A neural machine translation method based on a user dictionary

Abstract

The invention discloses a neural machine translation method based on a user dictionary, and the method comprises the steps: crawling mass data from a network through a web crawler technology, and obtaining a bilingual word pair from a corpus through an extraction technology to construct the user dictionary; Using a user dictionary to perform dictionalization on the training corpus, extracting thetraining corpus, and mixing the extracted training corpus with the original corpus to serve as input of neural network model training; Performing consistency detection on the user dictionary placeholders contained in the sentence pairs; Processing the training data by using a user dictionary, inputting the training data into a neural network model, and starting to train the model until the model converges; And inputting a sentence containing the user dictionary, obtaining dictionary information to replace the placeholder, and translating at the same time to obtain a high-precision translated text matched with the information in the user dictionary. During translation, the high-precision requirements of different users on noun phrases and named entities are well met, and the user-defined dictionary library is added according to the requirements and translation habits of the users, so that the high-quality requirements of the different users on translated texts are met.

Description

technical field [0001] The invention relates to a machine translation method, in particular to a neural machine translation method based on a user dictionary. Background technique [0002] Machine translation realizes the function of mutual conversion between two languages, and its translation ability is automatically learned from a large number of parallel corpora. In the neural machine translation task, a high-performance neural machine translation system needs a large amount of bilingual corpus as input. The bilingual parallel sentence pairs in these corpora are processed into multiple continuous word sequences after word segmentation, and then converted into The vector form is input into the encoder for learning and training. This method can learn the connection between the source language and the target language very well, but it cannot meet all translation needs. In some scenarios that require high-precision translations, the following translations may appear inaccura...

AI Technical Summary

Problems solved by technology

[0014] Aiming at the deficiencies in the prior art that directly introducing the user dictionary into the neural network-based machine translation model does not get a go

Images