Deep learning pathways for automatic sign language processing

This study provides a comprehensive review of the current state of the sign language processing (SLP) field, encompassing sign language recognition (SLR), translation (SLT), production (SLPn), and the associated datasets (SLD). It analyzes the advancements and challenges in each area, highlighting key methodologies and technologies. The authors explore feature extraction techniques, model architectures, and multimodal data integration in SLR. For SLT, they examine neural machine translation and sequence-to-sequence frameworks, emphasizing the need for context-aware systems. In SLPn, they review avatar-based systems and motion capture techniques, identifying gaps in generating natural and expressive sign language. The survey of SLD evaluates existing datasets and underscores the importance of comprehensive data collection. It also discusses current SLP systems’ limitations and proposes future research directions to enhance accuracy, naturalness, and user-centric applications. • Comprehensive survey of deep learning (DL) in automatic sign language processing (SLP). • Systematic analysis of DL models for SLP tasks: recognition, translation, and production. • In-depth evaluation of model features, performance, strengths, and limitations. • Identified performance gaps between SLP and spoken language processing. • Recommendations for future research to address current SLP limitations.