Exploring a Human-in-the-Loop Framework for Adaptive Sign Language Translation in Deaf Education

Sign Language Translation (SLT) plays a crucial role in enhancing accessibility and literacy for the deaf community. However, SLT models face significant challenges, including limited annotated datasets, sign language variability, and generalization issues. This study proposes a Human-in-theLoop (HITL) SLT conceptual framework to enhance model adaptability through continuous teacher and student interactions. By integrating Federated Learning (FL), schools can train SLT models locally while preserving data privacy and accommodating regional sign language variations. The proposed framework enables teachers to refine model predictions via Active Learning, introduce new sign gestures using Few-shot Learning, and enhance adaptability through Reinforcement Learning. Additionally, FL facilitates decentralized model updates, reducing computational burdens while improving model inclusivity across multiple schools. The synergy between HITL and FL creates an adaptive system that evolves based on real-time feedback, making AI-driven SLT more effective in educational settings. Future directions include deploying the HITL-SLT model in real-world classrooms to evaluate its practical impact. Conducting small-scale studies with teachers and students will be essential for validating its effectiveness, refining model adaptability, and enhancing interactive learning experiences.