Objective evaluation of second language learner’s translation proficiency using statistical translation measures
Hajime Tsubaki, Keiji Yasuda, Hirofumi Yamamoto, Yoshinori Sagisaka · ExLing Conferences · 2008
For objective evaluation of a second-language learner’s translation proficiency, we tested two measures commonly used in statistical machine translation. One is the word $n$-gram probability of a target language to measure the likelihood of translated sentences in that language. The other is the translation probability from source-language sentences to target-language sentences to measure translation accuracy between the sentence pairs. The subjective proficiency scores of Japanese learners were compared with these objective measures extracted from the English sentences they translated. Statistical analysis showed no correlation for the word $n$-gram probabilities, but a high correlation for the translation probabilities, which suggests their usefulness for the objective evaluation of learners' translation proficiency.