A UNL Deconverter for Chinese
Xiaodong Shi, Yidong Chen · 2005
Abstract. This paper describes the internal working of a novel UNL converter for the Chinese language. Three steps are involved in generating Chinese from UNL: first, the UNL expression is converted to a graph; second, the graph is converted to a number of trees. Third, a top-down tree walking is performed to translate each subtree and the results are composed to form a complete sentence. Because each node is visited exactly once, the algorithm is of linear time complexity and thus much faster than the standard deconverter provided by the UNL center. A manual evaluation effort was carried out which confirmed that the quality of the Deconverter output was better than that of the standard deconverter. 1