Pith. sign in

REVIEW 3 major objections 4 minor 49 references

Fine-Grained Chinese Hate Speech Understanding: Span-Level Resources, Coded Term Lexicon, and Enhanced Detection Frameworks

T0 review · 3 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read This paper claims that two-stage fine-tuning—first on a coded-hate-term lexicon, then on task data—improves Chinese hate speech detection beyond either stage alone.

desk verdict Valuable new Chinese hate-speech resources wrapped around a method claim that its own Table X contradicts. read the letter →

arxiv 2507.11292 v1 pith:GEUIURJ6 submitted 2025-07-15 cs.CL

Zewen Bai , Liang Yang , Shengdi Yin , Yuanyuan Sun , Hongfei Lin This is my paper
classification cs.CL
keywords hatespeechdetectionspan-levelannotationChinesecodedtermslexicontwo-stagefine-tuninglargelanguagemodelstarget-argumentextractionoffensive
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

Chinese hate speech is under-served by fine-grained resources, so this paper builds a span-level dataset, a coded-hate-term lexicon, and a training recipe that combines them. The dataset, STATE ToxiCN, annotates 8,029 posts with 9,533 target–argument–hateful–group quadruples to expose multi-target and implicit hate that sentence-level labels miss. The lexicon records 830 Chinese coded hate terms with target-group and explanation annotations. On the COLD and CDial-Bias benchmarks, the proposed two-stage fine-tuning—lexicon first, then task data—achieves the highest F1 scores (70.57 and 60.04), above both lexicon-only and task-only fine-tuning. The intended contribution is evidence that explicit coded-hate knowledge, injected before task fine-tuning, makes detection more sensitive and more precise.

What carries the argument

The load-bearing machinery is a two-stage fine-tuning pipeline: a LoRA adapter is trained first on the annotated lexicon of 830 Chinese coded hate terms, then fine-tuned on the target hate-speech dataset, so the first stage supplies high-recall term knowledge and the second stage recontextualizes it. The lexicon is organized into phono-graphemic variation terms (homophones, acronyms, character splitting or combining) and hateful semantic terms (slang, cultural allusions, subculture usages), each carrying targeted-group labels and a plain-language explanation. The evaluation scaffold around the pipeline is the Target-Argument-Hateful-Group quadruple schema of STATE ToxiCN, which provides span-level ground truth for measuring whether models locate exactly who is attacked and why.

What would settle it

Rerun Only LoRA, Only Lexicon, and Two Stage with multiple random seeds and a validation split for hyperparameter choice; if the F1 gap on COLD (70.57 versus 68.68) and on CDial-Bias (60.04 versus 58.27) falls within one standard deviation, the method's advantage is not established.

Watch

Extended reading notes

Core claim

The paper's central claim is that fine-grained Chinese hate speech understanding improves when span-level supervision and explicit coded-term knowledge are combined. STATE ToxiCN is presented as the first span-level Chinese hate speech dataset, using a Target-Argument-Hateful-Group quadruple schema to capture which group is attacked, with what argument, and whether the pair is hateful. The Chinese coded hate lexicon is presented as the first interpretable resource of its kind, dividing 830 terms into phono-graphemic variation terms (sound or character disguises) and hateful semantic terms (culturally loaded slurs), each with target-group and explanation annotations. The culminating method result is that two-stage fine-tuning—lexicon pre-tuning followed by dataset-specific LoRA fine-tuning—outperforms both Only LoRA and Only Lexicon on COLD and CDial-Bias, with F1 of 70.57 versus 68.68 on COLD and 60.04 versus 58.27 on CDial-Bias. Along the way, the paper reports that fine-tuned open models beat closed LLM APIs on span extraction, and that DeepSeek-v3 leads in coded-term identification and explanation.

Load-bearing premise

The central claim collapses if the two-stage improvement over the baselines is noise, because the paper chooses learning rates by highest test F1 and reports single runs with no variance estimate.

Editorial extensions

If this is right

  • The two-stage lexicon-then-fine-tuning recipe lifts F1 on COLD from 68.68 to 70.57 and on CDial-Bias from 58.27 to 60.04 relative to LoRA alone, so it is the recommended configuration among those tested.
  • Lexicon-only fine-tuning alone reaches 87.83 recall on COLD, meaning the lexicon can serve as a fast high-recall screening stage even without the second fine-tuning step.
  • Span-level supervision is the main differentiator in hard tasks: fine-tuned models stay near 50% soft F1 on quadruple extraction while API-only LLMs stay between 11% and 27%, identifying span-boundary precision as the bottleneck.
  • STATE ToxiCN can be used as a benchmark for semantic understanding of Chinese hate speech, not just binary label accuracy.
  • Coded hate term explanations are hardest for phono-graphemic variation terms across all models, so future interpretability work should target that category.

Reading between the lines

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

  • Editorial inference: the two-stage recipe likely transfers to other low-resource abusive-language settings, since a small annotated lexicon is inexpensive to build and the pipeline only needs a modest task corpus for the second stage.
  • Editorial inference: the observed asymmetry—fine-tuned models lose span F1 on coded-term posts while LLM APIs gain—suggests a hybrid that feeds lexicon explanations as context to a fine-tuned extraction model could beat both; the paper does not test this.
  • Editorial inference: because the lexicon is stored as JSON terms with groups and definitions, it could be updated continuously as new coded terms appear, and the two-stage framework re-run on each update; the paper lists adaptive lexicons only as future work.
  • Editorial inference: the language bias in the evaluation (Chinese-specific homophones and character forms) means the coded-term taxonomy is probably not portable as-is, but the two-category scheme likely is.
Share X Bluesky LinkedIn Reddit HN

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 presents STATE ToxiCN, a span-level Chinese hate speech dataset with 8,029 posts and 9,533 Target-Argument-Hateful-Group quadruples derived from TOXICN; a Chinese coded hate term lexicon of 830 terms with group and explanation annotations; and an evaluation protocol with three tasks (quadruple identification, coded term identification, coded term explanation) applied to twelve LLMs. The final contribution is a two-stage fine-tuning framework that first trains on the lexicon and then on the target detection dataset, evaluated on COLD and CDial-Bias. The paper claims this framework consistently achieves the highest F1 on both external datasets.

Significance. The resource contributions are potentially significant if released: STATE ToxiCN appears to be the first span-level Chinese hate speech dataset with quadruple annotations, and the 830-term coded hate lexicon with explanations is a novel resource for studying implicit hate in Chinese. The evaluation of LLMs, including open, safety-domain, and closed models, documents a clear performance gap between fine-tuned models and API models on span-level tasks and identifies DeepSeek-v3 as notably stronger on coded term tasks. The strength of the resource construction is undercut by the unsupported third contribution: the central claim that the two-stage framework 'significantly enhances' detection is contradicted by the paper's own Table X and rests on test-set hyperparameter selection with single runs and no significance testing. I do not see a circularity problem in the evaluation: the two-stage experiment is run on external COLD and CDial-Bias, and the lexicon's provenance from TOXICN does not leak those labels. The paper ships no machine-checked proofs, but the empirical claims are testable; the main issue is that they are not established as stated.

major comments (3)
  1. [Section VIII, Table X] The claim in Section VIII(b) that the two-stage framework 'consistently outperforms both Only LoRA and Only Lexicon by achieving the highest F1 scores on both COLD (70.57) and Cdial-Bias (60.04)' is contradicted by Table X: on COLD, Only Lexicon achieves F1 71.19, which is higher than Two Stage's 70.57. Consequently, the abstract's assertion that the proposed method 'significantly enhanc[es] hate speech detection performance' and the conclusion's 'demonstrably enhancing' are not supported by the reported data on one of the two datasets.
  2. [Section V-C, Tables VII, VIII, X] The learning rate is selected by highest test-set F1 ('ultimately selecting the one that yielded the highest F1 score on the test set'), and all reported metrics are single runs without standard deviations, confidence intervals, or significance tests. The advantages of Two Stage over Only LoRA are 1.89 F1 points on COLD and 1.77 F1 points on Cdial-Bias; under test-set selection and single-run noise, this ordering cannot be reliably attributed to the method. This concern applies not only to the two-stage claim but also to the fine-tuned-versus-API comparisons in Section VI.
  3. [Section VIII, Table XI] The training setup for 'Only Lexicon' and the first stage of 'Two Stage' is not specified precisely enough to be reproducible: a lexicon of (Term, Groups, Definition) entries is not a labeled binary classification corpus, and the manuscript does not state how these entries are converted into training examples, what labels are used, or how many examples result. Without this detail, the comparison between Only Lexicon and the other conditions cannot be interpreted.
minor comments (4)
  1. [Sections V-B and VII-A, Table IX] Task 2 is described as using recall as the primary metric, but Table IX reports F1 scores and the text says 'hard and soft F1 scores'; the metric should be stated consistently.
  2. [Figure 1] Figure 1 is not legible in the supplied manuscript: it appears as a long run of encoded tokens rather than a bar chart, so the win/tie/loss rates cannot be checked.
  3. [Throughout] There are inconsistencies in naming and formatting: 'STATE TOXICN' and 'STATE ToxiCN' are used interchangeably, and reference [16] appears as '[16]Zhou et al.' without a space.
  4. [Section V-C] The term 'weighted averaging' is unexplained; if multiple runs or learning rates are averaged, the procedure should be described.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: all central claims are empirical evaluations of new resources against external benchmarks, and no prediction is derived from its own input by construction.

full rationale

The paper makes three contributions: a span-level dataset (STATE ToxiCN), a Chinese coded hate lexicon, and a two-stage fine-tuning framework. The first two are resource-construction claims, not derivations. STATE ToxiCN is built by filtering and re-annotating the authors' own TOXICN dataset (Section III-B), and the lexicon is annotated from the same posts (Section III-C-3-b); this is transparent reuse of a previously published corpus, not a self-referential derivation. The third claim is validated on two external benchmarks, COLD and CDial-Bias (Section VIII), specifically chosen 'to avoid label leakage,' so the lexicon is not evaluated on labels that produced it. The only load-bearing citation to prior work by overlapping authors is TOXICN [11], which is used as the data source for further annotation rather than as an authority that forces the paper's conclusions. Section V-C's selection of the learning rate by highest test-set F1 is a methodological weakness that inflates reported Task 1 numbers, but it does not make any reported quantity equal to an input by construction, and Table XI for the two-stage experiments fixes hyperparameters rather than searching them. The observed contradiction in Table X (Only Lexicon F1 71.19 on COLD exceeds Two Stage F1 70.57, despite Section VIII-b claiming the two-stage framework 'consistently outperforms') is an internal consistency and statistical-support problem, not a circularity. For a purely empirical paper with no formal derivation chain, the appropriate circularity finding is 0.

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

No theoretical entities are invented; the ledger captures the dataset construction choices and domain assumptions the claims rest on. The paper's main free parameters are evaluation thresholds and hyperparameters, not fitted physical constants.

free parameters (4)
  • soft_match_threshold = 0.5
    Predictions count as correct when Target and Argument scores reach 0.5 (Section V-B); this hand-set threshold directly affects all soft-match F1 numbers.
  • learning_rate_task1 = 1e-5 to 5e-5 (selected by test F1)
    Section V-C: a range of learning rates was explored and the one yielding the highest F1 on the test set was chosen, creating a risk of test-set overfitting.
  • epochs = 10
    Section V-C and Table VI: fixed at 10 epochs because performance 'typically stabilizes', a manual choice affecting results.
  • two_stage_learning_rates = 5e-5 then 2e-5
    Table XI: stage-specific learning rates chosen without a validation-based justification.
assumptions (4)
  • domain assumption Posts from Zhihu and Tieba filtered from TOXICN can be meaningfully annotated for target and argument spans.
    Section III-B/C: the whole STATE ToxiCN dataset rests on the assumption that span-level hate structure is identifiable in these posts.
  • domain assumption The two-way categorization of coded hate terms (phono-graphemic variation vs. hateful semantic) is the right linguistic taxonomy for model training.
    Section III-C and VII: the lexicon and Task 2/3 evaluations depend on this partition; no independent evidence of exhaustiveness is given.
  • domain assumption Lexicon knowledge transfers to other hate/offensive detection datasets without label leakage.
    Section VIII: the two-stage framework's improvement assumes the lexicon learned on STATE ToxiCN posts generalizes to COLD and CDial-Bias.
  • domain assumption Manual filtering decisions by annotators do not systematically remove hard hate speech cases.
    Section III-B: Stage 2 manual filtering deletes samples judged to lack Target-Argument spans; if this correlates with difficulty, the dataset becomes easier than real-world hate speech.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Fine-Grained Chinese Hate Speech Understanding: Span-Level Resources, Coded Term Lexicon, and Enhanced Detection Frameworks." pith.science (2026). https://pith.science/paper/GEUIURJ6

@misc{pith2026250711292,
  author       = {Pith},
  title        = {Pith review of: Fine-Grained Chinese Hate Speech Understanding: Span-Level Resources, Coded Term Lexicon, and Enhanced Detection Frameworks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GEUIURJ6}},
  note         = {Machine review of arXiv:2507.11292}
}
read the original abstract

The proliferation of hate speech has inflicted significant societal harm, with its intensity and directionality closely tied to specific targets and arguments. In recent years, numerous machine learning-based methods have been developed to detect hateful comments on online platforms automatically. However, research on Chinese hate speech detection lags behind, and interpretability studies face two major challenges: first, the scarcity of span-level fine-grained annotated datasets limits models' deep semantic understanding of hate speech; second, insufficient research on identifying and interpreting coded hate speech restricts model explainability in complex real-world scenarios. To address these, we make the following contributions: (1) We introduce the Span-level Target-Aware Toxicity Extraction dataset (STATE ToxiCN), the first span-level Chinese hate speech dataset, and evaluate the hate semantic understanding of existing models using it. (2) We conduct the first comprehensive study on Chinese coded hate terms, LLMs' ability to interpret hate semantics. (3) We propose a method to integrate an annotated lexicon into models, significantly enhancing hate speech detection performance. Our work provides valuable resources and insights to advance the interpretability of Chinese hate speech detection research.

Figures

Figures reproduced from arXiv: 2507.11292 by the authors.

Figure 1
Figure 1. Results of pairwise comparison for coded hate term explanations. [PITH_FULL_IMAGE:figures/full_fig_p009_1.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

49 extracted references · 37 canonical work pages

  1. [1]

    Hate speech epidemic. the dynamic effects of derogatory language on intergroup relations and political radicalization,

    M. Bilewicz and W. Soral, “Hate speech epidemic. the dynamic effects of derogatory language on intergroup relations and political radicalization,” Political Psychology, vol. 41, pp. 3–33, 2020

    work page 2020

  2. [2]

    Analyzing the targets of hate in online social media,

    L. Silva, M. Mondal, D. Correa, F. Benevenuto, and I. Weber, “Analyzing the targets of hate in online social media,” in Proceedings of the International AAAI Conference on Web and Social Media , vol. 10, no. 1, 2016, pp. 687–690

    work page 2016

  3. [3]

    Sharedcon: Implicit hate speech detection using shared semantics,

    H. Ahn, Y . Kim, J. Kim, and Y .-S. Han, “Sharedcon: Implicit hate speech detection using shared semantics,” in Findings of the Association for Computational Linguistics ACL 2024 , 2024, pp. 10 444–10 455

    work page 2024

  4. [4]

    ToKen: Task decomposition and knowledge infusion for few-shot hate speech detection,

    B. AlKhamissi, F. Ladhak, S. Iyer, V . Stoyanov, Z. Kozareva, X. Li, P. Fung, L. Mathias, A. Celikyilmaz, and M. Diab, “ToKen: Task decomposition and knowledge infusion for few-shot hate speech detection,” in Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing , Y . Goldberg, Z. Kozareva, and Y . Zhang, Eds. Abu Dhabi, U...

    work page 2022

  5. [5]

    Semeval-2021 task 5: Toxic spans detection,

    J. Pavlopoulos, J. Sorensen, L. Laugier, and I. Androutsopoulos, “Semeval-2021 task 5: Toxic spans detection,” in Proceedings of the 15th international workshop on semantic evaluation (SemEval-2021) , 2021, pp. 59–69

    work page 2021

  6. [6]

    Hatexplain: A benchmark dataset for explainable hate speech detection,

    B. Mathew, P. Saha, S. M. Yimam, C. Biemann, P. Goyal, and A. Mukherjee, “Hatexplain: A benchmark dataset for explainable hate speech detection,” in Proceedings of the AAAI conference on artificial intelligence, vol. 35, no. 17, 2021, pp. 14 867–14 875

    work page 2021

  7. [7]

    Target-based offensive language identification,

    M. Zampieri, S. Morgan, K. North, T. Ranasinghe, A. Simmons, P. Khandelwal, S. Rosenthal, and P. Nakov, “Target-based offensive language identification,” 2023

    work page 2023

  8. [8]

    D. M. Eberhard, G. F. Simons, and C. D. Fennig, Ethnologue: Languages of the World, twenty-seventh ed. Dallas, Texas: SIL International, 2024, online version available at http://www.ethnologue.com

    work page 2024

Show all 49 references
  1. [9]

    What is a chinese

    V . H. Mair, “What is a chinese" dialect/topolect"?: Reflections on some key sino-english linguistic terms,” 1991

    1991

  2. [10]

    ToxiCloakCN: Evaluating robustness of offensive language detection in Chinese with cloaking perturbations,

    Y . Xiao, Y . Hu, K. T. W. Choo, and R. K.-W. Lee, “ToxiCloakCN: Evaluating robustness of offensive language detection in Chinese with cloaking perturbations,” in Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing , Y . Al-Onaizan, M. Bansal...

    2024

  3. [11]

    Facilitating fine-grained detection of Chinese toxic language: Hierarchical taxonomy, resources, and benchmarks,

    J. Lu, B. Xu, X. Zhang, C. Min, L. Yang, and H. Lin, “Facilitating fine-grained detection of Chinese toxic language: Hierarchical taxonomy, resources, and benchmarks,” in Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Pa...

    2023

  4. [12]

    Automated hate speech detection and the problem of offensive language,

    T. Davidson, D. Warmsley, M. Macy, and I. Weber, “Automated hate speech detection and the problem of offensive language,” in Proceedings of the international AAAI conference on web and social media , vol. 11, no. 1, 2017, pp. 512–515

    2017

  5. [13]

    Large scale crowdsourcing and characterization of twitter abusive behavior,

    A. Founta, C. Djouvas, D. Chatzakou, I. Leontiadis, J. Blackburn, G. Stringhini, A. Vakali, M. Sirivianos, and N. Kourtellis, “Large scale crowdsourcing and characterization of twitter abusive behavior,” in Proceedings of the international AAAI conference on web and social med...

    2018

  6. [14]

    Cold: A benchmark for chinese offensive language detection,

    J. Deng, J. Zhou, H. Sun, C. Zheng, F. Mi, H. Meng, and M. Huang, “Cold: A benchmark for chinese offensive language detection,” arXiv preprint arXiv:2201.06025, 2022

    2022 arXiv

  7. [15]

    Swsr: A chinese dataset and lexicon for online sexism detection,

    A. Jiang, X. Yang, Y . Liu, and A. Zubiaga, “Swsr: A chinese dataset and lexicon for online sexism detection,” Online Social Networks and Media, vol. 27, p. 100182, 2022

    2022

  8. [16]

    Towards identifying social bias in dialog systems: Framework, dataset, and benchmark,

    J. Zhou, J. Deng, F. Mi, Y . Li, Y . Wang, M. Huang, X. Jiang, Q. Liu, and H. Meng, “Towards identifying social bias in dialog systems: Framework, dataset, and benchmark,” in Findings of the Association for Computational Linguistics: EMNLP 2022 , 2022, pp. 3576–3591

    2022

  9. [17]

    Hatebert: Re- training bert for abusive language detection in english,

    T. Caselli, V . Basile, J. Mitrovi ´c, and M. Granitzer, “Hatebert: Re- training bert for abusive language detection in english,” arXiv preprint arXiv:2010.12472, 2020

    2010 arXiv

  10. [18]

    Detoxify,

    L. Hanu and Unitary team, “Detoxify,” Github. https://github.com/unitaryai/detoxify, 2020

    2020

  11. [19]

    Hate speech detection based on sentiment knowledge sharing,

    X. Zhou, Y . Yong, X. Fan, G. Ren, Y . Song, Y . Diao, L. Yang, and H. Lin, “Hate speech detection based on sentiment knowledge sharing,” in Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on N...

    2021

  12. [20]

    Hate speech detection on twitter using transfer learning,

    R. Ali, U. Farooq, U. Arshad, W. Shahzad, and M. O. Beg, “Hate speech detection on twitter using transfer learning,” Computer Speech & Language, vol. 74, p. 101365, 2022

    2022

  13. [21]

    Hateful symbols or hateful people? predictive features for hate speech detection on twitter,

    Z. Waseem and D. Hovy, “Hateful symbols or hateful people? predictive features for hate speech detection on twitter,” in Proceedings of the NAACL student research workshop , 2016, pp. 88–93

    2016

  14. [22]

    Toxigen: A large-scale machine-generated dataset for adversarial and implicit hate speech detection,

    T. Hartvigsen, S. Gabriel, H. Palangi, M. Sap, D. Ray, and E. Kamar, “Toxigen: A large-scale machine-generated dataset for adversarial and implicit hate speech detection,” arXiv preprint arXiv:2203.09509 , 2022

    2022 arXiv

  15. [23]

    Pars-off: a benchmark for offensive language detection on farsi social media,

    T. S. Ataei, K. Darvishi, S. Javdan, A. Pourdabiri, B. Minaei-Bidgoli, and M. T. Pilehvar, “Pars-off: a benchmark for offensive language detection on farsi social media,” IEEE Transactions on Affective Computing, vol. 14, no. 4, pp. 2787–2795, 2022

    2022

  16. [24]

    Tocab: A dataset for chinese abusive language processing,

    I. Chung and C.-J. Lin, “Tocab: A dataset for chinese abusive language processing,” in 2021 IEEE 22nd International Conference on Information Reuse and Integration for Data Science (IRI) . IEEE, 2021, pp. 445–452

    2021

  17. [25]

    Symbolic racism,

    J. B. McConahay and J. C. Hough Jr, “Symbolic racism,” Journal of social issues, vol. 32, no. 2, pp. 23–45, 1976

    1976

  18. [26]

    D. P. Macedo, L. I. Bartolomé, and D. Macedo, Dancing with bigotry: Beyond the politics of tolerance . Springer, 1999

    1999

  19. [27]

    Regulating the reproduction and mothering of poor women: The controlling image of the welfare mother in television news coverage of welfare reform,

    M. Kelly, “Regulating the reproduction and mothering of poor women: The controlling image of the welfare mother in television news coverage of welfare reform,” Journal of Poverty, vol. 14, no. 1, pp. 76–96, 2010

    2010

  20. [28]

    Humour and hatred: The racist jokes of the ku klux klan,

    M. Billig, “Humour and hatred: The racist jokes of the ku klux klan,” Discourse & Society , vol. 12, no. 3, pp. 267–289, 2001

    2001

  21. [29]

    Hatemoji: A test suite and adversarially-generated dataset for benchmarking and detecting emoji-based hate,

    H. R. Kirk, B. Vidgen, P. Röttger, T. Thrush, and S. A. Hale, “Hatemoji: A test suite and adversarially-generated dataset for benchmarking and detecting emoji-based hate,” arXiv preprint arXiv:2108.05921 , 2021

    2021 arXiv

  22. [30]

    BAE: BERT-based adversarial examples for text classification,

    S. Garg and G. Ramakrishnan, “BAE: BERT-based adversarial examples for text classification,” in Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP) , B. Webber, T. Cohn, Y . He, and Y . Liu, Eds. Online: Association for Computational ...

    2020

  23. [31]

    Generating natural language adversarial examples through probability weighted word saliency,

    S. Ren, Y . Deng, K. He, and W. Che, “Generating natural language adversarial examples through probability weighted word saliency,” in Proceedings of the 57th annual meeting of the association for computa- tional linguistics, 2019, pp. 1085–1097

    2019

  24. [32]

    Rocbert: Robust chinese bert with multimodal contrastive pretraining,

    H. Su, W. Shi, X. Shen, Z. Xiao, T. Ji, J. Fang, and J. Zhou, “Rocbert: Robust chinese bert with multimodal contrastive pretraining,” in Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , 2022, pp. 921– 931

    2022

  25. [33]

    Measuring nominal scale agreement among many raters

    J. L. Fleiss, “Measuring nominal scale agreement among many raters.” Psychological bulletin, vol. 76, no. 5, p. 378, 1971

    1971

  26. [34]

    mt5: A massively multilingual pre-trained text-to-text trans- former,

    L. Xue, “mt5: A massively multilingual pre-trained text-to-text trans- former,” arXiv preprint arXiv:2010.11934 , 2020

    2010 arXiv

  27. [35]

    Mistral 7b,

    A. Q. Jiang, A. Sablayrolles, A. Mensch, C. Bamford, D. S. Chaplot, D. d. l. Casas, F. Bressand, G. Lengyel, G. Lample, L. Saulnier et al., “Mistral 7b,” arXiv preprint arXiv:2310.06825 , 2023

    2023 arXiv

  28. [36]

    Llama 3.3 model card,

    AI@Meta, “Llama 3.3 model card,” 2024. [Online]. Avail- able: https://github.com/meta-llama/llama-models/blob/main/models/ llama3_3/MODEL_CARD.md

    2024

  29. [37]

    Qwen2.5: A party of foundation models,

    Q. Team, “Qwen2.5: A party of foundation models,” September 2024. [Online]. Available: https://qwenlm.github.io/blog/qwen2.5/ 12

    2024

  30. [38]

    Shieldlm: Empowering llms as aligned, customizable and explainable safety detectors,

    Z. Zhang, Y . Lu, J. Ma, D. Zhang, R. Li, P. Ke, H. Sun, L. Sha, Z. Sui, H. Wang et al., “Shieldlm: Empowering llms as aligned, customizable and explainable safety detectors,” arXiv preprint arXiv:2402.16444, 2024

    2024 arXiv

  31. [39]

    Shieldgemma: Generative ai content moderation based on gemma,

    W. Zeng, Y . Liu, R. Mullins, L. Peran, J. Fernandez, H. Harkous, K. Narasimhan, D. Proud, P. Kumar, B. Radharapu, O. Sturman, and O. Wahltinez, “Shieldgemma: Generative ai content moderation based on gemma,” 2024. [Online]. Available: https://arxiv.org/abs/2407.21772

    2024 arXiv

  32. [40]

    Gemini 2.5 flash,

    G. DeepMind, “Gemini 2.5 flash,” https://deepmind.google/models/ gemini/flash/, June 2025

    2025

  33. [41]

    Claude 3.7 sonnet and claude code,

    Anthropic, “Claude 3.7 sonnet and claude code,” https://www.anthropic. com/news/claude-3-7-sonnet, June 2024

    2024

  34. [42]

    Hello gpt-4o,

    OpenAI, “Hello gpt-4o,” https://openai.com/index/hello-gpt-4o/, 2024

    2024

  35. [43]

    Deepseek-v3 technical report,

    A. Liu, B. Feng, B. Xue, B. Wang, B. Wu, C. Lu, C. Zhao, C. Deng, C. Zhang, C. Ruan et al., “Deepseek-v3 technical report,” arXiv preprint arXiv:2412.19437, 2024

    2024 arXiv

  36. [44]

    Is information extraction solved by chatgpt? an analysis of performance, evaluation criteria, robustness and errors,

    R. Han, T. Peng, C. Yang, B. Wang, L. Liu, and X. Wan, “Is information extraction solved by chatgpt? an analysis of performance, evaluation criteria, robustness and errors,” arXiv preprint arXiv:2305.14450 , 2023

    2023 arXiv

  37. [45]

    Toolllm: Facilitating large language models to master 16000+ real-world apis,

    Y . Qin, S. Liang, Y . Ye, K. Zhu, L. Yan, Y . Lu, Y . Lin, X. Cong, X. Tang, B. Qian et al., “Toolllm: Facilitating large language models to master 16000+ real-world apis,” arXiv preprint arXiv:2307.16789 , 2023

    2023 arXiv

  38. [46]

    Translate meanings, not just words: Idiomkb’s role in optimizing idiomatic translation with language models,

    S. Li, J. Chen, S. Yuan, X. Wu, H. Yang, S. Tao, and Y . Xiao, “Translate meanings, not just words: Idiomkb’s role in optimizing idiomatic translation with language models,” in Proceedings of the AAAI Conference on Artificial Intelligence , vol. 38, no. 17, 2024, pp. 18 554–18 563

    2024

  39. [47]

    “a good pun is its own reword

    Z. Xu, S. Yuan, L. Chen, and D. Yang, ““a good pun is its own reword”: Can large language models understand puns?” in Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing , Y . Al-Onaizan, M. Bansal, and Y .-N. Chen, Eds. Miami, Florida, USA: ...

    2024

  40. [48]

    Llamafactory: Unified efficient fine-tuning of 100+ language models,

    Y . Zheng, R. Zhang, J. Zhang, Y . Ye, Z. Luo, Z. Feng, and Y . Ma, “Llamafactory: Unified efficient fine-tuning of 100+ language models,” in Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 3: System Demonstrations) . Bangkok, Th...

    2024 arXiv

  41. [49]

    Overview of nlpcc 2022 shared task 7: Fine-grained dialogue social bias measurement,

    J. Zhou, F. Mi, H. Meng, and J. Deng, “Overview of nlpcc 2022 shared task 7: Fine-grained dialogue social bias measurement,” in CCF International Conference on Natural Language Processing and Chinese Computing. Springer, 2022, pp. 342–350. UTF8gbsn

    2022

Pith tools

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