FloatinggOnion/mothertongue-demo

Demo of Mothertongue - Language acquisition for low-resource languages for the Gemini3 hackathon # Mothertongue: An Investigation into AI-Mediated Spoken Language Acquisition **Project Status**: Experimental MVP / Hackathon Submission (Groq) **Focus**: Low-resource languages (Yorùbá), Output Hypothesis, Cultural Alignment --- ## 📑 Abstract Mothertongue investigates the efficacy of Large Language Models (LLMs) as real-time conversational partners for language acquisition in low-resource contexts. Specifically, we explore how **Groq's Llama 3.3** can simulate culturally grounded immersion environments for Yorùbá learners, bridging the gap between passive understanding and active speaking fluency. This project serves as both a functional MVP for the hackathon and a proof-of-concept for scalable, culturally-aware AI tutoring systems. --- ## 🧪 Research Objectives ### 1. The Output Gap Traditional language apps focus on "Input" (reading/listening). Mothertongue operationalizes Swain's *Output Hypothesis*, positing that acquisition occurs when learners are forced to produce language to convey meaning. * **Hypothesis**: An AI partner that tolerates code-switching while gently encouraging comprehensive output can increase learner confidence faster than rigid drills. * **Implementation**: Real-time "Speaking Drills" where the AI prioritizes communicative success over grammatical perfection. ### 2. Cultural Alignment & Code-Switching Can an LLM authentically replicate specific sociolinguistic contexts? * **Experiment**: Simulating diverse Nigerian scenarios (e.g., *Agbero* conductors vs. *Mama Àgbà* elders) requires the model to handle distinct registers, honorifics, and the specific Yorùbá-English code-mixture ("Yorunglish") used in Lagos. * **Method**: We employ persona-driven prompting strategies to enforce context-specific linguistic behaviors. --- ## 🛠️ System Architecture (Methodology) To enable this immersion, we architected a resilient voice pipeline split across two providers by language: ### A. Speech Recognition (STT) All server-side transc …