Enhancing Machine Translation: Neural and Hybrid MT engines Directing Strategies

Abstract

MT direct engines are used excessively by language learners and in translation classrooms. Such MT engines don't allow human interference while processing the translation task. Thus, applying some translation strategies during the preparation stage is essential in directing MT engines. The researcher of the present study trained a group of EFL learners in Qassim University to apply some strategies on Arabic source texts before submitting them to MT engines. The researcher used an electronic metric "BLEU" to assess the effect of the proposed strategies in directing MT engines toward accuracy. Furthermore, the responses of two neural MT engines (Google Translate and Systran) and four Hybrid MT engines (Yandix, Reverso, Collins, and Microsoft) to the premodified text (modified through the proposed strategies) were compared. Results showed that the translation strategies are effective and direct the MT engines toward accuracy. The pre-translation strategies led to 37.8 enhancement in MT translation. As for comparing the response of the six MT engines, results indicated that all the six MT engines responded effectively to the modified text and Neural Google Translate had the highest response of all.