Elamur Research⌕
☰
home / papers / 2026 / Know the Known and the Unknown: Reasonable Answer Generation with Knowledge-Informed Citations

Know the Known and the Unknown: Reasonable Answer Generation with Knowledge-Informed Citations

Paper · 2026 · Yichi Zhang, Zhuo Chen, Lingbing Guo et al.

Summary

Citation-augmented question answering helps users verify LLM outputs. Existing benchmarks (ALCE, LongCite, SynSciQA, WebCiteS) have three weaknesses: they cite at the coarse chunk or sentence level, they assume the references always contain the answer, and they rely on plain SFT. The authors present KFC (Knowledge-inFormed Citation), a benchmark of about 42K automatically built QA samples drawn from the UltraDomain corpus.

An LLM generator extracts key entities from each reference chunk and then writes two kinds of questions: ones the references can answer and ones they cannot. Chain-of-thought prompts list the known and the unknown entities. Answers mark sourced entities with indices such as [1] and [2], and flag missing information with a special [?] marker as a warning.

The authors also propose SELF-KFC, a training scheme that blends SFT with preference alignment. The model learns to revise badly cited answers into better ones. Across several backbones, SELF-KFC beats SFT, DPO, ORPO and SimPO, and it holds up when data is scarce and models are small. This matters for trustworthy RAG systems, which must signal what their sources do not cover.

Key points

  • KFC contains ~42K samples from 19 domains, using UltraDomain textbook texts split into chunks of at most 256 words. It is the only benchmark in the authors' comparison that combines entity-level citations with awareness of unknown information.
  • The construction pipeline has six steps: corpus preparation, key entity extraction, question generation (separate templates for answerable and unanswerable questions), answer generation, rule-based citation labeling, and data instance organization.
  • New evaluation metrics score citation quality separately for known and unknown knowledge.
  • On the Qwen2.5 series (0.5B to 14B and a larger model), SELF-KFC outperforms the baselines at every size. The gains are largest for 1.5B–7B models, while 0.5B is capacity-limited.
  • A LLaMA3-8B case study shows SELF-KFC avoiding citation omissions that SFT and DPO make. The stated limitations are that the work is restricted to QA and the dataset size is moderate.

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

🎓Zhejiang UniversityHangzhou · China224 works · 13,587 citations🏢Ant GroupHangzhou · China51 works · 1,211 citations

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 worksComputer Vision and Pattern Recognition · 3,486 works

Cite

Cite as Yichi Zhang, Zhuo Chen, Lingbing Guo, Jun Xu, Mengshu Sun, Zhizhen Liu, Lei Liang, Wen Zhang and Huajun Chen. 2026. Know the Known and the Unknown: Reasonable Answer Generation with Knowledge-Informed Citations. 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.867

@inproceedings{zhang2026know,
  title = {{Know the Known and the Unknown: Reasonable Answer Generation with Knowledge-Informed Citations}},
  author = {Yichi Zhang, Zhuo Chen, Lingbing Guo, Jun Xu, Mengshu Sun, Zhizhen Liu, Lei Liang, Wen Zhang and Huajun Chen},
  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.867},
  url = {https://doi.org/10.18653/v1/2026.acl-long.867}
}

References

25 references, 11 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Yung-Sung Chuang, Benjamin Cohen-Wang, Zejiang Shen, Zhaofeng Wu, Hu Xu, Xi Victoria Lin, James R. Glass, Shang-Wen Li, and Wen tau Yih. 2025. Selfcite: Self-supervised alignment for context attribution in large language models. In Forty-second International Conference on Machine Learning. find
  2. Haolin Deng, Chang Wang, Xin Li, Dezhang Yuan, Junlang Zhan, Tianhua Zhou, Jin Ma, Jun Gao, and Ruifeng Xu. 2024. Webcites: Attributed query-focused summarization on chinese web search results with citations. In ACL (1), pages 15095–15114. Association for Computational Linguistics. find
  3. Tianyu Gao, Howard Yen, Jiatong Yu, and Danqi Chen. 2023. Enabling large language models to generate text with citations. In EMNLP, pages 6465–6488. Association for Computational Linguistics.
  4. Yunfan Gao, Yun Xiong, Xinyu Gao, Kangxiang Jia, Jinliu Pan, Yuxi Bi, Yi Dai, Jiawei Sun, Meng Wang, and Haofen Wang. 2024. Retrieval-augmented generation for large language models: A survey. Preprint, arXiv:2312.10997.
  5. Jiawei Gu, Xuhui Jiang, Zhichao Shi, Hexiang Tan, Xuehao Zhai, Chengjin Xu, Wei Li, Yinghan Shen, Shengjie Ma, Honghao Liu, Saizhuo Wang, Kun Zhang, Yuanzhuo Wang, Wen Gao, Lionel Ni, and Jian Guo. 2025. A survey on llm-as-a-judge. Preprint, arXiv:2411.15594. find
  6. Jiwoo Hong, Noah Lee, and James Thorne. 2024. ORPO: monolithic preference optimization without reference model. In EMNLP, pages 11170–11189. Association for Computational Linguistics. find
  7. Edward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. 2022. Lora: Low-rank adaptation of large language models. In ICLR. OpenReview.net.
  8. Chengyu Huang, Zeqiu Wu, Yushi Hu, and Wenya Wang. 2024. Training language models to generate text with citations via fine-grained rewards. In ACL (1), pages 2926–2949. Association for Computational Linguistics. find
  9. Jiaming Ji, Boyuan Chen, Hantao Lou, Donghai Hong, Borong Zhang, Xuehai Pan, Tianyi Qiu, Juntao Dai, and Yaodong Yang. 2024. Aligner: Efficient alignment by learning to correct. In NeurIPS.
  10. Woosuk Kwon, Zhuohan Li, Siyuan Zhuang, Ying Sheng, Lianmin Zheng, Cody Hao Yu, Joseph E. Gonzalez, Hao Zhang, and Ion Stoica. 2023. Efficient memory management for large language model serving with pagedattention. In Proceedings of the ACM SIGOPS 29th Symposium on Operating Systems Principles.
  11. Xiao-Yang Liu, Guoxuan Wang, and Daochen Zha. 2023. Fingpt: Democratizing internet-scale data for financial large language models. CoRR, abs/2307.10485. — [arXiv:2307.10485](https://arxiv.org/abs/2307.10485) find
  12. AI @ Meta Llama Team. 2024. The llama 3 herd of models. Preprint, arXiv:2407.21783.
  13. Yu Meng, Mengzhou Xia, and Danqi Chen. 2024. Simpo: Simple preference optimization with a reference-free reward. In NeurIPS. find
  14. Harsha Nori, Nicholas King, Scott Mayer McKinney, Dean Carignan, and Eric Horvitz. 2023. Capabilities of GPT-4 on medical challenge problems. CoRR, abs/2303.13375.
  15. Hongjin Qian, Zheng Liu, Peitian Zhang, Kelong Mao, Defu Lian, Zhicheng Dou, and Tiejun Huang. 2025. Memorag: Boosting long context processing with global memory-enhanced retrieval augmentation. In Proceedings of the ACM Web Conference 2025 (TheWebConf 2025), Sydney, Australia. ACM. ArXiv:2409.05591… — [arXiv:2409.05591](https://arxiv.org/abs/2409.05591) find
  16. QwenTeam. 2024. Qwen2.5: A party of foundation models. find
  17. QwenTeam. 2025. Qwen3 technical report. Preprint, arXiv:2505.09388.
  18. 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. In NeurIPS.
  19. Tobias Schimanski, Jingwei Ni, Mathias Kraus, Elliott Ash, and Markus Leippold. 2024. Towards faithful and robust LLM specialists for evidence-based question-answering. In ACL (1), pages 1913–1931. Association for Computational Linguistics. find
  20. Ling Team and Inclusion AI. 2025. Every activation boosted: Scaling general reasoner to 1 trillion open language foundation. CoRR, abs/2510.22115. — [arXiv:2510.22115](https://arxiv.org/abs/2510.22115) find
  21. Yidong Wang, Qi Guo, Wenjin Yao, Hongbo Zhang, Xin Zhang, Zhen Wu, Meishan Zhang, Xinyu Dai, Min Zhang, Qingsong Wen, Wei Ye, Shikun Zhang, and Yue Zhang. 2024. Autosurvey: Large language models can automatically write surveys. In NeurIPS.
  22. Xi Ye, Ruoxi Sun, Sercan Ö. Arik, and Tomas Pfister. 2024. Effective large language model adaptation for improved grounding and citation generation. In NAACL-HLT, pages 6237–6251. Association for Computational Linguistics. find
  23. Jiajie Zhang, Yushi Bai, Xin Lv, Wanjun Gu, Danqing Liu, Minhao Zou, Shulin Cao, Lei Hou, Yuxiao Dong, Ling Feng, and Juanzi Li. 2024a. Longcite: Enabling llms to generate fine-grained citations in long-context QA. CoRR, abs/2409.02897. find
  24. Yichi Zhang, Zhuo Chen, Yin Fang, Yanxi Lu, Fangming Li, Wen Zhang, and Huajun Chen. 2024b. Knowledgeable preference alignment for llms in domain-specific question answering. In ACL (Findings), pages 891–904. Association for Computational Linguistics. find
  25. Yaowei Zheng, Richong Zhang, Junhao Zhang, Yanhan Ye, Zheyan Luo, Zhangchi Feng, and Yongqiang Ma. 2024. Llamafactory: Unified efficient fine-tuning of 100+ language models. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 3: System Demonstrations),…

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/zhang-2026-know-known-unknown-reasonable-answer/