Hits:

Indexed by:会议论文

Date of Publication:2007-01-01

Included Journals:CPCI-S

Page Number:29-34

Key Words:Chinese dependency analysis; Nivre's algorithm; support vector machines; preference learning; lowly lexicalized

Abstract:This paper improved the long-distance dependency model and root node finder. Support vector machines (SVMs) are applied to identify Chinese dependency structure and preference learning (PL) is applied to find root node. This paper also discussed the effectiveness of lexicalization in root node finding and dependency analyzing. Experiments using the Harbin University of Technology Corpus show that minimal or low lexicalization is sufficient for root node finding accuracy and dependency analyzing accuracy. And low lexicalization could improve the finding and analyzing efficiency.