Abstract: This paper studies a novel approach to emend the result of the recognizer by applying the prior information on syllable confusion, named as Syllable Confusion Dictionary(SCD), which is estimated by the error of the recognizer. The approximated SCD is divided into two categories as context-dependent (CD) and context-independent (CI). As prior information, it characterizes the quality of pronunciation, recognition and their relation synthetically. In our experiments, we apply such a CD/CI SCD to a recognizer, which is used to recognize the putonghua with Shanghai accent, and get a relative improvement of 15%/20%. What is more, our study show that, combined with the MAP(Maximum a Posteriori) algorithm, this method can get better efficacy.