Stanford Language Co-Pilots: AI Tools for Supporting Multilingual Learners in Math

Subnetwork: ENCORE

Joined 2024

---

Partners

• Stanford Graduate School of Education

• EduNLP Lab

• Harvard University

• University of Maryland

• Oakland Unified School District (CA)

• Crane Schools (AZ)

• Fitchburg Public Schools (MA)

What are we building?

• Multilingual Learner Math Tools:

  • Dataset of multilingual learners talking about math (including expert-annotated data)

  • Algorithms based on this dataset to support teacher feedback and lesson planning

• AI Scaffolding Support:

  • Dataset of visual aids used in math classrooms

  • Dataset of teacher adaptation requests of Amplify Desmos curriculum

  • Algorithms to help teachers adapt instructional plans and scaffolding strategies for multilingual learners

• Teacher-facing app (M-Powering Teachers):

  • Provides automated feedback on lesson audio recording to support coaching sessions

  • Supports teacher reflection around high-leverage teaching moves

• Grounding in and Sharing Teacher Insights:

  • Building spaces for teacher experiences and insights to guide educational research and development

  • Empower teachers to be embedded in the development process and provide more than just feedback and annotation

What are we learning?

• How teachers engage with AI-powered tools to improve lesson planning and adaption for all learners.

• How to detect student-rich engagement in math learning, especially for multilingual students.

• How teachers evaluate and use LLM-powered tools to address challenges in their practice

Products

• Multilingual math transcript datasets

• Teacher curriculum adaptation datasets

• Automated Feedback tool to help teachers and coaches explore high leverage teaching practices

---

AIMS Collaboratory | Inventory of Public Goods

Public goods shared by Stanford Language Co-Pilots: AI Tools for Supporting Multilingual Learners in Math

For: Researchers, Developers

MathemaTikZ: A Dataset and Benchmark for Mathematical Diagram Generation

Introduces a dataset and benchmark for evaluating how well AI models generate mathematical diagrams from curriculum-based descriptions and problem contexts.

Academic Article

For: Researchers, Developers, Educators

Scaffolding middle school mathematics curricula with large language models

Explores how large language models can help teachers create curriculum scaffolds, including warmup tasks that support access to middle school mathematics.

Academic Article

For: Researchers, Developers

DiagramIR: An Automatic Pipeline for Educational Math Diagram Evaluation

Presents an automatic evaluation pipeline for educational math diagrams, using intermediate representations of TikZ code to compare AI-generated figures with human ratings.

Academic Article

---

Subnetwork: ENCORE Engagement Collaborative for Research & Equity (ENCORE), is a subnetwork of the AIMS Collaboratory.