Error Classification and Evaluation of Machine Translation Evaluation Metrics for Hindi as a Target Language

Abstract

AbstractEvaluation of Machine Translation (MT) is an onerous but a critical task. Automatic evaluation metrics evaluates the adequacy and fluency of a translated sentence. Automatic evaluation of machine translation is able to compare between two different translation systems but it doesn’t provide any insights into the kind of errors a translation system is making. Our error classification, inspired by Vilar et al., has extended categories more linguistically for Hindi language. In this paper, we will explore various evaluation metrics for machine translation and perform extensive linguistic and statistical analysis of the translation output to identify primary issues in existing framework of automated metrics for English-to-Hindi MT systems. This leads us to better insights for improvement of these metrics for English-to-Hindi automatic machine translation.KeywordsEvaluation metricEnglish-to-Hindi MTError classification in MT