Speaker
Description
Generative AI can produce lesson plans, activities, and differentiated materials within seconds. But useful educational AI requires more than better prompting: it needs meaningful educational context. This presentation examines an emerging approach to grounding AI-supported instructional decision-making in authentic curriculum language and documented teacher practice. Drawing on the development of a curriculum corpus and Teacher Practice Corpus, the presentation shows how instructional episodes can represent relationships among tasks, resources, language functions, disciplinary practices, assessment, feedback, and differentiation. Examples from multilingual science education illustrate how such evidence can help distinguish generic AI generation from recommendations grounded in what students are learning and what teachers actually do. Participants will consider how evidence-grounded AI architecture could complement existing AI tools rather than replace them, while preserving teacher judgment and adapting to different educational contexts.
Abstract section 1: Relevance
Generative AI is increasingly used to support K–12 instruction, while important pedagogical and systemic challenges remain (Huang et al., 2026). Research also shows that supplying LLMs with curriculum materials and expert-informed guidance can improve instructional scaffolding (Malik et al., 2025). Corpus-informed approaches provide an established means of grounding language-teaching materials in authentic language evidence (Friginal & Prado, 2025). Meanwhile, emerging science-education research emphasizes teachers’ professional judgment in preparing and contextualizing instructional data for AI-supported teaching (An & Martin, 2026). This presentation brings these strands together by examining curriculum language and documented teacher practice as complementary evidence for AI-supported instructional decision-making.
Abstract section 3: Content/method
The presentation draws on iterative development of two complementary resources: a 36-text middle-school science curriculum corpus and a Teacher Practice Corpus representing documented instructional episodes. Teacher-practice cases are coded for task, resource, language function, disciplinary practice, assessment, differentiation, and instructional relationships, while evidence status distinguishes planned from enacted practice. Examples from U.S. middle-school science and an enacted Thai high-school physics case are used to examine whether the representation transfers across instructional contexts.
Abstract section 2: Contribution/research questions
This practice-oriented presentation examines how educational evidence can inform AI-supported instructional design. It asks: (1) What information about curriculum and teacher practice should be represented for AI-supported decision-making? (2) How can relationships among instructional episodes, feedback, differentiation, assessment, and language demands be captured? (3) How might these representations help teachers use existing generative AI more responsively with multilingual learners?
Abstract section 4: Outcomes/results
Preliminary corpus development indicates that instructional context extends beyond lesson content alone. The cases reveal dependencies across episodes involving prior knowledge, feedback uptake, evidence reuse, differentiation, concept development, and assessment progression. Representing these relationships also requires attention to evidence provenance: a detailed lesson plan should not automatically be treated as enacted classroom practice. Comparison of the U.S. and Thai science cases suggests that a common episode-level architecture can represent substantially different pedagogical frameworks without redesigning the schema. The presentation considers how these findings could inform retrieval and constraint strategies for future AI-supported teacher decision-making.
Abstract section 5: References
Malik, R., Abdi, D., Wang, R., & Demszky, D. (2025). Scaffolding middle school mathematics curricula with large language models. British Journal of Educational Technology, 56(3), 999–1027. https://doi.org/10.1111/bjet.13571
Friginal, E., & Prado, M. (2025). Corpus-informed materials in language teaching. In L. McCallum & D. Tafazoli (Eds.), The Palgrave Encyclopedia of Computer-Assisted Language Learning. Palgrave Macmillan. https://doi.org/10.1007/978-3-031-51447-0_45-1
Huang, R., Yin, Y., Zhou, N., & Lang, F. (2026). Artificial intelligence in K–12 education: An umbrella review. Computers and Education: Artificial Intelligence, 10, 100519. https://doi.org/10.1016/j.caeai.2025.100519
An, T., & Martin, S. N. (2026). Reframing teacher professionalism in AI-supported science education: Digital integration and automation design as core competencies. Journal of Science Teacher Education. https://doi.org/10.1080/1046560X.2026.2670980
| Title | Beyond Generic AI: Grounding AI in Teacher Practice |
|---|---|
| Teaching Context | Teaching mature/lifelong learners |