Pith. sign in

REVIEW 3 major objections 6 minor 60 references

Counterspeech the ultimate shield! Multi-Conditioned Counterspeech Generation through Attributed Prefix Learning

T0 review · 3 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read Multi-attribute conditioning makes counterspeech more strategy-aligned and human-preferred

desk verdict A useful new dataset and a reasonable model combo, but the headline strategy-conformity gain is measured by a classifier trained on the same data; send to review, fix the evaluation, and the work stands as a modest contribution. read the letter →

arxiv 2505.11958 v3 pith:BTEXGJAD submitted 2025-05-17 cs.CL

Aswini Kumar , Anil Bandhakavi , Tanmoy Chakraborty This is my paper · ORCID
classification cs.CL
keywords counterspeechgenerationhatespeechprefixtuninghierarchicallearningpreferenceoptimizationmulti-attributeconditioningemotion-annotateddatasetcontrollabletext
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

This paper argues that counterspeech generation becomes more effective when the model is conditioned on two attributes at once — a response strategy such as 'informative' paired with an emotion such as 'joy' — rather than on strategy alone. To test this, it proposes HiPPrO, which learns attribute-specific prefix vectors in a hierarchy (strategy prefixes first, then emotion prefixes on top of them) on a frozen FLAN-T5, and then applies a reward-free and reference-free preference optimization (ORPO) to make outputs less toxic and closer to human-written references. The paper also contributes MultiCONAN, an emotion-annotated extension of IntentCONANv2 covering 13,973 counterspeech instances. It reports a ~38% gain in strategy conformity and small ROUGE gains over several baselines, with human raters preferring HiPPrO outputs over CoARL, GPT-3.5, and GPT-4 few-shot. If correct, this makes automated counterspeech more steerable by human moderators while tuning less than 0.01% of the model's parameters.

What carries the argument

The central object is the hierarchical prefix stack: two nested sets of tunable key-value prefix vectors, $(\alpha,\beta)$ for strategy and $(\gamma,\delta)$ for emotion, added to the encoder and decoder of a frozen FLAN-T5. The strategy prefixes are learned first, then the emotion prefixes are trained with the strategy prefixes frozen, so the two attribute conditions compose without retraining the base model. The second piece of machinery is the odds-ratio preference optimization (ORPO) loss, which computes the odds of the ground-truth counterspeech versus the model-generated one and penalizes low-odds outputs; because it works directly on the model's own probabilities, it needs no reward model and no reference model. Together, the hierarchy encodes attribute composition, and ORPO aligns the generated counterspeech with human-written quality while keeping the trainable footprint at 589,824 parameters (0.0052% of the total).

What would settle it

Have independent human raters (not the model's authors) label the strategy of HiPPrO outputs without seeing the conditioning token, and compare the match rate to the SC classifier's score; if human-assessed conformity for HiPPrO does not exceed the baselines by roughly 38 percentage points, the reported gain is an artifact of the in-distribution classifier.

Watch

Extended reading notes

Core claim

The paper's central claim is that multi-attribute conditioning, realized through hierarchical prefix learning followed by odds-ratio preference optimization, produces counterspeech that is more strategy-conformant, less toxic, and lexically closer to human references than single-attribute generation models and large instruction-tuned LLMs such as GPT-3.5 and GPT-4. In support, HiPPrO reaches a strategy-conformity score of 0.929 against 0.545 for GPT-4 zero-shot and 0.533 for GPT-3.5 zero-shot, ROUGE-1 of 0.273 against 0.239 and 0.242 for GPT-3.5 zero-shot and few-shot, and wins human evaluations on adequacy, relevance, and argumentative effectiveness against CoARL, GPT-4 few-shot, and GPT-3.5 few-shot (win rates 0.85–0.98). The ablations show that the stacked strategy-plus-emotion prefix space is the driver: flat prefix tuning on FLAN-T5 reaches only 0.470 strategy conformity, removing the ORPO component degrades toxicity and other metrics, and replacing ORPO with DPO slightly hurts performance.

Load-bearing premise

The 38% strategy-conformity gain is only meaningful if the RoBERTa-large classifier trained on MultiCONAN truly measures strategy conformity, rather than mostly rewarding outputs that mimic the training distribution.

Editorial extensions

If this is right

  • Multi-attribute conditioning (strategy plus emotion) can be composed through stacked prefixes without retraining the base model, so the same language model can be redirected to new attribute combinations by swapping or extending the prefix stack.
  • Because only 0.0052% of the model parameters are trained, counterspeech systems can be tailored to specific communities or moderation policies at very low computational cost.
  • ORPO's freedom from reward and reference models means preference tuning of counterspeech can proceed with just a base model and written references, removing a common scaling bottleneck in aligning generative models.
  • The strategy-conformity gains suggest that conditioning on intent alone under-specifies counterspeech; emotion is a second, load-bearing axis that affects not only tone but also perceived relevance and adequacy.
  • MultiCONAN, with 13,973 emotion-labeled counterspeech instances, provides a testbed for studying how emotional tone interacts with strategy in counterspeech generation.

Reading between the lines

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

  • The hierarchy encodes an implicit assumption that strategy is the primary attribute and emotion the secondary one; a natural test is whether reversing the order (emotion first, then strategy) preserves the gains, which would clarify whether the stack is compositional or merely additive.
  • Because the strategy-conformity classifier is trained on the same labels that the model is trained on, the 38% gain may partly reflect in-distribution mimicry; an independent human-labeled strategy set or a classifier trained on a different corpus could measure the true transfer.
  • If the compositional prefix stack transfers, the same two-stage prefix-adapter approach could be applied to other controllable generation tasks (e.g., style+formality, stance+emotion) where adding a new attribute is as cheap as training a small prefix adapter.
  • The toxicity scores are small in absolute terms (0.087 for HiPPrO vs 0.030 for GPT-4 zero-shot), so the main measurable benefit of ORPO may be on non-toxicity rather than on the lexical metrics; a dedicated detoxification evaluation on adversarial inputs would clarify this.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. The paper introduces HiPPrO, a two-stage framework for counterspeech generation conditioned jointly on strategy and emotion. In the first phase, the model learns hierarchical prefix vectors on top of FLAN-T5: strategy-specific prefixes first, then emotion-specific prefixes initialized from the strategy prefixes, with the base model frozen. In the second phase, it applies ORPO, a reference- and reward-model-free preference optimization, using the ground-truth counterspeech as the chosen response and the phase-one generation as the rejected response. The authors also release MultiCONAN, an extension of IntentCONANv2 in which all 13,973 counterspeech instances are annotated with one of five emotion labels. Evaluation includes ROUGE, METEOR, BERTScore, cosine similarity, RoBERTa-based strategy/emotion/target conformity scores, toxicity, and a human study; the abstract reports a roughly 38% improvement in strategy conformity and roughly 3%, 2%, and 3% improvements in ROUGE-1, ROUGE-2, and ROUGE-L over baseline models.

Significance. If the empirical claims hold, the paper makes a useful contribution: multi-attribute conditioning through hierarchical prefix learning is parameter-efficient, ORPO avoids the need for a reference model, and the MultiCONAN dataset is a concrete resource for counterspeech research. The paper is commendably transparent in releasing code and data, and it compares against a broad set of baselines including retrieval-based methods, zero-/few-shot LLMs, and prefix-tuning variants, with sensible ablations on virtual-token size and on DPO versus ORPO. However, the central quantitative claim depends on an evaluation classifier trained on the same data as the generator, and the headline improvement figures are reported as absolute differences labeled as percentages. The human evaluation, which the paper itself identifies as important because automatic metrics do not fully align with human perception, is based on only 30 examples. These issues need to be addressed before the headline claims can be accepted.

major comments (3)
  1. [Section 5.2, Table 2] The headline strategy-conformity gain is measured with RoBERTa-large classifiers fine-tuned on MultiCONAN, the same dataset from which HiPPrO's prefixes and preference pairs are derived. The reported SC of 0.929 for HiPPrO, and the +0.384 improvement over GPT-4 ZS, could therefore reflect in-distribution mimicry of the training labels rather than genuinely better adherence to the intended strategies. The paper's own Limitation paragraph states that the automatic metrics do not fully align with human perception. This concern is load-bearing because the abstract's central quantitative claim (~38% improvement) rests on SC. I request validation with an external classifier or with an independently annotated sample of generated outputs before this claim is accepted.
  2. [Abstract and Table 2] The manuscript reports "~38% improvement in intent conformity" and "~3%, ~2%, ~3% improvement" in ROUGE, but these numbers are absolute differences in metric values, not relative percentages. For example, SC 0.929 minus 0.545 (GPT-4 ZS) equals 0.384, which is an absolute gain of 0.384, not a 38% relative gain; the relative gain against that baseline would be roughly 70%. Similarly, the ROUGE deltas in the last row of Table 2 are absolute point differences. The terminology should be corrected, and the comparison baseline for each delta should be stated explicitly, since the best SC value in Table 2 among all baselines is DialoGPT at 0.714, not the baseline implied by the reported +0.384.
  3. [Section 6.3, Table 3] The human evaluation is based on only 30 randomly selected responses rated by 35 experts, and the reported win rates (e.g., 0.96 for HiPPrO vs. CoARL) are presented without confidence intervals, inter-rater agreement statistics, or significance tests. Given that the Limitation section acknowledges that automatic metrics do not fully align with human perception, the human evaluation is important evidence. At n=30, the evidence is too thin to support the strong claim that HiPPrO is superior on all four subjective dimensions. I recommend reporting the full rating distribution, agreement statistics, and a confidence interval or significance test for the win rates.
minor comments (6)
  1. [Throughout] The terms "Rouge" and "Meteor" should be "ROUGE" and "METEOR", and "odd ratio" in Equations (3)-(6) and in Figure 2 should be "odds ratio".
  2. [Table 2] The HiPPrO row contains "0.5670.929" without a separator, which makes the table difficult to read; this appears to be a missing space between CoSim and SC values.
  3. [Table 2] The ablation row labels use "PrifixTuning" in several places; this should be "PrefixTuning".
  4. [Abstract, Section 5.2] Terminology is inconsistent: the abstract says "intent conformity" while Section 5.2 and Table 2 use "strategy conformity" (SC); please pick one term and use it consistently.
  5. [Appendix 9.9] The computing section says "NVIDIA RTX A100 with80GB RAM"; there is a missing space before "80GB".
  6. [References] The reference list contains duplicate entries: Fanton et al. (2021a) and (2021b) are the same paper, and Masud et al. (2021a) and (2021b) are also identical; these should be merged.

Circularity Check

1 steps flagged · score 6.0 of 10

The headline ~38% strategy-conformity gain is measured by a RoBERTa classifier trained on the same MultiCONAN labels that the generator is trained to reproduce, making the central SC claim partially circular.

  1. fitted input called prediction [Section 5.2 (Evaluation Metrics) and Section 4 (Task Formulation); Table 2]
    "We first train three distinct RoBERTa-large models on our dataset to measure the SC, EC, and TC scores. The models achieve testing accuracy of 0.86, 0.75, and 0.88, respectively, before being considered as evaluation metrics."

    The SC classifier is fitted on MultiCONAN, the same dataset and strategy/emotion labels used to train HiPPrO (Eqs. 1-2, 5). The generator maximizes likelihood of ground-truth MultiCONAN counterspeech conditioned on those labels, so its outputs are optimized to match the training distribution; the classifier is optimized to label exactly that distribution. The reported SC of 0.929 and the headline ~38% gain over GPT-4 ZS's 0.545 therefore partly measure in-distribution mimicry of the classifier's training signal. The Limitation section concedes that 'the evaluation metrics employed do not fully align with human perceptions of counterspeech quality.'

full rationale

The paper's method (hierarchical prefix learning + ORPO) is a trained system, not a mathematical derivation, and most of its pipeline is self-contained: prefix adapters are optimized on MultiCONAN, and ORPO is a standard preference objective. The dataset extension (emotion labels) is a real annotation contribution with reported inter-annotator agreement. The principal circularity is in the evaluation of strategy conformity: the SC/EC/TC classifiers are trained on the same MultiCONAN dataset and labels that define the generator's conditioning and preference targets, so the SC metric rewards outputs that mimic the training distribution. This makes the headline 38% SC improvement partially self-referential. Independent signals—ROUGE, BERTScore, and the small human win-rate study—are not circular, but they do not rescue the SC-based headline, which is the paper's most prominent quantitative claim. The self-citations to Hengle et al. (2024) and Gupta et al. (2023) for the dataset and SC metric are not load-bearing for the method's derivation; the circularity comes from shared training data rather than from the citations themselves.

Assumptions & free parameters 2 free parameters · 6 assumptions · 0 invented entities

The central claims rest on the reliability of newly annotated emotion labels, on classifier-based conformity metrics that are trained on the same data as the model, and on a design choice (strategy prefixes before emotion prefixes) that is not ablated in reverse order. These are assumptions the reader must grant, not results the paper proves.

free parameters (2)
  • Number of virtual tokens (VT) = 3
    Chosen as optimal from {3,5,7,10} based on parameter count and performance; controls prefix capacity and the reported 589,824 trainable parameters.
  • ORPO loss mixing weight epsilon
    Equation 5 weights the ORPO loss by epsilon, but the value is never reported, leaving an unreported hand-tuned constant in the central optimization.
assumptions (6)
  • domain assumption Emotion labels from five annotators are accurate and consistent
    The MultiCONAN corpus is entirely built on these labels; inter-annotator agreement is reported for the annotation exercise, but no external validation of emotion assignments is provided.
  • domain assumption Strategy and emotion are sufficient conditioning attributes
    The task formulation in Section 4 defines counterspeech generation as a function of hate speech, strategy, and emotion, excluding other contextual factors such as conversation history or audience.
  • ad hoc to paper Hierarchical order with strategy prefixes frozen before emotion prefixes is beneficial
    Phase 1 assumes that optimizing strategy prefixes first and then emotion prefixes while freezing the strategy prefixes is the right inductive bias; no proof or alternative ordering is tested.
  • domain assumption Dataset counterspeech is always preferred over model output
    Phase 2 sets ground-truth counterspeech as the chosen response and model-generated text as rejected in ORPO, assuming the human-written response is always better for every hate speech instance.
  • domain assumption Classifier-based conformity evaluation is valid
    Section 5.2 uses RoBERTa-large SC/EC/TC classifiers trained on MultiCONAN to score attribute conformity; their accuracy is measured in-distribution, and the paper's own Limitation section concedes metrics diverge from human perception.
  • standard math ORPO odds-ratio loss is a valid alignment objective
    Equations 3 to 6 import the odds-ratio preference objective from Hong et al. (2024) without proving it; the paper relies on that prior result as background.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Counterspeech the ultimate shield! Multi-Conditioned Counterspeech Generation through Attributed Prefix Learning." pith.science (2026). https://pith.science/paper/BTEXGJAD

@misc{pith2026250511958,
  author       = {Pith},
  title        = {Pith review of: Counterspeech the ultimate shield! Multi-Conditioned Counterspeech Generation through Attributed Prefix Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BTEXGJAD}},
  note         = {Machine review of arXiv:2505.11958}
}
read the original abstract

Counterspeech has proven to be a powerful tool to combat hate speech online. Previous studies have focused on generating counterspeech conditioned only on specific intents (single attributed). However, a holistic approach considering multiple attributes simultaneously can yield more nuanced and effective responses. Here, we introduce HiPPrO, Hierarchical Prefix learning with Preference Optimization, a novel two-stage framework that utilizes the effectiveness of attribute-specific prefix embedding spaces hierarchically optimized during the counterspeech generation process in the first phase. Thereafter, we incorporate both reference and reward-free preference optimization to generate more constructive counterspeech. Furthermore, we extend IntentCONANv2 by annotating all 13,973 counterspeech instances with emotion labels by five annotators. HiPPrO leverages hierarchical prefix optimization to integrate these dual attributes effectively. An extensive evaluation demonstrates that HiPPrO achieves a ~38 % improvement in intent conformity and a ~3 %, ~2 %, ~3 % improvement in Rouge-1, Rouge-2, and Rouge-L, respectively, compared to several baseline models. Human evaluations further substantiate the superiority of our approach, highlighting the enhanced relevance and appropriateness of the generated counterspeech. This work underscores the potential of multi-attribute conditioning in advancing the efficacy of counterspeech generation systems. Our code is available on Github and dataset is open-sourced on Hugging-face.

Figures

Figures reproduced from arXiv: 2505.11958 by the authors.

Figure 1
Figure 1. An illustration of the output of existing meth [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Our proposed model, HiPPrO, follows a two-phase pipeline. In Phase 1, we train the prefix parameters (α, β, γ, δ) associated with prefix adapters (Fα, Fβ) on encoder side and (Fγ, Fδ) on the decoder side. For α and β, the model is trained with hate speech and strategy, separated by </s> as input. While training γ and δ, the optimal parameters, α ∗ and β ∗ , are kept fixed, and it includes hate speech, strategy, and … view at source ↗
Figure 3
Figure 3. Visual exploration of various attribute distribution present in the [PITH_FULL_IMAGE:figures/full_fig_p021_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

60 extracted references · 22 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]

    Chittaranjan Andrade. 2019. https://doi.org/10.4103/IJPSYM.IJPSYM_193_19 The p value and statistical significance: Misunderstandings, explanations, challenges, and alternatives . Indian Journal of Psychological Medicine, 41(3):210--215

    doi:10.4103/ijpsym.ijpsym_193_19 2019

  4. [4]

    Md Rabiul Awal, Rui Cao, Roy Ka-Wei Lee, and Sandra Mitrovi \' c . 2021. Angrybert: Joint learning target and emotion for hate speech detection. In Advances in Knowledge Discovery and Data Mining, pages 701--713, Cham. Springer International Publishing

    work page 2021

  5. [5]

    Satanjeev Banerjee and Alon Lavie. 2005. https://aclanthology.org/W05-0909 METEOR : An automatic metric for MT evaluation with improved correlation with human judgments . In Proceedings of the ACL Workshop on Intrinsic and Extrinsic Evaluation Measures for Machine Translation and/or Summarization , pages 65--72, Ann Arbor, Michigan. Association for Comput...

    2005

  6. [6]

    Susan Benesch, Derek Ruths, Kelly P Dillon, Haji Mohammad Saleem, and Lucas Wright. 2016 a . Considerations for successful counterspeech. a report for public safety canada under the kanishka project

    work page 2016

  7. [7]

    Susan Benesch, Derek Ruths, Kelly P Dillon, Haji Mohammad Saleem, and Lucas Wright. 2016 b . Counterspeech on twitter: A field study

    work page 2016

  8. [8]

    Susan Benesch, Derek Ruths, Kelly P Dillon, Haji Mohammad Saleem, and Lucas Wright. 2016 c . Counterspeech on twitter: A field study

    work page 2016

Show all 60 references
  1. [9]

    Pooja Bhojraj, Pratibha Mudgal, and Sipali Panda. 2024. Submental intubation in a case of panfacial fractures in a paediatric patient: A case report. Sri Lankan Journal of Anaesthesiology, 32(2)

    2024

  2. [10]

    Helena Bonaldi, Sara Dellantonio, Serra Sinem Tekiro g lu, and Marco Guerini. 2022. https://doi.org/10.18653/v1/2022.emnlp-main.549 Human-machine collaboration approaches to build a dialogue dataset for hate speech countering . In Proceedings of the 2022 Conference on Empirica...

    2022 doi

  3. [11]

    Tanmoy Chakraborty and Sarah Masud. 2022. https://doi.org/10.1145/3522598.3522601 Nipping in the bud: detection, diffusion and mitigation of hate speech on social media . SIGWEB Newsl., 2022(Winter)

    2022

  4. [12]

    Eshwar Chandrasekharan, Umashanthi Pavalanathan, Anirudh Srinivasan, Adam Glynn, Jacob Eisenstein, and Eric Gilbert. 2017. You can’t stay here: The efficacy of reddit’s 2015 ban examined through hate speech. Proc. ACM Hum.-Comput. Interact., 1(CSCW)

    2017

  5. [13]

    Chi, Jeff Dean, Jacob Devlin, Adam Roberts, Denny Zhou, Quoc V

    Hyung Won Chung, Le Hou, Shayne Longpre, Barret Zoph, Yi Tay, William Fedus, Yunxuan Li, Xuezhi Wang, Mostafa Dehghani, Siddhartha Brahma, Albert Webson, Shixiang Shane Gu, Zhuyun Dai, Mirac Suzgun, Xinyun Chen, Aakanksha Chowdhery, Alex Castro-Ros, Marie Pellat, Kevin Robinso...

    2022 arXiv

  6. [14]

    Yi-Ling Chung, Gavin Abercrombie, Florence Enock, Jonathan Bright, and Verena Rieser. 2023. Understanding counterspeech for online harm mitigation

    2023

  7. [15]

    Yi-Ling Chung, Elizaveta Kuzmenko, Serra Sinem Tekiro g lu, and Marco Guerini. 2019. Conan-counter narratives through nichesourcing: a multilingual dataset of responses to fight online hate speech. In Proceedings of the 57th Annual Meeting of the Association for Computational ...

    2019

  8. [16]

    Jacob Cohen. 1960. A coefficient of agreement for nominal scales. Educational and psychological measurement, 20(1):37--46

    1960

  9. [17]

    Zhang, Han Bao, Hanwei Xu, Haocheng Wang, Haowei Zhang, Honghui Ding, Huajian Xin, Huazuo Gao, Hui Li, Hui Qu, J

    DeepSeek-AI, Aixin Liu, Bei Feng, Bing Xue, Bingxuan Wang, Bochao Wu, Chengda Lu, Chenggang Zhao, Chengqi Deng, Chenyu Zhang, Chong Ruan, Damai Dai, Daya Guo, Dejian Yang, Deli Chen, Dongjie Ji, Erhang Li, Fangyun Lin, Fucong Dai, Fuli Luo, Guangbo Hao, Guanting Chen, Guowei L...

    2025 arXiv

  10. [18]

    Giovanni Di Leo and Francesco Sardanelli. 2020. https://doi.org/10.1186/s41747-020-0145-y Statistical significance: p value, 0.05 threshold, and applications to radiomics—reasons for a conservative approach . European Radiology Experimental, 4:18

    2020 doi

  11. [19]

    Matthijs Douze, Alexandr Guzhva, Chengqi Deng, Jeff Johnson, Gergely Szilvasy, Pierre-Emmanuel Mazaré, Maria Lomeli, Lucas Hosseini, and Hervé Jégou. 2025. http://arxiv.org/abs/2401.08281 The faiss library

    2025 arXiv

  12. [20]

    Margherita Fanton, Helena Bonaldi, Serra Sinem Tekiro g lu, and Marco Guerini. 2021 a . Human-in-the-loop for data collection: a multi-target counter narrative dataset to fight online hate speech. In Proceedings of the 59th Annual Meeting of the Association for Computational L...

    2021

  13. [21]

    Margherita Fanton, Helena Bonaldi, Serra Sinem Tekiroglu, and Marco Guerini. 2021 b . Human-in-the-loop for data collection: a multi-target counter narrative dataset to fight online hate speech. In Proceedings of the 59th Annual Meeting of the Association for Computational Lin...

    2021

  14. [22]

    Joseph L Fleiss. 1971. Measuring nominal scale agreement among many raters. Psychological bulletin, 76(5):378

    1971

  15. [23]

    Tanmay Garg, Sarah Masud, Tharun Suresh, and Tanmoy Chakraborty. 2023. Handling bias in toxic speech detection: A survey. ACM Computing Surveys, 55(13s):1--32

    2023

  16. [24]

    Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Alex Vaughan, Amy Yang, Angela Fan, Anirudh Goyal, Anthony Hartshorn, Aobo Yang, Archi Mitra, Archie Sravankumar, Artem Korenev, Art...

    2024 arXiv

  17. [25]

    Shad Akhtar

    Rishabh Gupta, Shaily Desai, Manvi Goel, Anil Bandhakavi, Tanmoy Chakraborty, and Md. Shad Akhtar. 2023. https://doi.org/10.18653/v1/2023.acl-long.318 Counterspeeches up my sleeve! intent distribution learning and persistent fusion for intent-conditioned counterspeech generati...

    2023 doi

  18. [26]

    Laura Hanu and Unitary team . 2020. Detoxify. Github. https://github.com/unitaryai/detoxify

    2020

  19. [27]

    Amey Hengle, Aswini Kumar, Anil Bandhakavi, and Tanmoy Chakraborty. 2025. http://arxiv.org/abs/2501.17581 Cseval: Towards automated, multi-dimensional, and reference-free counterspeech evaluation using auto-calibrated llms

    2025 arXiv

  20. [28]

    Amey Hengle, Aswini Kumar, Sahajpreet Singh, Anil Bandhakavi, Md Shad Akhtar, and Tanmoy Chakroborty. 2024. http://arxiv.org/abs/2403.10088 Intent-conditioned and non-toxic counterspeech generation using multi-task instruction tuning with rlaif

    2024 arXiv

  21. [29]

    Jiwoo Hong, Noah Lee, and James Thorne. 2024. Reference-free monolithic preference optimization with odds ratio. arXiv preprint arXiv:2403.07691

    2024 arXiv

  22. [30]

    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, Gianna Lengyel, Guillaume Bour, Guillaume Lample, Lélio Renard Lavaud, Lucile Saulnier, Marie-Anne...

    2024 arXiv

  23. [31]

    Jaylen Jones, Lingbo Mo, Eric Fosler-Lussier, and Huan Sun. 2024. https://doi.org/10.18653/v1/2024.naacl-short.14 A multi-aspect framework for counter narrative evaluation using large language models . In Proceedings of the 2024 Conference of the North American Chapter of the ...

    2024 doi

  24. [32]

    Kalev Leetaru. 2019. Online toxicity is as old as the web itself but the return to communities may help. forbes magazin

    2019

  25. [33]

    Brian Lester, Rami Al-Rfou, and Noah Constant. 2021. https://doi.org/10.18653/v1/2021.emnlp-main.243 The power of scale for parameter-efficient prompt tuning . In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 3045--3059, Online a...

    2021 doi

  26. [34]

    Xiang Lisa Li and Percy Liang. 2021. https://doi.org/10.18653/v1/2021.acl-long.353 Prefix-tuning: Optimizing continuous prompts for generation . In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conferen...

    2021 doi

  27. [35]

    Chin-Yew Lin. 2004. https://aclanthology.org/W04-1013 ROUGE : A package for automatic evaluation of summaries . In Text Summarization Branches Out, pages 74--81, Barcelona, Spain. Association for Computational Linguistics

    2004

  28. [36]

    Pengfei Liu, Weizhe Yuan, Jinlan Fu, Zhengbao Jiang, Hiroaki Hayashi, and Graham Neubig. 2023 a . Pre-train, prompt, and predict: A systematic survey of prompting methods in natural language processing. ACM Computing Surveys, 55(9):1--35

    2023

  29. [37]

    Yajing Liu, Yuning Lu, Hao Liu, Yaozu An, Zhuoran Xu, Zhuokun Yao, Baofeng Zhang, Zhiwei Xiong, and Chenguang Gui. 2023 b . https://doi.org/10.1109/CVPR52729.2023.01048 Hierarchical prompt learning for multi-task learning . In 2023 IEEE/CVF Conference on Computer Vision and Pa...

    2023

  30. [38]

    Shad Akhtar, and Tanmoy Chakraborty

    Sarah Masud, Manjot Bedi, Mohammad Aflah Khan, Md. Shad Akhtar, and Tanmoy Chakraborty. 2022. Proactively reducing the hate intensity of online posts via hate speech normalization. In KDD '22: The 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, pages 3524--3534. ACM

    2022

  31. [39]

    Sarah Masud, Subhabrata Dutta, Sakshi Makkar, Chhavi Jain, Vikram Goyal, Amitava Das, and Tanmoy Chakraborty. 2021 a . https://doi.org/10.1109/ICDE51399.2021.00050 Hate is the new infodemic: A topic-aware modeling of hate speech diffusion on twitter . In 2021 IEEE 37th Interna...

    2021

  32. [40]

    Sarah Masud, Subhabrata Dutta, Sakshi Makkar, Chhavi Jain, Vikram Goyal, Amitava Das, and Tanmoy Chakraborty. 2021 b . Hate is the new infodemic: A topic-aware modeling of hate speech diffusion on twitter. In 2021 IEEE 37th International Conference on Data Engineering (ICDE), ...

    2021

  33. [41]

    Shad Akhtar, and Tanmoy Chakraborty

    Sarah Masud, Mohammad Aflah Khan, Md. Shad Akhtar, and Tanmoy Chakraborty. 2023. http://arxiv.org/abs/2311.09834 Overview of the hasoc subtrack at fire 2023: Identification of tokens contributing to explicit hate in english by span detection

    2023 arXiv

  34. [42]

    Binny Mathew, Punyajoy Saha, Hardik Tharad, Subham Rajgaria, Prajwal Singhania, Suman Kalyan Maity, Pawan Goyal, and Animesh Mukherjee. 2019. Thou shalt not hate: Countering online hate speech. In Proceedings of the international AAAI conference on web and social media, volume...

    2019

  35. [44]

    Long Ouyang, Jeff 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 Christiano, Jan Leike, and...

    2022 arXiv

  36. [45]

    Sara Parker and Derek Ruths. 2023. https://doi.org/10.1073/pnas.2209384120 Is hate speech detection the solution the world wants? Proceedings of the National Academy of Sciences, 120(10):e2209384120

    2023 doi

  37. [46]

    Jing Qian, Anna Bethke, Yinyin Liu, Elizabeth Belding, and William Yang Wang. 2019. A benchmark dataset for learning to intervene in online hate speech. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Co...

    2019

  38. [47]

    Manning, and Chelsea Finn

    Rafael Rafailov, Archit Sharma, Eric Mitchell, Stefano Ermon, Christopher D. Manning, and Chelsea Finn. 2023. http://arxiv.org/abs/2305.18290 Direct preference optimization: Your language model is secretly a reward model

    2023 arXiv

  39. [48]

    Nils Reimers and Iryna Gurevych. 2019. https://doi.org/10.18653/v1/D19-1410 Sentence- BERT : Sentence embeddings using S iamese BERT -networks . In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference...

    2019 doi

  40. [49]

    Amrita Saha, Ghulam Ahmed Ansari, Abhishek Laddha, Karthik Sankaranarayanan, and Soumen Chakrabarti. 2019. Complex program induction for querying knowledge bases in the absence of gold programs. Transactions of the Association for Computational Linguistics, 7:185--200

    2019

  41. [50]

    Punyajoy Saha, Kanishk Singh, Adarsh Kumar, Binny Mathew, and Animesh Mukherjee. 2022. Countergedi: A controllable approach to generate polite, detoxified and emotional counterspeech. In Proceedings of the Thirty-First International Joint Conference on Artificial Intelligence,...

    2022

  42. [51]

    Carla Schieb and Mike Preuss. 2016. Governing hate speech by means of counterspeech on facebook. In 66th ICA Annual Conference, at Fukuoka, Japan, pages 1--23

    2016

  43. [52]

    Emily Sheng, Kai-Wei Chang, Prem Natarajan, and Nanyun Peng. 2020. Towards controllable biases in language generation. In Findings of the Association for Computational Linguistics: EMNLP 2020, pages 3239--3254. Association for Computational Linguistics

    2020

  44. [53]

    Sahajpreet Singh, Sarah Masud, and Tanmoy Chakraborty. 2024. http://arxiv.org/abs/2407.19498 Independent fact-checking organizations exhibit a departure from political neutrality

    2024 arXiv

  45. [54]

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. In Advances in Neural Information Processing Systems, volume 30. Curran Associates, Inc

    2017

  46. [55]

    Jiaan Wang, Yunlong Liang, Fandong Meng, Zengkui Sun, Haoxiang Shi, Zhixu Li, Jinan Xu, Jianfeng Qu, and Jie Zhou. 2023. https://doi.org/10.18653/v1/2023.newsum-1.1 Is C hat GPT a good NLG evaluator? a preliminary study . In Proceedings of the 4th New Frontiers in Summarizatio...

    2023 doi

  47. [56]

    Lucas Wright, Derek Ruths, Kelly P Dillon, Haji Mohammad Saleem, and Susan Benesch. 2017. Vectors for counterspeech on twitter. In Proceedings of the First Workshop on Abusive Language Online, pages 57--62

    2017

  48. [57]

    Neemesh Yadav, Sarah Masud, Vikram Goyal, Vikram Goyal, Md Shad Akhtar, and Tanmoy Chakraborty. 2024. http://arxiv.org/abs/2406.03953 Tox-bart: Leveraging toxicity attributes for explanation generation of implicit hate speech

    2024

  49. [58]

    Jingfeng Yang, Hongye Jin, Ruixiang Tang, Xiaotian Han, Qizhang Feng, Haoming Jiang, Shaochen Zhong, Bing Yin, and Xia Hu. 2024. Harnessing the power of llms in practice: A survey on chatgpt and beyond. ACM Transactions on Knowledge Discovery from Data, 18(6):1--32

    2024

  50. [59]

    Weinberger, and Yoav Artzi

    Tianyi Zhang, Varsha Kishore, Felix Wu, Kilian Q. Weinberger, and Yoav Artzi. 2020 a . http://arxiv.org/abs/1904.09675 Bertscore: Evaluating text generation with bert

    2020 arXiv

  51. [60]

    Yizhe Zhang, Siqi Sun, Michel Galley, Yen-Chun Chen, Chris Brockett, Xiang Gao, Jianfeng Gao, Jingjing Liu, and William B Dolan. 2020 b . Dialogpt: Large-scale generative pretraining for conversational response generation. In ACL: System Demonstrations

    2020

  52. [61]

    Wanzheng Zhu and Suma Bhat. 2021. https://doi.org/10.18653/v1/2021.findings-acl.12 Generate, prune, select: A pipeline for counterspeech generation against online hate speech . In Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021, pages 134--149, Onlin...

    2021 doi

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.