The design and implementation of Language Learning Chatbot with XAI using Ontology and Transfer Learning

The chatbot uses explainability tools to show how it generates responses: Rather than treating the AI as a "black box," the system uses an ontology graph (a visual map of connected concepts) to explain why the chatbot produces specific English learning suggestions, making it transparent how the AI reaches its conclusions.

Three-level learning design matches real language progression: The system is structured around phonetics (pronunciation), semantic understanding (domain-specific conversations), and free-style conversation, reflecting how people actually learn languages rather than treating all learning stages equally.

Transfer learning reduces the need for massive training datasets: By using GPT-2 and fine-tuning it on English learning data rather than training from scratch, the developers created a functional language learning tool more efficiently—an important consideration for responsible AI development that doesn't require unlimited computational resources.

Integration with existing platforms increases accessibility: By embedding the chatbot into WeChat's mini-program feature, the system reaches users through an app they already use, demonstrating how responsible AI design considers practical deployment and user accessibility rather than existing only in research environments.

Explainability is built into the core design, not added later: The ontology graph isn't a separate explanation layer but is fundamental to how the system works, suggesting that transparency in AI systems works best when designed from the beginning rather than retrofitted.