Multi-corpus mixed attention-assisted machine translation algorithm and system

The invention discloses a multi-corpus mixed attention-assisted machine translation algorithm, which comprises the following steps of: collecting three types of parallel corpora of different sources of English and Chinese contrasts, and mixing the parallel corpora into training corpora according to a ratio of 1: 1: 1; the neural network model is used for training a transformers neural network model based on the seq2seq architecture; inputting a to-be-translated English sentence into a manually collected professional word-English contrast word list of each field to search professional words and corresponding Chinese translations; inputting English sentences into the trained neural network model, translating word by word, selecting a result with the highest probability as a translation result, gradually calculating an attention matrix of professional words, calculating an attention value, and when the attention value exceeds a certain threshold value, executing the step-by-step translation; if the probability o