中文摘要：

复句关系词特征分析是复句关系词自动标识的关键.论文提出基于依存树的复句关系词特征分析方法,通过依存树核函数来捕获复句句法结构特征.对于复句句法分析所得依存树,选取最短路径包含树SPT（Short Path Tree）从树中选取关系词特征序列,设计依存树核函数对特征序列进行相似度的计算.实验证明,运用该方法进行关系词自动识别,具有较高的正确率.

英文摘要：

Feature analysis of relation words in Chinese complex sentences is the core work in automatic identification of rela-tion words in complex sentence. In this paper,a feature analysis method of relation words based on dependency tree kernel is pro-posed. In the method,grammar tree features of complex sentences are extracted by tree kernel function. A Short Path Tree（SPT）is used for features extract. A tree kernel function is designed for similar computation of features. Experiment shows the method achieves a very good effect.