Innovation in Language Learning

Evaluation of Machine Translation (MT) is an onerous but a critical task. It plays a vital role in checking the excellence of outputs from a Machine Translation system. Evaluation can be done manually as well as automatically. Manual evaluation involves human intervention making the evaluation process subjective and extensive. At the same time, it is time consuming and expensive, so automatic evaluation is becoming more popular nowadays. Automatic evaluation metrics evaluates the Comprehensiveness and Naturalness of a sentence. Several proposed metrics of MT Evaluation like BLEU have been criticized for their poor performance in evaluating machine translations. Languages like Hindi which have relatively free word-order and are morphologically rich pose additional problems in such evaluation. We explored characteristics of Hindi language and attempted here to make use of linguistic knowledge to classify errors in evaluating machine translations with Hindi as a target language. It is also important to identify different types of divergence in target language to obtain correct evaluation metrics. 

In this paper, we are exploring various evaluation metrics for machine translation. We have elaborated on the gaps in these metrics for English to Hindi Translation System. We made use of linguistic knowledge at various levels for MT Evaluation with Hindi as target language.