| In 2016,Google released its disruptive Neural Machine Translation(GNMT)system,propelling the increasingly extensive application of such system in conference interpretation.However,the empirical research seems lagged behind.Empirical research mostly focuses on neural network machine translation quality assessment,rarely involves interpretation quality assessment,and mainly uses evaluation criteria of computer science,which is abstract and has no linguistic relavance.Besides,researches on interpretation quality assessment mainly focus on text translation assessment,and rarely touches the impact of speech recognition on interpretation quality.Therefore,this paper selects original Chinese and English materials under the theme of energy and technology,uses Google,Sogou,Baidu and iFly's speech recognition and nural machine translation systems as tools,and conducts an experiment to compare the translation results of speech recognition text and precise original text.This paper uses the interpretation quality evaluation standard of linguistic significance.The purpose is to study the influence of speech recognition on the quality of neural network machine interpretation,and to put forward suggestions for improvement on its speech recognition and translation strategies.This paper finds out that the negative impact of speech recognition on the quality of neural network interpretation is threefold:first,the high error rate of term recognition leads to greater level of mistranslation.Second,the speech recognition results without filtering the spoken redundancy results in unsmooth expression or incorrect translation.Third,ircorrectly-recognized punctuation marks as well as unproperly-expressed nouns and numbers leads to grammatical errors in interpretation.Correspondingly,this paper raises several pieces of suggestion on voice recognition strategies by neural machine translation system as for improving the ultimate interpretation quality when put into practical use. |
PDF Full Text Request