Analysis on the development of the evaluation methods for machine translation

This paper studies the current evaluation methods for Machine Translation (MT), and reveals their core evaluation techniques. Through the comparative analysis on the differences of MT evaluation methods, this paper points out the present MT research difficulties and hotspots, and makes a qualitative review on the future MT evaluation research direction. In general, there are still some shortcomings on current MT evaluation methods, mainly characterized by the subjectivity of human evaluation and the incapability of quantitative evaluation. On the other hand, applying the evaluation method of the error type analysis can quantify the evaluation results, but lacks the evaluation for the readability of the translation results.