The invention relates to a
machine translation test method. According to the method, a dependency
syntax tree is constructed for sentences,
pruning is performed on the
syntax tree according to a specific rule, validity of the sentences is destroyed based on a group of deletion operators of a dependency
syntax tree level, and words or phrases are deleted from the original sentences to generate new sentences with effective grammar and
semantics. And inputting the original text and the newly generated sentences into a tested
machine translation system, calculating bag-of-word distances, sequencing and amplifying the sentences according to the bag-of-word distances, selecting five sentences with the largest distance, manually labeling the original
sentence and the translated
sentence results, marking wrong sentences, and completing the test of the
machine translation system. The invention aims to solve the problem that the
test performance is mainly limited by the maturity of an adopted
language model due to the fact that a
test case is mainly generated by replacing a part of words in a
sentence in the current
machine translation test. When the data is amplified, the invariance of the basic structure of the sentence is ensured, so that more pairs of errors are found, and most of the errors cannot be found by a previous
machine translation test technology.