Children's English Reading Story Generation via Supervised Fine-Tuning of Compact LLMs with Controllable Difficulty and Safety

What it is about

Story generators for early readers often break a curriculum's constraints. The authors start from a K–2 phonics curriculum of 129 lessons, each listing its target phonemes, and 2,580 stories (ten per lesson) that GPT-4o and Llama 3.3 70B wrote zero-shot. They fine-tune three 8B models (Llama 3, Apertus, Granite 3.3) with QLoRA in four ways: plain SFT; SFT on 996 "good" stories selected by five metrics; reward-weighted SFT; and SFT with simulated child reading errors. Judged on Spache readability, perplexity, coherence, syntactic complexity and toxicity, the fine-tuned outputs are reported to beat the zero-shot 70B and GPT-4o outputs on the difficulty metrics (Welch's t-test, p < 0.001). The authors list their own limits: the reward uses the same metrics as the evaluation, and the evaluation uses the training lessons.

Ideas

• I1. A hard lexical gate: each lesson lists its allowed graphemes; a deterministic check with en/src/tok.rs and en/src/morph.rs rejects generated words outside the set. The paper's grapheme-to-phoneme tools failed on curriculum notation, so a rule table written from the lesson is the first thing to test.
• I2. Keep the paper's selection rule as a plain filter: keep the stories whose readability is no worse than the corpus mean on every metric, and report how many survive.
• I3. Evaluate on lessons held out from building the story set; the paper does not do this, and the Mova release test must stay untouched.
• I4. Measure diversity as Self-BLEU within a lesson, which is a deterministic count, on the output of world/src/story.rs.

For Mova

Chapters: 06. Spelling rules: inflection, derivation, compounds and national standards (letter-to-sound), 10. Word classes, 19. The lexicon: vocabulary, multiword expressions and register. Code: en/src/tok.rs, en/src/morph.rs, world/src/story.rs. Measure: share of generated stories that pass the grapheme gate, and the difficulty measures on held-out lessons; two runs must be identical.

Cite

The paper itself, not this card. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.bea-1.51

@inproceedings{shen2026children,
  title = {{Children's English Reading Story Generation via Supervised Fine-Tuning of Compact LLMs with Controllable Difficulty and Safety}},
  author = {Shen, Qian and Cao, Fanghua and Yao, Min and Gilda, Shlok and Dorr, Bonnie and Leite, Walter},
  booktitle = {Proceedings of the 21st Workshop on Innovative Use of NLP for Building Educational Applications (BEA 2026)},
  year = {2026},
  doi = {10.18653/v1/2026.bea-1.51},
  url = {https://doi.org/10.18653/v1/2026.bea-1.51}
}

References

41 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Sandhini Agarwal, Lama Ahmad, Jason Ai, Sam Altman, Andy Applebaum, Edwin Arbus, Rahul K Arora, Yu Bai, Bowen Baker, Haiming Bao, et al. 2025. gpt-oss-120b & gpt-oss-20b model card. arXiv preprint arXiv:2508.10925. ↗
2. Salama Saif Hamed Al-Siyabi. 2024. Using chatgpt to enrich children literature and enhance their vocabulary repertoire. International Journal of Research and Studies Publishing, 5(58):278–296. find
3. Suha K. Assayed, Manar Alkhatib, and Khaled Shaalan. 2024. A transformer-based generative ai model in education: Fine-tuning bert for domain-specific in student advising. In Azza Basiouni and Claude Frasson, editors, Breaking Barriers with Generative Intelligence. Using GI to Improve Human Education and Well-Being, volume 2162 of Communications in Computer and Information Science, pages 165–174. Springer. ↗
4. Yuntao Bai, Andy Jones, Kamal Ndousse, Amanda Askell, Anna Chen, Nova DasSarma, Dawn Drain, Stanislav Fort, Deep Ganguli, Tom Henighan, et al. 2022. Training a helpful and harmless assistant with reinforcement learning from human feedback. arXiv preprint arXiv:2204.05862. ↗
5. Lichang Chen, Shiyang Li, Jun Yan, Hai Wang, Kalpa Gunaratna, Vikas Yadav, Zheng Tang, Vijay Srinivasan, Tianyi Zhou, Heng Huang, et al. 2024. Alpagasus: Training a better alpaca with fewer data. In The Twelfth International Conference on Learning Representations. of NLP for Building Educational Applications (BEA 2024), pages 299–308. find
6. Yong Chen, Hongpeng Chen, and Songzhi Su. 2023. Fine-tuning large language models in education. In Proceedings of the 2023 13th International Conference on Information Technology in Medicine and Education (ITME), pages 718–723. IEEE. ↗
7. Ariel Han and Zhenyao Cai. 2023. Design implications of generative ai systems for visual storytelling for young learners. In Proceedings of the 22nd annual ACM interaction design and children conference, pages 470–474. ↗
8. Sankalan Pal Chowdhury, Nico Daheim, Ekaterina Kochmar, Jakub Macina, Donya Rooein, Mrinmaya Sachan, and Shashank Sonkar. 2025. Large language models for education: Understanding the needs of stakeholders, current capabilities and the path forward. Proceedings of the 20th Workshop on Innovative Use of NLP for Building Educational Applications (BEA 2025). ↗
9. Laura Hanu and Unitary team. 2020. Detoxify. Github. https://github.com/unitaryai/detoxify. ↗
10. Hyung Won Chung, Le Hou, Shayne Longpre, Barret Zoph, Yi Tay, William Fedus, Yunxuan Li, Xuezhi Wang, Mostafa Dehghani, Siddhartha Brahma, et al. 2024. Scaling instruction-finetuned language models. Journal of Machine Learning Research, 25(70):1–53. find
11. Tim Dettmers, Artidoro Pagnoni, Ari Holtzman, and Luke Zettlemoyer. 2023. Qlora: Efficient finetuning of quantized llms. Advances in neural information processing systems, 36:10088–10115. ↗
12. Nghia Duong-Trung, Xia Wang, and Miloš Kravčík. 2024. Bloomllm: Large language models based question generation combining supervised fine-tuning and bloom’s taxonomy. In R. Ferreira Mello, N. Rummel, I. Jivet, G. Pishtari, and J. A. Ruipérez Valiente, editors, Technology Enhanced Learning for Inclusive and Equitable Quality Education, volume 15160 of Lecture Notes in Computer Science. Springer. ↗
13. Min Fan, Xinyue Cui, Wanqing Ma, Haiyan Li, Xin Tong, Lin Yang, and Yonghui Wang. 2025. From words to wonder: Designing and evaluating an ai-empowered creative storytelling system for elementary children. In Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems, pages 1–15. ↗
14. Asma Farajidizaji, Vatsal Raina, and Mark Gales. 2024. Is it possible to modify text to a target readability level? an initial investigation using zero-shot large language models. In Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024), pages 9325–9339. ↗
15. Lin Gao, Jing Lu, Zekai Shao, Ziyue Lin, Shengbin Yue, Chiokit Leong, Yi Sun, Rory James Zauner, Zhongyu Wei, and Siming Chen. 2025. Fine-tuned large language model for visualization system: A study on self-regulated learning in education. IEEE Transactions on Visualization and Computer Graphics, 31(1):514–524. ↗
16. Dominik Glandorf and Detmar Meurers. 2024. Towards fine-grained pedagogical control over english grammar complexity in educational text generation. In Proceedings of the 19th Workshop on Innovative Use Ari Holtzman, Jan Buys, Li Du, Maxwell Forbes, and Yejin Choi. 2020. The curious case of neural text degeneration. In International Conference on Learning Representations (ICLR). find our card
17. Chandra Irugalbandara, Ashish Mahendra, Roland Daynauth, Tharuka Kasthuri Arachchige, Jayanaka Dantanarayana, Krisztian Flautner, Lingjia Tang, Yiping Kang, and Jason Mars. 2024. Scaling down to scale up: A cost-benefit analysis of replacing openai’s llm with open source slms in production. In 2024 IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS), pages 280–291. IEEE. ↗
18. Qiao Jin and Ye Yuan. 2025. “they all look mad with each other”: Understanding the needs and preferences of children and parents in ai-generated images for stories. International Journal of Child-Computer Interaction, page 100787. find
19. Hyeongdon Kim et al. 2025. Readctrl: Personalizing text generation with readability-controlled instruction learning. In Proceedings of the 2025 Workshop on Intelligent and Interactive Writing Assistant (In2Writing), pages 1–15. find
20. Holly B Lane and Valentina Contesse. 2022. UFLI foundations: An explicit and systematic phonics program. Ventris Learning. find
21. Walter L Leite, Stephanie Hammerschmidt-Snidarich, A Corinne Huggins-Manley, Hatice K Karakis, Holly Lane, Seyedahmad Rahimi, Qian Shen, Xinyi Tai, and Matthew Schmidt. 2025. Storiza: A platform to support children’s oral reading fluency development with generative ai. In Proceedings of the 19th International Conference of the Learning Sciences-ICLS 2025, pp. 1574-1578. International Society of the Learning Sciences. ↗
22. Jasmine Lesner, Luke Murayama, Tony Guizar, Poom Phunjamaneechot, and Daniel Shapiro. 2024. Ai personalized interactive fiction for young children. In ECAI 2024, pages 4756–4763. IOS Press. ↗
23. Mengzhe Liao. 2025. Ai-powered storytelling relay: Designing a creative and interactive game for children and parents. find
24. Zhengyuan Liu, Stella Xin Yin, Dion Hoe-Lian Goh, and Nancy Chen. 2025. COGENT: A curriculum-oriented framework for generating grade-appropriate educational content. In Proceedings of the 20th Workshop on Innovative Use of NLP for Building Educational Applications (BEA 2025), pages 129–143, Vienna, Austria. Association for Computational Linguistics. ↗
25. Tuula Merisuo-Storm and Marjaana Soininen. 2014. Interesting reading materials and exercises encourage also reluctant boys to read. Procedia-Social and Behavioral Sciences, 116:2583–2588. ↗
26. Subhojyoti Mukherjee, Viet Dac Lai, Raghavendra Addanki, Ryan A Rossi, Seunghyun Yoon, Trung Bui, Anup Rao, Jayakumar Subramanian, and Branislav Kveton. 2025. Offline rl by reward-weighted fine-tuning for conversation optimization. In The Thirty-ninth Annual Conference on Neural Information Processing Systems. ↗
27. Long Ouyang, Jeff Wu, Xu Jiang, Diogo Almeida, Carroll L. Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, John Schulman, Jacob Hilton, Fraser Kelton, Luke Miller, Maddie Simens, Amanda Askell, Peter Welinder, Paul Christiano, Jan Leike, and Ryan Lowe. 2022. Training language models to follow instructions with human feedback. Advances in Neural Information Processing Systems, 35:27730–27744. ↗
28. Yuling Sun, Jiaju Chen, Bingsheng Yao, Jiali Liu, Dakuo Wang, Xiaojuan Ma, Yuxuan Lu, Ying Xu, and Liang He. 2024. Exploring parent’s needs for children-centered ai to support preschoolers’ interactive storytelling and reading activities. Proceedings of the ACM on Human-Computer Interaction, 8(CSCW2):1–25. ↗
29. Nathalie J Veenendaal, Margriet A Groen, and Ludo Verhoeven. 2015. What oral text reading fluency can reveal about reading comprehension. Journal of Research in Reading, 38(3):213–225. ↗
30. Yizhong Wang, Yeganeh Kordi, Swaroop Mishra, Alisa Liu, Noah A Smith, Daniel Khashabi, and Hannaneh Hajishirzi. 2023. Self-instruct: Aligning language models with self-generated instructions. In Proceedings of the 61st annual meeting of the association for computational linguistics (volume 1: long papers), pages 13484–13508. ↗
31. Eric Zelikman, Yuhuai Wu, Jesse Mu, and Noah Goodman. 2022. Star: Bootstrapping reasoning with reasoning. Advances in Neural Information Processing Systems, 35:15476–15488. ↗ our card
32. J Peters. 2006. Reinforcement learning by reward-weighted regression. In NIPS 2006 Workshop: Towards a New Reinforcement Learning? find
33. Shenao Zhang, Zhihan Liu, Boyi Liu, Yufeng Zhang, Yingxiang Yang, Yongfei Liu, Liyu Chen, Tao Sun, and Zhaoran Wang. 2025. Reward-augmented data enhances direct preference alignment of llms. In International Conference on Machine Learning, pages 75591–75614. PMLR. find
34. Yuxiao Qu, Tianjun Zhang, Naman Garg, and Aviral Kumar. 2024. Recursive introspection: Teaching language model agents how to self-improve. Advances in Neural Information Processing Systems, 37:55249– 55285. ↗
35. Chunting Zhou, Pengfei Liu, Puxin Xu, Srinivasan Iyer, Jiao Sun, Yuning Mao, Xuezhe Ma, Avia Efrat, Ping Yu, Lili Yu, et al. 2023. Lima: Less is more for alignment. Advances in Neural Information Processing Systems, 36:55006–55021. ↗
36. Timothy V Rasinski, William H Rupley, David D Pagie, and William Dee Nichols. 2016. Alternative text types to improve reading fluency for competent to struggling readers. International Journal of Instruction, 9(1):163–178. find
37. Yaoming Zhu, Sidi Lu, Lei Zheng, Jiaxian Guo, Weinan Zhang, Jun Wang, and Yong Yu. 2018. Texygen: A benchmarking platform for text generation models. In The 41st international ACM SIGIR conference on research & development in information retrieval, pages 1097–1100. ↗
38. Alireza Salemi, Sheshera Mysore, Michael Bendersky, and Hamed Zamani. 2024. Lamp: When large language models meet personalization. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 7370–7392. ↗
39. Shibani Santurkar, Esin Durmus, Faisal Ladhak, Cinoo Lee, Percy Liang, and Tatsunori Hashimoto. 2023. Whose opinions do language models reflect? In International Conference on Machine Learning, pages 29971–30004. PMLR. find
40. George Spache. 1953. A new readability formula for primary-grade reading materials. The Elementary School Journal, 53(7):410–413. ↗
41. Louise Spear-Swerling. 2006. Children’s reading comprehension and oral reading fluency in easy text. Reading and Writing, 19(2):199–220. ↗

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/shen-2026-children-s-english-reading-story/