Enhancing Two Multimodal Classroom Datasets to Advance R&D on Formative Assessment

Team: Dr. Jing Liu, Dr. Meghavarshini Krishnaswamy, Michael Chrzan

Overview: This project enhances two existing multimodal classroom datasets (EDSI and NCTE) to support AI-enabled formative assessment in K–8 mathematics. It transcribes small-group student interactions and enriches transcripts with turn-level discourse annotations and observation codes.

Subject Focus: Mathematics, Grades K-8

Targeted Universalism Focus: This dataset incorporates individual-level student demographic records to enable disaggregated equity analyses across diverse student populations

Public Goods & Deliverables:

  • Enriched EDSI Group-Work Corpus: Human-transcribed small-group audio with speaker diarization, timestamps, and demographic linkages.
  • Group-Work Detection Model: Open-source model and pipeline to automatically identify collaborative episodes in classroom audio.
  • Enhanced NCTE Corpus: Re-anonymized transcripts with word-level timestamps, turn-level discourse labels, and segment-level observation score linkages.
  • Unified Relational Schema: Data dictionary, linking documentation, and starter code (Python/SQL) under CC BY 4.0 (data) and Apache 2.0 (code).

Return to see all grantees.