Modern MT Systems and the Myth of Human Translation: Real World Status Quo

This paper objects to the current consensus that machine translation (MT) systems are generally inferior to human translation (HT) in terms of translation quality. In our opinion, this belief is erroneous for many reasons, the both most important being a lack of formalism in comparison methods and a certain supineness to recover from past experience. As a side effect, this paper will provide evidence for a much more favorable judgment of the performance of contemporary MT systems. We will present and discuss known methods of automatic MT evaluation, give real world examples of both machine and human translation and finally suggest an universal formal evaluation method to handle both human, as well as MT output in a comparable fashion.