Study on evaluation method of machine translation quality based on questionnaires and data analysis
Yiqun Sun, Minkang Zhou · 2016
In order to do a global evaluation of translation software and compare their differences within specific indicators, the study applied Euclidean distance and cosine similarity to evaluate 4 commonly used translation software tools, based on the analysis of the data from the questionnaire.The results suggest that software A and B are good in general, especially at the vocabulary translation.However, there are still some shortcomings in the discourse translation.Therefore, they should accumulate more materials in the Chinese corpus.It is concluded that Euclidean distance covers the shortage of randomness and the limitations to the micro level of the present evaluation methods such as BLEU.Meanwhile, Euclidean distance has higher accuracy and more convenient calculation than the cosine similarity.It can be used to evaluate translation software effectively.