Neutral machine translation method with copy mechanism

Abstract

The invention relates to a neutral machine translation method with a copy mechanism. The method is applied to an NMT model based on an attention mechanism and adopting an encoder-decoder framework. The method includes the steps that source phrases needing to be translated specifically and corresponding target phrases are marked in a source language and a target language in original training sentences; the source phrases, needing to be translated specifically, in the source language in the original training sentences are replaced with the corresponding target phrases in the target language in the original training sentences; the processed original training sentences are subjected to training of an NMT system; phrases needing to be translated specifically are marked in the to-be-translated source sentences. By means of the neutral machine translation method with the copy mechanism, specific phrases such as person names, place names, mechanism names and brand names can be well translated,the method can be compatible with any corpus processing technology, the translation effect is further improved, the structure of the NMT system does not need to be changed, and the method can be conveniently applied to any NMT system.

Description

technical field [0001] The neural machine translation involved in the present invention, in particular, relates to a neural machine translation method with a replication mechanism. 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 based on deep learning has attracted more and more attention. In the NMT field, one of the most commonly used translation models is the encoder-decoder model with an attention-based mechanism. The main idea is to encode the source 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 translation (hereinafter collectively referred to as 'target sentence'). In fact, this encoder-decoder framework is the core idea of ​​de...

AI Technical Summary

Problems solved by technology

This leads to a serious problem in the translation of NMT: when the source sentence to be translated contains words that are not in the vocabulary, UNK will be generated in the translation, resulting in low readability of the translation, especially in the training corpus. People's names, place names, brand words, etc. cannot be translated

This generation mode, when translating a given phrase, will lead to inconsistencies in the generated translation

Images