A global taxonomy and computational characterization of major sign languages for scalable AI recognition systems

Sign language is characterized as natural visual-gestural language with autonomous linguistic structures, varied orthographic influences, and legal status. Despite the recent advancements in deep learning-based recognition systems, these systems face the problem of regional biases, which limits the scalability of the systems and excludes millions of users from low-resource backgrounds. The study focuses on a comprehensive Global Taxonomy and Computational Characterization of several significant sign languages across five continents of the world. These languages are classified based on six crucial parameters: alphabetic adaptation, linguistic family, legal status, user population, manual spelling structures: one-handed or two-handed. The research proposes a Hierarchical Transformer Architecture capable of accommodating various linguistic variations. It follows a three-layer approach: a universal hand pose encoder, family-specific adaptation, and classifier head. Moreover, it presents a Global Sign Language Complexity Index (GSLCI) and a Fairness Index to measure the complexities of various sign language. Analysis of the proposed optimized framework indicates a maximum accuracy of 91% using the common features of various language families which can be considered a preferable mode of communication for people with hearing and speech impairments.