---
title: "CRAFT: Training-Free Cascaded Retrieval for Tabular QA (2026)"
description: "CRAFT addresses open-domain table question answering, where a system must locate the right table in… 29 references, cited by 1 paper here."
authors: ["Adarsh Singh", "Kushal Raj Bhandari", "Jianxi Gao", "Soham Dan", "Vivek Gupta"]
year: 2026
venue: "Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)"
source: "https://aclanthology.org/2026.acl-long.149/"
doi: "10.18653/v1/2026.acl-long.149"
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/singh-2026-craft-training-free-cascaded-retrieval/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/singh-2026-craft-training-free-cascaded-retrieval/)

[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/11/) / CRAFT: Training-Free Cascaded Retrieval for Tabular QA

# CRAFT: Training-Free Cascaded Retrieval for Tabular QA

**Authors:** Adarsh Singh, Kushal Raj Bhandari, Jianxi Gao, [Soham Dan](https://md.research.elamur.ai/free/en/authors/soham-dan/) and [Vivek Gupta](https://md.research.elamur.ai/free/en/authors/vivek-gupta/) · **Year:** 2026 · **Venue:** [Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)](https://md.research.elamur.ai/free/en/venues/acl/) ([ACL 2026](https://md.research.elamur.ai/free/en/venues/acl/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.acl-long.149/) · [PDF](https://aclanthology.org/2026.acl-long.149.pdf) · [DOI 10.18653/v1/2026.acl-long.149](https://doi.org/10.18653/v1/2026.acl-long.149) · [OpenAlex](https://openalex.org/W4415327343)

**Citations:** 0 citations (OpenAlex, 2026-10-14) · 1 in our library, of the 149,439 papers whose bibliographies we read

## Summary
CRAFT addresses open-domain table question answering, where a system must locate the right table in a very large collection before answering. Dense retrievers such as DTR and DPR are costly at scale and must be retrained for new data. CRAFT avoids training entirely and chains off-the-shelf models into a three-stage cascade. Gemini-1.5-Flash first adds generated titles, descriptions and sub-questions. SPLADE then shrinks the corpus to 5,000 candidates. Next, a dense encoder scores compact "mini-tables", each made of column headers plus the five most relevant rows. Finally, a stronger embedding model reranks the shortlist, and open LLMs produce the answers. On NQ-Tables, CRAFT sets a new state of the art for Recall@1. On OTT-QA, it falls behind fine-tuned hybrid systems at top-1 but comes close at deeper cutoffs. The work shows that careful pipeline design can rival dataset-specific fine-tuning. This makes it a practical, plug-and-play option for changing domains or settings with little labeled data.

## Key points
- **Benchmarks:** NQ-Tables (169,898 tables, 966 test queries, single-hop) and OTT-QA (419,183 tables, 2,214 test queries, mixing text and tables).
- **Models per stage:**
  - Stage 1 uses SPLADE on both datasets.
  - Stage 2 uses all-mpnet-base-v2 for NQ-Tables and Jina Embeddings v3 for OTT-QA.
  - Stage 3 uses text-embedding-3-small/large for NQ-Tables and gemini-embedding-001 for OTT-QA.
- **NQ-Tables recall:** R@1 41.13, R@10 87.16 and R@50 96.84. With text-embedding-3-large, R@1 rises to 49.84.
- **OTT-QA recall:** R@1 55.56, R@10 89.88 and R@50 96.07.
- **Efficiency and QA setup:** The mini-table design reduces the embedding calls made at query time by as much as 33× and shortens contexts by roughly 70%. Answers are generated by Llama3-8B, Qwen2.5-7B and Mistral-7B and scored with F1. The OTT-QA QA evaluation relied on gold passages, and the multi-stage design increases per-query latency.

*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

- [Microsoft](https://md.research.elamur.ai/free/en/companies/microsoft/) — Company, Redmond, United States · 325 works · 142,931 citations
- [Arizona State University](https://md.research.elamur.ai/free/en/universities/arizona-state-university/) — University, Tempe, United States · 50 works · 2,032 citations
- [Rensselaer Polytechnic Institute](https://md.research.elamur.ai/free/en/universities/rensselaer-polytechnic-institute/) — University, Troy, United States · 21 works · 4,933 citations

📚 Shelves (5): [🖥️ 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 · [🗄️ Information Systems](https://md.research.elamur.ai/free/en/shelves/computer-science/information-systems/) · 1,437 works

🧭 Topics (2): [❓ Question answering](https://md.research.elamur.ai/free/en/topics/question-answering/) · 441 works · [🔍 Information retrieval & search](https://md.research.elamur.ai/free/en/topics/information-retrieval/) · 473 works

## Cite

Cite as: Adarsh Singh, Kushal Raj Bhandari, Jianxi Gao, Soham Dan and Vivek Gupta. 2026. CRAFT: Training-Free Cascaded Retrieval for Tabular QA. 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.

```bibtex
@inproceedings{singh2026craft,
  title = {{CRAFT: Training-Free Cascaded Retrieval for Tabular QA}},
  author = {Adarsh Singh, Kushal Raj Bhandari, Jianxi Gao, Soham Dan and Vivek Gupta},
  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.149},
  url = {https://doi.org/10.18653/v1/2026.acl-long.149}
}
```

- [← previous CPTCoder: A Reliable LLM System for Medical Procedure Code Prediction](https://md.research.elamur.ai/free/en/papers/wang-2026-cptcoder-reliable-llm-system-medical/)
- [next → Cross-Cultural Transfer of Emoji Semantics and Sentiment in Financial Social Media](https://md.research.elamur.ai/free/en/papers/mahrous-2026-cross-cultural-transfer-emoji-semantics/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/11/)

## References

29 references, 12 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Gilbert Badaro, Mohammed Saeed, and Paolo Papotti. 2023. [Transformers for Tabular Data Representation: A Survey of Models and Applications](https://md.research.elamur.ai/free/en/papers/badaro-2023-transformers-tabular-data-representation-survey/). Transactions of the Association for Computational Linguistics, 11:227–249.
2. Wenhu Chen, Ming-Wei Chang, Eva Schlinger, William Wang, and William W Cohen. 2020. Open question answering over tables and text. arXiv preprint arXiv:2010.10439. [find](https://scholar.google.com/scholar?q=Wenhu+Chen%2C+Ming%2DWei+Chang%2C+Eva+Schlinger%2C+William+Wang%2C+and+William+W+Cohen%2E+2020%2E+Open+question+answering+over+tables+and+text%2E+arXiv+preprint+arXiv%3A2010%2E10439%2E)
3. Wenqi Fan, Yujuan Ding, Liangbo Ning, Shijie Wang, Hengyun Li, Dawei Yin, Tat-Seng Chua, and Qing Li. 2024. [A Survey on RAG Meeting LLMs: Towards Retrieval-Augmented Large Language Models](https://md.research.elamur.ai/free/en/papers/fan-2024-survey-rag-meeting-llms-towards/). In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, pages 6491–6501, Barce…
4. Thibault Formal, Benjamin Piwowarski, and Stéphane Clinchant. 2021. [SPLADE: Sparse Lexical and Expansion Model for First Stage Ranking](https://md.research.elamur.ai/free/en/papers/formal-2021-splade-sparse-lexical-expansion/). In Proceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval, pages 2288–2292, New York, NY, USA. Association…
5. Michael Glass, Gaetano Rossiello, Md Faisal Mahbub Chowdhury, Ankita Naik, Pengshan Cai, and Alfio Gliozzo. 2022. [Re2G: Retrieve, rerank, generate](https://md.research.elamur.ai/free/en/papers/glass-2022-re2g-retrieve-rerank-generate/). In Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages…
6. Aaron Grattafiori, Abhimanyu Dubey, and Abhinav Jauhri. 2024. [The llama 3 herd of models](https://md.research.elamur.ai/free/en/papers/grattafiori-2024-llama-3-herd-models/). Preprint, arXiv:2407.21783.
7. Jonathan Herzig, Thomas Müller, Syrine Krichene, and Julian Eisenschlos. 2021. [Open Domain Question Answering over Tables via Dense Retrieval](https://md.research.elamur.ai/free/en/papers/herzig-2021-open-domain-question-answering-over/). In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 512…
8. Jonathan Herzig, Paweł Krzysztof Nowak, Thomas Müller, Francesco Piccinno, and Julian Martin Eisenschlos. 2020. [Tapas: Weakly supervised table parsing via pre-training](https://md.research.elamur.ai/free/en/papers/herzig-2020-tapas-weakly-supervised-table-parsing/). arXiv preprint arXiv:2004.02349.
9. Minghao Hu, Yuxing Peng, Zhen Huang, and Dongsheng Li. 2019. Retrieve, read, rerank: Towards end-to-end multi-document reading comprehension. arXiv preprint arXiv:1906.04618. [find](https://scholar.google.com/scholar?q=Minghao+Hu%2C+Yuxing+Peng%2C+Zhen+Huang%2C+and+Dongsheng+Li%2E+2019%2E+Retrieve%2C+read%2C+rerank%3A+Towards+end%2Dto%2Dend+multi%2Ddocument+reading+comprehension%2E+arXiv+preprint+arXiv%3A1906%2E04618%2E)
10. Albert Q. Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, Devendra Singh Chaplot, Diego de Las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lucile Saulnier, Lélio Renard Lavaud, Marie-Anne Lachaux, Pierre Stock, Teven Le Scao, Thibaut Lavril, Thomas Wang, Timothée Lacroi… [find](https://scholar.google.com/scholar?q=Albert+Q%2E+Jiang%2C+Alexandre+Sablayrolles%2C+Arthur+Mensch%2C+Chris+Bamford%2C+Devendra+Singh+Chaplot%2C+Diego+de+Las+Casas%2C+Florian+Bressand%2C+Gianna+Lengyel%2C+Guillaume+Lample%2C+Lucile+Saulnier%2C+Le%CC%81lio+Renard+La)
11. Nengzheng Jin, Dongfang Li, Junying Chen, Joanna Siebert, and Qingcai Chen. 2023. Enhancing Open-Domain Table Question Answering via Syntax- and Structure-aware Dense Retrieval. In Proceedings of the 13th International Joint Conference on Natural Language Processing and the 3rd Conference of the Asi… [find](https://scholar.google.com/scholar?q=Nengzheng+Jin%2C+Dongfang+Li%2C+Junying+Chen%2C+Joanna+Siebert%2C+and+Qingcai+Chen%2E+2023%2E+Enhancing+Open%2DDomain+Table+Question+Answering+via+Syntax%2D+and+Structure%2Daware+Dense+Retrieval%2E+In+Proceedings+of+the+)
12. Vladimir Karpukhin, Barlas Oguz, Sewon Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih. 2020. [Dense passage retrieval for open-domain question answering](https://md.research.elamur.ai/free/en/papers/karpukhin-2020-dense-passage-retrieval-open-domain/). In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 6769– 6781, Onl…
13. Jinhyuk Lee, Feiyang Chen, Sahil Dua, Daniel Cer, Madhuri Shanbhogue, Iftekhar Naim, Gustavo Hernández Ábrego, Zhe Li, Kaifeng Chen, Henrique Schechter Vera, et al. 2025. [Gemini embedding: Generalizable embeddings from gemini](https://md.research.elamur.ai/free/en/papers/lee-2025-gemini-embedding-generalizable-embeddings-gemini/). arXiv preprint arXiv:2503.07891.
14. Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, Sebastian Riedel, and Douwe Kiela. 2020. [Retrieval-augmented generation for knowledge-intensive nlp tasks](https://md.research.elamur.ai/free/en/papers/lewis-2020-retrieval-augmented-generation-knowledge-intensive/). In Proceedings of the 34th Internati…
15. Stephen Robertson and Hugo Zaragoza. 2009. The Probabilistic Relevance Framework: BM25 and Beyond. Foundations and Trends® in Information Retrieval, 3(4):333–389. [find](https://scholar.google.com/scholar?q=Stephen+Robertson+and+Hugo+Zaragoza%2E+2009%2E+The+Probabilistic+Relevance+Framework%3A+BM25+and+Beyond%2E+Foundations+and+Trends%C2%AE+in+Information+Retrieval%2C+3%284%29%3A333%E2%80%93389%2E)
16. Da Li, Keping Bi, Jiafeng Guo, and Xueqi Cheng. 2025. Tailoring Table Retrieval from a Field-aware Hybrid Matching Perspective. Preprint, arXiv:2503.02251. [find](https://scholar.google.com/scholar?q=Da+Li%2C+Keping+Bi%2C+Jiafeng+Guo%2C+and+Xueqi+Cheng%2E+2025%2E+Tailoring+Table+Retrieval+from+a+Field%2Daware+Hybrid+Matching+Perspective%2E+Preprint%2C+arXiv%3A2503%2E02251%2E)
17. Saba Sturua, Isabelle Mohr, Mohammad Kalim Akram, Michael Günther, Bo Wang, Markus Krimmel, Feng Wang, Georgios Mastrapas, Andreas Koukounas, Nan Wang, and Han Xiao. 2024. jina-embeddings-v3: Multilingual embeddings with task lora. Preprint, arXiv:2409.10173. [find](https://scholar.google.com/scholar?q=Saba+Sturua%2C+Isabelle+Mohr%2C+Mohammad+Kalim+Akram%2C+Michael+Gu%CC%88nther%2C+Bo+Wang%2C+Markus+Krimmel%2C+Feng+Wang%2C+Georgios+Mastrapas%2C+Andreas+Koukounas%2C+Nan+Wang%2C+and+Han+Xiao%2E+2024%2E+jina%2Dembeddings%2Dv3%3A+Multili)
18. Xinyi Liang, Rui Hu, Yu Liu, and Konglin Zhu. 2024. Open-Domain Question Answering over Tables with Large Language Models. In De-Shuang Huang, Yijie Pan, and Jiayang Guo, editors, Advanced Intelligent Computing Technology and Applications, volume 14873, pages 347–358. Springer Nature Singapore, Sing… [find](https://scholar.google.com/scholar?q=Xinyi+Liang%2C+Rui+Hu%2C+Yu+Liu%2C+and+Konglin+Zhu%2E+2024%2E+Open%2DDomain+Question+Answering+over+Tables+with+Large+Language+Models%2E+In+De%2DShuang+Huang%2C+Yijie+Pan%2C+and+Jiayang+Guo%2C+editors%2C+Advanced+Intelligent)
19. Gemini Team, Rohan Anil, Sebastian Borgeaud, Jean-Baptiste Alayrac, Jiahui Yu, Radu Soricut, Johan Schalkwyk, Andrew M. Dai, Anja Hauth, and Katie Millican. 2025. [Gemini: A family of highly capable multimodal models](https://md.research.elamur.ai/free/en/papers/gemini-team-2023-gemini-family-highly-capable-multimodal/). Preprint, arXiv:2312.11805.
20. Sheng-Chieh Lin, Minghan Li, and Jimmy Lin. 2023a. Aggretriever: A simple approach to aggregate textual representations for robust dense passage retrieval. Transactions of the Association for Computational Linguistics, 11:436–452. [find](https://scholar.google.com/scholar?q=Sheng%2DChieh+Lin%2C+Minghan+Li%2C+and+Jimmy+Lin%2E+2023a%2E+Aggretriever%3A+A+simple+approach+to+aggregate+textual+representations+for+robust+dense+passage+retrieval%2E+Transactions+of+the+Association+for+Computat)
21. Lidan Wang, Jimmy Lin, and Donald Metzler. 2011. A cascade ranking model for efficient ranked retrieval. In Proceedings of the 34th international ACM SIGIR conference on Research and development in Information Retrieval, pages 105–114. [find](https://scholar.google.com/scholar?q=Lidan+Wang%2C+Jimmy+Lin%2C+and+Donald+Metzler%2E+2011%2E+A+cascade+ranking+model+for+efficient+ranked+retrieval%2E+In+Proceedings+of+the+34th+international+ACM+SIGIR+conference+on+Research+and+development+in+In)
22. Weizhe Lin, Rexhina Blloshmi, Bill Byrne, Adria de Gispert, and Gonzalo Iglesias. 2023b. LI-RAGE: Late Interaction Retrieval Augmented Generation with Explicit Signals for Open-Domain Table Question Answering. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics… [find](https://scholar.google.com/scholar?q=Weizhe+Lin%2C+Rexhina+Blloshmi%2C+Bill+Byrne%2C+Adria+de+Gispert%2C+and+Gonzalo+Iglesias%2E+2023b%2E+LI%2DRAGE%3A+Late+Interaction+Retrieval+Augmented+Generation+with+Explicit+Signals+for+Open%2DDomain+Table+Question+A)
23. Andrew Yates, Rodrigo Nogueira, and Jimmy Lin. 2021. Pretrained transformers for text ranking: BERT and beyond. In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies: Tutorials, pages 1–4, Online. Association… [find](https://scholar.google.com/scholar?q=Andrew+Yates%2C+Rodrigo+Nogueira%2C+and+Jimmy+Lin%2E+2021%2E+Pretrained+transformers+for+text+ranking%3A+BERT+and+beyond%2E+In+Proceedings+of+the+2021+Conference+of+the+North+American+Chapter+of+the+Association+f)
24. A Dehai Min, Nan Hu, Rihui Jin, Nuo Lin, Jiaoyan Chen, Yongrui Chen, Yu Li, Guilin Qi, Yun Li, Nijun Li, and Qianren Wang. 2024. Exploring the Impact of Table-to-Text Methods on Augmenting LLM-based Question Answering with Domain Hybrid Data. In Proceedings of the 2024 Conference of the North Americ… [find](https://scholar.google.com/scholar?q=A+Dehai+Min%2C+Nan+Hu%2C+Rihui+Jin%2C+Nuo+Lin%2C+Jiaoyan+Chen%2C+Yongrui+Chen%2C+Yu+Li%2C+Guilin+Qi%2C+Yun+Li%2C+Nijun+Li%2C+and+Qianren+Wang%2E+2024%2E+Exploring+the+Impact+of+Table%2Dto%2DText+Methods+on+Augmenting+LLM%2Dbased+Q)
25. OpenAI. 2024. text-embedding-3-small. Model card, accessed 19 May 2025. [find](https://scholar.google.com/scholar?q=OpenAI%2E+2024%2E+text%2Dembedding%2D3%2Dsmall%2E+Model+card%2C+accessed+19+May+2025%2E)
26. Feifei Pan, Mustafa Canim, Michael Glass, Alfio Gliozzo, and James Hendler. 2022. End-to-End Table Question Answering via Retrieval-Augmented Generation. Preprint, arXiv:2203.16714. — [arXiv:2203.16714](https://arxiv.org/abs/2203.16714) [find](https://scholar.google.com/scholar?q=Feifei+Pan%2C+Mustafa+Canim%2C+Michael+Glass%2C+Alfio+Gliozzo%2C+and+James+Hendler%2E+2022%2E+End%2Dto%2DEnd+Table+Question+Answering+via+Retrieval%2DAugmented+Generation%2E+Preprint%2C+arXiv%3A2203%2E16714%2E+%E2%80%94+%5BarXiv%3A2203%2E1671)
27. Qwen, An Yang, Baosong Yang, Beichen Zhang, and Hui. 2025. Qwen2.5 technical report. arXiv preprint arXiv:2412.15115. [find](https://scholar.google.com/scholar?q=Qwen%2C+An+Yang%2C+Baosong+Yang%2C+Beichen+Zhang%2C+and+Hui%2E+2025%2E+Qwen2%2E5+technical+report%2E+arXiv+preprint+arXiv%3A2412%2E15115%2E)
28. Nils Reimers and Iryna Gurevych. 2019. [Sentence-BERT: Sentence embeddings using siamese BERT-networks](https://md.research.elamur.ai/free/en/papers/reimers-2019-sentence-bert-sentence-embeddings-siamese/). In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing (EMNLP).
29. Hyperparameter Configuration Configuration Parameter Batch Size Learning Rate Training Epochs Optimizer Initialization THYME Dropout (p sem , p lex ) THYME FLOPS Reg. (λ q , λ t ) Value 144 1 × 10 −5 50 Adam BERT-base 0.15 1 × 10 −4 *Table 13: Training and inference configuration used for all baseli… [find](https://scholar.google.com/scholar?q=Hyperparameter+Configuration+Configuration+Parameter+Batch+Size+Learning+Rate+Training+Epochs+Optimizer+Initialization+THYME+Dropout+%28p+sem+%2C+p+lex+%29+THYME+FLOPS+Reg%2E+%28%CE%BB+q+%2C+%CE%BB+t+%29+Value+144+1+%C3%97+10+%E2%88%925+)

*Updated 31 October 2026*