Towards Trustworthy Sign Language Translation System: A Privacy-Preserving Edge–Cloud–Blockchain Approach

The growing Deaf and Hard-of-Hearing (DHH) community faces communication challenges due to a global shortage of certified sign language interpreters. Therefore, developing efficient and secure Sign Language Machine Translation (SLMT) systems is essential. This paper proposes a novel privacy-preserving end-to-end edge–cloud–blockchain system for SLMT that ensures real-time translation and enforces user consent management through immutable blockchain records. We evaluate our system by comparing the Encoder-Decoder Transformer, the mostly used model in SLMT, and our proposed Adaptive Transformer (ADAT) model. We evaluate the system on two datasets: RWTH-PHOENIX-Weather-2014T (PHOENIX14T) and MedASL, our newly developed medical-domain dataset. A comparative analysis of translation quality on PHOENIX14T shows that ADAT improves BLEU-4 by 0.02 absolute points and ROUGE-L by 0.11 compared to the Encoder-Decoder Transformer. On MedASL, ADAT surpasses the Encoder-Decoder Transformer with 0.01 absolute points in BLEU-4 and 0.02 in ROUGE-L. For runtime efficiency on MedASL, ADAT reduces training time by 50% and lowers both edge-cloud communication and overall end-to-end system times by 2%. These findings demonstrate that the proposed system offers a precise and efficient SLMT with low communication latency, establishing a foundation for responsible deployment in real-world domains such as healthcare, education, and public services.