Review of End-to-End Sign Language Recognition and Translation Methods based on Deep Learning

Web proceedings papers

Authors

Angela Madjar , Jovana Markovska , Jovana Dobreva , Kostadin Mishev , Monika Simjanoska and Dimitar Trajanov

Abstract

Sign languages are an extremely powerful communication tool for many deaf and hard-of-hearing people. It is important to concentrate on promoting better awareness of and sensitivity to their community. The significance of sign language is gaining momentum, finally, and it is evident more people are seeing the need for it in today’s society. In recent years, research has progressed steadily in regard to the use of different approaches that model the sign languages. The purpose of this paper is to review significant methods based on deep-learning in the fields of Sign Language Recognition and Translation. Most of the analyzed method performances are evaluated on the challenging RWTHPHOENIXWeather- 2014T (PHOENIX14T) dataset. Our analysis showed that, to this moment, the Transformer-based methods repeatedly report state-ofthe- art results, thus are the most competent methods at handling this type of demanding tasks.

Keywords

Sign Language · Translation · Recognition · Review · Natural Language Processing