Neuro MT: A Learning Machine Translation Web Service

Abstract

Communication through words is the most e ective way of exchanging information between people of the same language. But it may prove di cult if there's a language barrier. The advent of computers provided a solution that is an immediate and reliable substitute to a dictionary or a translator, the solution is called Machine Translation(MT). The problem with the common implementation of MT, Statistical Machine Translation(SMT), is that it requires discrete models that derives a part of the translation before learning a translation. The usage of Neural Machine Translation (NMT), removes the need to rely on discrete models to learn a translation. This special problem uses NMT for Filipino to English bidirectional translations. It also provides a web service that is easily integrable to mobile and web applications. The system also allows users to submit translation suggestions. This allows the NMT system to be improved with new data. The trained NMT models included in this special problem produces translations that are comparable against the current SMT used for Filipino to English bidirectional translations.