| In recent years,artificial intelligence has developed rapidly,redrawing the boundary lines and blueprints of translation and interpretation.Machine interpretation based on neural networks and deep learning has attracted much attention in the industry and has left a presence in major international conferences,but the quality of speech recognition and interpretation is not up to our expectations,thus there is still a certain distance between it and professional interpreters.This study selected accessible and commonly used Sogou Translation APP and Youdao Translator APP as research tools,and transcripts of the Prime Minister’s answers to questions from Chinese and foreign journalists in 2021 as the source language corpus,and the Chinese-English consecutive interpretation conducted by the Ministry of Foreign Affairs interpreter,Sogou Translation and Youdao Translator as target language corpus.Then,it utilized AntConc 3.5.7 and SPSS to compare the differences in the use of discourse cohesion devices between manual and machine interpretations from the perspectives of semantics and pragmatics.The results show that the frequencies of reference,substitution and ellipsis cohesion devices in ZV are higher than those in machine interpreting texts such as SV and YV.As for conjunction devices,there is no significant difference in the frequencies of adversative conjunction,causal conjunction,or temporal conjunction,except that the amount of additive conjunction devices in ZV is relatively less than SV and YV.Except for the repetition of lexical cohesion devices,the frequencies of synonym,near-synonym,and superordinate in ZV are higher than those of S V and YV.On this basis,the thesis further explored the inappropriateness of grammatical cohesion and lexical cohesion in machine interpreting based on the self-built English corpus bank of government work reports.These inappropriateness include inadequate reference cohesion,excessive additive conjunction and insufficient substitution cohesion at the grammatical level,and mechanical repetition at the lexical level,etc.Moreover,semi-structured interviews were conducted to further confirm the credibility of research findings and conclusions,which ensured the reliability and validity of the experiment.In the end,this thesis expounded the research deficiencies,such as the experimental environment different from that of the real interpretation site,and the occasional errors in the judgment of cohesion devices.At the same time,referring to the shortcomings of existing studies,the study suggested that the future researches should diversify the corpus type and resort to corpus-driven technologies to explore other specific language problems in machine interpretation,hoping to provide linguistic references for further improving the quality of machine interpretation. |
