Elamur Research⌕
☰
home / papers / 2026 / Benchmarking Deflection and Hallucination in Large Vision-Language Models

Benchmarking Deflection and Hallucination in Large Vision-Language Models

Paper · 2026 · Nicholas Moratelli, Christopher Davis, Leonardo F. R. Ribeiro et al.

Summary

Retrieval-augmented LVLMs are increasingly used for knowledge-based visual question answering (KB-VQA). Existing benchmarks have three gaps. They score only accuracy. They ignore clashes between image and text evidence. They become outdated as models memorize more facts. This paper argues that when retrieved evidence is missing or misleading, the right failure is a deflection (declining to answer), not a hallucination.

The authors built a curation pipeline with three steps:

  • It keeps only questions that four strong open-weight "gating" models cannot answer without retrieval.
  • It attaches gold and distractor contexts, both textual and visual, to each question.
  • It applies quality checks for solvability and difficulty.

The result is VLM-DeflectionBench, with 2,775 samples drawn from six existing datasets. Models are tested in four scenarios: parametric, oracle, realistic and adversarial. Answers are judged as correct, hallucinated or deflected.

Across 20 LVLMs, including proprietary ones, models seldom deflect reliably when evidence is noisy. Stricter prompts reduce hallucination but cause excessive refusals. Because the pipeline can be rerun with stronger models, the benchmark can stay difficult over time.

Key points

  • Sources: InfoSeek (1,646 samples), WebQA (826), E-VQA (185), MMDocRAG (58), MRAG-Bench (45) and ViQuAE (15). Only about 5.1% of the original samples survived filtering. The dataset has 1,246 unique questions and 2,717 unique query images. Each sample has on average 1.9 gold and 11.8 negative contexts.
  • Pipeline models:
    • Gating models: Gemma 3-27B, Qwen-2.5-VL-32B, InternVL3-38B and VL-Rethinker-72B.
    • Judge: GPT-4o with the SimpleQA prompt, whose labels CORRECT, INCORRECT and NOT ATTEMPTED map to accuracy, hallucination and deflection.
    • Distractor retrieval: EVA-CLIP, with Contriever re-ranking for text passages.
  • Prompt strictness is a blunt tool: Under the strictest prompt, Mistral-Small-3.1's deflection in the adversarial setting rose from 57.6% to 98.2%. However, its realistic-setting accuracy fell from 57.7% to 9.2%.
  • More distractors hurt: Going from 1 to 10 distractors cut Ovis2-16B's accuracy from 59.8% to 35.9%. Claude-Opus-4's accuracy dropped from 42.2% to 27.4%. Deflection barely increased to compensate.
  • Stated limitations:
    • Evaluation relies on an automatic judge rather than human annotation.
    • The source data is mostly text-centric.
    • Results come from a single run per query.
    • Only short-form QA is covered.

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

 Institutions

🎓University of CambridgeCambridge · United Kingdom243 works · 121,396 citations🎓University of Modena and Reggio EmiliaModena · Italy3 works · 3 citations

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksComputer Vision and Pattern Recognition · 2,869 works

🧭 Topics 1Research subjects and languages the work is about.Vision & language · 657 works

Cite

Cite as Nicholas Moratelli, Christopher Davis, Leonardo F. R. Ribeiro, Bill Byrne and Gonzalo Iglesias. 2026. Benchmarking Deflection and Hallucination in Large Vision-Language Models. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.acl-long.1307

@inproceedings{moratelli2026benchmarking,
  title = {{Benchmarking Deflection and Hallucination in Large Vision-Language Models}},
  author = {Nicholas Moratelli, Christopher Davis, Leonardo F. R. Ribeiro, Bill Byrne and Gonzalo Iglesias},
  booktitle = {Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  year = {2026},
  doi = {10.18653/v1/2026.acl-long.1307},
  url = {https://doi.org/10.18653/v1/2026.acl-long.1307}
}

References

44 references, 15 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Pravesh Agrawal, Szymon Antoniak, Emma Bou Hanna, Baptiste Bout, Devendra Chaplot, Jessica Chudnovsky, Diogo Costa, Baudouin De Monicault, Saurabh Garg, Theophile Gervet, and 1 others. 2024. Pixtral 12b. arXiv preprint arXiv:2410.07073. find
  2. Akari Asai, Zeqiu Wu, Yizhong Wang, Avi Sil, and Hannaneh Hajishirzi. 2024. Self-rag: Learning to retrieve, generate, and critique through self-reflection. In Proceedings of the International Conference on Learning Representations.
  3. Shuai Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Sibo Song, Kai Dang, Peng Wang, Shijie Wang, Jun Tang, Humen Zhong, Yuanzhi Zhu, Mingkun Yang, Zhaohai Li, Jianqiang Wan, Pengfei Wang, Wei Ding, Zheren Fu, Yiheng Xu, and 8 others. 2025. Qwen2.5-vl technical report. arXiv preprint arXiv:25…
  4. Davide Caffagni, Federico Cocchi, Luca Barsellotti, Nicholas Moratelli, Sara Sarto, Lorenzo Baraldi, Marcella Cornia, and Rita Cucchiara. 2024a. The revolution of multimodal large language models: A survey. In Proceedings of the Annual Meeting of the Association for Computational Linguistics.
  5. Davide Caffagni, Federico Cocchi, Nicholas Moratelli, Sara Sarto, Marcella Cornia, Lorenzo Baraldi, and Rita Cucchiara. 2024b. Wiki-llava: Hierarchical retrieval-augmented generation for multimodal llms. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops. find
  6. Yingshan Chang, Mridu Narang, Hisami Suzuki, Guihong Cao, Jianfeng Gao, and Yonatan Bisk. 2022. Webqa: Multihop and multimodal qa. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. find
  7. Wenhu Chen, Hexiang Hu, Xi Chen, Pat Verga, and William Cohen. 2022. Murag: Multimodal retrieval-augmented generator for open question answering over images and text. In Proceedings of the Conference on Empirical Methods in Natural Language Processing, pages 5558–5570.
  8. Yang Chen, Hexiang Hu, Yi Luan, Haitian Sun, Soravit Changpinyo, Alan Ritter, and Ming-Wei Chang. 2023. Can pre-trained vision and language models answer visual information-seeking questions? In Proceedings of the Conference on Empirical Methods in Natural Language Processing.
  9. Federico Cocchi, Nicholas Moratelli, Marcella Cornia, Lorenzo Baraldi, and Rita Cucchiara. 2025. Augmenting multimodal llms with self-reflective tokens for knowledge-based visual question answering. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. find
  10. Gheorghe Comanici, Eric Bieber, Mike Schaekermann, Ice Pasupat, Noveen Sachdeva, Inderjit Dhillon, Marcel Blistein, Ori Ram, Dan Zhang, Evan Rosen, and 1 others. 2025. Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities. arX…
  11. Lianghao Deng, Yuchong Sun, Shizhe Chen, Ning Yang, Yunfeng Wang, and Ruihua Song. 2025. Muka: Multimodal knowledge augmented visual information-seeking. In Proceedings of the Annual Meeting of the Association for Computational Linguistics. find
  12. Kuicai Dong, Yujing Chang, Shijie Huang, Yasheng Wang, Ruiming Tang, and Yong Liu. 2025. Benchmarking retrieval-augmented multimomal generation for document question answering. arXiv preprint arXiv:2505.16470. — [arXiv:2505.16470](https://arxiv.org/abs/2505.16470) find
  13. Lukas Haas, Gal Yona, Giovanni D’Antonio, Sasha Goldshtein, and Dipanjan Das. 2025. Simpleqa verified: A reliable factuality benchmark to measure parametric knowledge. arXiv preprint arXiv:2509.07968.
  14. Wenyi Hong, Wenmeng Yu, Xiaotao Gu, Guo Wang, Guobing Gan, Haomiao Tang, Jiale Cheng, Ji Qi, Junhui Ji, Lihang Pan, and 1 others. 2025. Glm-4.1 v-thinking: Towards versatile multimodal reasoning with scalable reinforcement learning. arXiv e-prints, pages arXiv–2507. find
  15. Wenbo Hu, Jia-Chen Gu, Zi-Yi Dou, Mohsen Fayyaz, Pan Lu, Kai-Wei Chang, and Nanyun Peng. 2025. Mrag-bench: Vision-centric evaluation for retrieval-augmented multimodal models. In Proceedings of the International Conference on Learning Representations. find
  16. Gautier Izacard, Mathilde Caron, Lucas Hosseini, Sebastian Riedel, Piotr Bojanowski, Armand Joulin, and Edouard Grave. 2021. Unsupervised dense information retrieval with contrastive learning. Transactions on Machine Learning Research.
  17. Paul Lerner, Olivier Ferret, Camille Guinaudeau, Hervé Le Borgne, Romaric Besançon, José G Moreno, and Jesús Lovón Melgarejo. 2022. Viquae, a dataset for knowledge-based visual question answering about named entities. In Proceedings of the ACM SIGIR Conference on Research and Development in Informat… find
  18. Bo Li, Yuanhan Zhang, Dong Guo, Renrui Zhang, Feng Li, Hao Zhang, Kaichen Zhang, Peiyuan Zhang, Yanwei Li, Ziwei Liu, and 1 others. 2024a. Llava-onevision: Easy visual task transfer. Transactions on Machine Learning Research.
  19. Dongxu Li, Yudong Liu, Haoning Wu, Yue Wang, Zhiqi Shen, Bowen Qu, Xinyao Niu, Fan Zhou, Chengen Huang, Yanpeng Li, and 1 others. 2024b. Aria: An open multimodal native mixture-of-experts model. arXiv preprint arXiv:2410.05993. — [arXiv:2410.05993](https://arxiv.org/abs/2410.05993) find
  20. Haotian Liu, Chunyuan Li, Yuheng Li, and Yong Jae Lee. 2024. Improved baselines with visual instruction tuning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.
  21. IEEE/CVF International Conference on Computer Vision. find
  22. Jielin Qiu, Andrea Madotto, Zhaojiang Lin, Paul A Crook, Yifan Xu, Babak Damavandi, Xin Luna Dong, Christos Faloutsos, Lei Li, and Seungwhan Moon. 2024. Snapntell: Enhancing entity-centric visual question answering with retrieval augmented multimodal llm. In Proceedings of the Conference on Empirica… find
  23. Dustin Schwenk, Apoorv Khandelwal, Christopher Clark, Kenneth Marino, and Roozbeh Mottaghi. 2022. A-okvqa: A benchmark for visual question answering using world knowledge. In Proceedings of the European Conference on Computer Vision.
  24. Sanket Shah, Anand Mishra, Naganand Yadati, and Partha Pratim Talukdar. 2019. Kvqa: Knowledge-aware visual question answering. In Proceedings of the AAAI Conference on Artificial Intelligence. find
  25. Ionut-Teodor Sorodoc, Leonardo FR Ribeiro, Rexhina Blloshmi, Christopher Davis, and Adrià de Gispert. 2025. Garage: A benchmark with grounding annotations for rag evaluation. In Proceedings of the Annual Meeting of the Association for Computational Linguistics. find
  26. Xin Su, Man Luo, Kris W Pan, Tien Pei Chou, Vasudev Lal, and Phillip Howard. 2025. Sk-vqa: Synthetic knowledge generation at scale for training context-augmented multimodal llms. In Proceedings of the International Conference on Machine Learning. find
  27. Quan Sun, Jinsheng Wang, Qiying Yu, Yufeng Cui, Fan Zhang, Xiaosong Zhang, and Xinlong Wang. 2024. Eva-clip-18b: Scaling clip to 18 billion parameters. arXiv preprint arXiv:2402.04252. — [arXiv:2402.04252](https://arxiv.org/abs/2402.04252) find
  28. Kelly Tang, Wei-Lin Chiang, and Anastasios N. Angelopoulos. 2025. Arena explorer: A topic modeling pipeline for llm evals & analytics. find
  29. Yixuan Tang and Yi Yang. 2024. Multihop-rag: Benchmarking retrieval-augmented generation for multi-hop queries. In First Conference on Language Modeling.
  30. Shiyin Lu, Yang Li, Qing-Guo Chen, Zhao Xu, Weihua Luo, Kaifu Zhang, and Han-Jia Ye. 2024. Ovis: Structural embedding alignment for multimodal large language model. arXiv preprint arXiv:2405.20797. — [arXiv:2405.20797](https://arxiv.org/abs/2405.20797) find
  31. Gemma Team, Aishwarya Kamath, Johan Ferret, Shreya Pathak, Nino Vieillard, Ramona Merhej, Sarah Perrin, Tatiana Matejovicova, Alexandre Ramé, Morgane Rivière, and 1 others. 2025. Gemma 3 technical report. arXiv preprint arXiv:2503.19786. find
  32. Kenneth Marino, Mohammad Rastegari, Ali Farhadi, and Roozbeh Mottaghi. 2019. Ok-vqa: A visual question answering benchmark requiring external knowledge. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. find
  33. Haozhe Wang, Chao Qu, Zuming Huang, Wei Chu, Fangzhen Lin, and Wenhu Chen. 2025. Vl-rethinker: Incentivizing self-reflection of vision-language models with reinforcement learning. arXiv preprint arXiv:2504.08837. find
  34. Thomas Mensink, Jasper Uijlings, Lluis Castrejon, Arushi Goel, Felipe Cadar, Howard Zhou, Fei Sha, André Araujo, and Vittorio Ferrari. 2023. Encyclopedic vqa: Visual questions about detailed properties of fine-grained categories. In Proceedings of the Junyang Wang, Yuhang Wang, Guohai Xu, Jing Zhang… find
  35. Zhecan Wang, Garrett Bingham, Adams Wei Yu, Quoc V Le, Thang Luong, and Golnaz Ghiasi. 2024. Haloquest: A visual hallucination dataset for advancing multimodal reasoning. In Proceedings of the European Conference on Computer Vision. find
  36. Jason Wei, Nguyen Karina, Hyung Won Chung, Yunxin Joy Jiao, Spencer Papay, Amelia Glaese, John Schulman, and William Fedus. 2024. Measuring short-form factuality in large language models. arXiv preprint arXiv:2411.04368.
  37. Yin Wu, Quanyu Long, Jing Li, Jianfei Yu, and Wenya Wang. 2025. Visual-rag: Benchmarking text-to-image retrieval augmented generation for visual knowledge intensive queries. arXiv preprint arXiv:2502.16636. — [arXiv:2502.16636](https://arxiv.org/abs/2502.16636) find
  38. Keyang Xuan, Li Yi, Fan Yang, Ruochen Wu, Yi R Fung, and Heng Ji. 2024. Lemma: towards lvlm-enhanced multimodal misinformation detection with external knowledge augmentation. arXiv preprint arXiv:2402.11943. — [arXiv:2402.11943](https://arxiv.org/abs/2402.11943) find
  39. Yibin Yan and Weidi Xie. 2024. Echosight: Advancing visual-language models with wiki knowledge. In Proceedings of the Conference on Empirical Methods in Natural Language Processing. find
  40. Biao Yang, Bin Wen, Boyang Ding, Changyi Liu, Chenglong Chu, Chengru Song, Chongling Rao, Chuan Yi, Da Li, Dunju Zang, and 1 others. 2025. Kwai keye-vl 1.5 technical report. arXiv preprint arXiv:2509.01563. — [arXiv:2509.01563](https://arxiv.org/abs/2509.01563) find
  41. Xiao Yang, Kai Sun, Hao Xin, Yushi Sun, Nikita Bhalla, Xiangsen Chen, Sajal Choudhary, Rongze Gui, Ziran Jiang, Ziyu Jiang, and 1 others. 2024. Crag-comprehensive rag benchmark. In Advances in Neural Information Processing Systems. find
  42. Yuan Yao, Tianyu Yu, Ao Zhang, Chongyi Wang, Junbo Cui, Hongji Zhu, Tianchi Cai, Haoyu Li, Weilin Zhao, Zhihui He, and 1 others. 2024. Minicpm-v: A gpt-4v level mllm on your phone. arXiv preprint arXiv:2408.01800.
  43. Xu Yuan, Liangbo Ning, Wenqi Fan, and Qing Li. 2025. mkg-rag: Multimodal knowledge graph-enhanced rag for visual question answering. arXiv preprint arXiv:2508.05318. — [arXiv:2508.05318](https://arxiv.org/abs/2508.05318) find
  44. Jinguo Zhu, Weiyun Wang, Zhe Chen, Zhaoyang Liu, Shenglong Ye, Lixin Gu, Hao Tian, Yuchen Duan, Weijie Su, Jie Shao, and 1 others. 2025. Internvl3: Exploring advanced training and test-time recipes for open-source multimodal models. arXiv preprint arXiv:2504.10479.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/moratelli-2026-benchmarking-deflection-hallucination-large-vision/