Elamur Research⌕
☰
home / papers / 2026 / TableMBR: Minimum Bayes Risk Table Generation Based on Structural Consistency

TableMBR: Minimum Bayes Risk Table Generation Based on Structural Consistency

Paper · 2026 · Daiki Yoshida, Hiroyuki Deguchi, Yusuke Sakai et al.

Summary

Text-to-table generation turns free text into structured tables. LLM-based pipelines such as Map&Make can induce a schema on the fly, but their outputs are unreliable: they add superfluous columns, miss values, and get numbers wrong. This paper adapts minimum Bayes risk (MBR) decoding, best known from machine translation, to table generation. TableMBR samples several candidate tables from Map&Make with temperature sampling and treats them as pseudo-references for one another. It then picks the candidate with the highest average agreement. Agreement is measured by a structural utility that averages F1 over row headers, column headers, and cells. Each cell is represented together with its row and column header, so its position counts. Element matching can be exact, lexical, or semantic. Experiments with Gemini-2.5 Flash and GPT-4.1 mini on Rotowire (open schema) and LiveSum (fixed schema) show consistent gains over Map&Make. The authors also find that weighting the utility toward precision works best, because a single spurious row or column badly hurts a table's usefulness.

Key points

  • Results: relative to the Map&Make baseline, the method improves Rotowire F1 by as much as 15% and LiveSum accuracy by as much as 23% (relative gains).
  • Utility design: the score is the mean of row-header, column-header, and cell F1. Cells are (row header, column header, value) triples, and their similarity is the product of sub-element similarities. Matching options are exact match, chrF-style lexical similarity (Popović, 2015), and BERTScore-style semantic similarity (Zhang et al., 2019).
  • Precision vs. recall: in the reported example for GPT-4.1 mini, cell F1 is 63.14 with the default exact-match utility. A precision-weighted Fβ with β = 0.25 raises it to 64.26, while a recall-weighted β = 4.0 lowers it to 60.70. Qualitatively, precision weighting removes hallucinated columns such as "Game Date" or "Double Double".
  • Limitations: inference cost and API latency grow roughly n-fold, where n is the number of hypotheses. Evaluation covers only the sports domain. With a weak base LLM, the method can converge on a "false consensus", i.e. an error that appears in most candidates.
  • Tools: the implementation uses the MIT-licensed mbrs toolkit, and the models are accessed through the official Google and OpenAI APIs.

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

🎓Nara Institute of Science and TechnologyIkoma · Japan48 works · 4,565 citations🏢NTTTokyo · Japan28 works · 5,199 citations

 Fields, topics and concepts

📚 Shelves 4The 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 works

Cite

Cite as Daiki Yoshida, Hiroyuki Deguchi, Yusuke Sakai, Hidetaka Kamigaito and Taro Watanabe. 2026. TableMBR: Minimum Bayes Risk Table Generation Based on Structural Consistency. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.acl-srw.95

@inproceedings{yoshida2026tablembr,
  title = {{TableMBR: Minimum Bayes Risk Table Generation Based on Structural Consistency}},
  author = {Daiki Yoshida, Hiroyuki Deguchi, Yusuke Sakai, Hidetaka Kamigaito and Taro Watanabe},
  booktitle = {Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop)},
  year = {2026},
  doi = {10.18653/v1/2026.acl-srw.95},
  url = {https://doi.org/10.18653/v1/2026.acl-srw.95}
}

References

19 references, 10 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Naman Ahuja, Fenil Bardoliya, Chitta Baral, and Vivek Gupta. 2025. Map&make: Schema guided text to table generation. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 30249– 30262, Vienna, Austria. Association for Computational… find
  2. Gheorghe Comanici, Eric Bieber, Mike Schaekermann, Ice Pasupat, Noveen Sachdeva, Inderjit Dhillon, Marcel Blistein, Ori Ram, Dan Zhang, Evan Rosen, Luke Marris, Sam Petulla, Colin Gaffney, Asaf Aharoni, Nathan Lintz, Tiago Cardal Pais, Henrik Jacobsson, Idan Szpektor, Nan-Jiang Jiang, and 3416 other…
  3. Hiroyuki Deguchi, Yusuke Sakai, Hidetaka Kamigaito, and Taro Watanabe. 2024. mbrs: A library for minimum Bayes risk decoding. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing: System Demonstrations, pages 351–362, Miami, Florida, USA. Association for Computat… find
  4. Zheye Deng, Chunkit Chan, Weiqi Wang, Yuxi Sun, Wei Fan, Tianshi Zheng, Yauwai Yim, and Yangqiu Song. 2024. Text-tuple-table: Towards information integration in text-to-table generation via global tuple extraction. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Proces… find
  5. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. BERT: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies,…
  6. Bryan Eikema and Wilker Aziz. 2020. Is MAP decoding all you need? the inadequacy of the mode in neural machine translation. In Proceedings of the 28th International Conference on Computational Linguistics, pages 4506–4520, Barcelona, Spain (Online). International Committee on Computational Linguisti…
  7. Philipp Koehn. 2004. Statistical significance tests for machine translation evaluation. In Proceedings of the 2004 Conference on Empirical Methods in Natural Language Processing, pages 388–395, Barcelona, Spain. Association for Computational Linguistics.
  8. Shankar Kumar and William Byrne. 2004. Minimum Bayes-risk decoding for statistical machine translation. In Proceedings of the Human Language Technology Conference of the North American Chapter of the Association for Computational Linguistics: HLT-NAACL 2004, pages 169–176, Boston, Massachusetts, USA…
  9. 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…
  10. Atsumoto Ohashi, Ukyo Honda, Tetsuro Morimura, and Yuu Jinnai. 2024. On the true distribution approximation of minimum Bayes-risk decoding. In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 2: Sho… find
  11. OpenAI, Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, Red Avila, Igor Babuschkin, Suchir Balaji, Valerie Balcom, Paul Baltescu, Haiming Bao, Mohammad Bavarian, Jeff Belgum, and 262 others.… — GPT-4 Technical Report
  12. Michał Pietruszka, Michał Turski, Łukasz Borchmann, Tomasz Dwojak, Gabriela Nowakowska, Karolina Szyndler, Dawid Jurkiewicz, and Łukasz Garncarek. 2024. STable: Table generation framework for encoder-decoder models. In Proceedings of the 18th Conference of the European Chapter of the Association for… find
  13. Maja Popović. 2015. chrF: character n-gram F-score for automatic MT evaluation. In Proceedings of the Tenth Workshop on Statistical Machine Translation, pages 392–395, Lisbon, Portugal. Association for Computational Linguistics.
  14. Freda Shi, Daniel Fried, Marjan Ghazvininejad, Luke Zettlemoyer, and Sida I. Wang. 2022. Natural language to code translation with execution. In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 3533–3546, Abu Dhabi, United Arab Emirates. Association for C…
  15. Anirudh S. Sundar, Christopher Richardson, and Larry Heck. 2024. gtbls: Generating tables from text by conditional question answering. ArXiv, abs/2403.14457. — [arXiv:2403.14457](https://arxiv.org/abs/2403.14457) find
  16. Sam Wiseman, Stuart Shieber, and Alexander Rush. 2017. Challenges in data-to-document generation. In Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pages 2253–2263, Copenhagen, Denmark. Association for Computational Linguistics.
  17. Xueqing Wu, Jiacheng Zhang, and Hang Li. 2022. Text-to-table: A new way of information extraction. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 2518–2533, Dublin, Ireland. Association for Computational Linguistics. find
  18. Mostafa Yadegari, Yuqiao Wen, Davood Rafiei, and Lili Mou. 2026. Exploring minimum bayes risk decoding for text-to-SQL ensemble. find
  19. Tianyi Zhang, Varsha Kishore, Felix Wu, Kilian Q. Weinberger, and Yoav Artzi. 2019. Bertscore: Evaluating text generation with bert. ArXiv, abs/1904.09675. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/yoshida-2026-tablembr-minimum-bayes-risk-table/