LEARNER TRANSLATION QUALITY ASSESSMENT BASED ON MACHINE LEARNING

Both learner translation and machine translation demand quality assessment. Several influential metrics used for machine translation evaluation have been applied to learner translation. Although multiple references are employed in learner translation assessment, no machine translation evaluation metric performs consistently well on all source topics. In this study, we train a new assessment model by combining the scores of several machine translation evaluation metrics into Support Vector Regression (SVR) learning model. With minimum language resources, the outputs of SVR evaluation model correlate with expert judgments significantly better than individual metric does and have better stability on our experimental corpus, which contains 9 topics EFL learners' Chinese-English translations and 9 topics English-Chinese translations respectively.