Research On Automatically Labeling Of CFN Semantic Roles

Along with the advent of the computer and the application in NLP improved the rate and quality of linguistic information. As deeply semantic parsing under the current technical conditions is difficult to achieve. The scholars began to focus on simple and practical tasks, and then shallow semantic parsing has been gradually paid attention by researchers. Semantic roles labeling is feasible of the program.We apply the CFN as corpus resource, which is constructed by Shanxi University. The CFN is based on FrameNet, referring to FrameNet and it is a knowledge database which is Chinese lexical semantic based on Chinese true corpus and offering computer applying.A conditional random fields classifier is used in the semantic role labeling system, which takes word as the labeled units. The conditional random fields classifier is trained to identify and classify the predicates' semantic roles at the same time. In the sequence labeling model, we choose CRF algorithm for labeling words with semantic role tags. At last, we get a group of the best features through comparing the result of different feature template experiment.Applying P, R, F_β=1 as evaluation criterion. The result shows that the fact is not true that applying feature more and result is good. For example applying base chunk makes P, R, F_β=1 lower. In this paper, we choose a group feature and make the arithmetic apply in actual problem better.