基于规则的中文零指代项识别研究

摘要

A rule-based approach for Chinese zero anaphor detection was proposed. Given a parse tree, the smallest IP sub-tree covering the current predicate was captured. Based on this IP sub-tree,some rules were proposed for detecting whether a Chinese zero anaphor exists. This paper also systematically evaluated the rule-based method on OntoNotes corpus. Using golden parse tree,our method achieves 82. 45 in F-measure. And the F-measure is 63. 84 using automatic parser. The experiment results show that our method is very effective on Chinese zero anaphor detection.%提出了一个基于规则的中文零指代项识别方法,即输入一个句法分析树,根据这个句法分析树得到当前词的最小IP子树,再依据得到的IP子树提出中文零指代识别的一些规则.所用的语料是Ontonotes.从实验结果可以看到,该方法在标准的句法分析树上F值能达到82.45％,在自动句法树上其也能达到66.45％.从实验结果可以看出,该方法在中文零指代识别上具有很好的性能.