Machine translation algorithm and device based on layer aggregation

Abstract

The invention relates to the technical field of text translation, and discloses a machine translation algorithm and device based on layer aggregation. The algorithm comprises the steps: obtaining a to-be-translated Chinese statement, and carrying out the preprocessing of the to-be-translated Chinese statement; the ATransformer encoder performs multilayer semantic feature information extraction onthe preprocessed statement on the basis of a multilayer information extraction algorithm; the ATransformer decoder is used for decoding the multi-layer semantic feature information and outputting a translation target language sequence; the judgment model D judges the translation target language sequence; if the translation target language sequence is judged to be a translation result; and if yes,taking the translation target language sequence as a final machine translation result and outputting the final machine translation result. Otherwise, parameter updating is conducted on the ATransformer model based on the strategy gradient algorithm, the preprocessed statement to be translated is input into the updated ATransformer encoder, and the machine translation algorithm is achieved again. The invention further provides a device of the machine translation algorithm based on layer aggregation. According to the invention, intelligent translation of the text is realized.

Description

technical field [0001] The present invention relates to the technical field of text translation, in particular to a machine translation algorithm and device based on layer aggregation. Background technique [0002] With the development of deep learning in the field of natural language processing, machine translation has transitioned from the early statistical machine translation research centered on shallow machine learning to neural machine translation research centered on deep learning technology. [0003] The disadvantage of traditional statistical machine translation is that it requires human experts to design features and corresponding translation process, it is difficult to deal with long-distance dependencies, and it will also cause serious data sparsity problems due to data dispersion; while neural machine translation models combine attention mechanisms, effectively Long-distance dependence is alleviated, and the effect is far better than statistical machine translat...

AI Technical Summary

Problems solved by technology

[0003] The disadvantage of traditional statistical machine translation is that it requires human experts to design features and corresponding translation process, it is difficult to deal with long-distance dependencies, and it will also cause serious data sparsity problems due to data dispersion; while neural machine translation models combine attention mechanisms, effectively Long-distance dependencies are alleviated, and the effect is far better than statistical machine translation models on large-scale parallel corpora

However, it has been found through research that different layers in the neural machine translation model can capture different types of grammatical and semantic information. To sum up, there is a lack of utilization of the information propagated by the middle layer. At the same time, the existing neural machine translation models usually adopt a single-model training method based on the maximum likelihood principle, that is, the current translation model is used as the training target, and the source language is maximized. It is difficult to guarantee the naturalness and accuracy of the translation results

Images