Candidate Scoring Using Web-Based Measure for Chinese Spelling Error Correction

Chinese character correction involves two major steps: 1) Providing candidate corrections for all or partially identified characters in a sentence, and 2) Scoring all altered sentences and identifying which is the best corrected sentence. In this paper a web-based measure is used to score candidate sentences, in which there exists one continuous error character in a sentence in almost all sentences in the Bakeoff corpora. The approach of using a web-based measure can be applied directly to sentences with multiple error characters, either consecutive or not, and is not optimized for one-character error correction of Chinese sentences. The results show that the approach achieved a fair precision score whereas the recall is low compared to results reported in this Bakeoff.