Neural machine translation method based on sample guidance

Abstract

The invention discloses a neural machine translation method based on sample guidance. The invention relates to a neural machine translation method based on sample guidance, which comprises the following steps of assisting the translation of a source statement through the translation of a sentence similar to the source statement, finding the sample statement similar to the source statement in a sample database, and marking the sample statement as (x, xm), wherein x represents the source statement. The method for guiding translation of the sample by introducing the sample into the neural machinetranslation model has the advantages that the useless noise information in a sample statement is masked off through a noise-masked encoder model, and the interference of the useless information on source statement translation can be effectively avoided; and 2, through the auxiliary decoder model, which information in the model sample statement can be used can be displayed and guided, so that theinformation in the target end sample statement can be fully used.

Description

technical field [0001] The invention relates to the neural field, in particular to an example-based neural machine translation method. Background technique [0002] With the improvement of computer computing power and the application of big data, deep learning has been further applied. Neural Machine Translation (NMT) based on deep learning has attracted more and more attention. In the field of NMT, one of the most commonly used translation models is an attention-based encoder-decoder model. The main idea is to encode the sentence to be translated (collectively referred to as 'source sentence' hereinafter) into a vector representation through an encoder, and then use a decoder to decode the vector representation of the source sentence and translate it into its Corresponding translations (collectively referred to as 'target sentences' hereinafter). [0003] In some special application scenarios, before translating the source sentence, a sentence similar to the source sente...

AI Technical Summary

Problems solved by technology

[0008] The method of combining statistical machine translation and translation memory involves intervention during translation, and NMT is a sequence-to-sequence structure. At present, intervention in NMT is still an unsolved problem, so the method in statistical machine translation is not applicable to NMT

[0009] The previous work combining NMT and translation memory can only match similar sentences when retrieving the translation memory, and then use the whole or part of the matched content to guide the translation, but if the whole information is used, because other parts of the sentence are different from The source sentence does not match, which inevitably introduces noise, and if only part of the matching content is used, then the information in the translation memory cannot be fully used

Images