Elamur Research⌕
☰
home / papers / 2026 / More Than Sum of Its Parts: Deciphering Intent Shifts in Multimodal Hate Speech Detection

More Than Sum of Its Parts: Deciphering Intent Shifts in Multimodal Hate Speech Detection

Paper · 2026 · Runze Sun, Yu Zheng, Zexuan Xiong et al.

Summary

The paper addresses implicit multimodal hate speech, where a text and an image look harmless on their own but turn hateful together. The reverse also happens: offensive-looking parts lose their toxicity in context. The authors argue that binary labels hide these shifts. They propose the Stratified Multimodal Interaction (SMI) paradigm, which labels the text alone, the image alone and the combined pair separately. This yields eight interaction patterns grouped into Easy, Normal and Hard levels. The detection task becomes six-class classification with natural-language explanations.

Using SMI, they build the H-VLI benchmark. They filter real samples from existing datasets, add LLM-generated "injected" texts, and verify everything with human annotators. They also present ARCADE, a debate framework modeled on a courtroom. A Prosecutor agent assumes the post is hateful, a Defender assumes it is not, and a Judge decides. A gating step routes each sample to a lighter or a full debate track.

ARCADE beats state-of-the-art baselines on H-VLI, especially on implicit cases, and stays competitive on older benchmarks. The work thus offers a harder and cleaner testbed and an interpretable way to reason about cross-modal intent.

Key points

  • H-VLI size and agreement: the raw pool of 8,438 image–text pairs is reduced to 5,569 final samples. Fleiss' κ rises from 0.59 to 0.94, compared with roughly 0.15 for MMHS150K.
  • Difficulty split: Easy covers 62.5% of samples (consistent modalities), Normal 23.6% (contextual neutralization: patterns 100 and 010), and Hard 13.8% (implicit emergence or inversion: patterns 001 and 110).
  • Data sources: the benchmark draws on MMHS150K, FHM and Google Search images. Qwen-plus generates supplementary texts, and Qwen3-VL-plus is used for intent alignment filtering.
  • ARCADE gating: the gating function Φ(S) sends over 90% of samples with an explicitly hateful single modality to Track I. Most 000 samples go to the full debate track or a summary dismissal.
  • Limitations:
    • ARCADE is English-only.
    • The debate is computationally costly.
    • Easy samples may see a slight performance drop.
    • The synthetic samples may not reflect real social-media noise.
    • The dataset will be released under a restrictive, research-only license.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksSocial Sciences · 1,772 worksSociology and Political Science · 577 works

🧭 Topics 3Research subjects and languages the work is about.Sentiment & emotion · 972 worksVision & language · 657 worksBias, fairness & safety · 517 works

Cite

Cite as Runze Sun, Yu Zheng, Zexuan Xiong, Zhongjin Qu, Lei Chen, Jie Zhou and Jiwen Lu. 2026. More Than Sum of Its Parts: Deciphering Intent Shifts in Multimodal Hate Speech Detection. Findings of the Association for Computational Linguistics: ACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.findings-acl.974

@inproceedings{sun2026more,
  title = {{More Than Sum of Its Parts: Deciphering Intent Shifts in Multimodal Hate Speech Detection}},
  author = {Runze Sun, Yu Zheng, Zexuan Xiong, Zhongjin Qu, Lei Chen, Jie Zhou and Jiwen Lu},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.974},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.974}
}

References

56 references, 24 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Hyeseon Ahn, Youngwook Kim, Jungin Kim, and Yo-Sub Han. 2024. SharedCon: Implicit hate speech detection using shared semantics. In Findings of the Association for Computational Linguistics: ACL 2024, pages 10444–10455, Bangkok, Thailand. Association for Computational Linguistics. find
  2. Rohan Anil, Sebastian Borgeaud, Jean-Baptiste Alayrac, Jiahui Yu, Radu Soricut, Johan Schalkwyk, Andrew M. Dai, Anja Hauth, Katie Millican, David Silver, Melvin Johnson, Ioannis Antonoglou, Julian Schrittwieser, Amelia Glaese, Jilin Chen, Emily Pitler, Timothy Lillicrap, Angeliki Lazaridou, Orhan Fi… — Gemini: A Family of Highly Capable Multimodal Models
  3. Jinze Bai, Shuai Bai, Yunfei Chu, Zeyu Cui, Kai Dang, Xiaodong Deng, Yang Fan, Wenbin Ge, Yu Han, Fei Huang, Binyuan Hui, Luo Ji, Mei Li, Junyang Lin, Runji Lin, Dayiheng Liu, Gao Liu, Chengqiang Lu, Keming Lu, and 29 others. 2023a. Qwen technical report. arXiv preprint arXiv:2309.16609.
  4. Jinze Bai, Shuai Bai, Shusheng Yang, Shijie Wang, Sinan Tan, Peng Wang, Junyang Lin, Chang Zhou, and Jingren Zhou. 2023b. Qwen-vl: A versatile vision-language model for understanding, localization, text reading, and beyond. arXiv preprint arXiv:2308.12966.
  5. Shuai Bai, Yuxuan Cai, Ruizhe Chen, Keqin Chen, Xionghui Chen, Zesen Cheng, Lianghao Deng, Wei Ding, Chang Gao, Chunjiang Ge, Wenbin Ge, Zhifang Guo, Qidong Huang, Jie Huang, Fei Huang, Binyuan Hui, Shutong Jiang, Zhaohai Li, Mingsheng Li, and 45 others. 2025. Qwen3-vl technical report. arXiv prepri…
  6. Rui Cao, Ming Shan Hee, Adriel Kuek, Wen-Haw Chong, Roy Ka-Wei Lee, and Jing Jiang. 2023. Pro-cap: Leveraging a frozen vision-language model for hateful meme detection. In ACM MM, pages 5244– 5252. find
  7. Rui Cao, Roy Ka-Wei Lee, Wen-Haw Chong, and Jing Jiang. 2022. Prompting for multimodal hateful meme classification. In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 321–332, Abu Dhabi, United Arab Emirates. Association for Computational Linguistics. find
  8. Tommaso Caselli, Valerio Basile, Jelena Mitrović, Inga Kartoziya, and Michael Granitzer. 2020. I feel offended, don’t be abusive! implicit/explicit messages in offensive and abusive language. In Proceedings of the Twelfth Language Resources and Evaluation Conference, pages 6193–6202, Marseille, Fra…
  9. Iginio Gagliardone, Danit Gal, Thiago Alves, and Gabriela Martinez. 2015. Countering online hate speech. UNESCO Publishing. find
  10. Sreyan Ghosh, Manan Suri, Purva Chiniya, Utkarsh Tyagi, Sonal Kumar, and Dinesh Manocha. 2023. Cosyn: Detecting implicit hate speech in online conversations using a context synergized hyperbolic network. In EMNLP, pages 6159–6173. find
  11. Raul Gomez, Jaume Gibert, Lluis Gomez, and Dimosthenis Karatzas. 2020. Exploring hate speech detection in multimodal publications. In WACV, pages 1470–1478. find
  12. Chen Han, Wenzhen Zheng, and Xijin Tang. 2025. Debate-to-detect: Reformulating misinformation detection as a real-world debate with large language models. In EMNLP, pages 15125–15140. find
  13. Nazanin Jafari, James Allan, and Sheikh Muhammad Sarwar. 2024. Target span detection for implicit harmful content. In SIGIR, pages 117–122. find
  14. Wenliang Dai, Junnan Li, Dongxu Li, Anthony Tiong, Junqi Zhao, Weisheng Wang, Boyang Li, Pascale N Fung, and Steven Hoi. 2023. Instructblip: Towards general-purpose vision-language models with instruction tuning. NeurIPS, 36:49250–49267.
  15. Weiqiang Jin, Dafu Su, Tao Tao, Xiujun Wang, Ningwei Wang, and Biao Zhao. 2025. Courtroom-fnd: a multi-role fake news detection method based on argument switching-based courtroom debate. Journal of King Saud University Computer and Information Sciences, 37(3):33. find
  16. Anthony Cortese. 2005. Bloomsbury Publishing. find
  17. Thomas Davidson, Dana Warmsley, Michael Macy, and Ingmar Weber. 2017. Automated hate speech detection and the problem of offensive language. In AAAI, pages 512–515.
  18. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. BERT: Pre-training of deep bidirectional transformers for language understanding. In NAACL-HLT.
  19. Alexey Dosovitskiy. 2021. An image is worth 16x16 words: Transformers for image recognition at scale. In ICLR.
  20. Vishwajeet Dwivedy and Pradeep Kumar Roy. 2023. Deep feature fusion for hate speech detection: a transfer learning approach. Multimedia Tools Appl., 82(23). find
  21. Elisabetta Fersini, Francesca Gasparini, Giulia Rizzi, Aurora Saibene, Berta Chulvi, Paolo Rosso, Alyssa Lees, and Jeffrey Sorensen. 2022a. Semeval-2022 task 5: Multimedia automatic misogyny identification. In Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval-2022), page…
  22. Elisabetta Fersini, Francesca Gasparini, Giulia Rizzi, Aurora Saibene, Berta Chulvi, Paolo Rosso, Alyssa Lees, and Jeffrey Sorensen. 2022b. SemEval-2022 task 5: Multimedia automatic misogyny identification. In Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval-2022), page…
  23. Zhenchao Jin. 2022. Imagedl: Search and download images from specific websites. https://github.com/CharlesPikachu/imagedl/. ↗
  24. Prashant Kapil and Asif Ekbal. 2025. A transformer based multi task learning approach to multimodal hate speech detection. Nat. Lang. Process. J., 11. find
  25. Douwe Kiela, Suvrat Bhooshan, Hamed Firooz, Ethan Perez, and Davide Testuggine. 2019. Supervised multimodal bitransformers for classifying images and text. arXiv preprint arXiv:1909.02950. find
  26. Douwe Kiela, Hamed Firooz, Aravind Mohan, Vedanuj Goswami, Amanpreet Singh, Pratik Ringshia, and Davide Testuggine. 2020. The hateful memes challenge: Detecting hate speech in multimodal memes. NeurIPS, 33.
  27. Sandeep Kumar, Abhijit A Nargund, and Vivek Sridhar. 2025. CourtEval: A courtroom-based multi-agent evaluation framework. In Findings of ACL, pages 25875–25887. find
  28. Hugo Laurençon, Lucile Saulnier, Leo Tronchon, Stas Bekman, Amanpreet Singh, Anton Lozhkov, Thomas Wang, Siddharth Karamcheti, Alexander Rush, Douwe Kiela, Matthieu Cord, and Victor Sanh. 2023. Obelics: An open web-scale filtered dataset of interleaved image-text documents. NeurIPS, 36:71683–71702.
  29. Junnan Li, Dongxu Li, Caiming Xiong, and Steven Hoi. 2022. Blip: Bootstrapping language-image pre-training for unified vision-language understanding and generation. In ICML, pages 12888–12900.
  30. Junnan Li, Ramprasaath Selvaraju, Akhilesh Gotmare, Shafiq Joty, Caiming Xiong, and Steven Chu Hong Hoi. 2021. Align before fuse: Vision and language representation learning with momentum distillation. NeurIPS, 34:9694–9705. find
  31. Someen Park, Jaehoon Kim, Seungwan Jin, Sohyun Park, and Kyungsik Han. 2024. PREDICT: Multi-agent-based debate simulation for generalized hate speech detection. In EMNLP, pages 20963–20987. find
  32. Yifei Li, Wenzhao Zheng, Yanran Zhang, Runze Sun, Yu Zheng, Lei Chen, Jie Zhou, and Jiwen Lu. 2025. Skyra: Ai-generated video detection via grounded artifact reasoning. arXiv preprint arXiv:2512.15693. — [arXiv:2512.15693](https://arxiv.org/abs/2512.15693) find
  33. Shraman Pramanick, Dimitar Dimitrov, Rituparna Mukherjee, Shivam Sharma, Md Shad Akhtar, Preslav Nakov, and Tanmoy Chakraborty. 2021a. Detecting harmful memes and their targets. arXiv preprint arXiv:2110.00413.
  34. Tian Liang, Zhiwei He, Wenxiang Jiao, Xing Wang, Yan Wang, Rui Wang, Yujiu Yang, Shuming Shi, and Zhaopeng Tu. 2024. Encouraging divergent thinking in large language models through multi-agent debate. In EMNLP, pages 17889–17904.
  35. Shraman Pramanick, Shivam Sharma, Dimitar Dimitrov, Md Shad Akhtar, Preslav Nakov, and Tanmoy Chakraborty. 2021b. Momenta: A multimodal framework for detecting harmful memes and their targets. In Findings of EMNLP, pages 4439–4455.
  36. Haotian Liu, Chunyuan Li, Yuheng Li, and Yong Jae Lee. 2024. Improved baselines with visual instruction tuning. In CVPR, pages 26296–26306.
  37. Naquee Rizwan and 1 others. 2025. Exploring the limits of zero shot vision language models for hate meme detection. In ICWSM, volume 19. find
  38. Rui Lu, Jinhe Bi, Yunpu Ma, Feng Xiao, Yuntao Du, and Yijun Tian. 2025. Mv-debate: Multi-view agent debate with dynamic reflection gating for multimodal harmful content detection in social media. arXiv preprint arXiv:2508.05557. — [arXiv:2508.05557](https://arxiv.org/abs/2508.05557) find
  39. Furqan Khan Saddozai, Sahar K Badri, Daniyal Alghaz-zawi, Asad Khattak, and Muhammad Zubair Asghar. 2025. Multimodal hate speech detection: a novel deep learning framework for multilingual text and images. PeerJ Computer Science, 11:e2801. find
  40. Xiaochen Ma, Guozheng Rao, Lina Xu, Xin Wang, Zaiming Fan, and Zhe Zhang. 2025. Guided and knowledgeable multi-agent debate for fact verification. Expert Systems with Applications, page 130103. find
  41. Anna Schmidt and Michael Wiegand. 2017. A survey on hate speech detection using natural language processing. In Proceedings of the Fifth International Workshop on Natural Language Processing for Social Media, pages 1–10.
  42. Mari J Matsuda. 2018. Words that wound: Critical race theory, assaultive speech, and the first amendment. Routledge. find
  43. Jingbiao Mei, Jinghong Chen, Weizhe Lin, Bill Byrne, and Marcus Tomalin. 2024. Improving hateful meme detection through retrieval-guided contrastive learning. In ACL, pages 5333–5347. find
  44. Chikashi Nobata, Joel Tetreault, Achint Thomas, Yashar Mehdad, and Yi Chang. 2016. Abusive language detection in online user content. In WWW, pages 145–153. find
  45. OpenAI. 2024. Gpt-4o system card. https://openai.com/research/gpt-4o-system-card. ↗
  46. OpenAI. 2025a. Gpt-5. https://platform.openai.com/docs/models/gpt-5. Accessed: 2026-01-04. ↗
  47. OpenAI. 2025b. Gpt-5-mini. https://platform.openai.com/docs/models/gpt-5-mini. Accessed: 2026-01-04. ↗
  48. OpenAI, Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, Red Avila, Igor Babuschkin, Suchir Balaji, Valerie Balcom, Paul Baltescu, Haiming Bao, Mohammad Bavarian, Jeff Belgum, and 262 others.… — GPT-4 Technical Report
  49. Piyush Sharma, Nan Ding, Sebastian Goodman, and Radu Soricut. 2018. Conceptual captions: A cleaned, hypernymed, image alt-text dataset for automatic image captioning. In ACL, pages 2556–2565.
  50. Shardul Suryawanshi, Bharathi Raja Chakravarthi, Mihael Arcan, and Paul Buitelaar. 2020. Multimodal meme dataset (multioff) for identifying offensive content in image and text. In Proceedings of the Second Workshop on Trolling, Aggression and Cyberbullying (TRAC-2020). Association for Computational…
  51. Alexander Tsesis. 2002. Destructive messages: How hate speech paves the way for harmful social movements, volume 27. NYU Press. find
  52. Xiuxian Wang, Lanjun Wang, Yuting Su, Hongshuo Tian, Guoqing Jin, and An-An Liu. 2025. Few-shot in-context learning for implicit semantic multimodal content detection and interpretation. IEEE TCSVT. find
  53. Yaqing Wang, Fenglong Ma, Zhiwei Jin, Ye Yuan, Guangxu Xun, Kishlay Jha, Lu Su, and Jing Gao. 2018. Eann: Event adversarial neural networks for multi-modal fake news detection. In Proceedings of the 24th acm sigkdd international conference on knowledge discovery & data mining, pages 849–857. find
  54. Zeerak Waseem and Dirk Hovy. 2016. Hateful symbols or hateful people? predictive features for hate speech detection on twitter. In NAACL SRW.
  55. Lu Wei, Liangzhi Li, Tong Xiang, Liu Xiao, and Noa Garcia. 2025. Cracking the code: Enhancing implicit hate speech detection through coding classification. In TrustNLP. find
  56. Jingjie Zeng, Liang Yang, Zekun Wang, Yuanyuan Sun, and Hongfei Lin. 2025. Sheep’s skin, wolf’s deeds: Are LLMs ready for metaphorical implicit hate speech? In ACL, pages 16657–16677. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/sun-2026-more-than-sum-its-parts/