Abstract
Sign language translation is one of the important issues in communicationbetween deaf and hearing people, as it expresses words through hand, body, andmouth movements. American Sign Language is one of the sign languages used, oneof which is the alphabetic sign. The development of neural machine translationtechnology is moving towards sign language translation. Transformer became thestate-of-the-art in natural language processing. This study compares theTransformer with the Sequence-to-Sequence (Seq2Seq) model in translating signlanguage to text. In addition, an experiment was conducted by adding ResidualLong Short-Term Memory (ResidualLSTM) in the Transformer. The addition ofResidualLSTM to the Transformer reduces the performance of the Transformermodel by 23.37% based on the BLEU Score value. In comparison, the Transformeritself increases the BLEU Score value by 28.14 compared to the Seq2Seq model.