Metric for Evaluation of Machine Translation Quality on the bases of Edit Distances and Reverse Translation

Abstract

The article describes a metric for evaluation of machine translation quality based on edit distances and reverse translation. The object of the research is texts in any alpha languages with different bases (alphabets), as well as their translations into other alphabetic languages. We consider the existing translation evaluation metrics, methods, and algorithms for improving the quality of translation. The developed methods and algorithms for evaluation and improvement of translation quality are presented in the article. We also determined a set of conditions necessary for automatic search for the optimal variant of hybrid machine translation of a text at the grapheme level, including maximum, minimum, and average values of translation lengths, reverse translations, and edit distances between pairs of texts that have the same meaning. We also did a correlation analysis of the compliance of the results of evaluation of the quality of translations of scientific and technical texts with existing methods and the developed method based on reverse translation.