Translation Quality Assessment On Neural Machine Translation Based On Multidimensional Quality Metrics

With the development of machine learning and the updating of neural networks,various machine translation tools have also been emerging.There are also various problems of MT,like its fluency and fidelity.There also lacks a comprehensive and feasible evaluation system for assessing machine translation quality.This thesis draws on the framework of multidimensional quality metrics(MQM),selects part of White Paper on 6G Vision and Candidate Technologies published by the China Academy of Information and Communications Technology as the research text,and compares the results of Google Translate with the official English version word by word.The thesis uses quantitative and qualitative analysis to evaluate the translation quality of Google Translate from four dimensions,accuracy,fluency,terminology and verity.The research results show that the common error types of Google Translate results belong to terminology,accuracy,mistranslation and addition.It did not well in word segmentation,cohesion,and logic.Further studies are needed in the field of translation quality assessment and machine translation development.