The compound sentence is an important unit of the Chinese sentence and its annotation is important to the research on comprehending Chinese texts.Identification of relation words is the basis of compound sentence annotation.Based on a comprehensive analysis of Chinese compounds corpus,this paper extracted features of relation words from their context and collocation.Those features are described in formulas.A combination of mutual information with information gains is used for selecting features and eliminating redundant features.The Bayesian model is used for training and testing feature sets....