Research On Joint Syntactic And Semantic Parsing For Chinese

Semantic parsing plays a critical role in natural language processing. Due to the difficulty in deep semantic parsing, previous research mainly focuses on shallow semantic parsing. Given a sentence and a predicate (either a verb or a noun) in it, the task of shallow semantic parsing is to recognize and map all the word sequences in the sentence into their corresponding semantic arguments (roles) or non-argument. As a particular case of shallow semantic parsing, the well-defined semantic role labeling (SRL) has been drawing more and more attention due to its importance in deep natural language processing applications, such as information extraction, question answering, and machine translation. According to the predicate types, SRL could be divided into SRL for verbal predicates (verbal SRL, in short) and SRL for nominal predicates (nominal SRL, in short).During the past few years, parse tree-based methods have dominated the research in SRL. Moreover, previous research has shown that the state-of-the-art SRL systems depend heavily on the qualities of parse trees, and that the performance of nominal SRL lags significantly behind that of verbal SRL. These two issues become more apparent when Chinese language is considered. For example, the evaluation on Chinese PropBank and NomBank has shown that the performance of Chinese verbal (nominal) SRL drops from ~92 (70) in F1 measure to ~67 (57) when replacing golden parse trees with automatic ones. To alleviate the heavy dependence of semantic parsing (in this dissertation, semantic role labeling) on syntactic parsing, this dissertation begins with exploration of syntactic parsing and semantic parsing, and then takes effort on their joint learning.In detail, the main content of this dissertation includes: 1. The research on syntactic parsing. In particular, a hierarchical syntactic parsing model is proposed. This model divides the whole task into three consequent sub-tasks: part-of-speech tagging, chunking, and structural parsing. The idea behind is to construct the parse tree level by level in a bottom-up way. In each level, it first recognizes easy constituents and then leaves those complex ones until more information is ready.2. The research on SRL. First, a Chinese verbal SRL system is built with focus on exploring various kinds of flat features and structural features from a parse tree. Then, nominal SRL is studied extensively with focus on how to benefit from verbal SRL. This is done by both augmenting nominal SRL training instances with verbal ones and employing various kinds of information generated from a verbal SRL system. Finally, we explore the issue of nominal predicate recognition. Evaluation shows that both our verbal and nominal SRL systems outperform the state-of-the-art ones.3. Joint syntactic and semantic parsing (in this dissertation, semantic role labeling). This is done from two levels. Firstly, an integrated parsing approach is proposed to integrate semantic parsing into the syntactic parsing process. Secondly, semantic information generated by semantic parsing is incorporated into the hierarchical syntactic parsing model to better capture semantic information in syntactic parsing. Evaluation shows that our method effectively alleviates the heavy dependence of semantic parsing on syntactic parsing and much improves the performance of both syntactic and semantic parsing (in particular, semantic parsing).The major contribution of this dissertation includes: the proposal of the hierarchical syntactic parsing model with great flexibility in integrating with other natural language processing tasks; the pioneer research on improving the performance of nominal SRL by properly integrating useful features derived from a state-of-the-art verbal SRL system; the proposal of a joint syntactic and semantic parsing approach which effectively reduces the heavy dependence of semantic parsing on syntactic parsing. Evaluation shows that this work much improves the state-of-the-art in SRL, thereby exhibits a great reference value to the future research in SRL and related tasks.