Key Problems in Conversion from Simplified to Traditional Chinese Characters

In this paper we tackle the problem of character conversion from simplified Chinese to traditional Chinese. Of those simplified characters that need conversion, about 9.5% of them have more than 2 counterparts in the traditional scripts. We improve upon the previous log-linear approach first used in (Chen et al 2011) by utilizing more data sets and better translation models. We also show that automatic classification and noise reduction of corpus can achieve better performance. As a proof of the validity of our approach, we scored No. 1 in a recent evaluation of simplified to traditional character conversion systems organized by the Chinese Information Processing Society of China.