Benefits and State of the Art of Automatic, Unsupervised Estimation of MT Quality.
Niko Papula · 2014
Automatic MT (machine translation) quality estimation system is briefly described and its results are compared to findings of 2012 Workshop on Statistical Machine Translation. The system is found to be better than other published systems, based on the criteria used in the mentioned conference. Practical applicability of the system is discussed based on conducted surveys and real-life test. Post-editing using generally available MT services is found to be usefuf in practice, contrary to findings in previous literature. The theoretical key finding is that comparing output from several generally available MT engines provides better than state-of-the-art results in automatic MT quality estimation.