Research On Neural Machine Translation Method Based On Semantic Concept

Machine translation is referred as the process of using a computer to convert one language into another.It is one of the core tasks of natural language processing(NLP)and has very important scientific research value.Neural machine translation(NMT)has become the mainstream machine translation method due to its superior performance,and it is one of the most successful applications of deep learning in the field of NLP.The high performance of neural machine translation systems typically relies on high quality,large-scale training data and powerful computing resources.Thus,it is a data-driven translation method,and this method faces a variety of problems in low-resource languages.In order to solve such problems and boost the theory of machine translation,inspired by the use of data and knowledge,this work focuses on how to integrate linguistic knowledge into the NMT model to improve the translation performance and translation quality.First,to address the low-frequency words and out-of-vocabulary words problems,this paper proposes a method that integrates the knowledge of semantic concept into NMT system.Second,this paper also proposes a read-type modeling and decoding method that integrates external knowledge base to address the problem of ambiguity understanding in NMT.The experimental results show that the method proposed in this paper can effectively improve the translation performance of NMT system.The innovations and main research results of this paper are as follows:(1)A "replace-translate-restore" rare word processing strategy based on sematic concept is proposed.Compared with the traditional method,this method models the low-frequency words and the out-of-vocabulary words from the semantic level,improving the translation accuracy of low-frequency words and out-of-vocabulary words,and effectively relieving the problem of under translation or incorrect translation of low frequency words and out-of-vocabulary words.This research designs three models that are used to calculate the semantic similarity in integrating the knowledge of semantic concept into NMT,and the accuracy of calculating the semantic similarity is improved.Furthermore,the translation of low-frequency words and out-of-vocabulary words in the NMT system are enhanced also.(2)A read-type modeling and decoding method based on external knowledge base is proposed.The method utilizes the graph attention mechanism and dynamically integrates the semantic knowledge triplet information into the NMT model.The experimental results show that this work can improve the semantic discrimination ability and the translation performance of the NMT model.The main contributions of this paper are as follows:This paper proposes to integrate the semantic concept and the knowledge of the external knowledge base into NMT system,and expands the related basic theories of NMT through method innovation and experimental verification.