Automatic Evaluation of Machine Translation Output for Slovak Language

Abstract

Future improvements of machine translation (MT) systems require reliable tools for automatic evaluation of machine translation system like tools for the identification and classification of machine translation errors. The aim of this paper is to summarise the metrics for automatic evaluation of machine translation systems and to propose a tool for automatic MT evaluation focusing on one translation direction, into Slovak language, that will be used in our project and in a dissertation thesis. We focus on a goal of the automatic MT evaluation; we describe the relevance of automatic evaluation of MT output in comparison to manual evaluation, basic methods for the automatic MT evaluation with the basic linguistic errors occurring frequently in the machine translation output. Our proposed tool for automatic MT evaluation comprises these methods and provides output data that can be used to improve existing machine translation systems like the Google translator API or the MT@EC, both are be used in our project.