Machine Translation Models Based On ANNs

The rapid development of modern science and technology has narrowed the space-time distance on the whole earth,and the international exchanges have become increasingly frequent and convenient,and the communication between different languages has become more frequent.However,the traditional manual translation and some inefficient translation methods can no longer meet people's needs for translation,people began to focus on the research of machine translation system.After more than sixty years of tortuous research,the accuracy of machine translation systems have made a qualitative leap,which is close to the level of human translation.However,the system of machine translation still faces many problems,and many aspects are still waiting to be further improved and explored.The translation performance of machine translation remains a serious challenge and needs to explore better new theories and techniques.This paper introduces rule-based,instance-based and statistics-based machine trans-lation systems in detail.The core issue of rule-based machine translation systems is to con-struct a complete rules system manually,but it is very difficult and extremely complicated to construct a complete rule system.The instance-based translation method is essentially a matching process whose quality of translation mainly depends on the size and coverage of the bilingual alignment corpus.Statistical machine translation has a good mathematical model,unsupervised learning ability and powerful knowledge of automatic acquisition,but it is very prone to data sparseness.Facing the challenge of SMT,a better solution is to build neural networks model.Machine translation using neural networks architecture is divided into two categories:one is still based on SMT as a framework,and the use of neural networks is to improve one of the key modules;the other is directly using neural networks to directly map the source sequence into the target language sequence.This paper introduces the most commonly used feedforward neural networks,recurrent neural networks(RNN),convolutional neural networks(CNN)and long short term memory networks(LSTM),as well as the language model or translation model based on a variety of neural networks model.Since 2016,the translation model based on neural networks has made rapid progress.Its translation perfor-mance is far beyond the SMT and has gradually become the core technology of commercial translation systems such as Google,Baidu and so on.We believe that people can overcome obstacles and achieve truly fully automated ma-chine translation systems,but until then we have a lot to continue researching.