Machine Translation Evaluation for Croatian-English and English-Croatian Language Pairs
Marija Brkić Bakarić, Sanja Seljan, Maja Matetić · 2011
This paper presents a bidirectional machine translation evaluation study for the Croatian- English and English-Croatian language pairs. Translations from Croatian into English have been obtained in four different domains from four online machine translation services, i.e. Google Translate, Stars21, Translation Guide and InterTran. These translations have been evaluated by three different automatic accuracy metrics, i.e. F-measure, BLEU and NIST, as well as by human evaluators. Evaluations are based on a single reference per sentence. In the reverse translation direction, Google Translate output has been analyzed in the same manner. System level correlation between F-measure, BLEU, NIST and human assessments is given and the significance of the results is discussed.