This paper describes details of NTOU Chinese spelling check system participating in CLP-2014 Bakeoff. Confusion sets were expanded by using two language resources, Shuowen and Four-Corner codes. A new method to find spelling errors in legal multi-character words was proposed. Comparison of sentence generation probabilities is the main information for error detection and correction. A rule-based classifier and a SVM-based classifier were trained to identify spelling errors. Two formal runs were submitted, and the rule-based classifier achieved better performance.
展开▼