Chinese CCG Syntactic Analysis Based On Heuristic Search And Pre - Marking

Combinatory Categorial Grammar (CCG) is a lexicalized grammar formalism. It provides a transparent interface between underlying semantic representation and sur-face syntax by first-order logic. So it is suitable for those fields which need semantic analysis. Meanwhile CCG provides flexible expression; it can express some compli-cated grammatical phenomenon and recover local or non-local dependency. Conse-quently, CCG plays an important role in these applications like semantic parsing, text inference and automatic question answering.Efficient parsing of Combinatory Categorial Grammar is a difficult problem in the NLP field. Compared with normal PCFG, the parsing of CCG is much harder. The main reason is that the number of nonterminal and rules is generally much larger than normal PCFG, so the complexity of parsing is much higher.Since Chinese CCG is difficult to parse, this paper investigates how to integrate two independent techniques on Chinese CCG parsing. Firstly supertagging is used within a log-linear model to eliminate these words whose probability is low, so the search space is pruned. Secondly A*search is applied to accelerate the parsing pro-cedure. At last evaluations are both done on effectiveness and efficiency.During the research, we mainly achieve the following research result.1. To the best of our knowledge, it is the first time to build a parser for Chinese CCG. We use generative models to make Chinese CCG as a PCFG.2. Use supertagging and A*search to accelerate the parsing procedure. Improve the efficiency of Chinese CCG parsing. Meanwhile we prove that supertagging can amend the inaccurate bias of the origin probability model and algorithm.3. Evaluate our method on Chinese CCG Bank. Make a quantitative analysis of the improvement of the algorithm. Experiments indicate parsing algorithm based on A*search and supertagging can significantly improve the efficiency and accuracy.