Research On Semantic Role Labeling Based On Chinese Function Words Usage

Semantic role labeling is one of the most important research areas of semantic analysis. According to the characteristics that Chinese function words can show some of the syntactic information of Chinese sentences, this thesis presented and realized two kinds of methods to put Chinese function words usage feature into the system of semantic role labeling to improve their performance. Firstly, this thesis attempts to use Chinese function words usage feature to optimize the results of dependency parsing, and then use the optimized results of dependency parsing as the input for semantic role labeling, which method is also named as improved semantic role labeling method based on dependency parsing optimized by Chinese function words. Secondly, this thesis attempts to dig a number of single and combined features which are related to Chinese function words usage feature, and then use greedy algorithm to analyze a most efficient Chinese function words related feature set, and then add that feature set into the feature set of semantic role labeling system directly to optimize the performance of the result of semantic role labeling system, which method is also named as improved semantic role labeling method based on Chinese function words usage feature set. Experiments illustrate that, improved semantic role labeling method based on dependency parsing optimized by Chinese function words and improved semantic role labeling method based on Chinese function words usage feature set are both effective and efficient improved methods for semantic role labeling, and Chinese function words is a kind of effective and efficient feature for semantic role labeling.The main research work of this thesis includes:To construct the basic semantic role labeling system used in this thesis based on relevant research and technical achievements at present which are popular and high performance as well as on the previous semantic role labeling system;To present and realize improved semantic role labeling method based on dependency parsing optimized by Chinese function words based on relevant research of improvement of dependency parsing based on Chinese function words, which is use Chinese function words usage feature to optimize the results of dependency parsing, and then use the optimized results of dependency parsing as the input for semantic role labeling;To dig and present a number of single and combined features which are related to Chinese function words usage feature and can add into semantic role labeling system directly, and then tested the influence for each feature to optimize the system of semantic role labeling by experiment, and then use greedy algorithm to analyze a most efficient Chinese function words related feature set;To present and realize improved semantic role labeling method based on Chinese function words usage feature set, which adds the feature set made by the previous work into the feature set of semantic role labeling system directly to optimize the performance of the result of semantic role labeling system; and then compare the results between improved semantic role labeling method based on dependency parsing optimized by Chinese function words and improved semantic role labeling method based on Chinese function words usage feature set.