A Corpus-driven Study Of Machine Translation Performance On Literary And Non-literary Text

Machine translation,as an emerging technology,has greatly facilitated people's lives,which gives rise to a widely held concern over whether machines have reached the intelligence level of human translators.Previous studies show that,to some extent,it is possible for machine translation to handle expository texts that mainly deliver objective information.But what opportunities and challenges may machine translation encounter when it comes to literary works that mainly convey human emotions? At present,there is not much systematic and quantitative research on this issue.To explore the current development of machine translation and assess its ability to translate literary works,the present study conducted a comparative study by testing machine translation performance on both literary text and non-literary text to explore if there was a significant gap between two types of machine translation.A corpus-driven analysis was carried out first to calculate three linguistic indicators,Type Token Ratio(TTR),Content Words Density and Average Sentence Length,then followed by case analysis to verify the results.The results show that the machine translated non-literary text is highly correlated with its raw text with significant positive correlation found in all three linguistic indicators,while between literary text and its machine translation only one linguistic indicator TTR reports significant positive correlation.Further,the case analysis result of many mistranslations and omissions existing in the output text indicates that machine translation performance on literary text is unable to meet users' expectations yet.Overall,the study concludes that the machine translation of non-literary text outperforms that of literary text.