Building a neural network-based English-to-Arabic transfer module from an unrestricted domain

Abstract

This paper presents a Transfer Module for an English-to-Arabic Machine Translation System (MTS) using an English-to-Arabic Bilingual Corpus. We propose an approach to build a transfer module by building a new transfer-based system for machine translation using Artificial Neural Networks (ANN). The idea is to allow the ANN-based transfer module to automatically learn correspondences between source and target language structures using a large set of English sentences and their Arabic translations. The paper presents the methodology for corpus building. It then introduces the approach that has been followed to develop the transfer module. It finally presents the experimental results which are very encouraging.