Pith. sign in

REVIEW 1 major objections 4 minor 60 references

A Federated Approach to Few-Shot Hate Speech Detection for Marginalized Communities

T0 review · 1 major / 4 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read Federated learning lets a server and a few devices jointly train few-shot hate speech filters for low-resource languages, without raw user text leaving the device.

desk verdict REACT is a genuinely useful low-resource hate speech dataset; the 'FL consistently improves' claim overreaches its own Table 2 and should be softened before publication. read the letter →

arxiv 2412.04942 v2 pith:FPY5M6CV submitted 2024-12-06 cs.CL cs.AI

Haotian Ye , Axel Wisiorek , Antonis Maronikolakis , Özge Alaçam , Hinrich Schütze This is my paper · ORCID
classification cs.CLcs.AI
keywords federatedlearningfew-shothatespeechdetectionlow-resourcelanguagesprivacy-preservingNLPmarginalizedcommunitiesclientpersonalizationmultilinguallanguagemodels
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 sets out to give marginalized communities in low-resource language regions a privacy-preserving way to filter online hate speech on their own devices, using federated learning instead of sending user text to a central server. To support this, it releases REACT, a collection of localized hate speech datasets in Afrikaans, Korean, Russian, and Ukrainian, covering six target groups (Black people, LGBTQ people, Russians, Russophone Ukrainians, Ukrainian war victims, and women), curated by speakers who know the local context. The experiments show that with only 3–15 labeled sentences per target group, federated training on lightweight multilingual models (mBERT and Distil-mBERT) consistently improves client and server F1 over training each group alone, with the clearest gains at 9–15 examples. Two personalization mechanisms, FedPer and adapters, provide no consistent accuracy gain but keep performance comparable while retaining more parameters on the client. If true, this points toward on-device hate speech filters that can be trained collectively, in low-resource languages, without exposing the messages they protect.

What carries the argument

The engine is FederatedAveraging (FedAvg): each client fine-tunes a shared multilingual Transformer on its own few-shot examples and sends only weight updates to a server, which averages them back into the shared model; the paper uses one client per language-target group (Afrikaans-Black, Afrikaans-LGBTQ, Russian-LGBTQ, Russian-war victims). Compact multilingual encoders, mBERT and Distil-mBERT, carry the representations, and a Levenshtein-ratio filter keeps near-duplicate training and test sentences from inflating evaluation. FedPer personalizes by keeping the classifier head and top Transformer layers client-private, while the adapter variant inserts small trainable blocks between Transformer layers; both are evaluated as alternatives to full sharing.

What would settle it

Run the same FL-versus-single-target comparison on a fresh set of unmodified, naturally occurring social-media posts in Afrikaans and Russian with 9 labeled sentences per target group; the claim would be contradicted if FL's average macro-F1 edge over single-target fine-tuning is no larger than seed-level variation.

Watch

Extended reading notes

Core claim

The paper's central empirical discovery is that federated learning improves few-shot hate speech detection for each participating client: across the four language-target pairs tested, FL raises macro-F1 over single-target fine-tuning most consistently when clients have 9–15 labeled sentences, and the aggregated server model also improves. The same experiments show that a standard centralized scorer, Perspective API, works reasonably on Russian but poorly on unsupported Afrikaans and misses culturally specific slurs, while client personalization via FedPer or adapters gives no consistent F1 gain over standard FL, although it keeps performance roughly equal and leaves more parameters private.

Load-bearing premise

The load-bearing premise is that REACT's sentences—including AI-generated and editorially modified examples, which reach 25–35% of the two Ukrainian subsets—represent naturally occurring online hate speech, because every FL conclusion is trained and tested on that mix.

Editorial extensions

If this is right

  • With only a handful of labeled examples per group, a shared hate speech filter can be trained across languages without raw user messages being uploaded.
  • The server model gains from the same federated rounds, so one aggregated filter can serve multiple target groups and languages.
  • Centralized toxicity scoring is not a sufficient substitute in low-resource settings, since it underperforms on unsupported languages and misses community-specific slurs.
  • Client personalization can be offered for privacy without a meaningful accuracy penalty, even though it does not consistently beat standard FL.
  • The released REACT dataset gives future work six-category, locally grounded material for Afrikaans, Korean, Russian, and Ukrainian.

Reading between the lines

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

  • Beyond the paper: the same federated few-shot recipe could be tried on other subjective classification tasks in low-resource settings, such as targeted harassment or mis- and disinformation, where per-community variation is large.
  • Beyond the paper: because two Ukrainian subsets are 25–35% AI-generated despite the text saying 'under 20%', users of REACT should re-validate on naturally occurring organic posts before trusting the reported effect sizes in deployment.
  • Beyond the paper: the paper's own stated limits—four simulated clients, no hyperparameter search, no real-device deployment—suggest the headline FL gains should be rechecked on-device before productizing.
  • Beyond the paper: a testable prediction is that with more, more heterogeneous clients, FedPer and adapter personalization will outperform standard FL, since the current null result may reflect the small client pool.
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

1 major / 4 minor

Summary. The paper introduces REACT, a collection of culturally specific hate speech detection datasets in low-resource languages (Afrikaans, Korean, Russian, Ukrainian) covering marginalized target groups, and evaluates a federated learning (FL) approach for few-shot hate speech detection. The authors compare FL-trained client and server models against single-target fine-tuning and the Perspective API, and additionally study two personalization methods (FedPer and adapters). The central claims are that FL consistently improves client and server performance, especially with 9–15 training samples, and that personalization is a promising direction.

Significance. The REACT dataset is a useful and timely resource: it targets marginalized communities in low-resource languages, involves native speakers and cross-annotation, and is released under a permissive license. The paper also tackles the important practical question of whether privacy-preserving FL can help in extremely low-resource hate speech detection. If the FL claim were rigorously supported, the contribution would be significant for content moderation on-device in the Global South. However, the manuscript's own reported numbers do not support the 'consistently improves' claim, and the personalization evaluation is partly circular because the optimal FedPer configuration is selected on the evaluation data. The dataset and the careful framing of limitations are genuine strengths, but the empirical conclusions require substantial revision before they can be accepted.

major comments (1)
  1. [§3 and Table 7] The text says 'a portion of the data (under 20% for most datasets) is generated using AI tools such as ChatGPT,' but Table 7 shows that two Ukrainian datasets contain 25.0% and 35.0% AI-generated sentences, contradicting the 'under 20%' phrasing. More fundamentally, the paper provides no evidence that AI-generated or curator-edited examples are representative of naturally occurring online hate speech; §H shows substantial editorial modification of examples. Since all model comparisons and FL conclusions in this paper are trained and evaluated on REACT, this is a validity concern for the study's external conclusions. The authors should report the fraction of AI-generated sentences in each train/dev/test split, analyze whether model performance differs on AI-generated versus naturally collected examples, and adjust the dataset description and claims accordingly.
minor comments (4)
  1. [Table 1 caption] The caption says 'In total, the data covers seven distinct target groups in eight languages,' but the table lists four languages (Afrikaans, Ukrainian, Russian, Korean) and six distinct target groups (Black people, LGBTQ, Russians, Russophones, War victims, Women); this should be corrected.
  2. [§1, RQ1] There is a typo: 'low-resourse' should be 'low-resource.'
  3. [§4.2] The description of Levenshtein filtering says the threshold is relaxed when the split is too small, but the exact relaxation rule and how manual checking was used for rus-lgbtq and rus-war is only described in the appendix; consider moving this important detail to the main text, since it affects the train/test overlap and therefore the validity of the comparisons.
  4. [§6, Table 10] The analysis of Perspective API thresholds is interesting, but the aggregate percentages in Table 10 would benefit from a breakdown by target group/language, since the main text reports that the API performs well on Russian and poorly on Afrikaans.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the FL-versus-baseline and personalization claims rest on direct held-out experiments, not on definitions, fitted predictions, or load-bearing self-citations.

full rationale

The paper's central claims are empirical: FL yields modest client/server F1 improvements over single-target fine-tuning and Perspective API, and personalization has context-dependent effects. These are evaluated on held-out test sets with multiple seeds and a fixed protocol; no parameter is fitted to the test outcome and then reported as a prediction. The REACT dataset is an external contribution, not derived from the conclusions. The only selection-on-outcome element is the FedPer 'optimal KP' (KP=4), chosen as the value with the highest average F1 improvement per client across training sizes; however, the paper transparently reports full per-KP results (Figures 5-6) and average improvements for all KP values in Table 3, with hedged conclusions, so this is a methodological weakness rather than a circular derivation. Existing self-citations (Maronikolakis et al. 2022, Alacam et al. 2024) appear only as related-work background and are not load-bearing. No equation, definition, or fitted parameter is equivalent by construction to a claimed prediction. Therefore no significant circularity is present.

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

The experiments rest on assumptions about data representativeness, annotation quality, simulation fidelity, and overlap control. The central claim depends on the representative-data assumption more than on any mathematical axiom. Free parameters are limited to hand-set evaluation and split choices, most importantly KP selected on test data.

free parameters (4)
  • Levenshtein similarity thresholds = default 0.5; relaxed to 0.7 for rus-lgbtq dev and 0.6 for rus-war train
    Hand-set in Appendix D to reduce train-test overlap; when splits are too small the thresholds are relaxed with manual checks, changing the evaluation set.
  • FedPer KP = KP=4 for both models (best of {1,2,3,4})
    Chosen in Section 5 as the value with the highest average F1 improvement per client across the four training sizes. Because the choice uses the test results that are then reported, it is a post hoc selection.
  • Perspective API toxicity thresholds = 0.7 and 0.9
    Baseline thresholds from the API's recommended range; results are threshold-sensitive, so they are an evaluation choice rather than a fit.
  • FL rounds and local epochs = 5 rounds, 1 local epoch per round; baseline 5 epochs
    Hand-set in Section 4.2. The comparison is only fair if 5 FL rounds equal 5 baseline epochs in effective training; no convergence analysis is given.
assumptions (4)
  • domain assumption Data collector annotations are a valid gold standard for hate speech.
    Inter-annotator agreement in Table 5 is moderate to substantial for Afrikaans, but labels are used as ground truth for training and testing all models.
  • ad hoc to paper AI-generated and curator-edited examples are representative of real online hate speech.
    Section 3 and Table 7: up to 35% of some datasets is ChatGPT-generated; no validation against naturally occurring hate speech is provided.
  • domain assumption Flower simulation with four clients on a single machine reproduces real federated learning behavior.
    Section 4.2; the Limitations section admits experiments are simulated and real deployment is future work.
  • domain assumption Levenshtein-ratio filtering prevents meaningful train-test overlap.
    Appendix D; thresholds are relaxed when splits are small, and no alternative leakage check is reported.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Federated Approach to Few-Shot Hate Speech Detection for Marginalized Communities." pith.science (2026). https://pith.science/paper/FPY5M6CV

@misc{pith2026241204942,
  author       = {Pith},
  title        = {Pith review of: A Federated Approach to Few-Shot Hate Speech Detection for Marginalized Communities},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FPY5M6CV}},
  note         = {Machine review of arXiv:2412.04942}
}
read the original abstract

Hate speech online remains an understudied issue for marginalized communities, particularly in the Global South, which includes developing societies with increasing internet penetration. In this paper, we aim to provide marginalized communities in societies where the dominant language is low-resource with a privacy-preserving tool to protect themselves from online hate speech by filtering offensive content in their native languages. Our contributions are twofold: 1) we release REACT (REsponsive hate speech datasets Across ConTexts), a collection of high-quality, culture-specific hate speech detection datasets comprising multiple target groups and low-resource languages, curated by experienced data collectors; 2) we propose a few-shot hate speech detection approach based on federated learning (FL), a privacy-preserving method for collaboratively training a central model that exhibits robustness when tackling different target groups and languages. By keeping training local to user devices, we ensure data privacy while leveraging the collective learning benefits of FL. Furthermore, we explore personalized client models tailored to specific target groups and evaluate their performance. Our findings indicate the overall effectiveness of FL across different target groups, and point to personalization as a promising direction.

Figures

Figures reproduced from arXiv: 2412.04942 by the authors.

Figure 1
Figure 1. Comparison of F1 scores using mBERT and Distil-mBERT across three training settings: FL (solid lines), single-target training (dashed lines), and Perspective API (horizontal dotted lines). Each subplot illustrates performance on a specific target group or the server. FL consistently improves client and server performance, especially with more (9-15) training samples. Training afr-black afr-lgbtq rus-lgbtq rus-war se… view at source ↗
Figure 3
Figure 3. F1 scores of client models customized us￾ing adapters and full-model fine-tuning (dashed lines), compared against those trained with standard FL (solid lines). Although a few clients see gains from adapter￾based personalization, the overall improvement is un￾clear. For adapter-based personalization, we find that full-model fine-tuning consistently outperforms adapter-only fine-tuning [PITH_FULL_IMAGE:figures/full_f… view at source ↗
Figure 4
Figure 4. Comparison of F1 scores of seven models, four multilingual and three monolingual. Each subplot shows performance on a specific target group or the server. The three monolingual models and multilingual MiniLM perform poorly across all target groups. Multilingual BERT and Distil-mBERT have the highest performance in most cases [PITH_FULL_IMAGE:figures/full_fig_p015_4.png] view at source ↗
Figures from the paper (3 more)
Figure 5
Figure 5. Figure 5: FedPer results for mBERT. Each plot shows [PITH_FULL_IMAGE:figures/full_fig_p017_5.png]
Figure 6
Figure 6. Figure 6: FedPer results for Distil-mBERT. Each plot shows [PITH_FULL_IMAGE:figures/full_fig_p017_6.png]
Figure 7
Figure 7. Figure 7: Adapter-based personalization results for mBERT and Distil-mBERT. Results are compared between [PITH_FULL_IMAGE:figures/full_fig_p017_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

60 extracted references · 19 canonical work pages

  1. [1]

    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.all mid.sentence after.sentence after.block STRINGS urlintro eprinturl eprintpr...

  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]

    \"O zge Alacam, Sanne Hoeken, and Sina Zarrie . 2024. Eyes don ' t lie: Subjective hate annotation and detection with gaze. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, Miami, United States. Association for Computational Linguistics

    work page 2024

  4. [4]

    Manoj Ghuhan Arivazhagan, Vinay Aggarwal, Aaditya Kumar Singh, and Sunav Choudhary. 2019. Federated learning with personalization layers. arXiv preprint arXiv:1912.00818

    arXiv 2019

  5. [5]

    Keith Bonawitz, Hubert Eichner, Wolfgang Grieskamp, Dzmitry Huba, Alex Ingerman, Vladimir Ivanov, Chlo\' e Kiddon, Jakub Kone c n\' y , Stefano Mazzocchi, Brendan McMahan, Timon Van Overveldt, David Petrou, Daniel Ramage, and Jason Roselander. 2019. https://proceedings.mlsys.org/paper_files/paper/2019/file/7b770da633baf74895be22a8807f1a8f-Paper.pdf Toward...

    work page 2019

  6. [6]

    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

  7. [7]

    David Byrd and Antigoni Polychroniadou. 2020. https://doi.org/10.1145/3383455.3422562 Differentially private secure multi-party computation for federated learning in financial applications . In ICAIF '20: The First ACM International Conference on AI in Finance, New York, NY, USA, October 15-16, 2020 , pages 16:1--16:9. ACM

    arXiv 2020

  8. [8]

    Tommaso Caselli, Valerio Basile, Jelena Mitrovi \'c , and Michael Granitzer. 2021. https://doi.org/10.18653/v1/2021.woah-1.3 H ate BERT : Retraining BERT for abusive language detection in E nglish . In Proceedings of the 5th Workshop on Online Abuse and Harms (WOAH 2021), pages 17--25, Online. Association for Computational Linguistics

    doi:10.18653/v1/2021.woah-1.3 2021

Show all 60 references
  1. [9]

    Alexis Conneau, Kartikay Khandelwal, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzm \'a n, Edouard Grave, Myle Ott, Luke Zettlemoyer, and Veselin Stoyanov. 2020. https://doi.org/10.18653/v1/2020.acl-main.747 Unsupervised cross-lingual representation learning ...

    2020 doi

  2. [10]

    Mithun Das, Somnath Banerjee, Punyajoy Saha, and Animesh Mukherjee. 2022. https://aclanthology.org/2022.aacl-main.23 Hate speech and offensive language detection in B engali . In Proceedings of the 2nd Conference of the Asia-Pacific Chapter of the Association for Computational...

    2022

  3. [11]

    Thomas Davidson, Debasmita Bhattacharya, and Ingmar Weber. 2019. https://doi.org/10.18653/v1/W19-3504 Racial bias in hate speech and abusive language detection datasets . In Proceedings of the Third Workshop on Abusive Language Online, pages 25--35, Florence, Italy. Associatio...

    2019 doi

  4. [12]

    Daryna Dementieva, Valeriia Khylenko, and Georg Groh. 2024. Ukrainian texts classification: Exploration of cross-lingual knowledge transfer approaches. arXiv preprint arXiv:2404.02043

    2024 arXiv

  5. [13]

    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 Proceedings of the 2019 Conference of the North A merican Chapter of the Associat...

    2019 doi

  6. [14]

    Lucas Dixon, John Li, Jeffrey Sorensen, Nithum Thain, and Lucy Vasserman. 2018. https://doi.org/10.1145/3278721.3278729 Measuring and mitigating unintended bias in text classification . In Proceedings of the 2018 AAAI/ACM Conference on AI, Ethics, and Society, AIES 2018, New O...

    2018

  7. [15]

    Eve Fleisig, Rediet Abebe, and Dan Klein. 2023. https://doi.org/10.18653/v1/2023.emnlp-main.415 When the majority is wrong: Modeling annotator disagreement for subjective tasks . In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 6...

    2023 doi

  8. [16]

    Jay Gala, Deep Gandhi, Jash Mehta, and Zeerak Talat. 2023. https://doi.org/10.18653/v1/2023.eacl-main.237 A federated approach for hate speech detection . In Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics, pages 3248...

    2023 doi

  9. [17]

    Tanmay Garg, Sarah Masud, Tharun Suresh, and Tanmoy Chakraborty. 2023. https://doi.org/10.1145/3580494 Handling bias in toxic speech detection: A survey . ACM Comput. Surv. , 55(13s):264:1--264:32

    2023 doi

  10. [18]

    Lara Grimminger and Roman Klinger. 2021. https://aclanthology.org/2021.wassa-1.18 Hate towards the political opponent: A T witter corpus study of the 2020 US elections on the basis of offensive speech and stance detection . In Proceedings of the Eleventh Workshop on Computatio...

    2021

  11. [19]

    Ella Guest, Bertie Vidgen, Alexandros Mittos, Nishanth Sastry, Gareth Tyson, and Helen Margetts. 2021. https://doi.org/10.18653/v1/2021.eacl-main.114 An expert annotated dataset for the detection of online misogyny . In Proceedings of the 16th Conference of the European Chapte...

    2021 doi

  12. [20]

    Lee, Gu - Yeon Wei, David Brooks, and Carole - Jean Wu

    Udit Gupta, Young Geun Kim, Sylvia Lee, Jordan Tse, Hsien - Hsin S. Lee, Gu - Yeon Wei, David Brooks, and Carole - Jean Wu. 2022. https://doi.org/10.1109/MM.2022.3163226 Chasing carbon: The elusive environmental footprint of computing . IEEE Micro , 42(4):37--47

    2022

  13. [21]

    Andrew Hard, Kanishka Rao, Rajiv Mathews, Swaroop Ramaswamy, Fran c oise Beaufays, Sean Augenstein, Hubert Eichner, Chlo \'e Kiddon, and Daniel Ramage. 2018. Federated learning for mobile keyboard prediction. arXiv preprint arXiv:1811.03604

    2018 arXiv

  14. [22]

    Sanne Hoeken, Sina Zarriess, and "O zge Alacam. 2024. Hateful word in context classification. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, Miami, Florida, USA. Association for Computational Linguistics

    2024

  15. [23]

    Neil Houlsby, Andrei Giurgiu, Stanislaw Jastrzebski, Bruna Morrone, Quentin de Laroussilhe, Andrea Gesmundo, Mona Attariyan, and Sylvain Gelly. 2019. http://proceedings.mlr.press/v97/houlsby19a.html Parameter-efficient transfer learning for NLP . In Proceedings of the 36th Int...

    2019

  16. [24]

    Yiqing Hua, Mor Naaman, and Thomas Ristenpart. 2020. https://doi.org/10.1145/3313831.3376548 Characterizing twitter users who engage in adversarial interactions against political candidates . In CHI '20: CHI Conference on Human Factors in Computing Systems, Honolulu, HI, USA, ...

    2020

  17. [25]

    Xiaoqi Jiao, Yichun Yin, Lifeng Shang, Xin Jiang, Xiao Chen, Linlin Li, Fang Wang, and Qun Liu. 2020. https://doi.org/10.18653/v1/2020.findings-emnlp.372 T iny BERT : Distilling BERT for natural language understanding . In Findings of the Association for Computational Linguist...

    2020 doi

  18. [26]

    Kamil Kanclerz, Marcin Gruza, Konrad Karanowski, Julita Bielaniewicz, Piotr Milkowski, Jan Kocon, and Przemyslaw Kazienko. 2022. https://aclanthology.org/2022.nlperspectives-1.6 What if ground truth is subjective? personalized deep neural hate speech detection . In Proceedings...

    2022

  19. [27]

    Reddi, Sebastian U

    Sai Praneeth Karimireddy, Satyen Kale, Mehryar Mohri, Sashank J. Reddi, Sebastian U. Stich, and Ananda Theertha Suresh. 2020. http://proceedings.mlr.press/v119/karimireddy20a.html SCAFFOLD: stochastic controlled averaging for federated learning . In Proceedings of the 37th Int...

    2020

  20. [28]

    Siwon Kim, Sangdoo Yun, Hwaran Lee, Martin Gubri, Sungroh Yoon, and Seong Joon Oh. 2023. http://papers.nips.cc/paper\_files/paper/2023/hash/420678bb4c8251ab30e765bc27c3b047-Abstract-Conference.html Propile: Probing privacy leakage in large language models . In Advances in Neur...

    2023

  21. [29]

    Zhenzhong Lan, Mingda Chen, Sebastian Goodman, Kevin Gimpel, Piyush Sharma, and Radu Soricut. 2020. https://openreview.net/forum?id=H1eA7AEtvS ALBERT: A lite BERT for self-supervised learning of language representations . In 8th International Conference on Learning Representat...

    2020

  22. [30]

    Xiang Li, Kaixuan Huang, Wenhao Yang, Shusen Wang, and Zhihua Zhang. 2020. https://openreview.net/forum?id=HJxNAnVtDS On the convergence of fedavg on non-iid data . In 8th International Conference on Learning Representations, ICLR 2020, Addis Ababa, Ethiopia, April 26-30, 2020...

    2020

  23. [31]

    Nils Lukas, Ahmed Salem, Robert Sim, Shruti Tople, Lukas Wutschitz, and Santiago Zanella B \' e guelin. 2023. https://doi.org/10.1109/SP46215.2023.10179300 Analyzing leakage of personally identifiable information in language models . In 44th IEEE Symposium on Security and Priv...

    2023

  24. [32]

    Thomas Mandl, Sandip Modha, Prasenjit Majumder, Daksh Patel, Mohana Dave, Chintak Mandalia, and Aditya Patel. 2019. https://doi.org/10.1145/3368567.3368584 Overview of the HASOC track at FIRE 2019: Hate speech and offensive content identification in indo-european languages . I...

    2019

  25. [33]

    Antonis Maronikolakis, Axel Wisiorek, Leah Nann, Haris Jabbar, Sahana Udupa, and Hinrich Schuetze. 2022. https://doi.org/10.18653/v1/2022.findings-acl.87 Listening to affected communities to define extreme speech: Dataset and experiments . In Findings of the Association for Co...

    2022 doi

  26. [34]

    Brendan McMahan, Eider Moore, Daniel Ramage, Seth Hampson, and Blaise Ag \" u era y Arcas. 2017. http://proceedings.mlr.press/v54/mcmahan17a.html Communication-efficient learning of deep networks from decentralized data . In Proceedings of the 20th International Conference on ...

    2017

  27. [35]

    Marzieh Mozafari, Reza Farahbakhsh, and No \" e l Crespi. 2019. https://doi.org/10.1007/978-3-030-36687-2\_77 A bert-based transfer learning approach for hate speech detection in online social media . In Complex Networks and Their Applications VIII - Volume 1 Proceedings of th...

    2019 doi

  28. [36]

    Shamsuddeen Hassan Muhammad, Idris Abdulmumin, Abinew Ali Ayele, David Ifeoluwa Adelani, Ibrahim Said Ahmad, Saminu Mohammad Aliyu, Nelson Odhiambo Onyango, Lilian D. A. Wanzare, Samuel Rutunda, Lukman Jibril Aliyu, Esubalew Alemneh, Oumaima Hourrane, Hagos Tesfahun Gebremicha...

    2025 doi arXiv

  29. [37]

    Hala Mulki, Hatem Haddad, Chedi Bechikh Ali, and Halima Alshabani. 2019. https://doi.org/10.18653/v1/W19-3512 L - HSAB : A L evantine T witter dataset for hate speech and abusive language . In Proceedings of the Third Workshop on Abusive Language Online, pages 111--118, Floren...

    2019 doi

  30. [38]

    Gianluca Nogara, Francesco Pierri, Stefano Cresci, Luca Luceri, Petter T \" o rnberg, and Silvia Giordano. 2023. https://doi.org/10.48550/ARXIV.2312.12651 Toxic bias: Perspective API misreads german as more toxic . CoRR, abs/2312.12651

    2023 doi arXiv

  31. [39]

    Debora Nozza. 2021. https://doi.org/10.18653/v1/2021.acl-short.114 Exposing the limits of zero-shot cross-lingual hate speech detection . In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on N...

    2021 doi

  32. [40]

    Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll L. Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, John Schulman, Jacob Hilton, Fraser Kelton, Luke Miller, Maddie Simens, Amanda Askell, Peter Welinder, Paul F. Christiano, Jan Leik...

    2022

  33. [41]

    Pratyush Patel, Esha Choukse, Chaojie Zhang, \'I \ n igo Goiri, Brijesh Warrier, Nithish Mahalingam, and Ricardo Bianchini. 2023. Polca: Power oversubscription in llm cloud providers. arXiv preprint arXiv:2308.12908

    2023 arXiv

  34. [42]

    Zesis Pitenis, Marcos Zampieri, and Tharindu Ranasinghe. 2020. https://aclanthology.org/2020.lrec-1.629 Offensive language identification in G reek . In Proceedings of the Twelfth Language Resources and Evaluation Conference, pages 5113--5119, Marseille, France. European Langu...

    2020

  35. [43]

    Tharindu Ranasinghe and Marcos Zampieri. 2021. https://doi.org/10.18653/v1/2021.naacl-demos.17 MUDES : Multilingual detection of offensive spans . In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Langua...

    2021 doi

  36. [44]

    Tharindu Ranasinghe and Marcos Zampieri. 2022. https://doi.org/10.1145/3457610 Multilingual offensive language identification for low-resource languages . ACM Trans. Asian Low Resour. Lang. Inf. Process. , 21(1):4:1--4:13

    2022 doi

  37. [45]

    Paul R \"o ttger, Haitham Seelawi, Debora Nozza, Zeerak Talat, and Bertie Vidgen. 2022. https://doi.org/10.18653/v1/2022.woah-1.15 Multilingual H ate C heck: Functional tests for multilingual hate speech detection models . In Proceedings of the Sixth Workshop on Online Abuse a...

    2022 doi

  38. [46]

    Paul R \"o ttger, Bertie Vidgen, Dong Nguyen, Zeerak Waseem, Helen Margetts, and Janet Pierrehumbert. 2021. https://doi.org/10.18653/v1/2021.acl-long.4 H ate C heck: Functional tests for hate speech detection models . In Proceedings of the 59th Annual Meeting of the Associatio...

    2021 doi

  39. [47]

    Victor Sanh, Lysandre Debut, Julien Chaumond, and Thomas Wolf. 2019. Distilbert, a distilled version of bert: smaller, faster, cheaper and lighter. ArXiv, abs/1910.01108

    2019 arXiv

  40. [48]

    Maarten Sap, Dallas Card, Saadia Gabriel, Yejin Choi, and Noah A. Smith. 2019. https://doi.org/10.18653/v1/P19-1163 The risk of racial bias in hate speech detection . In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 1668--1678, ...

    2019 doi

  41. [49]

    Smith, and Yejin Choi

    Maarten Sap, Saadia Gabriel, Lianhui Qin, Dan Jurafsky, Noah A. Smith, and Yejin Choi. 2020. https://doi.org/10.18653/v1/2020.acl-main.486 Social bias frames: Reasoning about social and power implications of language . In Proceedings of the 58th Annual Meeting of the Associati...

    2020 doi

  42. [50]

    Diptanu Sarkar, Marcos Zampieri, Tharindu Ranasinghe, and Alexander Ororbia. 2021. https://doi.org/10.18653/v1/2021.findings-emnlp.154 f BERT : A neural transformer for identifying offensive content . In Findings of the Association for Computational Linguistics: EMNLP 2021, pa...

    2021 doi

  43. [51]

    Micah J Sheller, Brandon Edwards, G Anthony Reina, Jason Martin, Sarthak Pati, Aikaterini Kotrotsou, Mikhail Milchenko, Weilin Xu, Daniel Marcus, Rivka R Colen, et al. 2020. Federated learning in medicine: facilitating multi-institutional collaborations without sharing patient...

    2020

  44. [52]

    Akshay Singh and Rahul Thakur. 2024. https://doi.org/10.18653/v1/2024.naacl-long.400 Generalizable multilingual hate speech detection on low resource I ndian languages using fair selection in federated learning . In Proceedings of the 2024 Conference of the North American Chap...

    2024 doi

  45. [53]

    Gomez, Lukasz Kaiser, and Illia Polosukhin

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. 2017. https://proceedings.neurips.cc/paper/2017/hash/3f5ee243547dee91fbd053c1c4a845aa-Abstract.html Attention is all you need . In Advances in Neural I...

    2017

  46. [54]

    Pablo Villalobos, Jaime Sevilla, Lennart Heim, Tamay Besiroglu, Marius Hobbhahn, and Anson Ho. 2022. Will we run out of data? an analysis of the limits of scaling datasets in machine learning. arXiv preprint arXiv:2211.04325

    2022 arXiv

  47. [55]

    Wenhui Wang, Furu Wei, Li Dong, Hangbo Bao, Nan Yang, and Ming Zhou. 2020. https://proceedings.neurips.cc/paper/2020/hash/3f5ee243547dee91fbd053c1c4a845aa-Abstract.html Minilm: Deep self-attention distillation for task-agnostic compression of pre-trained transformers . In Adva...

    2020

  48. [56]

    Zeerak Waseem and Dirk Hovy. 2016. https://doi.org/10.18653/v1/N16-2013 Hateful symbols or hateful people? predictive features for hate speech detection on T witter . In Proceedings of the NAACL Student Research Workshop , pages 88--93, San Diego, California. Association for C...

    2016 doi

  49. [57]

    Taylor Webb, Keith J Holyoak, and Hongjing Lu. 2023. Emergent analogical reasoning in large language models. Nature Human Behaviour, 7(9):1526--1541

    2023

  50. [58]

    Ellery Wulczyn, Nithum Thain, and Lucas Dixon. 2017. https://doi.org/10.1145/3038912.3052591 Ex machina: Personal attacks seen at scale . In Proceedings of the 26th International Conference on World Wide Web, WWW 2017, Perth, Australia, April 3-7, 2017 , pages 1391--1399. ACM

    2017

  51. [59]

    Marcos Zampieri, Damith Premasiri, and Tharindu Ranasinghe. 2024. A federated learning approach to privacy preserving offensive language identification. arXiv preprint arXiv:2404.11470

    2024 arXiv

  52. [60]

    Ziqi Zhang, David Robinson, and Jonathan A. Tepper. 2018. https://doi.org/10.1007/978-3-319-93417-4\_48 Detecting hate speech on twitter using a convolution-gru based deep neural network . In The Semantic Web - 15th International Conference, ESWC 2018, Heraklion, Crete, Greece...

    2018 doi

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.