Assessing The Machine Translation Output By Internet Literature-Specific Machine Translators Based On Multidimensional Quality Metrics

In recent years,China has been actively promoting its culture to a global audience,and Internet novels have become a significant component of this effort.Among various genres,fantasy Internet novels have gained particular popularity abroad.However,due to the sheer volume of Internet novels,which can be several million characters long,and the demand for frequent updates,it is difficult for translators to ensure both quantity and quality.To address this challenge,an increasing number of language service providers have turned to machine translation,and Baidu has also launched its domain-specific machine translator specifically designed for Internet novels.In this context,this paper aims to investigate the performance of the domain-specific engine in translating fantasy Internet novels,identify the common types of errors,explore their underlying causes,and propose methods to enhance the quality of machinetranslated texts.Following the Multidimensional Quality Metrics(MQM)framework,this research adopts a mixed-methods approach that combines quantitative and qualitative analyses.Fantasy novel texts that the authors translated for Company A’s web novel translation project are selected as the source text,and translations reviewed by professional translators are used as references.The study evaluates the output of Baidu Translate’s domain-specific translator by identifying and categorizing the errors.The word error rate was measured by dividing the penalty score by the total word count.To calculate the overall quality score,the word error rate was subtracted from a full score of1.The domain-specific translator achieved an overall quality score of 0.68.The results indicate that the accuracy and terminology of machine-translated texts still require significant improvement,particularly regarding untranslated,mistranslated,and terminology errors.The causes of these errors are twofold: bottlenecks in the machine translation engine itself(e.g.,heavy reliance on language resources,inability to understand the context)and the linguistic characteristics of the source text.Therefore,machine translation developers should improve the quality and quantity of training data,and translators can enhance the quality of machine-translated texts by conducting preediting,importing term bases,and utilizing translation memory.To conclude,while domain-specific machine translator can significantly enhance the efficiency of web novel translation,there remain noteworthy deficiencies and vast opportunities for improvement.The future calls for a collaborative and complementary approach between humans and machines.