Application of POS Tagging in Machine Translation Evaluation

Abstract

The aim of the paper is to present a process of natural language processing in its full extent as well as in machine translation from English language into Slovak as a representative of inflectional language. We aim at the data preparation phase for automatic evaluation of machine translation through POS tagging. The preparation phase for MT evaluation consists of several steps, but only the first step - creation of dataset-parallel corpus is deeply described. We focus on the source text collection of various styles and genres-dataset creation and machine translation collection. Two machine translation systems are used-web SMT Google translator API and MT@EC. As a morphology analyzing tool-TreeTagger is used. The process of dataset creation, which covers not only parallel corpora creation, but also creation of errors' database of Slovak words with morphological annotation, is analyzed. The main contribution consists of a novel approach to research of MT evaluation given by the POS tagging (machine learning methods), to identify differences between MT output and post-edited machine translation output. The ground essential of the research is machine translation errors analysis, their identification and classification, from English language into Slovak.