Machine translation from English to Malayalam using transfer approach

Abstract

A machine translation system converts text from a natural language to other while abiding to the syntax and semantics of the latter. The area of interest here is a Rule Based machine translation system that translates text from English to Malayalam using transfer approach. The system is designed to translate sentences from cricket domain related articles. The purpose behind making the system domain specific is to improve translation quality of sentences in the chosen domain. Our system depends on Stanford parser in the preprocessing stage. The input English sentence is tokenized, and parsed with the help of Stanford parser, and a parse tree is generated. According to the rules structured for the target language the source parse tree is reordered to produce the target parse tree. The words in target tree are mapped with bilingual English - Malayalam dictionary to obtain output Malayalam words. The system also maintains a separate verb dictionary that holds English verb in root form and its Malayalam meaning with 9 different inflections. Morphological generation phase provides necessary inflections for Malayalam words and thereby creates meaningful translations. The domain specific translation system was able to achieve correct translation results for about 86% of test sets.