Code-Switched Darija-French Tutoring Agents for STEM Accessibility
Code-Switched Darija-French Tutoring Agents for STEM Accessibility
Description
Details
Context and Problem Statement
The transition from Arabic-medium secondary education to French-medium scientific higher education creates a dual burden: understanding the concept and decoding the language of instruction. Moroccan classroom explanations often alternate naturally between Darija and French.
Research Question
Does controlled code-switching improve conceptual understanding without weakening acquisition of French scientific terminology, and how can this effect be measured reliably?
Proposed Approach
Adapt a Darija-capable language model to scientific tutoring, define an explicit code-switching policy that preserves technical terms in French, and evaluate linguistic quality, pre-test/post-test learning gains, and student preference.
Expected Contribution
A first benchmark for Darija-French STEM tutoring and empirical evidence on the educational effect of controlled code-switching.
Expected Prototype
A text-and-voice tutoring agent for a first-year STEM module with adjustable code-switching intensity.
Datasets
Darija-SFT-Mixture, DarijaMMLU, a bilingual pedagogical corpus, and consent-based tutoring transcripts.
Challenges
Dialectal variability, lack of standardized written Darija, potential delay in formal-language acquisition, and experimental confounding.
Research Question
Innovation
Expected Deliverable
Technologies
Required Skills
- Multilingual NLP and Model Adaptation
- Educational Sciences and Experimental Methodology
- Speech Processing
- Statistical Analysis
Datasets
- Darija-SFT-Mixture and DarijaMMLU (MBZUAI-Paris)
- Bilingual pedagogical explanation corpus
- Consent-based tutoring-session transcripts