Adaptive neural network machine translation method based on unsupervised field

Abstract

The invention provides an adaptive neural network machine translation method based on an unsupervised field. The method comprises the following steps that: taking the vector representation of the last word and the first word of a source end sentence in a bilingual corpora training sample as the input of a Softmax classifier and a translation module to be subjected to training; and according to a domain number generated by the Softmax classifier, generating the number of translation network decoders, and generating a target end and a corresponding domain on the basis of the decoder of the target end. By use of the method, the problem in the prior art that labelled domain data is in shortage is overcome, time and cost are saved, adaption between translation and domains can be efficiently finished, and the method is high in practicality and has a good applicable range and expandability.

Description

technical field [0001] The present invention relates to the technical fields of machine learning and machine translation, and more particularly, relates to a neural network machine translation method based on unsupervised domain self-adaptation. Background technique [0002] At present, with the deepening of international exchanges, people's demand for language translation is increasing day by day. However, there are many kinds of languages ​​in the world, and the Internet has become the most convenient platform for obtaining information. Users have increasingly urgent needs for online translation. Each has its own characteristics and flexible forms, making automatic language processing, including machine translation between languages, a difficult problem to be solved. At the same time, how to provide users with high-quality translation services has become a difficult problem to be solved. There are many kinds of languages ​​on the Internet, each language has a lot of poly...

AI Technical Summary

Problems solved by technology

[0005] However, in the data weighting method in the prior art, the method assigns weights to those sentences according to the similarity with the corpus in the domain; the above-mentioned prior art is inseparable from the serious problem of labeling the corpus, and the original training corpus needs to be divided into several A small component leads to complex operations such as increasing the number of model parameters, which in turn reduces the performance of neural network machine translation, translates inefficiently, and cannot accurately obtain self-adaptation between various fields.

Images