Pith. sign in

REVIEW 4 major objections 10 minor 1 cited by

IYKYK: Using language models to decode extremist cryptolects

T0 review · 4 major / 10 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read The paper argues that general-purpose LLMs cannot consistently detect or decode the coded language of extremist forums, and that prompting context and domain adaptation substantially improve both detection and decoding.

desk verdict Worth reading for the 19.4M-post corpus and the solid domain-adaptation results; the zero-shot F1 numbers are provisional because the gold labels and model prompts share the same inclusion-biased instruction. read the letter →

arxiv 2506.05635 v1 pith:B6VY5OCN submitted 2025-06-05 cs.CL

Christine de Kock , Arij Riabi , Zeerak Talat , Michael Sejr Schlichtkrull , Pranava Madhyastha , Ed Hovy This is my paper
classification cs.CL
keywords cryptolectsin-grouplanguageonlineextremismlargemodelsdomainadaptationhatespeechdetectionpromptengineeringStormfront
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 show that off-the-shelf language models are not equipped to detect or decode the deliberately coded vocabularies ('cryptolects') of extremist online communities, and that performance on these tasks can be substantially recovered by adding context to prompts and by domain-adapting models on forum data. To test this, it builds a corpus of 19.4 million posts from Stormfront and Incels, creates expert-validated lexicons of in-group terms, and evaluates eight models across six tasks. The headline results are that the best zero-shot model reaches F1 of 64.4% on Stormfront and 80.0% on Incels for word classification, and that providing ten example posts lifts correct definition generation for Incels from 44.7% to 90.3%. Domain adaptation of the multilingual encoder xlmt improves radical content detection, reaching state-of-the-art macro-F1 on the English Call for Action task and improving hate speech detection on the matching ideology. If correct, these findings imply that moderation systems cannot simply point a general LLM at extremist forums; they need platform-aware prompting or specialised adaptation, and the released datasets give the field a place to build those tools.

What carries the argument

Three components carry the argument. (1) The IYKYK corpus: 19.4M public posts from Stormfront and Incels, with metadata, used both as adaptation data and as the source of test cases. (2) A lexicon pipeline that combines LISTN-C, a socio-temporal method that jointly embeds words and users and scores words by distance to community clusters, with paid expert annotation to produce validated positive/negative lexicons for each platform. (3) A two-track evaluation design: instruction-tuned LLMs (Gemma, Llama, Mixtral, Qwen) are probed with four prompt framings (definition only, full instructions, one example post, ten example posts); the encoder model xlmt is continued-pretrained on the corpus and then fine-tuned on the Counter radical-content tasks and a hate speech task. The prompt framing and the continued-pretraining step are the levers that, per the paper, close most of the gap.

What would settle it

Re-annotate the Task 1 and Task 3 test sets with multiple independent experts per platform, then recompute zero-shot F1 and definition-correctness rates against majority-vote labels; if inter-annotator agreement is low or reported scores drop sharply, the paper's central comparison does not survive. A complementary check is to run the same classifiers on a random sample of frequent forum words that LISTN-C did not propose: if many of those are in-group terms, the gold standard misses part of the vocabulary and the reported miss rates understate the problem.

Watch

Extended reading notes

Core claim

The paper's central claim is that generalised language technologies—LLMs trained on broadly filtered internet text—are not directly reliable on extremist cryptolects, and that specialised techniques (prompt context and continued pretraining on platform data) are required to make them useful. On zero-shot classification of in-group words, the best model (Llama-3.3 70B) achieves F1 of 64.4% on Stormfront and 80.0% on Incels; adding the full annotation instructions raises F1 by about 10 percentage points, and including ten example posts raises the share of correct definitions for Incels from 44.7% to 90.3%. Domain-adapting the xlmt encoder on 19.4M forum posts significantly improves macro-F1 on the Counter tasks, reaching state-of-the-art results on English Call for Action prediction and establishing a first baseline for Radicalization Level prediction. The same adaptation improves hate speech detection when the adaptation platform matches the hate speech target (Stormfront for anti-immigrant, Incels for misogynist).

Load-bearing premise

Everything about model performance on in-group words rests on gold labels produced by a single expert annotator per platform, applied to candidate words selected by the authors' LISTN-C method, and the annotation instructions tell annotators to include borderline words, which pushes the positive class larger and can inflate LLM scores.

Editorial extensions

If this is right

  • Unadapted general LLMs are not safe arbiters of extremist content: the best zero-shot in-group word classifier still reaches only 64.4% F1 on Stormfront, meaning many coded terms slip through or are mislabelled in moderation pipelines.
  • Adding a few in-context example posts is an inexpensive remedy: correct decoding for Incels rises from 44.7% to 90.3% with ten examples, so API-based systems can be steered without retraining.
  • Continued pretraining on extremist forum text transfers beyond the original platforms: it improves radical-content detection even when the adaptation data comes from a different ideology, and it improves hate speech detection when the platform matches the target bias.
  • The new 19.4M-post corpus and expert-validated lexicons provide a shared testbed for moderation and intervention research, and the first baseline for Radicalization Level prediction gives future work a number to beat.

Reading between the lines

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

  • The paper's explanation—that toxic-content filtering during pretraining removes exactly the data these tasks need—implies a structural tension between safe generation and monitoring capability; a plausible design consequence is separate 'monitoring' models that keep toxic text for representation learning while general chatbots remain filtered.
  • The Incels results are inflated by productive morphemes such as '-cel', '-maxx', and '-oid'; a natural stress test is whether adversarial mutations of those suffixes (e.g., novel spellings or unusual compounds) cause the same models to fail, since cryptolects evolve by exactly this kind of recombination.
  • The pattern that example posts outperform explicit definitions for decoding suggests a general recipe for decoding any community-specific jargon, not just extremist groups: give the model distributional evidence (examples) rather than only a rule, because examples carry usage frequencies and collocational cues that definitions omit.
  • The combined-forums benefit suggests pooling data across many radical communities could cover rarer terms, but the paper does not test how long adapted models remain useful as these languages evolve; that temporal robustness is an open question.
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

4 major / 10 minor

Summary. The paper introduces IYKYK, a new dataset of 19.4 million public posts from the extremist forums Stormfront and Incels, and uses it to test the central hypothesis that general-purpose language technologies require specialised techniques when applied to extremist cryptolects. The authors construct gold-standard lexicons of in-group language using the LISTN-C candidate-generation method followed by expert annotation, and then evaluate seven instruction-tuned LLMs on three zero-shot tasks (classification, retrieval, and decoding of in-group terms) under four prompt framings. They also evaluate domain adaptation of the encoder model xlmt on radical-content prediction tasks from the Counter dataset and on hate-speech detection, reporting significant gains for English radical-content tasks and small but consistent gains for hate-speech tasks. The paper releases datasets and models to verified researchers and includes extensive ethical and limitation discussion.

Significance. If the reported results are valid, this is a substantial contribution to a relatively under-studied area of NLP: it provides a large-scale, long-range dataset from two extremist communities, a transparent lexicon-induction and annotation pipeline, and evidence that domain adaptation on in-domain data can improve radical-content and hate-speech classification. The supervised experiments in Section 7 are well-designed, use a standard toolkit, and include significance tests; the hate-speech transfer results, though small, are a useful sanity check. The paper also ships reproducibility details for hyperparameters and prompts. However, the zero-shot evidence base has several validity concerns that affect the strength of the central claim that LLMs cannot consistently detect or decode extremist language, so the contribution is currently somewhat uneven.

major comments (4)
  1. [§4.1, §6.2] Task 2, labeled 'Retrieval,' is not evaluated as retrieval: scoring only checks whether the single target word is included or excluded in the model's output, ignoring all other tokens the model might retrieve. The paper itself states in §6.2 that 'the same objective is evaluated in both' Task 1 and Task 2, and reports label coincidence of 71% (Incels) and 79% (Stormfront). As a result, the 'six tasks' framing overstates the diversity of the evaluation, and the claim that this task is 'more demanding than Task 1' is not supported by the metric. Please either re-frame Task 2 as a contextualized target-word classification variant or implement a true token-level retrieval metric that accounts for all retrieved and gold tokens.
  2. [§4.1, §6.3, Abstract] The decoding task (Task 3) is evaluated with only one model, Llama-3.3-70B, while the abstract and introduction describe 'eight models across six tasks.' This means that the conclusion that 'general purpose LLMs cannot consistently detect or decode extremist language' is supported by a single model for the decoding component. Please either expand the decoding evaluation to more models or revise the abstract and summary statements to accurately scope the decoding results as a case study with one model.
  3. [§5, App. B, App. C] The gold labels for Tasks 1–3 are produced by a single expert per platform on candidate words from the authors' own LISTN-C method, and the annotation instructions include an explicit inclusion bias ('in the case of borderline words, you should opt to include them'). The 'instructions' prompt framing then provides the model with the same positive/negative example lists and inclusion policy that the annotator used. This creates a potential alignment between the annotation rule and the model prompt, so the reported zero-shot F1 values (e.g., 80.0% for Incels) may not be a neutral measure of LLM ability to recognize cryptolects. The Limitations section acknowledges the single-annotator risk but does not discuss the reuse of annotation instructions as prompts. Please add an explicit analysis of this possible circularity, for example by ablating the instruction prompt or by reporting annotator-model agreement on the borderline cases.
  4. [§7, Table 3] The paper claims 'state-of-the-art results on the Call for Action task' without comparing against any published result other than the xlmt baseline from Riabi et al. (2025). If the Counter dataset has been used in other supervised systems, those should be cited and compared; otherwise, the claim should be qualified as 'best under this experimental configuration.'
minor comments (10)
  1. [Abstract and §4.1] The phrase 'eight models across six tasks' is misleading because Task 3 uses only one model; please specify the exact number of models per task in the abstract or in the task descriptions.
  2. [§2.2] There is a duplicated word in 'expressing expressing a radical perspective'; please remove the duplicate.
  3. [§4.2] The text reads 'usingxmlt'; there is a missing space between 'using' and 'xlmt'.
  4. [§6.1] The sentence 'discussed shown in Table 2' is grammatically awkward; consider rephrasing to 'as discussed and shown in Table 2.'
  5. [§7, Table 3] The table caption reports one-sided t-tests without correction for multiple comparisons; please add a note on whether the significance thresholds remain after any multiple-testing correction, or justify the use of uncorrected tests.
  6. [References] In Section 7 the text cites 'the model of Riabi et al. (2024)' but the Counter dataset reference appears as Riabi et al. (2025) in the bibliography; please check the intended year and citation.
  7. [App. B] The annotation instructions for Incels and Stormfront are nearly identical except for the platform name and ideology; a note that they were intentionally kept parallel would help readers understand the design choice.
  8. [Figure 1] The caption says the posts are 'in gray,' but the figure uses colored text for the translations; please ensure the visual distinction is clear in black-and-white printing.
  9. [Ethics] The statement that researcher risk management plans are in place is reassuring, but a one-sentence description of what those plans include (e.g., counseling, monitoring) would be more informative.
  10. [App. B.1.1] In the list of positive examples, 'linguistic exports' cites 'incel, mogging' as exports from the incel community, while 'linguistic imports' also lists 'mogging' as an import from elsewhere; clarify the intended status of 'mogging.'

Circularity Check

1 steps flagged · score 3.0 of 10

Zero-shot evaluation is largely self-contained, but the 'instructions' prompting condition reuses the exact annotation rubric that produced the gold labels, making part of the reported F1 gain self-referential.

  1. self definitional [Sec. 4.1 (Context framing), App. C.1.2/C.3.2, Sec. 5/App. B]
    "The instructions framing provides the full instructions as used for the human annotation (detailed in Sec. 5) which also includes the nature of the research (focusing on extremism), the platform name and ideology, and examples of positive and negative types. ... in the case of borderline words, you should opt to include them. ... The prompt is constructed based on the instructions in Sec. B.2, preceded by: You are an expert sociologist working on online extremist groups."

    The gold labels for Tasks 1 and 3 are produced by expert annotators applying the App. B instructions, which define in-group language and explicitly bias borderline cases toward inclusion. The 'instructions' prompt condition is built from those same instructions, including the positive/negative examples and the inclusion policy, and is prefixed by the same expert-sociologist role. The model is therefore evaluated against labels generated under the exact decision rule that is placed verbatim in its prompt.

full rationale

The paper's main quantitative claims are empirical evaluations rather than derivations from fitted parameters. The supervised domain-adaptation results (Sec. 7) compare xlmt baselines against models adapted on the IYKYK corpus and evaluated on the Counter benchmark; the labels and test splits come from prior work, and the improvements are controlled comparisons, so no circularity arises there. The zero-shot tasks use human expert labels, which is a standard evaluation design. The clearest circular element is the reuse of the annotation instructions as model prompts: the gold labels for Tasks 1 and 3 were generated by experts following App. B, and the 'instructions' framing feeds those same instructions, examples, and inclusion bias to the model. This makes the instruction-condition results partially self-referential. However, the definition-only and example framings do not share the full rubric, and the paper's Limitations section acknowledges the single-annotator and subjectivity concerns. The LISTN-C self-citation is a methodological tool rather than an unsupported load-bearing claim, and the Counter benchmark is externally grounded despite author overlap. Overall, the central hypothesis is supported by multiple independent comparisons, so the circularity is partial and localized rather than a collapse of the derivation chain.

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

The central claim does not rely on derived equations or fitted parameters. The main experimental levers are the candidate selection threshold and the annotation protocol, both of which shape the test sets. The paper's own limitations section concedes the single-annotator issue. No new particles, forces, or theoretical entities are introduced.

free parameters (2)
  • Top-3050 candidate threshold = 3050 words per community
    The number of candidate words extracted by LISTN-C per platform is chosen by hand to include both positive and negative test cases (Sec. 5). It determines the test set composition and class balance, directly affecting all zero-shot F1 scores.
  • Subword exclusion set = pill, mog, maxx, cel, chad, cuck, oid
    The robustness analysis in Sec. 6.1 excludes 570 positive Incels terms containing these substrings, changing reported F1 by 13.5%. This is a post-hoc stratification that affects interpretation of the main classification results.
assumptions (3)
  • domain assumption Single-annotator labels, produced under instructions that favor inclusion of borderline words, are an adequate gold standard for in-group language.
    App. B.1 instructs annotators to include borderline words; the Limitations section admits a single annotator per dataset was used. The entire zero-shot evaluation uses these labels as ground truth.
  • domain assumption The Counter dataset annotations (Riabi et al. 2025) provide valid labels for radical content detection, and macro-F1 on this dataset is the right measure of domain adaptation success.
    Section 4.2 adopts the Counter tasks and 70:20:10 split; the 'state-of-the-art' claim is relative to this internal benchmark, which the authors themselves introduced in prior work.
  • domain assumption LISTN-C candidate generation yields a representative sample of in-group language for both communities.
    Section 5 relies on the authors' LISTN-C method (de Kock 2024a) to produce the candidate word sets. If this method misses important in-group terms or over-selects statistically distinctive but not truly cryptolectal words, the test set is biased.

how reviews work

0 comments
Cite this review

Pith. "Pith review of IYKYK: Using language models to decode extremist cryptolects." pith.science (2026). https://pith.science/paper/B6VY5OCN

@misc{pith2026250605635,
  author       = {Pith},
  title        = {Pith review of: IYKYK: Using language models to decode extremist cryptolects},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/B6VY5OCN}},
  note         = {Machine review of arXiv:2506.05635}
}
read the original abstract

Extremist groups develop complex in-group language, also referred to as cryptolects, to exclude or mislead outsiders. We investigate the ability of current language technologies to detect and interpret the cryptolects of two online extremist platforms. Evaluating eight models across six tasks, our results indicate that general purpose LLMs cannot consistently detect or decode extremist language. However, performance can be significantly improved by domain adaptation and specialised prompting techniques. These results provide important insights to inform the development and deployment of automated moderation technologies. We further develop and release novel labelled and unlabelled datasets, including 19.4M posts from extremist platforms and lexicons validated by human experts.

Figures

Figures reproduced from arXiv: 2506.05635 by the authors.

Figure 1
Figure 1. Three posts from the IYKYK dataset (in gray) [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Tasks used for the zero-shot LLM evaluation. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Results for Task 1 (Classification) [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Results for Task 3 (Decoding). to that use case, sometimes performing worse than without it. For the term apache, a transphobic meme6 , all 10 examples refer to literal helicopters. Including more posts may mitigate this issue, but it is also an inherent challenge of t…
Figure 5
Figure 5. Figure 5: Precision scores for Task 1 [PITH_FULL_IMAGE:figures/full_fig_p016_5.png]
Figure 6
Figure 6. Figure 6: Recall scores for Task 1. Model CfA CfA+ideology CfA+NER CfA+rad. level xlmt 65.7 65.6 62.0 63.9 xlmt+incels 65.0 (−0.6) 68.7 (+3.2) 63.8 (+1.9) 64.2 (+0.3) xlmt+stormfront 65.9 (+0.3) 62.3 (−3.3) 61.7 (−0.3) 62.5 (−1.4) xlmt+stormfront+incels 65.8 (+0.2) 64.0 (−1.6) 6…
Figure 7
Figure 7. Figure 7: Results for Task 2 (Retrieval) [PITH_FULL_IMAGE:figures/full_fig_p017_7.png]
Figure 8
Figure 8. Figure 8: Results for Task 3, excluding common mor [PITH_FULL_IMAGE:figures/full_fig_p017_8.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. LLM Harms: A Taxonomy and Discussion

    cs.CY 2025-12 unverdicted novelty 3.0 of 10

    This paper proposes a taxonomy of LLM harms in five categories and suggests mitigation strategies plus a dynamic auditing system for responsible development.

Reference graph

Works this paper leans on

43 extracted references · 26 canonical work pages · cited by 1 Pith paper

  1. [1]

    Dai, Orhan Firat, Melvin Johnson, Dmitry Lepikhin, Alexandre Passos, Siamak Shakeri, Emanuel Taropa, Paige Bailey, Zhifeng Chen, Eric Chu, Jonathan H

    Rohan Anil, Andrew M. Dai, Orhan Firat, Melvin Johnson, Dmitry Lepikhin, Alexandre Passos, Siamak Shakeri, Emanuel Taropa, Paige Bailey, Zhifeng Chen, Eric Chu, Jonathan H. Clark, Laurent El Shafey, Yanping Huang, Kathy Meier-Hellstern, Gaurav Mishra, Erica Moreira, Mark Omernick, Kevin Robinson, and 109 others. 2023. https://arxiv.org/abs/2305.10403 Palm...

    arXiv 2023

  2. [2]

    incel” to “saint

    Stephane J Baele, Lewys Brace, and Travis G Coan. 2021. From “incel” to “saint”: Analyzing the violent worldview behind the 2018 toronto attack. Terrorism and political violence, 33(8):1667--1691

    work page 2021

  3. [3]

    Jinze Bai, Shuai Bai, Yunfei Chu, Zeyu Cui, Kai Dang, Xiaodong Deng, Yang Fan, Wenbin Ge, Yu Han, Fei Huang, and 1 others. 2023. Qwen technical report. arXiv preprint arXiv:2309.16609

    arXiv 2023

  4. [4]

    Francesco Barbieri, Luis Espinosa Anke, and Jose Camacho-Collados. 2022. 2022.lrec-1.27 XLM - T : Multilingual language models in T witter for sentiment analysis and beyond . In lrec:2022:1, pages 258--266, Marseille, France. European Language Resources Association

    work page 2022

  5. [5]

    Omran Berjawi, Giuseppe Fenza, and Vincenzo Loia. 2023. https://doi.org/10.1109/ACCESS.2023.3326995 A comprehensive survey of detection and prevention approaches for online radicalization: Identifying gaps and future directions . IEEE Access, 11:120463--120491

    arXiv 2023

  6. [6]

    Ksenija Bogeti \'c . 2023. Race and the language of incels: Figurative neologisms in an emerging english cryptolect. English Today, 39(2):89--99

    work page 2023

  7. [7]

    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 at scale . In acl:2020:main, pages 8440--8451, Online. acl

    doi:10.18653/v1/2020.acl-main.747 2020

  8. [8]

    Maura Conway. 2006. Terrorism and the internet: New media—new threat? Parliamentary Affairs, 59(2):283--298

    work page 2006

Show all 43 references
  1. [9]

    Christine de Kock. 2024 a . Inducing lexicons of in-group language with socio-temporal context. Accepted to ACL 2025. arXiv preprint arXiv:2409.19257

    2024 arXiv

  2. [10]

    Christine de Kock. 2024 b . https://arxiv.org/abs/2409.19243 Jointly modelling the evolution of community structure and language in online extremist groups . Preprint, arXiv:2409.19243

    2024 arXiv

  3. [11]

    Christine De Kock and Eduard Hovy. 2024. Investigating radicalisation indicators in online extremist communities. In Proceedings of the 8th Workshop on Online Abuse and Harms (WOAH 2024), pages 1--12

    2024

  4. [12]

    Donatella Della Porta and Gary LaFree. 2012. Processes of radicalization and de-radicalization. International Journal of Conflict and Violence (IJCV), 6(1):4--10

    2012

  5. [13]

    Glendon Frank Drake. 1980. The social role of slang. In Language, pages 63--70. Elsevier

    1980

  6. [14]

    Aleksandra Edwards and Jose Camacho-Collados. 2024. https://aclanthology.org/2024.lrec-main.879/ Language models for text classification: Is in-context learning enough? In Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources a...

    2024

  7. [15]

    Paula Fortuna, Juan Soler-Company, and Leo Wanner. 2021. https://doi.org/10.1016/j.ipm.2021.102524 How well do hate speech, toxicity, abusive and offensive language classification models generalize across datasets? Information Processing & Management, 58(3):102524

    2021

  8. [16]

    Gaikwad, Swati Ahirrao, Shraddha Phansalkar, and K

    M. Gaikwad, Swati Ahirrao, Shraddha Phansalkar, and K. Kotecha. 2021. Online extremism detection: A systematic literature review with emphasis on datasets, classification techniques, validation methods, and tools. IEEE Access, 9:48364--48404

    2021

  9. [17]

    Mayur Gaikwad, Swati Ahirrao, Shraddha Phansalkar, Ketan Kotecha, Shalli Rani, and Lorenzo Putzu. 2023. https://doi.org/10.1155/2023/4563145 Multi-ideology, multiclass online extremism dataset, and its evaluation using machine learning . Computational Intelligence and Neurosci...

    2023 doi

  10. [18]

    Gemma Team , Morgane Riviere, Shreya Pathak, Pier Giuseppe Sessa, Cassidy Hardin, Surya Bhupatiraju, L \'e onard Hussenot, Thomas Mesnard, Bobak Shahriari, Alexandre Ram \'e , and 1 others. 2024. Gemma 2: Improving open language models at a practical size. arXiv preprint arXiv...

    2024 arXiv

  11. [19]

    Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Alex Vaughan, and 1 others. 2024. The L lama 3 herd of models. arXiv preprint arXiv:2407.21783

    2024 arXiv

  12. [20]

    DS Hoffman. 1997. Web of hate: Extremists exploit the internet. www.ojp.gov/ncjrs/. Accessed 18-04-2025

    1997

  13. [21]

    Albert Q Jiang, Alexandre Sablayrolles, Antoine Roux, Arthur Mensch, Blanche Savary, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Emma Bou Hanna, Florian Bressand, and 1 others. 2024. Mixtral of experts. arXiv preprint arXiv:2401.04088

    2024 arXiv

  14. [22]

    Md Tawkat Islam Khondaker, Muhammad Abdul-mageed, and Laks Lakshmanan, V.s. 2023. https://doi.org/10.18653/v1/2023.woah-1.9 Cross-platform and cross-domain abusive language detection with supervised contrastive learning . In The 7th Workshop on Online Abuse and Harms (WOAH), p...

    2023 doi

  15. [23]

    Hannah Kirk, Abeba Birhane, Bertie Vidgen, and Leon Derczynski. 2022. https://doi.org/10.18653/v1/2022.findings-emnlp.35 Handling and presenting harmful text in NLP research . In Findings of the Association for Computational Linguistics: EMNLP 2022, pages 497--510, Abu Dhabi, ...

    2022 doi

  16. [24]

    Jasmine Latimore and John Coyne. 2023. Incels in australia: the ideology, the threat, and a way forward

    2023

  17. [25]

    Shayne Longpre, Gregory Yauney, Emily Reif, Katherine Lee, Adam Roberts, Barret Zoph, Denny Zhou, Jason Wei, Kevin Robinson, David Mimno, and Daphne Ippolito. 2024. https://doi.org/10.18653/v1/2024.naacl-long.179 A pretrainer`s guide to training data: Measuring the effects of ...

    2024 doi

  18. [26]

    Gretchen McCulloch. 2019. Because internet: Understanding how language is changing. Random House

    2019

  19. [27]

    Julia Mendelsohn, Ronan Le Bras, Yejin Choi, and Maarten Sap. 2023. From dogwhistles to bullhorns: Unveiling coded rhetoric with language models. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 15162--15180

    2023

  20. [28]

    Syrielle Montariol, Arij Riabi, and Djam \'e Seddah. 2022. https://doi.org/10.18653/v1/2022.findings-aacl.33 Multilingual auxiliary tasks training: Bridging the gap between languages for zero-shot transfer of hate speech detection models . In Findings of the Association for Co...

    2022 doi

  21. [29]

    PMO. 2025. Downing street opens doors to adolescence creators for vital discussion on protecting our children. www.gov.uk/. Accessed 18-04-2025

    2025

  22. [30]

    Jack W. Rae, Sebastian Borgeaud, Trevor Cai, Katie Millican, Jordan Hoffmann, Francis Song, John Aslanides, Sarah Henderson, Roman Ring, Susannah Young, Eliza Rutherford, Tom Hennigan, Jacob Menick, Albin Cassirer, Richard Powell, George van den Driessche, Lisa Anne Hendricks,...

    2022 arXiv

  23. [31]

    Arij Riabi, Menel Mahamdi, Virginie Mouilleron, and Djam \'e Seddah. 2024. https://aclanthology.org/2024.privatenlp-1.13 Cloaked classifiers: Pseudonymization strategies on sensitive classification tasks . In Proceedings of the Fifth Workshop on Privacy in Natural Language Pro...

    2024

  24. [32]

    Arij Riabi, Virginie Mouilleron, Menel Mahamdi, Wissam Antoun, and Djam \'e Seddah. 2025. Beyond dataset creation: Critical view of annotation variation and bias probing of a dataset for online radical content detection. In Proceedings of the 31st International Conference on C...

    2025

  25. [33]

    Ben Rich and Eva Bujalka. 2023. The draw of the ‘manosphere’: understanding andrew tate’s appeal to lost men. The Conversation

    2023

  26. [34]

    Paul Röttger, Bertie Vidgen, Dong Nguyen, Zeerak Talat, Helen Margetts, and Janet Pierrehumbert. 2021. https://doi.org/10.18653/v1/2021.acl-long.4 HateCheck : Functional Tests for Hate Speech Detection Models . In Proceedings of the 59th Annual Meeting of the Association for C...

    2021 doi

  27. [35]

    Aaron Shaw. 2023. Social media, extremism, and radicalization. Science advances, 9(35):eadk2031

    2023

  28. [36]

    Luca Soldaini, Rodney Kinney, Akshita Bhagia, Dustin Schwenk, David Atkinson, Russell Authur, Ben Bogin, Khyathi Chandu, Jennifer Dumas, Yanai Elazar, Valentin Hofmann, Ananya Jha, Sachin Kumar, Li Lucy, Xinxi Lyu, Nathan Lambert, Ian Magnusson, Jacob Morrison, Niklas Muennigh...

    2024 doi

  29. [37]

    Daniele Valentini. 2021. Digital mind invasion: Far-right individuals and socio-technical disgust. CYBERPOLITICS: Political Philosophy of the Future, pages 125--143

    2021

  30. [38]

    Rob van der Goot, Ahmet \"U st \"u n, Alan Ramponi, Ibrahim Sharaf, and Barbara Plank. 2021. https://doi.org/10.18653/v1/2021.eacl-demos.22 Massive choice, ample tasks ( M a C h A mp): A toolkit for multi-task learning in NLP . In Proceedings of the 16th Conference of the Euro...

    2021 doi

  31. [39]

    Gabriel Weimann and Ari Ben Am. 2020. Digital dog whistles: The new online language of extremism. International Journal of Security Studies, 2(1):4

    2020

  32. [40]

    Charlie Winter, Peter Neumann, Alexander Meleagrou-Hitchens, Magnus Ranstorp, Lorenzo Vidino, and Johanna Fürst. 2021. https://doi.org/10.4119/ijcv-3809 Online extremism: Research trends in internet activism, radicalization, and counter-strategies . International Journal of Co...

    2021 doi

  33. [41]

    Albert Xu, Eshaan Pathak, Eric Wallace, Suchin Gururangan, Maarten Sap, and Dan Klein. 2021. https://doi.org/10.18653/v1/2021.naacl-main.190 Detoxifying language models risks marginalizing minority voices . In Proceedings of the 2021 Conference of the North American Chapter of...

    2021 doi

  34. [42]

    Michael Yoder, Chloe Perry, David Brown, Kathleen M Carley, and Meredith Pruden. 2023. Identity construction in a misogynist incels forum. In The 7th Workshop on Online Abuse and Harms (WOAH), pages 1--13

    2023

  35. [43]

    Hao Yu, Zachary Yang, Kellin Pelrine, Jean-François Godbout, and Reihaneh Rabbany. 2023. https://doi.org/10.48550/arXiv.2308.10092 Open, closed, or small language models for text classification? CoRR, abs/2308.10092

    2023 doi arXiv

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.