Elamur Research⌕
☰
home / papers / 2026 / iTAG: Inverse Design for Natural Text Generation with Accurate Causal Graph Annotations

iTAG: Inverse Design for Natural Text Generation with Accurate Causal Graph Annotations

Paper · 2026 · Wenshuo Wang, Boyu Cao, Nan Zhuang et al.

Summary

Testing algorithms that infer causal structure from text requires corpora whose causal graphs are known, but manual causal annotation is expensive. Existing generators face a trade-off. Template-based methods produce accurate but stilted text. LLM-based methods turn graphs directly into fluent prose but may drop edges or invent new ones. iTAG addresses this by adding a concept-design step before text generation, using a three-phase pipeline:

  1. A parameterized sampler builds a DAG and its adjacency matrix.
  2. An LLM maps abstract nodes to real-world concepts. This is treated as an inverse problem: Chain-of-Thought reasoning, counterfactual checks and fallacy analysis repeatedly revise the assignment until the implied relations fit both the edges and the non-edges of the target graph.
  3. The concepts and their structure are rendered as natural text.

Experiments on business, medical and legal English texts show very high annotation accuracy. Human judges and trained classifiers struggle to tell the output from real text. Causal discovery algorithms perform on iTAG corpora much as they do on matched real corpora. iTAG is therefore a low-cost benchmarking aid, not a replacement for real-data evaluation.

Key points

  • Graph construction (Phase 1): an enhanced Erdős–Rényi DAG generator with several controls:
    • node count, expected density and parent/child degree caps;
    • ratios of confounders and colliders;
    • number of mediator chains.
  • Concept assignment loop (Algorithm 1): each round runs counterfactual verification, fallacy analysis and mismatch scoring. The loop stops once no fallacies remain; otherwise it returns the best assignment found within a fixed iteration budget.
  • Agreement with real data: across 24 algorithm–size pairs, after within-n centering, Pearson r and Spearman ρ are both ≥ 0.921 (p < 0.001) and R² ≥ 0.848.
  • Small gaps from real corpora: |ΔF1_G| ≤ 0.036, |ΔSHD| ≤ 1.33 and |ΔSID| ≤ 8.14. Algorithm rankings are preserved, with RuleBayes performing worst.
  • Limitations:
    • annotations cover adjacency-level structure only, with no SEM effect parameters;
    • tests use small graphs (3–10 variables) and English text only;
    • judging that a non-edge truly has no direct relation remains harder than confirming an edge.

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

🏛️Chinese Academy of SciencesBeijing · China194 works · 23,906 citations🔬Institute of Computing TechnologyBeijing · China51 works · 3,411 citations🎓South China University of TechnologyGuangzhou · China22 works · 1,554 citations🏢China MobileBeijing · China9 works · 21 citations

 Fields, topics and concepts

📚 Shelves 4The 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 works

🧭 Topics 1Research subjects and languages the work is about.Text generation · 202 works

Cite

Cite as Wenshuo Wang, Boyu Cao, Nan Zhuang and Wei Li. 2026. iTAG: Inverse Design for Natural Text Generation with Accurate Causal Graph Annotations. 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. DOI 10.18653/v1/2026.acl-long.364

@inproceedings{wang2026itag,
  title = {{iTAG: Inverse Design for Natural Text Generation with Accurate Causal Graph Annotations}},
  author = {Wenshuo Wang, Boyu Cao, Nan Zhuang and Wei Li},
  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.364},
  url = {https://doi.org/10.18653/v1/2026.acl-long.364}
}

References

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

  1. Mohammad Alali, Shaayan Syed, Mohammed Alsayed, Smit Patel, and Hemanth Bodala. 2021. Justice: A benchmark dataset for supreme court’s judgment prediction. arXiv preprint arXiv:2112.03414. find
  2. Alessandro Antonucci, Gregorio Piqué, and Marco Zaffalon. 2023. Zero-shot causal graph extrapolation from text via llms. arXiv preprint arXiv:2312.14670. — [arXiv:2312.14670](https://arxiv.org/abs/2312.14670) find
  3. Nabiha Asghar. 2016. Automatic extraction of causal relations from natural language texts: a comprehensive survey. arXiv preprint arXiv:1605.07895. — [arXiv:1605.07895](https://arxiv.org/abs/1605.07895) find
  4. Philippe Brouillard, Chandler Squires, Jonas Wahl, Konrad P Kording, Karen Sachs, Alexandre Drouin, and Dhanya Sridhar. 2024. The landscape of causal discovery data: Grounding causal discovery in real-world applications. arXiv preprint arXiv:2412.01953. — [arXiv:2412.01953](https://arxiv.org/abs/2412.01953) find
  5. Xiaohan Ding, Kaike Ping, Buse Çarık, and Eugenia Rho. 2025. A multi-level benchmark for causal language understanding in social media discourse. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 28764–28778. find
  6. Paul Erdős and Alfréd Rényi. 1959. On random graphs i. Publ. math. debrecen, 6(290-297):18. find
  7. Paul Erdős and Alfréd Rényi. 1960. On the evolution of random graphs. Publ. Math. Inst. Hungar. Acad. Sci, 5:17–61. find
  8. Philipp M Faller, Leena C Vankadara, Atalanti A Mas-takouri, Francesco Locatello, and Dominik Janzing. 2024. Self-compatibility: Evaluating causal discovery without ground truth. In International Conference on Artificial Intelligence and Statistics, pages 4132–4140. PMLR. find
  9. Tyler Gandee and Philippe Giabbanelli. 2024. Combining Natural Language Generation and Graph Algorithms to Explain Causal Maps Through Meaningful Paragraphs, pages 359–376. find
  10. Tyler J Gandee and Philippe J Giabbanelli. 2025. Faithful narratives from complex conceptual models: Should modelers or large language models simplify causal maps? Machine Learning and Knowledge Extraction, 7(4):116. find
  11. A. Goldberger, L. Amaral, L. Glass, J. Hausdorff, P. C. Ivanov, R. Mark, and H. E. Stanley. 2000. PhysioBank, PhysioToolkit, and PhysioNet: Components of a new research resource for complex physiologic signals. Circulation [Online], 101(23):e215–e220. RRID:SCR_007345. find
  12. Pranav Gujarathi, Manohar Reddy, Neha Tayade, and Sunandan Chakraborty. 2022. A study of extracting causal relationships from text. In Proceedings of SAI Intelligent Systems Conference, pages 807–828. Springer. find
  13. Sujai Hiremath, Kyra Gan, and Promit Ghosal. 2024. Losam: Local search in additive noise models with unmeasured confounders, a top-down global discovery approach. arXiv e-prints, pages arXiv–2410. find
  14. A. Johnson, T. Pollard, S. Horng, L. A. Celi, and R. Mark. 2023. MIMIC-IV-Note: Deidentified free-text clinical notes (version 2.2). RRID:SCR_007345. find
  15. Armand Joulin, Edouard Grave, Piotr Bojanowski, and Tomáš Mikolov. 2017. Bag of tricks for efficient text classification. In Proceedings of the 15th conference of the European chapter of the association for computational linguistics: volume 2, short papers, pages 427–431. find
  16. Byungsoo Kim, Vinicius C Azevedo, Nils Thuerey, Theodore Kim, Markus Gross, and Barbara Solen-thaler. 2019. Deep fluids: A generative network for parameterized fluid simulations. In Computer graphics forum, volume 38, pages 59–70. Wiley Online Library. find
  17. Yoon Kim. 2014. Convolutional neural networks for sentence classification. arXiv preprint arXiv:1408.5882.
  18. Zhaoning Li, Qi Li, Xiaotian Zou, and Jiangtao Ren. 2021. Causality extraction based on self-attentive bilstm-crf with transferred embeddings. Neurocomputing, 423:207–219. find
  19. Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, and Veselin Stoyanov. 2019. Roberta: A robustly optimized bert pretraining approach. arXiv preprint arXiv:1907.11692.
  20. Antonio Moreno-Sandoval, Jordi Porta Zamorano, Blanca Carbajo-Coronado, Yanco Torterolo, and Doaa Samy. 2025. The financial document causality detection shared task (fincausal 2025). In Proceedings of the Joint Workshop of the 9th Financial Technology and Natural Language Processing (FinNLP), the 6t… find
  21. 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 in large language models. Preprint, arXiv:2201.11903.
  22. Jie Yang, Soyeon Caren Han, and Josiah Poon. 2022. A survey on extraction of causal relations from natural language text. Knowledge and Information Systems, 64(5):1161–1186. find
  23. Zhongliang Yang, Nan Wei, Junyi Sheng, Yongfeng Huang, and Yu-Jin Zhang. 2018. Ts-cnn: Text steganalysis from semantic space based on convolutional neural network. arXiv preprint arXiv:1810.08136. — [arXiv:1810.08136](https://arxiv.org/abs/1810.08136) find
  24. Shunyu Yao, Dian Yu, Jeffrey Zhao, Izhak Shafran, Tom Griffiths, Yuan Cao, and Karthik Narasimhan. 2023. Tree of thoughts: Deliberate problem solving with large language models. Advances in neural information processing systems, 36:11809–11822. find
  25. Denny Zhou, Nathanael Schärli, Le Hou, Jason Wei, Nathan Scales, Xuezhi Wang, Dale Schuurmans, Claire Cui, Olivier Bousquet, Quoc Le, and Ed Chi. 2023. Least-to-most prompting enables complex reasoning in large language models. Preprint, arXiv:2205.10625.
  26. Tobias Pfaff, Meire Fortunato, Alvaro Sanchez-Gonzalez, and Peter Battaglia. 2020. Learning mesh-based simulation with graph networks. In International conference on learning representations. find
  27. Atharva Phatak, Vijay K Mago, Ameeta Agrawal, Aravind Inbasekaran, and Philippe J Giabbanelli. 2024. Narrating causal graphs with large language models. arXiv preprint arXiv:2403.07118. — [arXiv:2403.07118](https://arxiv.org/abs/2403.07118) find
  28. Anish Shrestha, Kyle Mielke, Tuong Anh Nguyen, and Philippe J Giabbanelli. 2022. Automatically explaining a model: Using deep neural networks to generate text from causal maps. In 2022 Winter simulation conference (WSC), pages 2629–2640. IEEE. find
  29. Antonio Sorgente, Giuseppe Vettigli, and Francesco Mele. 2013. Automatic extraction of cause-effect relations in natural language text. DART@ AI* IA, 2013:37–48. find
  30. Andrew M Stuart. 2010. Inverse problems: a bayesian perspective. Acta numerica, 19:451–559. find
  31. Meiyun Wang, Kiyoshi Izumi, and Hiroki Sakaji. 2025. Causalenhance: Knowledge-enhanced pre-training for causality identification and extraction. Knowledge-Based Systems, page 114447. find
  32. Xuezhi Wang, Jason Wei, Dale Schuurmans, Quoc Le, Ed Chi, Sharan Narang, Aakanksha Chowdhery, and Denny Zhou. 2022. Self-consistency improves chain of thought reasoning in language models. arXiv preprint arXiv:2203.11171.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/wang-2026-itag-inverse-design-natural-text/