Syllable-based Neural Machine Translation System for Myanmar-English Language Pair

Abstract

Without doubt, Internet has made for people to communicate easily with others because it is cheap and convenient. Besides, today's world has a growing demand for immediate and accurate information. However, language remains an important barrier that prevents all information from being spread across different cultures because of the high cost in terms of money and time that human translation implies. Therefore, the demands and challenges for translation from one language to another language is steadily growing. To overcome this demands and challenges, the researchers do research the machine translation to easily translate one language to several languages for the user. Therefore, machine translation is becoming a popular field in research trends and it make smaller the language barrier. Machine Translation (MT), which is also known as Computer Aided Translation, is the task of specifically designing to translate both verbal and written texts between natural languages by a computer system. MT uses a machine translation engine to perform substitution of words or phrases or any other in one language for words or phrases or any other in another language. MT was one of the first conceive computer applications in the 1950’s and there is still many challenges. MT is widely used in Natural Language Processing tasks such as online translation services applications in information extraction, document retrieval, intelligence analysis, electronic mail, and much more. Today, some machine translation applications are available in the market. The most widely used being applications are Statistical Machine Translation (SMT), Rule-Based Machine Translation (RBMT), Hybrid Systems, which combine RBMT and SMT and Neural Machine Translation (NMT). Today, there have been very few studies on the machine translation from Myanmar language to another language. And Myanmar machine translation is still in its early stages and researchers are faced with many difficulties such as the lack of resources and there is only less amount of data corpus. Furthermore, the techniques for performing as pre-processing step for Myanmar language, such as segmentation are also currently in the process of being developed. Existing research on Myanmar translation has been either rule-based or more recently phrase-based techniques have been tried. The research aims to develop Neural Machine Translation system for Myanmar to English language pair. In this work, there are two main parts. The firstiv one is the building the Myanmar-English parallel corpus and Myanmar monolingual corpus. Although Myanmar language is one of the low resource languages, bilingual sentences are collected from the website and eBooks by crawling or copying. The second part introduces the attention-based neural machine translation system for Myanmar-English language pair. The experiments of the proposed model are done based on word-based neural machine translation model, Myanmar character-based neural machine translation model and Myanmar syllable-based neural machine translation model. Moreover, Myanmar monolingual corpus is also used to improve Myanmar syllable-based neural machine translation model. The experimental results show that Myanmar syllable-based neural machine translation model outweighs over other models.