A Study On MT Quality Evaluation And PE Solutions

In the past decade,globalization and informatization have stimulated the demand for translation,bringing new opportunities as well as unprecedented challenges to the language service industry and promoting the development of the “CAT+MT+PE” mode.Differing from the traditional mode of using online machine translation systems,the new mode features high human-computer integration and can efficiently produce a translation with acceptable quality,which has been widely adopted by language service providers around the world.This thesis is a translation practice report on the English to Chinese translation of A Social History of the Media under the mode of “CAT+MT+PE”.As a monograph in the field of mass communication,the translation material is an informative text of historical facts,thus suitable to be translated in that mode.Given that the significance of CAT has been somewhat undervalued in previous studies whose emphasis is on“MT+PE”,the report has introduced some latest CAT tools involved in the translation practice and put forward some efficient editing techniques to better utilize those tools.The translation practice consists of three stages.Before translating,the translator reads some parallel texts,analyzes the source text,and selects the tools to be used in the practice.In the process of translating,the translator completes the PE practice with the help of the MT system and various CAT tools.After finishing translating,the quality of the translation is assured by Trados Quality Assurance and reviews from the supervisor and fellow students.During the reflection of the translation practice,a targeted evaluation framework including automatic evaluation metrics and customized error taxonomy has been established and adopted in the report for qualitative and quantitative analysis of the MT raw output.Referred to the post-edited translation,it is found that the BLEU and TER scores at the corpus level(51.28 and 39.15 respectively)reveal the moderate quality of machine translation and relatively low post-editing effort,showing the significance of post-editing in improving translation efficiency.As for the distribution of different MT errors,errors of fluency occupy the first place(54.5%)and errors of accuracy come second(33.0%),followed by locale violation errors(8.21%)as well as design errors(4.29%),among which fluency errors have the greatest impact on automatic scores.In view of the MT errors found in the translation practice report,corresponding post-editing solutions are proposed and elaborated with examples from the translation practice.It is to be hoped that this report can provide scholars and translators with some insights into machine translation and post-editing.