Mathematical and computational perspectives on next-generation neural networks for sign language recognition: A systematic review of advances, challenges, and assistive applications

Artificial intelligence (AI) and machine learning (ML) have revolutionized assistive technologies, particularly for individuals with hearing and speech impairments. This systematic review critically examines recent innovations in next-generation neural network architectures for sign language recognition (SLR), emphasizing their mathematical and computational foundations. Following PRISMA guidelines, we analyze state-of-the-art models, including convolutional neural networks (CNNs), recurrent neural networks (RNNs), long short-term memory (LSTM), and hybrid approaches integrating classical machine learning methods such as support vector machines (SVMs). We explore strategies for feature extraction, data augmentation, multimodal fusion, and optimization, highlighting their roles in improving accuracy, robustness, and real-time adaptability. Persistent challenges include dataset scarcity, limited generalizability, and computational trade-offs. From a mathematical perspective, optimization techniques, probabilistic modeling, and explainable AI frameworks are emerging as key enablers for safe and trustworthy SLR systems. This review identifies research gaps and proposes future directions toward responsible, mathematically grounded, and computationally efficient AI-powered assistive technologies.