Syntactic and Semantic Role Labeling for Chinese FrameNet Based on Cascaded Conditional Random Fields

The Chinese FrameNet Project is creating a lexical resource for Chinese, based on the principles of Frame Semantics and supported by corpus evidence. Due to the fact that syntactic and semantic role labeling (SSRL) is very necessary for deep natural language processing, a method based on cascaded conditional random fields (CCRFs) is proposed for the SSRL task, and the CCRFs model is trained to label the predicates' semantic roles, Phrase Types and Grammatical Functions in a sentence. The key of the methods is parameter estimation and feature selection. There are three category features for the CCRFs algorithm: features based on segmentation words, features based on the Part of Speech (POS) of the relative words, and features based on the position relative to the targets. Evaluation on the datasets of the pre-release version of Chinese FrameNet shows that the method can obtain satisfying performance and can achieve 70.45% F for syntactic & semantic role labeling.