The authors state an automatic method for semantic role labeling consistency calculation, based on the features of annotated corpus’ format, structure, content and user performances. The expriment shows that the proposed method is fast, stable and has high recall rate, and it can greatly improve the quality and efficiency.

It is necessary to grasp the main structure of the sentence through its predicate head for the sentence similarity calculation in EBMT. Taking 3000 tagged Chinese simple sentences as training set and the syntactic attributes and the contextual information as the classification features, this research acquires the knowledge of recognizing the predicate head through constructing a statistical decision tree model. The problem of applying the statistical decision tree to recognize the predicate head is also discussed.