A methodology for machine translation of simple sentences from Kannada to English language

Abstract

In this paper, a new approach for machine translation of simple sentences from Kannada to English language is presented. Initially, the method performs lexical analysis to process input transliterated Kannada sentence to recognize and classify tokens using newly constructed source language lexicon KannadaWordNet. Then, newly designed recursive descent parser for the domain validates the transliterated Kannada sentence with the syntactic constructs/grammar rules. Further, the phrase mapper and output generation mechanisms are employed to obtain translated English sentence. The proposed methodology is evaluated on a data set containing transliterated Kannada sentences of text written in low resolution images of display boards of Karnataka government offices in India. The method achieves an average translation accuracy of 97.56%. The proposed method is robust and efficient in dealing with various issues such as construction of source lexicon, syntactic/semantic representation of language constructs, design of efficient parser, resolving differences between syntactic/semantic constructs of source and target language, construction of bilingual dictionary. The method also resolves word sense ambiguity, translation ambiguity, and provides accurate translation without errors.