Elamur Research⌕
☰
home / papers / 2026 / Plan-Guided Text Simplification with Extended Contexts

Plan-Guided Text Simplification with Extended Contexts

Summary

Plan-guided simplification rewrites a document according to a plan. The plan gives a target reading level and one operation per sentence: copy, rephrase, split, or delete. This paper tests how well such systems follow their plans when they receive more text at once.

From Newsela-auto, the authors build datasets whose inputs span one to five paragraphs. On these they retrain the fine-tuned BART setup of Cripwell et al. (2023b) in three variants: with oracle plans, with plans from an existing planning model, and with no plans. Outputs are mapped back to the source sentences with a neural CRF aligner. Plan adherence is scored as micro-recall, and quality is measured with SARI, SMART and FKGL.

BART follows its plans much less reliably as inputs grow longer. As an alternative, the authors prompt instruction-tuned Qwen models with few-shot examples and numbered, operation-labelled sentences. These models stay robust at every length. Even at five paragraphs they beat BART's best single-paragraph score, and they produce better simplifications. Ablations show the approach needs a minimum model size and also works with another LLM family. This matters because document-level simplification depends on wider context.

Key points

  • Data: Newsela-auto contains 1,882 news articles, each rewritten at five reading levels. Pairing every version with every simpler version gives 18,820 document pairs, split 92.5/2.5/5.0% into train, validation and test.
  • Operation labels: labels come from sentence alignments, with copy assigned at a Levenshtein similarity of at least 0.92. The resulting distribution is roughly 20% copy, 27% rephrase, 17% split and 36% delete.
  • BART degradation:
    • With oracle plans (O-BART), micro-recall falls from 70.8% at one paragraph to 44.7% at five, a drop of 26.1 points.
    • With predicted plans (Ô-BART, from liamcripwell/pgdyn-plan), it falls from 58.5% to 41.0%.
    • Without plans, BART stays nearly flat, moving from 38.7% to 33.8%.
  • LLM alternative: prompted Qwen models (7B, 14B and 32B), served with vLLM, keep plan-following stable. At every context length they score above O-BART's single-paragraph result.
  • Resources: all code, prompts and settings are released at github.com/pascalmathas/plan_simp_extended. Fine-tuning runs use a fixed seed of 42 with deterministic Torch and CUDA behaviour.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

Fields: Machine translation · Natural language processing

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.63317/2gk3wxp2d525

@inproceedings{mathas2026plan,
  title = {{Plan-Guided Text Simplification with Extended Contexts}},
  author = {Mathas, Pascal and Bakker, Jan and Kamps, Jaap},
  booktitle = {Proceedings of the Joint Workshop on Readability and Text Simplification (READIxTSAR) @ LREC 2026},
  year = {2026},
  doi = {10.63317/2gk3wxp2d525},
  url = {https://doi.org/10.63317/2gk3wxp2d525}
}

References

15 references, 4 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Fernando Alva-Manchego, Carolina Scarton, and Lucia Specia. 2019. Cross-sentence transformations in text simplification. In Proceedings of the 2019 Workshop on Widening NLP, pages 181–184. find
  2. Reinald Kim Amplayo, Peter J. Liu, Yao Zhao, and Shashi Narayan. 2023. SMART: sentences as basic units for text evaluation. In The Eleventh International Conference on Learning Representations, ICLR 2023, Kigali, Rwanda, May 1-5, 2023. OpenReview.net. ↗
  3. Jan Bakker and Jaap Kamps. 2024. Beyond sentence-level text simplification: Reproducibility study of context-aware document simplification. find
  4. In Proceedings of the Workshop on DeTermIt! Evaluating Text Difficulty in a Multilingual Context@ LREC-COLING 2024, pages 27–38. find
  5. Liam Cripwell, Joël Legrand, and Claire Gardent. 2023a. Context-aware document simplification. In Findings of the Association for Computational Linguistics: ACL 2023, pages 13190–13206. find
  6. Liam Cripwell, Joël Legrand, and Claire Gardent. 2023b. Document-level planning for text simplification. In Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics, pages 993–1006. find
  7. Chao Jiang, Mounica Maddela, Wuwei Lan, Yang Zhong, and Wei Xu. 2020. Neural crf model for sentence alignment in text simplification. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 7943–7960. find our card
  8. J Peter Kincaid, Robert P Fishburne Jr, Richard L Rogers, and Brad S Chissom. 1975. Derivation of new readability formulas (automated readability index, fog count and flesch reading ease formula) for navy enlisted personnel. Technical report. ↗
  9. Woosuk Kwon, Zhuohan Li, Siyuan Zhuang, Ying Sheng, Lianmin Zheng, Cody Hao Yu, Joseph E. Gonzalez, Hao Zhang, and Ion Stoica. 2023. Efficient memory management for large language model serving with pagedattention. In Proceedings of the ACM SIGOPS 29th Symposium on Operating Systems Principles. find
  10. Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Veselin Stoyanov, and Luke Zettlemoyer. 2019. BART: denoising sequence-to-sequence pre-training for natural language generation, translation, and comprehension. CoRR, abs/1910.13461. ↗ our card
  11. Taiki Papandreou, Jan Bakker, and Jaap Kamps. 2025. Medical text simplification from jargon detection to jargon-aware prompting. In Proceedings of the Fourth Workshop on Text Simplification, Accessibility and Readability (TSAR 2025), pages 36–46. find
  12. Qwen Team. 2024. Qwen2.5: A party of foundation models. find
  13. Wei Xu, Chris Callison-Burch, and Courtney Napoles. 2015. Problems in current text simplification research: New data can help. Transactions of the Association for Computational Linguistics, 3:283–297. ↗ our card
  14. Wei Xu, Courtney Napoles, Ellie Pavlick, Quanze Chen, and Chris Callison-Burch. 2016. Optimizing statistical machine translation for text simplification. Transactions of the Association for Computational Linguistics, 4:401–415. ↗ our card
  15. An Yang, Baosong Yang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Zhou, Chengpeng Li, Chengyuan Li, Dayiheng Liu, Fei Huang, Guanting Dong, Haoran Wei, Huan Lin, Jialong Tang, Jialin Wang, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Ma, Jin Xu, Jingren Zhou, Jinze Bai, Jinzheng He, Junyang Lin, Kai Dang, Keming Lu, Keqin Chen, Kexin Yang, Mei Li, Mingfeng Xue, Na Ni, Pei Zhang, Peng Wang, Ru Peng, Rui Men, Ruize Gao, Runji Lin, Shijie Wang, Shuai Bai, Sinan Tan, Tianhang Zhu, Tianhao Li, Tianyu Liu, Wenbin Ge, Xiaodong Deng, Xiaohuan Zhou, Xingzhang Ren, Xinyu Zhang, Xipin Wei, Xuancheng Ren, Yang Fan, Yang Yao, Yichang Zhang, Yu Wan, Yunfei Chu, Yuqiong Liu, Zeyu Cui, Zhenru Zhang, and Zhihao Fan. 2024. Qwen2 technical report. arXiv preprint arXiv:2407.10671. ↗

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/mathas-2026-plan-guided-text-simplification-extended/