Skip to Main Content
In this paper, an approach is proposed for correcting article errors in English translation results in order to improve the performance of a MT system. We check the article and the singular/plural form of the headword in a NP at the same time. This is different from most of early researches in which only articles are considered. Our correcting algorithm is based on simple, viable n-gram model whose parameters can be obtained using the WWW search engine Google. Using much less features than those used in the early researches, we experimentally showed that our approach could perform the promising results with a precision of 86.2% on all classes of article errors.
Date of Conference: 30 Oct.-1 Nov. 2005