A Review on Sri Lankan Sign Language Interfaces for Large Language Model Interaction

The global Deaf and Hard of Hearing (DHH) population is projected to reach 2.5 billion by 2050, and currently, there are an estimated 1.5 billion DHH individuals in the global context. It clearly highlights the urgent need for them to have proper communication technologies. In Sri Lanka, there were over 300,000 DHH individuals who relied on Sri Lankan Sign Language (SLSL), which consists of Sinhala and Tamil sign variations, and technological solutions leveraging Artificial Intelligence (AI) remain limited. This paper presents a systematic review of research on SLSL recognition, emphasizing recent developments in Computer Vision, Deep Learning (DL), and Natural Language Processing (NLP) for SLSL translation. It highlighted technologies such as Convolutional Neural Network (CNN), MediaPipe-based feature extraction, few-shot learning, and multimodal systems, while identifying the present gaps, including the grammatically correct sentence level recognition, fingerspelling and lexical limitations and limited integration with Large Language Models (LLMs) like Generative Pre-training Transformer (GPT). This paper also explores and discusses the publicly available datasets, including the SSL50 and SSL400, highlighting their role in SLSL recognition and language translation using a critical analysis. Finally, it proposes a conceptual end-to-end framework that connects gesture recognition with LLM-based system to enable natural, context-aware communication for the Sri Lankan DHH community. The study concludes by outlining key research directions for developing scalable, inclusive, and linguistically grounded SLSL-LLM systems for accessibility in Sri Lanka and other Low- and Middle-Income Countries (LMICs).