Semantic Role Labeling For Questions Based On Chinese FrameNet

Semantic analysis plays an important role in natural language processing field and recently becomes a research hot spot and critical issue. As the deep semantic understanding of the context based on the current technology is difficult to achieve, especially the automatic parsing remains a bottleneck. Therefore many researchers begin to pay attention to the Shallow Semantic Parsing which is easier than deep semantic understanding but more practical. Semantic Role Labeling as an effective way of Shallow Semantic Parsing becomes more and more popular.With the popularity of search engines and extensive use of question answering system, semantic understanding of question sentence receives much more attention. Based on the Characteristics of Chinese Question and Chinese FrameNet (CFN), this paper proposed a method combined POS selection with Cascading Conditional random field to conduct semantic Role Labeling of Chinese Question. The main study of this paper includes the following aspects:Firstly, analyze the Characteristics of Chinese Question and introduce the CFN in details. From 3011 Shanxi Tourism Corpus and question Corpus of Harbin Institute of Technology, we find that Question has its own special trait, such as short length compared with declarative sentence, relative fixed interrogative, similar expressing structure and so on.Secondly, this paper proposed a method combined rule-based POS selection model with Statistics-based Cascading Conditional random field to conduct semantic Role Labeling of Chinese Question. Besides, a method using target words cooccurrence degree was presented to identify the correct frame for the sentence. Ideal experimental results were obtained.Finally, Effective features of semantic Role Labeling of Questions were analyzed according to the experimental results, and the difference between question and declarative sentence was discussed. Additionally, semantic Role Labeling of Chinese Question was tried to apply in Question Answering System for the Shan xi Tourism, which achieved good performance in question upstanding and answer extraction phase.