A Hybrid Approach for Machine Translation to Low Resource Indian Language: A Case Study on Chhattisgarhi

Today most of the machine translations of natural languages use machine learning (ML) techniques that rely heavily on the training dataset. Due to the unavailability of vast, diverse datasets with annotated corpus capturing every feature of a low resource language, the accuracy in the results obtained for machine translation of these languages using machine learning are extremely low. For low resource Indian language like Chhattisgarhi, today it is quite difficult to obtain a suitable corpus which will produce highly accurate results for machine translation. Moreover, the presence of homonyms is found to decrease the accuracy to a considerable extent. Considering these aspects this article proposes a hybrid approach involving both rule-based and machine learning based techniques for machine translation to such low resource languages. The proposed approach has been used to perform a case study for the translation of a high resource language like English to a low-resource language like Chhattisgarhi. From the case study it is found that the proposed approach provided an accuracy which is more than 75%.