Method for fusing bilingual predefined translation pair into neural machine translation model

Abstract

The invention discloses a method for integrating a bilingual predefined translation pair into a neural machine translation model. The invention relates to a method for fusing bilingual predefined translation pairs into a neural machine translation model, which is applied to an NMT model based on an attention mechanism and using an encoder-decoder framework, and includes a target to fuse bilingualpredefined translation pairs (p, q) into a neural machine translation model, wherein the p appears in a sentence of a source end and the p needs to be correctly translated into q and appears in a sentence of a target end, and other words in the source end are correctly translated. The method has the beneficial effects that a sample is introduced into the neural machine translation model so as to guide the translation of the neural machine translation model, and the method has the following advantages that 1, the model learns the mode by using a tagging method, the ground connection of predefined phrases of a source end and a target end is established, and the possibility that the source end and the target end are successfully translated is improved.

Description

technical field [0001] The invention relates to the field of neural machine translation, in particular to a method for integrating bilingual predefined translation pairs into a neural machine translation model. Background technique [0002] Machine translation (Machine Translation, MT) is an important field in the field of natural language processing (NLP), which aims to use machines to translate one language into another. After years of development, MT has evolved from a rule-based method to a statistical method, and then to today's neural network-based neural machine translation (Neural Machine Translation, NMT). Generally speaking, like many other mainstream NLP tasks, NMT also adopts a sequence to sequence structure (Sequence to sequence, seq2seq), consisting of an encoder and a decoder. The encoder (encoder) encodes the sentence at the source into a vector representation, and then the decoder (decoder) generates the corresponding translation word by word according to t...

AI Technical Summary

Problems solved by technology

Modifying the model architecture solves the problem of phrases, but it does not guarantee that the predefined phrases must appear on the target side, and the modification is more complicated and difficult to reproduce

Although the method of modifying beam search (Beam Search) does not need to modify the structure of the network, when generating each word, it is necessary to decide whether to use the information in the bilingual predefined translation pair according to the current translation, which greatly reduces the cost of translation. speed

Images