Conditional Random Field-based Parser and Language Model for Tradi-tional Chinese Spelling Checker

This paper describes our Chinese spelling check system submitted to SIGHAN Bake-off 2013 evaluation. The main idea is to exchange potential error character with its confusable ones and rescore the modified sentence using a conditional random field (CRF)-based word segmentation/part of speech (POS) tagger and a tri-gram language model (LM) to detect and correct possible spelling errors. Experimental results on the Bakeoff 2013 tasks showed the proposed method achieved 0.50 location detection and 0.24 error location F-scores in subtask1 and 0.49 location and 0.40 correction accuracies and 0.40 correction precision in subtask2.