---
title: "Rethinking Retrieval-Augmented Generation as a Cooperative"
description: "Most RAG pipelines treat reranking and generation as a one-way chain. 27 references. Among them: 0."
authors: ["Lichang Song", "Ting Long", "Yi Chang"]
year: 2026
venue: "Findings of the Association for Computational Linguistics: ACL 2026"
source: "https://aclanthology.org/2026.findings-acl.2157/"
doi: "10.18653/v1/2026.findings-acl.2157"
licence: "CC BY 4.0 (ACL Anthology)"
licence_url: "https://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/song-2026-rethinking-retrieval-augmented-generation-as/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/song-2026-rethinking-retrieval-augmented-generation-as/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2026](https://md.research.elamur.ai/free/en/papers/2026/page/41/) / Rethinking Retrieval-Augmented Generation as a Cooperative Decision-Making Problem

# Rethinking Retrieval-Augmented Generation as a Cooperative Decision-Making Problem

**Authors:** Lichang Song, Ting Long and [Yi Chang](https://md.research.elamur.ai/free/en/authors/yi-chang/) · **Year:** 2026 · **Venue:** [Findings of the Association for Computational Linguistics: ACL 2026](https://md.research.elamur.ai/free/en/venues/findings/) ([Findings of the Association for Computational Linguistics 2026](https://md.research.elamur.ai/free/en/venues/findings/2026/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2026.findings-acl.2157/) · [PDF](https://aclanthology.org/2026.findings-acl.2157.pdf) · [DOI 10.18653/v1/2026.findings-acl.2157](https://doi.org/10.18653/v1/2026.findings-acl.2157) · [OpenAlex](https://openalex.org/W7166839394)

**Citations:** 0 citations (OpenAlex, 2026-10-14)

## Summary
Most RAG pipelines treat reranking and generation as a one-way chain. The reranker fixes a document order, and the generator depends heavily on that order. A single misranked passage can therefore derail the answer even when the right evidence sits among the top candidates. The authors recast RAG as a cooperative multi-agent problem. The reranker and the generator act as peer agents trained toward one shared reward: whether the produced answer contains the gold answer. Their framework, CoRAG, uses BGE-Reranker as the reranker and an LLM generator tuned with GRPO. Standard MARL optimizers fit ranking poorly because of an MDP-ranking mismatch. For example, separate agents may pick the same document for several positions. CoRAG avoids this with a GRPO-style reranker objective. The task-level reward is converted into smoothed, sampled preference labels for each document, and the reranker learns from them through a margin-based pairwise loss. Training uses only about 10K PopQA examples. The model still beats baselines and transfers to other datasets, including code and table QA. This suggests that joint, outcome-driven training makes RAG output more stable.

## Key points
- **Credit assignment:** each document gets a binary success signal, averaged over past iterations. This average is smoothed into a Bernoulli parameter (α ∈ (0, 0.5)) and sampled to produce preference labels. The GRPO objective is then reduced to a deterministic pairwise learning-to-rank loss.
- **Training data:** about 10K PopQA samples, with a reward of 1 if the response contains the ground-truth answer and 0 otherwise.
- **LLM-as-a-judge on PopQA:** CoRAG scores 71.35 on average, against 64.60 for InstructRAG and 41.38 for RetRobust (recomputed; the paper's table prints 45.69). Judges were Llama-3-Instruct 8B, gpt-4o, deepseek-v3.2 and qwen3-vl-235b-a22b. The pattern-based score is 71.26 versus 66.19 for InstructRAG.
- **Code and table QA:** the generator gets the best pass@10 on HumanEval/HumanEval+ and the best pass@1 on HumanEval+. On WikiTable Questions it nearly matches the top baseline.
- **Limitation:** making the generator less sensitive to ranking may also reduce the payoff from future reranker improvements. Code is available at github.com/CoderrrSong/CoRAG.

*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

- [Jilin University](https://md.research.elamur.ai/free/en/universities/jilin-university/) — University, Changchun, China · 25 works · 5,252 citations

📚 Shelves (6): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 16,194 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 16,039 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 15,891 works · [👁️ Computer Vision and Pattern Recognition](https://md.research.elamur.ai/free/en/shelves/computer-science/computer-vision-and-pattern-recognition/) · 2,869 works · [🗄️ Information Systems](https://md.research.elamur.ai/free/en/shelves/computer-science/information-systems/) · 1,437 works

🧭 Topics (1): [🔍 Information retrieval & search](https://md.research.elamur.ai/free/en/topics/information-retrieval/) · 473 works

## Cite

Cite as: Lichang Song, Ting Long and Yi Chang. 2026. Rethinking Retrieval-Augmented Generation as a Cooperative Decision-Making Problem. Findings of the Association for Computational Linguistics: ACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@inproceedings{song2026rethinking,
  title = {{Rethinking Retrieval-Augmented Generation as a Cooperative Decision-Making Problem}},
  author = {Lichang Song, Ting Long and Yi Chang},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.2157},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.2157}
}
```

- [← previous Rethinking Research on Stereotypes: An Analysis through Social Psychological and Computational Perspectives](https://md.research.elamur.ai/free/en/papers/shejole-2026-rethinking-research-stereotypes-analysis-through/)
- [next → Rethinking Role-Playing Evaluation: Anonymous Benchmarking and A Systematic Study of Personality Effects](https://md.research.elamur.ai/free/en/papers/peng-2026-rethinking-role-playing-evaluation-anonymous/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/41/)

## References

27 references, 17 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Akari Asai, Zeqiu Wu, Yizhong Wang, Avirup Sil, and Hannaneh Hajishirzi. 2024. [Self-rag: Learning to retrieve, generate, and critique through self-reflection](https://md.research.elamur.ai/free/en/papers/asai-2023-self-rag-learning-retrieve-generate/).
2. Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, and 1 others. 2020. [Language models are few-shot learners](https://md.research.elamur.ai/free/en/papers/brown-2020-language-models-few-shot-learners/). Advances in neural information processing systems, 33:1877–1901.
3. Francesco Casalegno. 2022. Learning to rank: A complete guide to ranking using machine learning. Towards Data Science. [find](https://scholar.google.com/scholar?q=Francesco+Casalegno%2E+2022%2E+Learning+to+rank%3A+A+complete+guide+to+ranking+using+machine+learning%2E+Towards+Data+Science%2E)
4. Mark Chen. 2021. [Evaluating large language models trained on code](https://md.research.elamur.ai/free/en/papers/chen-2021-evaluating-large-language-models-trained/). arXiv preprint arXiv:2107.03374.
5. Yiqun Chen, Lingyong Yan, Weiwei Sun, Xinyu Ma, Yi Zhang, Shuaiqiang Wang, Dawei Yin, Yiming Yang, and Jiaxin Mao. 2025. Improving retrieval-augmented generation through multi-agent reinforcement learning. arXiv preprint arXiv:2501.15228. — [arXiv:2501.15228](https://arxiv.org/abs/2501.15228) [find](https://scholar.google.com/scholar?q=Yiqun+Chen%2C+Lingyong+Yan%2C+Weiwei+Sun%2C+Xinyu+Ma%2C+Yi+Zhang%2C+Shuaiqiang+Wang%2C+Dawei+Yin%2C+Yiming+Yang%2C+and+Jiaxin+Mao%2E+2025%2E+Improving+retrieval%2Daugmented+generation+through+multi%2Dagent+reinforcement+lear)
6. Yuqin Dai, Guoqing Wang, Yuan Wang, Kairan Dou, Kaichen Zhou, Zhanwei Zhang, Shuo Yang, Fei Tang, Jun Yin, Pengyu Zeng, and 1 others. 2025. Evinoterag: Enhancing rag models via answer-supportive evidence notes. arXiv preprint arXiv:2509.00877. — [arXiv:2509.00877](https://arxiv.org/abs/2509.00877) [find](https://scholar.google.com/scholar?q=Yuqin+Dai%2C+Guoqing+Wang%2C+Yuan+Wang%2C+Kairan+Dou%2C+Kaichen+Zhou%2C+Zhanwei+Zhang%2C+Shuo+Yang%2C+Fei+Tang%2C+Jun+Yin%2C+Pengyu+Zeng%2C+and+1+others%2E+2025%2E+Evinoterag%3A+Enhancing+rag+models+via+answer%2Dsupportive+evide)
7. Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Amy Yang, Angela Fan, and 1 others. 2024. [The llama 3 herd of models](https://md.research.elamur.ai/free/en/papers/grattafiori-2024-llama-3-herd-models/). arXiv preprint arXiv:2407.21783.
8. Mandar Joshi, Eunsol Choi, Daniel S Weld, and Luke Zettlemoyer. 2017. [Triviaqa: A large scale distantly supervised challenge dataset for reading comprehension](https://md.research.elamur.ai/free/en/papers/joshi-2017-triviaqa-large-scale-distantly-supervised/). arXiv preprint arXiv:1705.03551.
9. Tushar Khot, Harsh Trivedi, Matthew Finlayson, Yao Fu, Kyle Richardson, Peter Clark, and Ashish Sabharwal. 2022. Decomposed prompting: A modular approach for solving complex tasks. arXiv preprint arXiv:2210.02406. [find](https://scholar.google.com/scholar?q=Tushar+Khot%2C+Harsh+Trivedi%2C+Matthew+Finlayson%2C+Yao+Fu%2C+Kyle+Richardson%2C+Peter+Clark%2C+and+Ashish+Sabharwal%2E+2022%2E+Decomposed+prompting%3A+A+modular+approach+for+solving+complex+tasks%2E+arXiv+preprint+arXi)
10. Tom Kwiatkowski, Jennimaria Palomaki, Olivia Redfield, Michael Collins, Ankur Parikh, Chris Alberti, Danielle Epstein, Illia Polosukhin, Jacob Devlin, Kenton Lee, and 1 others. 2019. [Natural questions: a benchmark for question answering research](https://md.research.elamur.ai/free/en/papers/kwiatkowski-2019-natural-questions-benchmark-question-answering/). Transactions of the Association for Computational Lin…
11. Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, and 1 others. 2020. [Retrieval-augmented generation for knowledge-intensive nlp tasks](https://md.research.elamur.ai/free/en/papers/lewis-2020-retrieval-augmented-generation-knowledge-intensive/). Advances in neural information processing systems, 33:9459–…
12. Alex Mallen, Akari Asai, Victor Zhong, Rajarshi Das, Daniel Khashabi, and Hannaneh Hajishirzi. 2023. When not to trust language models: Investigating effectiveness of parametric and non-parametric memories. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (V… [find](https://scholar.google.com/scholar?q=Alex+Mallen%2C+Akari+Asai%2C+Victor+Zhong%2C+Rajarshi+Das%2C+Daniel+Khashabi%2C+and+Hannaneh+Hajishirzi%2E+2023%2E+When+not+to+trust+language+models%3A+Investigating+effectiveness+of+parametric+and+non%2Dparametric+mem)
13. Tianyu Gao, Howard Yen, Jiatong Yu, and Danqi Chen. 2023a. [Enabling large language models to generate text with citations](https://md.research.elamur.ai/free/en/papers/gao-2023-enabling-large-language-models-generate/). arXiv preprint arXiv:2305.14627.
14. Multi-Linguality Multi-Functionality Multi-Granularity. 2024. [M3-embedding: Multi-linguality, multi-functionality, multi-granularity text embeddings through self-knowledge distillation](https://md.research.elamur.ai/free/en/papers/chen-2024-m3-embedding-multi-linguality/).
15. Yunfan Gao, Yun Xiong, Xinyu Gao, Kangxiang Jia, Jinliu Pan, Yuxi Bi, Yixin Dai, Jiawei Sun, Haofen Wang, and Haofen Wang. 2023b. [Retrieval-augmented generation for large language models: A survey](https://md.research.elamur.ai/free/en/papers/gao-2023-retrieval-augmented-generation-large-language/). arXiv preprint arXiv:2312.10997, 2(1).
16. Thang Nguyen, Peter Chin, and Yu-Wing Tai. 2025. [Ma-rag: Multi-agent retrieval-augmented generation via collaborative chain-of-thought reasoning](https://md.research.elamur.ai/free/en/papers/nguyen-2025-ma-rag-multi-agent-retrieval/). arXiv preprint arXiv:2505.20096.
17. Xanh Ho, Anh-Khoa Duong Nguyen, Saku Sugawara, and Akiko Aizawa. 2020. [Constructing a multi-hop qa dataset for comprehensive evaluation of reasoning steps](https://md.research.elamur.ai/free/en/papers/ho-2020-constructing-multi-hop-qa-dataset/). arXiv preprint arXiv:2011.01060.
18. AJ Oche, AG Folashade, T Ghosal, and A Biswas. A systematic review of key retrieval-augmented generation (rag) systems: Progress, gaps, and future directions. arxiv 2024. arXiv preprint arXiv:2409.15730. — [arXiv:2409.15730](https://arxiv.org/abs/2409.15730) [find](https://scholar.google.com/scholar?q=AJ+Oche%2C+AG+Folashade%2C+T+Ghosal%2C+and+A+Biswas%2E+A+systematic+review+of+key+retrieval%2Daugmented+generation+%28rag%29+systems%3A+Progress%2C+gaps%2C+and+future+directions%2E+arxiv+2024%2E+arXiv+preprint+arXiv%3A2409%2E157)
19. Pengyue Jia, Derong Xu, Xiaopeng Li, Zhaocheng Du, Xiangyang Li, Yichao Wang, Yuhao Wang, Qidong Liu, Maolin Wang, Huifeng Guo, and 1 others. 2025. Bridging relevance and reasoning: Rationale distillation in retrieval-augmented generation. In Findings of the Association for Computational Linguistics… [find](https://scholar.google.com/scholar?q=Pengyue+Jia%2C+Derong+Xu%2C+Xiaopeng+Li%2C+Zhaocheng+Du%2C+Xiangyang+Li%2C+Yichao+Wang%2C+Yuhao+Wang%2C+Qidong+Liu%2C+Maolin+Wang%2C+Huifeng+Guo%2C+and+1+others%2E+2025%2E+Bridging+relevance+and+reasoning%3A+Rationale+distilla)
20. Zhengbao Jiang, Frank F Xu, Luyu Gao, Zhiqing Sun, Qian Liu, Jane Dwivedi-Yu, Yiming Yang, Jamie Callan, and Graham Neubig. 2023. [Active retrieval augmented generation](https://md.research.elamur.ai/free/en/papers/jiang-2023-active-retrieval-augmented-generation/). In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 7969–7992.
21. Panupong Pasupat and Percy Liang. 2015. [Compositional semantic parsing on semi-structured tables](https://md.research.elamur.ai/free/en/papers/pasupat-2015-compositional-semantic-parsing-semi-structured/). arXiv preprint arXiv:1508.00305.
22. Alec Radford, Karthik Narasimhan, Tim Salimans, Ilya Sutskever, and 1 others. 2018. Improving language understanding by generative pre-training. [find](https://scholar.google.com/scholar?q=Alec+Radford%2C+Karthik+Narasimhan%2C+Tim+Salimans%2C+Ilya+Sutskever%2C+and+1+others%2E+2018%2E+Improving+language+understanding+by+generative+pre%2Dtraining%2E)
23. Ori Ram, Yoav Levine, Itay Dalmedigos, Dor Muhlgay, Amnon Shashua, Kevin Leyton-Brown, and Yoav Shoham. 2023. [In-context retrieval-augmented language models](https://md.research.elamur.ai/free/en/papers/ram-2023-context-retrieval-augmented-language-models/). Transactions of the Association for Computational Linguistics, 11:1316–1331.
24. Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Xiao Bi, Haowei Zhang, Mingchuan Zhang, YK Li, Yang Wu, and 1 others. 2024. [Deepseekmath: Pushing the limits of mathematical reasoning in open language models](https://md.research.elamur.ai/free/en/papers/shao-2024-deepseekmath-pushing-limits-mathematical-reasoning/). arXiv preprint arXiv:2402.03300.
25. Chaitanya Sharma. 2025. Retrieval-augmented generation: A comprehensive survey of architectures, enhancements, and robustness frontiers. arXiv preprint arXiv:2506.00054. — [arXiv:2506.00054](https://arxiv.org/abs/2506.00054) [find](https://scholar.google.com/scholar?q=Chaitanya+Sharma%2E+2025%2E+Retrieval%2Daugmented+generation%3A+A+comprehensive+survey+of+architectures%2C+enhancements%2C+and+robustness+frontiers%2E+arXiv+preprint+arXiv%3A2506%2E00054%2E+%E2%80%94+%5BarXiv%3A2506%2E00054%5D%28https%3A%2F%2Fa)
26. Weizhou Shen, Yeyun Gong, Yelong Shen, Song Wang, Xiaojun Quan, Nan Duan, and Weizhu Chen. 2023. Joint generator-ranker learning for natural language generation. In Findings of the Association for Computational Linguistics: ACL 2023, pages 7681–7699. [find](https://scholar.google.com/scholar?q=Weizhou+Shen%2C+Yeyun+Gong%2C+Yelong+Shen%2C+Song+Wang%2C+Xiaojun+Quan%2C+Nan+Duan%2C+and+Weizhu+Chen%2E+2023%2E+Joint+generator%2Dranker+learning+for+natural+language+generation%2E+In+Findings+of+the+Association+for+Com)
27. Ivan Stelmakh, Yi Luan, Bhuwan Dhingra, and Ming-Wei Chang. 2022. [Asqa: Factoid questions meet long-form answers](https://md.research.elamur.ai/free/en/papers/stelmakh-2022-asqa-factoid-questions-meet-long/). arXiv preprint arXiv:2204.06092.

*Updated 1 November 2026*