (W-NUT 2020), pages 132--142

    work page 2020

  3. [3]

    Dimitrios Alikaniotis and Vipul Raheja. 2019. The unreasonable effectiveness of transformer language models in grammatical error correction. arXiv preprint arXiv:1906.01733

    work page Pith review arXiv 2019

  4. [4]

    Archit Bansal, Abhay Kaushik, and Ashutosh Modi. 2021. Iitk@ detox at semeval-2021 task 5: Semi-supervised learning and dice loss for toxic spans detection. arXiv preprint arXiv:2104.01566

    work page Pith review arXiv 2021

  5. [5]

    Sohel Rahman, and Rifat Shahriyar

    Abhik Bhattacharjee, Tahmid Hasan, Wasi Ahmad, Kazi Samin Mubasshir, Md Saiful Islam, Anindya Iqbal, M. Sohel Rahman, and Rifat Shahriyar. 2022. https://aclanthology.org/2022.findings-naacl.98 B angla BERT : Language model pretraining and benchmarks for low-resource language understanding evaluation in B angla . In Findings of the Association for Computat...

    2022

  6. [6]

    Mehedi Hasan Bijoy, Nahid Hossain, Salekul Islam, and Swakkhar Shatabda. 2022. Dpcspell: A transformer-based detector-purificator-corrector framework for spelling error correction of bangla and resource scarce indic languages. arXiv preprint arXiv:2211.03730

    work page Pith review arXiv 2022

  7. [7]

    CJ Bryant, Mariano Felice, and Edward Briscoe. 2017. Automatic annotation and evaluation of error types for grammatical error correction. Association for Computational Linguistics

    work page 2017

  8. [8]

    Bidyut Baran Chaudhuri. 2001. Reversed word dictionary and phonetically similar word grouping based spell-checker to bangla text. In Proc. LESAL Workshop, Mumbai

    work page 2001

Show all 37 references
  1. [9]

    Hugh Chen, Scott Lundberg, and Su-In Lee. 2017. Checkpoint ensembles: Ensemble methods from a single training process. arXiv preprint arXiv:1710.03282

    2017 arXiv

  2. [10]

    Ruijun Chen, Jin Wang, and Xuejie Zhang. 2021. Ynu-hpcc at semeval-2021 task 5: Using a transformer-based model with auxiliary information for toxic span detection. In Proceedings of the 15th International Workshop on Semantic Evaluation (SemEval-2021), pages 841--845

    2021

  3. [11]

    Gunjan Chhablani, Abheesht Sharma, Harshit Pandey, Yash Bhartia, and Shan Suthaharan. 2021. Nlrg at semeval-2021 task 5: toxic spans detection leveraging bert-based token classification and span prediction techniques. arXiv preprint arXiv:2102.12254

    2021 arXiv

  4. [12]

    Jan Chorowski and Navdeep Jaitly. 2016. Towards better decoding and language model integration in sequence to sequence models. arXiv preprint arXiv:1612.02695

    2016 arXiv

  5. [13]

    Jesse Dodge, Gabriel Ilharco, Roy Schwartz, Ali Farhadi, Hannaneh Hajishirzi, and Noah Smith. 2020. Fine-tuning pretrained language models: Weight initializations, data orders, and early stopping. arXiv preprint arXiv:2002.06305

    2020 arXiv

  6. [14]

    Sreyan Ghosh and Sonal Kumar. 2021. Cisco at semeval-2021 task 5: What's toxic?: Leveraging transformers for multiple toxic span extraction from online comments. arXiv preprint arXiv:2105.13959

    2021 arXiv

  7. [15]

    KM Azharul Hasan, Muhammad Hozaifa, and Sanjoy Dutta. 2014. Detection of semantic errors from simple bangla sentences. In 2014 17th International Conference on Computer and Information Technology (ICCIT), pages 296--299. IEEE

    2014

  8. [16]

    Sohel Rahman, and Rifat Shahriyar

    Tahmid Hasan, Abhik Bhattacharjee, Kazi Samin, Masum Hasan, Madhusudan Basak, M. Sohel Rahman, and Rifat Shahriyar. 2020. https://doi.org/10.18653/v1/2020.emnlp-main.207 Not low-resource anymore: Aligner ensembling, batch filtering, and new datasets for B engali- E nglish mach...

    2020 doi

  9. [17]

    Dawid Jurkiewicz, ukasz Borchmann, Izabela Kosmala, and Filip Grali \'n ski. 2020. Applicaai at semeval-2020 task 11: On roberta-crf, span cls and whether self-training helps them. arXiv preprint arXiv:2005.07934

    2020 arXiv

  10. [18]

    Nur Hossain Khan, Gonesh Chandra Saha, Bappa Sarker, and Md Habibur Rahman. 2014. Checking the correctness of bangla words using n-gram. International Journal of Computer Application, 89(11)

    2014

  11. [19]

    Ilya Loshchilov and Frank Hutter. 2017. Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101

    2017 arXiv

  12. [20]

    Michal Lukasik, Srinadh Bhojanapalli, Aditya Menon, and Sanjiv Kumar. 2020. Does label smoothing mitigate label noise? In International Conference on Machine Learning, pages 6448--6458. PMLR

    2020

  13. [21]

    Prianka Mandal and BM Mainul Hossain. 2017. Clustering-based bangla spell checker. In 2017 IEEE International Conference on Imaging, Vision & Pattern Recognition (icIVPR), pages 1--6. IEEE

    2017

  14. [22]

    Rafael M \"u ller, Simon Kornblith, and Geoffrey E Hinton. 2019. When does label smoothing help? Advances in neural information processing systems, 32

    2019

  15. [23]

    Md Boktiar Mahbub Murad, Sushmit, and Tasnim Nishat Islam. 2023. https://kaggle.com/competitions/bengali-ged Apurba presents bhashabhrom: Eee day 2023 datathon

    2023

  16. [24]

    Viet Anh Nguyen, Tam Minh Nguyen, Huy Quang Dao, and Quang Huu Pham. 2021. S-nlp at semeval-2021 task 5: An analysis of dual networks for sequence tagging. In Proceedings of the 15th international workshop on semantic evaluation (SemEval-2021), pages 888--897

    2021

  17. [25]

    John Pavlopoulos, Jeffrey Sorensen, L \'e o Laugier, and Ion Androutsopoulos. 2021. Semeval-2021 task 5: Toxic spans detection. In Proceedings of the 15th international workshop on semantic evaluation (SemEval-2021), pages 59--69

    2021

  18. [26]

    Chowdhury Rafeed Rahman, MD Rahman, Samiha Zakir, Mohammad Rafsan, and Mohammed Eunus Ali. 2022. Bspell: A cnn-blended bert based bengali spell checker. arXiv preprint arXiv:2208.09709

    2022 arXiv

  19. [27]

    F \'a bio Souza, Rodrigo Nogueira, and Roberto Lotufo. 2019. Portuguese named entity recognition using bert-crf. arXiv preprint arXiv:1909.10649

    2019 arXiv

  20. [28]

    Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jon Shlens, and Zbigniew Wojna. 2016. Rethinking the inception architecture for computer vision. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 2818--2826

    2016

  21. [29]

    Naushad UzZaman and Mumit Khan. 2006. A comprehensive bangla spelling checker

    2006

  22. [30]

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, ukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. Advances in neural information processing systems, 30

    2017

  23. [31]

    Chenyi Wang, Tianshu Liu, and Tiejun Zhao. 2021. Hitmi&t at semeval-2021 task 5: integrating transformer and crf for toxic spans detection. In Proceedings of the 15th International Workshop on Semantic Evaluation (SemEval-2021), pages 870--874

    2021

  24. [32]

    Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pierric Cistac, Tim Rault, R \'e mi Louf, Morgan Funtowicz, et al. 2020. Transformers: State-of-the-art natural language processing. In Proceedings of the 2020 conference on empirical met...

    2020

  25. [33]

    Enwei Zhu and Jinpeng Li. 2022. Boundary smoothing for named entity recognition. arXiv preprint arXiv:2204.12031

    2022 arXiv

  26. [34]

    Qinglin Zhu, Zijie Lin, Yice Zhang, Jingyi Sun, Xiang Li, Qihui Lin, Yixue Dang, and Ruifeng Xu. 2021. Hitsz-hlt at semeval-2021 task 5: Ensemble sequence labeling and span boundary detection for toxic span detection. In Proceedings of the 15th international workshop on semant...

    2021

  27. [35]

    Barret Zoph, Vijay Vasudevan, Jonathon Shlens, and Quoc V Le. 2018. Learning transferable architectures for scalable image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 8697--8710

    2018

  28. [36]

    URL: " 'urlintro :=

    ENTRY address author booktitle chapter edition editor howpublished institution journal key month note number organization pages publisher school series title type volume year eprint doi pubmed url lastchecked label extra.label sort.label short.list INTEGERS output.state before...

  29. [37]

    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 gl...

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.