Research And Implementation Of The Word Representation Model For Neural Machine Translation

With the rapid development of computer text processing,word representation has also become one of the research hotspots.Many research works have shown that word representation can improve the performance of many natural language processing tasks,such as neural machine translation.As the basis of many tasks,word representation is receiving more and more attention.To this end,this thesis explores two methods to enhance word meaning representation to improve neural machine translation.The first one is the word representation that introduces internal structure of words.Traditional methods mostly use words or characters as the basic unit.To receive the advantages of both,some researchers have proposed a mixed method,but ignoring the internal hierarchy of Chinese words.Different from previous work,this thesis explores more reasonable semantic units of words.This thesis proposes a new standard that defines the internal hierarchical structure of Chinese words.Based on this,a corresponding annotation guideline is proposed and a Chinese word corpus is manually annotated,used to improve neural machine translation.The second one is the word meaning that introduces BERT a pre-training model.Recently,pre-training methods have performed well in many natural language tasks,in order to enhance the semantic representation of words,this thesis uses pre-trained model to learn the semantic representation of words,then added into neural machine translation model.This thesis has initially verified the effectiveness of the proposed method in the experiments.The work of this thesis provides new ideas for the study of word representation,which has important scientific significance and practical value for the research work of further exploring of natural language processing.