Unsupervised vocabulary level translation quality evaluation method and system based on context expectation

The invention discloses an unsupervised vocabulary level translation quality evaluation method and system based on context expectation, and relates to the field of unsupervised vocabulary level translation quality evaluation. The problems that existing vocabulary level QE needs to depend on large-scale annotation data, and machine translation quality evaluation is inaccurate are solved. The method comprises the following steps: inputting a statement to be evaluated into a multi-language model, and covering a certain vocabulary at a translation end of the multi-language model; inputting the context information of the to-be-evaluated statement and the masked translated text into a multi-language model, and predicting masked words according to the multi-language model; processing the relation between each sub-word in the masked word by using conditional probability to obtain the generation probability of each sub-word; and performing vocabulary-level machine translation quality evaluation by taking the generatio