Error Analysis Of Chinese-English Machine Translation Under The Framework Of Clause Complex Theory

The studies of machine translation began at 1940 s.Machine translation was first implemented with the rule-based approach.After 1990 s,the statistical approach gradually took the place of the rule-based approach to become the mainstream technology.In 2016,Google put forward neural network machine translation,making the machine-learning-based approach a new hot topic.Neural network machine translation has been proved to have significantly reduced the errors in machine translation outputs,and have improved the fluency of machine translation.Hence,many internet enterprises have made public their own neural machine translation systems online.From the rule-based approach to the statistical approach and then to the machine-learning-based approach,machine translation systems depend less and less on language analysis.Despite the latter two approaches' efficiency in improving the quality of machine translation,it is believed that there should be some deficiencies in them due to the lack of language analysis.The fact that there are still many errors in machine translation constitutes the best proof.Based on the observation of machine translation outputs,the current machine translation systems still cannot properly deal with long sentences.This is because the structures of long sentences are usually complicated and are different across languages.Hence,the thesis aims to explore the influence of Chinese-English grammatical differences on machine translation through error analysis in light of the Clause Complex Theory.The objective of this study is to explore the relationship between clause-complex level errors in Chinese-English machine translation and Chinese-English grammatical differences,so as to provide suggestions for improving the quality of machine translation.The study is about to answer 3 questions: how should clause-complex level errors be classified;what's the relationship between clause-complex level errors and Chinese-English grammatical differences;and,what are distributional features of clause-complex level errors.The research subject is 21 Chinese texts from 3 different fields,including news,encyclopedia and politics.The research procedure is as follows: segment texts into clause complexes under the guidance of Clause Complex Theory;input a clause complex into 4 machine translation systems,namely Google,Baidu,Sogou and NiuTrans;tag clause-complex level errors in 4 output sentences;classify all the errors and analyze each type of errors from the perspective of Chinese-English grammatical differences;do statistics and analyze statistical results.After qualitative analysis,it is found that: clause-complex level errors can be divided into morphological errors,sharing structure errors and logical relation errors;clause-complex level errors are related to Chinese-English grammatical differences,with the fundamental difference in parataxis and hypostasis resulting in differences at all levels,which further lead to clause-complex level errors.The statistical results show: the probability of sharing structure errors and logical relation errors are relatively high,and stack pattern errors account for the largest percentage in sharing structure errors;the number of clause-complex level errors is positively related to the size of clause complex;the probabilities of clause-complex level errors are different across different genres.The results of the study show that researchers in the field of machine translation should pay attention to the study of inter-clausal relationship,so as to further improve the quality of machine translation.