基于树核函数,提出了从使用中心理论、集成竞争者信息和融入语义角色相关信息这3个方面对结构化句法树进行动态扩展来提升中英文代词消解的性能.首先探索了3种基本结构化句法树捕获方案,并使用SVMLight中提供的卷积树核函数直接进行基于结构化句法树的相似度计算,从而完成指代消解任务;其次,在分析3种结构化句法树捕获方案的基础上,从中心理论、竞争者信息和语义角色相关信息等几方面对捕获的结构化句法树进行了扩展;最后,通过ACE 2004 NWIRE英文语料和ACE 2005 NWIRE中文语料上的实验,说明了这些扩展能够提升代词消解的性能.
This paper proposes a tree kernel method to anaphora resolution of pronouns in both English and Chinese.First,several basic structured tree spans are proposed according to linguistic intuition.The similarity between two structured objects is computed directly using SVMLight.Then,a dynamic-expansion scheme is proposed to automatically determine a proper tree span for pronoun resolution by the centering theory,antecedent competitor-related information,and semantic role-related information.Evaluation on both the ACE 2004 English NWIRE corpus and the ACE 2005 Chinese NWIRE corpus justified the effectiveness of this method.