Research On Quality Evaluation Of English-Chinese Artificial Translation Based On Language Model

In today's world,no country can develop independently.The mutual influence and penetration of various cultures are increasing,and it has even promoted the progress of human society.Therefore,translation between different languages has become a key point in the process of economic,political and cultural exchange among countries.With the deepening of globalization,various translation activities are in full swing.Moreover,people now not only need efficient and fast translation,they also have higher and higher requirements for translation quality,which requires evaluation of translation quality.In the past,the evaluation of translation quality was mainly carried out manually,which was highly subjective.There were no definite standards for translation quality evaluation,and it was difficult to give timely results when facing a large number of translations.This inefficient process has gradually become a hindrance to people's translation research,so it is necessary to automatically evaluate the translation quality.At present,many research results of automatic translation evaluation come from the field of machine translation,but there are relatively few researches on automatic evaluation of manual translation.In order to improve the efficiency of manual translation and realize automatic evaluation of translation quality,this article will further develop this problem Explore.This article first summarizes the current status of the research on the evaluation of artificial translations in domestic and foreign academic circles by literature analysis.Then,by comparing the previous research content and research methods of scholars and combining the research tasks of this article,this paper proposes a new perspective.A translation quality evaluation method.We can observe that in some professional fields,if the translator does not have certain professional knowledge,even if the language level is high,the resulting translation is blunt and does not conform to the expression habits of the professional field,that is,each professional field has its own Express your habits.According to this feature,which is different from previous research,this article takes English-Chinese financial vocabulary translation as an example,and proposes a translation quality evaluation method based on neural network language model to evaluate the quality of manual translation results from the perspective of conforming to the expression habits of Chinese financial terms..The main method is to establish a translation quality evaluation model.The neural language model is constructed with LSTM,BiLSTM,GRU three structures to prevent the situation that the historical data cannot affect the current data during the training process and cause poor results.The performance of the neural network language model is evaluated through perplexity.After obtaining the language model,as the model learns the vocabulary expression habits in the Chinese financial field,feed the human translation results to this model to obtain an output sequence.Use edit distance to calculate the degree of difference between input and output strings or sequences.Through this difference,the translation quality of financial vocabulary can be evaluated.In the experimental part,first design experiments to examine the performance of various models.Experimental results prove that the neural language model based on BiLSTM structure has the best performance.Further,in the research of this paper,the test set was manually marked to evaluate the evaluation accuracy of the translation quality evaluation model.Experiments show that the accuracy of the translation quality evaluation model is 95.1%.It can be seen from the experimental results that since the neural language model is used to better learn the expression habits of financial terms,the translation quality is evaluated by calculating the editing distance.The final translation quality evaluation model yields the optimal translation and the optimal human annotation.The translation is basically consistent,which proves the practicability and effectiveness of the model.