Elamur Research⌕
☰
home / papers / 2026 / Natural-Language Policies to Executable Decisions: An Interpretable Large Language Model Framework

Natural-Language Policies to Executable Decisions: An Interpretable Large Language Model Framework

Paper · 2026 · Ziqiang Zhang, Jing Ma, Zilong Wang et al.

Summary

The paper addresses automated price quoting for a large tourism operator. Incoming travel orders are messy, mostly chat screenshots, and pricing rules sit in policy documents that change constantly. Hand-coded rule engines need developer work for every new clause, while letting an LLM compute prices freely is too risky for financial decisions. The authors build a deployed system that splits the work. LLMs only extract structured facts, ground resources in a catalog, and pick among discrete options. A deterministic engine calculates every number. Policies are compiled into condition trees: nodes hold natural-language conditions, and each root-to-leaf path maps to exactly one price specification. Trees must pass a validator suite before publication. Operators can intervene at three override levels (L1–L3), and logged traces feed periodic fine-tuning. Over six months of operation at a city-owned public tourism company, the system made no fatal numeric errors and shrank the order team from 15–20 people to 3. Non-programmers can now publish policies themselves. The work offers a reusable pattern for safe, auditable LLM use in policy-driven domains.

Key points

  • Deployment scale: 7 scenic sites, 12 business categories, 1,500+ operators, 1,000+ active policies; 3,960 orders processed in six months (2025H2). About 97% of orders arrived as instant-messaging screenshots.
  • Pipeline: policy onboarding (A0 OCR/text extraction, A1 catalog grounding, A2 node extraction, A3 assembly and validation) feeds a price-calculation module (B0–D) with tiered human overrides (L1 path, L2 tree, L3 resource/bundle edits).
  • Workload on 3,842 screenshot orders: 47.9% needed L3-a resource edits and 2.8% needed L3-b bundle switching. Of the remaining orders, only 5.7% required L2 tree switching.
  • Correctness on 1,349 information-consistent orders (2,598 product instances): 0% fatal numeric errors, 100% unit correctness, and 85.2% unit-price accuracy. The remaining errors are L1 path choices that can be fixed with one click.
  • Operational gains: per-order handling dropped from ~10 to under 2 minutes, and policy onboarding from ~20 to ~1 minute without developers. Limitations include a single deployment site and no offline benchmark for tree induction.

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

🎓Eastern Institute of Technology, NingboNingbo · China4 works · 23 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

🧭 Topics 1Research subjects and languages the work is about.Language models & pretraining · 3,894 works

Cite

Cite as Ziqiang Zhang, Jing Ma, Zilong Wang, Jiayuan Chen, Yi Qiao, Yu He, Wei Zhang, Dai Cheng and Xiaoyu Shen. 2026. Natural-Language Policies to Executable Decisions: An Interpretable Large Language Model Framework. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 6: Industry Track).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.acl-industry.114

@inproceedings{zhang2026natural,
  title = {{Natural-Language Policies to Executable Decisions: An Interpretable Large Language Model Framework}},
  author = {Ziqiang Zhang, Jing Ma, Zilong Wang, Jiayuan Chen, Yi Qiao, Yu He, Wei Zhang, Dai Cheng and Xiaoyu Shen},
  booktitle = {Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 6: Industry Track)},
  year = {2026},
  doi = {10.18653/v1/2026.acl-industry.114},
  url = {https://doi.org/10.18653/v1/2026.acl-industry.114}
}

References

14 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, and 1 others. 2023. Gpt-4 technical report. arXiv preprint arXiv:2303.08774.
  2. Chirag Agarwal, Sree Harsha Tanneru, and Himabindu Lakkaraju. 2024. Faithfulness vs. plausibility: On the (un)reliability of explanations from large language models. ArXiv, abs/2402.04614. — [arXiv:2402.04614](https://arxiv.org/abs/2402.04614) find
  3. Wenhu Chen, Xueguang Ma, Xinyi Wang, and William W. Cohen. 2022. Program of thoughts prompting: Disentangling computation from reasoning for numerical reasoning tasks. find
  4. Trans. Mach. Learn. Res., 2023. find
  5. Michael Desmond, Evelyn Duesterwald, Vatche Isahagian, and Vinod Muthusamy. 2022. A no-code low-code paradigm for authoring business automations using natural language. ArXiv, abs/2207.10648. — [arXiv:2207.10648](https://arxiv.org/abs/2207.10648) find
  6. Longwei Ding, Anhao Zhao, Fanghua Ye, Ziyang Chen, and Xiaoyu Shen. 2026. From llms to lrms: Rethinking pruning for reasoning-centric models. arXiv preprint arXiv:2601.18091. — [arXiv:2601.18091](https://arxiv.org/abs/2601.18091) find
  7. Luyu Gao, Aman Madaan, Shuyan Zhou, Uri Alon, Pengfei Liu, Yiming Yang, Jamie Callan, and Graham Neubig. 2022. Pal: Program-aided language models. ArXiv, abs/2211.10435. find
  8. Dan Hendrycks, Collin Burns, Anya Chen, and Spencer Ball. 2021. Cuad: An expert-annotated nlp dataset for legal contract review. ArXiv, abs/2103.06268.
  9. Divyansh Kaushik, Shashank Gupta, Chakradhar Raju, R. A. Dias, and Sanjib Ghosh. 2017. Making travel smarter: Extracting travel information from email itineraries using named entity recognition. In Recent Advances in Natural Language Processing. find
  10. Aixin Liu, Bei Feng, Bing Xue, Bingxuan Wang, Bochao Wu, Chengda Lu, Chenggang Zhao, Chengqi Deng, Chenyu Zhang, Chong Ruan, and 1 others. 2024. Deepseek-v3 technical report. arXiv preprint arXiv:2412.19437. find
  11. Sichao Xiong, Yigit Ihlamur, Fuat Alican, and Aaron Ontoyin Yin. 2024. Gptree: Towards explainable decision-making via llm-powered decision trees. arXiv preprint arXiv:2411.08257. — [arXiv:2411.08257](https://arxiv.org/abs/2411.08257) find
  12. Fengli Xu, Qianyue Hao, Zefang Zong, Jingwei Wang, Yunke Zhang, Jingyi Wang, Xiaochong Lan, Jiahui Gong, Tianjian Ouyang, Fanjin Meng, and 1 others. 2025. Towards large reasoning models: A survey of reinforced reasoning with large language models. arXiv preprint arXiv:2501.09686.
  13. Yiheng Xu, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, and Ming Zhou. 2019. Layoutlm: Pre-training of text and layout for document image understanding. Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining.
  14. Ruiwen Zhou, Wenyue Hua, Liangming Pan, Sitao Cheng, Xiaobao Wu, En Yu, and William Yang Wang. 2025. Rulearena: A benchmark for rule-guided reasoning with llms in real-world scenarios. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers),…

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/zhang-2026-natural-language-policies-executable-decisions/