Research On Integration Strategy For Chinese Semantic Role Labeling

A main purpose of natural language understanding is that computers can understand human languages and communicate with human beings without trouble. As a shallow semantic analysis aiming at analyzing the semantic structure, semantic role labeling recognizes and classifies the arguments of a given verb. It has a lot of successful achievements in high level tasks of natural language processing, such as question and answer system, information extraction and etc.This paper proposes an integrated strategy based on the study of the original advantages and disadvantages of syntacitc analysis. CRFs are chosen as the model of machine learning. Shallow parsing has high precision and recall rate while providing relatively simple information for sematic role labeling. Our strategy promotes shallow parsing by adding more features extracted from full parsing which has lower precision and recall rate but can provide rich information for sematic role labeling. And the three systems which are shallow parsing, semantic role labeling and the improved semantic role labeling proposed in this paper are implemented.From the results a conclusion could be drawn that the performance of a semantic role labeling system has gained a great promotion by using the integrated strategy, whether based on automatic parsing or golden parsing. Particularly, the F score arises by over 10% under the condition of golden parsing. The improved system using the integrated strategy can obtain a bigger rise in performance when syntactic parsing is developed.