Research On Feature-based Semantic Role Labeling Of Nominalized Predicates In Chinese

Semantic Role Labeling (SRL) plays a critical role in natural language processing and it's a particular case of shallow semantic parsing, which attempts to label predicate-related (either a verb or a noun) constituents in a sentence with semantic roles, such as agent, patient, temporal, locative, etc. According to the predicate type, SRL could be divided into SRL for verbal predicate and nominal predicate.Recently, There has been rather limited published work in nominal SRL and some issues were found in current approaches on nominal SRL. For example, those state-of-the-art SRL systems depend on the qualities of parse trees heavily, and the performance of nominal SRL lags behind that of verbal SRL singificantly, etc. These related issues become more apparent when Chinese language is considered. The feature-based method is the popular one in SRL, so that we also focus on the feature-based method and try to explore various kinds features and propose a novel method to improve the performance of nominal SRL.In detail, the main content of this dissertation includes:1. The research on constituent-based SRL of nominalized predicates in Chinese. Compared to that of verbal predicates,the structure between nominal predicates and their roles is more flexible and complex. Therefore, some new nominal predicate-specific features are explored to capture the structure information for nominal SRL. The experimental results show that our nominal SRL system outperforms the state-of-the-art ones. Finally, we augment the training data with verbal SRL instances to examine whether nominal SRL could benefit from verbal instances, however, the improvement is not statistically significant. 2. The research on how to combine constituent and dependency syntactic structures for SRL of nominalized predicates in Chinese. In addition to widely adopted features from constituent-based syntactic structure, dependency-based syntactic structures are introduced into nominal SRL. Comparing to the traditional SRL method relying on the constituent-based syntactic structure, our approach has a much richer set of syntactic features. We also discuss the complementary information between above two syntactic structures. The experimental results show that our nominal SRL system outperforms the state-of-the-art ones.The major contribution of this dissertation includes: The proposal of the effective features for SRL of nominalized predicates in Chinese from the constituent structure;the proposal of combining constituent and dependency syntactic structures for SRL of nominalized predicates in Chinese.Evaluation shows that this work much improves the-state-of-the-art in SRL of nominalized predicates in Chinese, thereby exhibits a great reference value to the future research in SRL of nominalized predicates in Chinese and related tasks.