A long complex sentence is in general composed of a number of clauses, and the dependencies among the clauses have a great effect on understanding of the sentence. In Korean or Japanese where a head word is always located at the end of a clause, it is...
A long complex sentence is in general composed of a number of clauses, and the dependencies among the clauses have a great effect on understanding of the sentence. In Korean or Japanese where a head word is always located at the end of a clause, it is especially difficult to analyze the clausal dependencies, while it is simple to parse a clause. The verbal endings are considered in these languages a key source in determining the dependency between two clauses, but are insufficient since some syntactic and semantic information within the clauses is not expressed with the endings. In this we propose a novel method to parse a long complex Korean sentence by segmenting it into a set of clauses and then analyzing their dependencies using their structural information. To get successful results when analyzing their dependencies, this task has been the target of various machine learning methods including a support vector machine. Kernel methods are usually used to analyze dependency relation and it is ed that they show high performance.
This proposes a method for analyzing dependencies among clauses in a long complex sentence. The proposed method consists in two steps. One is to detect clauses by analyzing dependencies in a long complex sentence, and it works on support vector machines with polynomial kernel. Next is to analyze dependencies among detected clauses using the proposed expressions and a composite kernel. The expressions are designed for dependency analysis of Korean clauses. The expressions adopt a composite kernel to detect any similarity among the clauses. The composite kernel consists of a parse tree kernel and a polynomial kernel. A parse tree kernel is used for treating structure information and a polynomial kernel is applied when using lexical information. In addition, the parse tree kernel for handling parse trees has the benefit of minimizing the information loss occurring when transforming a parse tree into a feature vector, can occur, As a result, very accurate similarity between parse trees. The proposed expressions are defined as three types. They are about within expressing layers in clause, relations between clauses, and inner clauses the each clause.
The experiment is processed in two steps : the first experiment is to detect clauses, and to analyze dependency between detected clauses.
The results of the experiment are that clause detection is achieved with 87.32% accuracy, and dependancy of clauses is detected with 83.32% accuracy.
，韩语论文题目，韩语论文网站 |
