Error analysis of Chinese-English machine translation on the clause-complex level

Error analysis is an important way to provide feedbacks for the performance of MT systems. This paper analyzes clause-complex level errors under the analytical framework of Clause Complex model, and accounts for the errors from the perspective of Chinese-English grammatical differences. The statistics reveals that probability of clause-complex level errors is relatively high and that the occurrence of errors is different across genres and positively related to the length of clause complex. It is hoped that the study could bring enlightenment for MT systems to process inter-clausal relationship in clause complexes.

I. Introduction

Machine translation (MT) technology has greatly developed over the past dozens of years. Along with its development, the methods for evaluating MT outputs have been developing. In the early stage of MT development, manual evaluation was the predominant method. However, as the statistical approach to machine translation is widely adopted, automatic evaluation by calculating the similarity between MT outputs and human translation becomes widely applied [1] [2].