Evaluation of the performance of Moses statistical engine adapted to English-Arabic language combination

Abstract

Statistical Machine Translation (SMT) is considered as sub-field of computational linguistics; and the latter is regarded as a branch of Artificial Intelligence (AI) dedicated to Natural Language Processing (NLP). The main purpose of this paper is shortening the distance between the Language and the most recent cutting edge technology dedicated to Machine Translation (MT). On the one hand, Statistical Machine Translation (SMT) considers the translation as a human craft, hence it uses linguistics monolingual and bilingual corpora translated by professional translators; the monolingual are used to train the Language Models (LM) and the bilingual are used to train the Translation Models (TM). On the other hand, it takes advantage of the processing high performance of computers by integrating Statistical Methods to select the best translation. This paper presents the basic concepts and approaches of Machine Translation (MT), and focuses on SMT, then introducing the features of the open source Moses Decoder. This system has been experimented and adapted to translate from English into Arabic. The English-Arabic prototype has been evaluated using MEDAR MT package and the obtained results were very encouraging.

Keywords: machine translation, statistical machine translation, Arabic machine process, Moses Decoder, Giza alignment tool