Elamur Research⌕
☰
home / papers / 2026 / Beyond Overlap Metrics: Rewarding Reasoning and Preferences for Faithful Multi-Role Dialogue Summarization

Beyond Overlap Metrics: Rewarding Reasoning and Preferences for Faithful Multi-Role Dialogue Summarization

Paper · 2026 · Xiaoyong Mei, Tingting Zuo, Da Chen et al.

Summary

The paper addresses the problem that summarizers of multi-speaker conversations are usually tuned toward overlap metrics such as ROUGE and BERTScore. Those metrics reward matching the reference wording but do not reliably prevent hallucinations, such as assigning an action to the wrong speaker. The authors propose a two-stage training pipeline. First, reasoning traces are distilled from a large teacher model (DeepSeek-V3). These traces cover dialogue examination, motive inference, content abstraction and a fidelity check, and they are used for staged supervised fine-tuning of a smaller model. Second, GRPO refines the model with a dual-principle reward. One component judges the reasoning on coverage, implicit inference, faithfulness and conciseness. The other scores the final summary with lexical and semantic metrics plus a length term. Experiments on the Chinese CSDS and English SAMSum datasets show overlap scores comparable to strong baselines. On SAMSum, the added principle reward clearly improves faithfulness and model-judged preference. This suggests that rewarding reasoning quality, not just overlap, yields more trustworthy dialogue summaries.

Key points

  • Reward weights. The reasoning ("thinking") reward weights key information coverage at 40%, inference and implicit understanding at 30%, faithfulness and precision at 20%, and conciseness at 10%. The summary-side reward combines BERTScore, ROUGE and length signals.
  • Models and resources. The student models are Qwen3-base trained with LoRA, with Qwen2.5 variants also evaluated. Checkpoints and data are published as the SummOrchestra collection on Hugging Face (NebulaPixel).
  • Preference result. On SAMSum, the model's preference score (0.6683) exceeds that of the human reference summaries (0.4973). The authors attribute this to noise in the gold summaries.
  • CSDS limitation. The principle reward was not used for CSDS because the faithfulness evaluator HHEM-2.1-Open does not support Chinese. Faithfulness gains are therefore shown only for English.
  • Ablation findings. Adding the principle reward slightly lowers conventional metrics while substantially raising faithfulness and preference scores. Optimizing for BERTScore alone reduces surface overlap, whereas combining it with ROUGE and length rewards gives more balanced results.

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

🎓Hong Kong University of Science and TechnologyHong Kong · China142 works · 25,136 citations🎓Chinese University of Hong KongHong Kong · China126 works · 14,073 citations🏢Huawei TechnologiesShenzhen · China75 works · 12,975 citations🎓Sun Yat-sen UniversityGuangzhou · China53 works · 9,185 citations🎓Zhejiang Normal UniversityJinhua · China4 works · 16 citations

 Fields, topics and concepts

📚 Shelves 4The 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 works

🧭 Topics 4Research subjects and languages the work is about.Reasoning · 1,426 worksEvaluation & benchmarks · 1,220 worksDialogue systems · 624 worksSummarization · 232 works

Cite

Cite as Xiaoyong Mei, Tingting Zuo, Da Chen, Guangyu Hu, Xiangyu Wen, Chao Duan, Mingyan Zhang and Fudan Zheng. 2026. Beyond Overlap Metrics: Rewarding Reasoning and Preferences for Faithful Multi-Role Dialogue Summarization. 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.1161

@inproceedings{mei2026beyond,
  title = {{Beyond Overlap Metrics: Rewarding Reasoning and Preferences for Faithful Multi-Role Dialogue Summarization}},
  author = {Xiaoyong Mei, Tingting Zuo, Da Chen, Guangyu Hu, Xiangyu Wen, Chao Duan, Mingyan Zhang and Fudan Zheng},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.1161},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.1161}
}

References

54 references, 13 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Eunice Akani, Benoît Favre, Frédéric Béchet, and Romain Gemignani. 2024. Increasing faithfulness in human-human dialog summarization with spoken language understanding tasks. CoRR, abs/2409.10070. — [arXiv:2409.10070](https://arxiv.org/abs/2409.10070) find
  2. Afra Amini, Tim Vieira, and Ryan Cotterell. 2024. Direct preference optimization with an offset. In Findings of the Association for Computational Linguistics: ACL 2024, pages 9954–9972, Bangkok, Thailand. Association for Computational Linguistics. find
  3. Hritik Bansal, Ashima Suvarna, Gantavya Bhatt, Nanyun Peng, Kai-Wei Chang, and Aditya Grover. 2025. Comparing bad apples to good oranges aligning large language models via joint preference optimization. In Findings of the Association for Computational Linguistics: ACL 2025, pages 701–723, Vienna, Au… find
  4. Jiaao Chen and Diyi Yang. 2020. Multi-view sequence-to-sequence models with conversational structure for abstractive dialogue summarization. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 4106– 4118, Online. Association for Computational Ling…
  5. Paul F Christiano, Jan Leike, Tom Brown, Miljan Martic, Shane Legg, and Dario Amodei. 2017. Deep reinforcement learning from human preferences. Advances in neural information processing systems, 30. find
  6. Luyang Fang, Xiao-Xing Yu, Jiazhang Cai, Yongkai Chen, Shushan Wu, Zheng Liu, Zhenyuan Yang, Haoran Lu, Xilin Gong, Yufang Liu, Terry Ma, Wei Ruan, Ali Abbasi, Jing Zhang, Tao Wang, Ehsan Latif, Wei Liu, Wei Zhang, Soheil Kolouri, and 5 others. 2025. Knowledge distillation and dataset distillation o… find
  7. Guy Feigenblat, Chulaka Gunasekara, Benjamin Sznajder, Sachindra Joshi, David Konopnicki, and Ranit Aharonov. 2021. TWEETSUMM - a dialog summarization dataset for customer service. In Findings of the Association for Computational Linguistics: EMNLP 2021, pages 245–260, Punta Cana, Dominican Republic… find
  8. Huawen Feng, Yan Fan, Xiong Liu, Ting-En Lin, Zekun Yao, Yuchuan Wu, Fei Huang, Yongbin Li, and Qianli Ma. 2024. Improving factual consistency of news summarization by contrastive preference optimization. In Findings of the Association for Computational Linguistics: EMNLP 2024, pages 11084– 11100, M… find
  9. Xiachong Feng, Xiaocheng Feng, and Bing Qin. 2022. A survey on dialogue summarization: Recent advances and new frontiers. In Proceedings of the Thirty-First International Joint Conference on Artificial Intelligence, IJCAI-22, pages 5453–5460. International Joint Conferences on Artificial Intelligenc… find
  10. Xiachong Feng, Xiaocheng Feng, Bing Qin, and Xinwei Geng. 2021. Dialogue discourse-aware graph model and data augmentation for meeting summarization. In Proceedings of the Thirtieth International Joint Conference on Artificial Intelligence, IJCAI-21, pages 3808–3814. International Joint Conferences… find
  11. Bogdan Gliwa, Iwona Mochol, Maciej Biesek, and Aleksander Wawer. 2019. SAMSum corpus: A human-annotated dialogue dataset for abstractive summarization. In Proceedings of the 2nd Workshop on New Frontiers in Summarization, pages 70–79, Hong Kong, China. Association for Computational Linguistics.
  12. Geoffrey E. Hinton, Oriol Vinyals, and Jeffrey Dean. 2015. Distilling the knowledge in a neural network. ArXiv, abs/1503.02531.
  13. Ziwei Ji, Zihan Liu, Nayeon Lee, Tiezheng Yu, Bryan Wilie, Min Zeng, and Pascale Fung. 2023. Rho: Reducing hallucination in open-domain dialogues with knowledge grounding. In Findings of the association for computational linguistics: ACL 2023, pages 4504–4522. find
  14. Keyan Jin, Yapeng Wang, Leonel Santos, Tao Fang, Xu Yang, Sio Kei Im, and Hugo Gonçalo Oliveira. 2026. Reasoning or not? A comprehensive evaluation of reasoning llms for dialogue summarization. Expert Syst. Appl., 299:129831. find
  15. Wojciech Kryscinski, Bryan McCann, Caiming Xiong, and Richard Socher. 2019. Evaluating the factual consistency of abstractive text summarization. In Conference on Empirical Methods in Natural Language Processing.
  16. Philippe Laban, Tobias Schnabel, Paul N. Bennett, and Marti A. Hearst. 2021. Summac: Re-visiting nli-based models for inconsistency detection in summarization. Transactions of the Association for Computational Linguistics, 10:163–177. find
  17. Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Veselin Stoyanov, and Luke Zettlemoyer. 2020. BART: Denoising sequence-to-sequence pre-training for natural language generation, translation, and comprehension. In Proceedings of the 58th Annual Meeting of the…
  18. Dawei Li, Bohan Jiang, Liangjie Huang, Alimohammad Beigi, Chengshuai Zhao, Zhen Tan, Amrita Bhattacharjee, Yuxuan Jiang, Canyu Chen, Tianhao Wu, Kai Shu, Lu Cheng, and Huan Liu. 2025a. From generation to judgment: Opportunities and challenges of LLM-as-a-judge. In Proceedings of the 2025 Conference… find
  19. Manling Li, Lingyu Zhang, Heng Ji, and Richard J. Radke. 2019. Keep meeting summaries on topic: Abstractive multi-modal meeting summarization. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 2190– 2196, Florence, Italy. Association for Computational…
  20. Miaoran Li, Rogger Luo, and Ofer Mendelevitch. 2024. HHEM-2.1-Open. find
  21. Raymond Li, Chuyuan Li, Gabriel Murray, and Giuseppe Carenini. 2025b. Hierarchical attention adapter for abstractive dialogue summarization. In Proceedings of The 5th New Frontiers in Summarization Workshop, pages 17–30, Hybrid. Association for Computational Linguistics. find
  22. Zhongguo Li and Maosong Sun. 2009. Punctuation as implicit annotations for chinese word segmentation. Computational Linguistics, 35(4):505–512. find
  23. Chin-Yew Lin. 2004. ROUGE: A package for automatic evaluation of summaries. In Text Summarization Branches Out, pages 74–81, Barcelona, Spain. Association for Computational Linguistics.
  24. Haitao Lin, Liqun Ma, Junnan Zhu, Lu Xiang, Yu Zhou, Jiajun Zhang, and Chengqing Zong. 2021. CSDS: A fine-grained Chinese dataset for customer service dialogue summarization. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 4436–4451, Online and Punta… find
  25. Aixin Liu, Bei Feng, Bing Xue, Bingxuan Wang, Bochao Wu, Chengda Lu, Chenggang Zhao, Chengqi Deng, Chenyu Zhang, Chong Ruan, and 1 others. 2024. Deepseek-v3 technical report. arXiv preprint arXiv:2412.19437. find
  26. Yen-Ju Lu, Ting-Yao Hu, Hema Swetha Koppula, Hadi Pouransari, Jen-Hao Rick Chang, Yin Xia, Xiang Kong, Qi Zhu, Xiaoming Simon Wang, Oncel Tuzel, and Raviteja Vemulapalli. 2025. Mutual reinforcement of LLM dialogue synthesis and summarization capabilities for few-shot dialogue summarization. In Findi… find
  27. Potsawee Manakul, Adian Liusie, and Mark Gales. 2023. Selfcheckgpt: Zero-resource black-box hallucination detection for generative large language models. In Proceedings of the 2023 conference on empirical methods in natural language processing, pages 9004– 9017. find
  28. Gabriel Murray, Steve Renals, and Jean Carletta. 2005. Extractive summarization of meeting recordings. In 9th European Conference on Speech Communication and Technology, INTERSPEECH-Eurospeech 2005, Lisbon, Portugal, September 4-8, 2005, pages 593– 596. ISCA. find
  29. Ryan Park, Rafael Rafailov, Stefano Ermon, and Chelsea Finn. 2024. Disentangling length from quality in direct preference optimization. In Findings of the Association for Computational Linguistics: ACL 2024, pages 4998–5017, Bangkok, Thailand. Association for Computational Linguistics.
  30. Rafael Rafailov, Archit Sharma, Eric Mitchell, Christopher D Manning, Stefano Ermon, and Chelsea Finn. 2023. Direct preference optimization: Your language model is secretly a reward model. Advances in neural information processing systems, 36:53728–53741.
  31. Sanjana Ramprasad, Elisa Ferracane, and Zachary Lipton. 2024. Analyzing LLM behavior in dialogue summarization: Unveiling circumstantial hallucination trends. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 12549–12561, Bangko… find
  32. Virgile Rennard, Guokan Shang, Julie Hunter, and Michalis Vazirgiannis. 2023. Abstractive meeting summarization: A survey. Transactions of the Association for Computational Linguistics, 11:861–884.
  33. Korbinian Riedhammer, Benoît Favre, and Dilek Hakkani-Tür. 2008. A keyphrase based approach to interactive meeting summarization. In 2008 IEEE Spoken Language Technology Workshop, SLT 2008, Goa, India, December 15-19, 2008, pages 153–156. IEEE. find
  34. Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Mingchuan Zhang, Y. K. Li, Y. Wu, and Daya Guo. 2024. Deepseekmath: Pushing the limits of mathematical reasoning in open language models. CoRR, abs/2402.03300.
  35. Hwanjun Song, Taewon Yun, Yuho Lee, Jihwan Oh, Gihun Lee, Jason Cai, and Hang Su. 2025. Learning to summarize from LLM-generated feedback. In Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volum… find
  36. Manato Tajiri and Michimasa Inaba. 2025. Refining text generation for realistic conversational recommendation via direct preference optimization. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 28628–28649, Suzhou, China. Association for Computational… find
  37. Jing Ye, Rui Wang, Yuchuan Wu, Victor Ma, Feiteng Fang, Fei Huang, and Yongbin Li. 2025a. CPO: Addressing reward ambiguity in role-playing dialogue via comparative policy optimization. In Findings of the Association for Computational Linguistics: EMNLP 2025, pages 297–323, Suzhou, China. Association… find
  38. Yuanhe Tian, Fei Xia, and Yan Song. 2024. Dialogue summarization with mixture of experts based on large language models. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 7143–7155, Bangkok, Thailand. Association for Computation… find
  39. Yuxuan Ye, Raul Santos-Rodriguez, and Edwin Simpson. 2025b. Optimising factual consistency in summarisation via preference learning from multiple imperfect metrics. In Findings of the Association for Computational Linguistics: EMNLP 2025, pages 17342–17355, Suzhou, China. Association for Computation… find
  40. Antoine Tixier, Polykarpos Meladianos, and Michalis Vazirgiannis. 2017. Combining graph degeneracy and submodularity for unsupervised extractive summarization. In Proceedings of the Workshop on New Frontiers in Summarization, pages 48–58, Copenhagen, Denmark. Association for Computational Linguistic… find
  41. Tian Yu, Shaolei Zhang, and Yang Feng. 2024. Truth-aware context selection: Mitigating hallucinations of large language models being misled by untruthful contexts. In Findings of the Association for Computational Linguistics: ACL 2024, pages 10862–10884, Bangkok, Thailand. Association for Computatio… find
  42. Binghai Wang, Runji Lin, Keming Lu, Le Yu, Zhenru Zhang, Fei Huang, Chujie Zheng, Kai Dang, Yang Fan, Xingzhang Ren, and 1 others. 2025. Worldpm: Scaling human preference modeling. arXiv preprint arXiv:2505.10527. find
  43. Zhuohao Yu, Jiali Zeng, Weizheng Gu, Yidong Wang, Jindong Wang, Fandong Meng, Jie Zhou, Yue Zhang, Shikun Zhang, and Wei Ye. 2025. Rewardanything: Generalizable principle-following reward models. CoRR, abs/2506.03637. — [arXiv:2506.03637](https://arxiv.org/abs/2506.03637) find
  44. Yiming Wang, Zhuosheng Zhang, and Rui Wang. 2023. Element-aware summarization with large language models: Expert-aligned evaluation and chain-of-thought method. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 8640–8665, Toront… find
  45. Jingqing Zhang, Yao Zhao, Mohammad Saleh, and Peter J. Liu. 2020. Pegasus: pre-training with extracted gap-sentences for abstractive summarization. In Proceedings of the 37th International Conference on Machine Learning, ICML’20. JMLR.org.
  46. Xiangyu Wen, Junhua Huang, Zeju Li, Min Li, Jianyuan Zhong, Zhijian Xu, Mingxuan Yuan, Yongxiang Huang, and Qiang Xu. 2025a. Reasoning scaffolding: Distilling the flow of thought from llms. ArXiv, abs/2509.23619. — [arXiv:2509.23619](https://arxiv.org/abs/2509.23619) find
  47. Xiangyu Wen, Jianyuan Zhong, Zhijian Xu, and Qiang Xu. 2025b. Guideline compliance in task-oriented dialogue: The chained prior approach. In North American Chapter of the Association for Computational Linguistics, pages 6750–6776. find
  48. Jie Wu, Yu Gao, Zi-Nuo Ye, Ming Li, Liang Li, Hanzhong Guo, Jie Liu, Zeyue Xue, Xiaoxia Hou, Wei Liu, Yangyang Zeng, and Weilin Huang. 2025. Rewarddance: Reward scaling in visual generation. ArXiv, abs/2509.08826. — [arXiv:2509.08826](https://arxiv.org/abs/2509.08826) find
  49. Joonho Yang, Seunghyun Yoon, Byeongjeong Kim, and Hwanhee Lee. 2024. FIZZ: factual inconsistency detection by zoom-in summary and zoom-out document. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, EMNLP 2024, Miami, FL, USA, November 12-16, 2024, pages 30–4… find
  50. Shengyu Zhang, Linfeng Dong, Xiaoya Li, Sen Zhang, Xiaofei Sun, Shuhe Wang, Jiwei Li, Runyi Hu, Tianwei Zhang, Guoyin Wang, and Fei Wu. 2025. Instruction tuning for large language models: A survey. ACM Comput. Surv. Just Accepted.
  51. Tianyi Zhang, Varsha Kishore, Felix Wu, Kilian Q Weinberger, and Yoav Artzi. 2019. Bertscore: Evaluating text generation with bert. arXiv preprint arXiv:1904.09675. find
  52. Yusen Zhang, Ansong Ni, Ziming Mao, Chen Henry Wu, Chenguang Zhu, Budhaditya Deb, Ahmed Awadallah, Dragomir Radev, and Rui Zhang. 2022. Summ n : A multi-stage summarization framework for long input dialogues and documents. In Proceedings of the 60th Annual Meeting of the Association for Computationa… find
  53. Lulu Zhao, Fujia Zheng, Keqing He, Weihao Zeng, Yuejie Lei, Huixing Jiang, Wei Wu, Weiran Xu, Jun Guo, and Fanyu Meng. 2021. Todsum: Task-oriented dialogue summarization with state tracking. CoRR, abs/2110.12680. — [arXiv:2110.12680](https://arxiv.org/abs/2110.12680) find
  54. Ming Zhong, Yang Liu, Yichong Xu, Chenguang Zhu, and Michael Zeng. 2022. Dialoglm: Pre-trained find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/mei-2026-beyond-overlap-metrics-rewarding-reasoning/