This paper proposes a semi-automatic method to detect segmentation errors in a manually annotated Chinese corpus in order to improve its quality further. A particular Chinese character string occurring more than once in a corpus may be assigned different segmentations during a segmentation process. Based on these differences our approach outputs the segmentation error candidates found in a segmented corpus and then on which the segmentation errors are identified manually. Segmentation error

more »

... of a gold standard corpus can be given using our method. In Peking University (PK) and Academic Sinica (AS) test corpora of Special Interest Group for Chinese Language Processing (SIGHAN) Bakeoff1, 1.29% and 2.26% segmentation error rates are detected by our method. These errors decrease the F-measure of SIGHAN Bakeoff1 baseline test by 1.36% in PK test data and 1.93% in AS test data respectively.