Data Augmentation Techniques in Automatic Translation of Vietnamese Sign Language for the Deaf

Automatic translation of Vietnamese sign language is a new and meaningful problem. Automatic translation of Vietnamese sign language is a meaningful solution as it promises to eliminate communication obstacles and improve the lives of deaf people. With the support of automatic translation technologies, many effective translation methods exist. However, the important problem in the current problem is that there is not a large enough amount of data available to evaluate and develop translation models. With Wordnet, automatic data augmentation is possible. By applying the hyponym and hypernym in Wordnet with the criteria of this study, we enrich the data based on the original data built. The measure of data similarity between the sentences generated from the original sentence is evaluated accordingly based on the combination of the cosine measure and meets the data requirements of the Vietnamese sign language automatic translation problem. Experiments show that BLEU scores on some translation models achieve high results after data augmentation.