YunahJang/query-robustness

How You Ask Matters: Human Query Rewrites

Dataset for How You Ask Matters! Adaptive RAG Robustness to Query Variations (EMNLP 2026).

Parallel query variations for measuring how retrieval-augmented generation (RAG) systems respond to different phrasings of the same information need.

Each config holds 500 questions from one QA benchmark. Every question comes with its original text, its gold answers, and six machine-generated rewrites. About 100 questions per config also have up to two independent human rewrites. Gold answers are shared across all forms, so any change in system output comes from the query wording alone.

Configs

Config	Source benchmark	Hop	Questions	With human rewrites	Human rewrites
nq	Natural Questions	single	500	99	194
triviaqa	TriviaQA	single	500	100	200
squad	SQuAD v1.1	single	500	100	200
hotpotqa	HotpotQA	multi	500	100	199
2wikimultihopqa	2WikiMultiHopQA	multi	500	98	190
musique	MuSiQue	multi	500	100	200

from datasets import load_dataset

ds = load_dataset("YunahJang/query-robustness", "hotpotqa", split="test")
human = ds.filter(lambda r: r["human_rewrite_1"] is not None)

Fields

Field	Type	Description
id	string	Question ID from the source subset
original_question	string	Original question, verbatim
answers	list[string]	Gold answers, verbatim
human_rewrite_1, human_rewrite_2	string or null	Human rewrites; null if none
annotator_1, annotator_2	string or null	Anonymous annotator ID for each rewrite
formality	string	Casual, informal rewrite (LLM)
readability	string	Rewrite with harder vocabulary and syntax (LLM)
declarative	string	The question restated as a declarative sentence (LLM)
imperative	string	The question restated as a command (LLM)
spelling_error	string	Misspellings added with nlpaug
grammar_error	string	Grammatical errors added by an LLM and checked with LanguageTool

LLM rewrites were generated with gpt-5-mini-2025-08-07.

Human rewrites. Annotators saw the original question and wrote the same information need in their own words. Rewrites are kept as written, including lowercase, keyword-style phrasing, and typos, because that variation is what the dataset is for. Annotator IDs are anonymous integers. No names or other personal information are included.

Passages, evidence documents, and retrieval contexts are not included. To get them, use the original benchmarks or a standard Wikipedia corpus.

Licensing

This dataset contains human-written rewrites of questions from six existing QA datasets, along with the original questions and answers. The rewrites are adaptations of the originals. The full dataset is released under CC BY-SA 4.0, as required by the ShareAlike terms of NQ, SQuAD, and HotpotQA.

Subset	Source	Original license
nq	Kwiatkowski et al., 2019	CC BY-SA 3.0
squad	Rajpurkar et al., 2016 (v1.1)	CC BY-SA 4.0
hotpotqa	Yang et al., 2018	CC BY-SA 4.0
musique	Trivedi et al., 2022	CC BY 4.0
2wikimultihopqa	Ho et al., 2020	Apache-2.0
triviaqa	Joshi et al., 2017	Apache-2.0 (the University of Washington does not own the copyright of the original questions)

The 500-question test subsets follow prior work (Moskvoretskii et al., 2025; Jeong et al., 2024; Trivedi et al., 2023), which sampled from the original test/dev splits without altering the question or answer text. If you use a subset, please also cite the original dataset.

Changes to the source material

• Original questions and gold answers are copied verbatim.
• Six machine-generated rewrites were added to every question.
• Up to two human rewrites were added to about 100 questions per subset.
• Passages, evidence, and training splits are not included.

Data sources

The 500-question subsets were taken from the release repositories of AdaRAGUE (Moskvoretskii et al., 2025), Adaptive-RAG (Jeong et al., 2024), and IRCoT (Trivedi et al., 2023). Their single-hop data is based on the DPR preprocessing (Karpukhin et al., 2020). These repositories only sampled the data, so each question is still covered by the license of its original benchmark, listed above.

Citation

If you use this dataset, please cite our paper and the original benchmark for each subset you use.

@inproceedings{jang2026howyouask,
  title         = {How You Ask Matters! Adaptive {RAG} Robustness to Query Variations},
  author        = {Jang, Yunah and Sundriyal, Megha and Jung, Kyomin and Cha, Meeyoung},
  booktitle     = {Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing (EMNLP)},
  year          = {2026},
  eprint        = {2604.10745},
  archivePrefix = {arXiv},
  url           = {https://arxiv.org/abs/2604.10745}
}

Source datasets

@article{kwiatkowski2019natural,
  title   = {Natural Questions: A Benchmark for Question Answering Research},
  author  = {Kwiatkowski, Tom and Palomaki, Jennimaria and Redfield, Olivia and Collins, Michael and Parikh, Ankur and Alberti, Chris and Epstein, Danielle and Polosukhin, Illia and Devlin, Jacob and Lee, Kenton and Toutanova, Kristina and Jones, Llion and Kelcey, Matthew and Chang, Ming-Wei and Dai, Andrew M. and Uszkoreit, Jakob and Le, Quoc and Petrov, Slav},
  journal = {Transactions of the Association for Computational Linguistics},
  volume  = {7},
  pages   = {452--466},
  year    = {2019}
}

@inproceedings{joshi2017triviaqa,
  title     = {{T}rivia{QA}: A Large Scale Distantly Supervised Challenge Dataset for Reading Comprehension},
  author    = {Joshi, Mandar and Choi, Eunsol and Weld, Daniel and Zettlemoyer, Luke},
  booktitle = {Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (ACL)},
  pages     = {1601--1611},
  year      = {2017}
}

@inproceedings{rajpurkar2016squad,
  title     = {{SQ}u{AD}: 100,000+ Questions for Machine Comprehension of Text},
  author    = {Rajpurkar, Pranav and Zhang, Jian and Lopyrev, Konstantin and Liang, Percy},
  booktitle = {Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing (EMNLP)},
  pages     = {2383--2392},
  year      = {2016}
}

@inproceedings{yang2018hotpotqa,
  title     = {{H}otpot{QA}: A Dataset for Diverse, Explainable Multi-hop Question Answering},
  author    = {Yang, Zhilin and Qi, Peng and Zhang, Saizheng and Bengio, Yoshua and Cohen, William and Salakhutdinov, Ruslan and Manning, Christopher D.},
  booktitle = {Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing (EMNLP)},
  pages     = {2369--2380},
  year      = {2018}
}

@inproceedings{ho2020constructing,
  title     = {Constructing A Multi-hop {QA} Dataset for Comprehensive Evaluation of Reasoning Steps},
  author    = {Ho, Xanh and Duong Nguyen, Anh-Khoa and Sugawara, Saku and Aizawa, Akiko},
  booktitle = {Proceedings of the 28th International Conference on Computational Linguistics (COLING)},
  pages     = {6609--6625},
  year      = {2020}
}

@article{trivedi2022musique,
  title   = {{M}u{S}i{Q}ue: Multihop Questions via Single-hop Question Composition},
  author  = {Trivedi, Harsh and Balasubramanian, Niranjan and Khot, Tushar and Sabharwal, Ashish},
  journal = {Transactions of the Association for Computational Linguistics},
  volume  = {10},
  pages   = {539--554},
  year    = {2022}
}

Subset sources

@inproceedings{moskvoretskii2025adaptive,
  title     = {Adaptive Retrieval Without Self-Knowledge? Bringing Uncertainty Back Home},
  author    = {Moskvoretskii, Viktor and Lysyuk, Maria and Salnikov, Mikhail and Ivanov, Nikolay and Pletenev, Sergey and Galimzianova, Daria and Krayko, Nikita and Konovalov, Vasily and Nikishina, Irina and Panchenko, Alexander},
  booktitle = {Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (ACL)},
  year      = {2025}
}

@inproceedings{jeong2024adaptive,
  title     = {Adaptive-{RAG}: Learning to Adapt Retrieval-Augmented Large Language Models through Question Complexity},
  author    = {Jeong, Soyeong and Baek, Jinheon and Cho, Sukmin and Hwang, Sung Ju and Park, Jong C.},
  booktitle = {Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics (NAACL)},
  pages     = {7036--7050},
  year      = {2024}
}

@inproceedings{trivedi2023interleaving,
  title     = {Interleaving Retrieval with Chain-of-Thought Reasoning for Knowledge-Intensive Multi-Step Questions},
  author    = {Trivedi, Harsh and Balasubramanian, Niranjan and Khot, Tushar and Sabharwal, Ashish},
  booktitle = {Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (ACL)},
  pages     = {10014--10037},
  year      = {2023}
}

@inproceedings{karpukhin2020dense,
  title     = {Dense Passage Retrieval for Open-Domain Question Answering},
  author    = {Karpukhin, Vladimir and Oguz, Barlas and Min, Sewon and Lewis, Patrick and Wu, Ledell and Edunov, Sergey and Chen, Danqi and Yih, Wen-tau},
  booktitle = {Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP)},
  pages     = {6769--6781},
  year      = {2020}
}