Chinese to English Machine Translation using Boundary-Assistant Encoder-Decoder Network with Embeddings from Language Models

In recent years, translation of text from one language to another without human involvement is done automatically through Artificial Intelligence (AI) which is defined as English Machine Translation (EMT). The Human Machine Translation (HMT) struggles to understand context of sentence, tone and nuance which leading to inaccurate translations for conveying original text. Hence, this research proposes Boundary-Assistant Encoder-Decoder NetworkEmbeddings from Language Models (BANet-ELMO) for encoding English sentence into a sequence of vectors through context-aware representation to generate output sentence. Initially, ParaCrawl (PC) corpus dataset is used in parallel sentence pairs which contains multiple languages with range of texts, articles and websites. After that, preprocessing step includes lower casing and stemming through Porter Stemmer (PS) for converting and removing suffixes of same word to lower case regardless of capitalization within the code. Finally, BANetELMO is proposed for contextual understanding of input text and enhanced semantic representation with improved translation accuracy between words and phrases. The proposed BANet-ELMO achieved better results such as accuracy (acc) is (34.46%), Bi-Lingual Evaluation Understudy (BLEU) score is (28.4%) and Error Rate (ER) is (22.3%) when compared to existing Neural Machine Translation (NMT).