---
title: "From Knowing to Teaching — paper by Wang et al. (2026)"
description: "For evaluation, the authors release LessonBench: 40 expert-designed lessons, each built from between two and five source… 24 references."
authors: ["Yucheng Wang", "Shen Yang", "Jifan Yu", "Haoxuan Li", "Joy Jia Yin Lim", "Daniel Zhang-Li", "Huiqin Liu", "Lei Hou", "Juanzi Li", "Bin Xu"]
year: 2026
venue: "Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)"
source: "https://aclanthology.org/2026.acl-long.1328/"
doi: "10.18653/v1/2026.acl-long.1328"
licence: "CC BY 4.0 (ACL Anthology)"
licence_url: "https://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/wang-2026-knowing-teaching-scaffolding-pedagogical-decisions/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/wang-2026-knowing-teaching-scaffolding-pedagogical-decisions/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2026](https://md.research.elamur.ai/free/en/papers/2026/page/34/) / From Knowing to Teaching: Scaffolding Pedagogical Decisions for LLM Agent

# From Knowing to Teaching: Scaffolding Pedagogical Decisions for LLM Agent

**Authors:** [Yucheng Wang](https://md.research.elamur.ai/free/en/authors/yucheng-wang/), Shen Yang, [Jifan Yu](https://md.research.elamur.ai/free/en/authors/jifan-yu/), [Haoxuan Li](https://md.research.elamur.ai/free/en/authors/haoxuan-li/), Joy Jia Yin Lim, [Daniel Zhang-Li](https://md.research.elamur.ai/free/en/authors/daniel-zhang-li/), [Huiqin Liu](https://md.research.elamur.ai/free/en/authors/huiqin-liu/), [Lei Hou](https://md.research.elamur.ai/free/en/authors/lei-hou/), [Juanzi Li](https://md.research.elamur.ai/free/en/authors/juanzi-li/) and [Bin Xu](https://md.research.elamur.ai/free/en/authors/bin-xu/) · **Year:** 2026 · **Venue:** [Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)](https://md.research.elamur.ai/free/en/venues/acl/) ([ACL 2026](https://md.research.elamur.ai/free/en/venues/acl/2026/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2026.acl-long.1328/) · [PDF](https://aclanthology.org/2026.acl-long.1328.pdf) · [DOI 10.18653/v1/2026.acl-long.1328](https://doi.org/10.18653/v1/2026.acl-long.1328) · [OpenAlex](https://openalex.org/W7166815617)

**Citations:** 0 citations (OpenAlex, 2026-10-14)

## Summary

LLMs know a great deal about many subjects, but they struggle to turn that knowledge into good lessons. Their generated lessons look tidy yet misuse sources, skip foundational steps, and quiz learners on material never presented. The authors frame this as a knowing-teaching gap, drawing on Shulman's notion of Pedagogical Content Knowledge. They isolate three decisions that expert teachers make: selection, sequencing, and synthesis.

Guided by interviews with K-12 teachers, they built TeachCraft, a three-agent pipeline for multi-document lesson synthesis:
- **Explorer** labels each source chunk with a pedagogical intent from a six-category taxonomy.
- **Planner** orders learning objectives along Bloom's taxonomy.
- **Generator** produces slides, scripts and quizzes through a shared schema that keeps them mutually consistent.

For evaluation, the authors release LessonBench: 40 expert-designed lessons, each built from between two and five source documents of varied types. In pairwise human and LLM judging, TeachCraft beats eight baselines, and ablations show that each decision helps on its own. The work suggests that teaching skill can be split into explicit, engineerable steps.

## Key points

- **Win rates:** TeachCraft wins 67.8% of human pairwise judgments (24 raters, 920 judgments) and 79.6% of LLM-judge comparisons. The eight baselines include direct prompting, general agents and educational systems such as NotebookLM.
- **LessonBench:** 40 lessons across four subject areas, built by retrieval-augmented curation from MIT OpenCourseWare and Khan Academy.
- **Ablations:** removing synthesis or sequencing hurts overall lesson quality the most.
- **SlidesBench:** visual quality is competitive with existing slide generators, with 100% execution reliability.
- **Limitations:** the multi-agent design needs many LLM calls, which raises latency and cost compared with direct prompting.

*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

- [Tsinghua University](https://md.research.elamur.ai/free/en/universities/tsinghua-university/) — University, Beijing, China · 415 works · 94,028 citations

📚 Shelves (5): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 18,614 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 17,965 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 17,450 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 17,296 works · [🗄️ Information Systems](https://md.research.elamur.ai/free/en/shelves/computer-science/information-systems/) · 1,730 works

## Cite

Cite as: Yucheng Wang, Shen Yang, Jifan Yu, Haoxuan Li, Joy Jia Yin Lim, Daniel Zhang-Li, Huiqin Liu, Lei Hou, Juanzi Li and Bin Xu. 2026. From Knowing to Teaching: Scaffolding Pedagogical Decisions for LLM Agent. 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.

```bibtex
@inproceedings{wang2026knowing,
  title = {{From Knowing to Teaching: Scaffolding Pedagogical Decisions for LLM Agent}},
  author = {Yucheng Wang, Shen Yang, Jifan Yu, Haoxuan Li, Joy Jia Yin Lim, Daniel Zhang-Li, Huiqin Liu, Lei Hou, Juanzi Li and Bin Xu},
  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.1328},
  url = {https://doi.org/10.18653/v1/2026.acl-long.1328}
}
```

- [← previous From Isolation to Entanglement: When Do Interpretability Methods Identify and Disentangle Known Concepts?](https://md.research.elamur.ai/free/en/papers/mueller-2026-isolation-entanglement-when-do-interpretability/)
- [next → From Language to Driving: A Dual-Loop SLM-Enhanced Framework for Multi-Planner Scheduling via a Domain-Specific Language](https://md.research.elamur.ai/free/en/papers/liu-2026-language-driving-dual-loop-slm/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/34/)

## References

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

1. Tushar Aggarwal and Aarohi Bhand. 2025. Pass: Presentation automation for slide generation and speech. arXiv preprint arXiv:2501.06497. — [arXiv:2501.06497](https://arxiv.org/abs/2501.06497) [find](https://scholar.google.com/scholar?q=Tushar+Aggarwal+and+Aarohi+Bhand%2E+2025%2E+Pass%3A+Presentation+automation+for+slide+generation+and+speech%2E+arXiv+preprint+arXiv%3A2501%2E06497%2E+%E2%80%94+%5BarXiv%3A2501%2E06497%5D%28https%3A%2F%2Farxiv%2Eorg%2Fabs%2F2501%2E06497%29)
2. Alvan R. Feinstein and Domenic V. Cicchetti. 1990. High agreement but low kappa: I. The problems of two paradoxes. Journal of Clinical Epidemiology, 43(6):543–549. [find](https://scholar.google.com/scholar?q=Alvan+R%2E+Feinstein+and+Domenic+V%2E+Cicchetti%2E+1990%2E+High+agreement+but+low+kappa%3A+I%2E+The+problems+of+two+paradoxes%2E+Journal+of+Clinical+Epidemiology%2C+43%286%29%3A543%E2%80%93549%2E)
3. Robert M. Gagné, Walter W. Wager, Katharine C. Golas, and John M. Keller. 2005. Principles of Instructional Design, 5 edition. Thomson/Wadsworth, Belmont, CA. [find](https://scholar.google.com/scholar?q=Robert+M%2E+Gagn%C3%A9%2C+Walter+W%2E+Wager%2C+Katharine+C%2E+Golas%2C+and+John+M%2E+Keller%2E+2005%2E+Principles+of+Instructional+Design%2C+5+edition%2E+Thomson%2FWadsworth%2C+Belmont%2C+CA%2E)
4. Jiaxin Ge, Zora Zhiruo Wang, Xuhui Zhou, Yi-Hao Peng, Sanjay Subramanian, Qinyue Tan, Maarten Sap, Alane Suhr, Daniel Fried, Graham Neubig, and 1 others. 2025. Autopresent: Designing structured visuals from scratch. In Proceedings of the Computer Vision and Pattern Recognition Conference, pages 2902… [find](https://scholar.google.com/scholar?q=Jiaxin+Ge%2C+Zora+Zhiruo+Wang%2C+Xuhui+Zhou%2C+Yi%2DHao+Peng%2C+Sanjay+Subramanian%2C+Qinyue+Tan%2C+Maarten+Sap%2C+Alane+Suhr%2C+Daniel+Fried%2C+Graham+Neubig%2C+and+1+others%2E+2025%2E+Autopresent%3A+Designing+structured+visual)
5. Google. 2024. Notebooklm. [https://notebooklm.google/](https://notebooklm.google/). Accessed: 2025-12-26. [↗](https://notebooklm.google/)
6. Google Cloud. 2025. Gemini 2.5 pro. Large language model. Accessed 2026-01-03. [find](https://scholar.google.com/scholar?q=Google+Cloud%2E+2025%2E+Gemini+2%2E5+pro%2E+Large+language+model%2E+Accessed+2026%2D01%2D03%2E)
7. Dan Hendrycks, Collin Burns, Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song, and Jacob Steinhardt. 2020. [Measuring massive multitask language understanding](https://md.research.elamur.ai/free/en/papers/hendrycks-2020-measuring-massive-multitask-language-understanding/). arXiv preprint arXiv:2009.03300.
8. Khan Academy. 2025. Khan academy. Accessed: 2025-12-20. [find](https://scholar.google.com/scholar?q=Khan+Academy%2E+2025%2E+Khan+academy%2E+Accessed%3A+2025%2D12%2D20%2E)
9. Anthropic. 2025a. Claude code overview. Documentation. Accessed 2026-01-03. [find](https://scholar.google.com/scholar?q=Anthropic%2E+2025a%2E+Claude+code+overview%2E+Documentation%2E+Accessed+2026%2D01%2D03%2E)
10. Kunze Li and Yu Zhang. 2024. Planning first, question second: An llm-guided method for controllable question generation. In Findings of the Association for Computational Linguistics ACL 2024, pages 4715– 4729. [find](https://scholar.google.com/scholar?q=Kunze+Li+and+Yu+Zhang%2E+2024%2E+Planning+first%2C+question+second%3A+An+llm%2Dguided+method+for+controllable+question+generation%2E+In+Findings+of+the+Association+for+Computational+Linguistics+ACL+2024%2C+pages+47)
11. Anthropic. 2025b. Introducing claude sonnet 4.5. Large language model. Accessed 2026-01-03. [find](https://scholar.google.com/scholar?q=Anthropic%2E+2025b%2E+Introducing+claude+sonnet+4%2E5%2E+Large+language+model%2E+Accessed+2026%2D01%2D03%2E)
12. Manus AI. 2025. Manus. AI agent; online service. Accessed 2026-01-03. [find](https://scholar.google.com/scholar?q=Manus+AI%2E+2025%2E+Manus%2E+AI+agent%3B+online+service%2E+Accessed+2026%2D01%2D03%2E)
13. Sambaran Bandyopadhyay, Himanshu Maheshwari, Anandhavelu Natarajan, and Apoorv Saxena. 2024. Enhancing presentation slide generation by llms with a multi-staged end-to-end approach. arXiv preprint arXiv:2406.06556. [find](https://scholar.google.com/scholar?q=Sambaran+Bandyopadhyay%2C+Himanshu+Maheshwari%2C+Anandhavelu+Natarajan%2C+and+Apoorv+Saxena%2E+2024%2E+Enhancing+presentation+slide+generation+by+llms+with+a+multi%2Dstaged+end%2Dto%2Dend+approach%2E+arXiv+preprint+arX)
14. Regina Barzilay, Kathleen McKeown, and Michael Elhadad. 1999. [Information fusion in the context of multi-document summarization](https://md.research.elamur.ai/free/en/papers/barzilay-1999-information-fusion-context-multi-document/). In Proceedings of the 37th annual meeting of the Association for Computational Linguistics, pages 550–557.
15. Benjamin S Bloom, Max D Engelhart, Edward J Furst, Walker H Hill, and David R Krathwohl. 1956. Taxonomy of Educational Objectives: The Classification of Educational Goals. Longmans, Green, New York. [find](https://scholar.google.com/scholar?q=Benjamin+S+Bloom%2C+Max+D+Engelhart%2C+Edward+J+Furst%2C+Walker+H+Hill%2C+and+David+R+Krathwohl%2E+1956%2E+Taxonomy+of+Educational+Objectives%3A+The+Classification+of+Educational+Goals%2E+Longmans%2C+Green%2C+New+York%2E)
16. Robert Maribe Branch. 2009. Instructional Design: The ADDIE Approach. Springer. [find](https://scholar.google.com/scholar?q=Robert+Maribe+Branch%2E+2009%2E+Instructional+Design%3A+The+ADDIE+Approach%2E+Springer%2E)
17. Jay DeYoung, Stephanie C Martinez, Iain J Marshall, and Byron C Wallace. 2024. Do multi-document summarization models synthesize? Transactions of the Association for Computational Linguistics, 12:1043–1062. [find](https://scholar.google.com/scholar?q=Jay+DeYoung%2C+Stephanie+C+Martinez%2C+Iain+J+Marshall%2C+and+Byron+C+Wallace%2E+2024%2E+Do+multi%2Ddocument+summarization+models+synthesize%3F+Transactions+of+the+Association+for+Computational+Linguistics%2C+12%3A1043)
18. Richard E Mayer. 2009. Multimedia Learning, 2 edition. Cambridge University Press, Cambridge, UK. [find](https://scholar.google.com/scholar?q=Richard+E+Mayer%2E+2009%2E+Multimedia+Learning%2C+2+edition%2E+Cambridge+University+Press%2C+Cambridge%2C+UK%2E)
19. MIT OpenCourseWare. 2025. Mit opencourseware. Accessed: 2025-12-20. [find](https://scholar.google.com/scholar?q=MIT+OpenCourseWare%2E+2025%2E+Mit+opencourseware%2E+Accessed%3A+2025%2D12%2D20%2E)
20. Ani Nenkova, Kathleen McKeown, and 1 others. 2011. Automatic summarization. Foundations and Trends® in Information Retrieval, 5(2–3):103–233. [find](https://scholar.google.com/scholar?q=Ani+Nenkova%2C+Kathleen+McKeown%2C+and+1+others%2E+2011%2E+Automatic+summarization%2E+Foundations+and+Trends%C2%AE+in+Information+Retrieval%2C+5%282%E2%80%933%29%3A103%E2%80%93233%2E)
21. OpenAI. 2025a. Gpt-5. Large language model. Accessed 2026-01-03. [find](https://scholar.google.com/scholar?q=OpenAI%2E+2025a%2E+Gpt%2D5%2E+Large+language+model%2E+Accessed+2026%2D01%2D03%2E)
22. OpenAI. 2025b. Introducing chatgpt agent: Bridging research and action. Product announcement. Accessed 2026-01-03. [find](https://scholar.google.com/scholar?q=OpenAI%2E+2025b%2E+Introducing+chatgpt+agent%3A+Bridging+research+and+action%2E+Product+announcement%2E+Accessed+2026%2D01%2D03%2E)
23. Nicy Scaria, Suma Dharani Chenna, and Deepak Subramani. 2024. Automated educational question generation at different bloom’s skill levels using large language models: Strategies and evaluation. In International Conference on Artificial Intelligence in Education, pages 165–179. Springer. [find](https://scholar.google.com/scholar?q=Nicy+Scaria%2C+Suma+Dharani+Chenna%2C+and+Deepak+Subramani%2E+2024%2E+Automated+educational+question+generation+at+different+bloom%E2%80%99s+skill+levels+using+large+language+models%3A+Strategies+and+evaluation%2E+In+Int)
24. Lee S Shulman. 1986. Those who understand: Knowledge growth in teaching. Educational researcher, 15(2):4–14. [find](https://scholar.google.com/scholar?q=Lee+S+Shulman%2E+1986%2E+Those+who+understand%3A+Knowledge+growth+in+teaching%2E+Educational+researcher%2C+15%282%29%3A4%E2%80%9314%2E)

*Updated 1 November 2026*