A Novel Hybrid Approach to Improve Neural Machine Translation Decoding using Phrase-Based Statistical Machine Translation

Abstract

Phrase-based models are among the best performing statistical machine translation (SMT) systems. These systems make translations phrase-by-phrase at a time. The decoding process is done locally in these systems. In addition, neural machine translation (NMT) systems have become very popular for the past four or five years with essential features such as more fluent translations. However, sometimes NMT systems give up accuracy for fluent translations due to the nature of the decoding technique they use. In this study, we aim to develop a hybrid system by guiding NMT decoding using the output sentences of the phrase-based SMT systems. According to the two-way translation experiments, German-to-English and English-to-German, and the results obtained in terms of two popular machine translation evaluation metrics: BLEU and METEOR, our method improves the quality of NMT system translations.