Error Analysis of English-Chinese Machine Translation

Abstract

In order to explore a practical way of improving machine translation (MT) quality, the error types and distribution of MT results have to be analyzed first. This paper analyzed English-Chinese MT errors from the perspective of naming-telling clause (NT clause, hereafter). Two types of text were input to get the MT output: one was to input the whole original English sentences into an MT engine; the other was to parse English sentences into English NT clauses, and then input these clauses into the MT engine in order. The errors of MT output are categorized into three classes: incorrect lexical choices, structural errors and component omissions. Structural errors are further divided into SV-structure errors and non-SV-structure errors. The analyzed data shows firstly, the major errors are structural errors, in which non-SV-structural errors account for a larger proportion; secondly, translation errors decrease significantly after English sentences are parsed into NT clauses. This result reveals that non-SV clauses are the main source of MT errors, and suggests that English long sentences should be parsed into NT clauses before they are translated.