On methods of machine translation quality assessment

The article discusses approaches to determining the quality of machine translation (MT) and several methods of translation quality assessment. The aim of the article is to review a number of methods and approaches to human and automatic assessment of MT quality. The first part of the article describes the methods of relative human evaluation (ranking of translations) and absolute evaluation based on penalties for errors in translation, as well as software and algorithms that simplify human assessment. Most attention is paid to the DQF/MQM (Dynamic Quality Framework/Multidimensional Quality Metrics) error typology which is not aimed at a limited subject area as the most flexible one. The second part of the article is devoted to a review of metrics for automatic quality assessment of MT that do not use linguistic data as well as the correlation coefficients of human and automatic evaluation.