Elamur Research⌕
☰
home / papers / 2026 / Exploring Generative Process Reward Modeling for Semi-Structured Data: A Case Study of Table Question Answering

Exploring Generative Process Reward Modeling for Semi-Structured Data: A Case Study of Table Question Answering

Paper · 2026 · Lei Tang, Wei Zhou and Mohsen Mesgar

Summary

Process reward models (PRMs) score each step of a candidate solution and then pick the best answer. They have worked well for math, but nobody had tested them on reasoning over semi-structured data. This short paper is the first systematic look at PRMs for table question answering (TQA). Tables add specific difficulties: much of their content is irrelevant to the question, reasoning steps are only loosely linked, and the reasoning patterns are domain-specific.

The authors build PRM training data from WTQ and TabFact. They sample reasoning paths with Qwen2.5-7B-Instruct and generate verification rationales with QwQ-32B. They then fine-tune DeepSeek-R1-Distill-Qwen-7B as several generative verifiers: GenPRM, Textual PRM, MDA and RRA. These are compared with majority voting and LLM-as-a-judge using best-of-N answer selection and a manually annotated set of steps.

GenPRM, which pairs textual checks with executed code, performs best among the PRMs, but it generalizes poorly to out-of-domain data. Step-level accuracy also correlates only weakly with final answer correctness. The authors trace this to uniform averaging of step scores and argue for aggregation that accounts for each step's relevance and causal role.

Key points

  • Evaluation data: In-domain tests use WTQ and TabFact; out-of-domain tests use CRT and SciTab. For process-level evaluation, 300 instances were manually annotated, giving 1916 steps (1366 correct, 550 incorrect).
  • Training setup: The policy model samples 8 paths per question at temperature 0.6. GenPRM labels come from relative progress estimation plus QwQ-32B rationales, and instances where the two disagree are discarded.
  • Answer-level results: GenPRM (7B and 32B) clearly beats Majority@N. The 7B version trails the 32B LLM-as-a-judge in answer selection, yet it matches or exceeds that judge in step-level verification on most datasets, except CRT. All methods remain well below the Oracle.
  • Better aggregation: Giving more weight to the final steps with a 32B code-based verifier improves EM by +3.56% on WTQ, +1.14% on TabFact, +4.55% on CRT and +2.69% on SciTab.
  • Code: Released at github.com/boschresearch/TQA-PRM.

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

🎓University of StuttgartStuttgart · Germany91 works · 46,376 citations🏢Robert BoschStuttgart · Germany11 works · 2,306 citations🎓University of AugsburgAugsburg · Germany8 works · 38,342 citations

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksComputer Vision and Pattern Recognition · 2,869 works

🧭 Topics 2Research subjects and languages the work is about.Reasoning · 1,426 worksQuestion answering · 441 works

Cite

Cite as Lei Tang, Wei Zhou and Mohsen Mesgar. 2026. Exploring Generative Process Reward Modeling for Semi-Structured Data: A Case Study of Table Question Answering. Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics (Volume 2: Short Papers).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.eacl-short.30

@inproceedings{tang2026exploring,
  title = {{Exploring Generative Process Reward Modeling for Semi-Structured Data: A Case Study of Table Question Answering}},
  author = {Lei Tang, Wei Zhou and Mohsen Mesgar},
  booktitle = {Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics (Volume 2: Short Papers)},
  year = {2026},
  doi = {10.18653/v1/2026.eacl-short.30},
  url = {https://doi.org/10.18653/v1/2026.eacl-short.30}
}

References

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

  1. Lin, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Yang, Jiaxi Yang, Jingren Zhou, and 25 others. 2025. Qwen2.5 technical report. Preprint, arXiv:2412.15115. find
  2. Wenhu Chen, Hongmin Wang, Jianshu Chen, Yunkai Zhang, Hong Wang, Shiyang Li, Xiyou Zhou, and William Yang Wang. 2020. Tabfact: A large-scale dataset for table-based fact verification. Preprint, arXiv:1909.02164.
  3. DeepSeek-AI, Daya Guo, Dejian Yang, Haowei Zhang, Jun-Mei Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shirong Ma, Peiyi Wang, Xiaoling Bi, Xiaokang Zhang, Xingkai Yu, Yu Wu, Z. F. Wu, Zhibin Gou, Zhihong Shao, Zhuoshu Li, Ziyi Gao, and 179 others. 2025. Deepseek-r1: Incentivizing reasoning capability i…
  4. Naihao Deng and Rada Mihalcea. 2025. Rethinking table instruction tuning. In Findings of the Association for Computational Linguistics: ACL 2025, pages 21757–21780, Vienna, Austria. Association for Computational Linguistics. find
  5. Yihe Deng, Weitong Zhang, Zixiang Chen, and Quanquan Gu. 2023. Rephrase and respond: Let large language models ask better questions for themselves. Preprint, arXiv:2311.04205. find
  6. Jiawei Gu, Xuhui Jiang, Zhichao Shi, Hexiang Tan, Xuehao Zhai, Chengjin Xu, Wei Li, Yinghan Shen, Shengjie Ma, Honghao Liu, Yuanzhuo Wang, and Jian Guo. 2024. A survey on llm-as-a-judge. ArXiv, abs/2411.15594. find
  7. Tianyang Liu, Fei Wang, and Muhao Chen. 2024. Rethinking tabular data understanding with large language models. In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers), pages 450–482, Me… find
  8. Shuaijie She, Junxiao Liu, Yifeng Liu, Jiajun Chen, Xin Huang, and Shujian Huang. 2025. R-prm: Reasoning-driven process reward modeling. Preprint, arXiv:2503.21295.
  9. Yi Su, Dian Yu, Linfeng Song, Juntao Li, Haitao Mi, Zhaopeng Tu, Min Zhang, and Dong Yu. 2025. Crossing the reward bridge: Expanding rl with verifiable rewards across diverse domains. ArXiv, abs/2503.23829. — [arXiv:2503.23829](https://arxiv.org/abs/2503.23829) find
  10. Qwen Team. 2025. Qwq-32b: Embracing the power of reinforcement learning. find
  11. Peiyi Wang, Lei Li, Zhihong Shao, R. X. Xu, Damai Dai, Yifei Li, Deli Chen, Y. Wu, and Zhifang Sui. 2024a. Math-shepherd: Verify and reinforce llms step-by-step without human annotations. Preprint, arXiv:2312.08935.
  12. Peiyi Wang, Lei Li, Zhihong Shao, Runxin Xu, Damai Dai, Yifei Li, Deli Chen, Yu Wu, and Zhifang Sui. 2024b. Math-shepherd: Verify and reinforce LLMs step-by-step without human annotations. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Pape…
  13. Weiyun Wang, Zhangwei Gao, Lianjie Chen, Zhe Chen, Jinguo Zhu, Xiangyu Zhao, Yangzhou Liu, Yue Cao, Shenglong Ye, Xizhou Zhu, Lewei Lu, Haodong Duan, Yu Qiao, Jifeng Dai, and Wenhai Wang. 2025. Visualprm: An effective process reward model for multimodal reasoning. ArXiv, abs/2503.10291. — [arXiv:2503.10291](https://arxiv.org/abs/2503.10291) find
  14. Xinyuan Lu, Liangming Pan, Qian Liu, Preslav Nakov, and Min-Yen Kan. 2023. SCITAB: A challenging benchmark for compositional reasoning and claim verification on scientific tables. find
  15. Xuezhi Wang, Jason Wei, Dale Schuurmans, Quoc Le, Ed H. Chi, and Denny Zhou. 2022. Self-consistency improves chain of thought reasoning in language models. ArXiv, abs/2203.11171.
  16. Brandon Ong, Tej Deep Pala, Vernon Toh, William-Chandra Tjhi, and Soujanya Poria. 2025. Training vision-language process reward models for test-time scaling in multimodal reasoning: Key insights and lessons learned. find
  17. Zilong Wang, Hao Zhang, Chun-Liang Li, Julian Martin Eisenschlos, Vincent Perot, Zifeng Wang, Lesly Miculicich, Yasuhisa Fujii, Jingbo Shang, Chen-Yu Lee, and Tomas Pfister. 2024c. Chain-of-table: Evolving tables in the reasoning chain for table understanding. Preprint, arXiv:2401.04398. find
  18. Panupong Pasupat and Percy Liang. 2015. Compositional semantic parsing on semi-structured tables. In Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing (Volume 1: Long Papers), pages 1470–…
  19. Qwen, :, An Yang, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chengyuan Li, Dayiheng Liu, Fei Huang, Haoran Wei, Huan Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Brian Ichter, Fei Xia, Ed Chi, Quoc Le, and Denny Zhou. 2023. Chain-of-thought prompting elicits reasoning i…
  20. Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. 2022. React: Synergizing reasoning and acting in language models. ArXiv, abs/2210.03629.
  21. Zhangyue Yin, Qiushi Sun, Zhiyuan Zeng, Qinyuan Cheng, Xipeng Qiu, and Xuanjing Huang. 2025. Dynamic and generalizable process reward modeling. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 4203–4233, Vienna, Austria. Associ… find
  22. Shimao Zhang, Xiao Liu, Xin Zhang, Junxiao Liu, Zheheng Luo, Shujian Huang, and Yeyun Gong. 2025a. Process-based self-rewarding language models. In Findings of the Association for Computational Linguistics: ACL 2025, pages 18097–18110, Vienna, Austria. Association for Computational Linguistics. find
  23. Tianshu Zhang, Xiang Yue, Yifei Li, and Huan Sun. 2024. TableLlama: Towards open large generalist models for tables. In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers), pages 6024–6… find
  24. Zhehao Zhang, Xitao Li, Yan Gao, and Jian-Guang Lou. 2023. CRT-QA: A dataset of complex reasoning question answering over tabular data. find
  25. Zhenru Zhang, Chujie Zheng, Yangzhen Wu, Beichen Zhang, Runji Lin, Bowen Yu, Dayiheng Liu, Jingren Zhou, and Junyang Lin. 2025b. The lessons of developing process reward models in mathematical reasoning. Preprint, arXiv:2501.07301.
  26. Jian Zhao, Runze Liu, Kaiyan Zhang, Zhimu Zhou, Junqi Gao, Dong Li, Jiafei Lyu, Zhouyi Qian, Biqing Qi, Xiu Li, and Bowen Zhou. 2025. Genprm: Scaling test-time compute of process reward models via generative reasoning. Preprint, arXiv:2504.00891. — [arXiv:2504.00891](https://arxiv.org/abs/2504.00891) find
  27. Wei Zhou, Mohsen Mesgar, Annemarie Friedrich, and Heike Adel. 2025. Efficient multi-agent collaboration with tool use for online planning in complex table question answering. Preprint, arXiv:2412.20145. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/tang-2026-exploring-generative-process-reward-modeling/