Introducing More Features to Improve Chinese Shift-Reduce Parsing
Recent years, shift-reduce parsing has gained popularity for its efficiency, but its performance still has a gap with the state of art parsers. In this paper, we construct a baseline Chinese shift-reduce parser using the common features. According to the error patterns in the baseline system, we explore the use of constituent label features and Lexical Relation Pair (LRP) information. The parser is trained and evaluated on the Tsinghua Chinese Treebank. After using the new features the boundary F-score arises from 85.19％ to 86.25％. Additionally, we investigate the use of LRP extracted from raw text automatically, and the parser can also get a competitive result.
Hongxian Wang Qiang Zhou Liou Chen
Center for Speech and Language Technologies, Research Institute of Information Technology, Tsinghua Center for Speech and Language Technologies,Research Institute of Information Technology, Tsinghua U
国际会议
2011亚太信号与信息处理协会年度峰会(APSIPAASC 2011)
西安
英文
1-7
2011-10-18（万方平台首次上网日期，不代表论文的发表时间）