Research On Optimization Of Language Model Based On Statistical Machine Translation

Recently,the research of statistical machine translation is becoming more and more popular.The effect of machine translation has improved a lot,which makes it become a fast and convenient service for many people in many fields.Language model as an important part of statistical machine translation,can be used to test the fluency of the target language.It influences the choice of words,reorder,and other decisions in decoding.The purpose of this thesis is to optimize the language model in statistical machine translation in order to make better translation performance.This thesis focuses on the art of the state of language model,making comparison of translation performance between the traditional language model and the neural network language model,and aims at solving the problem of large time expense in model training.It proposes a method to improve model training,which will enhance the convergence speed and the model efficiency.Neural networks language model spends much time at the expense of training,an improved measures was proposed to improve the performance of the model and the model convergence rate during training.We use word2vec take part in the training of neural network language model,that it can learn more semantic information to improve the performance of the model.Word2vec is introduced into the model training,which is the kind of existing knowledge incorporated into the neural network language model to improve its efficiency.In addition,the thesis studies on the influence of parameters of neural network statistical language model.By experiment,we found that using ReLU activation function can improve model performance.In short,we learn that neural network model performance good in machine translation,.The theory of neural networks adding in machine translation,will bring new ideas and a vast world to machine translation research.