Identification of reordering mark “de” in Chinese-English patent machine translation

DE is a ubiquitous character in Chinese, used for a wide range of modified construction. Chinese modifier and core connected by a reordering mark DE need to reorder when translated into English. As very flexible it becomes a major source of error in Chinese-English patent machine translation. Therefore the correct identification of reordering mark DE is very important for improving the translation performance. In this paper, basing on HNC theory we classify reordering mark DEs into two kinds and label them as MD1 and MD2. This paper presents a method based on semantic analysis of DE to exclude those DEs that will not cause modifier and core reordering when translated into English, then identifies the reordering mark MD1 and MD2. After integrating into a patent machine translation the result shows that our method can achieve accuracies 92.62% and 91.07%.