A Study On The Quality Of Neural Machine's C-E Interpretation In A Diagnostic Approach

Google released its Neural Machine Translation(NMT)system in November 2016.After more than a year of continuous improvement,coupled with speech recognition and synthesis technology,neural machine translation has moved beyond translation to interpretation.Meanwhile,though there are already some automatic evaluation standards for machine translation,they are very abstract and lack linguistic explanation.Other standards rarely focus on machine translation's interpretation quality.Therefore,this thesis aims to study the interpretation quality of neural machine translation and find out its weakness to help it improve.This thesis establishes four criteria for the evaluation of NMT's interpretation in a diagnostic approach,namely fluency,accuracy,fidelity and coherence and compared the four samples.It comes to the conclusion that due to the vagueness and loose structure of spoken Chinese,NMT has great difficulty in interpretation and still requires improvement.