A STUDY OF DATA AUGMENTATION AND ACCURACY IMPROVEMENT IN MACHINE TRANSLATION FOR VIETNAMESE SIGN LANGUAGE

Sign languages are independent of deaf communities. The translation from normal (i.e., Vietnamese Language - VL) as long other sign to language (VSL) is a meaningful task that breaks down communication barriers and improves the quality life for community. In this paper, we experimented with proposed several methods building improving models VL VSL task. We presented data augmentation method improve performance our neural machine models. Using an initial dataset 10k bilingual sentence pairs, were able obtain new 60k pairs perplexity score no more than 1.5 times original dataset. Experiments on showed rule-based achieved highest BLEU 68.02 among However, augmented dataset, Transformer model best 89.23, which significantly better conventional approach methods.