Pronoun co-reference resolution is an indispensable part for completely comprehending the spoken dialogue. Based on the characteristics of difference of spoken dialogue to written forms, and of antecedents of non-nominal word classes co-references, in this paper we propose a set of algorithms suitable to resolving non-nominal word classes co-reference in raw corpus of spoken dialogue on the basis of the works previously done by the others. The algorithms are based on the Right Frontier Rule of co-reference of non-nominal word classes. In the . Algorithms, we propose some rules about how to distinguish whether the antecedent is linearly adjacent or hierarchically adjacent, and the rule to filter candidate antecedents. The algorithms are tested on a public released spoken dialogue corpus Trains-93. Experimental results show that our algorithms improve the precision and recall of resolution, it can resolve more kinds of pronouns, and fits the raw corpus of spoken dialogue.%代词指代消解是全面理解口语对话不可缺少的一部分.根据口语不同于书面语的特点以及非名词指代先行项的特点,在前人工作的基础上提出了一套适合于在口语对话生语料上消解非名词指代的算法.算法基于非名词指代的右边界规则理论,给出了判断候选先行项属于“线性紧邻”还是“层次紧邻”的判别方法,同时给出了候选先行项的过滤规则.算法在公开发布的口语对话语料Tranis-93上进行了测试,实验结果表明,算法提高了消解的正确率和召回率,能消解更多不同的代词,且适用于口语对话生语料.
展开▼
