ESANA: Hybrid Machine Translation Approach for English-to-Sinhala Language Translation

In modern society Internet has become the most popular and efficient communication media. Most of the Internet resources are available on English language. However, English fluency rate in majority of the countries is not up to a satisfactory level. In Sri Lanka, it was observed that the English fluency rate has been reduced over past 30 years. Therefore the neediness of an English-to-Sinhala translator increases, especially due to the unavailability of such a tool in some of the popular resources like Google Translator. There are different types of translation tools exist, which are developed using different techniques. Translation tools that are developed using Rule based machine translation (RBMT) approach relies on countless built-in linguistic rules. Furthermore, it is not capable of handling the language ambiguity. Translation tools that are developed using Statistical Machine translation (SMT) approach require a huge bilingual text corpus. In this paper, we developed an English-to-Sinhala translator called Esana translator. This was developed using a Hybrid Machine Translation approach, which combines both RBMT and SMT approaches. The evaluation results show that Esana translator is capable of providing more accurate translation results with compared to the other approaches. We argue that the higher accuracy in Esana Translator is mainly due to its capability to handle the language ambiguity in a proper way. Our concept can easily be applied to any other language translations with slight modifications.