Chinese spelling check (CSC) is an essential issue in the research field of Chinese language processing (CLP). This paper describes the details of two CSC systems we developed to solve this problem. The first system was built based on CRF model, and the modules of such system include word segmentation, error detection and error correction. Another system was based on 2-Chars&&3-Chars model, and its modules include bigram segmentation, error detection and error correction. Using the final test data set provided by CLP2014, the final experimental result of the system based on 2-Chars&&3-Chars model was better, which achieved 0.403 detection accuracy with 0.3344 detection precision and 0.3964 correction accuracy with 0.3191 correction precision.
展开▼